xref: /freebsd/sys/netinet/tcp_input.c (revision 573bd1fcf532eae35ac30d6aa2c6ff4985a60fe8)
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 #include <netinet/tcp_debug.h>
121 #ifdef TCP_OFFLOAD
122 #include <netinet/tcp_offload.h>
123 #endif
124 #include <netinet/tcp_ecn.h>
125 #include <netinet/udp.h>
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(bool, blackhole_local) = false;
147 #define	V_blackhole_local	VNET(blackhole_local)
148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
149     CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
150     "Enforce net.inet.tcp.blackhole for locally originated packets");
151 
152 VNET_DEFINE(int, tcp_delack_enabled) = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
154     &VNET_NAME(tcp_delack_enabled), 0,
155     "Delay ACK to try and piggyback it onto a data packet");
156 
157 VNET_DEFINE(int, drop_synfin) = 0;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
159     &VNET_NAME(drop_synfin), 0,
160     "Drop TCP packets with SYN+FIN set");
161 
162 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
164     &VNET_NAME(tcp_do_prr_conservative), 0,
165     "Do conservative Proportional Rate Reduction");
166 
167 VNET_DEFINE(int, tcp_do_prr) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
169     &VNET_NAME(tcp_do_prr), 1,
170     "Enable Proportional Rate Reduction per RFC 6937");
171 
172 VNET_DEFINE(int, tcp_do_lrd) = 0;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
174     &VNET_NAME(tcp_do_lrd), 1,
175     "Perform Lost Retransmission Detection");
176 
177 VNET_DEFINE(int, tcp_do_newcwv) = 0;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
179     &VNET_NAME(tcp_do_newcwv), 0,
180     "Enable New Congestion Window Validation per RFC7661");
181 
182 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
184     &VNET_NAME(tcp_do_rfc3042), 0,
185     "Enable RFC 3042 (Limited Transmit)");
186 
187 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
189     &VNET_NAME(tcp_do_rfc3390), 0,
190     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
191 
192 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
194     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
195     "Slow-start flight size (initial congestion window) in number of segments");
196 
197 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
199     &VNET_NAME(tcp_do_rfc3465), 0,
200     "Enable RFC 3465 (Appropriate Byte Counting)");
201 
202 VNET_DEFINE(int, tcp_abc_l_var) = 2;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
204     &VNET_NAME(tcp_abc_l_var), 2,
205     "Cap the max cwnd increment during slow-start to this number of segments");
206 
207 VNET_DEFINE(int, tcp_insecure_syn) = 0;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
209     &VNET_NAME(tcp_insecure_syn), 0,
210     "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
211 
212 VNET_DEFINE(int, tcp_insecure_rst) = 0;
213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
214     &VNET_NAME(tcp_insecure_rst), 0,
215     "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
216 
217 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
218 #define	V_tcp_recvspace	VNET(tcp_recvspace)
219 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
220     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
221 
222 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
224     &VNET_NAME(tcp_do_autorcvbuf), 0,
225     "Enable automatic receive buffer sizing");
226 
227 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
229     &VNET_NAME(tcp_autorcvbuf_max), 0,
230     "Max size of automatic receive buffer");
231 
232 VNET_DEFINE(struct inpcbinfo, tcbinfo);
233 
234 /*
235  * TCP statistics are stored in an array of counter(9)s, which size matches
236  * size of struct tcpstat.  TCP running connection count is a regular array.
237  */
238 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
239 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
240     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
241 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
242 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
243     CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
244     "TCP connection counts by TCP state");
245 
246 /*
247  * Kernel module interface for updating tcpstat.  The first argument is an index
248  * into tcpstat treated as an array.
249  */
250 void
251 kmod_tcpstat_add(int statnum, int val)
252 {
253 
254 	counter_u64_add(VNET(tcpstat)[statnum], val);
255 }
256 
257 /*
258  * Make sure that we only start a SACK loss recovery when
259  * receiving a duplicate ACK with a SACK block, and also
260  * complete SACK loss recovery in case the other end
261  * reneges.
262  */
263 static bool inline
264 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
265 {
266 	return ((tp->t_flags & TF_SACK_PERMIT) &&
267 		((to->to_flags & TOF_SACK) ||
268 		(!TAILQ_EMPTY(&tp->snd_holes))));
269 }
270 
271 #ifdef TCP_HHOOK
272 /*
273  * Wrapper for the TCP established input helper hook.
274  */
275 void
276 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
277 {
278 	struct tcp_hhook_data hhook_data;
279 
280 	if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
281 		hhook_data.tp = tp;
282 		hhook_data.th = th;
283 		hhook_data.to = to;
284 
285 		hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
286 		    tp->osd);
287 	}
288 }
289 #endif
290 
291 /*
292  * CC wrapper hook functions
293  */
294 void
295 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
296     uint16_t type)
297 {
298 #ifdef STATS
299 	int32_t gput;
300 #endif
301 
302 	INP_WLOCK_ASSERT(tptoinpcb(tp));
303 
304 	tp->ccv->nsegs = nsegs;
305 	tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
306 	if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
307 	    (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
308 	     (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
309 		tp->ccv->flags |= CCF_CWND_LIMITED;
310 	else
311 		tp->ccv->flags &= ~CCF_CWND_LIMITED;
312 
313 	if (type == CC_ACK) {
314 #ifdef STATS
315 		stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
316 		    ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
317 		if (!IN_RECOVERY(tp->t_flags))
318 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
319 			   tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
320 		if ((tp->t_flags & TF_GPUTINPROG) &&
321 		    SEQ_GEQ(th->th_ack, tp->gput_ack)) {
322 			/*
323 			 * Compute goodput in bits per millisecond.
324 			 */
325 			gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
326 			    max(1, tcp_ts_getticks() - tp->gput_ts);
327 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
328 			    gput);
329 			/*
330 			 * XXXLAS: This is a temporary hack, and should be
331 			 * chained off VOI_TCP_GPUT when stats(9) grows an API
332 			 * to deal with chained VOIs.
333 			 */
334 			if (tp->t_stats_gput_prev > 0)
335 				stats_voi_update_abs_s32(tp->t_stats,
336 				    VOI_TCP_GPUT_ND,
337 				    ((gput - tp->t_stats_gput_prev) * 100) /
338 				    tp->t_stats_gput_prev);
339 			tp->t_flags &= ~TF_GPUTINPROG;
340 			tp->t_stats_gput_prev = gput;
341 		}
342 #endif /* STATS */
343 		if (tp->snd_cwnd > tp->snd_ssthresh) {
344 			tp->t_bytes_acked += tp->ccv->bytes_this_ack;
345 			if (tp->t_bytes_acked >= tp->snd_cwnd) {
346 				tp->t_bytes_acked -= tp->snd_cwnd;
347 				tp->ccv->flags |= CCF_ABC_SENTAWND;
348 			}
349 		} else {
350 				tp->ccv->flags &= ~CCF_ABC_SENTAWND;
351 				tp->t_bytes_acked = 0;
352 		}
353 	}
354 
355 	if (CC_ALGO(tp)->ack_received != NULL) {
356 		/* XXXLAS: Find a way to live without this */
357 		tp->ccv->curack = th->th_ack;
358 		CC_ALGO(tp)->ack_received(tp->ccv, type);
359 	}
360 #ifdef STATS
361 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
362 #endif
363 }
364 
365 void
366 cc_conn_init(struct tcpcb *tp)
367 {
368 	struct hc_metrics_lite metrics;
369 	struct inpcb *inp = tptoinpcb(tp);
370 	u_int maxseg;
371 	int rtt;
372 
373 	INP_WLOCK_ASSERT(inp);
374 
375 	tcp_hc_get(&inp->inp_inc, &metrics);
376 	maxseg = tcp_maxseg(tp);
377 
378 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
379 		tp->t_srtt = rtt;
380 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
381 		TCPSTAT_INC(tcps_usedrtt);
382 		if (metrics.rmx_rttvar) {
383 			tp->t_rttvar = metrics.rmx_rttvar;
384 			TCPSTAT_INC(tcps_usedrttvar);
385 		} else {
386 			/* default variation is +- 1 rtt */
387 			tp->t_rttvar =
388 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
389 		}
390 		TCPT_RANGESET(tp->t_rxtcur,
391 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
392 		    tp->t_rttmin, TCPTV_REXMTMAX);
393 	}
394 	if (metrics.rmx_ssthresh) {
395 		/*
396 		 * There's some sort of gateway or interface
397 		 * buffer limit on the path.  Use this to set
398 		 * the slow start threshold, but set the
399 		 * threshold to no less than 2*mss.
400 		 */
401 		tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
402 		TCPSTAT_INC(tcps_usedssthresh);
403 	}
404 
405 	/*
406 	 * Set the initial slow-start flight size.
407 	 *
408 	 * If a SYN or SYN/ACK was lost and retransmitted, we have to
409 	 * reduce the initial CWND to one segment as congestion is likely
410 	 * requiring us to be cautious.
411 	 */
412 	if (tp->snd_cwnd == 1)
413 		tp->snd_cwnd = maxseg;		/* SYN(-ACK) lost */
414 	else
415 		tp->snd_cwnd = tcp_compute_initwnd(maxseg);
416 
417 	if (CC_ALGO(tp)->conn_init != NULL)
418 		CC_ALGO(tp)->conn_init(tp->ccv);
419 }
420 
421 void inline
422 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
423 {
424 	INP_WLOCK_ASSERT(tptoinpcb(tp));
425 
426 #ifdef STATS
427 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
428 #endif
429 
430 	switch(type) {
431 	case CC_NDUPACK:
432 		if (!IN_FASTRECOVERY(tp->t_flags)) {
433 			tp->snd_recover = tp->snd_max;
434 			if (tp->t_flags2 & TF2_ECN_PERMIT)
435 				tp->t_flags2 |= TF2_ECN_SND_CWR;
436 		}
437 		break;
438 	case CC_ECN:
439 		if (!IN_CONGRECOVERY(tp->t_flags) ||
440 		    /*
441 		     * Allow ECN reaction on ACK to CWR, if
442 		     * that data segment was also CE marked.
443 		     */
444 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
445 			EXIT_CONGRECOVERY(tp->t_flags);
446 			TCPSTAT_INC(tcps_ecn_rcwnd);
447 			tp->snd_recover = tp->snd_max + 1;
448 			if (tp->t_flags2 & TF2_ECN_PERMIT)
449 				tp->t_flags2 |= TF2_ECN_SND_CWR;
450 		}
451 		break;
452 	case CC_RTO:
453 		tp->t_dupacks = 0;
454 		tp->t_bytes_acked = 0;
455 		EXIT_RECOVERY(tp->t_flags);
456 		if (tp->t_flags2 & TF2_ECN_PERMIT)
457 			tp->t_flags2 |= TF2_ECN_SND_CWR;
458 		break;
459 	case CC_RTO_ERR:
460 		TCPSTAT_INC(tcps_sndrexmitbad);
461 		/* RTO was unnecessary, so reset everything. */
462 		tp->snd_cwnd = tp->snd_cwnd_prev;
463 		tp->snd_ssthresh = tp->snd_ssthresh_prev;
464 		tp->snd_recover = tp->snd_recover_prev;
465 		if (tp->t_flags & TF_WASFRECOVERY)
466 			ENTER_FASTRECOVERY(tp->t_flags);
467 		if (tp->t_flags & TF_WASCRECOVERY)
468 			ENTER_CONGRECOVERY(tp->t_flags);
469 		tp->snd_nxt = tp->snd_max;
470 		tp->t_flags &= ~TF_PREVVALID;
471 		tp->t_badrxtwin = 0;
472 		break;
473 	}
474 
475 	if (CC_ALGO(tp)->cong_signal != NULL) {
476 		if (th != NULL)
477 			tp->ccv->curack = th->th_ack;
478 		CC_ALGO(tp)->cong_signal(tp->ccv, type);
479 	}
480 }
481 
482 void inline
483 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
484 {
485 	INP_WLOCK_ASSERT(tptoinpcb(tp));
486 
487 	/* XXXLAS: KASSERT that we're in recovery? */
488 
489 	if (CC_ALGO(tp)->post_recovery != NULL) {
490 		tp->ccv->curack = th->th_ack;
491 		CC_ALGO(tp)->post_recovery(tp->ccv);
492 	}
493 	/* XXXLAS: EXIT_RECOVERY ? */
494 	tp->t_bytes_acked = 0;
495 	tp->sackhint.delivered_data = 0;
496 	tp->sackhint.prr_out = 0;
497 }
498 
499 /*
500  * Indicate whether this ack should be delayed.  We can delay the ack if
501  * following conditions are met:
502  *	- There is no delayed ack timer in progress.
503  *	- Our last ack wasn't a 0-sized window. We never want to delay
504  *	  the ack that opens up a 0-sized window.
505  *	- LRO wasn't used for this segment. We make sure by checking that the
506  *	  segment size is not larger than the MSS.
507  */
508 #define DELAY_ACK(tp, tlen)						\
509 	((!tcp_timer_active(tp, TT_DELACK) &&				\
510 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
511 	    (tlen <= tp->t_maxseg) &&					\
512 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
513 
514 void inline
515 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
516 {
517 	INP_WLOCK_ASSERT(tptoinpcb(tp));
518 
519 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
520 		switch (iptos & IPTOS_ECN_MASK) {
521 		case IPTOS_ECN_CE:
522 			tp->ccv->flags |= CCF_IPHDR_CE;
523 			break;
524 		case IPTOS_ECN_ECT0:
525 			/* FALLTHROUGH */
526 		case IPTOS_ECN_ECT1:
527 			/* FALLTHROUGH */
528 		case IPTOS_ECN_NOTECT:
529 			tp->ccv->flags &= ~CCF_IPHDR_CE;
530 			break;
531 		}
532 
533 		if (flags & TH_CWR)
534 			tp->ccv->flags |= CCF_TCPHDR_CWR;
535 		else
536 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
537 
538 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
539 
540 		if (tp->ccv->flags & CCF_ACKNOW) {
541 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
542 			tp->t_flags |= TF_ACKNOW;
543 		}
544 	}
545 }
546 
547 void inline
548 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
549 {
550 	cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
551 }
552 
553 /*
554  * TCP input handling is split into multiple parts:
555  *   tcp6_input is a thin wrapper around tcp_input for the extended
556  *	ip6_protox[] call format in ip6_input
557  *   tcp_input handles primary segment validation, inpcb lookup and
558  *	SYN processing on listen sockets
559  *   tcp_do_segment processes the ACK and text of the segment for
560  *	establishing, established and closing connections
561  */
562 #ifdef INET6
563 int
564 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
565 {
566 	struct mbuf *m;
567 	struct in6_ifaddr *ia6;
568 	struct ip6_hdr *ip6;
569 
570 	m = *mp;
571 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
572 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
573 		if (m == NULL) {
574 			*mp = m;
575 			TCPSTAT_INC(tcps_rcvshort);
576 			return (IPPROTO_DONE);
577 		}
578 	}
579 
580 	/*
581 	 * draft-itojun-ipv6-tcp-to-anycast
582 	 * better place to put this in?
583 	 */
584 	ip6 = mtod(m, struct ip6_hdr *);
585 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
586 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
587 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
588 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
589 		*mp = NULL;
590 		return (IPPROTO_DONE);
591 	}
592 
593 	*mp = m;
594 	return (tcp_input_with_port(mp, offp, proto, port));
595 }
596 
597 int
598 tcp6_input(struct mbuf **mp, int *offp, int proto)
599 {
600 
601 	return(tcp6_input_with_port(mp, offp, proto, 0));
602 }
603 #endif /* INET6 */
604 
605 int
606 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
607 {
608 	struct mbuf *m = *mp;
609 	struct tcphdr *th = NULL;
610 	struct ip *ip = NULL;
611 	struct inpcb *inp = NULL;
612 	struct tcpcb *tp = NULL;
613 	struct socket *so = NULL;
614 	u_char *optp = NULL;
615 	int off0;
616 	int optlen = 0;
617 #ifdef INET
618 	int len;
619 	uint8_t ipttl;
620 #endif
621 	int tlen = 0, off;
622 	int drop_hdrlen;
623 	int thflags;
624 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
625 	int lookupflag;
626 	uint8_t iptos;
627 	struct m_tag *fwd_tag = NULL;
628 #ifdef INET6
629 	struct ip6_hdr *ip6 = NULL;
630 	int isipv6;
631 #else
632 	const void *ip6 = NULL;
633 #endif /* INET6 */
634 	struct tcpopt to;		/* options in this segment */
635 	char *s = NULL;			/* address and port logging */
636 #ifdef TCPDEBUG
637 	/*
638 	 * The size of tcp_saveipgen must be the size of the max ip header,
639 	 * now IPv6.
640 	 */
641 	u_char tcp_saveipgen[IP6_HDR_LEN];
642 	struct tcphdr tcp_savetcp;
643 	short ostate = 0;
644 #endif
645 
646 	NET_EPOCH_ASSERT();
647 
648 #ifdef INET6
649 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
650 #endif
651 
652 	off0 = *offp;
653 	m = *mp;
654 	*mp = NULL;
655 	to.to_flags = 0;
656 	TCPSTAT_INC(tcps_rcvtotal);
657 
658 #ifdef INET6
659 	if (isipv6) {
660 		ip6 = mtod(m, struct ip6_hdr *);
661 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
662 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
663 		if (port)
664 			goto skip6_csum;
665 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
666 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
667 				th->th_sum = m->m_pkthdr.csum_data;
668 			else
669 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
670 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
671 			th->th_sum ^= 0xffff;
672 		} else
673 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
674 		if (th->th_sum) {
675 			TCPSTAT_INC(tcps_rcvbadsum);
676 			goto drop;
677 		}
678 	skip6_csum:
679 		/*
680 		 * Be proactive about unspecified IPv6 address in source.
681 		 * As we use all-zero to indicate unbounded/unconnected pcb,
682 		 * unspecified IPv6 address can be used to confuse us.
683 		 *
684 		 * Note that packets with unspecified IPv6 destination is
685 		 * already dropped in ip6_input.
686 		 */
687 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
688 			/* XXX stat */
689 			goto drop;
690 		}
691 		iptos = IPV6_TRAFFIC_CLASS(ip6);
692 	}
693 #endif
694 #if defined(INET) && defined(INET6)
695 	else
696 #endif
697 #ifdef INET
698 	{
699 		/*
700 		 * Get IP and TCP header together in first mbuf.
701 		 * Note: IP leaves IP header in first mbuf.
702 		 */
703 		if (off0 > sizeof (struct ip)) {
704 			ip_stripoptions(m);
705 			off0 = sizeof(struct ip);
706 		}
707 		if (m->m_len < sizeof (struct tcpiphdr)) {
708 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
709 			    == NULL) {
710 				TCPSTAT_INC(tcps_rcvshort);
711 				return (IPPROTO_DONE);
712 			}
713 		}
714 		ip = mtod(m, struct ip *);
715 		th = (struct tcphdr *)((caddr_t)ip + off0);
716 		tlen = ntohs(ip->ip_len) - off0;
717 
718 		iptos = ip->ip_tos;
719 		if (port)
720 			goto skip_csum;
721 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
722 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
723 				th->th_sum = m->m_pkthdr.csum_data;
724 			else
725 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
726 				    ip->ip_dst.s_addr,
727 				    htonl(m->m_pkthdr.csum_data + tlen +
728 				    IPPROTO_TCP));
729 			th->th_sum ^= 0xffff;
730 		} else {
731 			struct ipovly *ipov = (struct ipovly *)ip;
732 
733 			/*
734 			 * Checksum extended TCP header and data.
735 			 */
736 			len = off0 + tlen;
737 			ipttl = ip->ip_ttl;
738 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
739 			ipov->ih_len = htons(tlen);
740 			th->th_sum = in_cksum(m, len);
741 			/* Reset length for SDT probes. */
742 			ip->ip_len = htons(len);
743 			/* Reset TOS bits */
744 			ip->ip_tos = iptos;
745 			/* Re-initialization for later version check */
746 			ip->ip_ttl = ipttl;
747 			ip->ip_v = IPVERSION;
748 			ip->ip_hl = off0 >> 2;
749 		}
750 	skip_csum:
751 		if (th->th_sum && (port == 0)) {
752 			TCPSTAT_INC(tcps_rcvbadsum);
753 			goto drop;
754 		}
755 	}
756 #endif /* INET */
757 
758 	/*
759 	 * Check that TCP offset makes sense,
760 	 * pull out TCP options and adjust length.		XXX
761 	 */
762 	off = th->th_off << 2;
763 	if (off < sizeof (struct tcphdr) || off > tlen) {
764 		TCPSTAT_INC(tcps_rcvbadoff);
765 		goto drop;
766 	}
767 	tlen -= off;	/* tlen is used instead of ti->ti_len */
768 	if (off > sizeof (struct tcphdr)) {
769 #ifdef INET6
770 		if (isipv6) {
771 			if (m->m_len < off0 + off) {
772 				m = m_pullup(m, off0 + off);
773 				if (m == NULL) {
774 					TCPSTAT_INC(tcps_rcvshort);
775 					return (IPPROTO_DONE);
776 				}
777 			}
778 			ip6 = mtod(m, struct ip6_hdr *);
779 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
780 		}
781 #endif
782 #if defined(INET) && defined(INET6)
783 		else
784 #endif
785 #ifdef INET
786 		{
787 			if (m->m_len < sizeof(struct ip) + off) {
788 				if ((m = m_pullup(m, sizeof (struct ip) + off))
789 				    == NULL) {
790 					TCPSTAT_INC(tcps_rcvshort);
791 					return (IPPROTO_DONE);
792 				}
793 				ip = mtod(m, struct ip *);
794 				th = (struct tcphdr *)((caddr_t)ip + off0);
795 			}
796 		}
797 #endif
798 		optlen = off - sizeof (struct tcphdr);
799 		optp = (u_char *)(th + 1);
800 	}
801 	thflags = tcp_get_flags(th);
802 
803 	/*
804 	 * Convert TCP protocol specific fields to host format.
805 	 */
806 	tcp_fields_to_host(th);
807 
808 	/*
809 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
810 	 */
811 	drop_hdrlen = off0 + off;
812 
813 	/*
814 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
815 	 */
816         if (
817 #ifdef INET6
818 	    (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
819 #ifdef INET
820 	    || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
821 #endif
822 #endif
823 #if defined(INET) && !defined(INET6)
824 	    (m->m_flags & M_IP_NEXTHOP)
825 #endif
826 	    )
827 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
828 
829 	/*
830 	 * For initial SYN packets we don't need write lock on matching
831 	 * PCB, be it a listening one or a synchronized one.  The packet
832 	 * shall not modify its state.
833 	 */
834 	lookupflag = INPLOOKUP_WILDCARD |
835 	    ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
836 	    INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB);
837 findpcb:
838 #ifdef INET6
839 	if (isipv6 && fwd_tag != NULL) {
840 		struct sockaddr_in6 *next_hop6;
841 
842 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
843 		/*
844 		 * Transparently forwarded. Pretend to be the destination.
845 		 * Already got one like this?
846 		 */
847 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
848 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
849 		    lookupflag, m->m_pkthdr.rcvif, m);
850 		if (!inp) {
851 			/*
852 			 * It's new.  Try to find the ambushing socket.
853 			 * Because we've rewritten the destination address,
854 			 * any hardware-generated hash is ignored.
855 			 */
856 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
857 			    th->th_sport, &next_hop6->sin6_addr,
858 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
859 			    th->th_dport, lookupflag, m->m_pkthdr.rcvif);
860 		}
861 	} else if (isipv6) {
862 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
863 		    th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag,
864 		    m->m_pkthdr.rcvif, m);
865 	}
866 #endif /* INET6 */
867 #if defined(INET6) && defined(INET)
868 	else
869 #endif
870 #ifdef INET
871 	if (fwd_tag != NULL) {
872 		struct sockaddr_in *next_hop;
873 
874 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
875 		/*
876 		 * Transparently forwarded. Pretend to be the destination.
877 		 * already got one like this?
878 		 */
879 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
880 		    ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
881 		if (!inp) {
882 			/*
883 			 * It's new.  Try to find the ambushing socket.
884 			 * Because we've rewritten the destination address,
885 			 * any hardware-generated hash is ignored.
886 			 */
887 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
888 			    th->th_sport, next_hop->sin_addr,
889 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
890 			    th->th_dport, lookupflag, m->m_pkthdr.rcvif);
891 		}
892 	} else
893 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
894 		    th->th_sport, ip->ip_dst, th->th_dport, lookupflag,
895 		    m->m_pkthdr.rcvif, m);
896 #endif /* INET */
897 
898 	/*
899 	 * If the INPCB does not exist then all data in the incoming
900 	 * segment is discarded and an appropriate RST is sent back.
901 	 * XXX MRT Send RST using which routing table?
902 	 */
903 	if (inp == NULL) {
904 		if (rstreason != 0) {
905 			/* We came here after second (safety) lookup. */
906 			MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0);
907 			goto dropwithreset;
908 		}
909 		/*
910 		 * Log communication attempts to ports that are not
911 		 * in use.
912 		 */
913 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
914 		    V_tcp_log_in_vain == 2) {
915 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
916 				log(LOG_INFO, "%s; %s: Connection attempt "
917 				    "to closed port\n", s, __func__);
918 		}
919 		/*
920 		 * When blackholing do not respond with a RST but
921 		 * completely ignore the segment and drop it.
922 		 */
923 		if (((V_blackhole == 1 && (thflags & TH_SYN)) ||
924 		    V_blackhole == 2) && (V_blackhole_local || (
925 #ifdef INET6
926 		    isipv6 ? !in6_localaddr(&ip6->ip6_src) :
927 #endif
928 #ifdef INET
929 		    !in_localip(ip->ip_src)
930 #else
931 		    true
932 #endif
933 		    )))
934 			goto dropunlock;
935 
936 		rstreason = BANDLIM_RST_CLOSEDPORT;
937 		goto dropwithreset;
938 	}
939 	INP_LOCK_ASSERT(inp);
940 
941 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
942 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
943 	    !SOLISTENING(inp->inp_socket)) {
944 		inp->inp_flowid = m->m_pkthdr.flowid;
945 		inp->inp_flowtype = M_HASHTYPE_GET(m);
946 	}
947 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
948 #ifdef INET6
949 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
950 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
951 		goto dropunlock;
952 	}
953 #ifdef INET
954 	else
955 #endif
956 #endif /* INET6 */
957 #ifdef INET
958 	if (IPSEC_ENABLED(ipv4) &&
959 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
960 		goto dropunlock;
961 	}
962 #endif /* INET */
963 #endif /* IPSEC */
964 
965 	/*
966 	 * Check the minimum TTL for socket.
967 	 */
968 	if (inp->inp_ip_minttl != 0) {
969 #ifdef INET6
970 		if (isipv6) {
971 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
972 				goto dropunlock;
973 		} else
974 #endif
975 		if (inp->inp_ip_minttl > ip->ip_ttl)
976 			goto dropunlock;
977 	}
978 
979 	tp = intotcpcb(inp);
980 	switch (tp->t_state) {
981 	case TCPS_TIME_WAIT:
982 		/*
983 		 * A previous connection in TIMEWAIT state is supposed to catch
984 		 * stray or duplicate segments arriving late.  If this segment
985 		 * was a legitimate new connection attempt, the old INPCB gets
986 		 * removed and we can try again to find a listening socket.
987 		 */
988 		tcp_dooptions(&to, optp, optlen,
989 		    (thflags & TH_SYN) ? TO_SYN : 0);
990 		/*
991 		 * tcp_twcheck unlocks the inp always, and frees the m if fails.
992 		 */
993 		if (tcp_twcheck(inp, &to, th, m, tlen))
994 			goto findpcb;
995 		return (IPPROTO_DONE);
996 	case TCPS_CLOSED:
997 		/*
998 		 * The TCPCB may no longer exist if the connection is winding
999 		 * down or it is in the CLOSED state.  Either way we drop the
1000 		 * segment and send an appropriate response.
1001 		 */
1002 		rstreason = BANDLIM_RST_CLOSEDPORT;
1003 		goto dropwithreset;
1004 	}
1005 
1006 	if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1007 		rstreason = BANDLIM_RST_CLOSEDPORT;
1008 		goto dropwithreset;
1009 	}
1010 
1011 #ifdef TCP_OFFLOAD
1012 	if (tp->t_flags & TF_TOE) {
1013 		tcp_offload_input(tp, m);
1014 		m = NULL;	/* consumed by the TOE driver */
1015 		goto dropunlock;
1016 	}
1017 #endif
1018 
1019 #ifdef MAC
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 || !SOLISTENING(so),
1043 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1044 	if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
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 inp.
1079 			 */
1080 			rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1081 			if (rstreason < 0) {
1082 				/*
1083 				 * A failing TCP MD5 signature comparison
1084 				 * must result in the segment being dropped
1085 				 * and must not produce any response back
1086 				 * to the sender.
1087 				 */
1088 				goto dropunlock;
1089 			} else if (rstreason == 0) {
1090 				/*
1091 				 * No syncache entry, or ACK was not for our
1092 				 * SYN/ACK.  Do our protection against double
1093 				 * ACK.  If peer sent us 2 ACKs, then for the
1094 				 * first one syncache_expand() successfully
1095 				 * converted syncache entry into a socket,
1096 				 * while we were waiting on the inpcb lock.  We
1097 				 * don't want to sent RST for the second ACK,
1098 				 * so we perform second lookup without wildcard
1099 				 * match, hoping to find the new socket.  If
1100 				 * the ACK is stray indeed, rstreason would
1101 				 * hint the above code that the lookup was a
1102 				 * second attempt.
1103 				 *
1104 				 * NB: syncache did its own logging
1105 				 * of the failure cause.
1106 				 */
1107 				INP_WUNLOCK(inp);
1108 				rstreason = BANDLIM_RST_OPENPORT;
1109 				lookupflag &= ~INPLOOKUP_WILDCARD;
1110 				goto findpcb;
1111 			}
1112 tfo_socket_result:
1113 			if (so == NULL) {
1114 				/*
1115 				 * We completed the 3-way handshake
1116 				 * but could not allocate a socket
1117 				 * either due to memory shortage,
1118 				 * listen queue length limits or
1119 				 * global socket limits.  Send RST
1120 				 * or wait and have the remote end
1121 				 * retransmit the ACK for another
1122 				 * try.
1123 				 */
1124 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1125 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1126 					    "Socket allocation failed due to "
1127 					    "limits or memory shortage, %s\n",
1128 					    s, __func__,
1129 					    V_tcp_sc_rst_sock_fail ?
1130 					    "sending RST" : "try again");
1131 				if (V_tcp_sc_rst_sock_fail) {
1132 					rstreason = BANDLIM_UNLIMITED;
1133 					goto dropwithreset;
1134 				} else
1135 					goto dropunlock;
1136 			}
1137 			/*
1138 			 * Socket is created in state SYN_RECEIVED.
1139 			 * Unlock the listen socket, lock the newly
1140 			 * created socket and update the tp variable.
1141 			 * If we came here via jump to tfo_socket_result,
1142 			 * then listening socket is read-locked.
1143 			 */
1144 			INP_UNLOCK(inp);	/* listen socket */
1145 			inp = sotoinpcb(so);
1146 			/*
1147 			 * New connection inpcb is already locked by
1148 			 * syncache_expand().
1149 			 */
1150 			INP_WLOCK_ASSERT(inp);
1151 			tp = intotcpcb(inp);
1152 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1153 			    ("%s: ", __func__));
1154 			/*
1155 			 * Process the segment and the data it
1156 			 * contains.  tcp_do_segment() consumes
1157 			 * the mbuf chain and unlocks the inpcb.
1158 			 */
1159 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1160 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1161 			    iptos);
1162 			return (IPPROTO_DONE);
1163 		}
1164 		/*
1165 		 * Segment flag validation for new connection attempts:
1166 		 *
1167 		 * Our (SYN|ACK) response was rejected.
1168 		 * Check with syncache and remove entry to prevent
1169 		 * retransmits.
1170 		 *
1171 		 * NB: syncache_chkrst does its own logging of failure
1172 		 * causes.
1173 		 */
1174 		if (thflags & TH_RST) {
1175 			syncache_chkrst(&inc, th, m, port);
1176 			goto dropunlock;
1177 		}
1178 		/*
1179 		 * We can't do anything without SYN.
1180 		 */
1181 		if ((thflags & TH_SYN) == 0) {
1182 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1183 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1184 				    "SYN is missing, segment ignored\n",
1185 				    s, __func__);
1186 			TCPSTAT_INC(tcps_badsyn);
1187 			goto dropunlock;
1188 		}
1189 		/*
1190 		 * (SYN|ACK) is bogus on a listen socket.
1191 		 */
1192 		if (thflags & TH_ACK) {
1193 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1194 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1195 				    "SYN|ACK invalid, segment rejected\n",
1196 				    s, __func__);
1197 			syncache_badack(&inc, port);	/* XXX: Not needed! */
1198 			TCPSTAT_INC(tcps_badsyn);
1199 			rstreason = BANDLIM_RST_OPENPORT;
1200 			goto dropwithreset;
1201 		}
1202 		/*
1203 		 * If the drop_synfin option is enabled, drop all
1204 		 * segments with both the SYN and FIN bits set.
1205 		 * This prevents e.g. nmap from identifying the
1206 		 * TCP/IP stack.
1207 		 * XXX: Poor reasoning.  nmap has other methods
1208 		 * and is constantly refining its stack detection
1209 		 * strategies.
1210 		 * XXX: This is a violation of the TCP specification
1211 		 * and was used by RFC1644.
1212 		 */
1213 		if ((thflags & TH_FIN) && V_drop_synfin) {
1214 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1215 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1216 				    "SYN|FIN segment ignored (based on "
1217 				    "sysctl setting)\n", s, __func__);
1218 			TCPSTAT_INC(tcps_badsyn);
1219 			goto dropunlock;
1220 		}
1221 		/*
1222 		 * Segment's flags are (SYN) or (SYN|FIN).
1223 		 *
1224 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1225 		 * as they do not affect the state of the TCP FSM.
1226 		 * The data pointed to by TH_URG and th_urp is ignored.
1227 		 */
1228 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1229 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1230 		KASSERT(thflags & (TH_SYN),
1231 		    ("%s: Listen socket: TH_SYN not set", __func__));
1232 		INP_RLOCK_ASSERT(inp);
1233 #ifdef INET6
1234 		/*
1235 		 * If deprecated address is forbidden,
1236 		 * we do not accept SYN to deprecated interface
1237 		 * address to prevent any new inbound connection from
1238 		 * getting established.
1239 		 * When we do not accept SYN, we send a TCP RST,
1240 		 * with deprecated source address (instead of dropping
1241 		 * it).  We compromise it as it is much better for peer
1242 		 * to send a RST, and RST will be the final packet
1243 		 * for the exchange.
1244 		 *
1245 		 * If we do not forbid deprecated addresses, we accept
1246 		 * the SYN packet.  RFC2462 does not suggest dropping
1247 		 * SYN in this case.
1248 		 * If we decipher RFC2462 5.5.4, it says like this:
1249 		 * 1. use of deprecated addr with existing
1250 		 *    communication is okay - "SHOULD continue to be
1251 		 *    used"
1252 		 * 2. use of it with new communication:
1253 		 *   (2a) "SHOULD NOT be used if alternate address
1254 		 *        with sufficient scope is available"
1255 		 *   (2b) nothing mentioned otherwise.
1256 		 * Here we fall into (2b) case as we have no choice in
1257 		 * our source address selection - we must obey the peer.
1258 		 *
1259 		 * The wording in RFC2462 is confusing, and there are
1260 		 * multiple description text for deprecated address
1261 		 * handling - worse, they are not exactly the same.
1262 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1263 		 */
1264 		if (isipv6 && !V_ip6_use_deprecated) {
1265 			struct in6_ifaddr *ia6;
1266 
1267 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1268 			if (ia6 != NULL &&
1269 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1270 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1271 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1272 					"Connection attempt to deprecated "
1273 					"IPv6 address rejected\n",
1274 					s, __func__);
1275 				rstreason = BANDLIM_RST_OPENPORT;
1276 				goto dropwithreset;
1277 			}
1278 		}
1279 #endif /* INET6 */
1280 		/*
1281 		 * Basic sanity checks on incoming SYN requests:
1282 		 *   Don't respond if the destination is a link layer
1283 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1284 		 *   If it is from this socket it must be forged.
1285 		 *   Don't respond if the source or destination is a
1286 		 *	global or subnet broad- or multicast address.
1287 		 *   Note that it is quite possible to receive unicast
1288 		 *	link-layer packets with a broadcast IP address. Use
1289 		 *	in_broadcast() to find them.
1290 		 */
1291 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1292 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1293 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1294 				"Connection attempt from broad- or multicast "
1295 				"link layer address ignored\n", s, __func__);
1296 			goto dropunlock;
1297 		}
1298 #ifdef INET6
1299 		if (isipv6) {
1300 			if (th->th_dport == th->th_sport &&
1301 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1302 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1303 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1304 					"Connection attempt to/from self "
1305 					"ignored\n", s, __func__);
1306 				goto dropunlock;
1307 			}
1308 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1309 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1310 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1311 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1312 					"Connection attempt from/to multicast "
1313 					"address ignored\n", s, __func__);
1314 				goto dropunlock;
1315 			}
1316 		}
1317 #endif
1318 #if defined(INET) && defined(INET6)
1319 		else
1320 #endif
1321 #ifdef INET
1322 		{
1323 			if (th->th_dport == th->th_sport &&
1324 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1325 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1326 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1327 					"Connection attempt from/to self "
1328 					"ignored\n", s, __func__);
1329 				goto dropunlock;
1330 			}
1331 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1332 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1333 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1334 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1335 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1336 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1337 					"Connection attempt from/to broad- "
1338 					"or multicast address ignored\n",
1339 					s, __func__);
1340 				goto dropunlock;
1341 			}
1342 		}
1343 #endif
1344 		/*
1345 		 * SYN appears to be valid.  Create compressed TCP state
1346 		 * for syncache.
1347 		 */
1348 #ifdef TCPDEBUG
1349 		if (so->so_options & SO_DEBUG)
1350 			tcp_trace(TA_INPUT, ostate, tp,
1351 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1352 #endif
1353 		TCP_PROBE3(debug__input, tp, th, m);
1354 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1355 		if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1356 		    iptos, port)) != NULL)
1357 			goto tfo_socket_result;
1358 
1359 		/*
1360 		 * Entry added to syncache and mbuf consumed.
1361 		 * Only the listen socket is unlocked by syncache_add().
1362 		 */
1363 		return (IPPROTO_DONE);
1364 	} else if (tp->t_state == TCPS_LISTEN) {
1365 		/*
1366 		 * When a listen socket is torn down the SO_ACCEPTCONN
1367 		 * flag is removed first while connections are drained
1368 		 * from the accept queue in a unlock/lock cycle of the
1369 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1370 		 * attempt go through unhandled.
1371 		 */
1372 		goto dropunlock;
1373 	}
1374 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1375 	if (tp->t_flags & TF_SIGNATURE) {
1376 		tcp_dooptions(&to, optp, optlen, thflags);
1377 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1378 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1379 			goto dropunlock;
1380 		}
1381 		if (!TCPMD5_ENABLED() ||
1382 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1383 			goto dropunlock;
1384 	}
1385 #endif
1386 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1387 
1388 	/*
1389 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1390 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1391 	 * the inpcb, and unlocks pcbinfo.
1392 	 *
1393 	 * XXXGL: in case of a pure SYN arriving on existing connection
1394 	 * TCP stacks won't need to modify the PCB, they would either drop
1395 	 * the segment silently, or send a challenge ACK.  However, we try
1396 	 * to upgrade the lock, because calling convention for stacks is
1397 	 * write-lock on PCB.  If upgrade fails, drop the SYN.
1398 	 */
1399 	if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0)
1400 		goto dropunlock;
1401 
1402 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1403 	return (IPPROTO_DONE);
1404 
1405 dropwithreset:
1406 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1407 
1408 	if (inp != NULL) {
1409 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1410 		INP_UNLOCK(inp);
1411 	} else
1412 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1413 	m = NULL;	/* mbuf chain got consumed. */
1414 	goto drop;
1415 
1416 dropunlock:
1417 	if (m != NULL)
1418 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1419 
1420 	if (inp != NULL)
1421 		INP_UNLOCK(inp);
1422 
1423 drop:
1424 	if (s != NULL)
1425 		free(s, M_TCPLOG);
1426 	if (m != NULL)
1427 		m_freem(m);
1428 	return (IPPROTO_DONE);
1429 }
1430 
1431 /*
1432  * Automatic sizing of receive socket buffer.  Often the send
1433  * buffer size is not optimally adjusted to the actual network
1434  * conditions at hand (delay bandwidth product).  Setting the
1435  * buffer size too small limits throughput on links with high
1436  * bandwidth and high delay (eg. trans-continental/oceanic links).
1437  *
1438  * On the receive side the socket buffer memory is only rarely
1439  * used to any significant extent.  This allows us to be much
1440  * more aggressive in scaling the receive socket buffer.  For
1441  * the case that the buffer space is actually used to a large
1442  * extent and we run out of kernel memory we can simply drop
1443  * the new segments; TCP on the sender will just retransmit it
1444  * later.  Setting the buffer size too big may only consume too
1445  * much kernel memory if the application doesn't read() from
1446  * the socket or packet loss or reordering makes use of the
1447  * reassembly queue.
1448  *
1449  * The criteria to step up the receive buffer one notch are:
1450  *  1. Application has not set receive buffer size with
1451  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1452  *  2. the number of bytes received during 1/2 of an sRTT
1453  *     is at least 3/8 of the current socket buffer size.
1454  *  3. receive buffer size has not hit maximal automatic size;
1455  *
1456  * If all of the criteria are met we increaset the socket buffer
1457  * by a 1/2 (bounded by the max). This allows us to keep ahead
1458  * of slow-start but also makes it so our peer never gets limited
1459  * by our rwnd which we then open up causing a burst.
1460  *
1461  * This algorithm does two steps per RTT at most and only if
1462  * we receive a bulk stream w/o packet losses or reorderings.
1463  * Shrinking the buffer during idle times is not necessary as
1464  * it doesn't consume any memory when idle.
1465  *
1466  * TODO: Only step up if the application is actually serving
1467  * the buffer to better manage the socket buffer resources.
1468  */
1469 int
1470 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1471     struct tcpcb *tp, int tlen)
1472 {
1473 	int newsize = 0;
1474 
1475 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1476 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1477 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1478 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1479 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1480 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1481 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1482 		}
1483 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1484 
1485 		/* Start over with next RTT. */
1486 		tp->rfbuf_ts = 0;
1487 		tp->rfbuf_cnt = 0;
1488 	} else {
1489 		tp->rfbuf_cnt += tlen;	/* add up */
1490 	}
1491 	return (newsize);
1492 }
1493 
1494 int
1495 tcp_input(struct mbuf **mp, int *offp, int proto)
1496 {
1497 	return(tcp_input_with_port(mp, offp, proto, 0));
1498 }
1499 
1500 static void
1501 tcp_handle_wakeup(struct tcpcb *tp)
1502 {
1503 
1504 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1505 
1506 	if (tp->t_flags & TF_WAKESOR) {
1507 		struct socket *so = tptosocket(tp);
1508 
1509 		tp->t_flags &= ~TF_WAKESOR;
1510 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1511 		sorwakeup_locked(so);
1512 	}
1513 }
1514 
1515 void
1516 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1517     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1518 {
1519 	uint16_t thflags;
1520 	int acked, ourfinisacked, needoutput = 0, sack_changed;
1521 	int rstreason, todrop, win, incforsyn = 0;
1522 	uint32_t tiwin;
1523 	uint16_t nsegs;
1524 	char *s;
1525 	struct inpcb *inp = tptoinpcb(tp);
1526 	struct in_conninfo *inc = &inp->inp_inc;
1527 	struct mbuf *mfree;
1528 	struct tcpopt to;
1529 	int tfo_syn;
1530 	u_int maxseg;
1531 
1532 #ifdef TCPDEBUG
1533 	/*
1534 	 * The size of tcp_saveipgen must be the size of the max ip header,
1535 	 * now IPv6.
1536 	 */
1537 	u_char tcp_saveipgen[IP6_HDR_LEN];
1538 	struct tcphdr tcp_savetcp;
1539 	short ostate = 0;
1540 #endif
1541 	thflags = tcp_get_flags(th);
1542 	tp->sackhint.last_sack_ack = 0;
1543 	sack_changed = 0;
1544 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1545 
1546 	NET_EPOCH_ASSERT();
1547 	INP_WLOCK_ASSERT(inp);
1548 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1549 	    __func__));
1550 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1551 	    __func__));
1552 
1553 #ifdef TCPPCAP
1554 	/* Save segment, if requested. */
1555 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1556 #endif
1557 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1558 	    tlen, NULL, true);
1559 
1560 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1561 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1562 			log(LOG_DEBUG, "%s; %s: "
1563 			    "SYN|FIN segment ignored (based on "
1564 			    "sysctl setting)\n", s, __func__);
1565 			free(s, M_TCPLOG);
1566 		}
1567 		goto drop;
1568 	}
1569 
1570 	/*
1571 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1572 	 * check SEQ.ACK first.
1573 	 */
1574 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1575 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1576 		rstreason = BANDLIM_UNLIMITED;
1577 		tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1578 		goto dropwithreset;
1579 	}
1580 
1581 	/*
1582 	 * Segment received on connection.
1583 	 * Reset idle time and keep-alive timer.
1584 	 * XXX: This should be done after segment
1585 	 * validation to ignore broken/spoofed segs.
1586 	 */
1587 	if  (tp->t_idle_reduce &&
1588 	     (tp->snd_max == tp->snd_una) &&
1589 	     ((ticks - tp->t_rcvtime) >= tp->t_rxtcur))
1590 		cc_after_idle(tp);
1591 	tp->t_rcvtime = ticks;
1592 
1593 	if (thflags & TH_FIN)
1594 		tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN);
1595 	/*
1596 	 * Scale up the window into a 32-bit value.
1597 	 * For the SYN_SENT state the scale is zero.
1598 	 */
1599 	tiwin = th->th_win << tp->snd_scale;
1600 #ifdef STATS
1601 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1602 #endif
1603 
1604 	/*
1605 	 * TCP ECN processing.
1606 	 */
1607 	if (tcp_ecn_input_segment(tp, thflags, tlen,
1608 	    tcp_packets_this_ack(tp, th->th_ack),
1609 	    iptos))
1610 		cc_cong_signal(tp, th, CC_ECN);
1611 
1612 	/*
1613 	 * Parse options on any incoming segment.
1614 	 */
1615 	tcp_dooptions(&to, (u_char *)(th + 1),
1616 	    (th->th_off << 2) - sizeof(struct tcphdr),
1617 	    (thflags & TH_SYN) ? TO_SYN : 0);
1618 
1619 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1620 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1621 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1622 		TCPSTAT_INC(tcps_sig_err_sigopt);
1623 		/* XXX: should drop? */
1624 	}
1625 #endif
1626 	/*
1627 	 * If echoed timestamp is later than the current time,
1628 	 * fall back to non RFC1323 RTT calculation.  Normalize
1629 	 * timestamp if syncookies were used when this connection
1630 	 * was established.
1631 	 */
1632 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1633 		to.to_tsecr -= tp->ts_offset;
1634 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1635 			to.to_tsecr = 0;
1636 		else if (tp->t_rxtshift == 1 &&
1637 			 tp->t_flags & TF_PREVVALID &&
1638 			 tp->t_badrxtwin != 0 &&
1639 			 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1640 			cc_cong_signal(tp, th, CC_RTO_ERR);
1641 	}
1642 	/*
1643 	 * Process options only when we get SYN/ACK back. The SYN case
1644 	 * for incoming connections is handled in tcp_syncache.
1645 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1646 	 * or <SYN,ACK>) segment itself is never scaled.
1647 	 * XXX this is traditional behavior, may need to be cleaned up.
1648 	 */
1649 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1650 		/* Handle parallel SYN for ECN */
1651 		tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1652 		if ((to.to_flags & TOF_SCALE) &&
1653 		    (tp->t_flags & TF_REQ_SCALE) &&
1654 		    !(tp->t_flags & TF_NOOPT)) {
1655 			tp->t_flags |= TF_RCVD_SCALE;
1656 			tp->snd_scale = to.to_wscale;
1657 		} else
1658 			tp->t_flags &= ~TF_REQ_SCALE;
1659 		/*
1660 		 * Initial send window.  It will be updated with
1661 		 * the next incoming segment to the scaled value.
1662 		 */
1663 		tp->snd_wnd = th->th_win;
1664 		if ((to.to_flags & TOF_TS) &&
1665 		    (tp->t_flags & TF_REQ_TSTMP) &&
1666 		    !(tp->t_flags & TF_NOOPT)) {
1667 			tp->t_flags |= TF_RCVD_TSTMP;
1668 			tp->ts_recent = to.to_tsval;
1669 			tp->ts_recent_age = tcp_ts_getticks();
1670 		} else
1671 			tp->t_flags &= ~TF_REQ_TSTMP;
1672 		if (to.to_flags & TOF_MSS)
1673 			tcp_mss(tp, to.to_mss);
1674 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1675 		    (!(to.to_flags & TOF_SACKPERM) ||
1676 		    (tp->t_flags & TF_NOOPT)))
1677 			tp->t_flags &= ~TF_SACK_PERMIT;
1678 		if (IS_FASTOPEN(tp->t_flags)) {
1679 			if ((to.to_flags & TOF_FASTOPEN) &&
1680 			    !(tp->t_flags & TF_NOOPT)) {
1681 				uint16_t mss;
1682 
1683 				if (to.to_flags & TOF_MSS)
1684 					mss = to.to_mss;
1685 				else
1686 					if ((inp->inp_vflag & INP_IPV6) != 0)
1687 						mss = TCP6_MSS;
1688 					else
1689 						mss = TCP_MSS;
1690 				tcp_fastopen_update_cache(tp, mss,
1691 				    to.to_tfo_len, to.to_tfo_cookie);
1692 			} else
1693 				tcp_fastopen_disable_path(tp);
1694 		}
1695 	}
1696 
1697 	/*
1698 	 * If timestamps were negotiated during SYN/ACK and a
1699 	 * segment without a timestamp is received, silently drop
1700 	 * the segment, unless it is a RST segment or missing timestamps are
1701 	 * tolerated.
1702 	 * See section 3.2 of RFC 7323.
1703 	 */
1704 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1705 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1706 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1707 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1708 				    "segment processed normally\n",
1709 				    s, __func__);
1710 				free(s, M_TCPLOG);
1711 			}
1712 		} else {
1713 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1714 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1715 				    "segment silently dropped\n", s, __func__);
1716 				free(s, M_TCPLOG);
1717 			}
1718 			goto drop;
1719 		}
1720 	}
1721 	/*
1722 	 * If timestamps were not negotiated during SYN/ACK and a
1723 	 * segment with a timestamp is received, ignore the
1724 	 * timestamp and process the packet normally.
1725 	 * See section 3.2 of RFC 7323.
1726 	 */
1727 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1728 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1729 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1730 			    "segment processed normally\n", s, __func__);
1731 			free(s, M_TCPLOG);
1732 		}
1733 	}
1734 
1735 	/*
1736 	 * Header prediction: check for the two common cases
1737 	 * of a uni-directional data xfer.  If the packet has
1738 	 * no control flags, is in-sequence, the window didn't
1739 	 * change and we're not retransmitting, it's a
1740 	 * candidate.  If the length is zero and the ack moved
1741 	 * forward, we're the sender side of the xfer.  Just
1742 	 * free the data acked & wake any higher level process
1743 	 * that was blocked waiting for space.  If the length
1744 	 * is non-zero and the ack didn't move, we're the
1745 	 * receiver side.  If we're getting packets in-order
1746 	 * (the reassembly queue is empty), add the data to
1747 	 * the socket buffer and note that we need a delayed ack.
1748 	 * Make sure that the hidden state-flags are also off.
1749 	 * Since we check for TCPS_ESTABLISHED first, it can only
1750 	 * be TH_NEEDSYN.
1751 	 */
1752 	if (tp->t_state == TCPS_ESTABLISHED &&
1753 	    th->th_seq == tp->rcv_nxt &&
1754 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1755 	    tp->snd_nxt == tp->snd_max &&
1756 	    tiwin && tiwin == tp->snd_wnd &&
1757 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1758 	    SEGQ_EMPTY(tp) &&
1759 	    ((to.to_flags & TOF_TS) == 0 ||
1760 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1761 		/*
1762 		 * If last ACK falls within this segment's sequence numbers,
1763 		 * record the timestamp.
1764 		 * NOTE that the test is modified according to the latest
1765 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1766 		 */
1767 		if ((to.to_flags & TOF_TS) != 0 &&
1768 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1769 			tp->ts_recent_age = tcp_ts_getticks();
1770 			tp->ts_recent = to.to_tsval;
1771 		}
1772 
1773 		if (tlen == 0) {
1774 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1775 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1776 			    !IN_RECOVERY(tp->t_flags) &&
1777 			    (to.to_flags & TOF_SACK) == 0 &&
1778 			    TAILQ_EMPTY(&tp->snd_holes)) {
1779 				/*
1780 				 * This is a pure ack for outstanding data.
1781 				 */
1782 				TCPSTAT_INC(tcps_predack);
1783 
1784 				/*
1785 				 * "bad retransmit" recovery without timestamps.
1786 				 */
1787 				if ((to.to_flags & TOF_TS) == 0 &&
1788 				    tp->t_rxtshift == 1 &&
1789 				    tp->t_flags & TF_PREVVALID &&
1790 				    tp->t_badrxtwin != 0 &&
1791 				    TSTMP_LT(ticks, tp->t_badrxtwin)) {
1792 					cc_cong_signal(tp, th, CC_RTO_ERR);
1793 				}
1794 
1795 				/*
1796 				 * Recalculate the transmit timer / rtt.
1797 				 *
1798 				 * Some boxes send broken timestamp replies
1799 				 * during the SYN+ACK phase, ignore
1800 				 * timestamps of 0 or we could calculate a
1801 				 * huge RTT and blow up the retransmit timer.
1802 				 */
1803 				if ((to.to_flags & TOF_TS) != 0 &&
1804 				    to.to_tsecr) {
1805 					uint32_t t;
1806 
1807 					t = tcp_ts_getticks() - to.to_tsecr;
1808 					if (!tp->t_rttlow || tp->t_rttlow > t)
1809 						tp->t_rttlow = t;
1810 					tcp_xmit_timer(tp,
1811 					    TCP_TS_TO_TICKS(t) + 1);
1812 				} else if (tp->t_rtttime &&
1813 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1814 					if (!tp->t_rttlow ||
1815 					    tp->t_rttlow > ticks - tp->t_rtttime)
1816 						tp->t_rttlow = ticks - tp->t_rtttime;
1817 					tcp_xmit_timer(tp,
1818 							ticks - tp->t_rtttime);
1819 				}
1820 				acked = BYTES_THIS_ACK(tp, th);
1821 
1822 #ifdef TCP_HHOOK
1823 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1824 				hhook_run_tcp_est_in(tp, th, &to);
1825 #endif
1826 
1827 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1828 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1829 				sbdrop(&so->so_snd, acked);
1830 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1831 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1832 					tp->snd_recover = th->th_ack - 1;
1833 
1834 				/*
1835 				 * Let the congestion control algorithm update
1836 				 * congestion control related information. This
1837 				 * typically means increasing the congestion
1838 				 * window.
1839 				 */
1840 				cc_ack_received(tp, th, nsegs, CC_ACK);
1841 
1842 				tp->snd_una = th->th_ack;
1843 				/*
1844 				 * Pull snd_wl2 up to prevent seq wrap relative
1845 				 * to th_ack.
1846 				 */
1847 				tp->snd_wl2 = th->th_ack;
1848 				tp->t_dupacks = 0;
1849 				m_freem(m);
1850 
1851 				/*
1852 				 * If all outstanding data are acked, stop
1853 				 * retransmit timer, otherwise restart timer
1854 				 * using current (possibly backed-off) value.
1855 				 * If process is waiting for space,
1856 				 * wakeup/selwakeup/signal.  If data
1857 				 * are ready to send, let tcp_output
1858 				 * decide between more output or persist.
1859 				 */
1860 #ifdef TCPDEBUG
1861 				if (so->so_options & SO_DEBUG)
1862 					tcp_trace(TA_INPUT, ostate, tp,
1863 					    (void *)tcp_saveipgen,
1864 					    &tcp_savetcp, 0);
1865 #endif
1866 				TCP_PROBE3(debug__input, tp, th, m);
1867 				/*
1868 				 * Clear t_acktime if remote side has ACKd
1869 				 * all data in the socket buffer.
1870 				 * Otherwise, update t_acktime if we received
1871 				 * a sufficiently large ACK.
1872 				 */
1873 				if (sbavail(&so->so_snd) == 0)
1874 					tp->t_acktime = 0;
1875 				else if (acked > 1)
1876 					tp->t_acktime = ticks;
1877 				if (tp->snd_una == tp->snd_max)
1878 					tcp_timer_activate(tp, TT_REXMT, 0);
1879 				else if (!tcp_timer_active(tp, TT_PERSIST))
1880 					tcp_timer_activate(tp, TT_REXMT,
1881 					    TP_RXTCUR(tp));
1882 				sowwakeup(so);
1883 				if (sbavail(&so->so_snd))
1884 					(void) tcp_output(tp);
1885 				goto check_delack;
1886 			}
1887 		} else if (th->th_ack == tp->snd_una &&
1888 		    tlen <= sbspace(&so->so_rcv)) {
1889 			int newsize = 0;	/* automatic sockbuf scaling */
1890 
1891 			/*
1892 			 * This is a pure, in-sequence data packet with
1893 			 * nothing on the reassembly queue and we have enough
1894 			 * buffer space to take it.
1895 			 */
1896 			/* Clean receiver SACK report if present */
1897 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1898 				tcp_clean_sackreport(tp);
1899 			TCPSTAT_INC(tcps_preddat);
1900 			tp->rcv_nxt += tlen;
1901 			if (tlen &&
1902 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1903 			    (tp->t_fbyte_in == 0)) {
1904 				tp->t_fbyte_in = ticks;
1905 				if (tp->t_fbyte_in == 0)
1906 					tp->t_fbyte_in = 1;
1907 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1908 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1909 			}
1910 			/*
1911 			 * Pull snd_wl1 up to prevent seq wrap relative to
1912 			 * th_seq.
1913 			 */
1914 			tp->snd_wl1 = th->th_seq;
1915 			/*
1916 			 * Pull rcv_up up to prevent seq wrap relative to
1917 			 * rcv_nxt.
1918 			 */
1919 			tp->rcv_up = tp->rcv_nxt;
1920 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1921 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1922 #ifdef TCPDEBUG
1923 			if (so->so_options & SO_DEBUG)
1924 				tcp_trace(TA_INPUT, ostate, tp,
1925 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1926 #endif
1927 			TCP_PROBE3(debug__input, tp, th, m);
1928 
1929 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1930 
1931 			/* Add data to socket buffer. */
1932 			SOCKBUF_LOCK(&so->so_rcv);
1933 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1934 				m_freem(m);
1935 			} else {
1936 				/*
1937 				 * Set new socket buffer size.
1938 				 * Give up when limit is reached.
1939 				 */
1940 				if (newsize)
1941 					if (!sbreserve_locked(so, SO_RCV,
1942 					    newsize, NULL))
1943 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1944 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1945 				sbappendstream_locked(&so->so_rcv, m, 0);
1946 			}
1947 			/* NB: sorwakeup_locked() does an implicit unlock. */
1948 			sorwakeup_locked(so);
1949 			if (DELAY_ACK(tp, tlen)) {
1950 				tp->t_flags |= TF_DELACK;
1951 			} else {
1952 				tp->t_flags |= TF_ACKNOW;
1953 				tcp_output(tp);
1954 			}
1955 			goto check_delack;
1956 		}
1957 	}
1958 
1959 	/*
1960 	 * Calculate amount of space in receive window,
1961 	 * and then do TCP input processing.
1962 	 * Receive window is amount of space in rcv queue,
1963 	 * but not less than advertised window.
1964 	 */
1965 	win = sbspace(&so->so_rcv);
1966 	if (win < 0)
1967 		win = 0;
1968 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1969 
1970 	switch (tp->t_state) {
1971 	/*
1972 	 * If the state is SYN_RECEIVED:
1973 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1974 	 */
1975 	case TCPS_SYN_RECEIVED:
1976 		if ((thflags & TH_ACK) &&
1977 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1978 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1979 				rstreason = BANDLIM_RST_OPENPORT;
1980 				tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1981 				goto dropwithreset;
1982 		}
1983 		if (IS_FASTOPEN(tp->t_flags)) {
1984 			/*
1985 			 * When a TFO connection is in SYN_RECEIVED, the
1986 			 * only valid packets are the initial SYN, a
1987 			 * retransmit/copy of the initial SYN (possibly with
1988 			 * a subset of the original data), a valid ACK, a
1989 			 * FIN, or a RST.
1990 			 */
1991 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1992 				rstreason = BANDLIM_RST_OPENPORT;
1993 				tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1994 				goto dropwithreset;
1995 			} else if (thflags & TH_SYN) {
1996 				/* non-initial SYN is ignored */
1997 				if ((tcp_timer_active(tp, TT_DELACK) ||
1998 				     tcp_timer_active(tp, TT_REXMT)))
1999 					goto drop;
2000 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2001 				goto drop;
2002 			}
2003 		}
2004 		break;
2005 
2006 	/*
2007 	 * If the state is SYN_SENT:
2008 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
2009 	 *	    been verified), then drop the connection.
2010 	 *	if seg contains a RST without an ACK, drop the seg.
2011 	 *	if seg does not contain SYN, then drop the seg.
2012 	 * Otherwise this is an acceptable SYN segment
2013 	 *	initialize tp->rcv_nxt and tp->irs
2014 	 *	if seg contains ack then advance tp->snd_una
2015 	 *	if seg contains an ECE and ECN support is enabled, the stream
2016 	 *	    is ECN capable.
2017 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2018 	 *	arrange for segment to be acked (eventually)
2019 	 *	continue processing rest of data/controls, beginning with URG
2020 	 */
2021 	case TCPS_SYN_SENT:
2022 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2023 			TCP_PROBE5(connect__refused, NULL, tp,
2024 			    m, tp, th);
2025 			tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2026 			tp = tcp_drop(tp, ECONNREFUSED);
2027 		}
2028 		if (thflags & TH_RST)
2029 			goto drop;
2030 		if (!(thflags & TH_SYN))
2031 			goto drop;
2032 
2033 		tp->irs = th->th_seq;
2034 		tcp_rcvseqinit(tp);
2035 		if (thflags & TH_ACK) {
2036 			int tfo_partial_ack = 0;
2037 
2038 			TCPSTAT_INC(tcps_connects);
2039 			soisconnected(so);
2040 #ifdef MAC
2041 			mac_socketpeer_set_from_mbuf(m, so);
2042 #endif
2043 			/* Do window scaling on this connection? */
2044 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2045 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2046 				tp->rcv_scale = tp->request_r_scale;
2047 			}
2048 			tp->rcv_adv += min(tp->rcv_wnd,
2049 			    TCP_MAXWIN << tp->rcv_scale);
2050 			tp->snd_una++;		/* SYN is acked */
2051 			/*
2052 			 * If not all the data that was sent in the TFO SYN
2053 			 * has been acked, resend the remainder right away.
2054 			 */
2055 			if (IS_FASTOPEN(tp->t_flags) &&
2056 			    (tp->snd_una != tp->snd_max)) {
2057 				tp->snd_nxt = th->th_ack;
2058 				tfo_partial_ack = 1;
2059 			}
2060 			/*
2061 			 * If there's data, delay ACK; if there's also a FIN
2062 			 * ACKNOW will be turned on later.
2063 			 */
2064 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2065 				tcp_timer_activate(tp, TT_DELACK,
2066 				    tcp_delacktime);
2067 			else
2068 				tp->t_flags |= TF_ACKNOW;
2069 
2070 			tcp_ecn_input_syn_sent(tp, thflags, iptos);
2071 
2072 			/*
2073 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2074 			 * Transitions:
2075 			 *	SYN_SENT  --> ESTABLISHED
2076 			 *	SYN_SENT* --> FIN_WAIT_1
2077 			 */
2078 			tp->t_starttime = ticks;
2079 			if (tp->t_flags & TF_NEEDFIN) {
2080 				tp->t_acktime = ticks;
2081 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2082 				tp->t_flags &= ~TF_NEEDFIN;
2083 				thflags &= ~TH_SYN;
2084 			} else {
2085 				tcp_state_change(tp, TCPS_ESTABLISHED);
2086 				TCP_PROBE5(connect__established, NULL, tp,
2087 				    m, tp, th);
2088 				cc_conn_init(tp);
2089 				tcp_timer_activate(tp, TT_KEEP,
2090 				    TP_KEEPIDLE(tp));
2091 			}
2092 		} else {
2093 			/*
2094 			 * Received initial SYN in SYN-SENT[*] state =>
2095 			 * simultaneous open.
2096 			 * If it succeeds, connection is * half-synchronized.
2097 			 * Otherwise, do 3-way handshake:
2098 			 *        SYN-SENT -> SYN-RECEIVED
2099 			 *        SYN-SENT* -> SYN-RECEIVED*
2100 			 */
2101 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN);
2102 			tcp_timer_activate(tp, TT_REXMT, 0);
2103 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2104 		}
2105 
2106 		/*
2107 		 * Advance th->th_seq to correspond to first data byte.
2108 		 * If data, trim to stay within window,
2109 		 * dropping FIN if necessary.
2110 		 */
2111 		th->th_seq++;
2112 		if (tlen > tp->rcv_wnd) {
2113 			todrop = tlen - tp->rcv_wnd;
2114 			m_adj(m, -todrop);
2115 			tlen = tp->rcv_wnd;
2116 			thflags &= ~TH_FIN;
2117 			TCPSTAT_INC(tcps_rcvpackafterwin);
2118 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2119 		}
2120 		tp->snd_wl1 = th->th_seq - 1;
2121 		tp->rcv_up = th->th_seq;
2122 		/*
2123 		 * Client side of transaction: already sent SYN and data.
2124 		 * If the remote host used T/TCP to validate the SYN,
2125 		 * our data will be ACK'd; if so, enter normal data segment
2126 		 * processing in the middle of step 5, ack processing.
2127 		 * Otherwise, goto step 6.
2128 		 */
2129 		if (thflags & TH_ACK)
2130 			goto process_ACK;
2131 
2132 		goto step6;
2133 	}
2134 
2135 	/*
2136 	 * States other than LISTEN or SYN_SENT.
2137 	 * First check the RST flag and sequence number since reset segments
2138 	 * are exempt from the timestamp and connection count tests.  This
2139 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2140 	 * below which allowed reset segments in half the sequence space
2141 	 * to fall though and be processed (which gives forged reset
2142 	 * segments with a random sequence number a 50 percent chance of
2143 	 * killing a connection).
2144 	 * Then check timestamp, if present.
2145 	 * Then check the connection count, if present.
2146 	 * Then check that at least some bytes of segment are within
2147 	 * receive window.  If segment begins before rcv_nxt,
2148 	 * drop leading data (and SYN); if nothing left, just ack.
2149 	 */
2150 	if (thflags & TH_RST) {
2151 		/*
2152 		 * RFC5961 Section 3.2
2153 		 *
2154 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2155 		 * - If RST is in window, we send challenge ACK.
2156 		 *
2157 		 * Note: to take into account delayed ACKs, we should
2158 		 *   test against last_ack_sent instead of rcv_nxt.
2159 		 * Note 2: we handle special case of closed window, not
2160 		 *   covered by the RFC.
2161 		 */
2162 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2163 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2164 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2165 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2166 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2167 			    __func__, th, tp));
2168 
2169 			if (V_tcp_insecure_rst ||
2170 			    tp->last_ack_sent == th->th_seq) {
2171 				TCPSTAT_INC(tcps_drops);
2172 				/* Drop the connection. */
2173 				switch (tp->t_state) {
2174 				case TCPS_SYN_RECEIVED:
2175 					so->so_error = ECONNREFUSED;
2176 					goto close;
2177 				case TCPS_ESTABLISHED:
2178 				case TCPS_FIN_WAIT_1:
2179 				case TCPS_FIN_WAIT_2:
2180 				case TCPS_CLOSE_WAIT:
2181 				case TCPS_CLOSING:
2182 				case TCPS_LAST_ACK:
2183 					so->so_error = ECONNRESET;
2184 				close:
2185 					/* FALLTHROUGH */
2186 				default:
2187 					tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
2188 					tp = tcp_close(tp);
2189 				}
2190 			} else {
2191 				TCPSTAT_INC(tcps_badrst);
2192 				/* Send challenge ACK. */
2193 				tcp_respond(tp, mtod(m, void *), th, m,
2194 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2195 				tp->last_ack_sent = tp->rcv_nxt;
2196 				m = NULL;
2197 			}
2198 		}
2199 		goto drop;
2200 	}
2201 
2202 	/*
2203 	 * RFC5961 Section 4.2
2204 	 * Send challenge ACK for any SYN in synchronized state.
2205 	 */
2206 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2207 	    tp->t_state != TCPS_SYN_RECEIVED) {
2208 		TCPSTAT_INC(tcps_badsyn);
2209 		if (V_tcp_insecure_syn &&
2210 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2211 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2212 			tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2213 			tp = tcp_drop(tp, ECONNRESET);
2214 			rstreason = BANDLIM_UNLIMITED;
2215 		} else {
2216 			tcp_ecn_input_syn_sent(tp, thflags, iptos);
2217 			/* Send challenge ACK. */
2218 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2219 			    tp->snd_nxt, TH_ACK);
2220 			tp->last_ack_sent = tp->rcv_nxt;
2221 			m = NULL;
2222 		}
2223 		goto drop;
2224 	}
2225 
2226 	/*
2227 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2228 	 * and it's less than ts_recent, drop it.
2229 	 */
2230 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2231 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2232 		/* Check to see if ts_recent is over 24 days old.  */
2233 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2234 			/*
2235 			 * Invalidate ts_recent.  If this segment updates
2236 			 * ts_recent, the age will be reset later and ts_recent
2237 			 * will get a valid value.  If it does not, setting
2238 			 * ts_recent to zero will at least satisfy the
2239 			 * requirement that zero be placed in the timestamp
2240 			 * echo reply when ts_recent isn't valid.  The
2241 			 * age isn't reset until we get a valid ts_recent
2242 			 * because we don't want out-of-order segments to be
2243 			 * dropped when ts_recent is old.
2244 			 */
2245 			tp->ts_recent = 0;
2246 		} else {
2247 			TCPSTAT_INC(tcps_rcvduppack);
2248 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2249 			TCPSTAT_INC(tcps_pawsdrop);
2250 			if (tlen)
2251 				goto dropafterack;
2252 			goto drop;
2253 		}
2254 	}
2255 
2256 	/*
2257 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2258 	 * this connection before trimming the data to fit the receive
2259 	 * window.  Check the sequence number versus IRS since we know
2260 	 * the sequence numbers haven't wrapped.  This is a partial fix
2261 	 * for the "LAND" DoS attack.
2262 	 */
2263 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2264 		rstreason = BANDLIM_RST_OPENPORT;
2265 		tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2266 		goto dropwithreset;
2267 	}
2268 
2269 	todrop = tp->rcv_nxt - th->th_seq;
2270 	if (todrop > 0) {
2271 		if (thflags & TH_SYN) {
2272 			thflags &= ~TH_SYN;
2273 			th->th_seq++;
2274 			if (th->th_urp > 1)
2275 				th->th_urp--;
2276 			else
2277 				thflags &= ~TH_URG;
2278 			todrop--;
2279 		}
2280 		/*
2281 		 * Following if statement from Stevens, vol. 2, p. 960.
2282 		 */
2283 		if (todrop > tlen
2284 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2285 			/*
2286 			 * Any valid FIN must be to the left of the window.
2287 			 * At this point the FIN must be a duplicate or out
2288 			 * of sequence; drop it.
2289 			 */
2290 			thflags &= ~TH_FIN;
2291 
2292 			/*
2293 			 * Send an ACK to resynchronize and drop any data.
2294 			 * But keep on processing for RST or ACK.
2295 			 */
2296 			tp->t_flags |= TF_ACKNOW;
2297 			todrop = tlen;
2298 			TCPSTAT_INC(tcps_rcvduppack);
2299 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2300 		} else {
2301 			TCPSTAT_INC(tcps_rcvpartduppack);
2302 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2303 		}
2304 		/*
2305 		 * DSACK - add SACK block for dropped range
2306 		 */
2307 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2308 			tcp_update_sack_list(tp, th->th_seq,
2309 			    th->th_seq + todrop);
2310 			/*
2311 			 * ACK now, as the next in-sequence segment
2312 			 * will clear the DSACK block again
2313 			 */
2314 			tp->t_flags |= TF_ACKNOW;
2315 		}
2316 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2317 		th->th_seq += todrop;
2318 		tlen -= todrop;
2319 		if (th->th_urp > todrop)
2320 			th->th_urp -= todrop;
2321 		else {
2322 			thflags &= ~TH_URG;
2323 			th->th_urp = 0;
2324 		}
2325 	}
2326 
2327 	/*
2328 	 * If new data are received on a connection after the
2329 	 * user processes are gone, then RST the other end.
2330 	 */
2331 	if ((tp->t_flags & TF_CLOSED) && tlen) {
2332 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2333 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2334 			    "after socket was closed, "
2335 			    "sending RST and removing tcpcb\n",
2336 			    s, __func__, tcpstates[tp->t_state], tlen);
2337 			free(s, M_TCPLOG);
2338 		}
2339 		tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE);
2340 		/* tcp_close will kill the inp pre-log the Reset */
2341 		tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2342 		tp = tcp_close(tp);
2343 		TCPSTAT_INC(tcps_rcvafterclose);
2344 		rstreason = BANDLIM_UNLIMITED;
2345 		goto dropwithreset;
2346 	}
2347 
2348 	/*
2349 	 * If segment ends after window, drop trailing data
2350 	 * (and PUSH and FIN); if nothing left, just ACK.
2351 	 */
2352 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2353 	if (todrop > 0) {
2354 		TCPSTAT_INC(tcps_rcvpackafterwin);
2355 		if (todrop >= tlen) {
2356 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2357 			/*
2358 			 * If window is closed can only take segments at
2359 			 * window edge, and have to drop data and PUSH from
2360 			 * incoming segments.  Continue processing, but
2361 			 * remember to ack.  Otherwise, drop segment
2362 			 * and ack.
2363 			 */
2364 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2365 				tp->t_flags |= TF_ACKNOW;
2366 				TCPSTAT_INC(tcps_rcvwinprobe);
2367 			} else
2368 				goto dropafterack;
2369 		} else
2370 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2371 		m_adj(m, -todrop);
2372 		tlen -= todrop;
2373 		thflags &= ~(TH_PUSH|TH_FIN);
2374 	}
2375 
2376 	/*
2377 	 * If last ACK falls within this segment's sequence numbers,
2378 	 * record its timestamp.
2379 	 * NOTE:
2380 	 * 1) That the test incorporates suggestions from the latest
2381 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2382 	 * 2) That updating only on newer timestamps interferes with
2383 	 *    our earlier PAWS tests, so this check should be solely
2384 	 *    predicated on the sequence space of this segment.
2385 	 * 3) That we modify the segment boundary check to be
2386 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2387 	 *    instead of RFC1323's
2388 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2389 	 *    This modified check allows us to overcome RFC1323's
2390 	 *    limitations as described in Stevens TCP/IP Illustrated
2391 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2392 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2393 	 */
2394 	if ((to.to_flags & TOF_TS) != 0 &&
2395 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2396 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2397 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2398 		tp->ts_recent_age = tcp_ts_getticks();
2399 		tp->ts_recent = to.to_tsval;
2400 	}
2401 
2402 	/*
2403 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2404 	 * flag is on (half-synchronized state), then queue data for
2405 	 * later processing; else drop segment and return.
2406 	 */
2407 	if ((thflags & TH_ACK) == 0) {
2408 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2409 		    (tp->t_flags & TF_NEEDSYN)) {
2410 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2411 			    IS_FASTOPEN(tp->t_flags)) {
2412 				tp->snd_wnd = tiwin;
2413 				cc_conn_init(tp);
2414 			}
2415 			goto step6;
2416 		} else if (tp->t_flags & TF_ACKNOW)
2417 			goto dropafterack;
2418 		else
2419 			goto drop;
2420 	}
2421 
2422 	/*
2423 	 * Ack processing.
2424 	 */
2425 	switch (tp->t_state) {
2426 	/*
2427 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2428 	 * ESTABLISHED state and continue processing.
2429 	 * The ACK was checked above.
2430 	 */
2431 	case TCPS_SYN_RECEIVED:
2432 
2433 		TCPSTAT_INC(tcps_connects);
2434 		if (tp->t_flags & TF_SONOTCONN) {
2435 			/*
2436 			 * Usually SYN_RECEIVED had been created from a LISTEN,
2437 			 * and solisten_enqueue() has already marked the socket
2438 			 * layer as connected.  If it didn't, which can happen
2439 			 * only with an accept_filter(9), then the tp is marked
2440 			 * with TF_SONOTCONN.  The other reason for this mark
2441 			 * to be set is a simultaneous open, a SYN_RECEIVED
2442 			 * that had been created from SYN_SENT.
2443 			 */
2444 			tp->t_flags &= ~TF_SONOTCONN;
2445 			soisconnected(so);
2446 		}
2447 		/* Do window scaling? */
2448 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2449 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2450 			tp->rcv_scale = tp->request_r_scale;
2451 		}
2452 		tp->snd_wnd = tiwin;
2453 		/*
2454 		 * Make transitions:
2455 		 *      SYN-RECEIVED  -> ESTABLISHED
2456 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2457 		 */
2458 		tp->t_starttime = ticks;
2459 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2460 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2461 			tp->t_tfo_pending = NULL;
2462 		}
2463 		if (tp->t_flags & TF_NEEDFIN) {
2464 			tp->t_acktime = ticks;
2465 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2466 			tp->t_flags &= ~TF_NEEDFIN;
2467 		} else {
2468 			tcp_state_change(tp, TCPS_ESTABLISHED);
2469 			TCP_PROBE5(accept__established, NULL, tp,
2470 			    m, tp, th);
2471 			/*
2472 			 * TFO connections call cc_conn_init() during SYN
2473 			 * processing.  Calling it again here for such
2474 			 * connections is not harmless as it would undo the
2475 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2476 			 * is retransmitted.
2477 			 */
2478 			if (!IS_FASTOPEN(tp->t_flags))
2479 				cc_conn_init(tp);
2480 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2481 		}
2482 		/*
2483 		 * Account for the ACK of our SYN prior to
2484 		 * regular ACK processing below, except for
2485 		 * simultaneous SYN, which is handled later.
2486 		 */
2487 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2488 			incforsyn = 1;
2489 		/*
2490 		 * If segment contains data or ACK, will call tcp_reass()
2491 		 * later; if not, do so now to pass queued data to user.
2492 		 */
2493 		if (tlen == 0 && (thflags & TH_FIN) == 0) {
2494 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2495 			    (struct mbuf *)0);
2496 			tcp_handle_wakeup(tp);
2497 		}
2498 		tp->snd_wl1 = th->th_seq - 1;
2499 		/* FALLTHROUGH */
2500 
2501 	/*
2502 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2503 	 * ACKs.  If the ack is in the range
2504 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2505 	 * then advance tp->snd_una to th->th_ack and drop
2506 	 * data from the retransmission queue.  If this ACK reflects
2507 	 * more up to date window information we update our window information.
2508 	 */
2509 	case TCPS_ESTABLISHED:
2510 	case TCPS_FIN_WAIT_1:
2511 	case TCPS_FIN_WAIT_2:
2512 	case TCPS_CLOSE_WAIT:
2513 	case TCPS_CLOSING:
2514 	case TCPS_LAST_ACK:
2515 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2516 			TCPSTAT_INC(tcps_rcvacktoomuch);
2517 			goto dropafterack;
2518 		}
2519 		if (tcp_is_sack_recovery(tp, &to)) {
2520 			if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2521 			    (tp->t_flags & TF_LRD)) {
2522 				tcp_sack_lost_retransmission(tp, th);
2523 			}
2524 		} else
2525 			/*
2526 			 * Reset the value so that previous (valid) value
2527 			 * from the last ack with SACK doesn't get used.
2528 			 */
2529 			tp->sackhint.sacked_bytes = 0;
2530 
2531 #ifdef TCP_HHOOK
2532 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2533 		hhook_run_tcp_est_in(tp, th, &to);
2534 #endif
2535 
2536 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2537 			maxseg = tcp_maxseg(tp);
2538 			if (tlen == 0 &&
2539 			    (tiwin == tp->snd_wnd ||
2540 			    (tp->t_flags & TF_SACK_PERMIT))) {
2541 				/*
2542 				 * If this is the first time we've seen a
2543 				 * FIN from the remote, this is not a
2544 				 * duplicate and it needs to be processed
2545 				 * normally.  This happens during a
2546 				 * simultaneous close.
2547 				 */
2548 				if ((thflags & TH_FIN) &&
2549 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2550 					tp->t_dupacks = 0;
2551 					break;
2552 				}
2553 				TCPSTAT_INC(tcps_rcvdupack);
2554 				/*
2555 				 * If we have outstanding data (other than
2556 				 * a window probe), this is a completely
2557 				 * duplicate ack (ie, window info didn't
2558 				 * change and FIN isn't set),
2559 				 * the ack is the biggest we've
2560 				 * seen and we've seen exactly our rexmt
2561 				 * threshold of them, assume a packet
2562 				 * has been dropped and retransmit it.
2563 				 * Kludge snd_nxt & the congestion
2564 				 * window so we send only this one
2565 				 * packet.
2566 				 *
2567 				 * We know we're losing at the current
2568 				 * window size so do congestion avoidance
2569 				 * (set ssthresh to half the current window
2570 				 * and pull our congestion window back to
2571 				 * the new ssthresh).
2572 				 *
2573 				 * Dup acks mean that packets have left the
2574 				 * network (they're now cached at the receiver)
2575 				 * so bump cwnd by the amount in the receiver
2576 				 * to keep a constant cwnd packets in the
2577 				 * network.
2578 				 *
2579 				 * When using TCP ECN, notify the peer that
2580 				 * we reduced the cwnd.
2581 				 */
2582 				/*
2583 				 * Following 2 kinds of acks should not affect
2584 				 * dupack counting:
2585 				 * 1) Old acks
2586 				 * 2) Acks with SACK but without any new SACK
2587 				 * information in them. These could result from
2588 				 * any anomaly in the network like a switch
2589 				 * duplicating packets or a possible DoS attack.
2590 				 */
2591 				if (th->th_ack != tp->snd_una ||
2592 				    (tcp_is_sack_recovery(tp, &to) &&
2593 				    !sack_changed))
2594 					break;
2595 				else if (!tcp_timer_active(tp, TT_REXMT))
2596 					tp->t_dupacks = 0;
2597 				else if (++tp->t_dupacks > tcprexmtthresh ||
2598 				     IN_FASTRECOVERY(tp->t_flags)) {
2599 					cc_ack_received(tp, th, nsegs,
2600 					    CC_DUPACK);
2601 					if (V_tcp_do_prr &&
2602 					    IN_FASTRECOVERY(tp->t_flags)) {
2603 						tcp_do_prr_ack(tp, th, &to);
2604 					} else if (tcp_is_sack_recovery(tp, &to) &&
2605 					    IN_FASTRECOVERY(tp->t_flags)) {
2606 						int awnd;
2607 
2608 						/*
2609 						 * Compute the amount of data in flight first.
2610 						 * We can inject new data into the pipe iff
2611 						 * we have less than 1/2 the original window's
2612 						 * worth of data in flight.
2613 						 */
2614 						if (V_tcp_do_newsack)
2615 							awnd = tcp_compute_pipe(tp);
2616 						else
2617 							awnd = (tp->snd_nxt - tp->snd_fack) +
2618 								tp->sackhint.sack_bytes_rexmit;
2619 
2620 						if (awnd < tp->snd_ssthresh) {
2621 							tp->snd_cwnd += maxseg;
2622 							if (tp->snd_cwnd > tp->snd_ssthresh)
2623 								tp->snd_cwnd = tp->snd_ssthresh;
2624 						}
2625 					} else
2626 						tp->snd_cwnd += maxseg;
2627 					(void) tcp_output(tp);
2628 					goto drop;
2629 				} else if (tp->t_dupacks == tcprexmtthresh ||
2630 					    (tp->t_flags & TF_SACK_PERMIT &&
2631 					     V_tcp_do_newsack &&
2632 					     tp->sackhint.sacked_bytes >
2633 					     (tcprexmtthresh - 1) * maxseg)) {
2634 enter_recovery:
2635 					/*
2636 					 * Above is the RFC6675 trigger condition of
2637 					 * more than (dupthresh-1)*maxseg sacked data.
2638 					 * If the count of holes in the
2639 					 * scoreboard is >= dupthresh, we could
2640 					 * also enter loss recovery, but don't
2641 					 * have that value readily available.
2642 					 */
2643 					tp->t_dupacks = tcprexmtthresh;
2644 					tcp_seq onxt = tp->snd_nxt;
2645 
2646 					/*
2647 					 * If we're doing sack, or prr, check
2648 					 * to see if we're already in sack
2649 					 * recovery. If we're not doing sack,
2650 					 * check to see if we're in newreno
2651 					 * recovery.
2652 					 */
2653 					if (V_tcp_do_prr ||
2654 					    (tp->t_flags & TF_SACK_PERMIT)) {
2655 						if (IN_FASTRECOVERY(tp->t_flags)) {
2656 							tp->t_dupacks = 0;
2657 							break;
2658 						}
2659 					} else {
2660 						if (SEQ_LEQ(th->th_ack,
2661 						    tp->snd_recover)) {
2662 							tp->t_dupacks = 0;
2663 							break;
2664 						}
2665 					}
2666 					/* Congestion signal before ack. */
2667 					cc_cong_signal(tp, th, CC_NDUPACK);
2668 					cc_ack_received(tp, th, nsegs,
2669 					    CC_DUPACK);
2670 					tcp_timer_activate(tp, TT_REXMT, 0);
2671 					tp->t_rtttime = 0;
2672 					if (V_tcp_do_prr) {
2673 						/*
2674 						 * snd_ssthresh is already updated by
2675 						 * cc_cong_signal.
2676 						 */
2677 						if (tcp_is_sack_recovery(tp, &to)) {
2678 							tp->sackhint.prr_delivered =
2679 							    tp->sackhint.sacked_bytes;
2680 						} else {
2681 							tp->sackhint.prr_delivered =
2682 							    imin(tp->snd_max - tp->snd_una,
2683 							    imin(INT_MAX / 65536,
2684 								tp->t_dupacks) * maxseg);
2685 						}
2686 						tp->sackhint.recover_fs = max(1,
2687 						    tp->snd_nxt - tp->snd_una);
2688 					}
2689 					if (tcp_is_sack_recovery(tp, &to)) {
2690 						TCPSTAT_INC(
2691 						    tcps_sack_recovery_episode);
2692 						tp->snd_recover = tp->snd_nxt;
2693 						tp->snd_cwnd = maxseg;
2694 						(void) tcp_output(tp);
2695 						if (SEQ_GT(th->th_ack, tp->snd_una))
2696 							goto resume_partialack;
2697 						goto drop;
2698 					}
2699 					tp->snd_nxt = th->th_ack;
2700 					tp->snd_cwnd = maxseg;
2701 					(void) tcp_output(tp);
2702 					KASSERT(tp->snd_limited <= 2,
2703 					    ("%s: tp->snd_limited too big",
2704 					    __func__));
2705 					tp->snd_cwnd = tp->snd_ssthresh +
2706 					     maxseg *
2707 					     (tp->t_dupacks - tp->snd_limited);
2708 					if (SEQ_GT(onxt, tp->snd_nxt))
2709 						tp->snd_nxt = onxt;
2710 					goto drop;
2711 				} else if (V_tcp_do_rfc3042) {
2712 					/*
2713 					 * Process first and second duplicate
2714 					 * ACKs. Each indicates a segment
2715 					 * leaving the network, creating room
2716 					 * for more. Make sure we can send a
2717 					 * packet on reception of each duplicate
2718 					 * ACK by increasing snd_cwnd by one
2719 					 * segment. Restore the original
2720 					 * snd_cwnd after packet transmission.
2721 					 */
2722 					cc_ack_received(tp, th, nsegs,
2723 					    CC_DUPACK);
2724 					uint32_t oldcwnd = tp->snd_cwnd;
2725 					tcp_seq oldsndmax = tp->snd_max;
2726 					u_int sent;
2727 					int avail;
2728 
2729 					KASSERT(tp->t_dupacks == 1 ||
2730 					    tp->t_dupacks == 2,
2731 					    ("%s: dupacks not 1 or 2",
2732 					    __func__));
2733 					if (tp->t_dupacks == 1)
2734 						tp->snd_limited = 0;
2735 					tp->snd_cwnd =
2736 					    (tp->snd_nxt - tp->snd_una) +
2737 					    (tp->t_dupacks - tp->snd_limited) *
2738 					    maxseg;
2739 					/*
2740 					 * Only call tcp_output when there
2741 					 * is new data available to be sent
2742 					 * or we need to send an ACK.
2743 					 */
2744 					SOCKBUF_LOCK(&so->so_snd);
2745 					avail = sbavail(&so->so_snd) -
2746 					    (tp->snd_nxt - tp->snd_una);
2747 					SOCKBUF_UNLOCK(&so->so_snd);
2748 					if (avail > 0 || tp->t_flags & TF_ACKNOW)
2749 						(void) tcp_output(tp);
2750 					sent = tp->snd_max - oldsndmax;
2751 					if (sent > maxseg) {
2752 						KASSERT((tp->t_dupacks == 2 &&
2753 						    tp->snd_limited == 0) ||
2754 						   (sent == maxseg + 1 &&
2755 						    tp->t_flags & TF_SENTFIN),
2756 						    ("%s: sent too much",
2757 						    __func__));
2758 						tp->snd_limited = 2;
2759 					} else if (sent > 0)
2760 						++tp->snd_limited;
2761 					tp->snd_cwnd = oldcwnd;
2762 					goto drop;
2763 				}
2764 			}
2765 			break;
2766 		} else {
2767 			/*
2768 			 * This ack is advancing the left edge, reset the
2769 			 * counter.
2770 			 */
2771 			tp->t_dupacks = 0;
2772 			/*
2773 			 * If this ack also has new SACK info, increment the
2774 			 * counter as per rfc6675. The variable
2775 			 * sack_changed tracks all changes to the SACK
2776 			 * scoreboard, including when partial ACKs without
2777 			 * SACK options are received, and clear the scoreboard
2778 			 * from the left side. Such partial ACKs should not be
2779 			 * counted as dupacks here.
2780 			 */
2781 			if (tcp_is_sack_recovery(tp, &to) &&
2782 			    sack_changed) {
2783 				tp->t_dupacks++;
2784 				/* limit overhead by setting maxseg last */
2785 				if (!IN_FASTRECOVERY(tp->t_flags) &&
2786 				    (tp->sackhint.sacked_bytes >
2787 				    ((tcprexmtthresh - 1) *
2788 				    (maxseg = tcp_maxseg(tp))))) {
2789 					goto enter_recovery;
2790 				}
2791 			}
2792 		}
2793 
2794 resume_partialack:
2795 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2796 		    ("%s: th_ack <= snd_una", __func__));
2797 
2798 		/*
2799 		 * If the congestion window was inflated to account
2800 		 * for the other side's cached packets, retract it.
2801 		 */
2802 		if (IN_FASTRECOVERY(tp->t_flags)) {
2803 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2804 				if (tp->t_flags & TF_SACK_PERMIT)
2805 					if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2806 						tcp_timer_activate(tp, TT_REXMT, 0);
2807 						tp->t_rtttime = 0;
2808 						tcp_do_prr_ack(tp, th, &to);
2809 						tp->t_flags |= TF_ACKNOW;
2810 						(void) tcp_output(tp);
2811 					} else
2812 						tcp_sack_partialack(tp, th);
2813 				else
2814 					tcp_newreno_partial_ack(tp, th);
2815 			} else
2816 				cc_post_recovery(tp, th);
2817 		} else if (IN_CONGRECOVERY(tp->t_flags)) {
2818 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2819 				if (V_tcp_do_prr) {
2820 					tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2821 					tp->snd_fack = th->th_ack;
2822 					tcp_do_prr_ack(tp, th, &to);
2823 					(void) tcp_output(tp);
2824 				}
2825 			} else
2826 				cc_post_recovery(tp, th);
2827 		}
2828 		/*
2829 		 * If we reach this point, ACK is not a duplicate,
2830 		 *     i.e., it ACKs something we sent.
2831 		 */
2832 		if (tp->t_flags & TF_NEEDSYN) {
2833 			/*
2834 			 * T/TCP: Connection was half-synchronized, and our
2835 			 * SYN has been ACK'd (so connection is now fully
2836 			 * synchronized).  Go to non-starred state,
2837 			 * increment snd_una for ACK of SYN, and check if
2838 			 * we can do window scaling.
2839 			 */
2840 			tp->t_flags &= ~TF_NEEDSYN;
2841 			tp->snd_una++;
2842 			/* Do window scaling? */
2843 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2844 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2845 				tp->rcv_scale = tp->request_r_scale;
2846 				/* Send window already scaled. */
2847 			}
2848 		}
2849 
2850 process_ACK:
2851 		INP_WLOCK_ASSERT(inp);
2852 
2853 		/*
2854 		 * Adjust for the SYN bit in sequence space,
2855 		 * but don't account for it in cwnd calculations.
2856 		 * This is for the SYN_RECEIVED, non-simultaneous
2857 		 * SYN case. SYN_SENT and simultaneous SYN are
2858 		 * treated elsewhere.
2859 		 */
2860 		if (incforsyn)
2861 			tp->snd_una++;
2862 		acked = BYTES_THIS_ACK(tp, th);
2863 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2864 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2865 		    tp->snd_una, th->th_ack, tp, m));
2866 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2867 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2868 
2869 		/*
2870 		 * If we just performed our first retransmit, and the ACK
2871 		 * arrives within our recovery window, then it was a mistake
2872 		 * to do the retransmit in the first place.  Recover our
2873 		 * original cwnd and ssthresh, and proceed to transmit where
2874 		 * we left off.
2875 		 */
2876 		if (tp->t_rxtshift == 1 &&
2877 		    tp->t_flags & TF_PREVVALID &&
2878 		    tp->t_badrxtwin != 0 &&
2879 		    to.to_flags & TOF_TS &&
2880 		    to.to_tsecr != 0 &&
2881 		    TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2882 			cc_cong_signal(tp, th, CC_RTO_ERR);
2883 
2884 		/*
2885 		 * If we have a timestamp reply, update smoothed
2886 		 * round trip time.  If no timestamp is present but
2887 		 * transmit timer is running and timed sequence
2888 		 * number was acked, update smoothed round trip time.
2889 		 * Since we now have an rtt measurement, cancel the
2890 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2891 		 * Recompute the initial retransmit timer.
2892 		 *
2893 		 * Some boxes send broken timestamp replies
2894 		 * during the SYN+ACK phase, ignore
2895 		 * timestamps of 0 or we could calculate a
2896 		 * huge RTT and blow up the retransmit timer.
2897 		 */
2898 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2899 			uint32_t t;
2900 
2901 			t = tcp_ts_getticks() - to.to_tsecr;
2902 			if (!tp->t_rttlow || tp->t_rttlow > t)
2903 				tp->t_rttlow = t;
2904 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2905 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2906 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2907 				tp->t_rttlow = ticks - tp->t_rtttime;
2908 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2909 		}
2910 
2911 		SOCKBUF_LOCK(&so->so_snd);
2912 		/*
2913 		 * Clear t_acktime if remote side has ACKd all data in the
2914 		 * socket buffer and FIN (if applicable).
2915 		 * Otherwise, update t_acktime if we received a sufficiently
2916 		 * large ACK.
2917 		 */
2918 		if ((tp->t_state <= TCPS_CLOSE_WAIT &&
2919 		    acked == sbavail(&so->so_snd)) ||
2920 		    acked > sbavail(&so->so_snd))
2921 			tp->t_acktime = 0;
2922 		else if (acked > 1)
2923 			tp->t_acktime = ticks;
2924 
2925 		/*
2926 		 * If all outstanding data is acked, stop retransmit
2927 		 * timer and remember to restart (more output or persist).
2928 		 * If there is more data to be acked, restart retransmit
2929 		 * timer, using current (possibly backed-off) value.
2930 		 */
2931 		if (th->th_ack == tp->snd_max) {
2932 			tcp_timer_activate(tp, TT_REXMT, 0);
2933 			needoutput = 1;
2934 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2935 			tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
2936 
2937 		/*
2938 		 * If no data (only SYN) was ACK'd,
2939 		 *    skip rest of ACK processing.
2940 		 */
2941 		if (acked == 0) {
2942 			SOCKBUF_UNLOCK(&so->so_snd);
2943 			goto step6;
2944 		}
2945 
2946 		/*
2947 		 * Let the congestion control algorithm update congestion
2948 		 * control related information. This typically means increasing
2949 		 * the congestion window.
2950 		 */
2951 		cc_ack_received(tp, th, nsegs, CC_ACK);
2952 
2953 		if (acked > sbavail(&so->so_snd)) {
2954 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2955 				tp->snd_wnd -= sbavail(&so->so_snd);
2956 			else
2957 				tp->snd_wnd = 0;
2958 			mfree = sbcut_locked(&so->so_snd,
2959 			    (int)sbavail(&so->so_snd));
2960 			ourfinisacked = 1;
2961 		} else {
2962 			mfree = sbcut_locked(&so->so_snd, acked);
2963 			if (tp->snd_wnd >= (uint32_t) acked)
2964 				tp->snd_wnd -= acked;
2965 			else
2966 				tp->snd_wnd = 0;
2967 			ourfinisacked = 0;
2968 		}
2969 		/* NB: sowwakeup_locked() does an implicit unlock. */
2970 		sowwakeup_locked(so);
2971 		m_freem(mfree);
2972 		/* Detect una wraparound. */
2973 		if (!IN_RECOVERY(tp->t_flags) &&
2974 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2975 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2976 			tp->snd_recover = th->th_ack - 1;
2977 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2978 		if (IN_RECOVERY(tp->t_flags) &&
2979 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2980 			EXIT_RECOVERY(tp->t_flags);
2981 		}
2982 		tp->snd_una = th->th_ack;
2983 		if (tp->t_flags & TF_SACK_PERMIT) {
2984 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2985 				tp->snd_recover = tp->snd_una;
2986 		}
2987 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2988 			tp->snd_nxt = tp->snd_una;
2989 
2990 		switch (tp->t_state) {
2991 		/*
2992 		 * In FIN_WAIT_1 STATE in addition to the processing
2993 		 * for the ESTABLISHED state if our FIN is now acknowledged
2994 		 * then enter FIN_WAIT_2.
2995 		 */
2996 		case TCPS_FIN_WAIT_1:
2997 			if (ourfinisacked) {
2998 				/*
2999 				 * If we can't receive any more
3000 				 * data, then closing user can proceed.
3001 				 * Starting the timer is contrary to the
3002 				 * specification, but if we don't get a FIN
3003 				 * we'll hang forever.
3004 				 */
3005 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3006 					soisdisconnected(so);
3007 					tcp_timer_activate(tp, TT_2MSL,
3008 					    (tcp_fast_finwait2_recycle ?
3009 					    tcp_finwait2_timeout :
3010 					    TP_MAXIDLE(tp)));
3011 				}
3012 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
3013 			}
3014 			break;
3015 
3016 		/*
3017 		 * In CLOSING STATE in addition to the processing for
3018 		 * the ESTABLISHED state if the ACK acknowledges our FIN
3019 		 * then enter the TIME-WAIT state, otherwise ignore
3020 		 * the segment.
3021 		 */
3022 		case TCPS_CLOSING:
3023 			if (ourfinisacked) {
3024 				tcp_twstart(tp);
3025 				m_freem(m);
3026 				return;
3027 			}
3028 			break;
3029 
3030 		/*
3031 		 * In LAST_ACK, we may still be waiting for data to drain
3032 		 * and/or to be acked, as well as for the ack of our FIN.
3033 		 * If our FIN is now acknowledged, delete the TCB,
3034 		 * enter the closed state and return.
3035 		 */
3036 		case TCPS_LAST_ACK:
3037 			if (ourfinisacked) {
3038 				tp = tcp_close(tp);
3039 				goto drop;
3040 			}
3041 			break;
3042 		}
3043 	}
3044 
3045 step6:
3046 	INP_WLOCK_ASSERT(inp);
3047 
3048 	/*
3049 	 * Update window information.
3050 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3051 	 */
3052 	if ((thflags & TH_ACK) &&
3053 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3054 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3055 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3056 		/* keep track of pure window updates */
3057 		if (tlen == 0 &&
3058 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3059 			TCPSTAT_INC(tcps_rcvwinupd);
3060 		tp->snd_wnd = tiwin;
3061 		tp->snd_wl1 = th->th_seq;
3062 		tp->snd_wl2 = th->th_ack;
3063 		if (tp->snd_wnd > tp->max_sndwnd)
3064 			tp->max_sndwnd = tp->snd_wnd;
3065 		needoutput = 1;
3066 	}
3067 
3068 	/*
3069 	 * Process segments with URG.
3070 	 */
3071 	if ((thflags & TH_URG) && th->th_urp &&
3072 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3073 		/*
3074 		 * This is a kludge, but if we receive and accept
3075 		 * random urgent pointers, we'll crash in
3076 		 * soreceive.  It's hard to imagine someone
3077 		 * actually wanting to send this much urgent data.
3078 		 */
3079 		SOCKBUF_LOCK(&so->so_rcv);
3080 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3081 			th->th_urp = 0;			/* XXX */
3082 			thflags &= ~TH_URG;		/* XXX */
3083 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3084 			goto dodata;			/* XXX */
3085 		}
3086 		/*
3087 		 * If this segment advances the known urgent pointer,
3088 		 * then mark the data stream.  This should not happen
3089 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3090 		 * a FIN has been received from the remote side.
3091 		 * In these states we ignore the URG.
3092 		 *
3093 		 * According to RFC961 (Assigned Protocols),
3094 		 * the urgent pointer points to the last octet
3095 		 * of urgent data.  We continue, however,
3096 		 * to consider it to indicate the first octet
3097 		 * of data past the urgent section as the original
3098 		 * spec states (in one of two places).
3099 		 */
3100 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3101 			tp->rcv_up = th->th_seq + th->th_urp;
3102 			so->so_oobmark = sbavail(&so->so_rcv) +
3103 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3104 			if (so->so_oobmark == 0)
3105 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3106 			sohasoutofband(so);
3107 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3108 		}
3109 		SOCKBUF_UNLOCK(&so->so_rcv);
3110 		/*
3111 		 * Remove out of band data so doesn't get presented to user.
3112 		 * This can happen independent of advancing the URG pointer,
3113 		 * but if two URG's are pending at once, some out-of-band
3114 		 * data may creep in... ick.
3115 		 */
3116 		if (th->th_urp <= (uint32_t)tlen &&
3117 		    !(so->so_options & SO_OOBINLINE)) {
3118 			/* hdr drop is delayed */
3119 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3120 		}
3121 	} else {
3122 		/*
3123 		 * If no out of band data is expected,
3124 		 * pull receive urgent pointer along
3125 		 * with the receive window.
3126 		 */
3127 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3128 			tp->rcv_up = tp->rcv_nxt;
3129 	}
3130 dodata:							/* XXX */
3131 	INP_WLOCK_ASSERT(inp);
3132 
3133 	/*
3134 	 * Process the segment text, merging it into the TCP sequencing queue,
3135 	 * and arranging for acknowledgment of receipt if necessary.
3136 	 * This process logically involves adjusting tp->rcv_wnd as data
3137 	 * is presented to the user (this happens in tcp_usrreq.c,
3138 	 * case PRU_RCVD).  If a FIN has already been received on this
3139 	 * connection then we just ignore the text.
3140 	 */
3141 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3142 		   IS_FASTOPEN(tp->t_flags));
3143 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3144 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3145 		tcp_seq save_start = th->th_seq;
3146 		tcp_seq save_rnxt  = tp->rcv_nxt;
3147 		int     save_tlen  = tlen;
3148 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3149 		/*
3150 		 * Insert segment which includes th into TCP reassembly queue
3151 		 * with control block tp.  Set thflags to whether reassembly now
3152 		 * includes a segment with FIN.  This handles the common case
3153 		 * inline (segment is the next to be received on an established
3154 		 * connection, and the queue is empty), avoiding linkage into
3155 		 * and removal from the queue and repetition of various
3156 		 * conversions.
3157 		 * Set DELACK for segments received in order, but ack
3158 		 * immediately when segments are out of order (so
3159 		 * fast retransmit can work).
3160 		 */
3161 		if (th->th_seq == tp->rcv_nxt &&
3162 		    SEGQ_EMPTY(tp) &&
3163 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3164 		     tfo_syn)) {
3165 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3166 				tp->t_flags |= TF_DELACK;
3167 			else
3168 				tp->t_flags |= TF_ACKNOW;
3169 			tp->rcv_nxt += tlen;
3170 			if (tlen &&
3171 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3172 			    (tp->t_fbyte_in == 0)) {
3173 				tp->t_fbyte_in = ticks;
3174 				if (tp->t_fbyte_in == 0)
3175 					tp->t_fbyte_in = 1;
3176 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3177 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3178 			}
3179 			thflags = tcp_get_flags(th) & TH_FIN;
3180 			TCPSTAT_INC(tcps_rcvpack);
3181 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3182 			SOCKBUF_LOCK(&so->so_rcv);
3183 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3184 				m_freem(m);
3185 			else
3186 				sbappendstream_locked(&so->so_rcv, m, 0);
3187 			tp->t_flags |= TF_WAKESOR;
3188 		} else {
3189 			/*
3190 			 * XXX: Due to the header drop above "th" is
3191 			 * theoretically invalid by now.  Fortunately
3192 			 * m_adj() doesn't actually frees any mbufs
3193 			 * when trimming from the head.
3194 			 */
3195 			tcp_seq temp = save_start;
3196 
3197 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3198 			tp->t_flags |= TF_ACKNOW;
3199 		}
3200 		if ((tp->t_flags & TF_SACK_PERMIT) &&
3201 		    (save_tlen > 0) &&
3202 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
3203 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3204 				/*
3205 				 * DSACK actually handled in the fastpath
3206 				 * above.
3207 				 */
3208 				tcp_update_sack_list(tp, save_start,
3209 				    save_start + save_tlen);
3210 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3211 				if ((tp->rcv_numsacks >= 1) &&
3212 				    (tp->sackblks[0].end == save_start)) {
3213 					/*
3214 					 * Partial overlap, recorded at todrop
3215 					 * above.
3216 					 */
3217 					tcp_update_sack_list(tp,
3218 					    tp->sackblks[0].start,
3219 					    tp->sackblks[0].end);
3220 				} else {
3221 					tcp_update_dsack_list(tp, save_start,
3222 					    save_start + save_tlen);
3223 				}
3224 			} else if (tlen >= save_tlen) {
3225 				/* Update of sackblks. */
3226 				tcp_update_dsack_list(tp, save_start,
3227 				    save_start + save_tlen);
3228 			} else if (tlen > 0) {
3229 				tcp_update_dsack_list(tp, save_start,
3230 				    save_start + tlen);
3231 			}
3232 		}
3233 		tcp_handle_wakeup(tp);
3234 #if 0
3235 		/*
3236 		 * Note the amount of data that peer has sent into
3237 		 * our window, in order to estimate the sender's
3238 		 * buffer size.
3239 		 * XXX: Unused.
3240 		 */
3241 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3242 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3243 		else
3244 			len = so->so_rcv.sb_hiwat;
3245 #endif
3246 	} else {
3247 		m_freem(m);
3248 		thflags &= ~TH_FIN;
3249 	}
3250 
3251 	/*
3252 	 * If FIN is received ACK the FIN and let the user know
3253 	 * that the connection is closing.
3254 	 */
3255 	if (thflags & TH_FIN) {
3256 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3257 			/* The socket upcall is handled by socantrcvmore. */
3258 			socantrcvmore(so);
3259 			/*
3260 			 * If connection is half-synchronized
3261 			 * (ie NEEDSYN flag on) then delay ACK,
3262 			 * so it may be piggybacked when SYN is sent.
3263 			 * Otherwise, since we received a FIN then no
3264 			 * more input can be expected, send ACK now.
3265 			 */
3266 			if (tp->t_flags & TF_NEEDSYN)
3267 				tp->t_flags |= TF_DELACK;
3268 			else
3269 				tp->t_flags |= TF_ACKNOW;
3270 			tp->rcv_nxt++;
3271 		}
3272 		switch (tp->t_state) {
3273 		/*
3274 		 * In SYN_RECEIVED and ESTABLISHED STATES
3275 		 * enter the CLOSE_WAIT state.
3276 		 */
3277 		case TCPS_SYN_RECEIVED:
3278 			tp->t_starttime = ticks;
3279 			/* FALLTHROUGH */
3280 		case TCPS_ESTABLISHED:
3281 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3282 			break;
3283 
3284 		/*
3285 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3286 		 * enter the CLOSING state.
3287 		 */
3288 		case TCPS_FIN_WAIT_1:
3289 			tcp_state_change(tp, TCPS_CLOSING);
3290 			break;
3291 
3292 		/*
3293 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3294 		 * starting the time-wait timer, turning off the other
3295 		 * standard timers.
3296 		 */
3297 		case TCPS_FIN_WAIT_2:
3298 			tcp_twstart(tp);
3299 			return;
3300 		}
3301 	}
3302 #ifdef TCPDEBUG
3303 	if (so->so_options & SO_DEBUG)
3304 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3305 			  &tcp_savetcp, 0);
3306 #endif
3307 	TCP_PROBE3(debug__input, tp, th, m);
3308 
3309 	/*
3310 	 * Return any desired output.
3311 	 */
3312 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3313 		(void) tcp_output(tp);
3314 
3315 check_delack:
3316 	INP_WLOCK_ASSERT(inp);
3317 
3318 	if (tp->t_flags & TF_DELACK) {
3319 		tp->t_flags &= ~TF_DELACK;
3320 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3321 	}
3322 	INP_WUNLOCK(inp);
3323 	return;
3324 
3325 dropafterack:
3326 	/*
3327 	 * Generate an ACK dropping incoming segment if it occupies
3328 	 * sequence space, where the ACK reflects our state.
3329 	 *
3330 	 * We can now skip the test for the RST flag since all
3331 	 * paths to this code happen after packets containing
3332 	 * RST have been dropped.
3333 	 *
3334 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3335 	 * segment we received passes the SYN-RECEIVED ACK test.
3336 	 * If it fails send a RST.  This breaks the loop in the
3337 	 * "LAND" DoS attack, and also prevents an ACK storm
3338 	 * between two listening ports that have been sent forged
3339 	 * SYN segments, each with the source address of the other.
3340 	 */
3341 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3342 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3343 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3344 		rstreason = BANDLIM_RST_OPENPORT;
3345 		tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
3346 		goto dropwithreset;
3347 	}
3348 #ifdef TCPDEBUG
3349 	if (so->so_options & SO_DEBUG)
3350 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3351 			  &tcp_savetcp, 0);
3352 #endif
3353 	TCP_PROBE3(debug__input, tp, th, m);
3354 	tp->t_flags |= TF_ACKNOW;
3355 	(void) tcp_output(tp);
3356 	INP_WUNLOCK(inp);
3357 	m_freem(m);
3358 	return;
3359 
3360 dropwithreset:
3361 	if (tp != NULL) {
3362 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3363 		INP_WUNLOCK(inp);
3364 	} else
3365 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3366 	return;
3367 
3368 drop:
3369 	/*
3370 	 * Drop space held by incoming segment and return.
3371 	 */
3372 #ifdef TCPDEBUG
3373 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
3374 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3375 			  &tcp_savetcp, 0);
3376 #endif
3377 	TCP_PROBE3(debug__input, tp, th, m);
3378 	if (tp != NULL) {
3379 		INP_WUNLOCK(inp);
3380 	}
3381 	m_freem(m);
3382 }
3383 
3384 /*
3385  * Issue RST and make ACK acceptable to originator of segment.
3386  * The mbuf must still include the original packet header.
3387  * tp may be NULL.
3388  */
3389 void
3390 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3391     int tlen, int rstreason)
3392 {
3393 #ifdef INET
3394 	struct ip *ip;
3395 #endif
3396 #ifdef INET6
3397 	struct ip6_hdr *ip6;
3398 #endif
3399 
3400 	if (tp != NULL) {
3401 		INP_LOCK_ASSERT(tptoinpcb(tp));
3402 	}
3403 
3404 	/* Don't bother if destination was broadcast/multicast. */
3405 	if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3406 		goto drop;
3407 #ifdef INET6
3408 	if (mtod(m, struct ip *)->ip_v == 6) {
3409 		ip6 = mtod(m, struct ip6_hdr *);
3410 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3411 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3412 			goto drop;
3413 		/* IPv6 anycast check is done at tcp6_input() */
3414 	}
3415 #endif
3416 #if defined(INET) && defined(INET6)
3417 	else
3418 #endif
3419 #ifdef INET
3420 	{
3421 		ip = mtod(m, struct ip *);
3422 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3423 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3424 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3425 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3426 			goto drop;
3427 	}
3428 #endif
3429 
3430 	/* Perform bandwidth limiting. */
3431 	if (badport_bandlim(rstreason) < 0)
3432 		goto drop;
3433 
3434 	/* tcp_respond consumes the mbuf chain. */
3435 	if (tcp_get_flags(th) & TH_ACK) {
3436 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3437 		    th->th_ack, TH_RST);
3438 	} else {
3439 		if (tcp_get_flags(th) & TH_SYN)
3440 			tlen++;
3441 		if (tcp_get_flags(th) & TH_FIN)
3442 			tlen++;
3443 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3444 		    (tcp_seq)0, TH_RST|TH_ACK);
3445 	}
3446 	return;
3447 drop:
3448 	m_freem(m);
3449 }
3450 
3451 /*
3452  * Parse TCP options and place in tcpopt.
3453  */
3454 void
3455 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3456 {
3457 	int opt, optlen;
3458 
3459 	to->to_flags = 0;
3460 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3461 		opt = cp[0];
3462 		if (opt == TCPOPT_EOL)
3463 			break;
3464 		if (opt == TCPOPT_NOP)
3465 			optlen = 1;
3466 		else {
3467 			if (cnt < 2)
3468 				break;
3469 			optlen = cp[1];
3470 			if (optlen < 2 || optlen > cnt)
3471 				break;
3472 		}
3473 		switch (opt) {
3474 		case TCPOPT_MAXSEG:
3475 			if (optlen != TCPOLEN_MAXSEG)
3476 				continue;
3477 			if (!(flags & TO_SYN))
3478 				continue;
3479 			to->to_flags |= TOF_MSS;
3480 			bcopy((char *)cp + 2,
3481 			    (char *)&to->to_mss, sizeof(to->to_mss));
3482 			to->to_mss = ntohs(to->to_mss);
3483 			break;
3484 		case TCPOPT_WINDOW:
3485 			if (optlen != TCPOLEN_WINDOW)
3486 				continue;
3487 			if (!(flags & TO_SYN))
3488 				continue;
3489 			to->to_flags |= TOF_SCALE;
3490 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3491 			break;
3492 		case TCPOPT_TIMESTAMP:
3493 			if (optlen != TCPOLEN_TIMESTAMP)
3494 				continue;
3495 			to->to_flags |= TOF_TS;
3496 			bcopy((char *)cp + 2,
3497 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3498 			to->to_tsval = ntohl(to->to_tsval);
3499 			bcopy((char *)cp + 6,
3500 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3501 			to->to_tsecr = ntohl(to->to_tsecr);
3502 			break;
3503 		case TCPOPT_SIGNATURE:
3504 			/*
3505 			 * In order to reply to a host which has set the
3506 			 * TCP_SIGNATURE option in its initial SYN, we have
3507 			 * to record the fact that the option was observed
3508 			 * here for the syncache code to perform the correct
3509 			 * response.
3510 			 */
3511 			if (optlen != TCPOLEN_SIGNATURE)
3512 				continue;
3513 			to->to_flags |= TOF_SIGNATURE;
3514 			to->to_signature = cp + 2;
3515 			break;
3516 		case TCPOPT_SACK_PERMITTED:
3517 			if (optlen != TCPOLEN_SACK_PERMITTED)
3518 				continue;
3519 			if (!(flags & TO_SYN))
3520 				continue;
3521 			if (!V_tcp_do_sack)
3522 				continue;
3523 			to->to_flags |= TOF_SACKPERM;
3524 			break;
3525 		case TCPOPT_SACK:
3526 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3527 				continue;
3528 			if (flags & TO_SYN)
3529 				continue;
3530 			to->to_flags |= TOF_SACK;
3531 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3532 			to->to_sacks = cp + 2;
3533 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3534 			break;
3535 		case TCPOPT_FAST_OPEN:
3536 			/*
3537 			 * Cookie length validation is performed by the
3538 			 * server side cookie checking code or the client
3539 			 * side cookie cache update code.
3540 			 */
3541 			if (!(flags & TO_SYN))
3542 				continue;
3543 			if (!V_tcp_fastopen_client_enable &&
3544 			    !V_tcp_fastopen_server_enable)
3545 				continue;
3546 			to->to_flags |= TOF_FASTOPEN;
3547 			to->to_tfo_len = optlen - 2;
3548 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3549 			break;
3550 		default:
3551 			continue;
3552 		}
3553 	}
3554 }
3555 
3556 /*
3557  * Pull out of band byte out of a segment so
3558  * it doesn't appear in the user's data queue.
3559  * It is still reflected in the segment length for
3560  * sequencing purposes.
3561  */
3562 void
3563 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3564     int off)
3565 {
3566 	int cnt = off + th->th_urp - 1;
3567 
3568 	while (cnt >= 0) {
3569 		if (m->m_len > cnt) {
3570 			char *cp = mtod(m, caddr_t) + cnt;
3571 			struct tcpcb *tp = sototcpcb(so);
3572 
3573 			INP_WLOCK_ASSERT(tptoinpcb(tp));
3574 
3575 			tp->t_iobc = *cp;
3576 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3577 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3578 			m->m_len--;
3579 			if (m->m_flags & M_PKTHDR)
3580 				m->m_pkthdr.len--;
3581 			return;
3582 		}
3583 		cnt -= m->m_len;
3584 		m = m->m_next;
3585 		if (m == NULL)
3586 			break;
3587 	}
3588 	panic("tcp_pulloutofband");
3589 }
3590 
3591 /*
3592  * Collect new round-trip time estimate
3593  * and update averages and current timeout.
3594  */
3595 void
3596 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3597 {
3598 	int delta;
3599 
3600 	INP_WLOCK_ASSERT(tptoinpcb(tp));
3601 
3602 	TCPSTAT_INC(tcps_rttupdated);
3603 	tp->t_rttupdated++;
3604 #ifdef STATS
3605 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3606 	    imax(0, rtt * 1000 / hz));
3607 #endif
3608 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3609 		/*
3610 		 * srtt is stored as fixed point with 5 bits after the
3611 		 * binary point (i.e., scaled by 8).  The following magic
3612 		 * is equivalent to the smoothing algorithm in rfc793 with
3613 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3614 		 * point).  Adjust rtt to origin 0.
3615 		 */
3616 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3617 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3618 
3619 		if ((tp->t_srtt += delta) <= 0)
3620 			tp->t_srtt = 1;
3621 
3622 		/*
3623 		 * We accumulate a smoothed rtt variance (actually, a
3624 		 * smoothed mean difference), then set the retransmit
3625 		 * timer to smoothed rtt + 4 times the smoothed variance.
3626 		 * rttvar is stored as fixed point with 4 bits after the
3627 		 * binary point (scaled by 16).  The following is
3628 		 * equivalent to rfc793 smoothing with an alpha of .75
3629 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3630 		 * rfc793's wired-in beta.
3631 		 */
3632 		if (delta < 0)
3633 			delta = -delta;
3634 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3635 		if ((tp->t_rttvar += delta) <= 0)
3636 			tp->t_rttvar = 1;
3637 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3638 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3639 	} else {
3640 		/*
3641 		 * No rtt measurement yet - use the unsmoothed rtt.
3642 		 * Set the variance to half the rtt (so our first
3643 		 * retransmit happens at 3*rtt).
3644 		 */
3645 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3646 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3647 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3648 	}
3649 	tp->t_rtttime = 0;
3650 	tp->t_rxtshift = 0;
3651 
3652 	/*
3653 	 * the retransmit should happen at rtt + 4 * rttvar.
3654 	 * Because of the way we do the smoothing, srtt and rttvar
3655 	 * will each average +1/2 tick of bias.  When we compute
3656 	 * the retransmit timer, we want 1/2 tick of rounding and
3657 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3658 	 * firing of the timer.  The bias will give us exactly the
3659 	 * 1.5 tick we need.  But, because the bias is
3660 	 * statistical, we have to test that we don't drop below
3661 	 * the minimum feasible timer (which is 2 ticks).
3662 	 */
3663 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3664 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3665 
3666 	/*
3667 	 * We received an ack for a packet that wasn't retransmitted;
3668 	 * it is probably safe to discard any error indications we've
3669 	 * received recently.  This isn't quite right, but close enough
3670 	 * for now (a route might have failed after we sent a segment,
3671 	 * and the return path might not be symmetrical).
3672 	 */
3673 	tp->t_softerror = 0;
3674 }
3675 
3676 /*
3677  * Determine a reasonable value for maxseg size.
3678  * If the route is known, check route for mtu.
3679  * If none, use an mss that can be handled on the outgoing interface
3680  * without forcing IP to fragment.  If no route is found, route has no mtu,
3681  * or the destination isn't local, use a default, hopefully conservative
3682  * size (usually 512 or the default IP max size, but no more than the mtu
3683  * of the interface), as we can't discover anything about intervening
3684  * gateways or networks.  We also initialize the congestion/slow start
3685  * window to be a single segment if the destination isn't local.
3686  * While looking at the routing entry, we also initialize other path-dependent
3687  * parameters from pre-set or cached values in the routing entry.
3688  *
3689  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3690  * IP options, e.g. IPSEC data, since length of this data may vary, and
3691  * thus it is calculated for every segment separately in tcp_output().
3692  *
3693  * NOTE that this routine is only called when we process an incoming
3694  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3695  * settings are handled in tcp_mssopt().
3696  */
3697 void
3698 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3699     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3700 {
3701 	int mss = 0;
3702 	uint32_t maxmtu = 0;
3703 	struct inpcb *inp = tptoinpcb(tp);
3704 	struct hc_metrics_lite metrics;
3705 #ifdef INET6
3706 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3707 	size_t min_protoh = isipv6 ?
3708 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3709 			    sizeof (struct tcpiphdr);
3710 #else
3711 	 size_t min_protoh = sizeof(struct tcpiphdr);
3712 #endif
3713 
3714 	INP_WLOCK_ASSERT(inp);
3715 
3716 	if (tp->t_port)
3717 		min_protoh += V_tcp_udp_tunneling_overhead;
3718 	if (mtuoffer != -1) {
3719 		KASSERT(offer == -1, ("%s: conflict", __func__));
3720 		offer = mtuoffer - min_protoh;
3721 	}
3722 
3723 	/* Initialize. */
3724 #ifdef INET6
3725 	if (isipv6) {
3726 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3727 		tp->t_maxseg = V_tcp_v6mssdflt;
3728 	}
3729 #endif
3730 #if defined(INET) && defined(INET6)
3731 	else
3732 #endif
3733 #ifdef INET
3734 	{
3735 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3736 		tp->t_maxseg = V_tcp_mssdflt;
3737 	}
3738 #endif
3739 
3740 	/*
3741 	 * No route to sender, stay with default mss and return.
3742 	 */
3743 	if (maxmtu == 0) {
3744 		/*
3745 		 * In case we return early we need to initialize metrics
3746 		 * to a defined state as tcp_hc_get() would do for us
3747 		 * if there was no cache hit.
3748 		 */
3749 		if (metricptr != NULL)
3750 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3751 		return;
3752 	}
3753 
3754 	/* What have we got? */
3755 	switch (offer) {
3756 		case 0:
3757 			/*
3758 			 * Offer == 0 means that there was no MSS on the SYN
3759 			 * segment, in this case we use tcp_mssdflt as
3760 			 * already assigned to t_maxseg above.
3761 			 */
3762 			offer = tp->t_maxseg;
3763 			break;
3764 
3765 		case -1:
3766 			/*
3767 			 * Offer == -1 means that we didn't receive SYN yet.
3768 			 */
3769 			/* FALLTHROUGH */
3770 
3771 		default:
3772 			/*
3773 			 * Prevent DoS attack with too small MSS. Round up
3774 			 * to at least minmss.
3775 			 */
3776 			offer = max(offer, V_tcp_minmss);
3777 	}
3778 
3779 	/*
3780 	 * rmx information is now retrieved from tcp_hostcache.
3781 	 */
3782 	tcp_hc_get(&inp->inp_inc, &metrics);
3783 	if (metricptr != NULL)
3784 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3785 
3786 	/*
3787 	 * If there's a discovered mtu in tcp hostcache, use it.
3788 	 * Else, use the link mtu.
3789 	 */
3790 	if (metrics.rmx_mtu)
3791 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3792 	else {
3793 #ifdef INET6
3794 		if (isipv6) {
3795 			mss = maxmtu - min_protoh;
3796 			if (!V_path_mtu_discovery &&
3797 			    !in6_localaddr(&inp->in6p_faddr))
3798 				mss = min(mss, V_tcp_v6mssdflt);
3799 		}
3800 #endif
3801 #if defined(INET) && defined(INET6)
3802 		else
3803 #endif
3804 #ifdef INET
3805 		{
3806 			mss = maxmtu - min_protoh;
3807 			if (!V_path_mtu_discovery &&
3808 			    !in_localaddr(inp->inp_faddr))
3809 				mss = min(mss, V_tcp_mssdflt);
3810 		}
3811 #endif
3812 		/*
3813 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3814 		 * probably violates the TCP spec.
3815 		 * The problem is that, since we don't know the
3816 		 * other end's MSS, we are supposed to use a conservative
3817 		 * default.  But, if we do that, then MTU discovery will
3818 		 * never actually take place, because the conservative
3819 		 * default is much less than the MTUs typically seen
3820 		 * on the Internet today.  For the moment, we'll sweep
3821 		 * this under the carpet.
3822 		 *
3823 		 * The conservative default might not actually be a problem
3824 		 * if the only case this occurs is when sending an initial
3825 		 * SYN with options and data to a host we've never talked
3826 		 * to before.  Then, they will reply with an MSS value which
3827 		 * will get recorded and the new parameters should get
3828 		 * recomputed.  For Further Study.
3829 		 */
3830 	}
3831 	mss = min(mss, offer);
3832 
3833 	/*
3834 	 * Sanity check: make sure that maxseg will be large
3835 	 * enough to allow some data on segments even if the
3836 	 * all the option space is used (40bytes).  Otherwise
3837 	 * funny things may happen in tcp_output.
3838 	 *
3839 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3840 	 */
3841 	mss = max(mss, 64);
3842 
3843 	tp->t_maxseg = mss;
3844 }
3845 
3846 void
3847 tcp_mss(struct tcpcb *tp, int offer)
3848 {
3849 	int mss;
3850 	uint32_t bufsize;
3851 	struct inpcb *inp = tptoinpcb(tp);
3852 	struct socket *so;
3853 	struct hc_metrics_lite metrics;
3854 	struct tcp_ifcap cap;
3855 
3856 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3857 
3858 	bzero(&cap, sizeof(cap));
3859 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3860 
3861 	mss = tp->t_maxseg;
3862 
3863 	/*
3864 	 * If there's a pipesize, change the socket buffer to that size,
3865 	 * don't change if sb_hiwat is different than default (then it
3866 	 * has been changed on purpose with setsockopt).
3867 	 * Make the socket buffers an integral number of mss units;
3868 	 * if the mss is larger than the socket buffer, decrease the mss.
3869 	 */
3870 	so = inp->inp_socket;
3871 	SOCKBUF_LOCK(&so->so_snd);
3872 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3873 		bufsize = metrics.rmx_sendpipe;
3874 	else
3875 		bufsize = so->so_snd.sb_hiwat;
3876 	if (bufsize < mss)
3877 		mss = bufsize;
3878 	else {
3879 		bufsize = roundup(bufsize, mss);
3880 		if (bufsize > sb_max)
3881 			bufsize = sb_max;
3882 		if (bufsize > so->so_snd.sb_hiwat)
3883 			(void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3884 	}
3885 	SOCKBUF_UNLOCK(&so->so_snd);
3886 	/*
3887 	 * Sanity check: make sure that maxseg will be large
3888 	 * enough to allow some data on segments even if the
3889 	 * all the option space is used (40bytes).  Otherwise
3890 	 * funny things may happen in tcp_output.
3891 	 *
3892 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3893 	 */
3894 	tp->t_maxseg = max(mss, 64);
3895 
3896 	SOCKBUF_LOCK(&so->so_rcv);
3897 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3898 		bufsize = metrics.rmx_recvpipe;
3899 	else
3900 		bufsize = so->so_rcv.sb_hiwat;
3901 	if (bufsize > mss) {
3902 		bufsize = roundup(bufsize, mss);
3903 		if (bufsize > sb_max)
3904 			bufsize = sb_max;
3905 		if (bufsize > so->so_rcv.sb_hiwat)
3906 			(void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
3907 	}
3908 	SOCKBUF_UNLOCK(&so->so_rcv);
3909 
3910 	/* Check the interface for TSO capabilities. */
3911 	if (cap.ifcap & CSUM_TSO) {
3912 		tp->t_flags |= TF_TSO;
3913 		tp->t_tsomax = cap.tsomax;
3914 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3915 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3916 	}
3917 }
3918 
3919 /*
3920  * Determine the MSS option to send on an outgoing SYN.
3921  */
3922 int
3923 tcp_mssopt(struct in_conninfo *inc)
3924 {
3925 	int mss = 0;
3926 	uint32_t thcmtu = 0;
3927 	uint32_t maxmtu = 0;
3928 	size_t min_protoh;
3929 
3930 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3931 
3932 #ifdef INET6
3933 	if (inc->inc_flags & INC_ISIPV6) {
3934 		mss = V_tcp_v6mssdflt;
3935 		maxmtu = tcp_maxmtu6(inc, NULL);
3936 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3937 	}
3938 #endif
3939 #if defined(INET) && defined(INET6)
3940 	else
3941 #endif
3942 #ifdef INET
3943 	{
3944 		mss = V_tcp_mssdflt;
3945 		maxmtu = tcp_maxmtu(inc, NULL);
3946 		min_protoh = sizeof(struct tcpiphdr);
3947 	}
3948 #endif
3949 #if defined(INET6) || defined(INET)
3950 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3951 #endif
3952 
3953 	if (maxmtu && thcmtu)
3954 		mss = min(maxmtu, thcmtu) - min_protoh;
3955 	else if (maxmtu || thcmtu)
3956 		mss = max(maxmtu, thcmtu) - min_protoh;
3957 
3958 	return (mss);
3959 }
3960 
3961 void
3962 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3963 {
3964 	int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3965 	int maxseg = tcp_maxseg(tp);
3966 
3967 	INP_WLOCK_ASSERT(tptoinpcb(tp));
3968 
3969 	/*
3970 	 * Compute the amount of data that this ACK is indicating
3971 	 * (del_data) and an estimate of how many bytes are in the
3972 	 * network.
3973 	 */
3974 	if (tcp_is_sack_recovery(tp, to) ||
3975 	    (IN_CONGRECOVERY(tp->t_flags) &&
3976 	     !IN_FASTRECOVERY(tp->t_flags))) {
3977 		del_data = tp->sackhint.delivered_data;
3978 		if (V_tcp_do_newsack)
3979 			pipe = tcp_compute_pipe(tp);
3980 		else
3981 			pipe = (tp->snd_nxt - tp->snd_fack) +
3982 				tp->sackhint.sack_bytes_rexmit;
3983 	} else {
3984 		if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3985 					     tp->snd_recover - tp->snd_una))
3986 			del_data = maxseg;
3987 		pipe = imax(0, tp->snd_max - tp->snd_una -
3988 			    imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3989 	}
3990 	tp->sackhint.prr_delivered += del_data;
3991 	/*
3992 	 * Proportional Rate Reduction
3993 	 */
3994 	if (pipe >= tp->snd_ssthresh) {
3995 		if (tp->sackhint.recover_fs == 0)
3996 			tp->sackhint.recover_fs =
3997 			    imax(1, tp->snd_nxt - tp->snd_una);
3998 		snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3999 			    tp->snd_ssthresh, tp->sackhint.recover_fs) -
4000 			    tp->sackhint.prr_out;
4001 	} else {
4002 		if (V_tcp_do_prr_conservative || (del_data == 0))
4003 			limit = tp->sackhint.prr_delivered -
4004 				tp->sackhint.prr_out;
4005 		else
4006 			limit = imax(tp->sackhint.prr_delivered -
4007 				    tp->sackhint.prr_out, del_data) +
4008 				    maxseg;
4009 		snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4010 	}
4011 	snd_cnt = imax(snd_cnt, 0) / maxseg;
4012 	/*
4013 	 * Send snd_cnt new data into the network in response to this ack.
4014 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
4015 	 * accordingly.
4016 	 */
4017 	if (IN_FASTRECOVERY(tp->t_flags)) {
4018 		if (tcp_is_sack_recovery(tp, to)) {
4019 			tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4020 					    tp->sackhint.sack_bytes_rexmit +
4021 					    (snd_cnt * maxseg);
4022 		} else {
4023 			tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4024 					    (snd_cnt * maxseg);
4025 		}
4026 	} else if (IN_CONGRECOVERY(tp->t_flags))
4027 		tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4028 	tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4029 }
4030 
4031 /*
4032  * On a partial ack arrives, force the retransmission of the
4033  * next unacknowledged segment.  Do not clear tp->t_dupacks.
4034  * By setting snd_nxt to ti_ack, this forces retransmission timer to
4035  * be started again.
4036  */
4037 void
4038 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4039 {
4040 	tcp_seq onxt = tp->snd_nxt;
4041 	uint32_t ocwnd = tp->snd_cwnd;
4042 	u_int maxseg = tcp_maxseg(tp);
4043 
4044 	INP_WLOCK_ASSERT(tptoinpcb(tp));
4045 
4046 	tcp_timer_activate(tp, TT_REXMT, 0);
4047 	tp->t_rtttime = 0;
4048 	tp->snd_nxt = th->th_ack;
4049 	/*
4050 	 * Set snd_cwnd to one segment beyond acknowledged offset.
4051 	 * (tp->snd_una has not yet been updated when this function is called.)
4052 	 */
4053 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4054 	tp->t_flags |= TF_ACKNOW;
4055 	(void) tcp_output(tp);
4056 	tp->snd_cwnd = ocwnd;
4057 	if (SEQ_GT(onxt, tp->snd_nxt))
4058 		tp->snd_nxt = onxt;
4059 	/*
4060 	 * Partial window deflation.  Relies on fact that tp->snd_una
4061 	 * not updated yet.
4062 	 */
4063 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4064 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4065 	else
4066 		tp->snd_cwnd = 0;
4067 	tp->snd_cwnd += maxseg;
4068 }
4069 
4070 int
4071 tcp_compute_pipe(struct tcpcb *tp)
4072 {
4073 	if (tp->t_fb->tfb_compute_pipe == NULL) {
4074 		return (tp->snd_max - tp->snd_una +
4075 			tp->sackhint.sack_bytes_rexmit -
4076 			tp->sackhint.sacked_bytes);
4077 	} else {
4078 		return((*tp->t_fb->tfb_compute_pipe)(tp));
4079 	}
4080 }
4081 
4082 uint32_t
4083 tcp_compute_initwnd(uint32_t maxseg)
4084 {
4085 	/*
4086 	 * Calculate the Initial Window, also used as Restart Window
4087 	 *
4088 	 * RFC5681 Section 3.1 specifies the default conservative values.
4089 	 * RFC3390 specifies slightly more aggressive values.
4090 	 * RFC6928 increases it to ten segments.
4091 	 * Support for user specified value for initial flight size.
4092 	 */
4093 	if (V_tcp_initcwnd_segments)
4094 		return min(V_tcp_initcwnd_segments * maxseg,
4095 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4096 	else if (V_tcp_do_rfc3390)
4097 		return min(4 * maxseg, max(2 * maxseg, 4380));
4098 	else {
4099 		/* Per RFC5681 Section 3.1 */
4100 		if (maxseg > 2190)
4101 			return (2 * maxseg);
4102 		else if (maxseg > 1095)
4103 			return (3 * maxseg);
4104 		else
4105 			return (4 * maxseg);
4106 	}
4107 }
4108