xref: /freebsd/sys/netinet/tcp_input.c (revision 79ac3c12a714bcd3f2354c52d948aed9575c46d6)
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 #include <netinet/udp.h>
127 
128 #include <netipsec/ipsec_support.h>
129 
130 #include <machine/in_cksum.h>
131 
132 #include <security/mac/mac_framework.h>
133 
134 const int tcprexmtthresh = 3;
135 
136 VNET_DEFINE(int, tcp_log_in_vain) = 0;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
138     &VNET_NAME(tcp_log_in_vain), 0,
139     "Log all incoming TCP segments to closed ports");
140 
141 VNET_DEFINE(int, blackhole) = 0;
142 #define	V_blackhole		VNET(blackhole)
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
144     &VNET_NAME(blackhole), 0,
145     "Do not send RST on segments to closed ports");
146 
147 VNET_DEFINE(int, tcp_delack_enabled) = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
149     &VNET_NAME(tcp_delack_enabled), 0,
150     "Delay ACK to try and piggyback it onto a data packet");
151 
152 VNET_DEFINE(int, drop_synfin) = 0;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
154     &VNET_NAME(drop_synfin), 0,
155     "Drop TCP packets with SYN+FIN set");
156 
157 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
159     &VNET_NAME(tcp_do_prr_conservative), 0,
160     "Do conservative Proportional Rate Reduction");
161 
162 VNET_DEFINE(int, tcp_do_prr) = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
164     &VNET_NAME(tcp_do_prr), 1,
165     "Enable Proportional Rate Reduction per RFC 6937");
166 
167 VNET_DEFINE(int, tcp_do_lrd) = 0;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
169     &VNET_NAME(tcp_do_lrd), 1,
170     "Perform Lost Retransmission Detection");
171 
172 VNET_DEFINE(int, tcp_do_newcwv) = 0;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
174     &VNET_NAME(tcp_do_newcwv), 0,
175     "Enable New Congestion Window Validation per RFC7661");
176 
177 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
179     &VNET_NAME(tcp_do_rfc3042), 0,
180     "Enable RFC 3042 (Limited Transmit)");
181 
182 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
184     &VNET_NAME(tcp_do_rfc3390), 0,
185     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
186 
187 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
189     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
190     "Slow-start flight size (initial congestion window) in number of segments");
191 
192 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
194     &VNET_NAME(tcp_do_rfc3465), 0,
195     "Enable RFC 3465 (Appropriate Byte Counting)");
196 
197 VNET_DEFINE(int, tcp_abc_l_var) = 2;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
199     &VNET_NAME(tcp_abc_l_var), 2,
200     "Cap the max cwnd increment during slow-start to this number of segments");
201 
202 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
203     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
204     "TCP ECN");
205 
206 VNET_DEFINE(int, tcp_do_ecn) = 2;
207 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
208     &VNET_NAME(tcp_do_ecn), 0,
209     "TCP ECN support");
210 
211 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
213     &VNET_NAME(tcp_ecn_maxretries), 0,
214     "Max retries before giving up on ECN");
215 
216 VNET_DEFINE(int, tcp_insecure_syn) = 0;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
218     &VNET_NAME(tcp_insecure_syn), 0,
219     "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
220 
221 VNET_DEFINE(int, tcp_insecure_rst) = 0;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
223     &VNET_NAME(tcp_insecure_rst), 0,
224     "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
225 
226 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
227 #define	V_tcp_recvspace	VNET(tcp_recvspace)
228 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
229     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
230 
231 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
233     &VNET_NAME(tcp_do_autorcvbuf), 0,
234     "Enable automatic receive buffer sizing");
235 
236 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
238     &VNET_NAME(tcp_autorcvbuf_max), 0,
239     "Max size of automatic receive buffer");
240 
241 VNET_DEFINE(struct inpcbhead, tcb);
242 #define	tcb6	tcb  /* for KAME src sync over BSD*'s */
243 VNET_DEFINE(struct inpcbinfo, tcbinfo);
244 
245 /*
246  * TCP statistics are stored in an array of counter(9)s, which size matches
247  * size of struct tcpstat.  TCP running connection count is a regular array.
248  */
249 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
251     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
252 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
253 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
254     CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
255     "TCP connection counts by TCP state");
256 
257 static void
258 tcp_vnet_init(const void *unused)
259 {
260 
261 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
262 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
263 }
264 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265     tcp_vnet_init, NULL);
266 
267 #ifdef VIMAGE
268 static void
269 tcp_vnet_uninit(const void *unused)
270 {
271 
272 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
273 	VNET_PCPUSTAT_FREE(tcpstat);
274 }
275 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
276     tcp_vnet_uninit, NULL);
277 #endif /* VIMAGE */
278 
279 /*
280  * Kernel module interface for updating tcpstat.  The first argument is an index
281  * into tcpstat treated as an array.
282  */
283 void
284 kmod_tcpstat_add(int statnum, int val)
285 {
286 
287 	counter_u64_add(VNET(tcpstat)[statnum], val);
288 }
289 
290 #ifdef TCP_HHOOK
291 /*
292  * Wrapper for the TCP established input helper hook.
293  */
294 void
295 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
296 {
297 	struct tcp_hhook_data hhook_data;
298 
299 	if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
300 		hhook_data.tp = tp;
301 		hhook_data.th = th;
302 		hhook_data.to = to;
303 
304 		hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
305 		    tp->osd);
306 	}
307 }
308 #endif
309 
310 /*
311  * CC wrapper hook functions
312  */
313 void
314 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
315     uint16_t type)
316 {
317 #ifdef STATS
318 	int32_t gput;
319 #endif
320 
321 	INP_WLOCK_ASSERT(tp->t_inpcb);
322 
323 	tp->ccv->nsegs = nsegs;
324 	tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
325 	if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
326 	    (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
327 	     (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
328 		tp->ccv->flags |= CCF_CWND_LIMITED;
329 	else
330 		tp->ccv->flags &= ~CCF_CWND_LIMITED;
331 
332 	if (type == CC_ACK) {
333 #ifdef STATS
334 		stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
335 		    ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
336 		if (!IN_RECOVERY(tp->t_flags))
337 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
338 			   tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
339 		if ((tp->t_flags & TF_GPUTINPROG) &&
340 		    SEQ_GEQ(th->th_ack, tp->gput_ack)) {
341 			/*
342 			 * Compute goodput in bits per millisecond.
343 			 */
344 			gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
345 			    max(1, tcp_ts_getticks() - tp->gput_ts);
346 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
347 			    gput);
348 			/*
349 			 * XXXLAS: This is a temporary hack, and should be
350 			 * chained off VOI_TCP_GPUT when stats(9) grows an API
351 			 * to deal with chained VOIs.
352 			 */
353 			if (tp->t_stats_gput_prev > 0)
354 				stats_voi_update_abs_s32(tp->t_stats,
355 				    VOI_TCP_GPUT_ND,
356 				    ((gput - tp->t_stats_gput_prev) * 100) /
357 				    tp->t_stats_gput_prev);
358 			tp->t_flags &= ~TF_GPUTINPROG;
359 			tp->t_stats_gput_prev = gput;
360 		}
361 #endif /* STATS */
362 		if (tp->snd_cwnd > tp->snd_ssthresh) {
363 			tp->t_bytes_acked += tp->ccv->bytes_this_ack;
364 			if (tp->t_bytes_acked >= tp->snd_cwnd) {
365 				tp->t_bytes_acked -= tp->snd_cwnd;
366 				tp->ccv->flags |= CCF_ABC_SENTAWND;
367 			}
368 		} else {
369 				tp->ccv->flags &= ~CCF_ABC_SENTAWND;
370 				tp->t_bytes_acked = 0;
371 		}
372 	}
373 
374 	if (CC_ALGO(tp)->ack_received != NULL) {
375 		/* XXXLAS: Find a way to live without this */
376 		tp->ccv->curack = th->th_ack;
377 		CC_ALGO(tp)->ack_received(tp->ccv, type);
378 	}
379 #ifdef STATS
380 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
381 #endif
382 }
383 
384 void
385 cc_conn_init(struct tcpcb *tp)
386 {
387 	struct hc_metrics_lite metrics;
388 	struct inpcb *inp = tp->t_inpcb;
389 	u_int maxseg;
390 	int rtt;
391 
392 	INP_WLOCK_ASSERT(tp->t_inpcb);
393 
394 	tcp_hc_get(&inp->inp_inc, &metrics);
395 	maxseg = tcp_maxseg(tp);
396 
397 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
398 		tp->t_srtt = rtt;
399 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
400 		TCPSTAT_INC(tcps_usedrtt);
401 		if (metrics.rmx_rttvar) {
402 			tp->t_rttvar = metrics.rmx_rttvar;
403 			TCPSTAT_INC(tcps_usedrttvar);
404 		} else {
405 			/* default variation is +- 1 rtt */
406 			tp->t_rttvar =
407 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
408 		}
409 		TCPT_RANGESET(tp->t_rxtcur,
410 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
411 		    tp->t_rttmin, TCPTV_REXMTMAX);
412 	}
413 	if (metrics.rmx_ssthresh) {
414 		/*
415 		 * There's some sort of gateway or interface
416 		 * buffer limit on the path.  Use this to set
417 		 * the slow start threshold, but set the
418 		 * threshold to no less than 2*mss.
419 		 */
420 		tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
421 		TCPSTAT_INC(tcps_usedssthresh);
422 	}
423 
424 	/*
425 	 * Set the initial slow-start flight size.
426 	 *
427 	 * If a SYN or SYN/ACK was lost and retransmitted, we have to
428 	 * reduce the initial CWND to one segment as congestion is likely
429 	 * requiring us to be cautious.
430 	 */
431 	if (tp->snd_cwnd == 1)
432 		tp->snd_cwnd = maxseg;		/* SYN(-ACK) lost */
433 	else
434 		tp->snd_cwnd = tcp_compute_initwnd(maxseg);
435 
436 	if (CC_ALGO(tp)->conn_init != NULL)
437 		CC_ALGO(tp)->conn_init(tp->ccv);
438 }
439 
440 void inline
441 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
442 {
443 	INP_WLOCK_ASSERT(tp->t_inpcb);
444 
445 #ifdef STATS
446 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
447 #endif
448 
449 	switch(type) {
450 	case CC_NDUPACK:
451 		if (!IN_FASTRECOVERY(tp->t_flags)) {
452 			tp->snd_recover = tp->snd_max;
453 			if (tp->t_flags2 & TF2_ECN_PERMIT)
454 				tp->t_flags2 |= TF2_ECN_SND_CWR;
455 		}
456 		break;
457 	case CC_ECN:
458 		if (!IN_CONGRECOVERY(tp->t_flags) ||
459 		    /*
460 		     * Allow ECN reaction on ACK to CWR, if
461 		     * that data segment was also CE marked.
462 		     */
463 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
464 			EXIT_CONGRECOVERY(tp->t_flags);
465 			TCPSTAT_INC(tcps_ecn_rcwnd);
466 			tp->snd_recover = tp->snd_max + 1;
467 			if (tp->t_flags2 & TF2_ECN_PERMIT)
468 				tp->t_flags2 |= TF2_ECN_SND_CWR;
469 		}
470 		break;
471 	case CC_RTO:
472 		tp->t_dupacks = 0;
473 		tp->t_bytes_acked = 0;
474 		EXIT_RECOVERY(tp->t_flags);
475 		if (tp->t_flags2 & TF2_ECN_PERMIT)
476 			tp->t_flags2 |= TF2_ECN_SND_CWR;
477 		break;
478 	case CC_RTO_ERR:
479 		TCPSTAT_INC(tcps_sndrexmitbad);
480 		/* RTO was unnecessary, so reset everything. */
481 		tp->snd_cwnd = tp->snd_cwnd_prev;
482 		tp->snd_ssthresh = tp->snd_ssthresh_prev;
483 		tp->snd_recover = tp->snd_recover_prev;
484 		if (tp->t_flags & TF_WASFRECOVERY)
485 			ENTER_FASTRECOVERY(tp->t_flags);
486 		if (tp->t_flags & TF_WASCRECOVERY)
487 			ENTER_CONGRECOVERY(tp->t_flags);
488 		tp->snd_nxt = tp->snd_max;
489 		tp->t_flags &= ~TF_PREVVALID;
490 		tp->t_badrxtwin = 0;
491 		break;
492 	}
493 
494 	if (CC_ALGO(tp)->cong_signal != NULL) {
495 		if (th != NULL)
496 			tp->ccv->curack = th->th_ack;
497 		CC_ALGO(tp)->cong_signal(tp->ccv, type);
498 	}
499 }
500 
501 void inline
502 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
503 {
504 	INP_WLOCK_ASSERT(tp->t_inpcb);
505 
506 	/* XXXLAS: KASSERT that we're in recovery? */
507 
508 	if (CC_ALGO(tp)->post_recovery != NULL) {
509 		tp->ccv->curack = th->th_ack;
510 		CC_ALGO(tp)->post_recovery(tp->ccv);
511 	}
512 	/* XXXLAS: EXIT_RECOVERY ? */
513 	tp->t_bytes_acked = 0;
514 	tp->sackhint.delivered_data = 0;
515 	tp->sackhint.prr_out = 0;
516 }
517 
518 /*
519  * Indicate whether this ack should be delayed.  We can delay the ack if
520  * following conditions are met:
521  *	- There is no delayed ack timer in progress.
522  *	- Our last ack wasn't a 0-sized window. We never want to delay
523  *	  the ack that opens up a 0-sized window.
524  *	- LRO wasn't used for this segment. We make sure by checking that the
525  *	  segment size is not larger than the MSS.
526  */
527 #define DELAY_ACK(tp, tlen)						\
528 	((!tcp_timer_active(tp, TT_DELACK) &&				\
529 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
530 	    (tlen <= tp->t_maxseg) &&					\
531 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
532 
533 void inline
534 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
535 {
536 	INP_WLOCK_ASSERT(tp->t_inpcb);
537 
538 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
539 		switch (iptos & IPTOS_ECN_MASK) {
540 		case IPTOS_ECN_CE:
541 			tp->ccv->flags |= CCF_IPHDR_CE;
542 			break;
543 		case IPTOS_ECN_ECT0:
544 			/* FALLTHROUGH */
545 		case IPTOS_ECN_ECT1:
546 			/* FALLTHROUGH */
547 		case IPTOS_ECN_NOTECT:
548 			tp->ccv->flags &= ~CCF_IPHDR_CE;
549 			break;
550 		}
551 
552 		if (flags & TH_CWR)
553 			tp->ccv->flags |= CCF_TCPHDR_CWR;
554 		else
555 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
556 
557 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
558 
559 		if (tp->ccv->flags & CCF_ACKNOW) {
560 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
561 			tp->t_flags |= TF_ACKNOW;
562 		}
563 	}
564 }
565 
566 void inline
567 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
568 {
569 	cc_ecnpkt_handler_flags(tp, th->th_flags, iptos);
570 }
571 
572 /*
573  * TCP input handling is split into multiple parts:
574  *   tcp6_input is a thin wrapper around tcp_input for the extended
575  *	ip6_protox[] call format in ip6_input
576  *   tcp_input handles primary segment validation, inpcb lookup and
577  *	SYN processing on listen sockets
578  *   tcp_do_segment processes the ACK and text of the segment for
579  *	establishing, established and closing connections
580  */
581 #ifdef INET6
582 int
583 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
584 {
585 	struct mbuf *m;
586 	struct in6_ifaddr *ia6;
587 	struct ip6_hdr *ip6;
588 
589 	m = *mp;
590 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
591 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
592 		if (m == NULL) {
593 			*mp = m;
594 			TCPSTAT_INC(tcps_rcvshort);
595 			return (IPPROTO_DONE);
596 		}
597 	}
598 
599 	/*
600 	 * draft-itojun-ipv6-tcp-to-anycast
601 	 * better place to put this in?
602 	 */
603 	ip6 = mtod(m, struct ip6_hdr *);
604 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
605 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
606 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
607 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
608 		*mp = NULL;
609 		return (IPPROTO_DONE);
610 	}
611 
612 	*mp = m;
613 	return (tcp_input_with_port(mp, offp, proto, port));
614 }
615 
616 int
617 tcp6_input(struct mbuf **mp, int *offp, int proto)
618 {
619 
620 	return(tcp6_input_with_port(mp, offp, proto, 0));
621 }
622 #endif /* INET6 */
623 
624 int
625 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
626 {
627 	struct mbuf *m = *mp;
628 	struct tcphdr *th = NULL;
629 	struct ip *ip = NULL;
630 	struct inpcb *inp = NULL;
631 	struct tcpcb *tp = NULL;
632 	struct socket *so = NULL;
633 	u_char *optp = NULL;
634 	int off0;
635 	int optlen = 0;
636 #ifdef INET
637 	int len;
638 	uint8_t ipttl;
639 #endif
640 	int tlen = 0, off;
641 	int drop_hdrlen;
642 	int thflags;
643 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
644 	int lookupflag;
645 	uint8_t iptos;
646 	struct m_tag *fwd_tag = NULL;
647 #ifdef INET6
648 	struct ip6_hdr *ip6 = NULL;
649 	int isipv6;
650 #else
651 	const void *ip6 = NULL;
652 #endif /* INET6 */
653 	struct tcpopt to;		/* options in this segment */
654 	char *s = NULL;			/* address and port logging */
655 #ifdef TCPDEBUG
656 	/*
657 	 * The size of tcp_saveipgen must be the size of the max ip header,
658 	 * now IPv6.
659 	 */
660 	u_char tcp_saveipgen[IP6_HDR_LEN];
661 	struct tcphdr tcp_savetcp;
662 	short ostate = 0;
663 #endif
664 
665 	NET_EPOCH_ASSERT();
666 
667 #ifdef INET6
668 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
669 #endif
670 
671 	off0 = *offp;
672 	m = *mp;
673 	*mp = NULL;
674 	to.to_flags = 0;
675 	TCPSTAT_INC(tcps_rcvtotal);
676 
677 #ifdef INET6
678 	if (isipv6) {
679 		ip6 = mtod(m, struct ip6_hdr *);
680 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
681 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
682 		if (port)
683 			goto skip6_csum;
684 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
685 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
686 				th->th_sum = m->m_pkthdr.csum_data;
687 			else
688 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
689 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
690 			th->th_sum ^= 0xffff;
691 		} else
692 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
693 		if (th->th_sum) {
694 			TCPSTAT_INC(tcps_rcvbadsum);
695 			goto drop;
696 		}
697 	skip6_csum:
698 		/*
699 		 * Be proactive about unspecified IPv6 address in source.
700 		 * As we use all-zero to indicate unbounded/unconnected pcb,
701 		 * unspecified IPv6 address can be used to confuse us.
702 		 *
703 		 * Note that packets with unspecified IPv6 destination is
704 		 * already dropped in ip6_input.
705 		 */
706 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
707 			/* XXX stat */
708 			goto drop;
709 		}
710 		iptos = IPV6_TRAFFIC_CLASS(ip6);
711 	}
712 #endif
713 #if defined(INET) && defined(INET6)
714 	else
715 #endif
716 #ifdef INET
717 	{
718 		/*
719 		 * Get IP and TCP header together in first mbuf.
720 		 * Note: IP leaves IP header in first mbuf.
721 		 */
722 		if (off0 > sizeof (struct ip)) {
723 			ip_stripoptions(m);
724 			off0 = sizeof(struct ip);
725 		}
726 		if (m->m_len < sizeof (struct tcpiphdr)) {
727 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
728 			    == NULL) {
729 				TCPSTAT_INC(tcps_rcvshort);
730 				return (IPPROTO_DONE);
731 			}
732 		}
733 		ip = mtod(m, struct ip *);
734 		th = (struct tcphdr *)((caddr_t)ip + off0);
735 		tlen = ntohs(ip->ip_len) - off0;
736 
737 		iptos = ip->ip_tos;
738 		if (port)
739 			goto skip_csum;
740 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
741 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
742 				th->th_sum = m->m_pkthdr.csum_data;
743 			else
744 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
745 				    ip->ip_dst.s_addr,
746 				    htonl(m->m_pkthdr.csum_data + tlen +
747 				    IPPROTO_TCP));
748 			th->th_sum ^= 0xffff;
749 		} else {
750 			struct ipovly *ipov = (struct ipovly *)ip;
751 
752 			/*
753 			 * Checksum extended TCP header and data.
754 			 */
755 			len = off0 + tlen;
756 			ipttl = ip->ip_ttl;
757 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
758 			ipov->ih_len = htons(tlen);
759 			th->th_sum = in_cksum(m, len);
760 			/* Reset length for SDT probes. */
761 			ip->ip_len = htons(len);
762 			/* Reset TOS bits */
763 			ip->ip_tos = iptos;
764 			/* Re-initialization for later version check */
765 			ip->ip_ttl = ipttl;
766 			ip->ip_v = IPVERSION;
767 			ip->ip_hl = off0 >> 2;
768 		}
769 	skip_csum:
770 		if (th->th_sum && (port == 0)) {
771 			TCPSTAT_INC(tcps_rcvbadsum);
772 			goto drop;
773 		}
774 	}
775 #endif /* INET */
776 
777 	/*
778 	 * Check that TCP offset makes sense,
779 	 * pull out TCP options and adjust length.		XXX
780 	 */
781 	off = th->th_off << 2;
782 	if (off < sizeof (struct tcphdr) || off > tlen) {
783 		TCPSTAT_INC(tcps_rcvbadoff);
784 		goto drop;
785 	}
786 	tlen -= off;	/* tlen is used instead of ti->ti_len */
787 	if (off > sizeof (struct tcphdr)) {
788 #ifdef INET6
789 		if (isipv6) {
790 			if (m->m_len < off0 + off) {
791 				m = m_pullup(m, off0 + off);
792 				if (m == NULL) {
793 					TCPSTAT_INC(tcps_rcvshort);
794 					return (IPPROTO_DONE);
795 				}
796 			}
797 			ip6 = mtod(m, struct ip6_hdr *);
798 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
799 		}
800 #endif
801 #if defined(INET) && defined(INET6)
802 		else
803 #endif
804 #ifdef INET
805 		{
806 			if (m->m_len < sizeof(struct ip) + off) {
807 				if ((m = m_pullup(m, sizeof (struct ip) + off))
808 				    == NULL) {
809 					TCPSTAT_INC(tcps_rcvshort);
810 					return (IPPROTO_DONE);
811 				}
812 				ip = mtod(m, struct ip *);
813 				th = (struct tcphdr *)((caddr_t)ip + off0);
814 			}
815 		}
816 #endif
817 		optlen = off - sizeof (struct tcphdr);
818 		optp = (u_char *)(th + 1);
819 	}
820 	thflags = th->th_flags;
821 
822 	/*
823 	 * Convert TCP protocol specific fields to host format.
824 	 */
825 	tcp_fields_to_host(th);
826 
827 	/*
828 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
829 	 */
830 	drop_hdrlen = off0 + off;
831 
832 	/*
833 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
834 	 */
835         if (
836 #ifdef INET6
837 	    (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
838 #ifdef INET
839 	    || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
840 #endif
841 #endif
842 #if defined(INET) && !defined(INET6)
843 	    (m->m_flags & M_IP_NEXTHOP)
844 #endif
845 	    )
846 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
847 
848 	/*
849 	 * For initial SYN packets we don't need write lock on matching
850 	 * PCB, be it a listening one or a synchronized one.  The packet
851 	 * shall not modify its state.
852 	 */
853 	lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
854 	    INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB;
855 findpcb:
856 #ifdef INET6
857 	if (isipv6 && fwd_tag != NULL) {
858 		struct sockaddr_in6 *next_hop6;
859 
860 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
861 		/*
862 		 * Transparently forwarded. Pretend to be the destination.
863 		 * Already got one like this?
864 		 */
865 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
866 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
867 		    lookupflag, m->m_pkthdr.rcvif, m);
868 		if (!inp) {
869 			/*
870 			 * It's new.  Try to find the ambushing socket.
871 			 * Because we've rewritten the destination address,
872 			 * any hardware-generated hash is ignored.
873 			 */
874 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
875 			    th->th_sport, &next_hop6->sin6_addr,
876 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
877 			    th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
878 			    m->m_pkthdr.rcvif);
879 		}
880 	} else if (isipv6) {
881 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
882 		    th->th_sport, &ip6->ip6_dst, th->th_dport,
883 		    INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
884 	}
885 #endif /* INET6 */
886 #if defined(INET6) && defined(INET)
887 	else
888 #endif
889 #ifdef INET
890 	if (fwd_tag != NULL) {
891 		struct sockaddr_in *next_hop;
892 
893 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
894 		/*
895 		 * Transparently forwarded. Pretend to be the destination.
896 		 * already got one like this?
897 		 */
898 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
899 		    ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
900 		if (!inp) {
901 			/*
902 			 * It's new.  Try to find the ambushing socket.
903 			 * Because we've rewritten the destination address,
904 			 * any hardware-generated hash is ignored.
905 			 */
906 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
907 			    th->th_sport, next_hop->sin_addr,
908 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
909 			    th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
910 			    m->m_pkthdr.rcvif);
911 		}
912 	} else
913 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
914 		    th->th_sport, ip->ip_dst, th->th_dport,
915 		    INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
916 #endif /* INET */
917 
918 	/*
919 	 * If the INPCB does not exist then all data in the incoming
920 	 * segment is discarded and an appropriate RST is sent back.
921 	 * XXX MRT Send RST using which routing table?
922 	 */
923 	if (inp == NULL) {
924 		/*
925 		 * Log communication attempts to ports that are not
926 		 * in use.
927 		 */
928 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
929 		    V_tcp_log_in_vain == 2) {
930 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
931 				log(LOG_INFO, "%s; %s: Connection attempt "
932 				    "to closed port\n", s, __func__);
933 		}
934 		/*
935 		 * When blackholing do not respond with a RST but
936 		 * completely ignore the segment and drop it.
937 		 */
938 		if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
939 		    V_blackhole == 2)
940 			goto dropunlock;
941 
942 		rstreason = BANDLIM_RST_CLOSEDPORT;
943 		goto dropwithreset;
944 	}
945 	INP_LOCK_ASSERT(inp);
946 	/*
947 	 * While waiting for inp lock during the lookup, another thread
948 	 * can have dropped the inpcb, in which case we need to loop back
949 	 * and try to find a new inpcb to deliver to.
950 	 */
951 	if (inp->inp_flags & INP_DROPPED) {
952 		INP_UNLOCK(inp);
953 		inp = NULL;
954 		goto findpcb;
955 	}
956 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
957 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
958 	    ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) {
959 		inp->inp_flowid = m->m_pkthdr.flowid;
960 		inp->inp_flowtype = M_HASHTYPE_GET(m);
961 	}
962 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
963 #ifdef INET6
964 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
965 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
966 		goto dropunlock;
967 	}
968 #ifdef INET
969 	else
970 #endif
971 #endif /* INET6 */
972 #ifdef INET
973 	if (IPSEC_ENABLED(ipv4) &&
974 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
975 		goto dropunlock;
976 	}
977 #endif /* INET */
978 #endif /* IPSEC */
979 
980 	/*
981 	 * Check the minimum TTL for socket.
982 	 */
983 	if (inp->inp_ip_minttl != 0) {
984 #ifdef INET6
985 		if (isipv6) {
986 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
987 				goto dropunlock;
988 		} else
989 #endif
990 		if (inp->inp_ip_minttl > ip->ip_ttl)
991 			goto dropunlock;
992 	}
993 
994 	/*
995 	 * A previous connection in TIMEWAIT state is supposed to catch stray
996 	 * or duplicate segments arriving late.  If this segment was a
997 	 * legitimate new connection attempt, the old INPCB gets removed and
998 	 * we can try again to find a listening socket.
999 	 */
1000 	if (inp->inp_flags & INP_TIMEWAIT) {
1001 		tcp_dooptions(&to, optp, optlen,
1002 		    (thflags & TH_SYN) ? TO_SYN : 0);
1003 		/*
1004 		 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1005 		 */
1006 		if (tcp_twcheck(inp, &to, th, m, tlen))
1007 			goto findpcb;
1008 		return (IPPROTO_DONE);
1009 	}
1010 	/*
1011 	 * The TCPCB may no longer exist if the connection is winding
1012 	 * down or it is in the CLOSED state.  Either way we drop the
1013 	 * segment and send an appropriate response.
1014 	 */
1015 	tp = intotcpcb(inp);
1016 	if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1017 		rstreason = BANDLIM_RST_CLOSEDPORT;
1018 		goto dropwithreset;
1019 	}
1020 
1021 	if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1022 		rstreason = BANDLIM_RST_CLOSEDPORT;
1023 		goto dropwithreset;
1024 	}
1025 
1026 #ifdef TCP_OFFLOAD
1027 	if (tp->t_flags & TF_TOE) {
1028 		tcp_offload_input(tp, m);
1029 		m = NULL;	/* consumed by the TOE driver */
1030 		goto dropunlock;
1031 	}
1032 #endif
1033 
1034 #ifdef MAC
1035 	if (mac_inpcb_check_deliver(inp, m))
1036 		goto dropunlock;
1037 #endif
1038 	so = inp->inp_socket;
1039 	KASSERT(so != NULL, ("%s: so == NULL", __func__));
1040 #ifdef TCPDEBUG
1041 	if (so->so_options & SO_DEBUG) {
1042 		ostate = tp->t_state;
1043 #ifdef INET6
1044 		if (isipv6) {
1045 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1046 		} else
1047 #endif
1048 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1049 		tcp_savetcp = *th;
1050 	}
1051 #endif /* TCPDEBUG */
1052 	/*
1053 	 * When the socket is accepting connections (the INPCB is in LISTEN
1054 	 * state) we look into the SYN cache if this is a new connection
1055 	 * attempt or the completion of a previous one.
1056 	 */
1057 	KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1058 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1059 	if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1060 		struct in_conninfo inc;
1061 
1062 		bzero(&inc, sizeof(inc));
1063 #ifdef INET6
1064 		if (isipv6) {
1065 			inc.inc_flags |= INC_ISIPV6;
1066 			if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1067 				inc.inc_flags |= INC_IPV6MINMTU;
1068 			inc.inc6_faddr = ip6->ip6_src;
1069 			inc.inc6_laddr = ip6->ip6_dst;
1070 		} else
1071 #endif
1072 		{
1073 			inc.inc_faddr = ip->ip_src;
1074 			inc.inc_laddr = ip->ip_dst;
1075 		}
1076 		inc.inc_fport = th->th_sport;
1077 		inc.inc_lport = th->th_dport;
1078 		inc.inc_fibnum = so->so_fibnum;
1079 
1080 		/*
1081 		 * Check for an existing connection attempt in syncache if
1082 		 * the flag is only ACK.  A successful lookup creates a new
1083 		 * socket appended to the listen queue in SYN_RECEIVED state.
1084 		 */
1085 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1086 			/*
1087 			 * Parse the TCP options here because
1088 			 * syncookies need access to the reflected
1089 			 * timestamp.
1090 			 */
1091 			tcp_dooptions(&to, optp, optlen, 0);
1092 			/*
1093 			 * NB: syncache_expand() doesn't unlock
1094 			 * inp and tcpinfo locks.
1095 			 */
1096 			rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1097 			if (rstreason < 0) {
1098 				/*
1099 				 * A failing TCP MD5 signature comparison
1100 				 * must result in the segment being dropped
1101 				 * and must not produce any response back
1102 				 * to the sender.
1103 				 */
1104 				goto dropunlock;
1105 			} else if (rstreason == 0) {
1106 				/*
1107 				 * No syncache entry or ACK was not
1108 				 * for our SYN/ACK.  Send a RST.
1109 				 * NB: syncache did its own logging
1110 				 * of the failure cause.
1111 				 */
1112 				rstreason = BANDLIM_RST_OPENPORT;
1113 				goto dropwithreset;
1114 			}
1115 tfo_socket_result:
1116 			if (so == NULL) {
1117 				/*
1118 				 * We completed the 3-way handshake
1119 				 * but could not allocate a socket
1120 				 * either due to memory shortage,
1121 				 * listen queue length limits or
1122 				 * global socket limits.  Send RST
1123 				 * or wait and have the remote end
1124 				 * retransmit the ACK for another
1125 				 * try.
1126 				 */
1127 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1128 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1129 					    "Socket allocation failed due to "
1130 					    "limits or memory shortage, %s\n",
1131 					    s, __func__,
1132 					    V_tcp_sc_rst_sock_fail ?
1133 					    "sending RST" : "try again");
1134 				if (V_tcp_sc_rst_sock_fail) {
1135 					rstreason = BANDLIM_UNLIMITED;
1136 					goto dropwithreset;
1137 				} else
1138 					goto dropunlock;
1139 			}
1140 			/*
1141 			 * Socket is created in state SYN_RECEIVED.
1142 			 * Unlock the listen socket, lock the newly
1143 			 * created socket and update the tp variable.
1144 			 * If we came here via jump to tfo_socket_result,
1145 			 * then listening socket is read-locked.
1146 			 */
1147 			INP_UNLOCK(inp);	/* listen socket */
1148 			inp = sotoinpcb(so);
1149 			/*
1150 			 * New connection inpcb is already locked by
1151 			 * syncache_expand().
1152 			 */
1153 			INP_WLOCK_ASSERT(inp);
1154 			tp = intotcpcb(inp);
1155 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1156 			    ("%s: ", __func__));
1157 			/*
1158 			 * Process the segment and the data it
1159 			 * contains.  tcp_do_segment() consumes
1160 			 * the mbuf chain and unlocks the inpcb.
1161 			 */
1162 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1163 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1164 			    iptos);
1165 			return (IPPROTO_DONE);
1166 		}
1167 		/*
1168 		 * Segment flag validation for new connection attempts:
1169 		 *
1170 		 * Our (SYN|ACK) response was rejected.
1171 		 * Check with syncache and remove entry to prevent
1172 		 * retransmits.
1173 		 *
1174 		 * NB: syncache_chkrst does its own logging of failure
1175 		 * causes.
1176 		 */
1177 		if (thflags & TH_RST) {
1178 			syncache_chkrst(&inc, th, m, port);
1179 			goto dropunlock;
1180 		}
1181 		/*
1182 		 * We can't do anything without SYN.
1183 		 */
1184 		if ((thflags & TH_SYN) == 0) {
1185 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1186 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1187 				    "SYN is missing, segment ignored\n",
1188 				    s, __func__);
1189 			TCPSTAT_INC(tcps_badsyn);
1190 			goto dropunlock;
1191 		}
1192 		/*
1193 		 * (SYN|ACK) is bogus on a listen socket.
1194 		 */
1195 		if (thflags & TH_ACK) {
1196 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1197 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1198 				    "SYN|ACK invalid, segment rejected\n",
1199 				    s, __func__);
1200 			syncache_badack(&inc, port);	/* XXX: Not needed! */
1201 			TCPSTAT_INC(tcps_badsyn);
1202 			rstreason = BANDLIM_RST_OPENPORT;
1203 			goto dropwithreset;
1204 		}
1205 		/*
1206 		 * If the drop_synfin option is enabled, drop all
1207 		 * segments with both the SYN and FIN bits set.
1208 		 * This prevents e.g. nmap from identifying the
1209 		 * TCP/IP stack.
1210 		 * XXX: Poor reasoning.  nmap has other methods
1211 		 * and is constantly refining its stack detection
1212 		 * strategies.
1213 		 * XXX: This is a violation of the TCP specification
1214 		 * and was used by RFC1644.
1215 		 */
1216 		if ((thflags & TH_FIN) && V_drop_synfin) {
1217 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1218 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1219 				    "SYN|FIN segment ignored (based on "
1220 				    "sysctl setting)\n", s, __func__);
1221 			TCPSTAT_INC(tcps_badsyn);
1222 			goto dropunlock;
1223 		}
1224 		/*
1225 		 * Segment's flags are (SYN) or (SYN|FIN).
1226 		 *
1227 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1228 		 * as they do not affect the state of the TCP FSM.
1229 		 * The data pointed to by TH_URG and th_urp is ignored.
1230 		 */
1231 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1232 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1233 		KASSERT(thflags & (TH_SYN),
1234 		    ("%s: Listen socket: TH_SYN not set", __func__));
1235 		INP_RLOCK_ASSERT(inp);
1236 #ifdef INET6
1237 		/*
1238 		 * If deprecated address is forbidden,
1239 		 * we do not accept SYN to deprecated interface
1240 		 * address to prevent any new inbound connection from
1241 		 * getting established.
1242 		 * When we do not accept SYN, we send a TCP RST,
1243 		 * with deprecated source address (instead of dropping
1244 		 * it).  We compromise it as it is much better for peer
1245 		 * to send a RST, and RST will be the final packet
1246 		 * for the exchange.
1247 		 *
1248 		 * If we do not forbid deprecated addresses, we accept
1249 		 * the SYN packet.  RFC2462 does not suggest dropping
1250 		 * SYN in this case.
1251 		 * If we decipher RFC2462 5.5.4, it says like this:
1252 		 * 1. use of deprecated addr with existing
1253 		 *    communication is okay - "SHOULD continue to be
1254 		 *    used"
1255 		 * 2. use of it with new communication:
1256 		 *   (2a) "SHOULD NOT be used if alternate address
1257 		 *        with sufficient scope is available"
1258 		 *   (2b) nothing mentioned otherwise.
1259 		 * Here we fall into (2b) case as we have no choice in
1260 		 * our source address selection - we must obey the peer.
1261 		 *
1262 		 * The wording in RFC2462 is confusing, and there are
1263 		 * multiple description text for deprecated address
1264 		 * handling - worse, they are not exactly the same.
1265 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1266 		 */
1267 		if (isipv6 && !V_ip6_use_deprecated) {
1268 			struct in6_ifaddr *ia6;
1269 
1270 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1271 			if (ia6 != NULL &&
1272 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1273 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1274 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1275 					"Connection attempt to deprecated "
1276 					"IPv6 address rejected\n",
1277 					s, __func__);
1278 				rstreason = BANDLIM_RST_OPENPORT;
1279 				goto dropwithreset;
1280 			}
1281 		}
1282 #endif /* INET6 */
1283 		/*
1284 		 * Basic sanity checks on incoming SYN requests:
1285 		 *   Don't respond if the destination is a link layer
1286 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1287 		 *   If it is from this socket it must be forged.
1288 		 *   Don't respond if the source or destination is a
1289 		 *	global or subnet broad- or multicast address.
1290 		 *   Note that it is quite possible to receive unicast
1291 		 *	link-layer packets with a broadcast IP address. Use
1292 		 *	in_broadcast() to find them.
1293 		 */
1294 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1295 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1296 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1297 				"Connection attempt from broad- or multicast "
1298 				"link layer address ignored\n", s, __func__);
1299 			goto dropunlock;
1300 		}
1301 #ifdef INET6
1302 		if (isipv6) {
1303 			if (th->th_dport == th->th_sport &&
1304 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1305 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1306 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1307 					"Connection attempt to/from self "
1308 					"ignored\n", s, __func__);
1309 				goto dropunlock;
1310 			}
1311 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1312 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1313 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 					"Connection attempt from/to multicast "
1316 					"address ignored\n", s, __func__);
1317 				goto dropunlock;
1318 			}
1319 		}
1320 #endif
1321 #if defined(INET) && defined(INET6)
1322 		else
1323 #endif
1324 #ifdef INET
1325 		{
1326 			if (th->th_dport == th->th_sport &&
1327 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1328 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1329 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1330 					"Connection attempt from/to self "
1331 					"ignored\n", s, __func__);
1332 				goto dropunlock;
1333 			}
1334 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1335 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1336 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1337 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1338 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1339 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1340 					"Connection attempt from/to broad- "
1341 					"or multicast address ignored\n",
1342 					s, __func__);
1343 				goto dropunlock;
1344 			}
1345 		}
1346 #endif
1347 		/*
1348 		 * SYN appears to be valid.  Create compressed TCP state
1349 		 * for syncache.
1350 		 */
1351 #ifdef TCPDEBUG
1352 		if (so->so_options & SO_DEBUG)
1353 			tcp_trace(TA_INPUT, ostate, tp,
1354 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1355 #endif
1356 		TCP_PROBE3(debug__input, tp, th, m);
1357 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1358 		if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1359 		    iptos, port)) != NULL)
1360 			goto tfo_socket_result;
1361 
1362 		/*
1363 		 * Entry added to syncache and mbuf consumed.
1364 		 * Only the listen socket is unlocked by syncache_add().
1365 		 */
1366 		return (IPPROTO_DONE);
1367 	} else if (tp->t_state == TCPS_LISTEN) {
1368 		/*
1369 		 * When a listen socket is torn down the SO_ACCEPTCONN
1370 		 * flag is removed first while connections are drained
1371 		 * from the accept queue in a unlock/lock cycle of the
1372 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1373 		 * attempt go through unhandled.
1374 		 */
1375 		goto dropunlock;
1376 	}
1377 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1378 	if (tp->t_flags & TF_SIGNATURE) {
1379 		tcp_dooptions(&to, optp, optlen, thflags);
1380 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1381 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1382 			goto dropunlock;
1383 		}
1384 		if (!TCPMD5_ENABLED() ||
1385 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1386 			goto dropunlock;
1387 	}
1388 #endif
1389 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1390 
1391 	/*
1392 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1393 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1394 	 * the inpcb, and unlocks pcbinfo.
1395 	 *
1396 	 * XXXGL: in case of a pure SYN arriving on existing connection
1397 	 * TCP stacks won't need to modify the PCB, they would either drop
1398 	 * the segment silently, or send a challenge ACK.  However, we try
1399 	 * to upgrade the lock, because calling convention for stacks is
1400 	 * write-lock on PCB.  If upgrade fails, drop the SYN.
1401 	 */
1402 	if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0)
1403 		goto dropunlock;
1404 
1405 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1406 	return (IPPROTO_DONE);
1407 
1408 dropwithreset:
1409 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1410 
1411 	if (inp != NULL) {
1412 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1413 		INP_UNLOCK(inp);
1414 	} else
1415 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1416 	m = NULL;	/* mbuf chain got consumed. */
1417 	goto drop;
1418 
1419 dropunlock:
1420 	if (m != NULL)
1421 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1422 
1423 	if (inp != NULL)
1424 		INP_UNLOCK(inp);
1425 
1426 drop:
1427 	if (s != NULL)
1428 		free(s, M_TCPLOG);
1429 	if (m != NULL)
1430 		m_freem(m);
1431 	return (IPPROTO_DONE);
1432 }
1433 
1434 /*
1435  * Automatic sizing of receive socket buffer.  Often the send
1436  * buffer size is not optimally adjusted to the actual network
1437  * conditions at hand (delay bandwidth product).  Setting the
1438  * buffer size too small limits throughput on links with high
1439  * bandwidth and high delay (eg. trans-continental/oceanic links).
1440  *
1441  * On the receive side the socket buffer memory is only rarely
1442  * used to any significant extent.  This allows us to be much
1443  * more aggressive in scaling the receive socket buffer.  For
1444  * the case that the buffer space is actually used to a large
1445  * extent and we run out of kernel memory we can simply drop
1446  * the new segments; TCP on the sender will just retransmit it
1447  * later.  Setting the buffer size too big may only consume too
1448  * much kernel memory if the application doesn't read() from
1449  * the socket or packet loss or reordering makes use of the
1450  * reassembly queue.
1451  *
1452  * The criteria to step up the receive buffer one notch are:
1453  *  1. Application has not set receive buffer size with
1454  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1455  *  2. the number of bytes received during 1/2 of an sRTT
1456  *     is at least 3/8 of the current socket buffer size.
1457  *  3. receive buffer size has not hit maximal automatic size;
1458  *
1459  * If all of the criteria are met we increaset the socket buffer
1460  * by a 1/2 (bounded by the max). This allows us to keep ahead
1461  * of slow-start but also makes it so our peer never gets limited
1462  * by our rwnd which we then open up causing a burst.
1463  *
1464  * This algorithm does two steps per RTT at most and only if
1465  * we receive a bulk stream w/o packet losses or reorderings.
1466  * Shrinking the buffer during idle times is not necessary as
1467  * it doesn't consume any memory when idle.
1468  *
1469  * TODO: Only step up if the application is actually serving
1470  * the buffer to better manage the socket buffer resources.
1471  */
1472 int
1473 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1474     struct tcpcb *tp, int tlen)
1475 {
1476 	int newsize = 0;
1477 
1478 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1479 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1480 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1481 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1482 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1483 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1484 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1485 		}
1486 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1487 
1488 		/* Start over with next RTT. */
1489 		tp->rfbuf_ts = 0;
1490 		tp->rfbuf_cnt = 0;
1491 	} else {
1492 		tp->rfbuf_cnt += tlen;	/* add up */
1493 	}
1494 	return (newsize);
1495 }
1496 
1497 int
1498 tcp_input(struct mbuf **mp, int *offp, int proto)
1499 {
1500 	return(tcp_input_with_port(mp, offp, proto, 0));
1501 }
1502 
1503 void
1504 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1505 {
1506 	/*
1507 	 * Since tp might be gone if the session entered
1508 	 * the TIME_WAIT state before coming here, we need
1509 	 * to check if the socket is still connected.
1510 	 */
1511 	if (tp == NULL) {
1512 		return;
1513 	}
1514 	if (so == NULL) {
1515 		return;
1516 	}
1517 	INP_LOCK_ASSERT(tp->t_inpcb);
1518 	if (tp->t_flags & TF_WAKESOR) {
1519 		tp->t_flags &= ~TF_WAKESOR;
1520 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1521 		sorwakeup_locked(so);
1522 	}
1523 }
1524 
1525 void
1526 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1527     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1528 {
1529 	int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1530 	int rstreason, todrop, win, incforsyn = 0;
1531 	uint32_t tiwin;
1532 	uint16_t nsegs;
1533 	char *s;
1534 	struct in_conninfo *inc;
1535 	struct mbuf *mfree;
1536 	struct tcpopt to;
1537 	int tfo_syn;
1538 	u_int maxseg;
1539 
1540 #ifdef TCPDEBUG
1541 	/*
1542 	 * The size of tcp_saveipgen must be the size of the max ip header,
1543 	 * now IPv6.
1544 	 */
1545 	u_char tcp_saveipgen[IP6_HDR_LEN];
1546 	struct tcphdr tcp_savetcp;
1547 	short ostate = 0;
1548 #endif
1549 	thflags = th->th_flags;
1550 	inc = &tp->t_inpcb->inp_inc;
1551 	tp->sackhint.last_sack_ack = 0;
1552 	sack_changed = 0;
1553 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1554 
1555 	NET_EPOCH_ASSERT();
1556 	INP_WLOCK_ASSERT(tp->t_inpcb);
1557 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1558 	    __func__));
1559 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1560 	    __func__));
1561 
1562 #ifdef TCPPCAP
1563 	/* Save segment, if requested. */
1564 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1565 #endif
1566 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1567 	    tlen, NULL, true);
1568 
1569 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1570 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1571 			log(LOG_DEBUG, "%s; %s: "
1572 			    "SYN|FIN segment ignored (based on "
1573 			    "sysctl setting)\n", s, __func__);
1574 			free(s, M_TCPLOG);
1575 		}
1576 		goto drop;
1577 	}
1578 
1579 	/*
1580 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1581 	 * check SEQ.ACK first.
1582 	 */
1583 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1584 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1585 		rstreason = BANDLIM_UNLIMITED;
1586 		goto dropwithreset;
1587 	}
1588 
1589 	/*
1590 	 * Segment received on connection.
1591 	 * Reset idle time and keep-alive timer.
1592 	 * XXX: This should be done after segment
1593 	 * validation to ignore broken/spoofed segs.
1594 	 */
1595 	tp->t_rcvtime = ticks;
1596 
1597 	/*
1598 	 * Scale up the window into a 32-bit value.
1599 	 * For the SYN_SENT state the scale is zero.
1600 	 */
1601 	tiwin = th->th_win << tp->snd_scale;
1602 #ifdef STATS
1603 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1604 #endif
1605 
1606 	/*
1607 	 * TCP ECN processing.
1608 	 */
1609 	if (tp->t_flags2 & TF2_ECN_PERMIT) {
1610 		if (thflags & TH_CWR) {
1611 			tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1612 			tp->t_flags |= TF_ACKNOW;
1613 		}
1614 		switch (iptos & IPTOS_ECN_MASK) {
1615 		case IPTOS_ECN_CE:
1616 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1617 			TCPSTAT_INC(tcps_ecn_ce);
1618 			break;
1619 		case IPTOS_ECN_ECT0:
1620 			TCPSTAT_INC(tcps_ecn_ect0);
1621 			break;
1622 		case IPTOS_ECN_ECT1:
1623 			TCPSTAT_INC(tcps_ecn_ect1);
1624 			break;
1625 		}
1626 
1627 		/* Process a packet differently from RFC3168. */
1628 		cc_ecnpkt_handler(tp, th, iptos);
1629 
1630 		/* Congestion experienced. */
1631 		if (thflags & TH_ECE) {
1632 			cc_cong_signal(tp, th, CC_ECN);
1633 		}
1634 	}
1635 
1636 	/*
1637 	 * Parse options on any incoming segment.
1638 	 */
1639 	tcp_dooptions(&to, (u_char *)(th + 1),
1640 	    (th->th_off << 2) - sizeof(struct tcphdr),
1641 	    (thflags & TH_SYN) ? TO_SYN : 0);
1642 
1643 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1644 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1645 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1646 		TCPSTAT_INC(tcps_sig_err_sigopt);
1647 		/* XXX: should drop? */
1648 	}
1649 #endif
1650 	/*
1651 	 * If echoed timestamp is later than the current time,
1652 	 * fall back to non RFC1323 RTT calculation.  Normalize
1653 	 * timestamp if syncookies were used when this connection
1654 	 * was established.
1655 	 */
1656 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1657 		to.to_tsecr -= tp->ts_offset;
1658 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1659 			to.to_tsecr = 0;
1660 		else if (tp->t_flags & TF_PREVVALID &&
1661 			 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1662 			cc_cong_signal(tp, th, CC_RTO_ERR);
1663 	}
1664 	/*
1665 	 * Process options only when we get SYN/ACK back. The SYN case
1666 	 * for incoming connections is handled in tcp_syncache.
1667 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1668 	 * or <SYN,ACK>) segment itself is never scaled.
1669 	 * XXX this is traditional behavior, may need to be cleaned up.
1670 	 */
1671 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1672 		/* Handle parallel SYN for ECN */
1673 		if (!(thflags & TH_ACK) &&
1674 		    ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1675 		    ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1676 			tp->t_flags2 |= TF2_ECN_PERMIT;
1677 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1678 			TCPSTAT_INC(tcps_ecn_shs);
1679 		}
1680 		if ((to.to_flags & TOF_SCALE) &&
1681 		    (tp->t_flags & TF_REQ_SCALE) &&
1682 		    !(tp->t_flags & TF_NOOPT)) {
1683 			tp->t_flags |= TF_RCVD_SCALE;
1684 			tp->snd_scale = to.to_wscale;
1685 		} else
1686 			tp->t_flags &= ~TF_REQ_SCALE;
1687 		/*
1688 		 * Initial send window.  It will be updated with
1689 		 * the next incoming segment to the scaled value.
1690 		 */
1691 		tp->snd_wnd = th->th_win;
1692 		if ((to.to_flags & TOF_TS) &&
1693 		    (tp->t_flags & TF_REQ_TSTMP) &&
1694 		    !(tp->t_flags & TF_NOOPT)) {
1695 			tp->t_flags |= TF_RCVD_TSTMP;
1696 			tp->ts_recent = to.to_tsval;
1697 			tp->ts_recent_age = tcp_ts_getticks();
1698 		} else
1699 			tp->t_flags &= ~TF_REQ_TSTMP;
1700 		if (to.to_flags & TOF_MSS)
1701 			tcp_mss(tp, to.to_mss);
1702 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1703 		    (!(to.to_flags & TOF_SACKPERM) ||
1704 		    (tp->t_flags & TF_NOOPT)))
1705 			tp->t_flags &= ~TF_SACK_PERMIT;
1706 		if (IS_FASTOPEN(tp->t_flags)) {
1707 			if ((to.to_flags & TOF_FASTOPEN) &&
1708 			    !(tp->t_flags & TF_NOOPT)) {
1709 				uint16_t mss;
1710 
1711 				if (to.to_flags & TOF_MSS)
1712 					mss = to.to_mss;
1713 				else
1714 					if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1715 						mss = TCP6_MSS;
1716 					else
1717 						mss = TCP_MSS;
1718 				tcp_fastopen_update_cache(tp, mss,
1719 				    to.to_tfo_len, to.to_tfo_cookie);
1720 			} else
1721 				tcp_fastopen_disable_path(tp);
1722 		}
1723 	}
1724 
1725 	/*
1726 	 * If timestamps were negotiated during SYN/ACK and a
1727 	 * segment without a timestamp is received, silently drop
1728 	 * the segment, unless it is a RST segment or missing timestamps are
1729 	 * tolerated.
1730 	 * See section 3.2 of RFC 7323.
1731 	 */
1732 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1733 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1734 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1735 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1736 				    "segment processed normally\n",
1737 				    s, __func__);
1738 				free(s, M_TCPLOG);
1739 			}
1740 		} else {
1741 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1742 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1743 				    "segment silently dropped\n", s, __func__);
1744 				free(s, M_TCPLOG);
1745 			}
1746 			goto drop;
1747 		}
1748 	}
1749 	/*
1750 	 * If timestamps were not negotiated during SYN/ACK and a
1751 	 * segment with a timestamp is received, ignore the
1752 	 * timestamp and process the packet normally.
1753 	 * See section 3.2 of RFC 7323.
1754 	 */
1755 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1756 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1757 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1758 			    "segment processed normally\n", s, __func__);
1759 			free(s, M_TCPLOG);
1760 		}
1761 	}
1762 
1763 	/*
1764 	 * Header prediction: check for the two common cases
1765 	 * of a uni-directional data xfer.  If the packet has
1766 	 * no control flags, is in-sequence, the window didn't
1767 	 * change and we're not retransmitting, it's a
1768 	 * candidate.  If the length is zero and the ack moved
1769 	 * forward, we're the sender side of the xfer.  Just
1770 	 * free the data acked & wake any higher level process
1771 	 * that was blocked waiting for space.  If the length
1772 	 * is non-zero and the ack didn't move, we're the
1773 	 * receiver side.  If we're getting packets in-order
1774 	 * (the reassembly queue is empty), add the data to
1775 	 * the socket buffer and note that we need a delayed ack.
1776 	 * Make sure that the hidden state-flags are also off.
1777 	 * Since we check for TCPS_ESTABLISHED first, it can only
1778 	 * be TH_NEEDSYN.
1779 	 */
1780 	if (tp->t_state == TCPS_ESTABLISHED &&
1781 	    th->th_seq == tp->rcv_nxt &&
1782 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1783 	    tp->snd_nxt == tp->snd_max &&
1784 	    tiwin && tiwin == tp->snd_wnd &&
1785 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1786 	    SEGQ_EMPTY(tp) &&
1787 	    ((to.to_flags & TOF_TS) == 0 ||
1788 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1789 		/*
1790 		 * If last ACK falls within this segment's sequence numbers,
1791 		 * record the timestamp.
1792 		 * NOTE that the test is modified according to the latest
1793 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1794 		 */
1795 		if ((to.to_flags & TOF_TS) != 0 &&
1796 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1797 			tp->ts_recent_age = tcp_ts_getticks();
1798 			tp->ts_recent = to.to_tsval;
1799 		}
1800 
1801 		if (tlen == 0) {
1802 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1803 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1804 			    !IN_RECOVERY(tp->t_flags) &&
1805 			    (to.to_flags & TOF_SACK) == 0 &&
1806 			    TAILQ_EMPTY(&tp->snd_holes)) {
1807 				/*
1808 				 * This is a pure ack for outstanding data.
1809 				 */
1810 				TCPSTAT_INC(tcps_predack);
1811 
1812 				/*
1813 				 * "bad retransmit" recovery without timestamps.
1814 				 */
1815 				if ((to.to_flags & TOF_TS) == 0 &&
1816 				    tp->t_rxtshift == 1 &&
1817 				    tp->t_flags & TF_PREVVALID &&
1818 				    (int)(ticks - tp->t_badrxtwin) < 0) {
1819 					cc_cong_signal(tp, th, CC_RTO_ERR);
1820 				}
1821 
1822 				/*
1823 				 * Recalculate the transmit timer / rtt.
1824 				 *
1825 				 * Some boxes send broken timestamp replies
1826 				 * during the SYN+ACK phase, ignore
1827 				 * timestamps of 0 or we could calculate a
1828 				 * huge RTT and blow up the retransmit timer.
1829 				 */
1830 				if ((to.to_flags & TOF_TS) != 0 &&
1831 				    to.to_tsecr) {
1832 					uint32_t t;
1833 
1834 					t = tcp_ts_getticks() - to.to_tsecr;
1835 					if (!tp->t_rttlow || tp->t_rttlow > t)
1836 						tp->t_rttlow = t;
1837 					tcp_xmit_timer(tp,
1838 					    TCP_TS_TO_TICKS(t) + 1);
1839 				} else if (tp->t_rtttime &&
1840 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1841 					if (!tp->t_rttlow ||
1842 					    tp->t_rttlow > ticks - tp->t_rtttime)
1843 						tp->t_rttlow = ticks - tp->t_rtttime;
1844 					tcp_xmit_timer(tp,
1845 							ticks - tp->t_rtttime);
1846 				}
1847 				acked = BYTES_THIS_ACK(tp, th);
1848 
1849 #ifdef TCP_HHOOK
1850 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1851 				hhook_run_tcp_est_in(tp, th, &to);
1852 #endif
1853 
1854 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1855 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1856 				sbdrop(&so->so_snd, acked);
1857 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1858 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1859 					tp->snd_recover = th->th_ack - 1;
1860 
1861 				/*
1862 				 * Let the congestion control algorithm update
1863 				 * congestion control related information. This
1864 				 * typically means increasing the congestion
1865 				 * window.
1866 				 */
1867 				cc_ack_received(tp, th, nsegs, CC_ACK);
1868 
1869 				tp->snd_una = th->th_ack;
1870 				/*
1871 				 * Pull snd_wl2 up to prevent seq wrap relative
1872 				 * to th_ack.
1873 				 */
1874 				tp->snd_wl2 = th->th_ack;
1875 				tp->t_dupacks = 0;
1876 				m_freem(m);
1877 
1878 				/*
1879 				 * If all outstanding data are acked, stop
1880 				 * retransmit timer, otherwise restart timer
1881 				 * using current (possibly backed-off) value.
1882 				 * If process is waiting for space,
1883 				 * wakeup/selwakeup/signal.  If data
1884 				 * are ready to send, let tcp_output
1885 				 * decide between more output or persist.
1886 				 */
1887 #ifdef TCPDEBUG
1888 				if (so->so_options & SO_DEBUG)
1889 					tcp_trace(TA_INPUT, ostate, tp,
1890 					    (void *)tcp_saveipgen,
1891 					    &tcp_savetcp, 0);
1892 #endif
1893 				TCP_PROBE3(debug__input, tp, th, m);
1894 				if (tp->snd_una == tp->snd_max)
1895 					tcp_timer_activate(tp, TT_REXMT, 0);
1896 				else if (!tcp_timer_active(tp, TT_PERSIST))
1897 					tcp_timer_activate(tp, TT_REXMT,
1898 						      tp->t_rxtcur);
1899 				sowwakeup(so);
1900 				if (sbavail(&so->so_snd))
1901 					(void) tp->t_fb->tfb_tcp_output(tp);
1902 				goto check_delack;
1903 			}
1904 		} else if (th->th_ack == tp->snd_una &&
1905 		    tlen <= sbspace(&so->so_rcv)) {
1906 			int newsize = 0;	/* automatic sockbuf scaling */
1907 
1908 			/*
1909 			 * This is a pure, in-sequence data packet with
1910 			 * nothing on the reassembly queue and we have enough
1911 			 * buffer space to take it.
1912 			 */
1913 			/* Clean receiver SACK report if present */
1914 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1915 				tcp_clean_sackreport(tp);
1916 			TCPSTAT_INC(tcps_preddat);
1917 			tp->rcv_nxt += tlen;
1918 			if (tlen &&
1919 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1920 			    (tp->t_fbyte_in == 0)) {
1921 				tp->t_fbyte_in = ticks;
1922 				if (tp->t_fbyte_in == 0)
1923 					tp->t_fbyte_in = 1;
1924 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1925 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1926 			}
1927 			/*
1928 			 * Pull snd_wl1 up to prevent seq wrap relative to
1929 			 * th_seq.
1930 			 */
1931 			tp->snd_wl1 = th->th_seq;
1932 			/*
1933 			 * Pull rcv_up up to prevent seq wrap relative to
1934 			 * rcv_nxt.
1935 			 */
1936 			tp->rcv_up = tp->rcv_nxt;
1937 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1938 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1939 #ifdef TCPDEBUG
1940 			if (so->so_options & SO_DEBUG)
1941 				tcp_trace(TA_INPUT, ostate, tp,
1942 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1943 #endif
1944 			TCP_PROBE3(debug__input, tp, th, m);
1945 
1946 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1947 
1948 			/* Add data to socket buffer. */
1949 			SOCKBUF_LOCK(&so->so_rcv);
1950 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1951 				m_freem(m);
1952 			} else {
1953 				/*
1954 				 * Set new socket buffer size.
1955 				 * Give up when limit is reached.
1956 				 */
1957 				if (newsize)
1958 					if (!sbreserve_locked(&so->so_rcv,
1959 					    newsize, so, NULL))
1960 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1961 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1962 				sbappendstream_locked(&so->so_rcv, m, 0);
1963 			}
1964 			/* NB: sorwakeup_locked() does an implicit unlock. */
1965 			sorwakeup_locked(so);
1966 			if (DELAY_ACK(tp, tlen)) {
1967 				tp->t_flags |= TF_DELACK;
1968 			} else {
1969 				tp->t_flags |= TF_ACKNOW;
1970 				tp->t_fb->tfb_tcp_output(tp);
1971 			}
1972 			goto check_delack;
1973 		}
1974 	}
1975 
1976 	/*
1977 	 * Calculate amount of space in receive window,
1978 	 * and then do TCP input processing.
1979 	 * Receive window is amount of space in rcv queue,
1980 	 * but not less than advertised window.
1981 	 */
1982 	win = sbspace(&so->so_rcv);
1983 	if (win < 0)
1984 		win = 0;
1985 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1986 
1987 	switch (tp->t_state) {
1988 	/*
1989 	 * If the state is SYN_RECEIVED:
1990 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1991 	 */
1992 	case TCPS_SYN_RECEIVED:
1993 		if ((thflags & TH_ACK) &&
1994 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1995 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1996 				rstreason = BANDLIM_RST_OPENPORT;
1997 				goto dropwithreset;
1998 		}
1999 		if (IS_FASTOPEN(tp->t_flags)) {
2000 			/*
2001 			 * When a TFO connection is in SYN_RECEIVED, the
2002 			 * only valid packets are the initial SYN, a
2003 			 * retransmit/copy of the initial SYN (possibly with
2004 			 * a subset of the original data), a valid ACK, a
2005 			 * FIN, or a RST.
2006 			 */
2007 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2008 				rstreason = BANDLIM_RST_OPENPORT;
2009 				goto dropwithreset;
2010 			} else if (thflags & TH_SYN) {
2011 				/* non-initial SYN is ignored */
2012 				if ((tcp_timer_active(tp, TT_DELACK) ||
2013 				     tcp_timer_active(tp, TT_REXMT)))
2014 					goto drop;
2015 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2016 				goto drop;
2017 			}
2018 		}
2019 		break;
2020 
2021 	/*
2022 	 * If the state is SYN_SENT:
2023 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
2024 	 *	    been verified), then drop the connection.
2025 	 *	if seg contains a RST without an ACK, drop the seg.
2026 	 *	if seg does not contain SYN, then drop the seg.
2027 	 * Otherwise this is an acceptable SYN segment
2028 	 *	initialize tp->rcv_nxt and tp->irs
2029 	 *	if seg contains ack then advance tp->snd_una
2030 	 *	if seg contains an ECE and ECN support is enabled, the stream
2031 	 *	    is ECN capable.
2032 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2033 	 *	arrange for segment to be acked (eventually)
2034 	 *	continue processing rest of data/controls, beginning with URG
2035 	 */
2036 	case TCPS_SYN_SENT:
2037 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2038 			TCP_PROBE5(connect__refused, NULL, tp,
2039 			    m, tp, th);
2040 			tp = tcp_drop(tp, ECONNREFUSED);
2041 		}
2042 		if (thflags & TH_RST)
2043 			goto drop;
2044 		if (!(thflags & TH_SYN))
2045 			goto drop;
2046 
2047 		tp->irs = th->th_seq;
2048 		tcp_rcvseqinit(tp);
2049 		if (thflags & TH_ACK) {
2050 			int tfo_partial_ack = 0;
2051 
2052 			TCPSTAT_INC(tcps_connects);
2053 			soisconnected(so);
2054 #ifdef MAC
2055 			mac_socketpeer_set_from_mbuf(m, so);
2056 #endif
2057 			/* Do window scaling on this connection? */
2058 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2059 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2060 				tp->rcv_scale = tp->request_r_scale;
2061 			}
2062 			tp->rcv_adv += min(tp->rcv_wnd,
2063 			    TCP_MAXWIN << tp->rcv_scale);
2064 			tp->snd_una++;		/* SYN is acked */
2065 			/*
2066 			 * If not all the data that was sent in the TFO SYN
2067 			 * has been acked, resend the remainder right away.
2068 			 */
2069 			if (IS_FASTOPEN(tp->t_flags) &&
2070 			    (tp->snd_una != tp->snd_max)) {
2071 				tp->snd_nxt = th->th_ack;
2072 				tfo_partial_ack = 1;
2073 			}
2074 			/*
2075 			 * If there's data, delay ACK; if there's also a FIN
2076 			 * ACKNOW will be turned on later.
2077 			 */
2078 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2079 				tcp_timer_activate(tp, TT_DELACK,
2080 				    tcp_delacktime);
2081 			else
2082 				tp->t_flags |= TF_ACKNOW;
2083 
2084 			if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2085 			    (V_tcp_do_ecn == 1)) {
2086 				tp->t_flags2 |= TF2_ECN_PERMIT;
2087 				TCPSTAT_INC(tcps_ecn_shs);
2088 			}
2089 
2090 			/*
2091 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2092 			 * Transitions:
2093 			 *	SYN_SENT  --> ESTABLISHED
2094 			 *	SYN_SENT* --> FIN_WAIT_1
2095 			 */
2096 			tp->t_starttime = ticks;
2097 			if (tp->t_flags & TF_NEEDFIN) {
2098 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2099 				tp->t_flags &= ~TF_NEEDFIN;
2100 				thflags &= ~TH_SYN;
2101 			} else {
2102 				tcp_state_change(tp, TCPS_ESTABLISHED);
2103 				TCP_PROBE5(connect__established, NULL, tp,
2104 				    m, tp, th);
2105 				cc_conn_init(tp);
2106 				tcp_timer_activate(tp, TT_KEEP,
2107 				    TP_KEEPIDLE(tp));
2108 			}
2109 		} else {
2110 			/*
2111 			 * Received initial SYN in SYN-SENT[*] state =>
2112 			 * simultaneous open.
2113 			 * If it succeeds, connection is * half-synchronized.
2114 			 * Otherwise, do 3-way handshake:
2115 			 *        SYN-SENT -> SYN-RECEIVED
2116 			 *        SYN-SENT* -> SYN-RECEIVED*
2117 			 */
2118 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2119 			tcp_timer_activate(tp, TT_REXMT, 0);
2120 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2121 		}
2122 
2123 		INP_WLOCK_ASSERT(tp->t_inpcb);
2124 
2125 		/*
2126 		 * Advance th->th_seq to correspond to first data byte.
2127 		 * If data, trim to stay within window,
2128 		 * dropping FIN if necessary.
2129 		 */
2130 		th->th_seq++;
2131 		if (tlen > tp->rcv_wnd) {
2132 			todrop = tlen - tp->rcv_wnd;
2133 			m_adj(m, -todrop);
2134 			tlen = tp->rcv_wnd;
2135 			thflags &= ~TH_FIN;
2136 			TCPSTAT_INC(tcps_rcvpackafterwin);
2137 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2138 		}
2139 		tp->snd_wl1 = th->th_seq - 1;
2140 		tp->rcv_up = th->th_seq;
2141 		/*
2142 		 * Client side of transaction: already sent SYN and data.
2143 		 * If the remote host used T/TCP to validate the SYN,
2144 		 * our data will be ACK'd; if so, enter normal data segment
2145 		 * processing in the middle of step 5, ack processing.
2146 		 * Otherwise, goto step 6.
2147 		 */
2148 		if (thflags & TH_ACK)
2149 			goto process_ACK;
2150 
2151 		goto step6;
2152 
2153 	/*
2154 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2155 	 *      do normal processing.
2156 	 *
2157 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2158 	 */
2159 	case TCPS_LAST_ACK:
2160 	case TCPS_CLOSING:
2161 		break;  /* continue normal processing */
2162 	}
2163 
2164 	/*
2165 	 * States other than LISTEN or SYN_SENT.
2166 	 * First check the RST flag and sequence number since reset segments
2167 	 * are exempt from the timestamp and connection count tests.  This
2168 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2169 	 * below which allowed reset segments in half the sequence space
2170 	 * to fall though and be processed (which gives forged reset
2171 	 * segments with a random sequence number a 50 percent chance of
2172 	 * killing a connection).
2173 	 * Then check timestamp, if present.
2174 	 * Then check the connection count, if present.
2175 	 * Then check that at least some bytes of segment are within
2176 	 * receive window.  If segment begins before rcv_nxt,
2177 	 * drop leading data (and SYN); if nothing left, just ack.
2178 	 */
2179 	if (thflags & TH_RST) {
2180 		/*
2181 		 * RFC5961 Section 3.2
2182 		 *
2183 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2184 		 * - If RST is in window, we send challenge ACK.
2185 		 *
2186 		 * Note: to take into account delayed ACKs, we should
2187 		 *   test against last_ack_sent instead of rcv_nxt.
2188 		 * Note 2: we handle special case of closed window, not
2189 		 *   covered by the RFC.
2190 		 */
2191 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2192 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2193 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2194 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2195 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2196 			    __func__, th, tp));
2197 
2198 			if (V_tcp_insecure_rst ||
2199 			    tp->last_ack_sent == th->th_seq) {
2200 				TCPSTAT_INC(tcps_drops);
2201 				/* Drop the connection. */
2202 				switch (tp->t_state) {
2203 				case TCPS_SYN_RECEIVED:
2204 					so->so_error = ECONNREFUSED;
2205 					goto close;
2206 				case TCPS_ESTABLISHED:
2207 				case TCPS_FIN_WAIT_1:
2208 				case TCPS_FIN_WAIT_2:
2209 				case TCPS_CLOSE_WAIT:
2210 				case TCPS_CLOSING:
2211 				case TCPS_LAST_ACK:
2212 					so->so_error = ECONNRESET;
2213 				close:
2214 					/* FALLTHROUGH */
2215 				default:
2216 					tp = tcp_close(tp);
2217 				}
2218 			} else {
2219 				TCPSTAT_INC(tcps_badrst);
2220 				/* Send challenge ACK. */
2221 				tcp_respond(tp, mtod(m, void *), th, m,
2222 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2223 				tp->last_ack_sent = tp->rcv_nxt;
2224 				m = NULL;
2225 			}
2226 		}
2227 		goto drop;
2228 	}
2229 
2230 	/*
2231 	 * RFC5961 Section 4.2
2232 	 * Send challenge ACK for any SYN in synchronized state.
2233 	 */
2234 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2235 	    tp->t_state != TCPS_SYN_RECEIVED) {
2236 		TCPSTAT_INC(tcps_badsyn);
2237 		if (V_tcp_insecure_syn &&
2238 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2239 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2240 			tp = tcp_drop(tp, ECONNRESET);
2241 			rstreason = BANDLIM_UNLIMITED;
2242 		} else {
2243 			/* Send challenge ACK. */
2244 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2245 			    tp->snd_nxt, TH_ACK);
2246 			tp->last_ack_sent = tp->rcv_nxt;
2247 			m = NULL;
2248 		}
2249 		goto drop;
2250 	}
2251 
2252 	/*
2253 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2254 	 * and it's less than ts_recent, drop it.
2255 	 */
2256 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2257 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2258 		/* Check to see if ts_recent is over 24 days old.  */
2259 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2260 			/*
2261 			 * Invalidate ts_recent.  If this segment updates
2262 			 * ts_recent, the age will be reset later and ts_recent
2263 			 * will get a valid value.  If it does not, setting
2264 			 * ts_recent to zero will at least satisfy the
2265 			 * requirement that zero be placed in the timestamp
2266 			 * echo reply when ts_recent isn't valid.  The
2267 			 * age isn't reset until we get a valid ts_recent
2268 			 * because we don't want out-of-order segments to be
2269 			 * dropped when ts_recent is old.
2270 			 */
2271 			tp->ts_recent = 0;
2272 		} else {
2273 			TCPSTAT_INC(tcps_rcvduppack);
2274 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2275 			TCPSTAT_INC(tcps_pawsdrop);
2276 			if (tlen)
2277 				goto dropafterack;
2278 			goto drop;
2279 		}
2280 	}
2281 
2282 	/*
2283 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2284 	 * this connection before trimming the data to fit the receive
2285 	 * window.  Check the sequence number versus IRS since we know
2286 	 * the sequence numbers haven't wrapped.  This is a partial fix
2287 	 * for the "LAND" DoS attack.
2288 	 */
2289 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2290 		rstreason = BANDLIM_RST_OPENPORT;
2291 		goto dropwithreset;
2292 	}
2293 
2294 	todrop = tp->rcv_nxt - th->th_seq;
2295 	if (todrop > 0) {
2296 		if (thflags & TH_SYN) {
2297 			thflags &= ~TH_SYN;
2298 			th->th_seq++;
2299 			if (th->th_urp > 1)
2300 				th->th_urp--;
2301 			else
2302 				thflags &= ~TH_URG;
2303 			todrop--;
2304 		}
2305 		/*
2306 		 * Following if statement from Stevens, vol. 2, p. 960.
2307 		 */
2308 		if (todrop > tlen
2309 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2310 			/*
2311 			 * Any valid FIN must be to the left of the window.
2312 			 * At this point the FIN must be a duplicate or out
2313 			 * of sequence; drop it.
2314 			 */
2315 			thflags &= ~TH_FIN;
2316 
2317 			/*
2318 			 * Send an ACK to resynchronize and drop any data.
2319 			 * But keep on processing for RST or ACK.
2320 			 */
2321 			tp->t_flags |= TF_ACKNOW;
2322 			todrop = tlen;
2323 			TCPSTAT_INC(tcps_rcvduppack);
2324 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2325 		} else {
2326 			TCPSTAT_INC(tcps_rcvpartduppack);
2327 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2328 		}
2329 		/*
2330 		 * DSACK - add SACK block for dropped range
2331 		 */
2332 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2333 			tcp_update_sack_list(tp, th->th_seq,
2334 			    th->th_seq + todrop);
2335 			/*
2336 			 * ACK now, as the next in-sequence segment
2337 			 * will clear the DSACK block again
2338 			 */
2339 			tp->t_flags |= TF_ACKNOW;
2340 		}
2341 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2342 		th->th_seq += todrop;
2343 		tlen -= todrop;
2344 		if (th->th_urp > todrop)
2345 			th->th_urp -= todrop;
2346 		else {
2347 			thflags &= ~TH_URG;
2348 			th->th_urp = 0;
2349 		}
2350 	}
2351 
2352 	/*
2353 	 * If new data are received on a connection after the
2354 	 * user processes are gone, then RST the other end.
2355 	 */
2356 	if ((so->so_state & SS_NOFDREF) &&
2357 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2358 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2359 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2360 			    "after socket was closed, "
2361 			    "sending RST and removing tcpcb\n",
2362 			    s, __func__, tcpstates[tp->t_state], tlen);
2363 			free(s, M_TCPLOG);
2364 		}
2365 		tp = tcp_close(tp);
2366 		TCPSTAT_INC(tcps_rcvafterclose);
2367 		rstreason = BANDLIM_UNLIMITED;
2368 		goto dropwithreset;
2369 	}
2370 
2371 	/*
2372 	 * If segment ends after window, drop trailing data
2373 	 * (and PUSH and FIN); if nothing left, just ACK.
2374 	 */
2375 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2376 	if (todrop > 0) {
2377 		TCPSTAT_INC(tcps_rcvpackafterwin);
2378 		if (todrop >= tlen) {
2379 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2380 			/*
2381 			 * If window is closed can only take segments at
2382 			 * window edge, and have to drop data and PUSH from
2383 			 * incoming segments.  Continue processing, but
2384 			 * remember to ack.  Otherwise, drop segment
2385 			 * and ack.
2386 			 */
2387 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2388 				tp->t_flags |= TF_ACKNOW;
2389 				TCPSTAT_INC(tcps_rcvwinprobe);
2390 			} else
2391 				goto dropafterack;
2392 		} else
2393 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2394 		m_adj(m, -todrop);
2395 		tlen -= todrop;
2396 		thflags &= ~(TH_PUSH|TH_FIN);
2397 	}
2398 
2399 	/*
2400 	 * If last ACK falls within this segment's sequence numbers,
2401 	 * record its timestamp.
2402 	 * NOTE:
2403 	 * 1) That the test incorporates suggestions from the latest
2404 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2405 	 * 2) That updating only on newer timestamps interferes with
2406 	 *    our earlier PAWS tests, so this check should be solely
2407 	 *    predicated on the sequence space of this segment.
2408 	 * 3) That we modify the segment boundary check to be
2409 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2410 	 *    instead of RFC1323's
2411 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2412 	 *    This modified check allows us to overcome RFC1323's
2413 	 *    limitations as described in Stevens TCP/IP Illustrated
2414 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2415 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2416 	 */
2417 	if ((to.to_flags & TOF_TS) != 0 &&
2418 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2419 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2420 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2421 		tp->ts_recent_age = tcp_ts_getticks();
2422 		tp->ts_recent = to.to_tsval;
2423 	}
2424 
2425 	/*
2426 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2427 	 * flag is on (half-synchronized state), then queue data for
2428 	 * later processing; else drop segment and return.
2429 	 */
2430 	if ((thflags & TH_ACK) == 0) {
2431 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2432 		    (tp->t_flags & TF_NEEDSYN)) {
2433 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2434 			    IS_FASTOPEN(tp->t_flags)) {
2435 				tp->snd_wnd = tiwin;
2436 				cc_conn_init(tp);
2437 			}
2438 			goto step6;
2439 		} else if (tp->t_flags & TF_ACKNOW)
2440 			goto dropafterack;
2441 		else
2442 			goto drop;
2443 	}
2444 
2445 	/*
2446 	 * Ack processing.
2447 	 */
2448 	switch (tp->t_state) {
2449 	/*
2450 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2451 	 * ESTABLISHED state and continue processing.
2452 	 * The ACK was checked above.
2453 	 */
2454 	case TCPS_SYN_RECEIVED:
2455 
2456 		TCPSTAT_INC(tcps_connects);
2457 		soisconnected(so);
2458 		/* Do window scaling? */
2459 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2460 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2461 			tp->rcv_scale = tp->request_r_scale;
2462 		}
2463 		tp->snd_wnd = tiwin;
2464 		/*
2465 		 * Make transitions:
2466 		 *      SYN-RECEIVED  -> ESTABLISHED
2467 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2468 		 */
2469 		tp->t_starttime = ticks;
2470 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2471 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2472 			tp->t_tfo_pending = NULL;
2473 		}
2474 		if (tp->t_flags & TF_NEEDFIN) {
2475 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2476 			tp->t_flags &= ~TF_NEEDFIN;
2477 		} else {
2478 			tcp_state_change(tp, TCPS_ESTABLISHED);
2479 			TCP_PROBE5(accept__established, NULL, tp,
2480 			    m, tp, th);
2481 			/*
2482 			 * TFO connections call cc_conn_init() during SYN
2483 			 * processing.  Calling it again here for such
2484 			 * connections is not harmless as it would undo the
2485 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2486 			 * is retransmitted.
2487 			 */
2488 			if (!IS_FASTOPEN(tp->t_flags))
2489 				cc_conn_init(tp);
2490 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2491 		}
2492 		/*
2493 		 * Account for the ACK of our SYN prior to
2494 		 * regular ACK processing below, except for
2495 		 * simultaneous SYN, which is handled later.
2496 		 */
2497 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2498 			incforsyn = 1;
2499 		/*
2500 		 * If segment contains data or ACK, will call tcp_reass()
2501 		 * later; if not, do so now to pass queued data to user.
2502 		 */
2503 		if (tlen == 0 && (thflags & TH_FIN) == 0) {
2504 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2505 			    (struct mbuf *)0);
2506 			tcp_handle_wakeup(tp, so);
2507 		}
2508 		tp->snd_wl1 = th->th_seq - 1;
2509 		/* FALLTHROUGH */
2510 
2511 	/*
2512 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2513 	 * ACKs.  If the ack is in the range
2514 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2515 	 * then advance tp->snd_una to th->th_ack and drop
2516 	 * data from the retransmission queue.  If this ACK reflects
2517 	 * more up to date window information we update our window information.
2518 	 */
2519 	case TCPS_ESTABLISHED:
2520 	case TCPS_FIN_WAIT_1:
2521 	case TCPS_FIN_WAIT_2:
2522 	case TCPS_CLOSE_WAIT:
2523 	case TCPS_CLOSING:
2524 	case TCPS_LAST_ACK:
2525 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2526 			TCPSTAT_INC(tcps_rcvacktoomuch);
2527 			goto dropafterack;
2528 		}
2529 		if ((tp->t_flags & TF_SACK_PERMIT) &&
2530 		    ((to.to_flags & TOF_SACK) ||
2531 		     !TAILQ_EMPTY(&tp->snd_holes))) {
2532 			if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2533 			    (tp->t_flags & TF_LRD)) {
2534 				tcp_sack_lost_retransmission(tp, th);
2535 			}
2536 		} else
2537 			/*
2538 			 * Reset the value so that previous (valid) value
2539 			 * from the last ack with SACK doesn't get used.
2540 			 */
2541 			tp->sackhint.sacked_bytes = 0;
2542 
2543 #ifdef TCP_HHOOK
2544 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2545 		hhook_run_tcp_est_in(tp, th, &to);
2546 #endif
2547 
2548 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2549 			maxseg = tcp_maxseg(tp);
2550 			if (tlen == 0 &&
2551 			    (tiwin == tp->snd_wnd ||
2552 			    (tp->t_flags & TF_SACK_PERMIT))) {
2553 				/*
2554 				 * If this is the first time we've seen a
2555 				 * FIN from the remote, this is not a
2556 				 * duplicate and it needs to be processed
2557 				 * normally.  This happens during a
2558 				 * simultaneous close.
2559 				 */
2560 				if ((thflags & TH_FIN) &&
2561 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2562 					tp->t_dupacks = 0;
2563 					break;
2564 				}
2565 				TCPSTAT_INC(tcps_rcvdupack);
2566 				/*
2567 				 * If we have outstanding data (other than
2568 				 * a window probe), this is a completely
2569 				 * duplicate ack (ie, window info didn't
2570 				 * change and FIN isn't set),
2571 				 * the ack is the biggest we've
2572 				 * seen and we've seen exactly our rexmt
2573 				 * threshold of them, assume a packet
2574 				 * has been dropped and retransmit it.
2575 				 * Kludge snd_nxt & the congestion
2576 				 * window so we send only this one
2577 				 * packet.
2578 				 *
2579 				 * We know we're losing at the current
2580 				 * window size so do congestion avoidance
2581 				 * (set ssthresh to half the current window
2582 				 * and pull our congestion window back to
2583 				 * the new ssthresh).
2584 				 *
2585 				 * Dup acks mean that packets have left the
2586 				 * network (they're now cached at the receiver)
2587 				 * so bump cwnd by the amount in the receiver
2588 				 * to keep a constant cwnd packets in the
2589 				 * network.
2590 				 *
2591 				 * When using TCP ECN, notify the peer that
2592 				 * we reduced the cwnd.
2593 				 */
2594 				/*
2595 				 * Following 2 kinds of acks should not affect
2596 				 * dupack counting:
2597 				 * 1) Old acks
2598 				 * 2) Acks with SACK but without any new SACK
2599 				 * information in them. These could result from
2600 				 * any anomaly in the network like a switch
2601 				 * duplicating packets or a possible DoS attack.
2602 				 */
2603 				if (th->th_ack != tp->snd_una ||
2604 				    ((tp->t_flags & TF_SACK_PERMIT) &&
2605 				    (to.to_flags & TOF_SACK) &&
2606 				    !sack_changed))
2607 					break;
2608 				else if (!tcp_timer_active(tp, TT_REXMT))
2609 					tp->t_dupacks = 0;
2610 				else if (++tp->t_dupacks > tcprexmtthresh ||
2611 				     IN_FASTRECOVERY(tp->t_flags)) {
2612 					cc_ack_received(tp, th, nsegs,
2613 					    CC_DUPACK);
2614 					if (V_tcp_do_prr &&
2615 					    IN_FASTRECOVERY(tp->t_flags) &&
2616 					    (tp->t_flags & TF_SACK_PERMIT)) {
2617 						tcp_do_prr_ack(tp, th, &to);
2618 					} else if ((tp->t_flags & TF_SACK_PERMIT) &&
2619 					    (to.to_flags & TOF_SACK) &&
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_newsack)
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_newsack &&
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 					    (to.to_flags & TOF_SACK)) {
2699 						TCPSTAT_INC(
2700 						    tcps_sack_recovery_episode);
2701 						tp->snd_recover = tp->snd_nxt;
2702 						tp->snd_cwnd = maxseg;
2703 						(void) tp->t_fb->tfb_tcp_output(tp);
2704 						if (SEQ_GT(th->th_ack, tp->snd_una))
2705 							goto resume_partialack;
2706 						goto drop;
2707 					}
2708 					tp->snd_nxt = th->th_ack;
2709 					tp->snd_cwnd = maxseg;
2710 					(void) tp->t_fb->tfb_tcp_output(tp);
2711 					KASSERT(tp->snd_limited <= 2,
2712 					    ("%s: tp->snd_limited too big",
2713 					    __func__));
2714 					tp->snd_cwnd = tp->snd_ssthresh +
2715 					     maxseg *
2716 					     (tp->t_dupacks - tp->snd_limited);
2717 					if (SEQ_GT(onxt, tp->snd_nxt))
2718 						tp->snd_nxt = onxt;
2719 					goto drop;
2720 				} else if (V_tcp_do_rfc3042) {
2721 					/*
2722 					 * Process first and second duplicate
2723 					 * ACKs. Each indicates a segment
2724 					 * leaving the network, creating room
2725 					 * for more. Make sure we can send a
2726 					 * packet on reception of each duplicate
2727 					 * ACK by increasing snd_cwnd by one
2728 					 * segment. Restore the original
2729 					 * snd_cwnd after packet transmission.
2730 					 */
2731 					cc_ack_received(tp, th, nsegs,
2732 					    CC_DUPACK);
2733 					uint32_t oldcwnd = tp->snd_cwnd;
2734 					tcp_seq oldsndmax = tp->snd_max;
2735 					u_int sent;
2736 					int avail;
2737 
2738 					KASSERT(tp->t_dupacks == 1 ||
2739 					    tp->t_dupacks == 2,
2740 					    ("%s: dupacks not 1 or 2",
2741 					    __func__));
2742 					if (tp->t_dupacks == 1)
2743 						tp->snd_limited = 0;
2744 					tp->snd_cwnd =
2745 					    (tp->snd_nxt - tp->snd_una) +
2746 					    (tp->t_dupacks - tp->snd_limited) *
2747 					    maxseg;
2748 					/*
2749 					 * Only call tcp_output when there
2750 					 * is new data available to be sent.
2751 					 * Otherwise we would send pure ACKs.
2752 					 */
2753 					SOCKBUF_LOCK(&so->so_snd);
2754 					avail = sbavail(&so->so_snd) -
2755 					    (tp->snd_nxt - tp->snd_una);
2756 					SOCKBUF_UNLOCK(&so->so_snd);
2757 					if (avail > 0)
2758 						(void) tp->t_fb->tfb_tcp_output(tp);
2759 					sent = tp->snd_max - oldsndmax;
2760 					if (sent > maxseg) {
2761 						KASSERT((tp->t_dupacks == 2 &&
2762 						    tp->snd_limited == 0) ||
2763 						   (sent == maxseg + 1 &&
2764 						    tp->t_flags & TF_SENTFIN),
2765 						    ("%s: sent too much",
2766 						    __func__));
2767 						tp->snd_limited = 2;
2768 					} else if (sent > 0)
2769 						++tp->snd_limited;
2770 					tp->snd_cwnd = oldcwnd;
2771 					goto drop;
2772 				}
2773 			}
2774 			break;
2775 		} else {
2776 			/*
2777 			 * This ack is advancing the left edge, reset the
2778 			 * counter.
2779 			 */
2780 			tp->t_dupacks = 0;
2781 			/*
2782 			 * If this ack also has new SACK info, increment the
2783 			 * counter as per rfc6675. The variable
2784 			 * sack_changed tracks all changes to the SACK
2785 			 * scoreboard, including when partial ACKs without
2786 			 * SACK options are received, and clear the scoreboard
2787 			 * from the left side. Such partial ACKs should not be
2788 			 * counted as dupacks here.
2789 			 */
2790 			if ((tp->t_flags & TF_SACK_PERMIT) &&
2791 			    (to.to_flags & TOF_SACK) &&
2792 			    sack_changed) {
2793 				tp->t_dupacks++;
2794 				/* limit overhead by setting maxseg last */
2795 				if (!IN_FASTRECOVERY(tp->t_flags) &&
2796 				    (tp->sackhint.sacked_bytes >
2797 				    ((tcprexmtthresh - 1) *
2798 				    (maxseg = tcp_maxseg(tp))))) {
2799 					goto enter_recovery;
2800 				}
2801 			}
2802 		}
2803 
2804 resume_partialack:
2805 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2806 		    ("%s: th_ack <= snd_una", __func__));
2807 
2808 		/*
2809 		 * If the congestion window was inflated to account
2810 		 * for the other side's cached packets, retract it.
2811 		 */
2812 		if (IN_FASTRECOVERY(tp->t_flags)) {
2813 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2814 				if (tp->t_flags & TF_SACK_PERMIT)
2815 					if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2816 						tcp_timer_activate(tp, TT_REXMT, 0);
2817 						tp->t_rtttime = 0;
2818 						tcp_do_prr_ack(tp, th, &to);
2819 						tp->t_flags |= TF_ACKNOW;
2820 						(void) tcp_output(tp);
2821 					} else
2822 						tcp_sack_partialack(tp, th);
2823 				else
2824 					tcp_newreno_partial_ack(tp, th);
2825 			} else
2826 				cc_post_recovery(tp, th);
2827 		} else if (IN_CONGRECOVERY(tp->t_flags)) {
2828 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2829 				if (V_tcp_do_prr) {
2830 					tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2831 					tp->snd_fack = th->th_ack;
2832 					tcp_do_prr_ack(tp, th, &to);
2833 					(void) tcp_output(tp);
2834 				}
2835 			} else
2836 				cc_post_recovery(tp, th);
2837 		}
2838 		/*
2839 		 * If we reach this point, ACK is not a duplicate,
2840 		 *     i.e., it ACKs something we sent.
2841 		 */
2842 		if (tp->t_flags & TF_NEEDSYN) {
2843 			/*
2844 			 * T/TCP: Connection was half-synchronized, and our
2845 			 * SYN has been ACK'd (so connection is now fully
2846 			 * synchronized).  Go to non-starred state,
2847 			 * increment snd_una for ACK of SYN, and check if
2848 			 * we can do window scaling.
2849 			 */
2850 			tp->t_flags &= ~TF_NEEDSYN;
2851 			tp->snd_una++;
2852 			/* Do window scaling? */
2853 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2854 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2855 				tp->rcv_scale = tp->request_r_scale;
2856 				/* Send window already scaled. */
2857 			}
2858 		}
2859 
2860 process_ACK:
2861 		INP_WLOCK_ASSERT(tp->t_inpcb);
2862 
2863 		/*
2864 		 * Adjust for the SYN bit in sequence space,
2865 		 * but don't account for it in cwnd calculations.
2866 		 * This is for the SYN_RECEIVED, non-simultaneous
2867 		 * SYN case. SYN_SENT and simultaneous SYN are
2868 		 * treated elsewhere.
2869 		 */
2870 		if (incforsyn)
2871 			tp->snd_una++;
2872 		acked = BYTES_THIS_ACK(tp, th);
2873 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2874 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2875 		    tp->snd_una, th->th_ack, tp, m));
2876 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2877 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2878 
2879 		/*
2880 		 * If we just performed our first retransmit, and the ACK
2881 		 * arrives within our recovery window, then it was a mistake
2882 		 * to do the retransmit in the first place.  Recover our
2883 		 * original cwnd and ssthresh, and proceed to transmit where
2884 		 * we left off.
2885 		 */
2886 		if (tp->t_rxtshift == 1 &&
2887 		    tp->t_flags & TF_PREVVALID &&
2888 		    tp->t_badrxtwin &&
2889 		    SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2890 			cc_cong_signal(tp, th, CC_RTO_ERR);
2891 
2892 		/*
2893 		 * If we have a timestamp reply, update smoothed
2894 		 * round trip time.  If no timestamp is present but
2895 		 * transmit timer is running and timed sequence
2896 		 * number was acked, update smoothed round trip time.
2897 		 * Since we now have an rtt measurement, cancel the
2898 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2899 		 * Recompute the initial retransmit timer.
2900 		 *
2901 		 * Some boxes send broken timestamp replies
2902 		 * during the SYN+ACK phase, ignore
2903 		 * timestamps of 0 or we could calculate a
2904 		 * huge RTT and blow up the retransmit timer.
2905 		 */
2906 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2907 			uint32_t t;
2908 
2909 			t = tcp_ts_getticks() - to.to_tsecr;
2910 			if (!tp->t_rttlow || tp->t_rttlow > t)
2911 				tp->t_rttlow = t;
2912 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2913 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2914 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2915 				tp->t_rttlow = ticks - tp->t_rtttime;
2916 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2917 		}
2918 
2919 		/*
2920 		 * If all outstanding data is acked, stop retransmit
2921 		 * timer and remember to restart (more output or persist).
2922 		 * If there is more data to be acked, restart retransmit
2923 		 * timer, using current (possibly backed-off) value.
2924 		 */
2925 		if (th->th_ack == tp->snd_max) {
2926 			tcp_timer_activate(tp, TT_REXMT, 0);
2927 			needoutput = 1;
2928 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2929 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2930 
2931 		/*
2932 		 * If no data (only SYN) was ACK'd,
2933 		 *    skip rest of ACK processing.
2934 		 */
2935 		if (acked == 0)
2936 			goto step6;
2937 
2938 		/*
2939 		 * Let the congestion control algorithm update congestion
2940 		 * control related information. This typically means increasing
2941 		 * the congestion window.
2942 		 */
2943 		cc_ack_received(tp, th, nsegs, CC_ACK);
2944 
2945 		SOCKBUF_LOCK(&so->so_snd);
2946 		if (acked > sbavail(&so->so_snd)) {
2947 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2948 				tp->snd_wnd -= sbavail(&so->so_snd);
2949 			else
2950 				tp->snd_wnd = 0;
2951 			mfree = sbcut_locked(&so->so_snd,
2952 			    (int)sbavail(&so->so_snd));
2953 			ourfinisacked = 1;
2954 		} else {
2955 			mfree = sbcut_locked(&so->so_snd, acked);
2956 			if (tp->snd_wnd >= (uint32_t) acked)
2957 				tp->snd_wnd -= acked;
2958 			else
2959 				tp->snd_wnd = 0;
2960 			ourfinisacked = 0;
2961 		}
2962 		/* NB: sowwakeup_locked() does an implicit unlock. */
2963 		sowwakeup_locked(so);
2964 		m_freem(mfree);
2965 		/* Detect una wraparound. */
2966 		if (!IN_RECOVERY(tp->t_flags) &&
2967 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2968 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2969 			tp->snd_recover = th->th_ack - 1;
2970 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2971 		if (IN_RECOVERY(tp->t_flags) &&
2972 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2973 			EXIT_RECOVERY(tp->t_flags);
2974 		}
2975 		tp->snd_una = th->th_ack;
2976 		if (tp->t_flags & TF_SACK_PERMIT) {
2977 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2978 				tp->snd_recover = tp->snd_una;
2979 		}
2980 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2981 			tp->snd_nxt = tp->snd_una;
2982 
2983 		switch (tp->t_state) {
2984 		/*
2985 		 * In FIN_WAIT_1 STATE in addition to the processing
2986 		 * for the ESTABLISHED state if our FIN is now acknowledged
2987 		 * then enter FIN_WAIT_2.
2988 		 */
2989 		case TCPS_FIN_WAIT_1:
2990 			if (ourfinisacked) {
2991 				/*
2992 				 * If we can't receive any more
2993 				 * data, then closing user can proceed.
2994 				 * Starting the timer is contrary to the
2995 				 * specification, but if we don't get a FIN
2996 				 * we'll hang forever.
2997 				 *
2998 				 * XXXjl:
2999 				 * we should release the tp also, and use a
3000 				 * compressed state.
3001 				 */
3002 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3003 					soisdisconnected(so);
3004 					tcp_timer_activate(tp, TT_2MSL,
3005 					    (tcp_fast_finwait2_recycle ?
3006 					    tcp_finwait2_timeout :
3007 					    TP_MAXIDLE(tp)));
3008 				}
3009 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
3010 			}
3011 			break;
3012 
3013 		/*
3014 		 * In CLOSING STATE in addition to the processing for
3015 		 * the ESTABLISHED state if the ACK acknowledges our FIN
3016 		 * then enter the TIME-WAIT state, otherwise ignore
3017 		 * the segment.
3018 		 */
3019 		case TCPS_CLOSING:
3020 			if (ourfinisacked) {
3021 				tcp_twstart(tp);
3022 				m_freem(m);
3023 				return;
3024 			}
3025 			break;
3026 
3027 		/*
3028 		 * In LAST_ACK, we may still be waiting for data to drain
3029 		 * and/or to be acked, as well as for the ack of our FIN.
3030 		 * If our FIN is now acknowledged, delete the TCB,
3031 		 * enter the closed state and return.
3032 		 */
3033 		case TCPS_LAST_ACK:
3034 			if (ourfinisacked) {
3035 				tp = tcp_close(tp);
3036 				goto drop;
3037 			}
3038 			break;
3039 		}
3040 	}
3041 
3042 step6:
3043 	INP_WLOCK_ASSERT(tp->t_inpcb);
3044 
3045 	/*
3046 	 * Update window information.
3047 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3048 	 */
3049 	if ((thflags & TH_ACK) &&
3050 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3051 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3052 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3053 		/* keep track of pure window updates */
3054 		if (tlen == 0 &&
3055 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3056 			TCPSTAT_INC(tcps_rcvwinupd);
3057 		tp->snd_wnd = tiwin;
3058 		tp->snd_wl1 = th->th_seq;
3059 		tp->snd_wl2 = th->th_ack;
3060 		if (tp->snd_wnd > tp->max_sndwnd)
3061 			tp->max_sndwnd = tp->snd_wnd;
3062 		needoutput = 1;
3063 	}
3064 
3065 	/*
3066 	 * Process segments with URG.
3067 	 */
3068 	if ((thflags & TH_URG) && th->th_urp &&
3069 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3070 		/*
3071 		 * This is a kludge, but if we receive and accept
3072 		 * random urgent pointers, we'll crash in
3073 		 * soreceive.  It's hard to imagine someone
3074 		 * actually wanting to send this much urgent data.
3075 		 */
3076 		SOCKBUF_LOCK(&so->so_rcv);
3077 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3078 			th->th_urp = 0;			/* XXX */
3079 			thflags &= ~TH_URG;		/* XXX */
3080 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3081 			goto dodata;			/* XXX */
3082 		}
3083 		/*
3084 		 * If this segment advances the known urgent pointer,
3085 		 * then mark the data stream.  This should not happen
3086 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3087 		 * a FIN has been received from the remote side.
3088 		 * In these states we ignore the URG.
3089 		 *
3090 		 * According to RFC961 (Assigned Protocols),
3091 		 * the urgent pointer points to the last octet
3092 		 * of urgent data.  We continue, however,
3093 		 * to consider it to indicate the first octet
3094 		 * of data past the urgent section as the original
3095 		 * spec states (in one of two places).
3096 		 */
3097 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3098 			tp->rcv_up = th->th_seq + th->th_urp;
3099 			so->so_oobmark = sbavail(&so->so_rcv) +
3100 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3101 			if (so->so_oobmark == 0)
3102 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3103 			sohasoutofband(so);
3104 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3105 		}
3106 		SOCKBUF_UNLOCK(&so->so_rcv);
3107 		/*
3108 		 * Remove out of band data so doesn't get presented to user.
3109 		 * This can happen independent of advancing the URG pointer,
3110 		 * but if two URG's are pending at once, some out-of-band
3111 		 * data may creep in... ick.
3112 		 */
3113 		if (th->th_urp <= (uint32_t)tlen &&
3114 		    !(so->so_options & SO_OOBINLINE)) {
3115 			/* hdr drop is delayed */
3116 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3117 		}
3118 	} else {
3119 		/*
3120 		 * If no out of band data is expected,
3121 		 * pull receive urgent pointer along
3122 		 * with the receive window.
3123 		 */
3124 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3125 			tp->rcv_up = tp->rcv_nxt;
3126 	}
3127 dodata:							/* XXX */
3128 	INP_WLOCK_ASSERT(tp->t_inpcb);
3129 
3130 	/*
3131 	 * Process the segment text, merging it into the TCP sequencing queue,
3132 	 * and arranging for acknowledgment of receipt if necessary.
3133 	 * This process logically involves adjusting tp->rcv_wnd as data
3134 	 * is presented to the user (this happens in tcp_usrreq.c,
3135 	 * case PRU_RCVD).  If a FIN has already been received on this
3136 	 * connection then we just ignore the text.
3137 	 */
3138 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3139 		   IS_FASTOPEN(tp->t_flags));
3140 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3141 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3142 		tcp_seq save_start = th->th_seq;
3143 		tcp_seq save_rnxt  = tp->rcv_nxt;
3144 		int     save_tlen  = tlen;
3145 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3146 		/*
3147 		 * Insert segment which includes th into TCP reassembly queue
3148 		 * with control block tp.  Set thflags to whether reassembly now
3149 		 * includes a segment with FIN.  This handles the common case
3150 		 * inline (segment is the next to be received on an established
3151 		 * connection, and the queue is empty), avoiding linkage into
3152 		 * and removal from the queue and repetition of various
3153 		 * conversions.
3154 		 * Set DELACK for segments received in order, but ack
3155 		 * immediately when segments are out of order (so
3156 		 * fast retransmit can work).
3157 		 */
3158 		if (th->th_seq == tp->rcv_nxt &&
3159 		    SEGQ_EMPTY(tp) &&
3160 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3161 		     tfo_syn)) {
3162 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3163 				tp->t_flags |= TF_DELACK;
3164 			else
3165 				tp->t_flags |= TF_ACKNOW;
3166 			tp->rcv_nxt += tlen;
3167 			if (tlen &&
3168 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3169 			    (tp->t_fbyte_in == 0)) {
3170 				tp->t_fbyte_in = ticks;
3171 				if (tp->t_fbyte_in == 0)
3172 					tp->t_fbyte_in = 1;
3173 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3174 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3175 			}
3176 			thflags = th->th_flags & TH_FIN;
3177 			TCPSTAT_INC(tcps_rcvpack);
3178 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3179 			SOCKBUF_LOCK(&so->so_rcv);
3180 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3181 				m_freem(m);
3182 			else
3183 				sbappendstream_locked(&so->so_rcv, m, 0);
3184 			tp->t_flags |= TF_WAKESOR;
3185 		} else {
3186 			/*
3187 			 * XXX: Due to the header drop above "th" is
3188 			 * theoretically invalid by now.  Fortunately
3189 			 * m_adj() doesn't actually frees any mbufs
3190 			 * when trimming from the head.
3191 			 */
3192 			tcp_seq temp = save_start;
3193 
3194 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3195 			tp->t_flags |= TF_ACKNOW;
3196 		}
3197 		if ((tp->t_flags & TF_SACK_PERMIT) &&
3198 		    (save_tlen > 0) &&
3199 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
3200 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3201 				/*
3202 				 * DSACK actually handled in the fastpath
3203 				 * above.
3204 				 */
3205 				tcp_update_sack_list(tp, save_start,
3206 				    save_start + save_tlen);
3207 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3208 				if ((tp->rcv_numsacks >= 1) &&
3209 				    (tp->sackblks[0].end == save_start)) {
3210 					/*
3211 					 * Partial overlap, recorded at todrop
3212 					 * above.
3213 					 */
3214 					tcp_update_sack_list(tp,
3215 					    tp->sackblks[0].start,
3216 					    tp->sackblks[0].end);
3217 				} else {
3218 					tcp_update_dsack_list(tp, save_start,
3219 					    save_start + save_tlen);
3220 				}
3221 			} else if (tlen >= save_tlen) {
3222 				/* Update of sackblks. */
3223 				tcp_update_dsack_list(tp, save_start,
3224 				    save_start + save_tlen);
3225 			} else if (tlen > 0) {
3226 				tcp_update_dsack_list(tp, save_start,
3227 				    save_start + tlen);
3228 			}
3229 		}
3230 		tcp_handle_wakeup(tp, so);
3231 #if 0
3232 		/*
3233 		 * Note the amount of data that peer has sent into
3234 		 * our window, in order to estimate the sender's
3235 		 * buffer size.
3236 		 * XXX: Unused.
3237 		 */
3238 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3239 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3240 		else
3241 			len = so->so_rcv.sb_hiwat;
3242 #endif
3243 	} else {
3244 		m_freem(m);
3245 		thflags &= ~TH_FIN;
3246 	}
3247 
3248 	/*
3249 	 * If FIN is received ACK the FIN and let the user know
3250 	 * that the connection is closing.
3251 	 */
3252 	if (thflags & TH_FIN) {
3253 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3254 			/* The socket upcall is handled by socantrcvmore. */
3255 			socantrcvmore(so);
3256 			/*
3257 			 * If connection is half-synchronized
3258 			 * (ie NEEDSYN flag on) then delay ACK,
3259 			 * so it may be piggybacked when SYN is sent.
3260 			 * Otherwise, since we received a FIN then no
3261 			 * more input can be expected, send ACK now.
3262 			 */
3263 			if (tp->t_flags & TF_NEEDSYN)
3264 				tp->t_flags |= TF_DELACK;
3265 			else
3266 				tp->t_flags |= TF_ACKNOW;
3267 			tp->rcv_nxt++;
3268 		}
3269 		switch (tp->t_state) {
3270 		/*
3271 		 * In SYN_RECEIVED and ESTABLISHED STATES
3272 		 * enter the CLOSE_WAIT state.
3273 		 */
3274 		case TCPS_SYN_RECEIVED:
3275 			tp->t_starttime = ticks;
3276 			/* FALLTHROUGH */
3277 		case TCPS_ESTABLISHED:
3278 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3279 			break;
3280 
3281 		/*
3282 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3283 		 * enter the CLOSING state.
3284 		 */
3285 		case TCPS_FIN_WAIT_1:
3286 			tcp_state_change(tp, TCPS_CLOSING);
3287 			break;
3288 
3289 		/*
3290 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3291 		 * starting the time-wait timer, turning off the other
3292 		 * standard timers.
3293 		 */
3294 		case TCPS_FIN_WAIT_2:
3295 			tcp_twstart(tp);
3296 			return;
3297 		}
3298 	}
3299 #ifdef TCPDEBUG
3300 	if (so->so_options & SO_DEBUG)
3301 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3302 			  &tcp_savetcp, 0);
3303 #endif
3304 	TCP_PROBE3(debug__input, tp, th, m);
3305 
3306 	/*
3307 	 * Return any desired output.
3308 	 */
3309 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3310 		(void) tp->t_fb->tfb_tcp_output(tp);
3311 
3312 check_delack:
3313 	INP_WLOCK_ASSERT(tp->t_inpcb);
3314 
3315 	if (tp->t_flags & TF_DELACK) {
3316 		tp->t_flags &= ~TF_DELACK;
3317 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3318 	}
3319 	INP_WUNLOCK(tp->t_inpcb);
3320 	return;
3321 
3322 dropafterack:
3323 	/*
3324 	 * Generate an ACK dropping incoming segment if it occupies
3325 	 * sequence space, where the ACK reflects our state.
3326 	 *
3327 	 * We can now skip the test for the RST flag since all
3328 	 * paths to this code happen after packets containing
3329 	 * RST have been dropped.
3330 	 *
3331 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3332 	 * segment we received passes the SYN-RECEIVED ACK test.
3333 	 * If it fails send a RST.  This breaks the loop in the
3334 	 * "LAND" DoS attack, and also prevents an ACK storm
3335 	 * between two listening ports that have been sent forged
3336 	 * SYN segments, each with the source address of the other.
3337 	 */
3338 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3339 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3340 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3341 		rstreason = BANDLIM_RST_OPENPORT;
3342 		goto dropwithreset;
3343 	}
3344 #ifdef TCPDEBUG
3345 	if (so->so_options & SO_DEBUG)
3346 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3347 			  &tcp_savetcp, 0);
3348 #endif
3349 	TCP_PROBE3(debug__input, tp, th, m);
3350 	tp->t_flags |= TF_ACKNOW;
3351 	(void) tp->t_fb->tfb_tcp_output(tp);
3352 	INP_WUNLOCK(tp->t_inpcb);
3353 	m_freem(m);
3354 	return;
3355 
3356 dropwithreset:
3357 	if (tp != NULL) {
3358 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3359 		INP_WUNLOCK(tp->t_inpcb);
3360 	} else
3361 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3362 	return;
3363 
3364 drop:
3365 	/*
3366 	 * Drop space held by incoming segment and return.
3367 	 */
3368 #ifdef TCPDEBUG
3369 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3370 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3371 			  &tcp_savetcp, 0);
3372 #endif
3373 	TCP_PROBE3(debug__input, tp, th, m);
3374 	if (tp != NULL) {
3375 		INP_WUNLOCK(tp->t_inpcb);
3376 	}
3377 	m_freem(m);
3378 }
3379 
3380 /*
3381  * Issue RST and make ACK acceptable to originator of segment.
3382  * The mbuf must still include the original packet header.
3383  * tp may be NULL.
3384  */
3385 void
3386 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3387     int tlen, int rstreason)
3388 {
3389 #ifdef INET
3390 	struct ip *ip;
3391 #endif
3392 #ifdef INET6
3393 	struct ip6_hdr *ip6;
3394 #endif
3395 
3396 	if (tp != NULL) {
3397 		INP_LOCK_ASSERT(tp->t_inpcb);
3398 	}
3399 
3400 	/* Don't bother if destination was broadcast/multicast. */
3401 	if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3402 		goto drop;
3403 #ifdef INET6
3404 	if (mtod(m, struct ip *)->ip_v == 6) {
3405 		ip6 = mtod(m, struct ip6_hdr *);
3406 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3407 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3408 			goto drop;
3409 		/* IPv6 anycast check is done at tcp6_input() */
3410 	}
3411 #endif
3412 #if defined(INET) && defined(INET6)
3413 	else
3414 #endif
3415 #ifdef INET
3416 	{
3417 		ip = mtod(m, struct ip *);
3418 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3419 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3420 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3421 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3422 			goto drop;
3423 	}
3424 #endif
3425 
3426 	/* Perform bandwidth limiting. */
3427 	if (badport_bandlim(rstreason) < 0)
3428 		goto drop;
3429 
3430 	/* tcp_respond consumes the mbuf chain. */
3431 	if (th->th_flags & TH_ACK) {
3432 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3433 		    th->th_ack, TH_RST);
3434 	} else {
3435 		if (th->th_flags & TH_SYN)
3436 			tlen++;
3437 		if (th->th_flags & TH_FIN)
3438 			tlen++;
3439 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3440 		    (tcp_seq)0, TH_RST|TH_ACK);
3441 	}
3442 	return;
3443 drop:
3444 	m_freem(m);
3445 }
3446 
3447 /*
3448  * Parse TCP options and place in tcpopt.
3449  */
3450 void
3451 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3452 {
3453 	int opt, optlen;
3454 
3455 	to->to_flags = 0;
3456 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3457 		opt = cp[0];
3458 		if (opt == TCPOPT_EOL)
3459 			break;
3460 		if (opt == TCPOPT_NOP)
3461 			optlen = 1;
3462 		else {
3463 			if (cnt < 2)
3464 				break;
3465 			optlen = cp[1];
3466 			if (optlen < 2 || optlen > cnt)
3467 				break;
3468 		}
3469 		switch (opt) {
3470 		case TCPOPT_MAXSEG:
3471 			if (optlen != TCPOLEN_MAXSEG)
3472 				continue;
3473 			if (!(flags & TO_SYN))
3474 				continue;
3475 			to->to_flags |= TOF_MSS;
3476 			bcopy((char *)cp + 2,
3477 			    (char *)&to->to_mss, sizeof(to->to_mss));
3478 			to->to_mss = ntohs(to->to_mss);
3479 			break;
3480 		case TCPOPT_WINDOW:
3481 			if (optlen != TCPOLEN_WINDOW)
3482 				continue;
3483 			if (!(flags & TO_SYN))
3484 				continue;
3485 			to->to_flags |= TOF_SCALE;
3486 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3487 			break;
3488 		case TCPOPT_TIMESTAMP:
3489 			if (optlen != TCPOLEN_TIMESTAMP)
3490 				continue;
3491 			to->to_flags |= TOF_TS;
3492 			bcopy((char *)cp + 2,
3493 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3494 			to->to_tsval = ntohl(to->to_tsval);
3495 			bcopy((char *)cp + 6,
3496 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3497 			to->to_tsecr = ntohl(to->to_tsecr);
3498 			break;
3499 		case TCPOPT_SIGNATURE:
3500 			/*
3501 			 * In order to reply to a host which has set the
3502 			 * TCP_SIGNATURE option in its initial SYN, we have
3503 			 * to record the fact that the option was observed
3504 			 * here for the syncache code to perform the correct
3505 			 * response.
3506 			 */
3507 			if (optlen != TCPOLEN_SIGNATURE)
3508 				continue;
3509 			to->to_flags |= TOF_SIGNATURE;
3510 			to->to_signature = cp + 2;
3511 			break;
3512 		case TCPOPT_SACK_PERMITTED:
3513 			if (optlen != TCPOLEN_SACK_PERMITTED)
3514 				continue;
3515 			if (!(flags & TO_SYN))
3516 				continue;
3517 			if (!V_tcp_do_sack)
3518 				continue;
3519 			to->to_flags |= TOF_SACKPERM;
3520 			break;
3521 		case TCPOPT_SACK:
3522 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3523 				continue;
3524 			if (flags & TO_SYN)
3525 				continue;
3526 			to->to_flags |= TOF_SACK;
3527 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3528 			to->to_sacks = cp + 2;
3529 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3530 			break;
3531 		case TCPOPT_FAST_OPEN:
3532 			/*
3533 			 * Cookie length validation is performed by the
3534 			 * server side cookie checking code or the client
3535 			 * side cookie cache update code.
3536 			 */
3537 			if (!(flags & TO_SYN))
3538 				continue;
3539 			if (!V_tcp_fastopen_client_enable &&
3540 			    !V_tcp_fastopen_server_enable)
3541 				continue;
3542 			to->to_flags |= TOF_FASTOPEN;
3543 			to->to_tfo_len = optlen - 2;
3544 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3545 			break;
3546 		default:
3547 			continue;
3548 		}
3549 	}
3550 }
3551 
3552 /*
3553  * Pull out of band byte out of a segment so
3554  * it doesn't appear in the user's data queue.
3555  * It is still reflected in the segment length for
3556  * sequencing purposes.
3557  */
3558 void
3559 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3560     int off)
3561 {
3562 	int cnt = off + th->th_urp - 1;
3563 
3564 	while (cnt >= 0) {
3565 		if (m->m_len > cnt) {
3566 			char *cp = mtod(m, caddr_t) + cnt;
3567 			struct tcpcb *tp = sototcpcb(so);
3568 
3569 			INP_WLOCK_ASSERT(tp->t_inpcb);
3570 
3571 			tp->t_iobc = *cp;
3572 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3573 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3574 			m->m_len--;
3575 			if (m->m_flags & M_PKTHDR)
3576 				m->m_pkthdr.len--;
3577 			return;
3578 		}
3579 		cnt -= m->m_len;
3580 		m = m->m_next;
3581 		if (m == NULL)
3582 			break;
3583 	}
3584 	panic("tcp_pulloutofband");
3585 }
3586 
3587 /*
3588  * Collect new round-trip time estimate
3589  * and update averages and current timeout.
3590  */
3591 void
3592 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3593 {
3594 	int delta;
3595 
3596 	INP_WLOCK_ASSERT(tp->t_inpcb);
3597 
3598 	TCPSTAT_INC(tcps_rttupdated);
3599 	tp->t_rttupdated++;
3600 #ifdef STATS
3601 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3602 	    imax(0, rtt * 1000 / hz));
3603 #endif
3604 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3605 		/*
3606 		 * srtt is stored as fixed point with 5 bits after the
3607 		 * binary point (i.e., scaled by 8).  The following magic
3608 		 * is equivalent to the smoothing algorithm in rfc793 with
3609 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3610 		 * point).  Adjust rtt to origin 0.
3611 		 */
3612 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3613 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3614 
3615 		if ((tp->t_srtt += delta) <= 0)
3616 			tp->t_srtt = 1;
3617 
3618 		/*
3619 		 * We accumulate a smoothed rtt variance (actually, a
3620 		 * smoothed mean difference), then set the retransmit
3621 		 * timer to smoothed rtt + 4 times the smoothed variance.
3622 		 * rttvar is stored as fixed point with 4 bits after the
3623 		 * binary point (scaled by 16).  The following is
3624 		 * equivalent to rfc793 smoothing with an alpha of .75
3625 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3626 		 * rfc793's wired-in beta.
3627 		 */
3628 		if (delta < 0)
3629 			delta = -delta;
3630 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3631 		if ((tp->t_rttvar += delta) <= 0)
3632 			tp->t_rttvar = 1;
3633 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3634 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3635 	} else {
3636 		/*
3637 		 * No rtt measurement yet - use the unsmoothed rtt.
3638 		 * Set the variance to half the rtt (so our first
3639 		 * retransmit happens at 3*rtt).
3640 		 */
3641 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3642 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3643 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3644 	}
3645 	tp->t_rtttime = 0;
3646 	tp->t_rxtshift = 0;
3647 
3648 	/*
3649 	 * the retransmit should happen at rtt + 4 * rttvar.
3650 	 * Because of the way we do the smoothing, srtt and rttvar
3651 	 * will each average +1/2 tick of bias.  When we compute
3652 	 * the retransmit timer, we want 1/2 tick of rounding and
3653 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3654 	 * firing of the timer.  The bias will give us exactly the
3655 	 * 1.5 tick we need.  But, because the bias is
3656 	 * statistical, we have to test that we don't drop below
3657 	 * the minimum feasible timer (which is 2 ticks).
3658 	 */
3659 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3660 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3661 
3662 	/*
3663 	 * We received an ack for a packet that wasn't retransmitted;
3664 	 * it is probably safe to discard any error indications we've
3665 	 * received recently.  This isn't quite right, but close enough
3666 	 * for now (a route might have failed after we sent a segment,
3667 	 * and the return path might not be symmetrical).
3668 	 */
3669 	tp->t_softerror = 0;
3670 }
3671 
3672 /*
3673  * Determine a reasonable value for maxseg size.
3674  * If the route is known, check route for mtu.
3675  * If none, use an mss that can be handled on the outgoing interface
3676  * without forcing IP to fragment.  If no route is found, route has no mtu,
3677  * or the destination isn't local, use a default, hopefully conservative
3678  * size (usually 512 or the default IP max size, but no more than the mtu
3679  * of the interface), as we can't discover anything about intervening
3680  * gateways or networks.  We also initialize the congestion/slow start
3681  * window to be a single segment if the destination isn't local.
3682  * While looking at the routing entry, we also initialize other path-dependent
3683  * parameters from pre-set or cached values in the routing entry.
3684  *
3685  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3686  * IP options, e.g. IPSEC data, since length of this data may vary, and
3687  * thus it is calculated for every segment separately in tcp_output().
3688  *
3689  * NOTE that this routine is only called when we process an incoming
3690  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3691  * settings are handled in tcp_mssopt().
3692  */
3693 void
3694 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3695     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3696 {
3697 	int mss = 0;
3698 	uint32_t maxmtu = 0;
3699 	struct inpcb *inp = tp->t_inpcb;
3700 	struct hc_metrics_lite metrics;
3701 #ifdef INET6
3702 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3703 	size_t min_protoh = isipv6 ?
3704 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3705 			    sizeof (struct tcpiphdr);
3706 #else
3707 	 size_t min_protoh = sizeof(struct tcpiphdr);
3708 #endif
3709 
3710 	INP_WLOCK_ASSERT(tp->t_inpcb);
3711 
3712 	if (tp->t_port)
3713 		min_protoh += V_tcp_udp_tunneling_overhead;
3714 	if (mtuoffer != -1) {
3715 		KASSERT(offer == -1, ("%s: conflict", __func__));
3716 		offer = mtuoffer - min_protoh;
3717 	}
3718 
3719 	/* Initialize. */
3720 #ifdef INET6
3721 	if (isipv6) {
3722 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3723 		tp->t_maxseg = V_tcp_v6mssdflt;
3724 	}
3725 #endif
3726 #if defined(INET) && defined(INET6)
3727 	else
3728 #endif
3729 #ifdef INET
3730 	{
3731 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3732 		tp->t_maxseg = V_tcp_mssdflt;
3733 	}
3734 #endif
3735 
3736 	/*
3737 	 * No route to sender, stay with default mss and return.
3738 	 */
3739 	if (maxmtu == 0) {
3740 		/*
3741 		 * In case we return early we need to initialize metrics
3742 		 * to a defined state as tcp_hc_get() would do for us
3743 		 * if there was no cache hit.
3744 		 */
3745 		if (metricptr != NULL)
3746 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3747 		return;
3748 	}
3749 
3750 	/* What have we got? */
3751 	switch (offer) {
3752 		case 0:
3753 			/*
3754 			 * Offer == 0 means that there was no MSS on the SYN
3755 			 * segment, in this case we use tcp_mssdflt as
3756 			 * already assigned to t_maxseg above.
3757 			 */
3758 			offer = tp->t_maxseg;
3759 			break;
3760 
3761 		case -1:
3762 			/*
3763 			 * Offer == -1 means that we didn't receive SYN yet.
3764 			 */
3765 			/* FALLTHROUGH */
3766 
3767 		default:
3768 			/*
3769 			 * Prevent DoS attack with too small MSS. Round up
3770 			 * to at least minmss.
3771 			 */
3772 			offer = max(offer, V_tcp_minmss);
3773 	}
3774 
3775 	/*
3776 	 * rmx information is now retrieved from tcp_hostcache.
3777 	 */
3778 	tcp_hc_get(&inp->inp_inc, &metrics);
3779 	if (metricptr != NULL)
3780 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3781 
3782 	/*
3783 	 * If there's a discovered mtu in tcp hostcache, use it.
3784 	 * Else, use the link mtu.
3785 	 */
3786 	if (metrics.rmx_mtu)
3787 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3788 	else {
3789 #ifdef INET6
3790 		if (isipv6) {
3791 			mss = maxmtu - min_protoh;
3792 			if (!V_path_mtu_discovery &&
3793 			    !in6_localaddr(&inp->in6p_faddr))
3794 				mss = min(mss, V_tcp_v6mssdflt);
3795 		}
3796 #endif
3797 #if defined(INET) && defined(INET6)
3798 		else
3799 #endif
3800 #ifdef INET
3801 		{
3802 			mss = maxmtu - min_protoh;
3803 			if (!V_path_mtu_discovery &&
3804 			    !in_localaddr(inp->inp_faddr))
3805 				mss = min(mss, V_tcp_mssdflt);
3806 		}
3807 #endif
3808 		/*
3809 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3810 		 * probably violates the TCP spec.
3811 		 * The problem is that, since we don't know the
3812 		 * other end's MSS, we are supposed to use a conservative
3813 		 * default.  But, if we do that, then MTU discovery will
3814 		 * never actually take place, because the conservative
3815 		 * default is much less than the MTUs typically seen
3816 		 * on the Internet today.  For the moment, we'll sweep
3817 		 * this under the carpet.
3818 		 *
3819 		 * The conservative default might not actually be a problem
3820 		 * if the only case this occurs is when sending an initial
3821 		 * SYN with options and data to a host we've never talked
3822 		 * to before.  Then, they will reply with an MSS value which
3823 		 * will get recorded and the new parameters should get
3824 		 * recomputed.  For Further Study.
3825 		 */
3826 	}
3827 	mss = min(mss, offer);
3828 
3829 	/*
3830 	 * Sanity check: make sure that maxseg will be large
3831 	 * enough to allow some data on segments even if the
3832 	 * all the option space is used (40bytes).  Otherwise
3833 	 * funny things may happen in tcp_output.
3834 	 *
3835 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3836 	 */
3837 	mss = max(mss, 64);
3838 
3839 	tp->t_maxseg = mss;
3840 }
3841 
3842 void
3843 tcp_mss(struct tcpcb *tp, int offer)
3844 {
3845 	int mss;
3846 	uint32_t bufsize;
3847 	struct inpcb *inp;
3848 	struct socket *so;
3849 	struct hc_metrics_lite metrics;
3850 	struct tcp_ifcap cap;
3851 
3852 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3853 
3854 	bzero(&cap, sizeof(cap));
3855 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3856 
3857 	mss = tp->t_maxseg;
3858 	inp = tp->t_inpcb;
3859 
3860 	/*
3861 	 * If there's a pipesize, change the socket buffer to that size,
3862 	 * don't change if sb_hiwat is different than default (then it
3863 	 * has been changed on purpose with setsockopt).
3864 	 * Make the socket buffers an integral number of mss units;
3865 	 * if the mss is larger than the socket buffer, decrease the mss.
3866 	 */
3867 	so = inp->inp_socket;
3868 	SOCKBUF_LOCK(&so->so_snd);
3869 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3870 		bufsize = metrics.rmx_sendpipe;
3871 	else
3872 		bufsize = so->so_snd.sb_hiwat;
3873 	if (bufsize < mss)
3874 		mss = bufsize;
3875 	else {
3876 		bufsize = roundup(bufsize, mss);
3877 		if (bufsize > sb_max)
3878 			bufsize = sb_max;
3879 		if (bufsize > so->so_snd.sb_hiwat)
3880 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3881 	}
3882 	SOCKBUF_UNLOCK(&so->so_snd);
3883 	/*
3884 	 * Sanity check: make sure that maxseg will be large
3885 	 * enough to allow some data on segments even if the
3886 	 * all the option space is used (40bytes).  Otherwise
3887 	 * funny things may happen in tcp_output.
3888 	 *
3889 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3890 	 */
3891 	tp->t_maxseg = max(mss, 64);
3892 
3893 	SOCKBUF_LOCK(&so->so_rcv);
3894 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3895 		bufsize = metrics.rmx_recvpipe;
3896 	else
3897 		bufsize = so->so_rcv.sb_hiwat;
3898 	if (bufsize > mss) {
3899 		bufsize = roundup(bufsize, mss);
3900 		if (bufsize > sb_max)
3901 			bufsize = sb_max;
3902 		if (bufsize > so->so_rcv.sb_hiwat)
3903 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3904 	}
3905 	SOCKBUF_UNLOCK(&so->so_rcv);
3906 
3907 	/* Check the interface for TSO capabilities. */
3908 	if (cap.ifcap & CSUM_TSO) {
3909 		tp->t_flags |= TF_TSO;
3910 		tp->t_tsomax = cap.tsomax;
3911 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3912 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3913 	}
3914 }
3915 
3916 /*
3917  * Determine the MSS option to send on an outgoing SYN.
3918  */
3919 int
3920 tcp_mssopt(struct in_conninfo *inc)
3921 {
3922 	int mss = 0;
3923 	uint32_t thcmtu = 0;
3924 	uint32_t maxmtu = 0;
3925 	size_t min_protoh;
3926 
3927 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3928 
3929 #ifdef INET6
3930 	if (inc->inc_flags & INC_ISIPV6) {
3931 		mss = V_tcp_v6mssdflt;
3932 		maxmtu = tcp_maxmtu6(inc, NULL);
3933 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3934 	}
3935 #endif
3936 #if defined(INET) && defined(INET6)
3937 	else
3938 #endif
3939 #ifdef INET
3940 	{
3941 		mss = V_tcp_mssdflt;
3942 		maxmtu = tcp_maxmtu(inc, NULL);
3943 		min_protoh = sizeof(struct tcpiphdr);
3944 	}
3945 #endif
3946 #if defined(INET6) || defined(INET)
3947 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3948 #endif
3949 
3950 	if (maxmtu && thcmtu)
3951 		mss = min(maxmtu, thcmtu) - min_protoh;
3952 	else if (maxmtu || thcmtu)
3953 		mss = max(maxmtu, thcmtu) - min_protoh;
3954 
3955 	return (mss);
3956 }
3957 
3958 void
3959 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3960 {
3961 	int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3962 	int maxseg = tcp_maxseg(tp);
3963 
3964 	INP_WLOCK_ASSERT(tp->t_inpcb);
3965 
3966 	/*
3967 	 * Compute the amount of data that this ACK is indicating
3968 	 * (del_data) and an estimate of how many bytes are in the
3969 	 * network.
3970 	 */
3971 	del_data = tp->sackhint.delivered_data;
3972 	if (V_tcp_do_newsack)
3973 		pipe = tcp_compute_pipe(tp);
3974 	else
3975 		pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3976 	tp->sackhint.prr_delivered += del_data;
3977 	/*
3978 	 * Proportional Rate Reduction
3979 	 */
3980 	if (pipe >= tp->snd_ssthresh) {
3981 		if (tp->sackhint.recover_fs == 0)
3982 			tp->sackhint.recover_fs =
3983 			    imax(1, tp->snd_nxt - tp->snd_una);
3984 		snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3985 			    tp->snd_ssthresh, tp->sackhint.recover_fs) -
3986 			    tp->sackhint.prr_out;
3987 	} else {
3988 		if (V_tcp_do_prr_conservative)
3989 			limit = tp->sackhint.prr_delivered -
3990 			    tp->sackhint.prr_out;
3991 		else
3992 			limit = imax(tp->sackhint.prr_delivered -
3993 				    tp->sackhint.prr_out, del_data) +
3994 				    maxseg;
3995 		snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3996 	}
3997 	snd_cnt = imax(snd_cnt, 0) / maxseg;
3998 	/*
3999 	 * Send snd_cnt new data into the network in response to this ack.
4000 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
4001 	 * accordingly.
4002 	 */
4003 	if (IN_FASTRECOVERY(tp->t_flags)) {
4004 		tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover +
4005 			tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
4006 	} else if (IN_CONGRECOVERY(tp->t_flags))
4007 		tp->snd_cwnd = imax(maxseg, pipe - del_data +
4008 				    (snd_cnt * maxseg));
4009 }
4010 
4011 /*
4012  * On a partial ack arrives, force the retransmission of the
4013  * next unacknowledged segment.  Do not clear tp->t_dupacks.
4014  * By setting snd_nxt to ti_ack, this forces retransmission timer to
4015  * be started again.
4016  */
4017 void
4018 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4019 {
4020 	tcp_seq onxt = tp->snd_nxt;
4021 	uint32_t ocwnd = tp->snd_cwnd;
4022 	u_int maxseg = tcp_maxseg(tp);
4023 
4024 	INP_WLOCK_ASSERT(tp->t_inpcb);
4025 
4026 	tcp_timer_activate(tp, TT_REXMT, 0);
4027 	tp->t_rtttime = 0;
4028 	tp->snd_nxt = th->th_ack;
4029 	/*
4030 	 * Set snd_cwnd to one segment beyond acknowledged offset.
4031 	 * (tp->snd_una has not yet been updated when this function is called.)
4032 	 */
4033 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4034 	tp->t_flags |= TF_ACKNOW;
4035 	(void) tp->t_fb->tfb_tcp_output(tp);
4036 	tp->snd_cwnd = ocwnd;
4037 	if (SEQ_GT(onxt, tp->snd_nxt))
4038 		tp->snd_nxt = onxt;
4039 	/*
4040 	 * Partial window deflation.  Relies on fact that tp->snd_una
4041 	 * not updated yet.
4042 	 */
4043 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4044 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4045 	else
4046 		tp->snd_cwnd = 0;
4047 	tp->snd_cwnd += maxseg;
4048 }
4049 
4050 int
4051 tcp_compute_pipe(struct tcpcb *tp)
4052 {
4053 	return (tp->snd_max - tp->snd_una +
4054 		tp->sackhint.sack_bytes_rexmit -
4055 		tp->sackhint.sacked_bytes);
4056 }
4057 
4058 uint32_t
4059 tcp_compute_initwnd(uint32_t maxseg)
4060 {
4061 	/*
4062 	 * Calculate the Initial Window, also used as Restart Window
4063 	 *
4064 	 * RFC5681 Section 3.1 specifies the default conservative values.
4065 	 * RFC3390 specifies slightly more aggressive values.
4066 	 * RFC6928 increases it to ten segments.
4067 	 * Support for user specified value for initial flight size.
4068 	 */
4069 	if (V_tcp_initcwnd_segments)
4070 		return min(V_tcp_initcwnd_segments * maxseg,
4071 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4072 	else if (V_tcp_do_rfc3390)
4073 		return min(4 * maxseg, max(2 * maxseg, 4380));
4074 	else {
4075 		/* Per RFC5681 Section 3.1 */
4076 		if (maxseg > 2190)
4077 			return (2 * maxseg);
4078 		else if (maxseg > 1095)
4079 			return (3 * maxseg);
4080 		else
4081 			return (4 * maxseg);
4082 	}
4083 }
4084