xref: /freebsd/sys/netinet/tcp_input.c (revision 0197ba465f275a32975d5c93b811e7caaec5b438)
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(bool, blackhole_local) = false;
148 #define	V_blackhole_local	VNET(blackhole_local)
149 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
150     CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
151     "Enforce net.inet.tcp.blackhole for locally originated packets");
152 
153 VNET_DEFINE(int, tcp_delack_enabled) = 1;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
155     &VNET_NAME(tcp_delack_enabled), 0,
156     "Delay ACK to try and piggyback it onto a data packet");
157 
158 VNET_DEFINE(int, drop_synfin) = 0;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
160     &VNET_NAME(drop_synfin), 0,
161     "Drop TCP packets with SYN+FIN set");
162 
163 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
165     &VNET_NAME(tcp_do_prr_conservative), 0,
166     "Do conservative Proportional Rate Reduction");
167 
168 VNET_DEFINE(int, tcp_do_prr) = 1;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
170     &VNET_NAME(tcp_do_prr), 1,
171     "Enable Proportional Rate Reduction per RFC 6937");
172 
173 VNET_DEFINE(int, tcp_do_lrd) = 0;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
175     &VNET_NAME(tcp_do_lrd), 1,
176     "Perform Lost Retransmission Detection");
177 
178 VNET_DEFINE(int, tcp_do_newcwv) = 0;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
180     &VNET_NAME(tcp_do_newcwv), 0,
181     "Enable New Congestion Window Validation per RFC7661");
182 
183 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
185     &VNET_NAME(tcp_do_rfc3042), 0,
186     "Enable RFC 3042 (Limited Transmit)");
187 
188 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
190     &VNET_NAME(tcp_do_rfc3390), 0,
191     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
192 
193 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
195     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
196     "Slow-start flight size (initial congestion window) in number of segments");
197 
198 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
200     &VNET_NAME(tcp_do_rfc3465), 0,
201     "Enable RFC 3465 (Appropriate Byte Counting)");
202 
203 VNET_DEFINE(int, tcp_abc_l_var) = 2;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
205     &VNET_NAME(tcp_abc_l_var), 2,
206     "Cap the max cwnd increment during slow-start to this number of segments");
207 
208 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
209     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
210     "TCP ECN");
211 
212 VNET_DEFINE(int, tcp_do_ecn) = 2;
213 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
214     &VNET_NAME(tcp_do_ecn), 0,
215     "TCP ECN support");
216 
217 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
218 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
219     &VNET_NAME(tcp_ecn_maxretries), 0,
220     "Max retries before giving up on ECN");
221 
222 VNET_DEFINE(int, tcp_insecure_syn) = 0;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
224     &VNET_NAME(tcp_insecure_syn), 0,
225     "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
226 
227 VNET_DEFINE(int, tcp_insecure_rst) = 0;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
229     &VNET_NAME(tcp_insecure_rst), 0,
230     "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
231 
232 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
233 #define	V_tcp_recvspace	VNET(tcp_recvspace)
234 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
235     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
236 
237 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
239     &VNET_NAME(tcp_do_autorcvbuf), 0,
240     "Enable automatic receive buffer sizing");
241 
242 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
243 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
244     &VNET_NAME(tcp_autorcvbuf_max), 0,
245     "Max size of automatic receive buffer");
246 
247 VNET_DEFINE(struct inpcbinfo, tcbinfo);
248 
249 /*
250  * TCP statistics are stored in an array of counter(9)s, which size matches
251  * size of struct tcpstat.  TCP running connection count is a regular array.
252  */
253 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
254 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
255     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
256 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
257 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
258     CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
259     "TCP connection counts by TCP state");
260 
261 /*
262  * Kernel module interface for updating tcpstat.  The first argument is an index
263  * into tcpstat treated as an array.
264  */
265 void
266 kmod_tcpstat_add(int statnum, int val)
267 {
268 
269 	counter_u64_add(VNET(tcpstat)[statnum], val);
270 }
271 
272 #ifdef TCP_HHOOK
273 /*
274  * Wrapper for the TCP established input helper hook.
275  */
276 void
277 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
278 {
279 	struct tcp_hhook_data hhook_data;
280 
281 	if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
282 		hhook_data.tp = tp;
283 		hhook_data.th = th;
284 		hhook_data.to = to;
285 
286 		hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
287 		    tp->osd);
288 	}
289 }
290 #endif
291 
292 /*
293  * CC wrapper hook functions
294  */
295 void
296 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
297     uint16_t type)
298 {
299 #ifdef STATS
300 	int32_t gput;
301 #endif
302 
303 	INP_WLOCK_ASSERT(tp->t_inpcb);
304 
305 	tp->ccv->nsegs = nsegs;
306 	tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
307 	if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
308 	    (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
309 	     (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
310 		tp->ccv->flags |= CCF_CWND_LIMITED;
311 	else
312 		tp->ccv->flags &= ~CCF_CWND_LIMITED;
313 
314 	if (type == CC_ACK) {
315 #ifdef STATS
316 		stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
317 		    ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
318 		if (!IN_RECOVERY(tp->t_flags))
319 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
320 			   tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
321 		if ((tp->t_flags & TF_GPUTINPROG) &&
322 		    SEQ_GEQ(th->th_ack, tp->gput_ack)) {
323 			/*
324 			 * Compute goodput in bits per millisecond.
325 			 */
326 			gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
327 			    max(1, tcp_ts_getticks() - tp->gput_ts);
328 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
329 			    gput);
330 			/*
331 			 * XXXLAS: This is a temporary hack, and should be
332 			 * chained off VOI_TCP_GPUT when stats(9) grows an API
333 			 * to deal with chained VOIs.
334 			 */
335 			if (tp->t_stats_gput_prev > 0)
336 				stats_voi_update_abs_s32(tp->t_stats,
337 				    VOI_TCP_GPUT_ND,
338 				    ((gput - tp->t_stats_gput_prev) * 100) /
339 				    tp->t_stats_gput_prev);
340 			tp->t_flags &= ~TF_GPUTINPROG;
341 			tp->t_stats_gput_prev = gput;
342 		}
343 #endif /* STATS */
344 		if (tp->snd_cwnd > tp->snd_ssthresh) {
345 			tp->t_bytes_acked += tp->ccv->bytes_this_ack;
346 			if (tp->t_bytes_acked >= tp->snd_cwnd) {
347 				tp->t_bytes_acked -= tp->snd_cwnd;
348 				tp->ccv->flags |= CCF_ABC_SENTAWND;
349 			}
350 		} else {
351 				tp->ccv->flags &= ~CCF_ABC_SENTAWND;
352 				tp->t_bytes_acked = 0;
353 		}
354 	}
355 
356 	if (CC_ALGO(tp)->ack_received != NULL) {
357 		/* XXXLAS: Find a way to live without this */
358 		tp->ccv->curack = th->th_ack;
359 		CC_ALGO(tp)->ack_received(tp->ccv, type);
360 	}
361 #ifdef STATS
362 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
363 #endif
364 }
365 
366 void
367 cc_conn_init(struct tcpcb *tp)
368 {
369 	struct hc_metrics_lite metrics;
370 	struct inpcb *inp = tp->t_inpcb;
371 	u_int maxseg;
372 	int rtt;
373 
374 	INP_WLOCK_ASSERT(tp->t_inpcb);
375 
376 	tcp_hc_get(&inp->inp_inc, &metrics);
377 	maxseg = tcp_maxseg(tp);
378 
379 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
380 		tp->t_srtt = rtt;
381 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
382 		TCPSTAT_INC(tcps_usedrtt);
383 		if (metrics.rmx_rttvar) {
384 			tp->t_rttvar = metrics.rmx_rttvar;
385 			TCPSTAT_INC(tcps_usedrttvar);
386 		} else {
387 			/* default variation is +- 1 rtt */
388 			tp->t_rttvar =
389 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
390 		}
391 		TCPT_RANGESET(tp->t_rxtcur,
392 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
393 		    tp->t_rttmin, TCPTV_REXMTMAX);
394 	}
395 	if (metrics.rmx_ssthresh) {
396 		/*
397 		 * There's some sort of gateway or interface
398 		 * buffer limit on the path.  Use this to set
399 		 * the slow start threshold, but set the
400 		 * threshold to no less than 2*mss.
401 		 */
402 		tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
403 		TCPSTAT_INC(tcps_usedssthresh);
404 	}
405 
406 	/*
407 	 * Set the initial slow-start flight size.
408 	 *
409 	 * If a SYN or SYN/ACK was lost and retransmitted, we have to
410 	 * reduce the initial CWND to one segment as congestion is likely
411 	 * requiring us to be cautious.
412 	 */
413 	if (tp->snd_cwnd == 1)
414 		tp->snd_cwnd = maxseg;		/* SYN(-ACK) lost */
415 	else
416 		tp->snd_cwnd = tcp_compute_initwnd(maxseg);
417 
418 	if (CC_ALGO(tp)->conn_init != NULL)
419 		CC_ALGO(tp)->conn_init(tp->ccv);
420 }
421 
422 void inline
423 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
424 {
425 	INP_WLOCK_ASSERT(tp->t_inpcb);
426 
427 #ifdef STATS
428 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
429 #endif
430 
431 	switch(type) {
432 	case CC_NDUPACK:
433 		if (!IN_FASTRECOVERY(tp->t_flags)) {
434 			tp->snd_recover = tp->snd_max;
435 			if (tp->t_flags2 & TF2_ECN_PERMIT)
436 				tp->t_flags2 |= TF2_ECN_SND_CWR;
437 		}
438 		break;
439 	case CC_ECN:
440 		if (!IN_CONGRECOVERY(tp->t_flags) ||
441 		    /*
442 		     * Allow ECN reaction on ACK to CWR, if
443 		     * that data segment was also CE marked.
444 		     */
445 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
446 			EXIT_CONGRECOVERY(tp->t_flags);
447 			TCPSTAT_INC(tcps_ecn_rcwnd);
448 			tp->snd_recover = tp->snd_max + 1;
449 			if (tp->t_flags2 & TF2_ECN_PERMIT)
450 				tp->t_flags2 |= TF2_ECN_SND_CWR;
451 		}
452 		break;
453 	case CC_RTO:
454 		tp->t_dupacks = 0;
455 		tp->t_bytes_acked = 0;
456 		EXIT_RECOVERY(tp->t_flags);
457 		if (tp->t_flags2 & TF2_ECN_PERMIT)
458 			tp->t_flags2 |= TF2_ECN_SND_CWR;
459 		break;
460 	case CC_RTO_ERR:
461 		TCPSTAT_INC(tcps_sndrexmitbad);
462 		/* RTO was unnecessary, so reset everything. */
463 		tp->snd_cwnd = tp->snd_cwnd_prev;
464 		tp->snd_ssthresh = tp->snd_ssthresh_prev;
465 		tp->snd_recover = tp->snd_recover_prev;
466 		if (tp->t_flags & TF_WASFRECOVERY)
467 			ENTER_FASTRECOVERY(tp->t_flags);
468 		if (tp->t_flags & TF_WASCRECOVERY)
469 			ENTER_CONGRECOVERY(tp->t_flags);
470 		tp->snd_nxt = tp->snd_max;
471 		tp->t_flags &= ~TF_PREVVALID;
472 		tp->t_badrxtwin = 0;
473 		break;
474 	}
475 
476 	if (CC_ALGO(tp)->cong_signal != NULL) {
477 		if (th != NULL)
478 			tp->ccv->curack = th->th_ack;
479 		CC_ALGO(tp)->cong_signal(tp->ccv, type);
480 	}
481 }
482 
483 void inline
484 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
485 {
486 	INP_WLOCK_ASSERT(tp->t_inpcb);
487 
488 	/* XXXLAS: KASSERT that we're in recovery? */
489 
490 	if (CC_ALGO(tp)->post_recovery != NULL) {
491 		tp->ccv->curack = th->th_ack;
492 		CC_ALGO(tp)->post_recovery(tp->ccv);
493 	}
494 	/* XXXLAS: EXIT_RECOVERY ? */
495 	tp->t_bytes_acked = 0;
496 	tp->sackhint.delivered_data = 0;
497 	tp->sackhint.prr_out = 0;
498 }
499 
500 /*
501  * Indicate whether this ack should be delayed.  We can delay the ack if
502  * following conditions are met:
503  *	- There is no delayed ack timer in progress.
504  *	- Our last ack wasn't a 0-sized window. We never want to delay
505  *	  the ack that opens up a 0-sized window.
506  *	- LRO wasn't used for this segment. We make sure by checking that the
507  *	  segment size is not larger than the MSS.
508  */
509 #define DELAY_ACK(tp, tlen)						\
510 	((!tcp_timer_active(tp, TT_DELACK) &&				\
511 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
512 	    (tlen <= tp->t_maxseg) &&					\
513 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
514 
515 void inline
516 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
517 {
518 	INP_WLOCK_ASSERT(tp->t_inpcb);
519 
520 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
521 		switch (iptos & IPTOS_ECN_MASK) {
522 		case IPTOS_ECN_CE:
523 			tp->ccv->flags |= CCF_IPHDR_CE;
524 			break;
525 		case IPTOS_ECN_ECT0:
526 			/* FALLTHROUGH */
527 		case IPTOS_ECN_ECT1:
528 			/* FALLTHROUGH */
529 		case IPTOS_ECN_NOTECT:
530 			tp->ccv->flags &= ~CCF_IPHDR_CE;
531 			break;
532 		}
533 
534 		if (flags & TH_CWR)
535 			tp->ccv->flags |= CCF_TCPHDR_CWR;
536 		else
537 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
538 
539 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
540 
541 		if (tp->ccv->flags & CCF_ACKNOW) {
542 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
543 			tp->t_flags |= TF_ACKNOW;
544 		}
545 	}
546 }
547 
548 void inline
549 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
550 {
551 	cc_ecnpkt_handler_flags(tp, th->th_flags, iptos);
552 }
553 
554 /*
555  * TCP input handling is split into multiple parts:
556  *   tcp6_input is a thin wrapper around tcp_input for the extended
557  *	ip6_protox[] call format in ip6_input
558  *   tcp_input handles primary segment validation, inpcb lookup and
559  *	SYN processing on listen sockets
560  *   tcp_do_segment processes the ACK and text of the segment for
561  *	establishing, established and closing connections
562  */
563 #ifdef INET6
564 int
565 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
566 {
567 	struct mbuf *m;
568 	struct in6_ifaddr *ia6;
569 	struct ip6_hdr *ip6;
570 
571 	m = *mp;
572 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
573 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
574 		if (m == NULL) {
575 			*mp = m;
576 			TCPSTAT_INC(tcps_rcvshort);
577 			return (IPPROTO_DONE);
578 		}
579 	}
580 
581 	/*
582 	 * draft-itojun-ipv6-tcp-to-anycast
583 	 * better place to put this in?
584 	 */
585 	ip6 = mtod(m, struct ip6_hdr *);
586 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
587 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
588 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
589 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
590 		*mp = NULL;
591 		return (IPPROTO_DONE);
592 	}
593 
594 	*mp = m;
595 	return (tcp_input_with_port(mp, offp, proto, port));
596 }
597 
598 int
599 tcp6_input(struct mbuf **mp, int *offp, int proto)
600 {
601 
602 	return(tcp6_input_with_port(mp, offp, proto, 0));
603 }
604 #endif /* INET6 */
605 
606 int
607 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
608 {
609 	struct mbuf *m = *mp;
610 	struct tcphdr *th = NULL;
611 	struct ip *ip = NULL;
612 	struct inpcb *inp = NULL;
613 	struct tcpcb *tp = NULL;
614 	struct socket *so = NULL;
615 	u_char *optp = NULL;
616 	int off0;
617 	int optlen = 0;
618 #ifdef INET
619 	int len;
620 	uint8_t ipttl;
621 #endif
622 	int tlen = 0, off;
623 	int drop_hdrlen;
624 	int thflags;
625 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
626 	int lookupflag;
627 	uint8_t iptos;
628 	struct m_tag *fwd_tag = NULL;
629 #ifdef INET6
630 	struct ip6_hdr *ip6 = NULL;
631 	int isipv6;
632 #else
633 	const void *ip6 = NULL;
634 #endif /* INET6 */
635 	struct tcpopt to;		/* options in this segment */
636 	char *s = NULL;			/* address and port logging */
637 #ifdef TCPDEBUG
638 	/*
639 	 * The size of tcp_saveipgen must be the size of the max ip header,
640 	 * now IPv6.
641 	 */
642 	u_char tcp_saveipgen[IP6_HDR_LEN];
643 	struct tcphdr tcp_savetcp;
644 	short ostate = 0;
645 #endif
646 
647 	NET_EPOCH_ASSERT();
648 
649 #ifdef INET6
650 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
651 #endif
652 
653 	off0 = *offp;
654 	m = *mp;
655 	*mp = NULL;
656 	to.to_flags = 0;
657 	TCPSTAT_INC(tcps_rcvtotal);
658 
659 #ifdef INET6
660 	if (isipv6) {
661 		ip6 = mtod(m, struct ip6_hdr *);
662 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
663 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
664 		if (port)
665 			goto skip6_csum;
666 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
667 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
668 				th->th_sum = m->m_pkthdr.csum_data;
669 			else
670 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
671 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
672 			th->th_sum ^= 0xffff;
673 		} else
674 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
675 		if (th->th_sum) {
676 			TCPSTAT_INC(tcps_rcvbadsum);
677 			goto drop;
678 		}
679 	skip6_csum:
680 		/*
681 		 * Be proactive about unspecified IPv6 address in source.
682 		 * As we use all-zero to indicate unbounded/unconnected pcb,
683 		 * unspecified IPv6 address can be used to confuse us.
684 		 *
685 		 * Note that packets with unspecified IPv6 destination is
686 		 * already dropped in ip6_input.
687 		 */
688 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
689 			/* XXX stat */
690 			goto drop;
691 		}
692 		iptos = IPV6_TRAFFIC_CLASS(ip6);
693 	}
694 #endif
695 #if defined(INET) && defined(INET6)
696 	else
697 #endif
698 #ifdef INET
699 	{
700 		/*
701 		 * Get IP and TCP header together in first mbuf.
702 		 * Note: IP leaves IP header in first mbuf.
703 		 */
704 		if (off0 > sizeof (struct ip)) {
705 			ip_stripoptions(m);
706 			off0 = sizeof(struct ip);
707 		}
708 		if (m->m_len < sizeof (struct tcpiphdr)) {
709 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
710 			    == NULL) {
711 				TCPSTAT_INC(tcps_rcvshort);
712 				return (IPPROTO_DONE);
713 			}
714 		}
715 		ip = mtod(m, struct ip *);
716 		th = (struct tcphdr *)((caddr_t)ip + off0);
717 		tlen = ntohs(ip->ip_len) - off0;
718 
719 		iptos = ip->ip_tos;
720 		if (port)
721 			goto skip_csum;
722 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
723 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
724 				th->th_sum = m->m_pkthdr.csum_data;
725 			else
726 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
727 				    ip->ip_dst.s_addr,
728 				    htonl(m->m_pkthdr.csum_data + tlen +
729 				    IPPROTO_TCP));
730 			th->th_sum ^= 0xffff;
731 		} else {
732 			struct ipovly *ipov = (struct ipovly *)ip;
733 
734 			/*
735 			 * Checksum extended TCP header and data.
736 			 */
737 			len = off0 + tlen;
738 			ipttl = ip->ip_ttl;
739 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
740 			ipov->ih_len = htons(tlen);
741 			th->th_sum = in_cksum(m, len);
742 			/* Reset length for SDT probes. */
743 			ip->ip_len = htons(len);
744 			/* Reset TOS bits */
745 			ip->ip_tos = iptos;
746 			/* Re-initialization for later version check */
747 			ip->ip_ttl = ipttl;
748 			ip->ip_v = IPVERSION;
749 			ip->ip_hl = off0 >> 2;
750 		}
751 	skip_csum:
752 		if (th->th_sum && (port == 0)) {
753 			TCPSTAT_INC(tcps_rcvbadsum);
754 			goto drop;
755 		}
756 	}
757 #endif /* INET */
758 
759 	/*
760 	 * Check that TCP offset makes sense,
761 	 * pull out TCP options and adjust length.		XXX
762 	 */
763 	off = th->th_off << 2;
764 	if (off < sizeof (struct tcphdr) || off > tlen) {
765 		TCPSTAT_INC(tcps_rcvbadoff);
766 		goto drop;
767 	}
768 	tlen -= off;	/* tlen is used instead of ti->ti_len */
769 	if (off > sizeof (struct tcphdr)) {
770 #ifdef INET6
771 		if (isipv6) {
772 			if (m->m_len < off0 + off) {
773 				m = m_pullup(m, off0 + off);
774 				if (m == NULL) {
775 					TCPSTAT_INC(tcps_rcvshort);
776 					return (IPPROTO_DONE);
777 				}
778 			}
779 			ip6 = mtod(m, struct ip6_hdr *);
780 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
781 		}
782 #endif
783 #if defined(INET) && defined(INET6)
784 		else
785 #endif
786 #ifdef INET
787 		{
788 			if (m->m_len < sizeof(struct ip) + off) {
789 				if ((m = m_pullup(m, sizeof (struct ip) + off))
790 				    == NULL) {
791 					TCPSTAT_INC(tcps_rcvshort);
792 					return (IPPROTO_DONE);
793 				}
794 				ip = mtod(m, struct ip *);
795 				th = (struct tcphdr *)((caddr_t)ip + off0);
796 			}
797 		}
798 #endif
799 		optlen = off - sizeof (struct tcphdr);
800 		optp = (u_char *)(th + 1);
801 	}
802 	thflags = th->th_flags;
803 
804 	/*
805 	 * Convert TCP protocol specific fields to host format.
806 	 */
807 	tcp_fields_to_host(th);
808 
809 	/*
810 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
811 	 */
812 	drop_hdrlen = off0 + off;
813 
814 	/*
815 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
816 	 */
817         if (
818 #ifdef INET6
819 	    (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
820 #ifdef INET
821 	    || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
822 #endif
823 #endif
824 #if defined(INET) && !defined(INET6)
825 	    (m->m_flags & M_IP_NEXTHOP)
826 #endif
827 	    )
828 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
829 
830 	/*
831 	 * For initial SYN packets we don't need write lock on matching
832 	 * PCB, be it a listening one or a synchronized one.  The packet
833 	 * shall not modify its state.
834 	 */
835 	lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
836 	    INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB;
837 findpcb:
838 #ifdef INET6
839 	if (isipv6 && fwd_tag != NULL) {
840 		struct sockaddr_in6 *next_hop6;
841 
842 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
843 		/*
844 		 * Transparently forwarded. Pretend to be the destination.
845 		 * Already got one like this?
846 		 */
847 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
848 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
849 		    lookupflag, m->m_pkthdr.rcvif, m);
850 		if (!inp) {
851 			/*
852 			 * It's new.  Try to find the ambushing socket.
853 			 * Because we've rewritten the destination address,
854 			 * any hardware-generated hash is ignored.
855 			 */
856 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
857 			    th->th_sport, &next_hop6->sin6_addr,
858 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
859 			    th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
860 			    m->m_pkthdr.rcvif);
861 		}
862 	} else if (isipv6) {
863 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
864 		    th->th_sport, &ip6->ip6_dst, th->th_dport,
865 		    INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
866 	}
867 #endif /* INET6 */
868 #if defined(INET6) && defined(INET)
869 	else
870 #endif
871 #ifdef INET
872 	if (fwd_tag != NULL) {
873 		struct sockaddr_in *next_hop;
874 
875 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
876 		/*
877 		 * Transparently forwarded. Pretend to be the destination.
878 		 * already got one like this?
879 		 */
880 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
881 		    ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
882 		if (!inp) {
883 			/*
884 			 * It's new.  Try to find the ambushing socket.
885 			 * Because we've rewritten the destination address,
886 			 * any hardware-generated hash is ignored.
887 			 */
888 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
889 			    th->th_sport, next_hop->sin_addr,
890 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
891 			    th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
892 			    m->m_pkthdr.rcvif);
893 		}
894 	} else
895 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
896 		    th->th_sport, ip->ip_dst, th->th_dport,
897 		    INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
898 #endif /* INET */
899 
900 	/*
901 	 * If the INPCB does not exist then all data in the incoming
902 	 * segment is discarded and an appropriate RST is sent back.
903 	 * XXX MRT Send RST using which routing table?
904 	 */
905 	if (inp == NULL) {
906 		/*
907 		 * Log communication attempts to ports that are not
908 		 * in use.
909 		 */
910 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
911 		    V_tcp_log_in_vain == 2) {
912 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
913 				log(LOG_INFO, "%s; %s: Connection attempt "
914 				    "to closed port\n", s, __func__);
915 		}
916 		/*
917 		 * When blackholing do not respond with a RST but
918 		 * completely ignore the segment and drop it.
919 		 */
920 		if (((V_blackhole == 1 && (thflags & TH_SYN)) ||
921 		    V_blackhole == 2) && (V_blackhole_local ||
922 #ifdef INET6
923 		    isipv6 ? !in6_localaddr(&ip6->ip6_src) :
924 #endif
925 #ifdef INET
926 		    !in_localip(ip->ip_src)
927 #else
928 		    true
929 #endif
930 		    ))
931 			goto dropunlock;
932 
933 		rstreason = BANDLIM_RST_CLOSEDPORT;
934 		goto dropwithreset;
935 	}
936 	INP_LOCK_ASSERT(inp);
937 	/*
938 	 * While waiting for inp lock during the lookup, another thread
939 	 * can have dropped the inpcb, in which case we need to loop back
940 	 * and try to find a new inpcb to deliver to.
941 	 */
942 	if (inp->inp_flags & INP_DROPPED) {
943 		INP_UNLOCK(inp);
944 		inp = NULL;
945 		goto findpcb;
946 	}
947 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
948 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
949 	    ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) {
950 		inp->inp_flowid = m->m_pkthdr.flowid;
951 		inp->inp_flowtype = M_HASHTYPE_GET(m);
952 	}
953 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
954 #ifdef INET6
955 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
956 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
957 		goto dropunlock;
958 	}
959 #ifdef INET
960 	else
961 #endif
962 #endif /* INET6 */
963 #ifdef INET
964 	if (IPSEC_ENABLED(ipv4) &&
965 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
966 		goto dropunlock;
967 	}
968 #endif /* INET */
969 #endif /* IPSEC */
970 
971 	/*
972 	 * Check the minimum TTL for socket.
973 	 */
974 	if (inp->inp_ip_minttl != 0) {
975 #ifdef INET6
976 		if (isipv6) {
977 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
978 				goto dropunlock;
979 		} else
980 #endif
981 		if (inp->inp_ip_minttl > ip->ip_ttl)
982 			goto dropunlock;
983 	}
984 
985 	/*
986 	 * A previous connection in TIMEWAIT state is supposed to catch stray
987 	 * or duplicate segments arriving late.  If this segment was a
988 	 * legitimate new connection attempt, the old INPCB gets removed and
989 	 * we can try again to find a listening socket.
990 	 */
991 	if (inp->inp_flags & INP_TIMEWAIT) {
992 		tcp_dooptions(&to, optp, optlen,
993 		    (thflags & TH_SYN) ? TO_SYN : 0);
994 		/*
995 		 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
996 		 */
997 		if (tcp_twcheck(inp, &to, th, m, tlen))
998 			goto findpcb;
999 		return (IPPROTO_DONE);
1000 	}
1001 	/*
1002 	 * The TCPCB may no longer exist if the connection is winding
1003 	 * down or it is in the CLOSED state.  Either way we drop the
1004 	 * segment and send an appropriate response.
1005 	 */
1006 	tp = intotcpcb(inp);
1007 	if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1008 		rstreason = BANDLIM_RST_CLOSEDPORT;
1009 		goto dropwithreset;
1010 	}
1011 
1012 	if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1013 		rstreason = BANDLIM_RST_CLOSEDPORT;
1014 		goto dropwithreset;
1015 	}
1016 
1017 #ifdef TCP_OFFLOAD
1018 	if (tp->t_flags & TF_TOE) {
1019 		tcp_offload_input(tp, m);
1020 		m = NULL;	/* consumed by the TOE driver */
1021 		goto dropunlock;
1022 	}
1023 #endif
1024 
1025 #ifdef MAC
1026 	if (mac_inpcb_check_deliver(inp, m))
1027 		goto dropunlock;
1028 #endif
1029 	so = inp->inp_socket;
1030 	KASSERT(so != NULL, ("%s: so == NULL", __func__));
1031 #ifdef TCPDEBUG
1032 	if (so->so_options & SO_DEBUG) {
1033 		ostate = tp->t_state;
1034 #ifdef INET6
1035 		if (isipv6) {
1036 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1037 		} else
1038 #endif
1039 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1040 		tcp_savetcp = *th;
1041 	}
1042 #endif /* TCPDEBUG */
1043 	/*
1044 	 * When the socket is accepting connections (the INPCB is in LISTEN
1045 	 * state) we look into the SYN cache if this is a new connection
1046 	 * attempt or the completion of a previous one.
1047 	 */
1048 	KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1049 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1050 	if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1051 		struct in_conninfo inc;
1052 
1053 		bzero(&inc, sizeof(inc));
1054 #ifdef INET6
1055 		if (isipv6) {
1056 			inc.inc_flags |= INC_ISIPV6;
1057 			if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1058 				inc.inc_flags |= INC_IPV6MINMTU;
1059 			inc.inc6_faddr = ip6->ip6_src;
1060 			inc.inc6_laddr = ip6->ip6_dst;
1061 		} else
1062 #endif
1063 		{
1064 			inc.inc_faddr = ip->ip_src;
1065 			inc.inc_laddr = ip->ip_dst;
1066 		}
1067 		inc.inc_fport = th->th_sport;
1068 		inc.inc_lport = th->th_dport;
1069 		inc.inc_fibnum = so->so_fibnum;
1070 
1071 		/*
1072 		 * Check for an existing connection attempt in syncache if
1073 		 * the flag is only ACK.  A successful lookup creates a new
1074 		 * socket appended to the listen queue in SYN_RECEIVED state.
1075 		 */
1076 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1077 			/*
1078 			 * Parse the TCP options here because
1079 			 * syncookies need access to the reflected
1080 			 * timestamp.
1081 			 */
1082 			tcp_dooptions(&to, optp, optlen, 0);
1083 			/*
1084 			 * NB: syncache_expand() doesn't unlock
1085 			 * inp and tcpinfo locks.
1086 			 */
1087 			rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1088 			if (rstreason < 0) {
1089 				/*
1090 				 * A failing TCP MD5 signature comparison
1091 				 * must result in the segment being dropped
1092 				 * and must not produce any response back
1093 				 * to the sender.
1094 				 */
1095 				goto dropunlock;
1096 			} else if (rstreason == 0) {
1097 				/*
1098 				 * No syncache entry or ACK was not
1099 				 * for our SYN/ACK.  Send a RST.
1100 				 * NB: syncache did its own logging
1101 				 * of the failure cause.
1102 				 */
1103 				rstreason = BANDLIM_RST_OPENPORT;
1104 				goto dropwithreset;
1105 			}
1106 tfo_socket_result:
1107 			if (so == NULL) {
1108 				/*
1109 				 * We completed the 3-way handshake
1110 				 * but could not allocate a socket
1111 				 * either due to memory shortage,
1112 				 * listen queue length limits or
1113 				 * global socket limits.  Send RST
1114 				 * or wait and have the remote end
1115 				 * retransmit the ACK for another
1116 				 * try.
1117 				 */
1118 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1119 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1120 					    "Socket allocation failed due to "
1121 					    "limits or memory shortage, %s\n",
1122 					    s, __func__,
1123 					    V_tcp_sc_rst_sock_fail ?
1124 					    "sending RST" : "try again");
1125 				if (V_tcp_sc_rst_sock_fail) {
1126 					rstreason = BANDLIM_UNLIMITED;
1127 					goto dropwithreset;
1128 				} else
1129 					goto dropunlock;
1130 			}
1131 			/*
1132 			 * Socket is created in state SYN_RECEIVED.
1133 			 * Unlock the listen socket, lock the newly
1134 			 * created socket and update the tp variable.
1135 			 * If we came here via jump to tfo_socket_result,
1136 			 * then listening socket is read-locked.
1137 			 */
1138 			INP_UNLOCK(inp);	/* listen socket */
1139 			inp = sotoinpcb(so);
1140 			/*
1141 			 * New connection inpcb is already locked by
1142 			 * syncache_expand().
1143 			 */
1144 			INP_WLOCK_ASSERT(inp);
1145 			tp = intotcpcb(inp);
1146 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1147 			    ("%s: ", __func__));
1148 			/*
1149 			 * Process the segment and the data it
1150 			 * contains.  tcp_do_segment() consumes
1151 			 * the mbuf chain and unlocks the inpcb.
1152 			 */
1153 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1154 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1155 			    iptos);
1156 			return (IPPROTO_DONE);
1157 		}
1158 		/*
1159 		 * Segment flag validation for new connection attempts:
1160 		 *
1161 		 * Our (SYN|ACK) response was rejected.
1162 		 * Check with syncache and remove entry to prevent
1163 		 * retransmits.
1164 		 *
1165 		 * NB: syncache_chkrst does its own logging of failure
1166 		 * causes.
1167 		 */
1168 		if (thflags & TH_RST) {
1169 			syncache_chkrst(&inc, th, m, port);
1170 			goto dropunlock;
1171 		}
1172 		/*
1173 		 * We can't do anything without SYN.
1174 		 */
1175 		if ((thflags & TH_SYN) == 0) {
1176 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 				    "SYN is missing, segment ignored\n",
1179 				    s, __func__);
1180 			TCPSTAT_INC(tcps_badsyn);
1181 			goto dropunlock;
1182 		}
1183 		/*
1184 		 * (SYN|ACK) is bogus on a listen socket.
1185 		 */
1186 		if (thflags & TH_ACK) {
1187 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1188 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1189 				    "SYN|ACK invalid, segment rejected\n",
1190 				    s, __func__);
1191 			syncache_badack(&inc, port);	/* XXX: Not needed! */
1192 			TCPSTAT_INC(tcps_badsyn);
1193 			rstreason = BANDLIM_RST_OPENPORT;
1194 			goto dropwithreset;
1195 		}
1196 		/*
1197 		 * If the drop_synfin option is enabled, drop all
1198 		 * segments with both the SYN and FIN bits set.
1199 		 * This prevents e.g. nmap from identifying the
1200 		 * TCP/IP stack.
1201 		 * XXX: Poor reasoning.  nmap has other methods
1202 		 * and is constantly refining its stack detection
1203 		 * strategies.
1204 		 * XXX: This is a violation of the TCP specification
1205 		 * and was used by RFC1644.
1206 		 */
1207 		if ((thflags & TH_FIN) && V_drop_synfin) {
1208 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1209 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1210 				    "SYN|FIN segment ignored (based on "
1211 				    "sysctl setting)\n", s, __func__);
1212 			TCPSTAT_INC(tcps_badsyn);
1213 			goto dropunlock;
1214 		}
1215 		/*
1216 		 * Segment's flags are (SYN) or (SYN|FIN).
1217 		 *
1218 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1219 		 * as they do not affect the state of the TCP FSM.
1220 		 * The data pointed to by TH_URG and th_urp is ignored.
1221 		 */
1222 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1223 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1224 		KASSERT(thflags & (TH_SYN),
1225 		    ("%s: Listen socket: TH_SYN not set", __func__));
1226 		INP_RLOCK_ASSERT(inp);
1227 #ifdef INET6
1228 		/*
1229 		 * If deprecated address is forbidden,
1230 		 * we do not accept SYN to deprecated interface
1231 		 * address to prevent any new inbound connection from
1232 		 * getting established.
1233 		 * When we do not accept SYN, we send a TCP RST,
1234 		 * with deprecated source address (instead of dropping
1235 		 * it).  We compromise it as it is much better for peer
1236 		 * to send a RST, and RST will be the final packet
1237 		 * for the exchange.
1238 		 *
1239 		 * If we do not forbid deprecated addresses, we accept
1240 		 * the SYN packet.  RFC2462 does not suggest dropping
1241 		 * SYN in this case.
1242 		 * If we decipher RFC2462 5.5.4, it says like this:
1243 		 * 1. use of deprecated addr with existing
1244 		 *    communication is okay - "SHOULD continue to be
1245 		 *    used"
1246 		 * 2. use of it with new communication:
1247 		 *   (2a) "SHOULD NOT be used if alternate address
1248 		 *        with sufficient scope is available"
1249 		 *   (2b) nothing mentioned otherwise.
1250 		 * Here we fall into (2b) case as we have no choice in
1251 		 * our source address selection - we must obey the peer.
1252 		 *
1253 		 * The wording in RFC2462 is confusing, and there are
1254 		 * multiple description text for deprecated address
1255 		 * handling - worse, they are not exactly the same.
1256 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1257 		 */
1258 		if (isipv6 && !V_ip6_use_deprecated) {
1259 			struct in6_ifaddr *ia6;
1260 
1261 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1262 			if (ia6 != NULL &&
1263 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1264 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1265 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1266 					"Connection attempt to deprecated "
1267 					"IPv6 address rejected\n",
1268 					s, __func__);
1269 				rstreason = BANDLIM_RST_OPENPORT;
1270 				goto dropwithreset;
1271 			}
1272 		}
1273 #endif /* INET6 */
1274 		/*
1275 		 * Basic sanity checks on incoming SYN requests:
1276 		 *   Don't respond if the destination is a link layer
1277 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1278 		 *   If it is from this socket it must be forged.
1279 		 *   Don't respond if the source or destination is a
1280 		 *	global or subnet broad- or multicast address.
1281 		 *   Note that it is quite possible to receive unicast
1282 		 *	link-layer packets with a broadcast IP address. Use
1283 		 *	in_broadcast() to find them.
1284 		 */
1285 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1286 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1287 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1288 				"Connection attempt from broad- or multicast "
1289 				"link layer address ignored\n", s, __func__);
1290 			goto dropunlock;
1291 		}
1292 #ifdef INET6
1293 		if (isipv6) {
1294 			if (th->th_dport == th->th_sport &&
1295 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1296 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1297 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1298 					"Connection attempt to/from self "
1299 					"ignored\n", s, __func__);
1300 				goto dropunlock;
1301 			}
1302 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1303 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1304 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1305 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1306 					"Connection attempt from/to multicast "
1307 					"address ignored\n", s, __func__);
1308 				goto dropunlock;
1309 			}
1310 		}
1311 #endif
1312 #if defined(INET) && defined(INET6)
1313 		else
1314 #endif
1315 #ifdef INET
1316 		{
1317 			if (th->th_dport == th->th_sport &&
1318 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1319 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 					"Connection attempt from/to self "
1322 					"ignored\n", s, __func__);
1323 				goto dropunlock;
1324 			}
1325 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1326 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1327 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1328 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1329 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1330 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1331 					"Connection attempt from/to broad- "
1332 					"or multicast address ignored\n",
1333 					s, __func__);
1334 				goto dropunlock;
1335 			}
1336 		}
1337 #endif
1338 		/*
1339 		 * SYN appears to be valid.  Create compressed TCP state
1340 		 * for syncache.
1341 		 */
1342 #ifdef TCPDEBUG
1343 		if (so->so_options & SO_DEBUG)
1344 			tcp_trace(TA_INPUT, ostate, tp,
1345 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1346 #endif
1347 		TCP_PROBE3(debug__input, tp, th, m);
1348 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1349 		if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1350 		    iptos, port)) != NULL)
1351 			goto tfo_socket_result;
1352 
1353 		/*
1354 		 * Entry added to syncache and mbuf consumed.
1355 		 * Only the listen socket is unlocked by syncache_add().
1356 		 */
1357 		return (IPPROTO_DONE);
1358 	} else if (tp->t_state == TCPS_LISTEN) {
1359 		/*
1360 		 * When a listen socket is torn down the SO_ACCEPTCONN
1361 		 * flag is removed first while connections are drained
1362 		 * from the accept queue in a unlock/lock cycle of the
1363 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1364 		 * attempt go through unhandled.
1365 		 */
1366 		goto dropunlock;
1367 	}
1368 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1369 	if (tp->t_flags & TF_SIGNATURE) {
1370 		tcp_dooptions(&to, optp, optlen, thflags);
1371 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1372 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1373 			goto dropunlock;
1374 		}
1375 		if (!TCPMD5_ENABLED() ||
1376 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1377 			goto dropunlock;
1378 	}
1379 #endif
1380 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1381 
1382 	/*
1383 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1384 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1385 	 * the inpcb, and unlocks pcbinfo.
1386 	 *
1387 	 * XXXGL: in case of a pure SYN arriving on existing connection
1388 	 * TCP stacks won't need to modify the PCB, they would either drop
1389 	 * the segment silently, or send a challenge ACK.  However, we try
1390 	 * to upgrade the lock, because calling convention for stacks is
1391 	 * write-lock on PCB.  If upgrade fails, drop the SYN.
1392 	 */
1393 	if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0)
1394 		goto dropunlock;
1395 
1396 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1397 	return (IPPROTO_DONE);
1398 
1399 dropwithreset:
1400 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1401 
1402 	if (inp != NULL) {
1403 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1404 		INP_UNLOCK(inp);
1405 	} else
1406 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1407 	m = NULL;	/* mbuf chain got consumed. */
1408 	goto drop;
1409 
1410 dropunlock:
1411 	if (m != NULL)
1412 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1413 
1414 	if (inp != NULL)
1415 		INP_UNLOCK(inp);
1416 
1417 drop:
1418 	if (s != NULL)
1419 		free(s, M_TCPLOG);
1420 	if (m != NULL)
1421 		m_freem(m);
1422 	return (IPPROTO_DONE);
1423 }
1424 
1425 /*
1426  * Automatic sizing of receive socket buffer.  Often the send
1427  * buffer size is not optimally adjusted to the actual network
1428  * conditions at hand (delay bandwidth product).  Setting the
1429  * buffer size too small limits throughput on links with high
1430  * bandwidth and high delay (eg. trans-continental/oceanic links).
1431  *
1432  * On the receive side the socket buffer memory is only rarely
1433  * used to any significant extent.  This allows us to be much
1434  * more aggressive in scaling the receive socket buffer.  For
1435  * the case that the buffer space is actually used to a large
1436  * extent and we run out of kernel memory we can simply drop
1437  * the new segments; TCP on the sender will just retransmit it
1438  * later.  Setting the buffer size too big may only consume too
1439  * much kernel memory if the application doesn't read() from
1440  * the socket or packet loss or reordering makes use of the
1441  * reassembly queue.
1442  *
1443  * The criteria to step up the receive buffer one notch are:
1444  *  1. Application has not set receive buffer size with
1445  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1446  *  2. the number of bytes received during 1/2 of an sRTT
1447  *     is at least 3/8 of the current socket buffer size.
1448  *  3. receive buffer size has not hit maximal automatic size;
1449  *
1450  * If all of the criteria are met we increaset the socket buffer
1451  * by a 1/2 (bounded by the max). This allows us to keep ahead
1452  * of slow-start but also makes it so our peer never gets limited
1453  * by our rwnd which we then open up causing a burst.
1454  *
1455  * This algorithm does two steps per RTT at most and only if
1456  * we receive a bulk stream w/o packet losses or reorderings.
1457  * Shrinking the buffer during idle times is not necessary as
1458  * it doesn't consume any memory when idle.
1459  *
1460  * TODO: Only step up if the application is actually serving
1461  * the buffer to better manage the socket buffer resources.
1462  */
1463 int
1464 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1465     struct tcpcb *tp, int tlen)
1466 {
1467 	int newsize = 0;
1468 
1469 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1470 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1471 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1472 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1473 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1474 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1475 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1476 		}
1477 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1478 
1479 		/* Start over with next RTT. */
1480 		tp->rfbuf_ts = 0;
1481 		tp->rfbuf_cnt = 0;
1482 	} else {
1483 		tp->rfbuf_cnt += tlen;	/* add up */
1484 	}
1485 	return (newsize);
1486 }
1487 
1488 int
1489 tcp_input(struct mbuf **mp, int *offp, int proto)
1490 {
1491 	return(tcp_input_with_port(mp, offp, proto, 0));
1492 }
1493 
1494 void
1495 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1496 {
1497 	/*
1498 	 * Since tp might be gone if the session entered
1499 	 * the TIME_WAIT state before coming here, we need
1500 	 * to check if the socket is still connected.
1501 	 */
1502 	if (tp == NULL) {
1503 		return;
1504 	}
1505 	if (so == NULL) {
1506 		return;
1507 	}
1508 	INP_LOCK_ASSERT(tp->t_inpcb);
1509 	if (tp->t_flags & TF_WAKESOR) {
1510 		tp->t_flags &= ~TF_WAKESOR;
1511 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1512 		sorwakeup_locked(so);
1513 	}
1514 }
1515 
1516 void
1517 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1518     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1519 {
1520 	int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1521 	int rstreason, todrop, win, incforsyn = 0;
1522 	uint32_t tiwin;
1523 	uint16_t nsegs;
1524 	char *s;
1525 	struct in_conninfo *inc;
1526 	struct mbuf *mfree;
1527 	struct tcpopt to;
1528 	int tfo_syn;
1529 	u_int maxseg;
1530 
1531 #ifdef TCPDEBUG
1532 	/*
1533 	 * The size of tcp_saveipgen must be the size of the max ip header,
1534 	 * now IPv6.
1535 	 */
1536 	u_char tcp_saveipgen[IP6_HDR_LEN];
1537 	struct tcphdr tcp_savetcp;
1538 	short ostate = 0;
1539 #endif
1540 	thflags = th->th_flags;
1541 	inc = &tp->t_inpcb->inp_inc;
1542 	tp->sackhint.last_sack_ack = 0;
1543 	sack_changed = 0;
1544 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1545 
1546 	NET_EPOCH_ASSERT();
1547 	INP_WLOCK_ASSERT(tp->t_inpcb);
1548 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1549 	    __func__));
1550 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1551 	    __func__));
1552 
1553 #ifdef TCPPCAP
1554 	/* Save segment, if requested. */
1555 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1556 #endif
1557 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1558 	    tlen, NULL, true);
1559 
1560 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1561 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1562 			log(LOG_DEBUG, "%s; %s: "
1563 			    "SYN|FIN segment ignored (based on "
1564 			    "sysctl setting)\n", s, __func__);
1565 			free(s, M_TCPLOG);
1566 		}
1567 		goto drop;
1568 	}
1569 
1570 	/*
1571 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1572 	 * check SEQ.ACK first.
1573 	 */
1574 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1575 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1576 		rstreason = BANDLIM_UNLIMITED;
1577 		goto dropwithreset;
1578 	}
1579 
1580 	/*
1581 	 * Segment received on connection.
1582 	 * Reset idle time and keep-alive timer.
1583 	 * XXX: This should be done after segment
1584 	 * validation to ignore broken/spoofed segs.
1585 	 */
1586 	tp->t_rcvtime = ticks;
1587 
1588 	/*
1589 	 * Scale up the window into a 32-bit value.
1590 	 * For the SYN_SENT state the scale is zero.
1591 	 */
1592 	tiwin = th->th_win << tp->snd_scale;
1593 #ifdef STATS
1594 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1595 #endif
1596 
1597 	/*
1598 	 * TCP ECN processing.
1599 	 */
1600 	if (tp->t_flags2 & TF2_ECN_PERMIT) {
1601 		if (thflags & TH_CWR) {
1602 			tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1603 			tp->t_flags |= TF_ACKNOW;
1604 		}
1605 		switch (iptos & IPTOS_ECN_MASK) {
1606 		case IPTOS_ECN_CE:
1607 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1608 			TCPSTAT_INC(tcps_ecn_ce);
1609 			break;
1610 		case IPTOS_ECN_ECT0:
1611 			TCPSTAT_INC(tcps_ecn_ect0);
1612 			break;
1613 		case IPTOS_ECN_ECT1:
1614 			TCPSTAT_INC(tcps_ecn_ect1);
1615 			break;
1616 		}
1617 
1618 		/* Process a packet differently from RFC3168. */
1619 		cc_ecnpkt_handler(tp, th, iptos);
1620 
1621 		/* Congestion experienced. */
1622 		if (thflags & TH_ECE) {
1623 			cc_cong_signal(tp, th, CC_ECN);
1624 		}
1625 	}
1626 
1627 	/*
1628 	 * Parse options on any incoming segment.
1629 	 */
1630 	tcp_dooptions(&to, (u_char *)(th + 1),
1631 	    (th->th_off << 2) - sizeof(struct tcphdr),
1632 	    (thflags & TH_SYN) ? TO_SYN : 0);
1633 
1634 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1635 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1636 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1637 		TCPSTAT_INC(tcps_sig_err_sigopt);
1638 		/* XXX: should drop? */
1639 	}
1640 #endif
1641 	/*
1642 	 * If echoed timestamp is later than the current time,
1643 	 * fall back to non RFC1323 RTT calculation.  Normalize
1644 	 * timestamp if syncookies were used when this connection
1645 	 * was established.
1646 	 */
1647 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1648 		to.to_tsecr -= tp->ts_offset;
1649 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1650 			to.to_tsecr = 0;
1651 		else if (tp->t_flags & TF_PREVVALID &&
1652 			 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1653 			cc_cong_signal(tp, th, CC_RTO_ERR);
1654 	}
1655 	/*
1656 	 * Process options only when we get SYN/ACK back. The SYN case
1657 	 * for incoming connections is handled in tcp_syncache.
1658 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1659 	 * or <SYN,ACK>) segment itself is never scaled.
1660 	 * XXX this is traditional behavior, may need to be cleaned up.
1661 	 */
1662 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1663 		/* Handle parallel SYN for ECN */
1664 		if (!(thflags & TH_ACK) &&
1665 		    ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1666 		    ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1667 			tp->t_flags2 |= TF2_ECN_PERMIT;
1668 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1669 			TCPSTAT_INC(tcps_ecn_shs);
1670 		}
1671 		if ((to.to_flags & TOF_SCALE) &&
1672 		    (tp->t_flags & TF_REQ_SCALE) &&
1673 		    !(tp->t_flags & TF_NOOPT)) {
1674 			tp->t_flags |= TF_RCVD_SCALE;
1675 			tp->snd_scale = to.to_wscale;
1676 		} else
1677 			tp->t_flags &= ~TF_REQ_SCALE;
1678 		/*
1679 		 * Initial send window.  It will be updated with
1680 		 * the next incoming segment to the scaled value.
1681 		 */
1682 		tp->snd_wnd = th->th_win;
1683 		if ((to.to_flags & TOF_TS) &&
1684 		    (tp->t_flags & TF_REQ_TSTMP) &&
1685 		    !(tp->t_flags & TF_NOOPT)) {
1686 			tp->t_flags |= TF_RCVD_TSTMP;
1687 			tp->ts_recent = to.to_tsval;
1688 			tp->ts_recent_age = tcp_ts_getticks();
1689 		} else
1690 			tp->t_flags &= ~TF_REQ_TSTMP;
1691 		if (to.to_flags & TOF_MSS)
1692 			tcp_mss(tp, to.to_mss);
1693 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1694 		    (!(to.to_flags & TOF_SACKPERM) ||
1695 		    (tp->t_flags & TF_NOOPT)))
1696 			tp->t_flags &= ~TF_SACK_PERMIT;
1697 		if (IS_FASTOPEN(tp->t_flags)) {
1698 			if ((to.to_flags & TOF_FASTOPEN) &&
1699 			    !(tp->t_flags & TF_NOOPT)) {
1700 				uint16_t mss;
1701 
1702 				if (to.to_flags & TOF_MSS)
1703 					mss = to.to_mss;
1704 				else
1705 					if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1706 						mss = TCP6_MSS;
1707 					else
1708 						mss = TCP_MSS;
1709 				tcp_fastopen_update_cache(tp, mss,
1710 				    to.to_tfo_len, to.to_tfo_cookie);
1711 			} else
1712 				tcp_fastopen_disable_path(tp);
1713 		}
1714 	}
1715 
1716 	/*
1717 	 * If timestamps were negotiated during SYN/ACK and a
1718 	 * segment without a timestamp is received, silently drop
1719 	 * the segment, unless it is a RST segment or missing timestamps are
1720 	 * tolerated.
1721 	 * See section 3.2 of RFC 7323.
1722 	 */
1723 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1724 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1725 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1726 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1727 				    "segment processed normally\n",
1728 				    s, __func__);
1729 				free(s, M_TCPLOG);
1730 			}
1731 		} else {
1732 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1733 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1734 				    "segment silently dropped\n", s, __func__);
1735 				free(s, M_TCPLOG);
1736 			}
1737 			goto drop;
1738 		}
1739 	}
1740 	/*
1741 	 * If timestamps were not negotiated during SYN/ACK and a
1742 	 * segment with a timestamp is received, ignore the
1743 	 * timestamp and process the packet normally.
1744 	 * See section 3.2 of RFC 7323.
1745 	 */
1746 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1747 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1748 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1749 			    "segment processed normally\n", s, __func__);
1750 			free(s, M_TCPLOG);
1751 		}
1752 	}
1753 
1754 	/*
1755 	 * Header prediction: check for the two common cases
1756 	 * of a uni-directional data xfer.  If the packet has
1757 	 * no control flags, is in-sequence, the window didn't
1758 	 * change and we're not retransmitting, it's a
1759 	 * candidate.  If the length is zero and the ack moved
1760 	 * forward, we're the sender side of the xfer.  Just
1761 	 * free the data acked & wake any higher level process
1762 	 * that was blocked waiting for space.  If the length
1763 	 * is non-zero and the ack didn't move, we're the
1764 	 * receiver side.  If we're getting packets in-order
1765 	 * (the reassembly queue is empty), add the data to
1766 	 * the socket buffer and note that we need a delayed ack.
1767 	 * Make sure that the hidden state-flags are also off.
1768 	 * Since we check for TCPS_ESTABLISHED first, it can only
1769 	 * be TH_NEEDSYN.
1770 	 */
1771 	if (tp->t_state == TCPS_ESTABLISHED &&
1772 	    th->th_seq == tp->rcv_nxt &&
1773 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1774 	    tp->snd_nxt == tp->snd_max &&
1775 	    tiwin && tiwin == tp->snd_wnd &&
1776 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1777 	    SEGQ_EMPTY(tp) &&
1778 	    ((to.to_flags & TOF_TS) == 0 ||
1779 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1780 		/*
1781 		 * If last ACK falls within this segment's sequence numbers,
1782 		 * record the timestamp.
1783 		 * NOTE that the test is modified according to the latest
1784 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1785 		 */
1786 		if ((to.to_flags & TOF_TS) != 0 &&
1787 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1788 			tp->ts_recent_age = tcp_ts_getticks();
1789 			tp->ts_recent = to.to_tsval;
1790 		}
1791 
1792 		if (tlen == 0) {
1793 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1794 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1795 			    !IN_RECOVERY(tp->t_flags) &&
1796 			    (to.to_flags & TOF_SACK) == 0 &&
1797 			    TAILQ_EMPTY(&tp->snd_holes)) {
1798 				/*
1799 				 * This is a pure ack for outstanding data.
1800 				 */
1801 				TCPSTAT_INC(tcps_predack);
1802 
1803 				/*
1804 				 * "bad retransmit" recovery without timestamps.
1805 				 */
1806 				if ((to.to_flags & TOF_TS) == 0 &&
1807 				    tp->t_rxtshift == 1 &&
1808 				    tp->t_flags & TF_PREVVALID &&
1809 				    (int)(ticks - tp->t_badrxtwin) < 0) {
1810 					cc_cong_signal(tp, th, CC_RTO_ERR);
1811 				}
1812 
1813 				/*
1814 				 * Recalculate the transmit timer / rtt.
1815 				 *
1816 				 * Some boxes send broken timestamp replies
1817 				 * during the SYN+ACK phase, ignore
1818 				 * timestamps of 0 or we could calculate a
1819 				 * huge RTT and blow up the retransmit timer.
1820 				 */
1821 				if ((to.to_flags & TOF_TS) != 0 &&
1822 				    to.to_tsecr) {
1823 					uint32_t t;
1824 
1825 					t = tcp_ts_getticks() - to.to_tsecr;
1826 					if (!tp->t_rttlow || tp->t_rttlow > t)
1827 						tp->t_rttlow = t;
1828 					tcp_xmit_timer(tp,
1829 					    TCP_TS_TO_TICKS(t) + 1);
1830 				} else if (tp->t_rtttime &&
1831 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1832 					if (!tp->t_rttlow ||
1833 					    tp->t_rttlow > ticks - tp->t_rtttime)
1834 						tp->t_rttlow = ticks - tp->t_rtttime;
1835 					tcp_xmit_timer(tp,
1836 							ticks - tp->t_rtttime);
1837 				}
1838 				acked = BYTES_THIS_ACK(tp, th);
1839 
1840 #ifdef TCP_HHOOK
1841 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1842 				hhook_run_tcp_est_in(tp, th, &to);
1843 #endif
1844 
1845 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1846 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1847 				sbdrop(&so->so_snd, acked);
1848 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1849 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1850 					tp->snd_recover = th->th_ack - 1;
1851 
1852 				/*
1853 				 * Let the congestion control algorithm update
1854 				 * congestion control related information. This
1855 				 * typically means increasing the congestion
1856 				 * window.
1857 				 */
1858 				cc_ack_received(tp, th, nsegs, CC_ACK);
1859 
1860 				tp->snd_una = th->th_ack;
1861 				/*
1862 				 * Pull snd_wl2 up to prevent seq wrap relative
1863 				 * to th_ack.
1864 				 */
1865 				tp->snd_wl2 = th->th_ack;
1866 				tp->t_dupacks = 0;
1867 				m_freem(m);
1868 
1869 				/*
1870 				 * If all outstanding data are acked, stop
1871 				 * retransmit timer, otherwise restart timer
1872 				 * using current (possibly backed-off) value.
1873 				 * If process is waiting for space,
1874 				 * wakeup/selwakeup/signal.  If data
1875 				 * are ready to send, let tcp_output
1876 				 * decide between more output or persist.
1877 				 */
1878 #ifdef TCPDEBUG
1879 				if (so->so_options & SO_DEBUG)
1880 					tcp_trace(TA_INPUT, ostate, tp,
1881 					    (void *)tcp_saveipgen,
1882 					    &tcp_savetcp, 0);
1883 #endif
1884 				TCP_PROBE3(debug__input, tp, th, m);
1885 				if (tp->snd_una == tp->snd_max)
1886 					tcp_timer_activate(tp, TT_REXMT, 0);
1887 				else if (!tcp_timer_active(tp, TT_PERSIST))
1888 					tcp_timer_activate(tp, TT_REXMT,
1889 						      tp->t_rxtcur);
1890 				sowwakeup(so);
1891 				if (sbavail(&so->so_snd))
1892 					(void) tcp_output(tp);
1893 				goto check_delack;
1894 			}
1895 		} else if (th->th_ack == tp->snd_una &&
1896 		    tlen <= sbspace(&so->so_rcv)) {
1897 			int newsize = 0;	/* automatic sockbuf scaling */
1898 
1899 			/*
1900 			 * This is a pure, in-sequence data packet with
1901 			 * nothing on the reassembly queue and we have enough
1902 			 * buffer space to take it.
1903 			 */
1904 			/* Clean receiver SACK report if present */
1905 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1906 				tcp_clean_sackreport(tp);
1907 			TCPSTAT_INC(tcps_preddat);
1908 			tp->rcv_nxt += tlen;
1909 			if (tlen &&
1910 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1911 			    (tp->t_fbyte_in == 0)) {
1912 				tp->t_fbyte_in = ticks;
1913 				if (tp->t_fbyte_in == 0)
1914 					tp->t_fbyte_in = 1;
1915 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1916 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1917 			}
1918 			/*
1919 			 * Pull snd_wl1 up to prevent seq wrap relative to
1920 			 * th_seq.
1921 			 */
1922 			tp->snd_wl1 = th->th_seq;
1923 			/*
1924 			 * Pull rcv_up up to prevent seq wrap relative to
1925 			 * rcv_nxt.
1926 			 */
1927 			tp->rcv_up = tp->rcv_nxt;
1928 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1929 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1930 #ifdef TCPDEBUG
1931 			if (so->so_options & SO_DEBUG)
1932 				tcp_trace(TA_INPUT, ostate, tp,
1933 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1934 #endif
1935 			TCP_PROBE3(debug__input, tp, th, m);
1936 
1937 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1938 
1939 			/* Add data to socket buffer. */
1940 			SOCKBUF_LOCK(&so->so_rcv);
1941 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1942 				m_freem(m);
1943 			} else {
1944 				/*
1945 				 * Set new socket buffer size.
1946 				 * Give up when limit is reached.
1947 				 */
1948 				if (newsize)
1949 					if (!sbreserve_locked(&so->so_rcv,
1950 					    newsize, so, NULL))
1951 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1952 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1953 				sbappendstream_locked(&so->so_rcv, m, 0);
1954 			}
1955 			/* NB: sorwakeup_locked() does an implicit unlock. */
1956 			sorwakeup_locked(so);
1957 			if (DELAY_ACK(tp, tlen)) {
1958 				tp->t_flags |= TF_DELACK;
1959 			} else {
1960 				tp->t_flags |= TF_ACKNOW;
1961 				tcp_output(tp);
1962 			}
1963 			goto check_delack;
1964 		}
1965 	}
1966 
1967 	/*
1968 	 * Calculate amount of space in receive window,
1969 	 * and then do TCP input processing.
1970 	 * Receive window is amount of space in rcv queue,
1971 	 * but not less than advertised window.
1972 	 */
1973 	win = sbspace(&so->so_rcv);
1974 	if (win < 0)
1975 		win = 0;
1976 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1977 
1978 	switch (tp->t_state) {
1979 	/*
1980 	 * If the state is SYN_RECEIVED:
1981 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1982 	 */
1983 	case TCPS_SYN_RECEIVED:
1984 		if ((thflags & TH_ACK) &&
1985 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1986 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1987 				rstreason = BANDLIM_RST_OPENPORT;
1988 				goto dropwithreset;
1989 		}
1990 		if (IS_FASTOPEN(tp->t_flags)) {
1991 			/*
1992 			 * When a TFO connection is in SYN_RECEIVED, the
1993 			 * only valid packets are the initial SYN, a
1994 			 * retransmit/copy of the initial SYN (possibly with
1995 			 * a subset of the original data), a valid ACK, a
1996 			 * FIN, or a RST.
1997 			 */
1998 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1999 				rstreason = BANDLIM_RST_OPENPORT;
2000 				goto dropwithreset;
2001 			} else if (thflags & TH_SYN) {
2002 				/* non-initial SYN is ignored */
2003 				if ((tcp_timer_active(tp, TT_DELACK) ||
2004 				     tcp_timer_active(tp, TT_REXMT)))
2005 					goto drop;
2006 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2007 				goto drop;
2008 			}
2009 		}
2010 		break;
2011 
2012 	/*
2013 	 * If the state is SYN_SENT:
2014 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
2015 	 *	    been verified), then drop the connection.
2016 	 *	if seg contains a RST without an ACK, drop the seg.
2017 	 *	if seg does not contain SYN, then drop the seg.
2018 	 * Otherwise this is an acceptable SYN segment
2019 	 *	initialize tp->rcv_nxt and tp->irs
2020 	 *	if seg contains ack then advance tp->snd_una
2021 	 *	if seg contains an ECE and ECN support is enabled, the stream
2022 	 *	    is ECN capable.
2023 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2024 	 *	arrange for segment to be acked (eventually)
2025 	 *	continue processing rest of data/controls, beginning with URG
2026 	 */
2027 	case TCPS_SYN_SENT:
2028 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2029 			TCP_PROBE5(connect__refused, NULL, tp,
2030 			    m, tp, th);
2031 			tp = tcp_drop(tp, ECONNREFUSED);
2032 		}
2033 		if (thflags & TH_RST)
2034 			goto drop;
2035 		if (!(thflags & TH_SYN))
2036 			goto drop;
2037 
2038 		tp->irs = th->th_seq;
2039 		tcp_rcvseqinit(tp);
2040 		if (thflags & TH_ACK) {
2041 			int tfo_partial_ack = 0;
2042 
2043 			TCPSTAT_INC(tcps_connects);
2044 			soisconnected(so);
2045 #ifdef MAC
2046 			mac_socketpeer_set_from_mbuf(m, so);
2047 #endif
2048 			/* Do window scaling on this connection? */
2049 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2050 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2051 				tp->rcv_scale = tp->request_r_scale;
2052 			}
2053 			tp->rcv_adv += min(tp->rcv_wnd,
2054 			    TCP_MAXWIN << tp->rcv_scale);
2055 			tp->snd_una++;		/* SYN is acked */
2056 			/*
2057 			 * If not all the data that was sent in the TFO SYN
2058 			 * has been acked, resend the remainder right away.
2059 			 */
2060 			if (IS_FASTOPEN(tp->t_flags) &&
2061 			    (tp->snd_una != tp->snd_max)) {
2062 				tp->snd_nxt = th->th_ack;
2063 				tfo_partial_ack = 1;
2064 			}
2065 			/*
2066 			 * If there's data, delay ACK; if there's also a FIN
2067 			 * ACKNOW will be turned on later.
2068 			 */
2069 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2070 				tcp_timer_activate(tp, TT_DELACK,
2071 				    tcp_delacktime);
2072 			else
2073 				tp->t_flags |= TF_ACKNOW;
2074 
2075 			if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2076 			    (V_tcp_do_ecn == 1)) {
2077 				tp->t_flags2 |= TF2_ECN_PERMIT;
2078 				TCPSTAT_INC(tcps_ecn_shs);
2079 			}
2080 
2081 			/*
2082 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2083 			 * Transitions:
2084 			 *	SYN_SENT  --> ESTABLISHED
2085 			 *	SYN_SENT* --> FIN_WAIT_1
2086 			 */
2087 			tp->t_starttime = ticks;
2088 			if (tp->t_flags & TF_NEEDFIN) {
2089 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2090 				tp->t_flags &= ~TF_NEEDFIN;
2091 				thflags &= ~TH_SYN;
2092 			} else {
2093 				tcp_state_change(tp, TCPS_ESTABLISHED);
2094 				TCP_PROBE5(connect__established, NULL, tp,
2095 				    m, tp, th);
2096 				cc_conn_init(tp);
2097 				tcp_timer_activate(tp, TT_KEEP,
2098 				    TP_KEEPIDLE(tp));
2099 			}
2100 		} else {
2101 			/*
2102 			 * Received initial SYN in SYN-SENT[*] state =>
2103 			 * simultaneous open.
2104 			 * If it succeeds, connection is * half-synchronized.
2105 			 * Otherwise, do 3-way handshake:
2106 			 *        SYN-SENT -> SYN-RECEIVED
2107 			 *        SYN-SENT* -> SYN-RECEIVED*
2108 			 */
2109 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2110 			tcp_timer_activate(tp, TT_REXMT, 0);
2111 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2112 		}
2113 
2114 		INP_WLOCK_ASSERT(tp->t_inpcb);
2115 
2116 		/*
2117 		 * Advance th->th_seq to correspond to first data byte.
2118 		 * If data, trim to stay within window,
2119 		 * dropping FIN if necessary.
2120 		 */
2121 		th->th_seq++;
2122 		if (tlen > tp->rcv_wnd) {
2123 			todrop = tlen - tp->rcv_wnd;
2124 			m_adj(m, -todrop);
2125 			tlen = tp->rcv_wnd;
2126 			thflags &= ~TH_FIN;
2127 			TCPSTAT_INC(tcps_rcvpackafterwin);
2128 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2129 		}
2130 		tp->snd_wl1 = th->th_seq - 1;
2131 		tp->rcv_up = th->th_seq;
2132 		/*
2133 		 * Client side of transaction: already sent SYN and data.
2134 		 * If the remote host used T/TCP to validate the SYN,
2135 		 * our data will be ACK'd; if so, enter normal data segment
2136 		 * processing in the middle of step 5, ack processing.
2137 		 * Otherwise, goto step 6.
2138 		 */
2139 		if (thflags & TH_ACK)
2140 			goto process_ACK;
2141 
2142 		goto step6;
2143 
2144 	/*
2145 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2146 	 *      do normal processing.
2147 	 *
2148 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2149 	 */
2150 	case TCPS_LAST_ACK:
2151 	case TCPS_CLOSING:
2152 		break;  /* continue normal processing */
2153 	}
2154 
2155 	/*
2156 	 * States other than LISTEN or SYN_SENT.
2157 	 * First check the RST flag and sequence number since reset segments
2158 	 * are exempt from the timestamp and connection count tests.  This
2159 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2160 	 * below which allowed reset segments in half the sequence space
2161 	 * to fall though and be processed (which gives forged reset
2162 	 * segments with a random sequence number a 50 percent chance of
2163 	 * killing a connection).
2164 	 * Then check timestamp, if present.
2165 	 * Then check the connection count, if present.
2166 	 * Then check that at least some bytes of segment are within
2167 	 * receive window.  If segment begins before rcv_nxt,
2168 	 * drop leading data (and SYN); if nothing left, just ack.
2169 	 */
2170 	if (thflags & TH_RST) {
2171 		/*
2172 		 * RFC5961 Section 3.2
2173 		 *
2174 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2175 		 * - If RST is in window, we send challenge ACK.
2176 		 *
2177 		 * Note: to take into account delayed ACKs, we should
2178 		 *   test against last_ack_sent instead of rcv_nxt.
2179 		 * Note 2: we handle special case of closed window, not
2180 		 *   covered by the RFC.
2181 		 */
2182 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2183 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2184 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2185 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2186 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2187 			    __func__, th, tp));
2188 
2189 			if (V_tcp_insecure_rst ||
2190 			    tp->last_ack_sent == th->th_seq) {
2191 				TCPSTAT_INC(tcps_drops);
2192 				/* Drop the connection. */
2193 				switch (tp->t_state) {
2194 				case TCPS_SYN_RECEIVED:
2195 					so->so_error = ECONNREFUSED;
2196 					goto close;
2197 				case TCPS_ESTABLISHED:
2198 				case TCPS_FIN_WAIT_1:
2199 				case TCPS_FIN_WAIT_2:
2200 				case TCPS_CLOSE_WAIT:
2201 				case TCPS_CLOSING:
2202 				case TCPS_LAST_ACK:
2203 					so->so_error = ECONNRESET;
2204 				close:
2205 					/* FALLTHROUGH */
2206 				default:
2207 					tp = tcp_close(tp);
2208 				}
2209 			} else {
2210 				TCPSTAT_INC(tcps_badrst);
2211 				/* Send challenge ACK. */
2212 				tcp_respond(tp, mtod(m, void *), th, m,
2213 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2214 				tp->last_ack_sent = tp->rcv_nxt;
2215 				m = NULL;
2216 			}
2217 		}
2218 		goto drop;
2219 	}
2220 
2221 	/*
2222 	 * RFC5961 Section 4.2
2223 	 * Send challenge ACK for any SYN in synchronized state.
2224 	 */
2225 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2226 	    tp->t_state != TCPS_SYN_RECEIVED) {
2227 		TCPSTAT_INC(tcps_badsyn);
2228 		if (V_tcp_insecure_syn &&
2229 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2230 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2231 			tp = tcp_drop(tp, ECONNRESET);
2232 			rstreason = BANDLIM_UNLIMITED;
2233 		} else {
2234 			/* Send challenge ACK. */
2235 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2236 			    tp->snd_nxt, TH_ACK);
2237 			tp->last_ack_sent = tp->rcv_nxt;
2238 			m = NULL;
2239 		}
2240 		goto drop;
2241 	}
2242 
2243 	/*
2244 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2245 	 * and it's less than ts_recent, drop it.
2246 	 */
2247 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2248 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2249 		/* Check to see if ts_recent is over 24 days old.  */
2250 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2251 			/*
2252 			 * Invalidate ts_recent.  If this segment updates
2253 			 * ts_recent, the age will be reset later and ts_recent
2254 			 * will get a valid value.  If it does not, setting
2255 			 * ts_recent to zero will at least satisfy the
2256 			 * requirement that zero be placed in the timestamp
2257 			 * echo reply when ts_recent isn't valid.  The
2258 			 * age isn't reset until we get a valid ts_recent
2259 			 * because we don't want out-of-order segments to be
2260 			 * dropped when ts_recent is old.
2261 			 */
2262 			tp->ts_recent = 0;
2263 		} else {
2264 			TCPSTAT_INC(tcps_rcvduppack);
2265 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2266 			TCPSTAT_INC(tcps_pawsdrop);
2267 			if (tlen)
2268 				goto dropafterack;
2269 			goto drop;
2270 		}
2271 	}
2272 
2273 	/*
2274 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2275 	 * this connection before trimming the data to fit the receive
2276 	 * window.  Check the sequence number versus IRS since we know
2277 	 * the sequence numbers haven't wrapped.  This is a partial fix
2278 	 * for the "LAND" DoS attack.
2279 	 */
2280 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2281 		rstreason = BANDLIM_RST_OPENPORT;
2282 		goto dropwithreset;
2283 	}
2284 
2285 	todrop = tp->rcv_nxt - th->th_seq;
2286 	if (todrop > 0) {
2287 		if (thflags & TH_SYN) {
2288 			thflags &= ~TH_SYN;
2289 			th->th_seq++;
2290 			if (th->th_urp > 1)
2291 				th->th_urp--;
2292 			else
2293 				thflags &= ~TH_URG;
2294 			todrop--;
2295 		}
2296 		/*
2297 		 * Following if statement from Stevens, vol. 2, p. 960.
2298 		 */
2299 		if (todrop > tlen
2300 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2301 			/*
2302 			 * Any valid FIN must be to the left of the window.
2303 			 * At this point the FIN must be a duplicate or out
2304 			 * of sequence; drop it.
2305 			 */
2306 			thflags &= ~TH_FIN;
2307 
2308 			/*
2309 			 * Send an ACK to resynchronize and drop any data.
2310 			 * But keep on processing for RST or ACK.
2311 			 */
2312 			tp->t_flags |= TF_ACKNOW;
2313 			todrop = tlen;
2314 			TCPSTAT_INC(tcps_rcvduppack);
2315 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2316 		} else {
2317 			TCPSTAT_INC(tcps_rcvpartduppack);
2318 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2319 		}
2320 		/*
2321 		 * DSACK - add SACK block for dropped range
2322 		 */
2323 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2324 			tcp_update_sack_list(tp, th->th_seq,
2325 			    th->th_seq + todrop);
2326 			/*
2327 			 * ACK now, as the next in-sequence segment
2328 			 * will clear the DSACK block again
2329 			 */
2330 			tp->t_flags |= TF_ACKNOW;
2331 		}
2332 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2333 		th->th_seq += todrop;
2334 		tlen -= todrop;
2335 		if (th->th_urp > todrop)
2336 			th->th_urp -= todrop;
2337 		else {
2338 			thflags &= ~TH_URG;
2339 			th->th_urp = 0;
2340 		}
2341 	}
2342 
2343 	/*
2344 	 * If new data are received on a connection after the
2345 	 * user processes are gone, then RST the other end.
2346 	 */
2347 	if ((so->so_state & SS_NOFDREF) &&
2348 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2349 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2350 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2351 			    "after socket was closed, "
2352 			    "sending RST and removing tcpcb\n",
2353 			    s, __func__, tcpstates[tp->t_state], tlen);
2354 			free(s, M_TCPLOG);
2355 		}
2356 		tp = tcp_close(tp);
2357 		TCPSTAT_INC(tcps_rcvafterclose);
2358 		rstreason = BANDLIM_UNLIMITED;
2359 		goto dropwithreset;
2360 	}
2361 
2362 	/*
2363 	 * If segment ends after window, drop trailing data
2364 	 * (and PUSH and FIN); if nothing left, just ACK.
2365 	 */
2366 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2367 	if (todrop > 0) {
2368 		TCPSTAT_INC(tcps_rcvpackafterwin);
2369 		if (todrop >= tlen) {
2370 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2371 			/*
2372 			 * If window is closed can only take segments at
2373 			 * window edge, and have to drop data and PUSH from
2374 			 * incoming segments.  Continue processing, but
2375 			 * remember to ack.  Otherwise, drop segment
2376 			 * and ack.
2377 			 */
2378 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2379 				tp->t_flags |= TF_ACKNOW;
2380 				TCPSTAT_INC(tcps_rcvwinprobe);
2381 			} else
2382 				goto dropafterack;
2383 		} else
2384 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2385 		m_adj(m, -todrop);
2386 		tlen -= todrop;
2387 		thflags &= ~(TH_PUSH|TH_FIN);
2388 	}
2389 
2390 	/*
2391 	 * If last ACK falls within this segment's sequence numbers,
2392 	 * record its timestamp.
2393 	 * NOTE:
2394 	 * 1) That the test incorporates suggestions from the latest
2395 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2396 	 * 2) That updating only on newer timestamps interferes with
2397 	 *    our earlier PAWS tests, so this check should be solely
2398 	 *    predicated on the sequence space of this segment.
2399 	 * 3) That we modify the segment boundary check to be
2400 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2401 	 *    instead of RFC1323's
2402 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2403 	 *    This modified check allows us to overcome RFC1323's
2404 	 *    limitations as described in Stevens TCP/IP Illustrated
2405 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2406 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2407 	 */
2408 	if ((to.to_flags & TOF_TS) != 0 &&
2409 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2410 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2411 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2412 		tp->ts_recent_age = tcp_ts_getticks();
2413 		tp->ts_recent = to.to_tsval;
2414 	}
2415 
2416 	/*
2417 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2418 	 * flag is on (half-synchronized state), then queue data for
2419 	 * later processing; else drop segment and return.
2420 	 */
2421 	if ((thflags & TH_ACK) == 0) {
2422 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2423 		    (tp->t_flags & TF_NEEDSYN)) {
2424 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2425 			    IS_FASTOPEN(tp->t_flags)) {
2426 				tp->snd_wnd = tiwin;
2427 				cc_conn_init(tp);
2428 			}
2429 			goto step6;
2430 		} else if (tp->t_flags & TF_ACKNOW)
2431 			goto dropafterack;
2432 		else
2433 			goto drop;
2434 	}
2435 
2436 	/*
2437 	 * Ack processing.
2438 	 */
2439 	switch (tp->t_state) {
2440 	/*
2441 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2442 	 * ESTABLISHED state and continue processing.
2443 	 * The ACK was checked above.
2444 	 */
2445 	case TCPS_SYN_RECEIVED:
2446 
2447 		TCPSTAT_INC(tcps_connects);
2448 		soisconnected(so);
2449 		/* Do window scaling? */
2450 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2451 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2452 			tp->rcv_scale = tp->request_r_scale;
2453 		}
2454 		tp->snd_wnd = tiwin;
2455 		/*
2456 		 * Make transitions:
2457 		 *      SYN-RECEIVED  -> ESTABLISHED
2458 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2459 		 */
2460 		tp->t_starttime = ticks;
2461 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2462 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2463 			tp->t_tfo_pending = NULL;
2464 		}
2465 		if (tp->t_flags & TF_NEEDFIN) {
2466 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2467 			tp->t_flags &= ~TF_NEEDFIN;
2468 		} else {
2469 			tcp_state_change(tp, TCPS_ESTABLISHED);
2470 			TCP_PROBE5(accept__established, NULL, tp,
2471 			    m, tp, th);
2472 			/*
2473 			 * TFO connections call cc_conn_init() during SYN
2474 			 * processing.  Calling it again here for such
2475 			 * connections is not harmless as it would undo the
2476 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2477 			 * is retransmitted.
2478 			 */
2479 			if (!IS_FASTOPEN(tp->t_flags))
2480 				cc_conn_init(tp);
2481 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2482 		}
2483 		/*
2484 		 * Account for the ACK of our SYN prior to
2485 		 * regular ACK processing below, except for
2486 		 * simultaneous SYN, which is handled later.
2487 		 */
2488 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2489 			incforsyn = 1;
2490 		/*
2491 		 * If segment contains data or ACK, will call tcp_reass()
2492 		 * later; if not, do so now to pass queued data to user.
2493 		 */
2494 		if (tlen == 0 && (thflags & TH_FIN) == 0) {
2495 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2496 			    (struct mbuf *)0);
2497 			tcp_handle_wakeup(tp, so);
2498 		}
2499 		tp->snd_wl1 = th->th_seq - 1;
2500 		/* FALLTHROUGH */
2501 
2502 	/*
2503 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2504 	 * ACKs.  If the ack is in the range
2505 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2506 	 * then advance tp->snd_una to th->th_ack and drop
2507 	 * data from the retransmission queue.  If this ACK reflects
2508 	 * more up to date window information we update our window information.
2509 	 */
2510 	case TCPS_ESTABLISHED:
2511 	case TCPS_FIN_WAIT_1:
2512 	case TCPS_FIN_WAIT_2:
2513 	case TCPS_CLOSE_WAIT:
2514 	case TCPS_CLOSING:
2515 	case TCPS_LAST_ACK:
2516 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2517 			TCPSTAT_INC(tcps_rcvacktoomuch);
2518 			goto dropafterack;
2519 		}
2520 		if ((tp->t_flags & TF_SACK_PERMIT) &&
2521 		    ((to.to_flags & TOF_SACK) ||
2522 		     !TAILQ_EMPTY(&tp->snd_holes))) {
2523 			if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2524 			    (tp->t_flags & TF_LRD)) {
2525 				tcp_sack_lost_retransmission(tp, th);
2526 			}
2527 		} else
2528 			/*
2529 			 * Reset the value so that previous (valid) value
2530 			 * from the last ack with SACK doesn't get used.
2531 			 */
2532 			tp->sackhint.sacked_bytes = 0;
2533 
2534 #ifdef TCP_HHOOK
2535 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2536 		hhook_run_tcp_est_in(tp, th, &to);
2537 #endif
2538 
2539 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2540 			maxseg = tcp_maxseg(tp);
2541 			if (tlen == 0 &&
2542 			    (tiwin == tp->snd_wnd ||
2543 			    (tp->t_flags & TF_SACK_PERMIT))) {
2544 				/*
2545 				 * If this is the first time we've seen a
2546 				 * FIN from the remote, this is not a
2547 				 * duplicate and it needs to be processed
2548 				 * normally.  This happens during a
2549 				 * simultaneous close.
2550 				 */
2551 				if ((thflags & TH_FIN) &&
2552 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2553 					tp->t_dupacks = 0;
2554 					break;
2555 				}
2556 				TCPSTAT_INC(tcps_rcvdupack);
2557 				/*
2558 				 * If we have outstanding data (other than
2559 				 * a window probe), this is a completely
2560 				 * duplicate ack (ie, window info didn't
2561 				 * change and FIN isn't set),
2562 				 * the ack is the biggest we've
2563 				 * seen and we've seen exactly our rexmt
2564 				 * threshold of them, assume a packet
2565 				 * has been dropped and retransmit it.
2566 				 * Kludge snd_nxt & the congestion
2567 				 * window so we send only this one
2568 				 * packet.
2569 				 *
2570 				 * We know we're losing at the current
2571 				 * window size so do congestion avoidance
2572 				 * (set ssthresh to half the current window
2573 				 * and pull our congestion window back to
2574 				 * the new ssthresh).
2575 				 *
2576 				 * Dup acks mean that packets have left the
2577 				 * network (they're now cached at the receiver)
2578 				 * so bump cwnd by the amount in the receiver
2579 				 * to keep a constant cwnd packets in the
2580 				 * network.
2581 				 *
2582 				 * When using TCP ECN, notify the peer that
2583 				 * we reduced the cwnd.
2584 				 */
2585 				/*
2586 				 * Following 2 kinds of acks should not affect
2587 				 * dupack counting:
2588 				 * 1) Old acks
2589 				 * 2) Acks with SACK but without any new SACK
2590 				 * information in them. These could result from
2591 				 * any anomaly in the network like a switch
2592 				 * duplicating packets or a possible DoS attack.
2593 				 */
2594 				if (th->th_ack != tp->snd_una ||
2595 				    ((tp->t_flags & TF_SACK_PERMIT) &&
2596 				    (to.to_flags & TOF_SACK) &&
2597 				    !sack_changed))
2598 					break;
2599 				else if (!tcp_timer_active(tp, TT_REXMT))
2600 					tp->t_dupacks = 0;
2601 				else if (++tp->t_dupacks > tcprexmtthresh ||
2602 				     IN_FASTRECOVERY(tp->t_flags)) {
2603 					cc_ack_received(tp, th, nsegs,
2604 					    CC_DUPACK);
2605 					if (V_tcp_do_prr &&
2606 					    IN_FASTRECOVERY(tp->t_flags)) {
2607 						tcp_do_prr_ack(tp, th, &to);
2608 					} else if ((tp->t_flags & TF_SACK_PERMIT) &&
2609 					    (to.to_flags & TOF_SACK) &&
2610 					    IN_FASTRECOVERY(tp->t_flags)) {
2611 						int awnd;
2612 
2613 						/*
2614 						 * Compute the amount of data in flight first.
2615 						 * We can inject new data into the pipe iff
2616 						 * we have less than 1/2 the original window's
2617 						 * worth of data in flight.
2618 						 */
2619 						if (V_tcp_do_newsack)
2620 							awnd = tcp_compute_pipe(tp);
2621 						else
2622 							awnd = (tp->snd_nxt - tp->snd_fack) +
2623 								tp->sackhint.sack_bytes_rexmit;
2624 
2625 						if (awnd < tp->snd_ssthresh) {
2626 							tp->snd_cwnd += maxseg;
2627 							if (tp->snd_cwnd > tp->snd_ssthresh)
2628 								tp->snd_cwnd = tp->snd_ssthresh;
2629 						}
2630 					} else
2631 						tp->snd_cwnd += maxseg;
2632 					(void) tcp_output(tp);
2633 					goto drop;
2634 				} else if (tp->t_dupacks == tcprexmtthresh ||
2635 					    (tp->t_flags & TF_SACK_PERMIT &&
2636 					     V_tcp_do_newsack &&
2637 					     tp->sackhint.sacked_bytes >
2638 					     (tcprexmtthresh - 1) * maxseg)) {
2639 enter_recovery:
2640 					/*
2641 					 * Above is the RFC6675 trigger condition of
2642 					 * more than (dupthresh-1)*maxseg sacked data.
2643 					 * If the count of holes in the
2644 					 * scoreboard is >= dupthresh, we could
2645 					 * also enter loss recovery, but don't
2646 					 * have that value readily available.
2647 					 */
2648 					tp->t_dupacks = tcprexmtthresh;
2649 					tcp_seq onxt = tp->snd_nxt;
2650 
2651 					/*
2652 					 * If we're doing sack, or prr, check
2653 					 * to see if we're already in sack
2654 					 * recovery. If we're not doing sack,
2655 					 * check to see if we're in newreno
2656 					 * recovery.
2657 					 */
2658 					if (V_tcp_do_prr ||
2659 					    (tp->t_flags & TF_SACK_PERMIT)) {
2660 						if (IN_FASTRECOVERY(tp->t_flags)) {
2661 							tp->t_dupacks = 0;
2662 							break;
2663 						}
2664 					} else {
2665 						if (SEQ_LEQ(th->th_ack,
2666 						    tp->snd_recover)) {
2667 							tp->t_dupacks = 0;
2668 							break;
2669 						}
2670 					}
2671 					/* Congestion signal before ack. */
2672 					cc_cong_signal(tp, th, CC_NDUPACK);
2673 					cc_ack_received(tp, th, nsegs,
2674 					    CC_DUPACK);
2675 					tcp_timer_activate(tp, TT_REXMT, 0);
2676 					tp->t_rtttime = 0;
2677 					if (V_tcp_do_prr) {
2678 						/*
2679 						 * snd_ssthresh is already updated by
2680 						 * cc_cong_signal.
2681 						 */
2682 						if ((tp->t_flags & TF_SACK_PERMIT) &&
2683 						    (to.to_flags & TOF_SACK)) {
2684 							tp->sackhint.prr_delivered =
2685 							    tp->sackhint.sacked_bytes;
2686 						} else {
2687 							tp->sackhint.prr_delivered =
2688 							    imin(tp->snd_max - tp->snd_una,
2689 							    imin(INT_MAX / 65536,
2690 								tp->t_dupacks) * maxseg);
2691 						}
2692 						tp->sackhint.recover_fs = max(1,
2693 						    tp->snd_nxt - tp->snd_una);
2694 					}
2695 					if ((tp->t_flags & TF_SACK_PERMIT) &&
2696 					    (to.to_flags & TOF_SACK)) {
2697 						TCPSTAT_INC(
2698 						    tcps_sack_recovery_episode);
2699 						tp->snd_recover = tp->snd_nxt;
2700 						tp->snd_cwnd = maxseg;
2701 						(void) tcp_output(tp);
2702 						if (SEQ_GT(th->th_ack, tp->snd_una))
2703 							goto resume_partialack;
2704 						goto drop;
2705 					}
2706 					tp->snd_nxt = th->th_ack;
2707 					tp->snd_cwnd = maxseg;
2708 					(void) tcp_output(tp);
2709 					KASSERT(tp->snd_limited <= 2,
2710 					    ("%s: tp->snd_limited too big",
2711 					    __func__));
2712 					tp->snd_cwnd = tp->snd_ssthresh +
2713 					     maxseg *
2714 					     (tp->t_dupacks - tp->snd_limited);
2715 					if (SEQ_GT(onxt, tp->snd_nxt))
2716 						tp->snd_nxt = onxt;
2717 					goto drop;
2718 				} else if (V_tcp_do_rfc3042) {
2719 					/*
2720 					 * Process first and second duplicate
2721 					 * ACKs. Each indicates a segment
2722 					 * leaving the network, creating room
2723 					 * for more. Make sure we can send a
2724 					 * packet on reception of each duplicate
2725 					 * ACK by increasing snd_cwnd by one
2726 					 * segment. Restore the original
2727 					 * snd_cwnd after packet transmission.
2728 					 */
2729 					cc_ack_received(tp, th, nsegs,
2730 					    CC_DUPACK);
2731 					uint32_t oldcwnd = tp->snd_cwnd;
2732 					tcp_seq oldsndmax = tp->snd_max;
2733 					u_int sent;
2734 					int avail;
2735 
2736 					KASSERT(tp->t_dupacks == 1 ||
2737 					    tp->t_dupacks == 2,
2738 					    ("%s: dupacks not 1 or 2",
2739 					    __func__));
2740 					if (tp->t_dupacks == 1)
2741 						tp->snd_limited = 0;
2742 					tp->snd_cwnd =
2743 					    (tp->snd_nxt - tp->snd_una) +
2744 					    (tp->t_dupacks - tp->snd_limited) *
2745 					    maxseg;
2746 					/*
2747 					 * Only call tcp_output when there
2748 					 * is new data available to be sent.
2749 					 * Otherwise we would send pure ACKs.
2750 					 */
2751 					SOCKBUF_LOCK(&so->so_snd);
2752 					avail = sbavail(&so->so_snd) -
2753 					    (tp->snd_nxt - tp->snd_una);
2754 					SOCKBUF_UNLOCK(&so->so_snd);
2755 					if (avail > 0)
2756 						(void) tcp_output(tp);
2757 					sent = tp->snd_max - oldsndmax;
2758 					if (sent > maxseg) {
2759 						KASSERT((tp->t_dupacks == 2 &&
2760 						    tp->snd_limited == 0) ||
2761 						   (sent == maxseg + 1 &&
2762 						    tp->t_flags & TF_SENTFIN),
2763 						    ("%s: sent too much",
2764 						    __func__));
2765 						tp->snd_limited = 2;
2766 					} else if (sent > 0)
2767 						++tp->snd_limited;
2768 					tp->snd_cwnd = oldcwnd;
2769 					goto drop;
2770 				}
2771 			}
2772 			break;
2773 		} else {
2774 			/*
2775 			 * This ack is advancing the left edge, reset the
2776 			 * counter.
2777 			 */
2778 			tp->t_dupacks = 0;
2779 			/*
2780 			 * If this ack also has new SACK info, increment the
2781 			 * counter as per rfc6675. The variable
2782 			 * sack_changed tracks all changes to the SACK
2783 			 * scoreboard, including when partial ACKs without
2784 			 * SACK options are received, and clear the scoreboard
2785 			 * from the left side. Such partial ACKs should not be
2786 			 * counted as dupacks here.
2787 			 */
2788 			if ((tp->t_flags & TF_SACK_PERMIT) &&
2789 			    (to.to_flags & TOF_SACK) &&
2790 			    sack_changed) {
2791 				tp->t_dupacks++;
2792 				/* limit overhead by setting maxseg last */
2793 				if (!IN_FASTRECOVERY(tp->t_flags) &&
2794 				    (tp->sackhint.sacked_bytes >
2795 				    ((tcprexmtthresh - 1) *
2796 				    (maxseg = tcp_maxseg(tp))))) {
2797 					goto enter_recovery;
2798 				}
2799 			}
2800 		}
2801 
2802 resume_partialack:
2803 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2804 		    ("%s: th_ack <= snd_una", __func__));
2805 
2806 		/*
2807 		 * If the congestion window was inflated to account
2808 		 * for the other side's cached packets, retract it.
2809 		 */
2810 		if (IN_FASTRECOVERY(tp->t_flags)) {
2811 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2812 				if (tp->t_flags & TF_SACK_PERMIT)
2813 					if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2814 						tcp_timer_activate(tp, TT_REXMT, 0);
2815 						tp->t_rtttime = 0;
2816 						tcp_do_prr_ack(tp, th, &to);
2817 						tp->t_flags |= TF_ACKNOW;
2818 						(void) tcp_output(tp);
2819 					} else
2820 						tcp_sack_partialack(tp, th);
2821 				else
2822 					tcp_newreno_partial_ack(tp, th);
2823 			} else
2824 				cc_post_recovery(tp, th);
2825 		} else if (IN_CONGRECOVERY(tp->t_flags)) {
2826 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2827 				if (V_tcp_do_prr) {
2828 					tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2829 					tp->snd_fack = th->th_ack;
2830 					tcp_do_prr_ack(tp, th, &to);
2831 					(void) tcp_output(tp);
2832 				}
2833 			} else
2834 				cc_post_recovery(tp, th);
2835 		}
2836 		/*
2837 		 * If we reach this point, ACK is not a duplicate,
2838 		 *     i.e., it ACKs something we sent.
2839 		 */
2840 		if (tp->t_flags & TF_NEEDSYN) {
2841 			/*
2842 			 * T/TCP: Connection was half-synchronized, and our
2843 			 * SYN has been ACK'd (so connection is now fully
2844 			 * synchronized).  Go to non-starred state,
2845 			 * increment snd_una for ACK of SYN, and check if
2846 			 * we can do window scaling.
2847 			 */
2848 			tp->t_flags &= ~TF_NEEDSYN;
2849 			tp->snd_una++;
2850 			/* Do window scaling? */
2851 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2852 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2853 				tp->rcv_scale = tp->request_r_scale;
2854 				/* Send window already scaled. */
2855 			}
2856 		}
2857 
2858 process_ACK:
2859 		INP_WLOCK_ASSERT(tp->t_inpcb);
2860 
2861 		/*
2862 		 * Adjust for the SYN bit in sequence space,
2863 		 * but don't account for it in cwnd calculations.
2864 		 * This is for the SYN_RECEIVED, non-simultaneous
2865 		 * SYN case. SYN_SENT and simultaneous SYN are
2866 		 * treated elsewhere.
2867 		 */
2868 		if (incforsyn)
2869 			tp->snd_una++;
2870 		acked = BYTES_THIS_ACK(tp, th);
2871 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2872 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2873 		    tp->snd_una, th->th_ack, tp, m));
2874 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2875 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2876 
2877 		/*
2878 		 * If we just performed our first retransmit, and the ACK
2879 		 * arrives within our recovery window, then it was a mistake
2880 		 * to do the retransmit in the first place.  Recover our
2881 		 * original cwnd and ssthresh, and proceed to transmit where
2882 		 * we left off.
2883 		 */
2884 		if (tp->t_rxtshift == 1 &&
2885 		    tp->t_flags & TF_PREVVALID &&
2886 		    tp->t_badrxtwin &&
2887 		    SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2888 			cc_cong_signal(tp, th, CC_RTO_ERR);
2889 
2890 		/*
2891 		 * If we have a timestamp reply, update smoothed
2892 		 * round trip time.  If no timestamp is present but
2893 		 * transmit timer is running and timed sequence
2894 		 * number was acked, update smoothed round trip time.
2895 		 * Since we now have an rtt measurement, cancel the
2896 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2897 		 * Recompute the initial retransmit timer.
2898 		 *
2899 		 * Some boxes send broken timestamp replies
2900 		 * during the SYN+ACK phase, ignore
2901 		 * timestamps of 0 or we could calculate a
2902 		 * huge RTT and blow up the retransmit timer.
2903 		 */
2904 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2905 			uint32_t t;
2906 
2907 			t = tcp_ts_getticks() - to.to_tsecr;
2908 			if (!tp->t_rttlow || tp->t_rttlow > t)
2909 				tp->t_rttlow = t;
2910 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2911 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2912 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2913 				tp->t_rttlow = ticks - tp->t_rtttime;
2914 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2915 		}
2916 
2917 		/*
2918 		 * If all outstanding data is acked, stop retransmit
2919 		 * timer and remember to restart (more output or persist).
2920 		 * If there is more data to be acked, restart retransmit
2921 		 * timer, using current (possibly backed-off) value.
2922 		 */
2923 		if (th->th_ack == tp->snd_max) {
2924 			tcp_timer_activate(tp, TT_REXMT, 0);
2925 			needoutput = 1;
2926 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2927 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2928 
2929 		/*
2930 		 * If no data (only SYN) was ACK'd,
2931 		 *    skip rest of ACK processing.
2932 		 */
2933 		if (acked == 0)
2934 			goto step6;
2935 
2936 		/*
2937 		 * Let the congestion control algorithm update congestion
2938 		 * control related information. This typically means increasing
2939 		 * the congestion window.
2940 		 */
2941 		cc_ack_received(tp, th, nsegs, CC_ACK);
2942 
2943 		SOCKBUF_LOCK(&so->so_snd);
2944 		if (acked > sbavail(&so->so_snd)) {
2945 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2946 				tp->snd_wnd -= sbavail(&so->so_snd);
2947 			else
2948 				tp->snd_wnd = 0;
2949 			mfree = sbcut_locked(&so->so_snd,
2950 			    (int)sbavail(&so->so_snd));
2951 			ourfinisacked = 1;
2952 		} else {
2953 			mfree = sbcut_locked(&so->so_snd, acked);
2954 			if (tp->snd_wnd >= (uint32_t) acked)
2955 				tp->snd_wnd -= acked;
2956 			else
2957 				tp->snd_wnd = 0;
2958 			ourfinisacked = 0;
2959 		}
2960 		/* NB: sowwakeup_locked() does an implicit unlock. */
2961 		sowwakeup_locked(so);
2962 		m_freem(mfree);
2963 		/* Detect una wraparound. */
2964 		if (!IN_RECOVERY(tp->t_flags) &&
2965 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2966 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2967 			tp->snd_recover = th->th_ack - 1;
2968 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2969 		if (IN_RECOVERY(tp->t_flags) &&
2970 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2971 			EXIT_RECOVERY(tp->t_flags);
2972 		}
2973 		tp->snd_una = th->th_ack;
2974 		if (tp->t_flags & TF_SACK_PERMIT) {
2975 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2976 				tp->snd_recover = tp->snd_una;
2977 		}
2978 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2979 			tp->snd_nxt = tp->snd_una;
2980 
2981 		switch (tp->t_state) {
2982 		/*
2983 		 * In FIN_WAIT_1 STATE in addition to the processing
2984 		 * for the ESTABLISHED state if our FIN is now acknowledged
2985 		 * then enter FIN_WAIT_2.
2986 		 */
2987 		case TCPS_FIN_WAIT_1:
2988 			if (ourfinisacked) {
2989 				/*
2990 				 * If we can't receive any more
2991 				 * data, then closing user can proceed.
2992 				 * Starting the timer is contrary to the
2993 				 * specification, but if we don't get a FIN
2994 				 * we'll hang forever.
2995 				 *
2996 				 * XXXjl:
2997 				 * we should release the tp also, and use a
2998 				 * compressed state.
2999 				 */
3000 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3001 					soisdisconnected(so);
3002 					tcp_timer_activate(tp, TT_2MSL,
3003 					    (tcp_fast_finwait2_recycle ?
3004 					    tcp_finwait2_timeout :
3005 					    TP_MAXIDLE(tp)));
3006 				}
3007 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
3008 			}
3009 			break;
3010 
3011 		/*
3012 		 * In CLOSING STATE in addition to the processing for
3013 		 * the ESTABLISHED state if the ACK acknowledges our FIN
3014 		 * then enter the TIME-WAIT state, otherwise ignore
3015 		 * the segment.
3016 		 */
3017 		case TCPS_CLOSING:
3018 			if (ourfinisacked) {
3019 				tcp_twstart(tp);
3020 				m_freem(m);
3021 				return;
3022 			}
3023 			break;
3024 
3025 		/*
3026 		 * In LAST_ACK, we may still be waiting for data to drain
3027 		 * and/or to be acked, as well as for the ack of our FIN.
3028 		 * If our FIN is now acknowledged, delete the TCB,
3029 		 * enter the closed state and return.
3030 		 */
3031 		case TCPS_LAST_ACK:
3032 			if (ourfinisacked) {
3033 				tp = tcp_close(tp);
3034 				goto drop;
3035 			}
3036 			break;
3037 		}
3038 	}
3039 
3040 step6:
3041 	INP_WLOCK_ASSERT(tp->t_inpcb);
3042 
3043 	/*
3044 	 * Update window information.
3045 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3046 	 */
3047 	if ((thflags & TH_ACK) &&
3048 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3049 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3050 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3051 		/* keep track of pure window updates */
3052 		if (tlen == 0 &&
3053 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3054 			TCPSTAT_INC(tcps_rcvwinupd);
3055 		tp->snd_wnd = tiwin;
3056 		tp->snd_wl1 = th->th_seq;
3057 		tp->snd_wl2 = th->th_ack;
3058 		if (tp->snd_wnd > tp->max_sndwnd)
3059 			tp->max_sndwnd = tp->snd_wnd;
3060 		needoutput = 1;
3061 	}
3062 
3063 	/*
3064 	 * Process segments with URG.
3065 	 */
3066 	if ((thflags & TH_URG) && th->th_urp &&
3067 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3068 		/*
3069 		 * This is a kludge, but if we receive and accept
3070 		 * random urgent pointers, we'll crash in
3071 		 * soreceive.  It's hard to imagine someone
3072 		 * actually wanting to send this much urgent data.
3073 		 */
3074 		SOCKBUF_LOCK(&so->so_rcv);
3075 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3076 			th->th_urp = 0;			/* XXX */
3077 			thflags &= ~TH_URG;		/* XXX */
3078 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3079 			goto dodata;			/* XXX */
3080 		}
3081 		/*
3082 		 * If this segment advances the known urgent pointer,
3083 		 * then mark the data stream.  This should not happen
3084 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3085 		 * a FIN has been received from the remote side.
3086 		 * In these states we ignore the URG.
3087 		 *
3088 		 * According to RFC961 (Assigned Protocols),
3089 		 * the urgent pointer points to the last octet
3090 		 * of urgent data.  We continue, however,
3091 		 * to consider it to indicate the first octet
3092 		 * of data past the urgent section as the original
3093 		 * spec states (in one of two places).
3094 		 */
3095 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3096 			tp->rcv_up = th->th_seq + th->th_urp;
3097 			so->so_oobmark = sbavail(&so->so_rcv) +
3098 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3099 			if (so->so_oobmark == 0)
3100 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3101 			sohasoutofband(so);
3102 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3103 		}
3104 		SOCKBUF_UNLOCK(&so->so_rcv);
3105 		/*
3106 		 * Remove out of band data so doesn't get presented to user.
3107 		 * This can happen independent of advancing the URG pointer,
3108 		 * but if two URG's are pending at once, some out-of-band
3109 		 * data may creep in... ick.
3110 		 */
3111 		if (th->th_urp <= (uint32_t)tlen &&
3112 		    !(so->so_options & SO_OOBINLINE)) {
3113 			/* hdr drop is delayed */
3114 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3115 		}
3116 	} else {
3117 		/*
3118 		 * If no out of band data is expected,
3119 		 * pull receive urgent pointer along
3120 		 * with the receive window.
3121 		 */
3122 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3123 			tp->rcv_up = tp->rcv_nxt;
3124 	}
3125 dodata:							/* XXX */
3126 	INP_WLOCK_ASSERT(tp->t_inpcb);
3127 
3128 	/*
3129 	 * Process the segment text, merging it into the TCP sequencing queue,
3130 	 * and arranging for acknowledgment of receipt if necessary.
3131 	 * This process logically involves adjusting tp->rcv_wnd as data
3132 	 * is presented to the user (this happens in tcp_usrreq.c,
3133 	 * case PRU_RCVD).  If a FIN has already been received on this
3134 	 * connection then we just ignore the text.
3135 	 */
3136 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3137 		   IS_FASTOPEN(tp->t_flags));
3138 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3139 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3140 		tcp_seq save_start = th->th_seq;
3141 		tcp_seq save_rnxt  = tp->rcv_nxt;
3142 		int     save_tlen  = tlen;
3143 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3144 		/*
3145 		 * Insert segment which includes th into TCP reassembly queue
3146 		 * with control block tp.  Set thflags to whether reassembly now
3147 		 * includes a segment with FIN.  This handles the common case
3148 		 * inline (segment is the next to be received on an established
3149 		 * connection, and the queue is empty), avoiding linkage into
3150 		 * and removal from the queue and repetition of various
3151 		 * conversions.
3152 		 * Set DELACK for segments received in order, but ack
3153 		 * immediately when segments are out of order (so
3154 		 * fast retransmit can work).
3155 		 */
3156 		if (th->th_seq == tp->rcv_nxt &&
3157 		    SEGQ_EMPTY(tp) &&
3158 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3159 		     tfo_syn)) {
3160 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3161 				tp->t_flags |= TF_DELACK;
3162 			else
3163 				tp->t_flags |= TF_ACKNOW;
3164 			tp->rcv_nxt += tlen;
3165 			if (tlen &&
3166 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3167 			    (tp->t_fbyte_in == 0)) {
3168 				tp->t_fbyte_in = ticks;
3169 				if (tp->t_fbyte_in == 0)
3170 					tp->t_fbyte_in = 1;
3171 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3172 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3173 			}
3174 			thflags = th->th_flags & TH_FIN;
3175 			TCPSTAT_INC(tcps_rcvpack);
3176 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3177 			SOCKBUF_LOCK(&so->so_rcv);
3178 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3179 				m_freem(m);
3180 			else
3181 				sbappendstream_locked(&so->so_rcv, m, 0);
3182 			tp->t_flags |= TF_WAKESOR;
3183 		} else {
3184 			/*
3185 			 * XXX: Due to the header drop above "th" is
3186 			 * theoretically invalid by now.  Fortunately
3187 			 * m_adj() doesn't actually frees any mbufs
3188 			 * when trimming from the head.
3189 			 */
3190 			tcp_seq temp = save_start;
3191 
3192 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3193 			tp->t_flags |= TF_ACKNOW;
3194 		}
3195 		if ((tp->t_flags & TF_SACK_PERMIT) &&
3196 		    (save_tlen > 0) &&
3197 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
3198 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3199 				/*
3200 				 * DSACK actually handled in the fastpath
3201 				 * above.
3202 				 */
3203 				tcp_update_sack_list(tp, save_start,
3204 				    save_start + save_tlen);
3205 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3206 				if ((tp->rcv_numsacks >= 1) &&
3207 				    (tp->sackblks[0].end == save_start)) {
3208 					/*
3209 					 * Partial overlap, recorded at todrop
3210 					 * above.
3211 					 */
3212 					tcp_update_sack_list(tp,
3213 					    tp->sackblks[0].start,
3214 					    tp->sackblks[0].end);
3215 				} else {
3216 					tcp_update_dsack_list(tp, save_start,
3217 					    save_start + save_tlen);
3218 				}
3219 			} else if (tlen >= save_tlen) {
3220 				/* Update of sackblks. */
3221 				tcp_update_dsack_list(tp, save_start,
3222 				    save_start + save_tlen);
3223 			} else if (tlen > 0) {
3224 				tcp_update_dsack_list(tp, save_start,
3225 				    save_start + tlen);
3226 			}
3227 		}
3228 		tcp_handle_wakeup(tp, so);
3229 #if 0
3230 		/*
3231 		 * Note the amount of data that peer has sent into
3232 		 * our window, in order to estimate the sender's
3233 		 * buffer size.
3234 		 * XXX: Unused.
3235 		 */
3236 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3237 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3238 		else
3239 			len = so->so_rcv.sb_hiwat;
3240 #endif
3241 	} else {
3242 		m_freem(m);
3243 		thflags &= ~TH_FIN;
3244 	}
3245 
3246 	/*
3247 	 * If FIN is received ACK the FIN and let the user know
3248 	 * that the connection is closing.
3249 	 */
3250 	if (thflags & TH_FIN) {
3251 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3252 			/* The socket upcall is handled by socantrcvmore. */
3253 			socantrcvmore(so);
3254 			/*
3255 			 * If connection is half-synchronized
3256 			 * (ie NEEDSYN flag on) then delay ACK,
3257 			 * so it may be piggybacked when SYN is sent.
3258 			 * Otherwise, since we received a FIN then no
3259 			 * more input can be expected, send ACK now.
3260 			 */
3261 			if (tp->t_flags & TF_NEEDSYN)
3262 				tp->t_flags |= TF_DELACK;
3263 			else
3264 				tp->t_flags |= TF_ACKNOW;
3265 			tp->rcv_nxt++;
3266 		}
3267 		switch (tp->t_state) {
3268 		/*
3269 		 * In SYN_RECEIVED and ESTABLISHED STATES
3270 		 * enter the CLOSE_WAIT state.
3271 		 */
3272 		case TCPS_SYN_RECEIVED:
3273 			tp->t_starttime = ticks;
3274 			/* FALLTHROUGH */
3275 		case TCPS_ESTABLISHED:
3276 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3277 			break;
3278 
3279 		/*
3280 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3281 		 * enter the CLOSING state.
3282 		 */
3283 		case TCPS_FIN_WAIT_1:
3284 			tcp_state_change(tp, TCPS_CLOSING);
3285 			break;
3286 
3287 		/*
3288 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3289 		 * starting the time-wait timer, turning off the other
3290 		 * standard timers.
3291 		 */
3292 		case TCPS_FIN_WAIT_2:
3293 			tcp_twstart(tp);
3294 			return;
3295 		}
3296 	}
3297 #ifdef TCPDEBUG
3298 	if (so->so_options & SO_DEBUG)
3299 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3300 			  &tcp_savetcp, 0);
3301 #endif
3302 	TCP_PROBE3(debug__input, tp, th, m);
3303 
3304 	/*
3305 	 * Return any desired output.
3306 	 */
3307 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3308 		(void) tcp_output(tp);
3309 
3310 check_delack:
3311 	INP_WLOCK_ASSERT(tp->t_inpcb);
3312 
3313 	if (tp->t_flags & TF_DELACK) {
3314 		tp->t_flags &= ~TF_DELACK;
3315 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3316 	}
3317 	INP_WUNLOCK(tp->t_inpcb);
3318 	return;
3319 
3320 dropafterack:
3321 	/*
3322 	 * Generate an ACK dropping incoming segment if it occupies
3323 	 * sequence space, where the ACK reflects our state.
3324 	 *
3325 	 * We can now skip the test for the RST flag since all
3326 	 * paths to this code happen after packets containing
3327 	 * RST have been dropped.
3328 	 *
3329 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3330 	 * segment we received passes the SYN-RECEIVED ACK test.
3331 	 * If it fails send a RST.  This breaks the loop in the
3332 	 * "LAND" DoS attack, and also prevents an ACK storm
3333 	 * between two listening ports that have been sent forged
3334 	 * SYN segments, each with the source address of the other.
3335 	 */
3336 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3337 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3338 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3339 		rstreason = BANDLIM_RST_OPENPORT;
3340 		goto dropwithreset;
3341 	}
3342 #ifdef TCPDEBUG
3343 	if (so->so_options & SO_DEBUG)
3344 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3345 			  &tcp_savetcp, 0);
3346 #endif
3347 	TCP_PROBE3(debug__input, tp, th, m);
3348 	tp->t_flags |= TF_ACKNOW;
3349 	(void) tcp_output(tp);
3350 	INP_WUNLOCK(tp->t_inpcb);
3351 	m_freem(m);
3352 	return;
3353 
3354 dropwithreset:
3355 	if (tp != NULL) {
3356 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3357 		INP_WUNLOCK(tp->t_inpcb);
3358 	} else
3359 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3360 	return;
3361 
3362 drop:
3363 	/*
3364 	 * Drop space held by incoming segment and return.
3365 	 */
3366 #ifdef TCPDEBUG
3367 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3368 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3369 			  &tcp_savetcp, 0);
3370 #endif
3371 	TCP_PROBE3(debug__input, tp, th, m);
3372 	if (tp != NULL) {
3373 		INP_WUNLOCK(tp->t_inpcb);
3374 	}
3375 	m_freem(m);
3376 }
3377 
3378 /*
3379  * Issue RST and make ACK acceptable to originator of segment.
3380  * The mbuf must still include the original packet header.
3381  * tp may be NULL.
3382  */
3383 void
3384 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3385     int tlen, int rstreason)
3386 {
3387 #ifdef INET
3388 	struct ip *ip;
3389 #endif
3390 #ifdef INET6
3391 	struct ip6_hdr *ip6;
3392 #endif
3393 
3394 	if (tp != NULL) {
3395 		INP_LOCK_ASSERT(tp->t_inpcb);
3396 	}
3397 
3398 	/* Don't bother if destination was broadcast/multicast. */
3399 	if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3400 		goto drop;
3401 #ifdef INET6
3402 	if (mtod(m, struct ip *)->ip_v == 6) {
3403 		ip6 = mtod(m, struct ip6_hdr *);
3404 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3405 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3406 			goto drop;
3407 		/* IPv6 anycast check is done at tcp6_input() */
3408 	}
3409 #endif
3410 #if defined(INET) && defined(INET6)
3411 	else
3412 #endif
3413 #ifdef INET
3414 	{
3415 		ip = mtod(m, struct ip *);
3416 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3417 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3418 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3419 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3420 			goto drop;
3421 	}
3422 #endif
3423 
3424 	/* Perform bandwidth limiting. */
3425 	if (badport_bandlim(rstreason) < 0)
3426 		goto drop;
3427 
3428 	/* tcp_respond consumes the mbuf chain. */
3429 	if (th->th_flags & TH_ACK) {
3430 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3431 		    th->th_ack, TH_RST);
3432 	} else {
3433 		if (th->th_flags & TH_SYN)
3434 			tlen++;
3435 		if (th->th_flags & TH_FIN)
3436 			tlen++;
3437 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3438 		    (tcp_seq)0, TH_RST|TH_ACK);
3439 	}
3440 	return;
3441 drop:
3442 	m_freem(m);
3443 }
3444 
3445 /*
3446  * Parse TCP options and place in tcpopt.
3447  */
3448 void
3449 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3450 {
3451 	int opt, optlen;
3452 
3453 	to->to_flags = 0;
3454 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3455 		opt = cp[0];
3456 		if (opt == TCPOPT_EOL)
3457 			break;
3458 		if (opt == TCPOPT_NOP)
3459 			optlen = 1;
3460 		else {
3461 			if (cnt < 2)
3462 				break;
3463 			optlen = cp[1];
3464 			if (optlen < 2 || optlen > cnt)
3465 				break;
3466 		}
3467 		switch (opt) {
3468 		case TCPOPT_MAXSEG:
3469 			if (optlen != TCPOLEN_MAXSEG)
3470 				continue;
3471 			if (!(flags & TO_SYN))
3472 				continue;
3473 			to->to_flags |= TOF_MSS;
3474 			bcopy((char *)cp + 2,
3475 			    (char *)&to->to_mss, sizeof(to->to_mss));
3476 			to->to_mss = ntohs(to->to_mss);
3477 			break;
3478 		case TCPOPT_WINDOW:
3479 			if (optlen != TCPOLEN_WINDOW)
3480 				continue;
3481 			if (!(flags & TO_SYN))
3482 				continue;
3483 			to->to_flags |= TOF_SCALE;
3484 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3485 			break;
3486 		case TCPOPT_TIMESTAMP:
3487 			if (optlen != TCPOLEN_TIMESTAMP)
3488 				continue;
3489 			to->to_flags |= TOF_TS;
3490 			bcopy((char *)cp + 2,
3491 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3492 			to->to_tsval = ntohl(to->to_tsval);
3493 			bcopy((char *)cp + 6,
3494 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3495 			to->to_tsecr = ntohl(to->to_tsecr);
3496 			break;
3497 		case TCPOPT_SIGNATURE:
3498 			/*
3499 			 * In order to reply to a host which has set the
3500 			 * TCP_SIGNATURE option in its initial SYN, we have
3501 			 * to record the fact that the option was observed
3502 			 * here for the syncache code to perform the correct
3503 			 * response.
3504 			 */
3505 			if (optlen != TCPOLEN_SIGNATURE)
3506 				continue;
3507 			to->to_flags |= TOF_SIGNATURE;
3508 			to->to_signature = cp + 2;
3509 			break;
3510 		case TCPOPT_SACK_PERMITTED:
3511 			if (optlen != TCPOLEN_SACK_PERMITTED)
3512 				continue;
3513 			if (!(flags & TO_SYN))
3514 				continue;
3515 			if (!V_tcp_do_sack)
3516 				continue;
3517 			to->to_flags |= TOF_SACKPERM;
3518 			break;
3519 		case TCPOPT_SACK:
3520 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3521 				continue;
3522 			if (flags & TO_SYN)
3523 				continue;
3524 			to->to_flags |= TOF_SACK;
3525 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3526 			to->to_sacks = cp + 2;
3527 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3528 			break;
3529 		case TCPOPT_FAST_OPEN:
3530 			/*
3531 			 * Cookie length validation is performed by the
3532 			 * server side cookie checking code or the client
3533 			 * side cookie cache update code.
3534 			 */
3535 			if (!(flags & TO_SYN))
3536 				continue;
3537 			if (!V_tcp_fastopen_client_enable &&
3538 			    !V_tcp_fastopen_server_enable)
3539 				continue;
3540 			to->to_flags |= TOF_FASTOPEN;
3541 			to->to_tfo_len = optlen - 2;
3542 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3543 			break;
3544 		default:
3545 			continue;
3546 		}
3547 	}
3548 }
3549 
3550 /*
3551  * Pull out of band byte out of a segment so
3552  * it doesn't appear in the user's data queue.
3553  * It is still reflected in the segment length for
3554  * sequencing purposes.
3555  */
3556 void
3557 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3558     int off)
3559 {
3560 	int cnt = off + th->th_urp - 1;
3561 
3562 	while (cnt >= 0) {
3563 		if (m->m_len > cnt) {
3564 			char *cp = mtod(m, caddr_t) + cnt;
3565 			struct tcpcb *tp = sototcpcb(so);
3566 
3567 			INP_WLOCK_ASSERT(tp->t_inpcb);
3568 
3569 			tp->t_iobc = *cp;
3570 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3571 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3572 			m->m_len--;
3573 			if (m->m_flags & M_PKTHDR)
3574 				m->m_pkthdr.len--;
3575 			return;
3576 		}
3577 		cnt -= m->m_len;
3578 		m = m->m_next;
3579 		if (m == NULL)
3580 			break;
3581 	}
3582 	panic("tcp_pulloutofband");
3583 }
3584 
3585 /*
3586  * Collect new round-trip time estimate
3587  * and update averages and current timeout.
3588  */
3589 void
3590 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3591 {
3592 	int delta;
3593 
3594 	INP_WLOCK_ASSERT(tp->t_inpcb);
3595 
3596 	TCPSTAT_INC(tcps_rttupdated);
3597 	tp->t_rttupdated++;
3598 #ifdef STATS
3599 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3600 	    imax(0, rtt * 1000 / hz));
3601 #endif
3602 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3603 		/*
3604 		 * srtt is stored as fixed point with 5 bits after the
3605 		 * binary point (i.e., scaled by 8).  The following magic
3606 		 * is equivalent to the smoothing algorithm in rfc793 with
3607 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3608 		 * point).  Adjust rtt to origin 0.
3609 		 */
3610 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3611 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3612 
3613 		if ((tp->t_srtt += delta) <= 0)
3614 			tp->t_srtt = 1;
3615 
3616 		/*
3617 		 * We accumulate a smoothed rtt variance (actually, a
3618 		 * smoothed mean difference), then set the retransmit
3619 		 * timer to smoothed rtt + 4 times the smoothed variance.
3620 		 * rttvar is stored as fixed point with 4 bits after the
3621 		 * binary point (scaled by 16).  The following is
3622 		 * equivalent to rfc793 smoothing with an alpha of .75
3623 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3624 		 * rfc793's wired-in beta.
3625 		 */
3626 		if (delta < 0)
3627 			delta = -delta;
3628 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3629 		if ((tp->t_rttvar += delta) <= 0)
3630 			tp->t_rttvar = 1;
3631 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3632 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3633 	} else {
3634 		/*
3635 		 * No rtt measurement yet - use the unsmoothed rtt.
3636 		 * Set the variance to half the rtt (so our first
3637 		 * retransmit happens at 3*rtt).
3638 		 */
3639 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3640 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3641 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3642 	}
3643 	tp->t_rtttime = 0;
3644 	tp->t_rxtshift = 0;
3645 
3646 	/*
3647 	 * the retransmit should happen at rtt + 4 * rttvar.
3648 	 * Because of the way we do the smoothing, srtt and rttvar
3649 	 * will each average +1/2 tick of bias.  When we compute
3650 	 * the retransmit timer, we want 1/2 tick of rounding and
3651 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3652 	 * firing of the timer.  The bias will give us exactly the
3653 	 * 1.5 tick we need.  But, because the bias is
3654 	 * statistical, we have to test that we don't drop below
3655 	 * the minimum feasible timer (which is 2 ticks).
3656 	 */
3657 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3658 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3659 
3660 	/*
3661 	 * We received an ack for a packet that wasn't retransmitted;
3662 	 * it is probably safe to discard any error indications we've
3663 	 * received recently.  This isn't quite right, but close enough
3664 	 * for now (a route might have failed after we sent a segment,
3665 	 * and the return path might not be symmetrical).
3666 	 */
3667 	tp->t_softerror = 0;
3668 }
3669 
3670 /*
3671  * Determine a reasonable value for maxseg size.
3672  * If the route is known, check route for mtu.
3673  * If none, use an mss that can be handled on the outgoing interface
3674  * without forcing IP to fragment.  If no route is found, route has no mtu,
3675  * or the destination isn't local, use a default, hopefully conservative
3676  * size (usually 512 or the default IP max size, but no more than the mtu
3677  * of the interface), as we can't discover anything about intervening
3678  * gateways or networks.  We also initialize the congestion/slow start
3679  * window to be a single segment if the destination isn't local.
3680  * While looking at the routing entry, we also initialize other path-dependent
3681  * parameters from pre-set or cached values in the routing entry.
3682  *
3683  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3684  * IP options, e.g. IPSEC data, since length of this data may vary, and
3685  * thus it is calculated for every segment separately in tcp_output().
3686  *
3687  * NOTE that this routine is only called when we process an incoming
3688  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3689  * settings are handled in tcp_mssopt().
3690  */
3691 void
3692 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3693     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3694 {
3695 	int mss = 0;
3696 	uint32_t maxmtu = 0;
3697 	struct inpcb *inp = tp->t_inpcb;
3698 	struct hc_metrics_lite metrics;
3699 #ifdef INET6
3700 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3701 	size_t min_protoh = isipv6 ?
3702 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3703 			    sizeof (struct tcpiphdr);
3704 #else
3705 	 size_t min_protoh = sizeof(struct tcpiphdr);
3706 #endif
3707 
3708 	INP_WLOCK_ASSERT(tp->t_inpcb);
3709 
3710 	if (tp->t_port)
3711 		min_protoh += V_tcp_udp_tunneling_overhead;
3712 	if (mtuoffer != -1) {
3713 		KASSERT(offer == -1, ("%s: conflict", __func__));
3714 		offer = mtuoffer - min_protoh;
3715 	}
3716 
3717 	/* Initialize. */
3718 #ifdef INET6
3719 	if (isipv6) {
3720 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3721 		tp->t_maxseg = V_tcp_v6mssdflt;
3722 	}
3723 #endif
3724 #if defined(INET) && defined(INET6)
3725 	else
3726 #endif
3727 #ifdef INET
3728 	{
3729 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3730 		tp->t_maxseg = V_tcp_mssdflt;
3731 	}
3732 #endif
3733 
3734 	/*
3735 	 * No route to sender, stay with default mss and return.
3736 	 */
3737 	if (maxmtu == 0) {
3738 		/*
3739 		 * In case we return early we need to initialize metrics
3740 		 * to a defined state as tcp_hc_get() would do for us
3741 		 * if there was no cache hit.
3742 		 */
3743 		if (metricptr != NULL)
3744 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3745 		return;
3746 	}
3747 
3748 	/* What have we got? */
3749 	switch (offer) {
3750 		case 0:
3751 			/*
3752 			 * Offer == 0 means that there was no MSS on the SYN
3753 			 * segment, in this case we use tcp_mssdflt as
3754 			 * already assigned to t_maxseg above.
3755 			 */
3756 			offer = tp->t_maxseg;
3757 			break;
3758 
3759 		case -1:
3760 			/*
3761 			 * Offer == -1 means that we didn't receive SYN yet.
3762 			 */
3763 			/* FALLTHROUGH */
3764 
3765 		default:
3766 			/*
3767 			 * Prevent DoS attack with too small MSS. Round up
3768 			 * to at least minmss.
3769 			 */
3770 			offer = max(offer, V_tcp_minmss);
3771 	}
3772 
3773 	/*
3774 	 * rmx information is now retrieved from tcp_hostcache.
3775 	 */
3776 	tcp_hc_get(&inp->inp_inc, &metrics);
3777 	if (metricptr != NULL)
3778 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3779 
3780 	/*
3781 	 * If there's a discovered mtu in tcp hostcache, use it.
3782 	 * Else, use the link mtu.
3783 	 */
3784 	if (metrics.rmx_mtu)
3785 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3786 	else {
3787 #ifdef INET6
3788 		if (isipv6) {
3789 			mss = maxmtu - min_protoh;
3790 			if (!V_path_mtu_discovery &&
3791 			    !in6_localaddr(&inp->in6p_faddr))
3792 				mss = min(mss, V_tcp_v6mssdflt);
3793 		}
3794 #endif
3795 #if defined(INET) && defined(INET6)
3796 		else
3797 #endif
3798 #ifdef INET
3799 		{
3800 			mss = maxmtu - min_protoh;
3801 			if (!V_path_mtu_discovery &&
3802 			    !in_localaddr(inp->inp_faddr))
3803 				mss = min(mss, V_tcp_mssdflt);
3804 		}
3805 #endif
3806 		/*
3807 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3808 		 * probably violates the TCP spec.
3809 		 * The problem is that, since we don't know the
3810 		 * other end's MSS, we are supposed to use a conservative
3811 		 * default.  But, if we do that, then MTU discovery will
3812 		 * never actually take place, because the conservative
3813 		 * default is much less than the MTUs typically seen
3814 		 * on the Internet today.  For the moment, we'll sweep
3815 		 * this under the carpet.
3816 		 *
3817 		 * The conservative default might not actually be a problem
3818 		 * if the only case this occurs is when sending an initial
3819 		 * SYN with options and data to a host we've never talked
3820 		 * to before.  Then, they will reply with an MSS value which
3821 		 * will get recorded and the new parameters should get
3822 		 * recomputed.  For Further Study.
3823 		 */
3824 	}
3825 	mss = min(mss, offer);
3826 
3827 	/*
3828 	 * Sanity check: make sure that maxseg will be large
3829 	 * enough to allow some data on segments even if the
3830 	 * all the option space is used (40bytes).  Otherwise
3831 	 * funny things may happen in tcp_output.
3832 	 *
3833 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3834 	 */
3835 	mss = max(mss, 64);
3836 
3837 	tp->t_maxseg = mss;
3838 }
3839 
3840 void
3841 tcp_mss(struct tcpcb *tp, int offer)
3842 {
3843 	int mss;
3844 	uint32_t bufsize;
3845 	struct inpcb *inp;
3846 	struct socket *so;
3847 	struct hc_metrics_lite metrics;
3848 	struct tcp_ifcap cap;
3849 
3850 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3851 
3852 	bzero(&cap, sizeof(cap));
3853 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3854 
3855 	mss = tp->t_maxseg;
3856 	inp = tp->t_inpcb;
3857 
3858 	/*
3859 	 * If there's a pipesize, change the socket buffer to that size,
3860 	 * don't change if sb_hiwat is different than default (then it
3861 	 * has been changed on purpose with setsockopt).
3862 	 * Make the socket buffers an integral number of mss units;
3863 	 * if the mss is larger than the socket buffer, decrease the mss.
3864 	 */
3865 	so = inp->inp_socket;
3866 	SOCKBUF_LOCK(&so->so_snd);
3867 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3868 		bufsize = metrics.rmx_sendpipe;
3869 	else
3870 		bufsize = so->so_snd.sb_hiwat;
3871 	if (bufsize < mss)
3872 		mss = bufsize;
3873 	else {
3874 		bufsize = roundup(bufsize, mss);
3875 		if (bufsize > sb_max)
3876 			bufsize = sb_max;
3877 		if (bufsize > so->so_snd.sb_hiwat)
3878 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3879 	}
3880 	SOCKBUF_UNLOCK(&so->so_snd);
3881 	/*
3882 	 * Sanity check: make sure that maxseg will be large
3883 	 * enough to allow some data on segments even if the
3884 	 * all the option space is used (40bytes).  Otherwise
3885 	 * funny things may happen in tcp_output.
3886 	 *
3887 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3888 	 */
3889 	tp->t_maxseg = max(mss, 64);
3890 
3891 	SOCKBUF_LOCK(&so->so_rcv);
3892 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3893 		bufsize = metrics.rmx_recvpipe;
3894 	else
3895 		bufsize = so->so_rcv.sb_hiwat;
3896 	if (bufsize > mss) {
3897 		bufsize = roundup(bufsize, mss);
3898 		if (bufsize > sb_max)
3899 			bufsize = sb_max;
3900 		if (bufsize > so->so_rcv.sb_hiwat)
3901 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3902 	}
3903 	SOCKBUF_UNLOCK(&so->so_rcv);
3904 
3905 	/* Check the interface for TSO capabilities. */
3906 	if (cap.ifcap & CSUM_TSO) {
3907 		tp->t_flags |= TF_TSO;
3908 		tp->t_tsomax = cap.tsomax;
3909 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3910 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3911 	}
3912 }
3913 
3914 /*
3915  * Determine the MSS option to send on an outgoing SYN.
3916  */
3917 int
3918 tcp_mssopt(struct in_conninfo *inc)
3919 {
3920 	int mss = 0;
3921 	uint32_t thcmtu = 0;
3922 	uint32_t maxmtu = 0;
3923 	size_t min_protoh;
3924 
3925 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3926 
3927 #ifdef INET6
3928 	if (inc->inc_flags & INC_ISIPV6) {
3929 		mss = V_tcp_v6mssdflt;
3930 		maxmtu = tcp_maxmtu6(inc, NULL);
3931 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3932 	}
3933 #endif
3934 #if defined(INET) && defined(INET6)
3935 	else
3936 #endif
3937 #ifdef INET
3938 	{
3939 		mss = V_tcp_mssdflt;
3940 		maxmtu = tcp_maxmtu(inc, NULL);
3941 		min_protoh = sizeof(struct tcpiphdr);
3942 	}
3943 #endif
3944 #if defined(INET6) || defined(INET)
3945 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3946 #endif
3947 
3948 	if (maxmtu && thcmtu)
3949 		mss = min(maxmtu, thcmtu) - min_protoh;
3950 	else if (maxmtu || thcmtu)
3951 		mss = max(maxmtu, thcmtu) - min_protoh;
3952 
3953 	return (mss);
3954 }
3955 
3956 void
3957 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3958 {
3959 	int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3960 	int maxseg = tcp_maxseg(tp);
3961 
3962 	INP_WLOCK_ASSERT(tp->t_inpcb);
3963 
3964 	/*
3965 	 * Compute the amount of data that this ACK is indicating
3966 	 * (del_data) and an estimate of how many bytes are in the
3967 	 * network.
3968 	 */
3969 	if (((tp->t_flags & TF_SACK_PERMIT) &&
3970 	    (to->to_flags & TOF_SACK)) ||
3971 	    (IN_CONGRECOVERY(tp->t_flags) &&
3972 	     !IN_FASTRECOVERY(tp->t_flags))) {
3973 		del_data = tp->sackhint.delivered_data;
3974 		if (V_tcp_do_newsack)
3975 			pipe = tcp_compute_pipe(tp);
3976 		else
3977 			pipe = (tp->snd_nxt - tp->snd_fack) +
3978 				tp->sackhint.sack_bytes_rexmit;
3979 	} else {
3980 		if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3981 					     tp->snd_recover - tp->snd_una))
3982 			del_data = maxseg;
3983 		pipe = imax(0, tp->snd_max - tp->snd_una -
3984 			    imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3985 	}
3986 	tp->sackhint.prr_delivered += del_data;
3987 	/*
3988 	 * Proportional Rate Reduction
3989 	 */
3990 	if (pipe >= tp->snd_ssthresh) {
3991 		if (tp->sackhint.recover_fs == 0)
3992 			tp->sackhint.recover_fs =
3993 			    imax(1, tp->snd_nxt - tp->snd_una);
3994 		snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3995 			    tp->snd_ssthresh, tp->sackhint.recover_fs) -
3996 			    tp->sackhint.prr_out;
3997 	} else {
3998 		if (V_tcp_do_prr_conservative || (del_data == 0))
3999 			limit = tp->sackhint.prr_delivered -
4000 				tp->sackhint.prr_out;
4001 		else
4002 			limit = imax(tp->sackhint.prr_delivered -
4003 				    tp->sackhint.prr_out, del_data) +
4004 				    maxseg;
4005 		snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4006 	}
4007 	snd_cnt = imax(snd_cnt, 0) / maxseg;
4008 	/*
4009 	 * Send snd_cnt new data into the network in response to this ack.
4010 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
4011 	 * accordingly.
4012 	 */
4013 	if (IN_FASTRECOVERY(tp->t_flags)) {
4014 		if ((tp->t_flags & TF_SACK_PERMIT) &&
4015 		    (to->to_flags & TOF_SACK)) {
4016 			tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4017 					    tp->sackhint.sack_bytes_rexmit +
4018 					    (snd_cnt * maxseg);
4019 		} else {
4020 			tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4021 					    (snd_cnt * maxseg);
4022 		}
4023 	} else if (IN_CONGRECOVERY(tp->t_flags))
4024 		tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4025 	tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4026 }
4027 
4028 /*
4029  * On a partial ack arrives, force the retransmission of the
4030  * next unacknowledged segment.  Do not clear tp->t_dupacks.
4031  * By setting snd_nxt to ti_ack, this forces retransmission timer to
4032  * be started again.
4033  */
4034 void
4035 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4036 {
4037 	tcp_seq onxt = tp->snd_nxt;
4038 	uint32_t ocwnd = tp->snd_cwnd;
4039 	u_int maxseg = tcp_maxseg(tp);
4040 
4041 	INP_WLOCK_ASSERT(tp->t_inpcb);
4042 
4043 	tcp_timer_activate(tp, TT_REXMT, 0);
4044 	tp->t_rtttime = 0;
4045 	tp->snd_nxt = th->th_ack;
4046 	/*
4047 	 * Set snd_cwnd to one segment beyond acknowledged offset.
4048 	 * (tp->snd_una has not yet been updated when this function is called.)
4049 	 */
4050 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4051 	tp->t_flags |= TF_ACKNOW;
4052 	(void) tcp_output(tp);
4053 	tp->snd_cwnd = ocwnd;
4054 	if (SEQ_GT(onxt, tp->snd_nxt))
4055 		tp->snd_nxt = onxt;
4056 	/*
4057 	 * Partial window deflation.  Relies on fact that tp->snd_una
4058 	 * not updated yet.
4059 	 */
4060 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4061 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4062 	else
4063 		tp->snd_cwnd = 0;
4064 	tp->snd_cwnd += maxseg;
4065 }
4066 
4067 int
4068 tcp_compute_pipe(struct tcpcb *tp)
4069 {
4070 	return (tp->snd_max - tp->snd_una +
4071 		tp->sackhint.sack_bytes_rexmit -
4072 		tp->sackhint.sacked_bytes);
4073 }
4074 
4075 uint32_t
4076 tcp_compute_initwnd(uint32_t maxseg)
4077 {
4078 	/*
4079 	 * Calculate the Initial Window, also used as Restart Window
4080 	 *
4081 	 * RFC5681 Section 3.1 specifies the default conservative values.
4082 	 * RFC3390 specifies slightly more aggressive values.
4083 	 * RFC6928 increases it to ten segments.
4084 	 * Support for user specified value for initial flight size.
4085 	 */
4086 	if (V_tcp_initcwnd_segments)
4087 		return min(V_tcp_initcwnd_segments * maxseg,
4088 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4089 	else if (V_tcp_do_rfc3390)
4090 		return min(4 * maxseg, max(2 * maxseg, 4380));
4091 	else {
4092 		/* Per RFC5681 Section 3.1 */
4093 		if (maxseg > 2190)
4094 			return (2 * maxseg);
4095 		else if (maxseg > 1095)
4096 			return (3 * maxseg);
4097 		else
4098 			return (4 * maxseg);
4099 	}
4100 }
4101