xref: /freebsd/sys/netinet/tcp_input.c (revision 69c5fa5cd1ec9b09ed88a086607a8a0993818db9)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 2007-2008,2010
7  *	Swinburne University of Technology, Melbourne, Australia.
8  * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9  * Copyright (c) 2010 The FreeBSD Foundation
10  * Copyright (c) 2010-2011 Juniper Networks, Inc.
11  * All rights reserved.
12  *
13  * Portions of this software were developed at the Centre for Advanced Internet
14  * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15  * James Healy and David Hayes, made possible in part by a grant from the Cisco
16  * University Research Program Fund at Community Foundation Silicon Valley.
17  *
18  * Portions of this software were developed at the Centre for Advanced
19  * Internet Architectures, Swinburne University of Technology, Melbourne,
20  * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
21  *
22  * Portions of this software were developed by Robert N. M. Watson under
23  * contract to Juniper Networks, Inc.
24  *
25  * Redistribution and use in source and binary forms, with or without
26  * modification, are permitted provided that the following conditions
27  * are met:
28  * 1. Redistributions of source code must retain the above copyright
29  *    notice, this list of conditions and the following disclaimer.
30  * 2. Redistributions in binary form must reproduce the above copyright
31  *    notice, this list of conditions and the following disclaimer in the
32  *    documentation and/or other materials provided with the distribution.
33  * 3. Neither the name of the University nor the names of its contributors
34  *    may be used to endorse or promote products derived from this software
35  *    without specific prior written permission.
36  *
37  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47  * SUCH DAMAGE.
48  *
49  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
50  */
51 
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
54 
55 #include "opt_inet.h"
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
59 
60 #include <sys/param.h>
61 #include <sys/arb.h>
62 #include <sys/kernel.h>
63 #ifdef TCP_HHOOK
64 #include <sys/hhook.h>
65 #endif
66 #include <sys/malloc.h>
67 #include <sys/mbuf.h>
68 #include <sys/proc.h>		/* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
71 #include <sys/sdt.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
79 
80 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
81 
82 #include <vm/uma.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 
89 #define TCPSTATES		/* for logging */
90 
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h>	/* required for icmp_var.h */
97 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
116 #ifdef TCPPCAP
117 #include <netinet/tcp_pcap.h>
118 #endif
119 #include <netinet/tcp_syncache.h>
120 #ifdef TCPDEBUG
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
123 #ifdef TCP_OFFLOAD
124 #include <netinet/tcp_offload.h>
125 #endif
126 
127 #include <netipsec/ipsec_support.h>
128 
129 #include <machine/in_cksum.h>
130 
131 #include <security/mac/mac_framework.h>
132 
133 const int tcprexmtthresh = 3;
134 
135 VNET_DEFINE(int, tcp_log_in_vain) = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
137     &VNET_NAME(tcp_log_in_vain), 0,
138     "Log all incoming TCP segments to closed ports");
139 
140 VNET_DEFINE(int, blackhole) = 0;
141 #define	V_blackhole		VNET(blackhole)
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
143     &VNET_NAME(blackhole), 0,
144     "Do not send RST on segments to closed ports");
145 
146 VNET_DEFINE(int, tcp_delack_enabled) = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
148     &VNET_NAME(tcp_delack_enabled), 0,
149     "Delay ACK to try and piggyback it onto a data packet");
150 
151 VNET_DEFINE(int, drop_synfin) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153     &VNET_NAME(drop_synfin), 0,
154     "Drop TCP packets with SYN+FIN set");
155 
156 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
158     &VNET_NAME(tcp_do_prr_conservative), 0,
159     "Do conservative Proportional Rate Reduction");
160 
161 VNET_DEFINE(int, tcp_do_prr) = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
163     &VNET_NAME(tcp_do_prr), 1,
164     "Enable Proportional Rate Reduction per RFC 6937");
165 
166 VNET_DEFINE(int, tcp_do_newcwv) = 0;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
168     &VNET_NAME(tcp_do_newcwv), 0,
169     "Enable New Congestion Window Validation per RFC7661");
170 
171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
173     &VNET_NAME(tcp_do_rfc6675_pipe), 0,
174     "Use calculated pipe/in-flight bytes per RFC 6675");
175 
176 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
178     &VNET_NAME(tcp_do_rfc3042), 0,
179     "Enable RFC 3042 (Limited Transmit)");
180 
181 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
183     &VNET_NAME(tcp_do_rfc3390), 0,
184     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
185 
186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
188     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
189     "Slow-start flight size (initial congestion window) in number of segments");
190 
191 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
193     &VNET_NAME(tcp_do_rfc3465), 0,
194     "Enable RFC 3465 (Appropriate Byte Counting)");
195 
196 VNET_DEFINE(int, tcp_abc_l_var) = 2;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
198     &VNET_NAME(tcp_abc_l_var), 2,
199     "Cap the max cwnd increment during slow-start to this number of segments");
200 
201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
202     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
203     "TCP ECN");
204 
205 VNET_DEFINE(int, tcp_do_ecn) = 2;
206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
207     &VNET_NAME(tcp_do_ecn), 0,
208     "TCP ECN support");
209 
210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
212     &VNET_NAME(tcp_ecn_maxretries), 0,
213     "Max retries before giving up on ECN");
214 
215 VNET_DEFINE(int, tcp_insecure_syn) = 0;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
217     &VNET_NAME(tcp_insecure_syn), 0,
218     "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
219 
220 VNET_DEFINE(int, tcp_insecure_rst) = 0;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
222     &VNET_NAME(tcp_insecure_rst), 0,
223     "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
224 
225 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
226 #define	V_tcp_recvspace	VNET(tcp_recvspace)
227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
228     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
229 
230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
232     &VNET_NAME(tcp_do_autorcvbuf), 0,
233     "Enable automatic receive buffer sizing");
234 
235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
237     &VNET_NAME(tcp_autorcvbuf_max), 0,
238     "Max size of automatic receive buffer");
239 
240 VNET_DEFINE(struct inpcbhead, tcb);
241 #define	tcb6	tcb  /* for KAME src sync over BSD*'s */
242 VNET_DEFINE(struct inpcbinfo, tcbinfo);
243 
244 /*
245  * TCP statistics are stored in an array of counter(9)s, which size matches
246  * size of struct tcpstat.  TCP running connection count is a regular array.
247  */
248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
250     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
253     CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
254     "TCP connection counts by TCP state");
255 
256 static void
257 tcp_vnet_init(const void *unused)
258 {
259 
260 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
261 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
262 }
263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
264     tcp_vnet_init, NULL);
265 
266 #ifdef VIMAGE
267 static void
268 tcp_vnet_uninit(const void *unused)
269 {
270 
271 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
272 	VNET_PCPUSTAT_FREE(tcpstat);
273 }
274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275     tcp_vnet_uninit, NULL);
276 #endif /* VIMAGE */
277 
278 /*
279  * Kernel module interface for updating tcpstat.  The first argument is an index
280  * into tcpstat treated as an array.
281  */
282 void
283 kmod_tcpstat_add(int statnum, int val)
284 {
285 
286 	counter_u64_add(VNET(tcpstat)[statnum], val);
287 }
288 
289 #ifdef TCP_HHOOK
290 /*
291  * Wrapper for the TCP established input helper hook.
292  */
293 void
294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
295 {
296 	struct tcp_hhook_data hhook_data;
297 
298 	if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
299 		hhook_data.tp = tp;
300 		hhook_data.th = th;
301 		hhook_data.to = to;
302 
303 		hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
304 		    tp->osd);
305 	}
306 }
307 #endif
308 
309 /*
310  * CC wrapper hook functions
311  */
312 void
313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
314     uint16_t type)
315 {
316 #ifdef STATS
317 	int32_t gput;
318 #endif
319 
320 	INP_WLOCK_ASSERT(tp->t_inpcb);
321 
322 	tp->ccv->nsegs = nsegs;
323 	tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
324 	if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
325 	    (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
326 	     (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
327 		tp->ccv->flags |= CCF_CWND_LIMITED;
328 	else
329 		tp->ccv->flags &= ~CCF_CWND_LIMITED;
330 
331 	if (type == CC_ACK) {
332 #ifdef STATS
333 		stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
334 		    ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
335 		if (!IN_RECOVERY(tp->t_flags))
336 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
337 			   tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
338 		if ((tp->t_flags & TF_GPUTINPROG) &&
339 		    SEQ_GEQ(th->th_ack, tp->gput_ack)) {
340 			/*
341 			 * Compute goodput in bits per millisecond.
342 			 */
343 			gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
344 			    max(1, tcp_ts_getticks() - tp->gput_ts);
345 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
346 			    gput);
347 			/*
348 			 * XXXLAS: This is a temporary hack, and should be
349 			 * chained off VOI_TCP_GPUT when stats(9) grows an API
350 			 * to deal with chained VOIs.
351 			 */
352 			if (tp->t_stats_gput_prev > 0)
353 				stats_voi_update_abs_s32(tp->t_stats,
354 				    VOI_TCP_GPUT_ND,
355 				    ((gput - tp->t_stats_gput_prev) * 100) /
356 				    tp->t_stats_gput_prev);
357 			tp->t_flags &= ~TF_GPUTINPROG;
358 			tp->t_stats_gput_prev = gput;
359 		}
360 #endif /* STATS */
361 		if (tp->snd_cwnd > tp->snd_ssthresh) {
362 			tp->t_bytes_acked += tp->ccv->bytes_this_ack;
363 			if (tp->t_bytes_acked >= tp->snd_cwnd) {
364 				tp->t_bytes_acked -= tp->snd_cwnd;
365 				tp->ccv->flags |= CCF_ABC_SENTAWND;
366 			}
367 		} else {
368 				tp->ccv->flags &= ~CCF_ABC_SENTAWND;
369 				tp->t_bytes_acked = 0;
370 		}
371 	}
372 
373 	if (CC_ALGO(tp)->ack_received != NULL) {
374 		/* XXXLAS: Find a way to live without this */
375 		tp->ccv->curack = th->th_ack;
376 		CC_ALGO(tp)->ack_received(tp->ccv, type);
377 	}
378 #ifdef STATS
379 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
380 #endif
381 }
382 
383 void
384 cc_conn_init(struct tcpcb *tp)
385 {
386 	struct hc_metrics_lite metrics;
387 	struct inpcb *inp = tp->t_inpcb;
388 	u_int maxseg;
389 	int rtt;
390 
391 	INP_WLOCK_ASSERT(tp->t_inpcb);
392 
393 	tcp_hc_get(&inp->inp_inc, &metrics);
394 	maxseg = tcp_maxseg(tp);
395 
396 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
397 		tp->t_srtt = rtt;
398 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
399 		TCPSTAT_INC(tcps_usedrtt);
400 		if (metrics.rmx_rttvar) {
401 			tp->t_rttvar = metrics.rmx_rttvar;
402 			TCPSTAT_INC(tcps_usedrttvar);
403 		} else {
404 			/* default variation is +- 1 rtt */
405 			tp->t_rttvar =
406 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
407 		}
408 		TCPT_RANGESET(tp->t_rxtcur,
409 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
410 		    tp->t_rttmin, TCPTV_REXMTMAX);
411 	}
412 	if (metrics.rmx_ssthresh) {
413 		/*
414 		 * There's some sort of gateway or interface
415 		 * buffer limit on the path.  Use this to set
416 		 * the slow start threshold, but set the
417 		 * threshold to no less than 2*mss.
418 		 */
419 		tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
420 		TCPSTAT_INC(tcps_usedssthresh);
421 	}
422 
423 	/*
424 	 * Set the initial slow-start flight size.
425 	 *
426 	 * If a SYN or SYN/ACK was lost and retransmitted, we have to
427 	 * reduce the initial CWND to one segment as congestion is likely
428 	 * requiring us to be cautious.
429 	 */
430 	if (tp->snd_cwnd == 1)
431 		tp->snd_cwnd = maxseg;		/* SYN(-ACK) lost */
432 	else
433 		tp->snd_cwnd = tcp_compute_initwnd(maxseg);
434 
435 	if (CC_ALGO(tp)->conn_init != NULL)
436 		CC_ALGO(tp)->conn_init(tp->ccv);
437 }
438 
439 void inline
440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
441 {
442 	INP_WLOCK_ASSERT(tp->t_inpcb);
443 
444 #ifdef STATS
445 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
446 #endif
447 
448 	switch(type) {
449 	case CC_NDUPACK:
450 		if (!IN_FASTRECOVERY(tp->t_flags)) {
451 			tp->snd_recover = tp->snd_max;
452 			if (tp->t_flags2 & TF2_ECN_PERMIT)
453 				tp->t_flags2 |= TF2_ECN_SND_CWR;
454 		}
455 		break;
456 	case CC_ECN:
457 		if (!IN_CONGRECOVERY(tp->t_flags) ||
458 		    /*
459 		     * Allow ECN reaction on ACK to CWR, if
460 		     * that data segment was also CE marked.
461 		     */
462 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
463 			EXIT_CONGRECOVERY(tp->t_flags);
464 			TCPSTAT_INC(tcps_ecn_rcwnd);
465 			tp->snd_recover = tp->snd_max + 1;
466 			if (tp->t_flags2 & TF2_ECN_PERMIT)
467 				tp->t_flags2 |= TF2_ECN_SND_CWR;
468 		}
469 		break;
470 	case CC_RTO:
471 		tp->t_dupacks = 0;
472 		tp->t_bytes_acked = 0;
473 		EXIT_RECOVERY(tp->t_flags);
474 		if (tp->t_flags2 & TF2_ECN_PERMIT)
475 			tp->t_flags2 |= TF2_ECN_SND_CWR;
476 		break;
477 	case CC_RTO_ERR:
478 		TCPSTAT_INC(tcps_sndrexmitbad);
479 		/* RTO was unnecessary, so reset everything. */
480 		tp->snd_cwnd = tp->snd_cwnd_prev;
481 		tp->snd_ssthresh = tp->snd_ssthresh_prev;
482 		tp->snd_recover = tp->snd_recover_prev;
483 		if (tp->t_flags & TF_WASFRECOVERY)
484 			ENTER_FASTRECOVERY(tp->t_flags);
485 		if (tp->t_flags & TF_WASCRECOVERY)
486 			ENTER_CONGRECOVERY(tp->t_flags);
487 		tp->snd_nxt = tp->snd_max;
488 		tp->t_flags &= ~TF_PREVVALID;
489 		tp->t_badrxtwin = 0;
490 		break;
491 	}
492 
493 	if (CC_ALGO(tp)->cong_signal != NULL) {
494 		if (th != NULL)
495 			tp->ccv->curack = th->th_ack;
496 		CC_ALGO(tp)->cong_signal(tp->ccv, type);
497 	}
498 }
499 
500 void inline
501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
502 {
503 	INP_WLOCK_ASSERT(tp->t_inpcb);
504 
505 	/* XXXLAS: KASSERT that we're in recovery? */
506 
507 	if (CC_ALGO(tp)->post_recovery != NULL) {
508 		tp->ccv->curack = th->th_ack;
509 		CC_ALGO(tp)->post_recovery(tp->ccv);
510 	}
511 	/* XXXLAS: EXIT_RECOVERY ? */
512 	tp->t_bytes_acked = 0;
513 }
514 
515 /*
516  * Indicate whether this ack should be delayed.  We can delay the ack if
517  * following conditions are met:
518  *	- There is no delayed ack timer in progress.
519  *	- Our last ack wasn't a 0-sized window. We never want to delay
520  *	  the ack that opens up a 0-sized window.
521  *	- LRO wasn't used for this segment. We make sure by checking that the
522  *	  segment size is not larger than the MSS.
523  */
524 #define DELAY_ACK(tp, tlen)						\
525 	((!tcp_timer_active(tp, TT_DELACK) &&				\
526 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
527 	    (tlen <= tp->t_maxseg) &&					\
528 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
529 
530 void inline
531 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
532 {
533 	INP_WLOCK_ASSERT(tp->t_inpcb);
534 
535 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
536 		switch (iptos & IPTOS_ECN_MASK) {
537 		case IPTOS_ECN_CE:
538 			tp->ccv->flags |= CCF_IPHDR_CE;
539 			break;
540 		case IPTOS_ECN_ECT0:
541 			/* FALLTHROUGH */
542 		case IPTOS_ECN_ECT1:
543 			/* FALLTHROUGH */
544 		case IPTOS_ECN_NOTECT:
545 			tp->ccv->flags &= ~CCF_IPHDR_CE;
546 			break;
547 		}
548 
549 		if (th->th_flags & TH_CWR)
550 			tp->ccv->flags |= CCF_TCPHDR_CWR;
551 		else
552 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
553 
554 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
555 
556 		if (tp->ccv->flags & CCF_ACKNOW) {
557 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
558 			tp->t_flags |= TF_ACKNOW;
559 		}
560 	}
561 }
562 
563 /*
564  * TCP input handling is split into multiple parts:
565  *   tcp6_input is a thin wrapper around tcp_input for the extended
566  *	ip6_protox[] call format in ip6_input
567  *   tcp_input handles primary segment validation, inpcb lookup and
568  *	SYN processing on listen sockets
569  *   tcp_do_segment processes the ACK and text of the segment for
570  *	establishing, established and closing connections
571  */
572 #ifdef INET6
573 int
574 tcp6_input(struct mbuf **mp, int *offp, int proto)
575 {
576 	struct mbuf *m;
577 	struct in6_ifaddr *ia6;
578 	struct ip6_hdr *ip6;
579 
580 	m = *mp;
581 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
582 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
583 		if (m == NULL) {
584 			*mp = m;
585 			TCPSTAT_INC(tcps_rcvshort);
586 			return (IPPROTO_DONE);
587 		}
588 	}
589 
590 	/*
591 	 * draft-itojun-ipv6-tcp-to-anycast
592 	 * better place to put this in?
593 	 */
594 	ip6 = mtod(m, struct ip6_hdr *);
595 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
596 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
597 		ifa_free(&ia6->ia_ifa);
598 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
599 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
600 		*mp = NULL;
601 		return (IPPROTO_DONE);
602 	}
603 	if (ia6)
604 		ifa_free(&ia6->ia_ifa);
605 
606 	*mp = m;
607 	return (tcp_input(mp, offp, proto));
608 }
609 #endif /* INET6 */
610 
611 int
612 tcp_input(struct mbuf **mp, int *offp, int proto)
613 {
614 	struct mbuf *m = *mp;
615 	struct tcphdr *th = NULL;
616 	struct ip *ip = NULL;
617 	struct inpcb *inp = NULL;
618 	struct tcpcb *tp = NULL;
619 	struct socket *so = NULL;
620 	u_char *optp = NULL;
621 	int off0;
622 	int optlen = 0;
623 #ifdef INET
624 	int len;
625 	uint8_t ipttl;
626 #endif
627 	int tlen = 0, off;
628 	int drop_hdrlen;
629 	int thflags;
630 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
631 	uint8_t iptos;
632 	struct m_tag *fwd_tag = NULL;
633 #ifdef INET6
634 	struct ip6_hdr *ip6 = NULL;
635 	int isipv6;
636 #else
637 	const void *ip6 = NULL;
638 #endif /* INET6 */
639 	struct tcpopt to;		/* options in this segment */
640 	char *s = NULL;			/* address and port logging */
641 #ifdef TCPDEBUG
642 	/*
643 	 * The size of tcp_saveipgen must be the size of the max ip header,
644 	 * now IPv6.
645 	 */
646 	u_char tcp_saveipgen[IP6_HDR_LEN];
647 	struct tcphdr tcp_savetcp;
648 	short ostate = 0;
649 #endif
650 
651 	NET_EPOCH_ASSERT();
652 
653 #ifdef INET6
654 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
655 #endif
656 
657 	off0 = *offp;
658 	m = *mp;
659 	*mp = NULL;
660 	to.to_flags = 0;
661 	TCPSTAT_INC(tcps_rcvtotal);
662 
663 #ifdef INET6
664 	if (isipv6) {
665 		ip6 = mtod(m, struct ip6_hdr *);
666 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
667 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
668 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
669 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
670 				th->th_sum = m->m_pkthdr.csum_data;
671 			else
672 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
673 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
674 			th->th_sum ^= 0xffff;
675 		} else
676 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
677 		if (th->th_sum) {
678 			TCPSTAT_INC(tcps_rcvbadsum);
679 			goto drop;
680 		}
681 
682 		/*
683 		 * Be proactive about unspecified IPv6 address in source.
684 		 * As we use all-zero to indicate unbounded/unconnected pcb,
685 		 * unspecified IPv6 address can be used to confuse us.
686 		 *
687 		 * Note that packets with unspecified IPv6 destination is
688 		 * already dropped in ip6_input.
689 		 */
690 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
691 			/* XXX stat */
692 			goto drop;
693 		}
694 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
695 	}
696 #endif
697 #if defined(INET) && defined(INET6)
698 	else
699 #endif
700 #ifdef INET
701 	{
702 		/*
703 		 * Get IP and TCP header together in first mbuf.
704 		 * Note: IP leaves IP header in first mbuf.
705 		 */
706 		if (off0 > sizeof (struct ip)) {
707 			ip_stripoptions(m);
708 			off0 = sizeof(struct ip);
709 		}
710 		if (m->m_len < sizeof (struct tcpiphdr)) {
711 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
712 			    == NULL) {
713 				TCPSTAT_INC(tcps_rcvshort);
714 				return (IPPROTO_DONE);
715 			}
716 		}
717 		ip = mtod(m, struct ip *);
718 		th = (struct tcphdr *)((caddr_t)ip + off0);
719 		tlen = ntohs(ip->ip_len) - off0;
720 
721 		iptos = ip->ip_tos;
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 
752 		if (th->th_sum) {
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 findpcb:
831 #ifdef INET6
832 	if (isipv6 && fwd_tag != NULL) {
833 		struct sockaddr_in6 *next_hop6;
834 
835 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
836 		/*
837 		 * Transparently forwarded. Pretend to be the destination.
838 		 * Already got one like this?
839 		 */
840 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
841 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
842 		    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
843 		if (!inp) {
844 			/*
845 			 * It's new.  Try to find the ambushing socket.
846 			 * Because we've rewritten the destination address,
847 			 * any hardware-generated hash is ignored.
848 			 */
849 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
850 			    th->th_sport, &next_hop6->sin6_addr,
851 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
852 			    th->th_dport, INPLOOKUP_WILDCARD |
853 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
854 		}
855 	} else if (isipv6) {
856 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
857 		    th->th_sport, &ip6->ip6_dst, th->th_dport,
858 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
859 		    m->m_pkthdr.rcvif, m);
860 	}
861 #endif /* INET6 */
862 #if defined(INET6) && defined(INET)
863 	else
864 #endif
865 #ifdef INET
866 	if (fwd_tag != NULL) {
867 		struct sockaddr_in *next_hop;
868 
869 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
870 		/*
871 		 * Transparently forwarded. Pretend to be the destination.
872 		 * already got one like this?
873 		 */
874 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
875 		    ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
876 		    m->m_pkthdr.rcvif, m);
877 		if (!inp) {
878 			/*
879 			 * It's new.  Try to find the ambushing socket.
880 			 * Because we've rewritten the destination address,
881 			 * any hardware-generated hash is ignored.
882 			 */
883 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
884 			    th->th_sport, next_hop->sin_addr,
885 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
886 			    th->th_dport, INPLOOKUP_WILDCARD |
887 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
888 		}
889 	} else
890 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
891 		    th->th_sport, ip->ip_dst, th->th_dport,
892 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
893 		    m->m_pkthdr.rcvif, m);
894 #endif /* INET */
895 
896 	/*
897 	 * If the INPCB does not exist then all data in the incoming
898 	 * segment is discarded and an appropriate RST is sent back.
899 	 * XXX MRT Send RST using which routing table?
900 	 */
901 	if (inp == NULL) {
902 		/*
903 		 * Log communication attempts to ports that are not
904 		 * in use.
905 		 */
906 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
907 		    V_tcp_log_in_vain == 2) {
908 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
909 				log(LOG_INFO, "%s; %s: Connection attempt "
910 				    "to closed port\n", s, __func__);
911 		}
912 		/*
913 		 * When blackholing do not respond with a RST but
914 		 * completely ignore the segment and drop it.
915 		 */
916 		if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
917 		    V_blackhole == 2)
918 			goto dropunlock;
919 
920 		rstreason = BANDLIM_RST_CLOSEDPORT;
921 		goto dropwithreset;
922 	}
923 	INP_WLOCK_ASSERT(inp);
924 	/*
925 	 * While waiting for inp lock during the lookup, another thread
926 	 * can have dropped the inpcb, in which case we need to loop back
927 	 * and try to find a new inpcb to deliver to.
928 	 */
929 	if (inp->inp_flags & INP_DROPPED) {
930 		INP_WUNLOCK(inp);
931 		inp = NULL;
932 		goto findpcb;
933 	}
934 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
935 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
936 	    ((inp->inp_socket == NULL) ||
937 	    (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
938 		inp->inp_flowid = m->m_pkthdr.flowid;
939 		inp->inp_flowtype = M_HASHTYPE_GET(m);
940 	}
941 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
942 #ifdef INET6
943 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
944 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
945 		goto dropunlock;
946 	}
947 #ifdef INET
948 	else
949 #endif
950 #endif /* INET6 */
951 #ifdef INET
952 	if (IPSEC_ENABLED(ipv4) &&
953 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
954 		goto dropunlock;
955 	}
956 #endif /* INET */
957 #endif /* IPSEC */
958 
959 	/*
960 	 * Check the minimum TTL for socket.
961 	 */
962 	if (inp->inp_ip_minttl != 0) {
963 #ifdef INET6
964 		if (isipv6) {
965 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
966 				goto dropunlock;
967 		} else
968 #endif
969 		if (inp->inp_ip_minttl > ip->ip_ttl)
970 			goto dropunlock;
971 	}
972 
973 	/*
974 	 * A previous connection in TIMEWAIT state is supposed to catch stray
975 	 * or duplicate segments arriving late.  If this segment was a
976 	 * legitimate new connection attempt, the old INPCB gets removed and
977 	 * we can try again to find a listening socket.
978 	 *
979 	 * At this point, due to earlier optimism, we may hold only an inpcb
980 	 * lock, and not the inpcbinfo write lock.  If so, we need to try to
981 	 * acquire it, or if that fails, acquire a reference on the inpcb,
982 	 * drop all locks, acquire a global write lock, and then re-acquire
983 	 * the inpcb lock.  We may at that point discover that another thread
984 	 * has tried to free the inpcb, in which case we need to loop back
985 	 * and try to find a new inpcb to deliver to.
986 	 *
987 	 * XXXRW: It may be time to rethink timewait locking.
988 	 */
989 	if (inp->inp_flags & INP_TIMEWAIT) {
990 		tcp_dooptions(&to, optp, optlen,
991 		    (thflags & TH_SYN) ? TO_SYN : 0);
992 		/*
993 		 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
994 		 */
995 		if (tcp_twcheck(inp, &to, th, m, tlen))
996 			goto findpcb;
997 		return (IPPROTO_DONE);
998 	}
999 	/*
1000 	 * The TCPCB may no longer exist if the connection is winding
1001 	 * down or it is in the CLOSED state.  Either way we drop the
1002 	 * segment and send an appropriate response.
1003 	 */
1004 	tp = intotcpcb(inp);
1005 	if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1006 		rstreason = BANDLIM_RST_CLOSEDPORT;
1007 		goto dropwithreset;
1008 	}
1009 
1010 #ifdef TCP_OFFLOAD
1011 	if (tp->t_flags & TF_TOE) {
1012 		tcp_offload_input(tp, m);
1013 		m = NULL;	/* consumed by the TOE driver */
1014 		goto dropunlock;
1015 	}
1016 #endif
1017 
1018 #ifdef MAC
1019 	INP_WLOCK_ASSERT(inp);
1020 	if (mac_inpcb_check_deliver(inp, m))
1021 		goto dropunlock;
1022 #endif
1023 	so = inp->inp_socket;
1024 	KASSERT(so != NULL, ("%s: so == NULL", __func__));
1025 #ifdef TCPDEBUG
1026 	if (so->so_options & SO_DEBUG) {
1027 		ostate = tp->t_state;
1028 #ifdef INET6
1029 		if (isipv6) {
1030 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1031 		} else
1032 #endif
1033 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1034 		tcp_savetcp = *th;
1035 	}
1036 #endif /* TCPDEBUG */
1037 	/*
1038 	 * When the socket is accepting connections (the INPCB is in LISTEN
1039 	 * state) we look into the SYN cache if this is a new connection
1040 	 * attempt or the completion of a previous one.
1041 	 */
1042 	KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1043 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1044 	if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1045 		struct in_conninfo inc;
1046 
1047 		bzero(&inc, sizeof(inc));
1048 #ifdef INET6
1049 		if (isipv6) {
1050 			inc.inc_flags |= INC_ISIPV6;
1051 			if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1052 				inc.inc_flags |= INC_IPV6MINMTU;
1053 			inc.inc6_faddr = ip6->ip6_src;
1054 			inc.inc6_laddr = ip6->ip6_dst;
1055 		} else
1056 #endif
1057 		{
1058 			inc.inc_faddr = ip->ip_src;
1059 			inc.inc_laddr = ip->ip_dst;
1060 		}
1061 		inc.inc_fport = th->th_sport;
1062 		inc.inc_lport = th->th_dport;
1063 		inc.inc_fibnum = so->so_fibnum;
1064 
1065 		/*
1066 		 * Check for an existing connection attempt in syncache if
1067 		 * the flag is only ACK.  A successful lookup creates a new
1068 		 * socket appended to the listen queue in SYN_RECEIVED state.
1069 		 */
1070 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1071 			/*
1072 			 * Parse the TCP options here because
1073 			 * syncookies need access to the reflected
1074 			 * timestamp.
1075 			 */
1076 			tcp_dooptions(&to, optp, optlen, 0);
1077 			/*
1078 			 * NB: syncache_expand() doesn't unlock
1079 			 * inp and tcpinfo locks.
1080 			 */
1081 			rstreason = syncache_expand(&inc, &to, th, &so, m);
1082 			if (rstreason < 0) {
1083 				/*
1084 				 * A failing TCP MD5 signature comparison
1085 				 * must result in the segment being dropped
1086 				 * and must not produce any response back
1087 				 * to the sender.
1088 				 */
1089 				goto dropunlock;
1090 			} else if (rstreason == 0) {
1091 				/*
1092 				 * No syncache entry or ACK was not
1093 				 * for our SYN/ACK.  Send a RST.
1094 				 * NB: syncache did its own logging
1095 				 * of the failure cause.
1096 				 */
1097 				rstreason = BANDLIM_RST_OPENPORT;
1098 				goto dropwithreset;
1099 			}
1100 tfo_socket_result:
1101 			if (so == NULL) {
1102 				/*
1103 				 * We completed the 3-way handshake
1104 				 * but could not allocate a socket
1105 				 * either due to memory shortage,
1106 				 * listen queue length limits or
1107 				 * global socket limits.  Send RST
1108 				 * or wait and have the remote end
1109 				 * retransmit the ACK for another
1110 				 * try.
1111 				 */
1112 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1113 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1114 					    "Socket allocation failed due to "
1115 					    "limits or memory shortage, %s\n",
1116 					    s, __func__,
1117 					    V_tcp_sc_rst_sock_fail ?
1118 					    "sending RST" : "try again");
1119 				if (V_tcp_sc_rst_sock_fail) {
1120 					rstreason = BANDLIM_UNLIMITED;
1121 					goto dropwithreset;
1122 				} else
1123 					goto dropunlock;
1124 			}
1125 			/*
1126 			 * Socket is created in state SYN_RECEIVED.
1127 			 * Unlock the listen socket, lock the newly
1128 			 * created socket and update the tp variable.
1129 			 */
1130 			INP_WUNLOCK(inp);	/* listen socket */
1131 			inp = sotoinpcb(so);
1132 			/*
1133 			 * New connection inpcb is already locked by
1134 			 * syncache_expand().
1135 			 */
1136 			INP_WLOCK_ASSERT(inp);
1137 			tp = intotcpcb(inp);
1138 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1139 			    ("%s: ", __func__));
1140 			/*
1141 			 * Process the segment and the data it
1142 			 * contains.  tcp_do_segment() consumes
1143 			 * the mbuf chain and unlocks the inpcb.
1144 			 */
1145 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1146 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1147 			    iptos);
1148 			return (IPPROTO_DONE);
1149 		}
1150 		/*
1151 		 * Segment flag validation for new connection attempts:
1152 		 *
1153 		 * Our (SYN|ACK) response was rejected.
1154 		 * Check with syncache and remove entry to prevent
1155 		 * retransmits.
1156 		 *
1157 		 * NB: syncache_chkrst does its own logging of failure
1158 		 * causes.
1159 		 */
1160 		if (thflags & TH_RST) {
1161 			syncache_chkrst(&inc, th, m);
1162 			goto dropunlock;
1163 		}
1164 		/*
1165 		 * We can't do anything without SYN.
1166 		 */
1167 		if ((thflags & TH_SYN) == 0) {
1168 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1169 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1170 				    "SYN is missing, segment ignored\n",
1171 				    s, __func__);
1172 			TCPSTAT_INC(tcps_badsyn);
1173 			goto dropunlock;
1174 		}
1175 		/*
1176 		 * (SYN|ACK) is bogus on a listen socket.
1177 		 */
1178 		if (thflags & TH_ACK) {
1179 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1180 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1181 				    "SYN|ACK invalid, segment rejected\n",
1182 				    s, __func__);
1183 			syncache_badack(&inc);	/* XXX: Not needed! */
1184 			TCPSTAT_INC(tcps_badsyn);
1185 			rstreason = BANDLIM_RST_OPENPORT;
1186 			goto dropwithreset;
1187 		}
1188 		/*
1189 		 * If the drop_synfin option is enabled, drop all
1190 		 * segments with both the SYN and FIN bits set.
1191 		 * This prevents e.g. nmap from identifying the
1192 		 * TCP/IP stack.
1193 		 * XXX: Poor reasoning.  nmap has other methods
1194 		 * and is constantly refining its stack detection
1195 		 * strategies.
1196 		 * XXX: This is a violation of the TCP specification
1197 		 * and was used by RFC1644.
1198 		 */
1199 		if ((thflags & TH_FIN) && V_drop_synfin) {
1200 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1201 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1202 				    "SYN|FIN segment ignored (based on "
1203 				    "sysctl setting)\n", s, __func__);
1204 			TCPSTAT_INC(tcps_badsyn);
1205 			goto dropunlock;
1206 		}
1207 		/*
1208 		 * Segment's flags are (SYN) or (SYN|FIN).
1209 		 *
1210 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1211 		 * as they do not affect the state of the TCP FSM.
1212 		 * The data pointed to by TH_URG and th_urp is ignored.
1213 		 */
1214 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1215 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1216 		KASSERT(thflags & (TH_SYN),
1217 		    ("%s: Listen socket: TH_SYN not set", __func__));
1218 #ifdef INET6
1219 		/*
1220 		 * If deprecated address is forbidden,
1221 		 * we do not accept SYN to deprecated interface
1222 		 * address to prevent any new inbound connection from
1223 		 * getting established.
1224 		 * When we do not accept SYN, we send a TCP RST,
1225 		 * with deprecated source address (instead of dropping
1226 		 * it).  We compromise it as it is much better for peer
1227 		 * to send a RST, and RST will be the final packet
1228 		 * for the exchange.
1229 		 *
1230 		 * If we do not forbid deprecated addresses, we accept
1231 		 * the SYN packet.  RFC2462 does not suggest dropping
1232 		 * SYN in this case.
1233 		 * If we decipher RFC2462 5.5.4, it says like this:
1234 		 * 1. use of deprecated addr with existing
1235 		 *    communication is okay - "SHOULD continue to be
1236 		 *    used"
1237 		 * 2. use of it with new communication:
1238 		 *   (2a) "SHOULD NOT be used if alternate address
1239 		 *        with sufficient scope is available"
1240 		 *   (2b) nothing mentioned otherwise.
1241 		 * Here we fall into (2b) case as we have no choice in
1242 		 * our source address selection - we must obey the peer.
1243 		 *
1244 		 * The wording in RFC2462 is confusing, and there are
1245 		 * multiple description text for deprecated address
1246 		 * handling - worse, they are not exactly the same.
1247 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1248 		 */
1249 		if (isipv6 && !V_ip6_use_deprecated) {
1250 			struct in6_ifaddr *ia6;
1251 
1252 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1253 			if (ia6 != NULL &&
1254 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1255 				ifa_free(&ia6->ia_ifa);
1256 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1257 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1258 					"Connection attempt to deprecated "
1259 					"IPv6 address rejected\n",
1260 					s, __func__);
1261 				rstreason = BANDLIM_RST_OPENPORT;
1262 				goto dropwithreset;
1263 			}
1264 			if (ia6)
1265 				ifa_free(&ia6->ia_ifa);
1266 		}
1267 #endif /* INET6 */
1268 		/*
1269 		 * Basic sanity checks on incoming SYN requests:
1270 		 *   Don't respond if the destination is a link layer
1271 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1272 		 *   If it is from this socket it must be forged.
1273 		 *   Don't respond if the source or destination is a
1274 		 *	global or subnet broad- or multicast address.
1275 		 *   Note that it is quite possible to receive unicast
1276 		 *	link-layer packets with a broadcast IP address. Use
1277 		 *	in_broadcast() to find them.
1278 		 */
1279 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1280 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1281 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1282 				"Connection attempt from broad- or multicast "
1283 				"link layer address ignored\n", s, __func__);
1284 			goto dropunlock;
1285 		}
1286 #ifdef INET6
1287 		if (isipv6) {
1288 			if (th->th_dport == th->th_sport &&
1289 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1290 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1291 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1292 					"Connection attempt to/from self "
1293 					"ignored\n", s, __func__);
1294 				goto dropunlock;
1295 			}
1296 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1297 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1298 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1300 					"Connection attempt from/to multicast "
1301 					"address ignored\n", s, __func__);
1302 				goto dropunlock;
1303 			}
1304 		}
1305 #endif
1306 #if defined(INET) && defined(INET6)
1307 		else
1308 #endif
1309 #ifdef INET
1310 		{
1311 			if (th->th_dport == th->th_sport &&
1312 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1313 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 					"Connection attempt from/to self "
1316 					"ignored\n", s, __func__);
1317 				goto dropunlock;
1318 			}
1319 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1320 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1321 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1322 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1323 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1324 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1325 					"Connection attempt from/to broad- "
1326 					"or multicast address ignored\n",
1327 					s, __func__);
1328 				goto dropunlock;
1329 			}
1330 		}
1331 #endif
1332 		/*
1333 		 * SYN appears to be valid.  Create compressed TCP state
1334 		 * for syncache.
1335 		 */
1336 #ifdef TCPDEBUG
1337 		if (so->so_options & SO_DEBUG)
1338 			tcp_trace(TA_INPUT, ostate, tp,
1339 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1340 #endif
1341 		TCP_PROBE3(debug__input, tp, th, m);
1342 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1343 		if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1344 			goto tfo_socket_result;
1345 
1346 		/*
1347 		 * Entry added to syncache and mbuf consumed.
1348 		 * Only the listen socket is unlocked by syncache_add().
1349 		 */
1350 		INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1351 		return (IPPROTO_DONE);
1352 	} else if (tp->t_state == TCPS_LISTEN) {
1353 		/*
1354 		 * When a listen socket is torn down the SO_ACCEPTCONN
1355 		 * flag is removed first while connections are drained
1356 		 * from the accept queue in a unlock/lock cycle of the
1357 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1358 		 * attempt go through unhandled.
1359 		 */
1360 		goto dropunlock;
1361 	}
1362 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1363 	if (tp->t_flags & TF_SIGNATURE) {
1364 		tcp_dooptions(&to, optp, optlen, thflags);
1365 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1366 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1367 			goto dropunlock;
1368 		}
1369 		if (!TCPMD5_ENABLED() ||
1370 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1371 			goto dropunlock;
1372 	}
1373 #endif
1374 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1375 
1376 	/*
1377 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1378 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1379 	 * the inpcb, and unlocks pcbinfo.
1380 	 */
1381 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1382 	return (IPPROTO_DONE);
1383 
1384 dropwithreset:
1385 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1386 
1387 	if (inp != NULL) {
1388 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1389 		INP_WUNLOCK(inp);
1390 	} else
1391 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1392 	m = NULL;	/* mbuf chain got consumed. */
1393 	goto drop;
1394 
1395 dropunlock:
1396 	if (m != NULL)
1397 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1398 
1399 	if (inp != NULL)
1400 		INP_WUNLOCK(inp);
1401 
1402 drop:
1403 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1404 	if (s != NULL)
1405 		free(s, M_TCPLOG);
1406 	if (m != NULL)
1407 		m_freem(m);
1408 	return (IPPROTO_DONE);
1409 }
1410 
1411 /*
1412  * Automatic sizing of receive socket buffer.  Often the send
1413  * buffer size is not optimally adjusted to the actual network
1414  * conditions at hand (delay bandwidth product).  Setting the
1415  * buffer size too small limits throughput on links with high
1416  * bandwidth and high delay (eg. trans-continental/oceanic links).
1417  *
1418  * On the receive side the socket buffer memory is only rarely
1419  * used to any significant extent.  This allows us to be much
1420  * more aggressive in scaling the receive socket buffer.  For
1421  * the case that the buffer space is actually used to a large
1422  * extent and we run out of kernel memory we can simply drop
1423  * the new segments; TCP on the sender will just retransmit it
1424  * later.  Setting the buffer size too big may only consume too
1425  * much kernel memory if the application doesn't read() from
1426  * the socket or packet loss or reordering makes use of the
1427  * reassembly queue.
1428  *
1429  * The criteria to step up the receive buffer one notch are:
1430  *  1. Application has not set receive buffer size with
1431  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1432  *  2. the number of bytes received during 1/2 of an sRTT
1433  *     is at least 3/8 of the current socket buffer size.
1434  *  3. receive buffer size has not hit maximal automatic size;
1435  *
1436  * If all of the criteria are met we increaset the socket buffer
1437  * by a 1/2 (bounded by the max). This allows us to keep ahead
1438  * of slow-start but also makes it so our peer never gets limited
1439  * by our rwnd which we then open up causing a burst.
1440  *
1441  * This algorithm does two steps per RTT at most and only if
1442  * we receive a bulk stream w/o packet losses or reorderings.
1443  * Shrinking the buffer during idle times is not necessary as
1444  * it doesn't consume any memory when idle.
1445  *
1446  * TODO: Only step up if the application is actually serving
1447  * the buffer to better manage the socket buffer resources.
1448  */
1449 int
1450 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1451     struct tcpcb *tp, int tlen)
1452 {
1453 	int newsize = 0;
1454 
1455 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1456 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1457 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1458 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1459 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1460 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1461 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1462 		}
1463 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1464 
1465 		/* Start over with next RTT. */
1466 		tp->rfbuf_ts = 0;
1467 		tp->rfbuf_cnt = 0;
1468 	} else {
1469 		tp->rfbuf_cnt += tlen;	/* add up */
1470 	}
1471 	return (newsize);
1472 }
1473 
1474 void
1475 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1476 {
1477 	/*
1478 	 * Since tp might be gone if the session entered
1479 	 * the TIME_WAIT state before coming here, we need
1480 	 * to check if the socket is still connected.
1481 	 */
1482 	if ((so->so_state & SS_ISCONNECTED) == 0)
1483 		return;
1484 	INP_LOCK_ASSERT(tp->t_inpcb);
1485 	if (tp->t_flags & TF_WAKESOR) {
1486 		tp->t_flags &= ~TF_WAKESOR;
1487 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1488 		sorwakeup(so);
1489 	}
1490 	if (tp->t_flags & TF_WAKESOW) {
1491 		tp->t_flags &= ~TF_WAKESOW;
1492 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1493 		sowwakeup(so);
1494 	}
1495 }
1496 
1497 void
1498 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1499     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1500 {
1501 	int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1502 	int rstreason, todrop, win, incforsyn = 0;
1503 	uint32_t tiwin;
1504 	uint16_t nsegs;
1505 	char *s;
1506 	struct in_conninfo *inc;
1507 	struct mbuf *mfree;
1508 	struct tcpopt to;
1509 	int tfo_syn;
1510 
1511 #ifdef TCPDEBUG
1512 	/*
1513 	 * The size of tcp_saveipgen must be the size of the max ip header,
1514 	 * now IPv6.
1515 	 */
1516 	u_char tcp_saveipgen[IP6_HDR_LEN];
1517 	struct tcphdr tcp_savetcp;
1518 	short ostate = 0;
1519 #endif
1520 	thflags = th->th_flags;
1521 	inc = &tp->t_inpcb->inp_inc;
1522 	tp->sackhint.last_sack_ack = 0;
1523 	sack_changed = 0;
1524 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1525 
1526 	NET_EPOCH_ASSERT();
1527 	INP_WLOCK_ASSERT(tp->t_inpcb);
1528 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1529 	    __func__));
1530 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1531 	    __func__));
1532 
1533 #ifdef TCPPCAP
1534 	/* Save segment, if requested. */
1535 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1536 #endif
1537 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1538 	    tlen, NULL, true);
1539 
1540 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1541 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1542 			log(LOG_DEBUG, "%s; %s: "
1543 			    "SYN|FIN segment ignored (based on "
1544 			    "sysctl setting)\n", s, __func__);
1545 			free(s, M_TCPLOG);
1546 		}
1547 		goto drop;
1548 	}
1549 
1550 	/*
1551 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1552 	 * check SEQ.ACK first.
1553 	 */
1554 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1555 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1556 		rstreason = BANDLIM_UNLIMITED;
1557 		goto dropwithreset;
1558 	}
1559 
1560 	/*
1561 	 * Segment received on connection.
1562 	 * Reset idle time and keep-alive timer.
1563 	 * XXX: This should be done after segment
1564 	 * validation to ignore broken/spoofed segs.
1565 	 */
1566 	tp->t_rcvtime = ticks;
1567 
1568 	/*
1569 	 * Scale up the window into a 32-bit value.
1570 	 * For the SYN_SENT state the scale is zero.
1571 	 */
1572 	tiwin = th->th_win << tp->snd_scale;
1573 #ifdef STATS
1574 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1575 #endif
1576 
1577 	/*
1578 	 * TCP ECN processing.
1579 	 */
1580 	if (tp->t_flags2 & TF2_ECN_PERMIT) {
1581 		if (thflags & TH_CWR) {
1582 			tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1583 			tp->t_flags |= TF_ACKNOW;
1584 		}
1585 		switch (iptos & IPTOS_ECN_MASK) {
1586 		case IPTOS_ECN_CE:
1587 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1588 			TCPSTAT_INC(tcps_ecn_ce);
1589 			break;
1590 		case IPTOS_ECN_ECT0:
1591 			TCPSTAT_INC(tcps_ecn_ect0);
1592 			break;
1593 		case IPTOS_ECN_ECT1:
1594 			TCPSTAT_INC(tcps_ecn_ect1);
1595 			break;
1596 		}
1597 
1598 		/* Process a packet differently from RFC3168. */
1599 		cc_ecnpkt_handler(tp, th, iptos);
1600 
1601 		/* Congestion experienced. */
1602 		if (thflags & TH_ECE) {
1603 			cc_cong_signal(tp, th, CC_ECN);
1604 		}
1605 	}
1606 
1607 	/*
1608 	 * Parse options on any incoming segment.
1609 	 */
1610 	tcp_dooptions(&to, (u_char *)(th + 1),
1611 	    (th->th_off << 2) - sizeof(struct tcphdr),
1612 	    (thflags & TH_SYN) ? TO_SYN : 0);
1613 
1614 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1615 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1616 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1617 		TCPSTAT_INC(tcps_sig_err_sigopt);
1618 		/* XXX: should drop? */
1619 	}
1620 #endif
1621 	/*
1622 	 * If echoed timestamp is later than the current time,
1623 	 * fall back to non RFC1323 RTT calculation.  Normalize
1624 	 * timestamp if syncookies were used when this connection
1625 	 * was established.
1626 	 */
1627 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1628 		to.to_tsecr -= tp->ts_offset;
1629 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1630 			to.to_tsecr = 0;
1631 		else if (tp->t_flags & TF_PREVVALID &&
1632 			 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1633 			cc_cong_signal(tp, th, CC_RTO_ERR);
1634 	}
1635 	/*
1636 	 * Process options only when we get SYN/ACK back. The SYN case
1637 	 * for incoming connections is handled in tcp_syncache.
1638 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1639 	 * or <SYN,ACK>) segment itself is never scaled.
1640 	 * XXX this is traditional behavior, may need to be cleaned up.
1641 	 */
1642 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1643 		/* Handle parallel SYN for ECN */
1644 		if (!(thflags & TH_ACK) &&
1645 		    ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1646 		    ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1647 			tp->t_flags2 |= TF2_ECN_PERMIT;
1648 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1649 			TCPSTAT_INC(tcps_ecn_shs);
1650 		}
1651 		if ((to.to_flags & TOF_SCALE) &&
1652 		    (tp->t_flags & TF_REQ_SCALE)) {
1653 			tp->t_flags |= TF_RCVD_SCALE;
1654 			tp->snd_scale = to.to_wscale;
1655 		} else
1656 			tp->t_flags &= ~TF_REQ_SCALE;
1657 		/*
1658 		 * Initial send window.  It will be updated with
1659 		 * the next incoming segment to the scaled value.
1660 		 */
1661 		tp->snd_wnd = th->th_win;
1662 		if ((to.to_flags & TOF_TS) &&
1663 		    (tp->t_flags & TF_REQ_TSTMP)) {
1664 			tp->t_flags |= TF_RCVD_TSTMP;
1665 			tp->ts_recent = to.to_tsval;
1666 			tp->ts_recent_age = tcp_ts_getticks();
1667 		} else
1668 			tp->t_flags &= ~TF_REQ_TSTMP;
1669 		if (to.to_flags & TOF_MSS)
1670 			tcp_mss(tp, to.to_mss);
1671 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1672 		    (to.to_flags & TOF_SACKPERM) == 0)
1673 			tp->t_flags &= ~TF_SACK_PERMIT;
1674 		if (IS_FASTOPEN(tp->t_flags)) {
1675 			if (to.to_flags & TOF_FASTOPEN) {
1676 				uint16_t mss;
1677 
1678 				if (to.to_flags & TOF_MSS)
1679 					mss = to.to_mss;
1680 				else
1681 					if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1682 						mss = TCP6_MSS;
1683 					else
1684 						mss = TCP_MSS;
1685 				tcp_fastopen_update_cache(tp, mss,
1686 				    to.to_tfo_len, to.to_tfo_cookie);
1687 			} else
1688 				tcp_fastopen_disable_path(tp);
1689 		}
1690 	}
1691 
1692 	/*
1693 	 * If timestamps were negotiated during SYN/ACK and a
1694 	 * segment without a timestamp is received, silently drop
1695 	 * the segment, unless it is a RST segment or missing timestamps are
1696 	 * tolerated.
1697 	 * See section 3.2 of RFC 7323.
1698 	 */
1699 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1700 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1701 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1702 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1703 				    "segment processed normally\n",
1704 				    s, __func__);
1705 				free(s, M_TCPLOG);
1706 			}
1707 		} else {
1708 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1709 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1710 				    "segment silently dropped\n", s, __func__);
1711 				free(s, M_TCPLOG);
1712 			}
1713 			goto drop;
1714 		}
1715 	}
1716 	/*
1717 	 * If timestamps were not negotiated during SYN/ACK and a
1718 	 * segment with a timestamp is received, ignore the
1719 	 * timestamp and process the packet normally.
1720 	 * See section 3.2 of RFC 7323.
1721 	 */
1722 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1723 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1724 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1725 			    "segment processed normally\n", s, __func__);
1726 			free(s, M_TCPLOG);
1727 		}
1728 	}
1729 
1730 	/*
1731 	 * Header prediction: check for the two common cases
1732 	 * of a uni-directional data xfer.  If the packet has
1733 	 * no control flags, is in-sequence, the window didn't
1734 	 * change and we're not retransmitting, it's a
1735 	 * candidate.  If the length is zero and the ack moved
1736 	 * forward, we're the sender side of the xfer.  Just
1737 	 * free the data acked & wake any higher level process
1738 	 * that was blocked waiting for space.  If the length
1739 	 * is non-zero and the ack didn't move, we're the
1740 	 * receiver side.  If we're getting packets in-order
1741 	 * (the reassembly queue is empty), add the data to
1742 	 * the socket buffer and note that we need a delayed ack.
1743 	 * Make sure that the hidden state-flags are also off.
1744 	 * Since we check for TCPS_ESTABLISHED first, it can only
1745 	 * be TH_NEEDSYN.
1746 	 */
1747 	if (tp->t_state == TCPS_ESTABLISHED &&
1748 	    th->th_seq == tp->rcv_nxt &&
1749 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1750 	    tp->snd_nxt == tp->snd_max &&
1751 	    tiwin && tiwin == tp->snd_wnd &&
1752 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1753 	    SEGQ_EMPTY(tp) &&
1754 	    ((to.to_flags & TOF_TS) == 0 ||
1755 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1756 		/*
1757 		 * If last ACK falls within this segment's sequence numbers,
1758 		 * record the timestamp.
1759 		 * NOTE that the test is modified according to the latest
1760 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1761 		 */
1762 		if ((to.to_flags & TOF_TS) != 0 &&
1763 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1764 			tp->ts_recent_age = tcp_ts_getticks();
1765 			tp->ts_recent = to.to_tsval;
1766 		}
1767 
1768 		if (tlen == 0) {
1769 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1770 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1771 			    !IN_RECOVERY(tp->t_flags) &&
1772 			    (to.to_flags & TOF_SACK) == 0 &&
1773 			    TAILQ_EMPTY(&tp->snd_holes)) {
1774 				/*
1775 				 * This is a pure ack for outstanding data.
1776 				 */
1777 				TCPSTAT_INC(tcps_predack);
1778 
1779 				/*
1780 				 * "bad retransmit" recovery without timestamps.
1781 				 */
1782 				if ((to.to_flags & TOF_TS) == 0 &&
1783 				    tp->t_rxtshift == 1 &&
1784 				    tp->t_flags & TF_PREVVALID &&
1785 				    (int)(ticks - tp->t_badrxtwin) < 0) {
1786 					cc_cong_signal(tp, th, CC_RTO_ERR);
1787 				}
1788 
1789 				/*
1790 				 * Recalculate the transmit timer / rtt.
1791 				 *
1792 				 * Some boxes send broken timestamp replies
1793 				 * during the SYN+ACK phase, ignore
1794 				 * timestamps of 0 or we could calculate a
1795 				 * huge RTT and blow up the retransmit timer.
1796 				 */
1797 				if ((to.to_flags & TOF_TS) != 0 &&
1798 				    to.to_tsecr) {
1799 					uint32_t t;
1800 
1801 					t = tcp_ts_getticks() - to.to_tsecr;
1802 					if (!tp->t_rttlow || tp->t_rttlow > t)
1803 						tp->t_rttlow = t;
1804 					tcp_xmit_timer(tp,
1805 					    TCP_TS_TO_TICKS(t) + 1);
1806 				} else if (tp->t_rtttime &&
1807 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1808 					if (!tp->t_rttlow ||
1809 					    tp->t_rttlow > ticks - tp->t_rtttime)
1810 						tp->t_rttlow = ticks - tp->t_rtttime;
1811 					tcp_xmit_timer(tp,
1812 							ticks - tp->t_rtttime);
1813 				}
1814 				acked = BYTES_THIS_ACK(tp, th);
1815 
1816 #ifdef TCP_HHOOK
1817 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1818 				hhook_run_tcp_est_in(tp, th, &to);
1819 #endif
1820 
1821 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1822 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1823 				sbdrop(&so->so_snd, acked);
1824 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1825 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1826 					tp->snd_recover = th->th_ack - 1;
1827 
1828 				/*
1829 				 * Let the congestion control algorithm update
1830 				 * congestion control related information. This
1831 				 * typically means increasing the congestion
1832 				 * window.
1833 				 */
1834 				cc_ack_received(tp, th, nsegs, CC_ACK);
1835 
1836 				tp->snd_una = th->th_ack;
1837 				/*
1838 				 * Pull snd_wl2 up to prevent seq wrap relative
1839 				 * to th_ack.
1840 				 */
1841 				tp->snd_wl2 = th->th_ack;
1842 				tp->t_dupacks = 0;
1843 				m_freem(m);
1844 
1845 				/*
1846 				 * If all outstanding data are acked, stop
1847 				 * retransmit timer, otherwise restart timer
1848 				 * using current (possibly backed-off) value.
1849 				 * If process is waiting for space,
1850 				 * wakeup/selwakeup/signal.  If data
1851 				 * are ready to send, let tcp_output
1852 				 * decide between more output or persist.
1853 				 */
1854 #ifdef TCPDEBUG
1855 				if (so->so_options & SO_DEBUG)
1856 					tcp_trace(TA_INPUT, ostate, tp,
1857 					    (void *)tcp_saveipgen,
1858 					    &tcp_savetcp, 0);
1859 #endif
1860 				TCP_PROBE3(debug__input, tp, th, m);
1861 				if (tp->snd_una == tp->snd_max)
1862 					tcp_timer_activate(tp, TT_REXMT, 0);
1863 				else if (!tcp_timer_active(tp, TT_PERSIST))
1864 					tcp_timer_activate(tp, TT_REXMT,
1865 						      tp->t_rxtcur);
1866 				tp->t_flags |= TF_WAKESOW;
1867 				if (sbavail(&so->so_snd))
1868 					(void) tp->t_fb->tfb_tcp_output(tp);
1869 				goto check_delack;
1870 			}
1871 		} else if (th->th_ack == tp->snd_una &&
1872 		    tlen <= sbspace(&so->so_rcv)) {
1873 			int newsize = 0;	/* automatic sockbuf scaling */
1874 
1875 			/*
1876 			 * This is a pure, in-sequence data packet with
1877 			 * nothing on the reassembly queue and we have enough
1878 			 * buffer space to take it.
1879 			 */
1880 			/* Clean receiver SACK report if present */
1881 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1882 				tcp_clean_sackreport(tp);
1883 			TCPSTAT_INC(tcps_preddat);
1884 			tp->rcv_nxt += tlen;
1885 			if (tlen &&
1886 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1887 			    (tp->t_fbyte_in == 0)) {
1888 				tp->t_fbyte_in = ticks;
1889 				if (tp->t_fbyte_in == 0)
1890 					tp->t_fbyte_in = 1;
1891 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1892 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1893 			}
1894 			/*
1895 			 * Pull snd_wl1 up to prevent seq wrap relative to
1896 			 * th_seq.
1897 			 */
1898 			tp->snd_wl1 = th->th_seq;
1899 			/*
1900 			 * Pull rcv_up up to prevent seq wrap relative to
1901 			 * rcv_nxt.
1902 			 */
1903 			tp->rcv_up = tp->rcv_nxt;
1904 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1905 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1906 #ifdef TCPDEBUG
1907 			if (so->so_options & SO_DEBUG)
1908 				tcp_trace(TA_INPUT, ostate, tp,
1909 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1910 #endif
1911 			TCP_PROBE3(debug__input, tp, th, m);
1912 
1913 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1914 
1915 			/* Add data to socket buffer. */
1916 			SOCKBUF_LOCK(&so->so_rcv);
1917 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1918 				m_freem(m);
1919 			} else {
1920 				/*
1921 				 * Set new socket buffer size.
1922 				 * Give up when limit is reached.
1923 				 */
1924 				if (newsize)
1925 					if (!sbreserve_locked(&so->so_rcv,
1926 					    newsize, so, NULL))
1927 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1928 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1929 				sbappendstream_locked(&so->so_rcv, m, 0);
1930 			}
1931 			SOCKBUF_UNLOCK(&so->so_rcv);
1932 			tp->t_flags |= TF_WAKESOR;
1933 			if (DELAY_ACK(tp, tlen)) {
1934 				tp->t_flags |= TF_DELACK;
1935 			} else {
1936 				tp->t_flags |= TF_ACKNOW;
1937 				tp->t_fb->tfb_tcp_output(tp);
1938 			}
1939 			goto check_delack;
1940 		}
1941 	}
1942 
1943 	/*
1944 	 * Calculate amount of space in receive window,
1945 	 * and then do TCP input processing.
1946 	 * Receive window is amount of space in rcv queue,
1947 	 * but not less than advertised window.
1948 	 */
1949 	win = sbspace(&so->so_rcv);
1950 	if (win < 0)
1951 		win = 0;
1952 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1953 
1954 	switch (tp->t_state) {
1955 	/*
1956 	 * If the state is SYN_RECEIVED:
1957 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1958 	 */
1959 	case TCPS_SYN_RECEIVED:
1960 		if ((thflags & TH_ACK) &&
1961 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1962 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1963 				rstreason = BANDLIM_RST_OPENPORT;
1964 				goto dropwithreset;
1965 		}
1966 		if (IS_FASTOPEN(tp->t_flags)) {
1967 			/*
1968 			 * When a TFO connection is in SYN_RECEIVED, the
1969 			 * only valid packets are the initial SYN, a
1970 			 * retransmit/copy of the initial SYN (possibly with
1971 			 * a subset of the original data), a valid ACK, a
1972 			 * FIN, or a RST.
1973 			 */
1974 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1975 				rstreason = BANDLIM_RST_OPENPORT;
1976 				goto dropwithreset;
1977 			} else if (thflags & TH_SYN) {
1978 				/* non-initial SYN is ignored */
1979 				if ((tcp_timer_active(tp, TT_DELACK) ||
1980 				     tcp_timer_active(tp, TT_REXMT)))
1981 					goto drop;
1982 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1983 				goto drop;
1984 			}
1985 		}
1986 		break;
1987 
1988 	/*
1989 	 * If the state is SYN_SENT:
1990 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
1991 	 *	    been verified), then drop the connection.
1992 	 *	if seg contains a RST without an ACK, drop the seg.
1993 	 *	if seg does not contain SYN, then drop the seg.
1994 	 * Otherwise this is an acceptable SYN segment
1995 	 *	initialize tp->rcv_nxt and tp->irs
1996 	 *	if seg contains ack then advance tp->snd_una
1997 	 *	if seg contains an ECE and ECN support is enabled, the stream
1998 	 *	    is ECN capable.
1999 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2000 	 *	arrange for segment to be acked (eventually)
2001 	 *	continue processing rest of data/controls, beginning with URG
2002 	 */
2003 	case TCPS_SYN_SENT:
2004 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2005 			TCP_PROBE5(connect__refused, NULL, tp,
2006 			    m, tp, th);
2007 			tp = tcp_drop(tp, ECONNREFUSED);
2008 		}
2009 		if (thflags & TH_RST)
2010 			goto drop;
2011 		if (!(thflags & TH_SYN))
2012 			goto drop;
2013 
2014 		tp->irs = th->th_seq;
2015 		tcp_rcvseqinit(tp);
2016 		if (thflags & TH_ACK) {
2017 			int tfo_partial_ack = 0;
2018 
2019 			TCPSTAT_INC(tcps_connects);
2020 			soisconnected(so);
2021 #ifdef MAC
2022 			mac_socketpeer_set_from_mbuf(m, so);
2023 #endif
2024 			/* Do window scaling on this connection? */
2025 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2026 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2027 				tp->rcv_scale = tp->request_r_scale;
2028 			}
2029 			tp->rcv_adv += min(tp->rcv_wnd,
2030 			    TCP_MAXWIN << tp->rcv_scale);
2031 			tp->snd_una++;		/* SYN is acked */
2032 			/*
2033 			 * If not all the data that was sent in the TFO SYN
2034 			 * has been acked, resend the remainder right away.
2035 			 */
2036 			if (IS_FASTOPEN(tp->t_flags) &&
2037 			    (tp->snd_una != tp->snd_max)) {
2038 				tp->snd_nxt = th->th_ack;
2039 				tfo_partial_ack = 1;
2040 			}
2041 			/*
2042 			 * If there's data, delay ACK; if there's also a FIN
2043 			 * ACKNOW will be turned on later.
2044 			 */
2045 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2046 				tcp_timer_activate(tp, TT_DELACK,
2047 				    tcp_delacktime);
2048 			else
2049 				tp->t_flags |= TF_ACKNOW;
2050 
2051 			if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2052 			    (V_tcp_do_ecn == 1)) {
2053 				tp->t_flags2 |= TF2_ECN_PERMIT;
2054 				TCPSTAT_INC(tcps_ecn_shs);
2055 			}
2056 
2057 			/*
2058 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2059 			 * Transitions:
2060 			 *	SYN_SENT  --> ESTABLISHED
2061 			 *	SYN_SENT* --> FIN_WAIT_1
2062 			 */
2063 			tp->t_starttime = ticks;
2064 			if (tp->t_flags & TF_NEEDFIN) {
2065 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2066 				tp->t_flags &= ~TF_NEEDFIN;
2067 				thflags &= ~TH_SYN;
2068 			} else {
2069 				tcp_state_change(tp, TCPS_ESTABLISHED);
2070 				TCP_PROBE5(connect__established, NULL, tp,
2071 				    m, tp, th);
2072 				cc_conn_init(tp);
2073 				tcp_timer_activate(tp, TT_KEEP,
2074 				    TP_KEEPIDLE(tp));
2075 			}
2076 		} else {
2077 			/*
2078 			 * Received initial SYN in SYN-SENT[*] state =>
2079 			 * simultaneous open.
2080 			 * If it succeeds, connection is * half-synchronized.
2081 			 * Otherwise, do 3-way handshake:
2082 			 *        SYN-SENT -> SYN-RECEIVED
2083 			 *        SYN-SENT* -> SYN-RECEIVED*
2084 			 */
2085 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2086 			tcp_timer_activate(tp, TT_REXMT, 0);
2087 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2088 		}
2089 
2090 		INP_WLOCK_ASSERT(tp->t_inpcb);
2091 
2092 		/*
2093 		 * Advance th->th_seq to correspond to first data byte.
2094 		 * If data, trim to stay within window,
2095 		 * dropping FIN if necessary.
2096 		 */
2097 		th->th_seq++;
2098 		if (tlen > tp->rcv_wnd) {
2099 			todrop = tlen - tp->rcv_wnd;
2100 			m_adj(m, -todrop);
2101 			tlen = tp->rcv_wnd;
2102 			thflags &= ~TH_FIN;
2103 			TCPSTAT_INC(tcps_rcvpackafterwin);
2104 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2105 		}
2106 		tp->snd_wl1 = th->th_seq - 1;
2107 		tp->rcv_up = th->th_seq;
2108 		/*
2109 		 * Client side of transaction: already sent SYN and data.
2110 		 * If the remote host used T/TCP to validate the SYN,
2111 		 * our data will be ACK'd; if so, enter normal data segment
2112 		 * processing in the middle of step 5, ack processing.
2113 		 * Otherwise, goto step 6.
2114 		 */
2115 		if (thflags & TH_ACK)
2116 			goto process_ACK;
2117 
2118 		goto step6;
2119 
2120 	/*
2121 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2122 	 *      do normal processing.
2123 	 *
2124 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2125 	 */
2126 	case TCPS_LAST_ACK:
2127 	case TCPS_CLOSING:
2128 		break;  /* continue normal processing */
2129 	}
2130 
2131 	/*
2132 	 * States other than LISTEN or SYN_SENT.
2133 	 * First check the RST flag and sequence number since reset segments
2134 	 * are exempt from the timestamp and connection count tests.  This
2135 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2136 	 * below which allowed reset segments in half the sequence space
2137 	 * to fall though and be processed (which gives forged reset
2138 	 * segments with a random sequence number a 50 percent chance of
2139 	 * killing a connection).
2140 	 * Then check timestamp, if present.
2141 	 * Then check the connection count, if present.
2142 	 * Then check that at least some bytes of segment are within
2143 	 * receive window.  If segment begins before rcv_nxt,
2144 	 * drop leading data (and SYN); if nothing left, just ack.
2145 	 */
2146 	if (thflags & TH_RST) {
2147 		/*
2148 		 * RFC5961 Section 3.2
2149 		 *
2150 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2151 		 * - If RST is in window, we send challenge ACK.
2152 		 *
2153 		 * Note: to take into account delayed ACKs, we should
2154 		 *   test against last_ack_sent instead of rcv_nxt.
2155 		 * Note 2: we handle special case of closed window, not
2156 		 *   covered by the RFC.
2157 		 */
2158 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2159 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2160 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2161 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2162 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2163 			    __func__, th, tp));
2164 
2165 			if (V_tcp_insecure_rst ||
2166 			    tp->last_ack_sent == th->th_seq) {
2167 				TCPSTAT_INC(tcps_drops);
2168 				/* Drop the connection. */
2169 				switch (tp->t_state) {
2170 				case TCPS_SYN_RECEIVED:
2171 					so->so_error = ECONNREFUSED;
2172 					goto close;
2173 				case TCPS_ESTABLISHED:
2174 				case TCPS_FIN_WAIT_1:
2175 				case TCPS_FIN_WAIT_2:
2176 				case TCPS_CLOSE_WAIT:
2177 				case TCPS_CLOSING:
2178 				case TCPS_LAST_ACK:
2179 					so->so_error = ECONNRESET;
2180 				close:
2181 					/* FALLTHROUGH */
2182 				default:
2183 					tp = tcp_close(tp);
2184 				}
2185 			} else {
2186 				TCPSTAT_INC(tcps_badrst);
2187 				/* Send challenge ACK. */
2188 				tcp_respond(tp, mtod(m, void *), th, m,
2189 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2190 				tp->last_ack_sent = tp->rcv_nxt;
2191 				m = NULL;
2192 			}
2193 		}
2194 		goto drop;
2195 	}
2196 
2197 	/*
2198 	 * RFC5961 Section 4.2
2199 	 * Send challenge ACK for any SYN in synchronized state.
2200 	 */
2201 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2202 	    tp->t_state != TCPS_SYN_RECEIVED) {
2203 		TCPSTAT_INC(tcps_badsyn);
2204 		if (V_tcp_insecure_syn &&
2205 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2206 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2207 			tp = tcp_drop(tp, ECONNRESET);
2208 			rstreason = BANDLIM_UNLIMITED;
2209 		} else {
2210 			/* Send challenge ACK. */
2211 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2212 			    tp->snd_nxt, TH_ACK);
2213 			tp->last_ack_sent = tp->rcv_nxt;
2214 			m = NULL;
2215 		}
2216 		goto drop;
2217 	}
2218 
2219 	/*
2220 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2221 	 * and it's less than ts_recent, drop it.
2222 	 */
2223 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2224 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2225 		/* Check to see if ts_recent is over 24 days old.  */
2226 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2227 			/*
2228 			 * Invalidate ts_recent.  If this segment updates
2229 			 * ts_recent, the age will be reset later and ts_recent
2230 			 * will get a valid value.  If it does not, setting
2231 			 * ts_recent to zero will at least satisfy the
2232 			 * requirement that zero be placed in the timestamp
2233 			 * echo reply when ts_recent isn't valid.  The
2234 			 * age isn't reset until we get a valid ts_recent
2235 			 * because we don't want out-of-order segments to be
2236 			 * dropped when ts_recent is old.
2237 			 */
2238 			tp->ts_recent = 0;
2239 		} else {
2240 			TCPSTAT_INC(tcps_rcvduppack);
2241 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2242 			TCPSTAT_INC(tcps_pawsdrop);
2243 			if (tlen)
2244 				goto dropafterack;
2245 			goto drop;
2246 		}
2247 	}
2248 
2249 	/*
2250 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2251 	 * this connection before trimming the data to fit the receive
2252 	 * window.  Check the sequence number versus IRS since we know
2253 	 * the sequence numbers haven't wrapped.  This is a partial fix
2254 	 * for the "LAND" DoS attack.
2255 	 */
2256 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2257 		rstreason = BANDLIM_RST_OPENPORT;
2258 		goto dropwithreset;
2259 	}
2260 
2261 	todrop = tp->rcv_nxt - th->th_seq;
2262 	if (todrop > 0) {
2263 		if (thflags & TH_SYN) {
2264 			thflags &= ~TH_SYN;
2265 			th->th_seq++;
2266 			if (th->th_urp > 1)
2267 				th->th_urp--;
2268 			else
2269 				thflags &= ~TH_URG;
2270 			todrop--;
2271 		}
2272 		/*
2273 		 * Following if statement from Stevens, vol. 2, p. 960.
2274 		 */
2275 		if (todrop > tlen
2276 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2277 			/*
2278 			 * Any valid FIN must be to the left of the window.
2279 			 * At this point the FIN must be a duplicate or out
2280 			 * of sequence; drop it.
2281 			 */
2282 			thflags &= ~TH_FIN;
2283 
2284 			/*
2285 			 * Send an ACK to resynchronize and drop any data.
2286 			 * But keep on processing for RST or ACK.
2287 			 */
2288 			tp->t_flags |= TF_ACKNOW;
2289 			todrop = tlen;
2290 			TCPSTAT_INC(tcps_rcvduppack);
2291 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2292 		} else {
2293 			TCPSTAT_INC(tcps_rcvpartduppack);
2294 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2295 		}
2296 		/*
2297 		 * DSACK - add SACK block for dropped range
2298 		 */
2299 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2300 			tcp_update_sack_list(tp, th->th_seq,
2301 			    th->th_seq + todrop);
2302 			/*
2303 			 * ACK now, as the next in-sequence segment
2304 			 * will clear the DSACK block again
2305 			 */
2306 			tp->t_flags |= TF_ACKNOW;
2307 		}
2308 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2309 		th->th_seq += todrop;
2310 		tlen -= todrop;
2311 		if (th->th_urp > todrop)
2312 			th->th_urp -= todrop;
2313 		else {
2314 			thflags &= ~TH_URG;
2315 			th->th_urp = 0;
2316 		}
2317 	}
2318 
2319 	/*
2320 	 * If new data are received on a connection after the
2321 	 * user processes are gone, then RST the other end.
2322 	 */
2323 	if ((so->so_state & SS_NOFDREF) &&
2324 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2325 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2326 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2327 			    "after socket was closed, "
2328 			    "sending RST and removing tcpcb\n",
2329 			    s, __func__, tcpstates[tp->t_state], tlen);
2330 			free(s, M_TCPLOG);
2331 		}
2332 		tp = tcp_close(tp);
2333 		TCPSTAT_INC(tcps_rcvafterclose);
2334 		rstreason = BANDLIM_UNLIMITED;
2335 		goto dropwithreset;
2336 	}
2337 
2338 	/*
2339 	 * If segment ends after window, drop trailing data
2340 	 * (and PUSH and FIN); if nothing left, just ACK.
2341 	 */
2342 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2343 	if (todrop > 0) {
2344 		TCPSTAT_INC(tcps_rcvpackafterwin);
2345 		if (todrop >= tlen) {
2346 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2347 			/*
2348 			 * If window is closed can only take segments at
2349 			 * window edge, and have to drop data and PUSH from
2350 			 * incoming segments.  Continue processing, but
2351 			 * remember to ack.  Otherwise, drop segment
2352 			 * and ack.
2353 			 */
2354 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2355 				tp->t_flags |= TF_ACKNOW;
2356 				TCPSTAT_INC(tcps_rcvwinprobe);
2357 			} else
2358 				goto dropafterack;
2359 		} else
2360 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2361 		m_adj(m, -todrop);
2362 		tlen -= todrop;
2363 		thflags &= ~(TH_PUSH|TH_FIN);
2364 	}
2365 
2366 	/*
2367 	 * If last ACK falls within this segment's sequence numbers,
2368 	 * record its timestamp.
2369 	 * NOTE:
2370 	 * 1) That the test incorporates suggestions from the latest
2371 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2372 	 * 2) That updating only on newer timestamps interferes with
2373 	 *    our earlier PAWS tests, so this check should be solely
2374 	 *    predicated on the sequence space of this segment.
2375 	 * 3) That we modify the segment boundary check to be
2376 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2377 	 *    instead of RFC1323's
2378 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2379 	 *    This modified check allows us to overcome RFC1323's
2380 	 *    limitations as described in Stevens TCP/IP Illustrated
2381 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2382 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2383 	 */
2384 	if ((to.to_flags & TOF_TS) != 0 &&
2385 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2386 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2387 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2388 		tp->ts_recent_age = tcp_ts_getticks();
2389 		tp->ts_recent = to.to_tsval;
2390 	}
2391 
2392 	/*
2393 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2394 	 * flag is on (half-synchronized state), then queue data for
2395 	 * later processing; else drop segment and return.
2396 	 */
2397 	if ((thflags & TH_ACK) == 0) {
2398 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2399 		    (tp->t_flags & TF_NEEDSYN)) {
2400 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2401 			    IS_FASTOPEN(tp->t_flags)) {
2402 				tp->snd_wnd = tiwin;
2403 				cc_conn_init(tp);
2404 			}
2405 			goto step6;
2406 		} else if (tp->t_flags & TF_ACKNOW)
2407 			goto dropafterack;
2408 		else
2409 			goto drop;
2410 	}
2411 
2412 	/*
2413 	 * Ack processing.
2414 	 */
2415 	switch (tp->t_state) {
2416 	/*
2417 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2418 	 * ESTABLISHED state and continue processing.
2419 	 * The ACK was checked above.
2420 	 */
2421 	case TCPS_SYN_RECEIVED:
2422 
2423 		TCPSTAT_INC(tcps_connects);
2424 		soisconnected(so);
2425 		/* Do window scaling? */
2426 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2427 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2428 			tp->rcv_scale = tp->request_r_scale;
2429 		}
2430 		tp->snd_wnd = tiwin;
2431 		/*
2432 		 * Make transitions:
2433 		 *      SYN-RECEIVED  -> ESTABLISHED
2434 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2435 		 */
2436 		tp->t_starttime = ticks;
2437 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2438 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2439 			tp->t_tfo_pending = NULL;
2440 		}
2441 		if (tp->t_flags & TF_NEEDFIN) {
2442 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2443 			tp->t_flags &= ~TF_NEEDFIN;
2444 		} else {
2445 			tcp_state_change(tp, TCPS_ESTABLISHED);
2446 			TCP_PROBE5(accept__established, NULL, tp,
2447 			    m, tp, th);
2448 			/*
2449 			 * TFO connections call cc_conn_init() during SYN
2450 			 * processing.  Calling it again here for such
2451 			 * connections is not harmless as it would undo the
2452 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2453 			 * is retransmitted.
2454 			 */
2455 			if (!IS_FASTOPEN(tp->t_flags))
2456 				cc_conn_init(tp);
2457 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2458 		}
2459 		/*
2460 		 * Account for the ACK of our SYN prior to
2461 		 * regular ACK processing below, except for
2462 		 * simultaneous SYN, which is handled later.
2463 		 */
2464 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2465 			incforsyn = 1;
2466 		/*
2467 		 * If segment contains data or ACK, will call tcp_reass()
2468 		 * later; if not, do so now to pass queued data to user.
2469 		 */
2470 		if (tlen == 0 && (thflags & TH_FIN) == 0)
2471 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2472 			    (struct mbuf *)0);
2473 		tp->snd_wl1 = th->th_seq - 1;
2474 		/* FALLTHROUGH */
2475 
2476 	/*
2477 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2478 	 * ACKs.  If the ack is in the range
2479 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2480 	 * then advance tp->snd_una to th->th_ack and drop
2481 	 * data from the retransmission queue.  If this ACK reflects
2482 	 * more up to date window information we update our window information.
2483 	 */
2484 	case TCPS_ESTABLISHED:
2485 	case TCPS_FIN_WAIT_1:
2486 	case TCPS_FIN_WAIT_2:
2487 	case TCPS_CLOSE_WAIT:
2488 	case TCPS_CLOSING:
2489 	case TCPS_LAST_ACK:
2490 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2491 			TCPSTAT_INC(tcps_rcvacktoomuch);
2492 			goto dropafterack;
2493 		}
2494 		if ((tp->t_flags & TF_SACK_PERMIT) &&
2495 		    ((to.to_flags & TOF_SACK) ||
2496 		     !TAILQ_EMPTY(&tp->snd_holes)))
2497 			sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2498 		else
2499 			/*
2500 			 * Reset the value so that previous (valid) value
2501 			 * from the last ack with SACK doesn't get used.
2502 			 */
2503 			tp->sackhint.sacked_bytes = 0;
2504 
2505 #ifdef TCP_HHOOK
2506 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2507 		hhook_run_tcp_est_in(tp, th, &to);
2508 #endif
2509 
2510 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2511 			u_int maxseg;
2512 
2513 			maxseg = tcp_maxseg(tp);
2514 			if (tlen == 0 &&
2515 			    (tiwin == tp->snd_wnd ||
2516 			    (tp->t_flags & TF_SACK_PERMIT))) {
2517 				/*
2518 				 * If this is the first time we've seen a
2519 				 * FIN from the remote, this is not a
2520 				 * duplicate and it needs to be processed
2521 				 * normally.  This happens during a
2522 				 * simultaneous close.
2523 				 */
2524 				if ((thflags & TH_FIN) &&
2525 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2526 					tp->t_dupacks = 0;
2527 					break;
2528 				}
2529 				TCPSTAT_INC(tcps_rcvdupack);
2530 				/*
2531 				 * If we have outstanding data (other than
2532 				 * a window probe), this is a completely
2533 				 * duplicate ack (ie, window info didn't
2534 				 * change and FIN isn't set),
2535 				 * the ack is the biggest we've
2536 				 * seen and we've seen exactly our rexmt
2537 				 * threshold of them, assume a packet
2538 				 * has been dropped and retransmit it.
2539 				 * Kludge snd_nxt & the congestion
2540 				 * window so we send only this one
2541 				 * packet.
2542 				 *
2543 				 * We know we're losing at the current
2544 				 * window size so do congestion avoidance
2545 				 * (set ssthresh to half the current window
2546 				 * and pull our congestion window back to
2547 				 * the new ssthresh).
2548 				 *
2549 				 * Dup acks mean that packets have left the
2550 				 * network (they're now cached at the receiver)
2551 				 * so bump cwnd by the amount in the receiver
2552 				 * to keep a constant cwnd packets in the
2553 				 * network.
2554 				 *
2555 				 * When using TCP ECN, notify the peer that
2556 				 * we reduced the cwnd.
2557 				 */
2558 				/*
2559 				 * Following 2 kinds of acks should not affect
2560 				 * dupack counting:
2561 				 * 1) Old acks
2562 				 * 2) Acks with SACK but without any new SACK
2563 				 * information in them. These could result from
2564 				 * any anomaly in the network like a switch
2565 				 * duplicating packets or a possible DoS attack.
2566 				 */
2567 				if (th->th_ack != tp->snd_una ||
2568 				    ((tp->t_flags & TF_SACK_PERMIT) &&
2569 				    !sack_changed))
2570 					break;
2571 				else if (!tcp_timer_active(tp, TT_REXMT))
2572 					tp->t_dupacks = 0;
2573 				else if (++tp->t_dupacks > tcprexmtthresh ||
2574 				     IN_FASTRECOVERY(tp->t_flags)) {
2575 					cc_ack_received(tp, th, nsegs,
2576 					    CC_DUPACK);
2577 					if (V_tcp_do_prr &&
2578 					    IN_FASTRECOVERY(tp->t_flags) &&
2579 					    (tp->t_flags & TF_SACK_PERMIT)) {
2580 						long snd_cnt = 0, limit = 0;
2581 						long del_data = 0, pipe = 0;
2582 						/*
2583 						 * In a duplicate ACK del_data is only the
2584 						 * diff_in_sack. If no SACK is used del_data
2585 						 * will be 0. Pipe is the amount of data we
2586 						 * estimate to be in the network.
2587 						 */
2588 						del_data = tp->sackhint.delivered_data;
2589 						pipe = (tp->snd_nxt - tp->snd_fack) +
2590 							tp->sackhint.sack_bytes_rexmit;
2591 						tp->sackhint.prr_delivered += del_data;
2592 						if (pipe > tp->snd_ssthresh) {
2593 							if (tp->sackhint.recover_fs == 0)
2594 								tp->sackhint.recover_fs =
2595 								    max(1, tp->snd_nxt - tp->snd_una);
2596 							snd_cnt = (tp->sackhint.prr_delivered *
2597 							    tp->snd_ssthresh /
2598 							    tp->sackhint.recover_fs) +
2599 							    1 - tp->sackhint.sack_bytes_rexmit;
2600 						} else {
2601 							if (V_tcp_do_prr_conservative)
2602 								limit = tp->sackhint.prr_delivered -
2603 									tp->sackhint.sack_bytes_rexmit;
2604 							else
2605 								if ((tp->sackhint.prr_delivered -
2606 								    tp->sackhint.sack_bytes_rexmit) >
2607 								    del_data)
2608 									limit = tp->sackhint.prr_delivered -
2609 									    tp->sackhint.sack_bytes_rexmit +
2610 									    maxseg;
2611 								else
2612 									limit = del_data + maxseg;
2613 							if ((tp->snd_ssthresh - pipe) < limit)
2614 								snd_cnt = tp->snd_ssthresh - pipe;
2615 							else
2616 								snd_cnt = limit;
2617 						}
2618 						snd_cnt = max((snd_cnt / maxseg), 0);
2619 						/*
2620 						 * Send snd_cnt new data into the network in
2621 						 * response to this ACK. If there is a going
2622 						 * to be a SACK retransmission, adjust snd_cwnd
2623 						 * accordingly.
2624 						 */
2625 						tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
2626 						    tp->sackhint.sack_bytes_rexmit +
2627 						    (snd_cnt * maxseg);
2628 					} else if ((tp->t_flags & TF_SACK_PERMIT) &&
2629 					    IN_FASTRECOVERY(tp->t_flags)) {
2630 						int awnd;
2631 
2632 						/*
2633 						 * Compute the amount of data in flight first.
2634 						 * We can inject new data into the pipe iff
2635 						 * we have less than 1/2 the original window's
2636 						 * worth of data in flight.
2637 						 */
2638 						if (V_tcp_do_rfc6675_pipe)
2639 							awnd = tcp_compute_pipe(tp);
2640 						else
2641 							awnd = (tp->snd_nxt - tp->snd_fack) +
2642 								tp->sackhint.sack_bytes_rexmit;
2643 
2644 						if (awnd < tp->snd_ssthresh) {
2645 							tp->snd_cwnd += maxseg;
2646 							if (tp->snd_cwnd > tp->snd_ssthresh)
2647 								tp->snd_cwnd = tp->snd_ssthresh;
2648 						}
2649 					} else
2650 						tp->snd_cwnd += maxseg;
2651 					(void) tp->t_fb->tfb_tcp_output(tp);
2652 					goto drop;
2653 				} else if (tp->t_dupacks == tcprexmtthresh) {
2654 					tcp_seq onxt = tp->snd_nxt;
2655 
2656 					/*
2657 					 * If we're doing sack, or prr, check
2658 					 * to see if we're already in sack
2659 					 * recovery. If we're not doing sack,
2660 					 * check to see if we're in newreno
2661 					 * recovery.
2662 					 */
2663 					if (V_tcp_do_prr ||
2664 					    (tp->t_flags & TF_SACK_PERMIT)) {
2665 						if (IN_FASTRECOVERY(tp->t_flags)) {
2666 							tp->t_dupacks = 0;
2667 							break;
2668 						}
2669 					} else {
2670 						if (SEQ_LEQ(th->th_ack,
2671 						    tp->snd_recover)) {
2672 							tp->t_dupacks = 0;
2673 							break;
2674 						}
2675 					}
2676 					/* Congestion signal before ack. */
2677 					cc_cong_signal(tp, th, CC_NDUPACK);
2678 					cc_ack_received(tp, th, nsegs,
2679 					    CC_DUPACK);
2680 					tcp_timer_activate(tp, TT_REXMT, 0);
2681 					tp->t_rtttime = 0;
2682 					if (V_tcp_do_prr) {
2683 						/*
2684 						 * snd_ssthresh is already updated by
2685 						 * cc_cong_signal.
2686 						 */
2687 						tp->sackhint.prr_delivered = 0;
2688 						tp->sackhint.sack_bytes_rexmit = 0;
2689 						tp->sackhint.recover_fs = max(1,
2690 						    tp->snd_nxt - tp->snd_una);
2691 					}
2692 					if (tp->t_flags & TF_SACK_PERMIT) {
2693 						TCPSTAT_INC(
2694 						    tcps_sack_recovery_episode);
2695 						tp->snd_recover = tp->snd_nxt;
2696 						tp->snd_cwnd = maxseg;
2697 						(void) tp->t_fb->tfb_tcp_output(tp);
2698 						goto drop;
2699 					}
2700 					tp->snd_nxt = th->th_ack;
2701 					tp->snd_cwnd = maxseg;
2702 					(void) tp->t_fb->tfb_tcp_output(tp);
2703 					KASSERT(tp->snd_limited <= 2,
2704 					    ("%s: tp->snd_limited too big",
2705 					    __func__));
2706 					tp->snd_cwnd = tp->snd_ssthresh +
2707 					     maxseg *
2708 					     (tp->t_dupacks - tp->snd_limited);
2709 					if (SEQ_GT(onxt, tp->snd_nxt))
2710 						tp->snd_nxt = onxt;
2711 					goto drop;
2712 				} else if (V_tcp_do_rfc3042) {
2713 					/*
2714 					 * Process first and second duplicate
2715 					 * ACKs. Each indicates a segment
2716 					 * leaving the network, creating room
2717 					 * for more. Make sure we can send a
2718 					 * packet on reception of each duplicate
2719 					 * ACK by increasing snd_cwnd by one
2720 					 * segment. Restore the original
2721 					 * snd_cwnd after packet transmission.
2722 					 */
2723 					cc_ack_received(tp, th, nsegs,
2724 					    CC_DUPACK);
2725 					uint32_t oldcwnd = tp->snd_cwnd;
2726 					tcp_seq oldsndmax = tp->snd_max;
2727 					u_int sent;
2728 					int avail;
2729 
2730 					KASSERT(tp->t_dupacks == 1 ||
2731 					    tp->t_dupacks == 2,
2732 					    ("%s: dupacks not 1 or 2",
2733 					    __func__));
2734 					if (tp->t_dupacks == 1)
2735 						tp->snd_limited = 0;
2736 					tp->snd_cwnd =
2737 					    (tp->snd_nxt - tp->snd_una) +
2738 					    (tp->t_dupacks - tp->snd_limited) *
2739 					    maxseg;
2740 					/*
2741 					 * Only call tcp_output when there
2742 					 * is new data available to be sent.
2743 					 * Otherwise we would send pure ACKs.
2744 					 */
2745 					SOCKBUF_LOCK(&so->so_snd);
2746 					avail = sbavail(&so->so_snd) -
2747 					    (tp->snd_nxt - tp->snd_una);
2748 					SOCKBUF_UNLOCK(&so->so_snd);
2749 					if (avail > 0)
2750 						(void) tp->t_fb->tfb_tcp_output(tp);
2751 					sent = tp->snd_max - oldsndmax;
2752 					if (sent > maxseg) {
2753 						KASSERT((tp->t_dupacks == 2 &&
2754 						    tp->snd_limited == 0) ||
2755 						   (sent == maxseg + 1 &&
2756 						    tp->t_flags & TF_SENTFIN),
2757 						    ("%s: sent too much",
2758 						    __func__));
2759 						tp->snd_limited = 2;
2760 					} else if (sent > 0)
2761 						++tp->snd_limited;
2762 					tp->snd_cwnd = oldcwnd;
2763 					goto drop;
2764 				}
2765 			}
2766 			break;
2767 		} else {
2768 			/*
2769 			 * This ack is advancing the left edge, reset the
2770 			 * counter.
2771 			 */
2772 			tp->t_dupacks = 0;
2773 			/*
2774 			 * If this ack also has new SACK info, increment the
2775 			 * counter as per rfc6675. The variable
2776 			 * sack_changed tracks all changes to the SACK
2777 			 * scoreboard, including when partial ACKs without
2778 			 * SACK options are received, and clear the scoreboard
2779 			 * from the left side. Such partial ACKs should not be
2780 			 * counted as dupacks here.
2781 			 */
2782 			if ((tp->t_flags & TF_SACK_PERMIT) &&
2783 			    (to.to_flags & TOF_SACK) &&
2784 			    sack_changed)
2785 				tp->t_dupacks++;
2786 		}
2787 
2788 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2789 		    ("%s: th_ack <= snd_una", __func__));
2790 
2791 		/*
2792 		 * If the congestion window was inflated to account
2793 		 * for the other side's cached packets, retract it.
2794 		 */
2795 		if (IN_FASTRECOVERY(tp->t_flags)) {
2796 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2797 				if (tp->t_flags & TF_SACK_PERMIT)
2798 					if (V_tcp_do_prr)
2799 						tcp_prr_partialack(tp, th);
2800 					else
2801 						tcp_sack_partialack(tp, th);
2802 				else
2803 					tcp_newreno_partial_ack(tp, th);
2804 			} else
2805 				cc_post_recovery(tp, th);
2806 		}
2807 		/*
2808 		 * If we reach this point, ACK is not a duplicate,
2809 		 *     i.e., it ACKs something we sent.
2810 		 */
2811 		if (tp->t_flags & TF_NEEDSYN) {
2812 			/*
2813 			 * T/TCP: Connection was half-synchronized, and our
2814 			 * SYN has been ACK'd (so connection is now fully
2815 			 * synchronized).  Go to non-starred state,
2816 			 * increment snd_una for ACK of SYN, and check if
2817 			 * we can do window scaling.
2818 			 */
2819 			tp->t_flags &= ~TF_NEEDSYN;
2820 			tp->snd_una++;
2821 			/* Do window scaling? */
2822 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2823 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2824 				tp->rcv_scale = tp->request_r_scale;
2825 				/* Send window already scaled. */
2826 			}
2827 		}
2828 
2829 process_ACK:
2830 		INP_WLOCK_ASSERT(tp->t_inpcb);
2831 
2832 		/*
2833 		 * Adjust for the SYN bit in sequence space,
2834 		 * but don't account for it in cwnd calculations.
2835 		 * This is for the SYN_RECEIVED, non-simultaneous
2836 		 * SYN case. SYN_SENT and simultaneous SYN are
2837 		 * treated elsewhere.
2838 		 */
2839 		if (incforsyn)
2840 			tp->snd_una++;
2841 		acked = BYTES_THIS_ACK(tp, th);
2842 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2843 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2844 		    tp->snd_una, th->th_ack, tp, m));
2845 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2846 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2847 
2848 		/*
2849 		 * If we just performed our first retransmit, and the ACK
2850 		 * arrives within our recovery window, then it was a mistake
2851 		 * to do the retransmit in the first place.  Recover our
2852 		 * original cwnd and ssthresh, and proceed to transmit where
2853 		 * we left off.
2854 		 */
2855 		if (tp->t_rxtshift == 1 &&
2856 		    tp->t_flags & TF_PREVVALID &&
2857 		    tp->t_badrxtwin &&
2858 		    SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2859 			cc_cong_signal(tp, th, CC_RTO_ERR);
2860 
2861 		/*
2862 		 * If we have a timestamp reply, update smoothed
2863 		 * round trip time.  If no timestamp is present but
2864 		 * transmit timer is running and timed sequence
2865 		 * number was acked, update smoothed round trip time.
2866 		 * Since we now have an rtt measurement, cancel the
2867 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2868 		 * Recompute the initial retransmit timer.
2869 		 *
2870 		 * Some boxes send broken timestamp replies
2871 		 * during the SYN+ACK phase, ignore
2872 		 * timestamps of 0 or we could calculate a
2873 		 * huge RTT and blow up the retransmit timer.
2874 		 */
2875 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2876 			uint32_t t;
2877 
2878 			t = tcp_ts_getticks() - to.to_tsecr;
2879 			if (!tp->t_rttlow || tp->t_rttlow > t)
2880 				tp->t_rttlow = t;
2881 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2882 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2883 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2884 				tp->t_rttlow = ticks - tp->t_rtttime;
2885 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2886 		}
2887 
2888 		/*
2889 		 * If all outstanding data is acked, stop retransmit
2890 		 * timer and remember to restart (more output or persist).
2891 		 * If there is more data to be acked, restart retransmit
2892 		 * timer, using current (possibly backed-off) value.
2893 		 */
2894 		if (th->th_ack == tp->snd_max) {
2895 			tcp_timer_activate(tp, TT_REXMT, 0);
2896 			needoutput = 1;
2897 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2898 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2899 
2900 		/*
2901 		 * If no data (only SYN) was ACK'd,
2902 		 *    skip rest of ACK processing.
2903 		 */
2904 		if (acked == 0)
2905 			goto step6;
2906 
2907 		/*
2908 		 * Let the congestion control algorithm update congestion
2909 		 * control related information. This typically means increasing
2910 		 * the congestion window.
2911 		 */
2912 		cc_ack_received(tp, th, nsegs, CC_ACK);
2913 
2914 		SOCKBUF_LOCK(&so->so_snd);
2915 		if (acked > sbavail(&so->so_snd)) {
2916 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2917 				tp->snd_wnd -= sbavail(&so->so_snd);
2918 			else
2919 				tp->snd_wnd = 0;
2920 			mfree = sbcut_locked(&so->so_snd,
2921 			    (int)sbavail(&so->so_snd));
2922 			ourfinisacked = 1;
2923 		} else {
2924 			mfree = sbcut_locked(&so->so_snd, acked);
2925 			if (tp->snd_wnd >= (uint32_t) acked)
2926 				tp->snd_wnd -= acked;
2927 			else
2928 				tp->snd_wnd = 0;
2929 			ourfinisacked = 0;
2930 		}
2931 		SOCKBUF_UNLOCK(&so->so_snd);
2932 		tp->t_flags |= TF_WAKESOW;
2933 		m_freem(mfree);
2934 		/* Detect una wraparound. */
2935 		if (!IN_RECOVERY(tp->t_flags) &&
2936 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2937 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2938 			tp->snd_recover = th->th_ack - 1;
2939 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2940 		if (IN_RECOVERY(tp->t_flags) &&
2941 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2942 			EXIT_RECOVERY(tp->t_flags);
2943 		}
2944 		tp->snd_una = th->th_ack;
2945 		if (tp->t_flags & TF_SACK_PERMIT) {
2946 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2947 				tp->snd_recover = tp->snd_una;
2948 		}
2949 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2950 			tp->snd_nxt = tp->snd_una;
2951 
2952 		switch (tp->t_state) {
2953 		/*
2954 		 * In FIN_WAIT_1 STATE in addition to the processing
2955 		 * for the ESTABLISHED state if our FIN is now acknowledged
2956 		 * then enter FIN_WAIT_2.
2957 		 */
2958 		case TCPS_FIN_WAIT_1:
2959 			if (ourfinisacked) {
2960 				/*
2961 				 * If we can't receive any more
2962 				 * data, then closing user can proceed.
2963 				 * Starting the timer is contrary to the
2964 				 * specification, but if we don't get a FIN
2965 				 * we'll hang forever.
2966 				 *
2967 				 * XXXjl:
2968 				 * we should release the tp also, and use a
2969 				 * compressed state.
2970 				 */
2971 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2972 					soisdisconnected(so);
2973 					tcp_timer_activate(tp, TT_2MSL,
2974 					    (tcp_fast_finwait2_recycle ?
2975 					    tcp_finwait2_timeout :
2976 					    TP_MAXIDLE(tp)));
2977 				}
2978 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
2979 			}
2980 			break;
2981 
2982 		/*
2983 		 * In CLOSING STATE in addition to the processing for
2984 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2985 		 * then enter the TIME-WAIT state, otherwise ignore
2986 		 * the segment.
2987 		 */
2988 		case TCPS_CLOSING:
2989 			if (ourfinisacked) {
2990 				tcp_twstart(tp);
2991 				m_freem(m);
2992 				return;
2993 			}
2994 			break;
2995 
2996 		/*
2997 		 * In LAST_ACK, we may still be waiting for data to drain
2998 		 * and/or to be acked, as well as for the ack of our FIN.
2999 		 * If our FIN is now acknowledged, delete the TCB,
3000 		 * enter the closed state and return.
3001 		 */
3002 		case TCPS_LAST_ACK:
3003 			if (ourfinisacked) {
3004 				tp = tcp_close(tp);
3005 				goto drop;
3006 			}
3007 			break;
3008 		}
3009 	}
3010 
3011 step6:
3012 	INP_WLOCK_ASSERT(tp->t_inpcb);
3013 
3014 	/*
3015 	 * Update window information.
3016 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3017 	 */
3018 	if ((thflags & TH_ACK) &&
3019 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3020 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3021 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3022 		/* keep track of pure window updates */
3023 		if (tlen == 0 &&
3024 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3025 			TCPSTAT_INC(tcps_rcvwinupd);
3026 		tp->snd_wnd = tiwin;
3027 		tp->snd_wl1 = th->th_seq;
3028 		tp->snd_wl2 = th->th_ack;
3029 		if (tp->snd_wnd > tp->max_sndwnd)
3030 			tp->max_sndwnd = tp->snd_wnd;
3031 		needoutput = 1;
3032 	}
3033 
3034 	/*
3035 	 * Process segments with URG.
3036 	 */
3037 	if ((thflags & TH_URG) && th->th_urp &&
3038 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3039 		/*
3040 		 * This is a kludge, but if we receive and accept
3041 		 * random urgent pointers, we'll crash in
3042 		 * soreceive.  It's hard to imagine someone
3043 		 * actually wanting to send this much urgent data.
3044 		 */
3045 		SOCKBUF_LOCK(&so->so_rcv);
3046 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3047 			th->th_urp = 0;			/* XXX */
3048 			thflags &= ~TH_URG;		/* XXX */
3049 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3050 			goto dodata;			/* XXX */
3051 		}
3052 		/*
3053 		 * If this segment advances the known urgent pointer,
3054 		 * then mark the data stream.  This should not happen
3055 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3056 		 * a FIN has been received from the remote side.
3057 		 * In these states we ignore the URG.
3058 		 *
3059 		 * According to RFC961 (Assigned Protocols),
3060 		 * the urgent pointer points to the last octet
3061 		 * of urgent data.  We continue, however,
3062 		 * to consider it to indicate the first octet
3063 		 * of data past the urgent section as the original
3064 		 * spec states (in one of two places).
3065 		 */
3066 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3067 			tp->rcv_up = th->th_seq + th->th_urp;
3068 			so->so_oobmark = sbavail(&so->so_rcv) +
3069 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3070 			if (so->so_oobmark == 0)
3071 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3072 			sohasoutofband(so);
3073 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3074 		}
3075 		SOCKBUF_UNLOCK(&so->so_rcv);
3076 		/*
3077 		 * Remove out of band data so doesn't get presented to user.
3078 		 * This can happen independent of advancing the URG pointer,
3079 		 * but if two URG's are pending at once, some out-of-band
3080 		 * data may creep in... ick.
3081 		 */
3082 		if (th->th_urp <= (uint32_t)tlen &&
3083 		    !(so->so_options & SO_OOBINLINE)) {
3084 			/* hdr drop is delayed */
3085 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3086 		}
3087 	} else {
3088 		/*
3089 		 * If no out of band data is expected,
3090 		 * pull receive urgent pointer along
3091 		 * with the receive window.
3092 		 */
3093 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3094 			tp->rcv_up = tp->rcv_nxt;
3095 	}
3096 dodata:							/* XXX */
3097 	INP_WLOCK_ASSERT(tp->t_inpcb);
3098 
3099 	/*
3100 	 * Process the segment text, merging it into the TCP sequencing queue,
3101 	 * and arranging for acknowledgment of receipt if necessary.
3102 	 * This process logically involves adjusting tp->rcv_wnd as data
3103 	 * is presented to the user (this happens in tcp_usrreq.c,
3104 	 * case PRU_RCVD).  If a FIN has already been received on this
3105 	 * connection then we just ignore the text.
3106 	 */
3107 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3108 		   IS_FASTOPEN(tp->t_flags));
3109 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3110 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3111 		tcp_seq save_start = th->th_seq;
3112 		tcp_seq save_rnxt  = tp->rcv_nxt;
3113 		int     save_tlen  = tlen;
3114 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3115 		/*
3116 		 * Insert segment which includes th into TCP reassembly queue
3117 		 * with control block tp.  Set thflags to whether reassembly now
3118 		 * includes a segment with FIN.  This handles the common case
3119 		 * inline (segment is the next to be received on an established
3120 		 * connection, and the queue is empty), avoiding linkage into
3121 		 * and removal from the queue and repetition of various
3122 		 * conversions.
3123 		 * Set DELACK for segments received in order, but ack
3124 		 * immediately when segments are out of order (so
3125 		 * fast retransmit can work).
3126 		 */
3127 		if (th->th_seq == tp->rcv_nxt &&
3128 		    SEGQ_EMPTY(tp) &&
3129 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3130 		     tfo_syn)) {
3131 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3132 				tp->t_flags |= TF_DELACK;
3133 			else
3134 				tp->t_flags |= TF_ACKNOW;
3135 			tp->rcv_nxt += tlen;
3136 			if (tlen &&
3137 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3138 			    (tp->t_fbyte_in == 0)) {
3139 				tp->t_fbyte_in = ticks;
3140 				if (tp->t_fbyte_in == 0)
3141 					tp->t_fbyte_in = 1;
3142 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3143 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3144 			}
3145 			thflags = th->th_flags & TH_FIN;
3146 			TCPSTAT_INC(tcps_rcvpack);
3147 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3148 			SOCKBUF_LOCK(&so->so_rcv);
3149 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3150 				m_freem(m);
3151 			else
3152 				sbappendstream_locked(&so->so_rcv, m, 0);
3153 			SOCKBUF_UNLOCK(&so->so_rcv);
3154 			tp->t_flags |= TF_WAKESOR;
3155 		} else {
3156 			/*
3157 			 * XXX: Due to the header drop above "th" is
3158 			 * theoretically invalid by now.  Fortunately
3159 			 * m_adj() doesn't actually frees any mbufs
3160 			 * when trimming from the head.
3161 			 */
3162 			tcp_seq temp = save_start;
3163 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3164 			tp->t_flags |= TF_ACKNOW;
3165 		}
3166 		if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3167 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3168 				/*
3169 				 * DSACK actually handled in the fastpath
3170 				 * above.
3171 				 */
3172 				tcp_update_sack_list(tp, save_start,
3173 				    save_start + save_tlen);
3174 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3175 				if ((tp->rcv_numsacks >= 1) &&
3176 				    (tp->sackblks[0].end == save_start)) {
3177 					/*
3178 					 * Partial overlap, recorded at todrop
3179 					 * above.
3180 					 */
3181 					tcp_update_sack_list(tp,
3182 					    tp->sackblks[0].start,
3183 					    tp->sackblks[0].end);
3184 				} else {
3185 					tcp_update_dsack_list(tp, save_start,
3186 					    save_start + save_tlen);
3187 				}
3188 			} else if (tlen >= save_tlen) {
3189 				/* Update of sackblks. */
3190 				tcp_update_dsack_list(tp, save_start,
3191 				    save_start + save_tlen);
3192 			} else if (tlen > 0) {
3193 				tcp_update_dsack_list(tp, save_start,
3194 				    save_start + tlen);
3195 			}
3196 		}
3197 #if 0
3198 		/*
3199 		 * Note the amount of data that peer has sent into
3200 		 * our window, in order to estimate the sender's
3201 		 * buffer size.
3202 		 * XXX: Unused.
3203 		 */
3204 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3205 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3206 		else
3207 			len = so->so_rcv.sb_hiwat;
3208 #endif
3209 	} else {
3210 		m_freem(m);
3211 		thflags &= ~TH_FIN;
3212 	}
3213 
3214 	/*
3215 	 * If FIN is received ACK the FIN and let the user know
3216 	 * that the connection is closing.
3217 	 */
3218 	if (thflags & TH_FIN) {
3219 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3220 			socantrcvmore(so);
3221 			/* The socket upcall is handled by socantrcvmore. */
3222 			tp->t_flags &= ~TF_WAKESOR;
3223 			/*
3224 			 * If connection is half-synchronized
3225 			 * (ie NEEDSYN flag on) then delay ACK,
3226 			 * so it may be piggybacked when SYN is sent.
3227 			 * Otherwise, since we received a FIN then no
3228 			 * more input can be expected, send ACK now.
3229 			 */
3230 			if (tp->t_flags & TF_NEEDSYN)
3231 				tp->t_flags |= TF_DELACK;
3232 			else
3233 				tp->t_flags |= TF_ACKNOW;
3234 			tp->rcv_nxt++;
3235 		}
3236 		switch (tp->t_state) {
3237 		/*
3238 		 * In SYN_RECEIVED and ESTABLISHED STATES
3239 		 * enter the CLOSE_WAIT state.
3240 		 */
3241 		case TCPS_SYN_RECEIVED:
3242 			tp->t_starttime = ticks;
3243 			/* FALLTHROUGH */
3244 		case TCPS_ESTABLISHED:
3245 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3246 			break;
3247 
3248 		/*
3249 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3250 		 * enter the CLOSING state.
3251 		 */
3252 		case TCPS_FIN_WAIT_1:
3253 			tcp_state_change(tp, TCPS_CLOSING);
3254 			break;
3255 
3256 		/*
3257 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3258 		 * starting the time-wait timer, turning off the other
3259 		 * standard timers.
3260 		 */
3261 		case TCPS_FIN_WAIT_2:
3262 			tcp_twstart(tp);
3263 			return;
3264 		}
3265 	}
3266 #ifdef TCPDEBUG
3267 	if (so->so_options & SO_DEBUG)
3268 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3269 			  &tcp_savetcp, 0);
3270 #endif
3271 	TCP_PROBE3(debug__input, tp, th, m);
3272 
3273 	/*
3274 	 * Return any desired output.
3275 	 */
3276 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3277 		(void) tp->t_fb->tfb_tcp_output(tp);
3278 
3279 check_delack:
3280 	INP_WLOCK_ASSERT(tp->t_inpcb);
3281 
3282 	if (tp->t_flags & TF_DELACK) {
3283 		tp->t_flags &= ~TF_DELACK;
3284 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3285 	}
3286 	tcp_handle_wakeup(tp, so);
3287 	INP_WUNLOCK(tp->t_inpcb);
3288 	return;
3289 
3290 dropafterack:
3291 	/*
3292 	 * Generate an ACK dropping incoming segment if it occupies
3293 	 * sequence space, where the ACK reflects our state.
3294 	 *
3295 	 * We can now skip the test for the RST flag since all
3296 	 * paths to this code happen after packets containing
3297 	 * RST have been dropped.
3298 	 *
3299 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3300 	 * segment we received passes the SYN-RECEIVED ACK test.
3301 	 * If it fails send a RST.  This breaks the loop in the
3302 	 * "LAND" DoS attack, and also prevents an ACK storm
3303 	 * between two listening ports that have been sent forged
3304 	 * SYN segments, each with the source address of the other.
3305 	 */
3306 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3307 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3308 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3309 		rstreason = BANDLIM_RST_OPENPORT;
3310 		goto dropwithreset;
3311 	}
3312 #ifdef TCPDEBUG
3313 	if (so->so_options & SO_DEBUG)
3314 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3315 			  &tcp_savetcp, 0);
3316 #endif
3317 	TCP_PROBE3(debug__input, tp, th, m);
3318 	tp->t_flags |= TF_ACKNOW;
3319 	(void) tp->t_fb->tfb_tcp_output(tp);
3320 	tcp_handle_wakeup(tp, so);
3321 	INP_WUNLOCK(tp->t_inpcb);
3322 	m_freem(m);
3323 	return;
3324 
3325 dropwithreset:
3326 	if (tp != NULL) {
3327 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3328 		tcp_handle_wakeup(tp, so);
3329 		INP_WUNLOCK(tp->t_inpcb);
3330 	} else
3331 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3332 	return;
3333 
3334 drop:
3335 	/*
3336 	 * Drop space held by incoming segment and return.
3337 	 */
3338 #ifdef TCPDEBUG
3339 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3340 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3341 			  &tcp_savetcp, 0);
3342 #endif
3343 	TCP_PROBE3(debug__input, tp, th, m);
3344 	if (tp != NULL) {
3345 		tcp_handle_wakeup(tp, so);
3346 		INP_WUNLOCK(tp->t_inpcb);
3347 	}
3348 	m_freem(m);
3349 }
3350 
3351 /*
3352  * Issue RST and make ACK acceptable to originator of segment.
3353  * The mbuf must still include the original packet header.
3354  * tp may be NULL.
3355  */
3356 void
3357 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3358     int tlen, int rstreason)
3359 {
3360 #ifdef INET
3361 	struct ip *ip;
3362 #endif
3363 #ifdef INET6
3364 	struct ip6_hdr *ip6;
3365 #endif
3366 
3367 	if (tp != NULL) {
3368 		INP_WLOCK_ASSERT(tp->t_inpcb);
3369 	}
3370 
3371 	/* Don't bother if destination was broadcast/multicast. */
3372 	if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3373 		goto drop;
3374 #ifdef INET6
3375 	if (mtod(m, struct ip *)->ip_v == 6) {
3376 		ip6 = mtod(m, struct ip6_hdr *);
3377 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3378 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3379 			goto drop;
3380 		/* IPv6 anycast check is done at tcp6_input() */
3381 	}
3382 #endif
3383 #if defined(INET) && defined(INET6)
3384 	else
3385 #endif
3386 #ifdef INET
3387 	{
3388 		ip = mtod(m, struct ip *);
3389 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3390 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3391 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3392 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3393 			goto drop;
3394 	}
3395 #endif
3396 
3397 	/* Perform bandwidth limiting. */
3398 	if (badport_bandlim(rstreason) < 0)
3399 		goto drop;
3400 
3401 	/* tcp_respond consumes the mbuf chain. */
3402 	if (th->th_flags & TH_ACK) {
3403 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3404 		    th->th_ack, TH_RST);
3405 	} else {
3406 		if (th->th_flags & TH_SYN)
3407 			tlen++;
3408 		if (th->th_flags & TH_FIN)
3409 			tlen++;
3410 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3411 		    (tcp_seq)0, TH_RST|TH_ACK);
3412 	}
3413 	return;
3414 drop:
3415 	m_freem(m);
3416 }
3417 
3418 /*
3419  * Parse TCP options and place in tcpopt.
3420  */
3421 void
3422 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3423 {
3424 	int opt, optlen;
3425 
3426 	to->to_flags = 0;
3427 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3428 		opt = cp[0];
3429 		if (opt == TCPOPT_EOL)
3430 			break;
3431 		if (opt == TCPOPT_NOP)
3432 			optlen = 1;
3433 		else {
3434 			if (cnt < 2)
3435 				break;
3436 			optlen = cp[1];
3437 			if (optlen < 2 || optlen > cnt)
3438 				break;
3439 		}
3440 		switch (opt) {
3441 		case TCPOPT_MAXSEG:
3442 			if (optlen != TCPOLEN_MAXSEG)
3443 				continue;
3444 			if (!(flags & TO_SYN))
3445 				continue;
3446 			to->to_flags |= TOF_MSS;
3447 			bcopy((char *)cp + 2,
3448 			    (char *)&to->to_mss, sizeof(to->to_mss));
3449 			to->to_mss = ntohs(to->to_mss);
3450 			break;
3451 		case TCPOPT_WINDOW:
3452 			if (optlen != TCPOLEN_WINDOW)
3453 				continue;
3454 			if (!(flags & TO_SYN))
3455 				continue;
3456 			to->to_flags |= TOF_SCALE;
3457 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3458 			break;
3459 		case TCPOPT_TIMESTAMP:
3460 			if (optlen != TCPOLEN_TIMESTAMP)
3461 				continue;
3462 			to->to_flags |= TOF_TS;
3463 			bcopy((char *)cp + 2,
3464 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3465 			to->to_tsval = ntohl(to->to_tsval);
3466 			bcopy((char *)cp + 6,
3467 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3468 			to->to_tsecr = ntohl(to->to_tsecr);
3469 			break;
3470 		case TCPOPT_SIGNATURE:
3471 			/*
3472 			 * In order to reply to a host which has set the
3473 			 * TCP_SIGNATURE option in its initial SYN, we have
3474 			 * to record the fact that the option was observed
3475 			 * here for the syncache code to perform the correct
3476 			 * response.
3477 			 */
3478 			if (optlen != TCPOLEN_SIGNATURE)
3479 				continue;
3480 			to->to_flags |= TOF_SIGNATURE;
3481 			to->to_signature = cp + 2;
3482 			break;
3483 		case TCPOPT_SACK_PERMITTED:
3484 			if (optlen != TCPOLEN_SACK_PERMITTED)
3485 				continue;
3486 			if (!(flags & TO_SYN))
3487 				continue;
3488 			if (!V_tcp_do_sack)
3489 				continue;
3490 			to->to_flags |= TOF_SACKPERM;
3491 			break;
3492 		case TCPOPT_SACK:
3493 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3494 				continue;
3495 			if (flags & TO_SYN)
3496 				continue;
3497 			to->to_flags |= TOF_SACK;
3498 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3499 			to->to_sacks = cp + 2;
3500 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3501 			break;
3502 		case TCPOPT_FAST_OPEN:
3503 			/*
3504 			 * Cookie length validation is performed by the
3505 			 * server side cookie checking code or the client
3506 			 * side cookie cache update code.
3507 			 */
3508 			if (!(flags & TO_SYN))
3509 				continue;
3510 			if (!V_tcp_fastopen_client_enable &&
3511 			    !V_tcp_fastopen_server_enable)
3512 				continue;
3513 			to->to_flags |= TOF_FASTOPEN;
3514 			to->to_tfo_len = optlen - 2;
3515 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3516 			break;
3517 		default:
3518 			continue;
3519 		}
3520 	}
3521 }
3522 
3523 /*
3524  * Pull out of band byte out of a segment so
3525  * it doesn't appear in the user's data queue.
3526  * It is still reflected in the segment length for
3527  * sequencing purposes.
3528  */
3529 void
3530 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3531     int off)
3532 {
3533 	int cnt = off + th->th_urp - 1;
3534 
3535 	while (cnt >= 0) {
3536 		if (m->m_len > cnt) {
3537 			char *cp = mtod(m, caddr_t) + cnt;
3538 			struct tcpcb *tp = sototcpcb(so);
3539 
3540 			INP_WLOCK_ASSERT(tp->t_inpcb);
3541 
3542 			tp->t_iobc = *cp;
3543 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3544 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3545 			m->m_len--;
3546 			if (m->m_flags & M_PKTHDR)
3547 				m->m_pkthdr.len--;
3548 			return;
3549 		}
3550 		cnt -= m->m_len;
3551 		m = m->m_next;
3552 		if (m == NULL)
3553 			break;
3554 	}
3555 	panic("tcp_pulloutofband");
3556 }
3557 
3558 /*
3559  * Collect new round-trip time estimate
3560  * and update averages and current timeout.
3561  */
3562 void
3563 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3564 {
3565 	int delta;
3566 
3567 	INP_WLOCK_ASSERT(tp->t_inpcb);
3568 
3569 	TCPSTAT_INC(tcps_rttupdated);
3570 	tp->t_rttupdated++;
3571 #ifdef STATS
3572 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3573 	    imax(0, rtt * 1000 / hz));
3574 #endif
3575 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3576 		/*
3577 		 * srtt is stored as fixed point with 5 bits after the
3578 		 * binary point (i.e., scaled by 8).  The following magic
3579 		 * is equivalent to the smoothing algorithm in rfc793 with
3580 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3581 		 * point).  Adjust rtt to origin 0.
3582 		 */
3583 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3584 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3585 
3586 		if ((tp->t_srtt += delta) <= 0)
3587 			tp->t_srtt = 1;
3588 
3589 		/*
3590 		 * We accumulate a smoothed rtt variance (actually, a
3591 		 * smoothed mean difference), then set the retransmit
3592 		 * timer to smoothed rtt + 4 times the smoothed variance.
3593 		 * rttvar is stored as fixed point with 4 bits after the
3594 		 * binary point (scaled by 16).  The following is
3595 		 * equivalent to rfc793 smoothing with an alpha of .75
3596 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3597 		 * rfc793's wired-in beta.
3598 		 */
3599 		if (delta < 0)
3600 			delta = -delta;
3601 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3602 		if ((tp->t_rttvar += delta) <= 0)
3603 			tp->t_rttvar = 1;
3604 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3605 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3606 	} else {
3607 		/*
3608 		 * No rtt measurement yet - use the unsmoothed rtt.
3609 		 * Set the variance to half the rtt (so our first
3610 		 * retransmit happens at 3*rtt).
3611 		 */
3612 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3613 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3614 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3615 	}
3616 	tp->t_rtttime = 0;
3617 	tp->t_rxtshift = 0;
3618 
3619 	/*
3620 	 * the retransmit should happen at rtt + 4 * rttvar.
3621 	 * Because of the way we do the smoothing, srtt and rttvar
3622 	 * will each average +1/2 tick of bias.  When we compute
3623 	 * the retransmit timer, we want 1/2 tick of rounding and
3624 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3625 	 * firing of the timer.  The bias will give us exactly the
3626 	 * 1.5 tick we need.  But, because the bias is
3627 	 * statistical, we have to test that we don't drop below
3628 	 * the minimum feasible timer (which is 2 ticks).
3629 	 */
3630 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3631 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3632 
3633 	/*
3634 	 * We received an ack for a packet that wasn't retransmitted;
3635 	 * it is probably safe to discard any error indications we've
3636 	 * received recently.  This isn't quite right, but close enough
3637 	 * for now (a route might have failed after we sent a segment,
3638 	 * and the return path might not be symmetrical).
3639 	 */
3640 	tp->t_softerror = 0;
3641 }
3642 
3643 /*
3644  * Determine a reasonable value for maxseg size.
3645  * If the route is known, check route for mtu.
3646  * If none, use an mss that can be handled on the outgoing interface
3647  * without forcing IP to fragment.  If no route is found, route has no mtu,
3648  * or the destination isn't local, use a default, hopefully conservative
3649  * size (usually 512 or the default IP max size, but no more than the mtu
3650  * of the interface), as we can't discover anything about intervening
3651  * gateways or networks.  We also initialize the congestion/slow start
3652  * window to be a single segment if the destination isn't local.
3653  * While looking at the routing entry, we also initialize other path-dependent
3654  * parameters from pre-set or cached values in the routing entry.
3655  *
3656  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3657  * IP options, e.g. IPSEC data, since length of this data may vary, and
3658  * thus it is calculated for every segment separately in tcp_output().
3659  *
3660  * NOTE that this routine is only called when we process an incoming
3661  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3662  * settings are handled in tcp_mssopt().
3663  */
3664 void
3665 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3666     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3667 {
3668 	int mss = 0;
3669 	uint32_t maxmtu = 0;
3670 	struct inpcb *inp = tp->t_inpcb;
3671 	struct hc_metrics_lite metrics;
3672 #ifdef INET6
3673 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3674 	size_t min_protoh = isipv6 ?
3675 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3676 			    sizeof (struct tcpiphdr);
3677 #else
3678 	const size_t min_protoh = sizeof(struct tcpiphdr);
3679 #endif
3680 
3681 	INP_WLOCK_ASSERT(tp->t_inpcb);
3682 
3683 	if (mtuoffer != -1) {
3684 		KASSERT(offer == -1, ("%s: conflict", __func__));
3685 		offer = mtuoffer - min_protoh;
3686 	}
3687 
3688 	/* Initialize. */
3689 #ifdef INET6
3690 	if (isipv6) {
3691 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3692 		tp->t_maxseg = V_tcp_v6mssdflt;
3693 	}
3694 #endif
3695 #if defined(INET) && defined(INET6)
3696 	else
3697 #endif
3698 #ifdef INET
3699 	{
3700 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3701 		tp->t_maxseg = V_tcp_mssdflt;
3702 	}
3703 #endif
3704 
3705 	/*
3706 	 * No route to sender, stay with default mss and return.
3707 	 */
3708 	if (maxmtu == 0) {
3709 		/*
3710 		 * In case we return early we need to initialize metrics
3711 		 * to a defined state as tcp_hc_get() would do for us
3712 		 * if there was no cache hit.
3713 		 */
3714 		if (metricptr != NULL)
3715 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3716 		return;
3717 	}
3718 
3719 	/* What have we got? */
3720 	switch (offer) {
3721 		case 0:
3722 			/*
3723 			 * Offer == 0 means that there was no MSS on the SYN
3724 			 * segment, in this case we use tcp_mssdflt as
3725 			 * already assigned to t_maxseg above.
3726 			 */
3727 			offer = tp->t_maxseg;
3728 			break;
3729 
3730 		case -1:
3731 			/*
3732 			 * Offer == -1 means that we didn't receive SYN yet.
3733 			 */
3734 			/* FALLTHROUGH */
3735 
3736 		default:
3737 			/*
3738 			 * Prevent DoS attack with too small MSS. Round up
3739 			 * to at least minmss.
3740 			 */
3741 			offer = max(offer, V_tcp_minmss);
3742 	}
3743 
3744 	/*
3745 	 * rmx information is now retrieved from tcp_hostcache.
3746 	 */
3747 	tcp_hc_get(&inp->inp_inc, &metrics);
3748 	if (metricptr != NULL)
3749 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3750 
3751 	/*
3752 	 * If there's a discovered mtu in tcp hostcache, use it.
3753 	 * Else, use the link mtu.
3754 	 */
3755 	if (metrics.rmx_mtu)
3756 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3757 	else {
3758 #ifdef INET6
3759 		if (isipv6) {
3760 			mss = maxmtu - min_protoh;
3761 			if (!V_path_mtu_discovery &&
3762 			    !in6_localaddr(&inp->in6p_faddr))
3763 				mss = min(mss, V_tcp_v6mssdflt);
3764 		}
3765 #endif
3766 #if defined(INET) && defined(INET6)
3767 		else
3768 #endif
3769 #ifdef INET
3770 		{
3771 			mss = maxmtu - min_protoh;
3772 			if (!V_path_mtu_discovery &&
3773 			    !in_localaddr(inp->inp_faddr))
3774 				mss = min(mss, V_tcp_mssdflt);
3775 		}
3776 #endif
3777 		/*
3778 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3779 		 * probably violates the TCP spec.
3780 		 * The problem is that, since we don't know the
3781 		 * other end's MSS, we are supposed to use a conservative
3782 		 * default.  But, if we do that, then MTU discovery will
3783 		 * never actually take place, because the conservative
3784 		 * default is much less than the MTUs typically seen
3785 		 * on the Internet today.  For the moment, we'll sweep
3786 		 * this under the carpet.
3787 		 *
3788 		 * The conservative default might not actually be a problem
3789 		 * if the only case this occurs is when sending an initial
3790 		 * SYN with options and data to a host we've never talked
3791 		 * to before.  Then, they will reply with an MSS value which
3792 		 * will get recorded and the new parameters should get
3793 		 * recomputed.  For Further Study.
3794 		 */
3795 	}
3796 	mss = min(mss, offer);
3797 
3798 	/*
3799 	 * Sanity check: make sure that maxseg will be large
3800 	 * enough to allow some data on segments even if the
3801 	 * all the option space is used (40bytes).  Otherwise
3802 	 * funny things may happen in tcp_output.
3803 	 *
3804 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3805 	 */
3806 	mss = max(mss, 64);
3807 
3808 	tp->t_maxseg = mss;
3809 }
3810 
3811 void
3812 tcp_mss(struct tcpcb *tp, int offer)
3813 {
3814 	int mss;
3815 	uint32_t bufsize;
3816 	struct inpcb *inp;
3817 	struct socket *so;
3818 	struct hc_metrics_lite metrics;
3819 	struct tcp_ifcap cap;
3820 
3821 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3822 
3823 	bzero(&cap, sizeof(cap));
3824 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3825 
3826 	mss = tp->t_maxseg;
3827 	inp = tp->t_inpcb;
3828 
3829 	/*
3830 	 * If there's a pipesize, change the socket buffer to that size,
3831 	 * don't change if sb_hiwat is different than default (then it
3832 	 * has been changed on purpose with setsockopt).
3833 	 * Make the socket buffers an integral number of mss units;
3834 	 * if the mss is larger than the socket buffer, decrease the mss.
3835 	 */
3836 	so = inp->inp_socket;
3837 	SOCKBUF_LOCK(&so->so_snd);
3838 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3839 		bufsize = metrics.rmx_sendpipe;
3840 	else
3841 		bufsize = so->so_snd.sb_hiwat;
3842 	if (bufsize < mss)
3843 		mss = bufsize;
3844 	else {
3845 		bufsize = roundup(bufsize, mss);
3846 		if (bufsize > sb_max)
3847 			bufsize = sb_max;
3848 		if (bufsize > so->so_snd.sb_hiwat)
3849 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3850 	}
3851 	SOCKBUF_UNLOCK(&so->so_snd);
3852 	/*
3853 	 * Sanity check: make sure that maxseg will be large
3854 	 * enough to allow some data on segments even if the
3855 	 * all the option space is used (40bytes).  Otherwise
3856 	 * funny things may happen in tcp_output.
3857 	 *
3858 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3859 	 */
3860 	tp->t_maxseg = max(mss, 64);
3861 
3862 	SOCKBUF_LOCK(&so->so_rcv);
3863 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3864 		bufsize = metrics.rmx_recvpipe;
3865 	else
3866 		bufsize = so->so_rcv.sb_hiwat;
3867 	if (bufsize > mss) {
3868 		bufsize = roundup(bufsize, mss);
3869 		if (bufsize > sb_max)
3870 			bufsize = sb_max;
3871 		if (bufsize > so->so_rcv.sb_hiwat)
3872 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3873 	}
3874 	SOCKBUF_UNLOCK(&so->so_rcv);
3875 
3876 	/* Check the interface for TSO capabilities. */
3877 	if (cap.ifcap & CSUM_TSO) {
3878 		tp->t_flags |= TF_TSO;
3879 		tp->t_tsomax = cap.tsomax;
3880 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3881 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3882 	}
3883 }
3884 
3885 /*
3886  * Determine the MSS option to send on an outgoing SYN.
3887  */
3888 int
3889 tcp_mssopt(struct in_conninfo *inc)
3890 {
3891 	int mss = 0;
3892 	uint32_t thcmtu = 0;
3893 	uint32_t maxmtu = 0;
3894 	size_t min_protoh;
3895 
3896 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3897 
3898 #ifdef INET6
3899 	if (inc->inc_flags & INC_ISIPV6) {
3900 		mss = V_tcp_v6mssdflt;
3901 		maxmtu = tcp_maxmtu6(inc, NULL);
3902 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3903 	}
3904 #endif
3905 #if defined(INET) && defined(INET6)
3906 	else
3907 #endif
3908 #ifdef INET
3909 	{
3910 		mss = V_tcp_mssdflt;
3911 		maxmtu = tcp_maxmtu(inc, NULL);
3912 		min_protoh = sizeof(struct tcpiphdr);
3913 	}
3914 #endif
3915 #if defined(INET6) || defined(INET)
3916 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3917 #endif
3918 
3919 	if (maxmtu && thcmtu)
3920 		mss = min(maxmtu, thcmtu) - min_protoh;
3921 	else if (maxmtu || thcmtu)
3922 		mss = max(maxmtu, thcmtu) - min_protoh;
3923 
3924 	return (mss);
3925 }
3926 
3927 void
3928 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th)
3929 {
3930 	long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3931 	int maxseg = tcp_maxseg(tp);
3932 
3933 	INP_WLOCK_ASSERT(tp->t_inpcb);
3934 
3935 	tcp_timer_activate(tp, TT_REXMT, 0);
3936 	tp->t_rtttime = 0;
3937 	/*
3938 	 * Compute the amount of data that this ACK is indicating
3939 	 * (del_data) and an estimate of how many bytes are in the
3940 	 * network.
3941 	 */
3942 	if (SEQ_GEQ(th->th_ack, tp->snd_una))
3943 		del_data = BYTES_THIS_ACK(tp, th);
3944 	del_data += tp->sackhint.delivered_data;
3945 	pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3946 	tp->sackhint.prr_delivered += del_data;
3947 	/*
3948 	 * Proportional Rate Reduction
3949 	 */
3950 	if (pipe > tp->snd_ssthresh) {
3951 		if (tp->sackhint.recover_fs == 0)
3952 			tp->sackhint.recover_fs =
3953 			    max(1, tp->snd_nxt - tp->snd_una);
3954 		snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh /
3955 		    tp->sackhint.recover_fs) - tp->sackhint.sack_bytes_rexmit;
3956 	} else {
3957 		if (V_tcp_do_prr_conservative)
3958 			limit = tp->sackhint.prr_delivered -
3959 			    tp->sackhint.sack_bytes_rexmit;
3960 		else
3961 			if ((tp->sackhint.prr_delivered -
3962 			    tp->sackhint.sack_bytes_rexmit) > del_data)
3963 				limit = tp->sackhint.prr_delivered -
3964 				    tp->sackhint.sack_bytes_rexmit + maxseg;
3965 			else
3966 				limit = del_data + maxseg;
3967 		snd_cnt = min((tp->snd_ssthresh - pipe), limit);
3968 	}
3969 	snd_cnt = max((snd_cnt / maxseg), 0);
3970 	/*
3971 	 * Send snd_cnt new data into the network in response to this ack.
3972 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
3973 	 * accordingly.
3974 	 */
3975 	tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3976 		tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg);
3977 	tp->t_flags |= TF_ACKNOW;
3978 	(void) tcp_output(tp);
3979 }
3980 
3981 /*
3982  * On a partial ack arrives, force the retransmission of the
3983  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3984  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3985  * be started again.
3986  */
3987 void
3988 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3989 {
3990 	tcp_seq onxt = tp->snd_nxt;
3991 	uint32_t ocwnd = tp->snd_cwnd;
3992 	u_int maxseg = tcp_maxseg(tp);
3993 
3994 	INP_WLOCK_ASSERT(tp->t_inpcb);
3995 
3996 	tcp_timer_activate(tp, TT_REXMT, 0);
3997 	tp->t_rtttime = 0;
3998 	tp->snd_nxt = th->th_ack;
3999 	/*
4000 	 * Set snd_cwnd to one segment beyond acknowledged offset.
4001 	 * (tp->snd_una has not yet been updated when this function is called.)
4002 	 */
4003 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4004 	tp->t_flags |= TF_ACKNOW;
4005 	(void) tp->t_fb->tfb_tcp_output(tp);
4006 	tp->snd_cwnd = ocwnd;
4007 	if (SEQ_GT(onxt, tp->snd_nxt))
4008 		tp->snd_nxt = onxt;
4009 	/*
4010 	 * Partial window deflation.  Relies on fact that tp->snd_una
4011 	 * not updated yet.
4012 	 */
4013 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4014 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4015 	else
4016 		tp->snd_cwnd = 0;
4017 	tp->snd_cwnd += maxseg;
4018 }
4019 
4020 int
4021 tcp_compute_pipe(struct tcpcb *tp)
4022 {
4023 	return (tp->snd_max - tp->snd_una +
4024 		tp->sackhint.sack_bytes_rexmit -
4025 		tp->sackhint.sacked_bytes);
4026 }
4027 
4028 uint32_t
4029 tcp_compute_initwnd(uint32_t maxseg)
4030 {
4031 	/*
4032 	 * Calculate the Initial Window, also used as Restart Window
4033 	 *
4034 	 * RFC5681 Section 3.1 specifies the default conservative values.
4035 	 * RFC3390 specifies slightly more aggressive values.
4036 	 * RFC6928 increases it to ten segments.
4037 	 * Support for user specified value for initial flight size.
4038 	 */
4039 	if (V_tcp_initcwnd_segments)
4040 		return min(V_tcp_initcwnd_segments * maxseg,
4041 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4042 	else if (V_tcp_do_rfc3390)
4043 		return min(4 * maxseg, max(2 * maxseg, 4380));
4044 	else {
4045 		/* Per RFC5681 Section 3.1 */
4046 		if (maxseg > 2190)
4047 			return (2 * maxseg);
4048 		else if (maxseg > 1095)
4049 			return (3 * maxseg);
4050 		else
4051 			return (4 * maxseg);
4052 	}
4053 }
4054