tcp: RFC3517bis is now officially RFC6675
[dragonfly.git] / sys / netinet / tcp_input.c
1 /*
2  * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2002, 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36  *      The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *      This product includes software developed by the University of
49  *      California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *      @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67  * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68  */
69
70 #include "opt_inet.h"
71 #include "opt_inet6.h"
72 #include "opt_ipsec.h"
73 #include "opt_tcpdebug.h"
74 #include "opt_tcp_input.h"
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 #include <sys/malloc.h>
81 #include <sys/mbuf.h>
82 #include <sys/proc.h>           /* for proc0 declaration */
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/syslog.h>
87 #include <sys/in_cksum.h>
88
89 #include <sys/socketvar2.h>
90
91 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
92 #include <machine/stdarg.h>
93
94 #include <net/if.h>
95 #include <net/route.h>
96
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip_icmp.h>    /* for ICMP_BANDLIM */
101 #include <netinet/in_var.h>
102 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
103 #include <netinet/in_pcb.h>
104 #include <netinet/ip_var.h>
105 #include <netinet/ip6.h>
106 #include <netinet/icmp6.h>
107 #include <netinet6/nd6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/in6_pcb.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
117 #include <netinet/tcpip.h>
118
119 #ifdef TCPDEBUG
120 #include <netinet/tcp_debug.h>
121
122 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
123 struct tcphdr tcp_savetcp;
124 #endif
125
126 #ifdef FAST_IPSEC
127 #include <netproto/ipsec/ipsec.h>
128 #include <netproto/ipsec/ipsec6.h>
129 #endif
130
131 #ifdef IPSEC
132 #include <netinet6/ipsec.h>
133 #include <netinet6/ipsec6.h>
134 #include <netproto/key/key.h>
135 #endif
136
137 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
138
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141     &log_in_vain, 0, "Log all incoming TCP connections");
142
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145     &blackhole, 0, "Do not send RST when dropping refused connections");
146
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149     &tcp_delack_enabled, 0,
150     "Delay ACK to try and piggyback it onto a data packet");
151
152 #ifdef TCP_DROP_SYNFIN
153 static int drop_synfin = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
156 #endif
157
158 static int tcp_do_limitedtransmit = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160     &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
161
162 static int tcp_do_early_retransmit = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164     &tcp_do_early_retransmit, 0, "Early retransmit");
165
166 int tcp_aggregate_acks = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168     &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
169
170 static int tcp_do_eifel_detect = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
172     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
173
174 static int tcp_do_abc = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
176     &tcp_do_abc, 0,
177     "TCP Appropriate Byte Counting (RFC 3465)");
178
179 /*
180  * The following value actually takes range [25ms, 250ms],
181  * given that most modern systems use 1ms ~ 10ms as the unit
182  * of timestamp option.
183  */
184 static u_int tcp_paws_tolerance = 25;
185 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, paws_tolerance, CTLFLAG_RW,
186     &tcp_paws_tolerance, 0, "RFC1323 PAWS tolerance");
187
188 /*
189  * Define as tunable for easy testing with SACK on and off.
190  * Warning:  do not change setting in the middle of an existing active TCP flow,
191  *   else strange things might happen to that flow.
192  */
193 int tcp_do_sack = 1;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
195     &tcp_do_sack, 0, "Enable SACK Algorithms");
196
197 int tcp_do_smartsack = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
199     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
200
201 int tcp_do_rescuesack = 1;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW,
203     &tcp_do_rescuesack, 0, "Rescue retransmission for SACK");
204
205 int tcp_aggressive_rescuesack = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW,
207     &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK");
208
209 static int tcp_force_sackrxt = 1;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, force_sackrxt, CTLFLAG_RW,
211     &tcp_force_sackrxt, 0, "Allowed forced SACK retransmit burst");
212
213 int tcp_do_rfc6675 = 1;
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675, CTLFLAG_RW,
215     &tcp_do_rfc6675, 0, "Enable RFC6675");
216
217 int tcp_rfc6675_rxt = 0;
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_rxt, CTLFLAG_RW,
219     &tcp_rfc6675_rxt, 0, "Enable RFC6675 retransmit");
220
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
222     "TCP Segment Reassembly Queue");
223
224 int tcp_reass_maxseg = 0;
225 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
226     &tcp_reass_maxseg, 0,
227     "Global maximum number of TCP Segments in Reassembly Queue");
228
229 int tcp_reass_qsize = 0;
230 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
231     &tcp_reass_qsize, 0,
232     "Global number of TCP Segments currently in Reassembly Queue");
233
234 static int tcp_reass_overflows = 0;
235 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
236     &tcp_reass_overflows, 0,
237     "Global number of TCP Segment Reassembly Queue Overflows");
238
239 int tcp_do_autorcvbuf = 1;
240 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
241     &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
242
243 int tcp_autorcvbuf_inc = 16*1024;
244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
245     &tcp_autorcvbuf_inc, 0,
246     "Incrementor step size of automatic receive buffer");
247
248 int tcp_autorcvbuf_max = 2*1024*1024;
249 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
250     &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
251
252 int tcp_sosend_agglim = 3;
253 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW,
254     &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit");
255
256 int tcp_sosend_async = 1;
257 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW,
258     &tcp_sosend_async, 0, "TCP asynchronized pru_send");
259
260 static int tcp_ignore_redun_dsack = 1;
261 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW,
262     &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK");
263
264 static void      tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t,
265                     tcp_seq);
266 static void      tcp_pulloutofband(struct socket *,
267                      struct tcphdr *, struct mbuf *, int);
268 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
269                      struct mbuf *);
270 static void      tcp_xmit_timer(struct tcpcb *, int, tcp_seq);
271 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
272 static void      tcp_sack_rexmt(struct tcpcb *, boolean_t);
273 static boolean_t tcp_sack_limitedxmit(struct tcpcb *);
274 static int       tcp_rmx_msl(const struct tcpcb *);
275 static void      tcp_established(struct tcpcb *);
276 static boolean_t tcp_recv_dupack(struct tcpcb *, tcp_seq, u_int);
277
278 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
279 #ifdef INET6
280 #define ND6_HINT(tp) \
281 do { \
282         if ((tp) && (tp)->t_inpcb && \
283             ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
284             (tp)->t_inpcb->in6p_route.ro_rt) \
285                 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
286 } while (0)
287 #else
288 #define ND6_HINT(tp)
289 #endif
290
291 /*
292  * Indicate whether this ack should be delayed.  We can delay the ack if
293  *      - delayed acks are enabled and
294  *      - there is no delayed ack timer in progress and
295  *      - our last ack wasn't a 0-sized window.  We never want to delay
296  *        the ack that opens up a 0-sized window.
297  */
298 #define DELAY_ACK(tp) \
299         (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
300         !(tp->t_flags & TF_RXWIN0SENT))
301
302 #define acceptable_window_update(tp, th, tiwin)                         \
303     (SEQ_LT(tp->snd_wl1, th->th_seq) ||                                 \
304      (tp->snd_wl1 == th->th_seq &&                                      \
305       (SEQ_LT(tp->snd_wl2, th->th_ack) ||                               \
306        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
307
308 #define iceildiv(n, d)          (((n)+(d)-1) / (d))
309 #define need_early_retransmit(tp, ownd) \
310     (tcp_do_early_retransmit && \
311      (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \
312      ownd < ((tp->t_rxtthresh + 1) * tp->t_maxseg) && \
313      tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \
314      (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \
315       tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg)))
316
317 /*
318  * Returns TRUE, if this segment can be merged with the last
319  * pending segment in the reassemble queue and this segment
320  * does not overlap with the pending segment immediately
321  * preceeding the last pending segment.
322  */
323 static __inline boolean_t
324 tcp_paws_canreasslast(const struct tcpcb *tp, const struct tcphdr *th, int tlen)
325 {
326         const struct tseg_qent *last, *prev;
327
328         last = TAILQ_LAST(&tp->t_segq, tsegqe_head);
329         if (last == NULL)
330                 return FALSE;
331
332         /* This segment comes immediately after the last pending segment */
333         if (last->tqe_th->th_seq + last->tqe_len == th->th_seq) {
334                 if (last->tqe_th->th_flags & TH_FIN) {
335                         /* No segments should follow segment w/ FIN */
336                         return FALSE;
337                 }
338                 return TRUE;
339         }
340
341         if (th->th_seq + tlen != last->tqe_th->th_seq)
342                 return FALSE;
343         /* This segment comes immediately before the last pending segment */
344
345         prev = TAILQ_PREV(last, tsegqe_head, tqe_q);
346         if (prev == NULL) {
347                 /*
348                  * No pending preceeding segment, we assume this segment
349                  * could be reassembled.
350                  */
351                 return TRUE;
352         }
353
354         /* This segment does not overlap with the preceeding segment */
355         if (SEQ_GEQ(th->th_seq, prev->tqe_th->th_seq + prev->tqe_len))
356                 return TRUE;
357
358         return FALSE;
359 }
360
361 static __inline void
362 tcp_ncr_update_rxtthresh(struct tcpcb *tp)
363 {
364         int old_rxtthresh = tp->t_rxtthresh;
365         uint32_t ownd = tp->snd_max - tp->snd_una;
366
367         tp->t_rxtthresh = max(3, ((ownd / tp->t_maxseg) >> 1));
368         if (tp->t_rxtthresh != old_rxtthresh) {
369                 tcp_sack_update_lostseq(&tp->scb, tp->snd_una,
370                     tp->t_maxseg, tp->t_rxtthresh);
371         }
372 }
373
374 static int
375 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
376 {
377         struct tseg_qent *q;
378         struct tseg_qent *p = NULL;
379         struct tseg_qent *te;
380         struct socket *so = tp->t_inpcb->inp_socket;
381         int flags;
382
383         /*
384          * Call with th == NULL after become established to
385          * force pre-ESTABLISHED data up to user socket.
386          */
387         if (th == NULL)
388                 goto present;
389
390         /*
391          * Limit the number of segments in the reassembly queue to prevent
392          * holding on to too many segments (and thus running out of mbufs).
393          * Make sure to let the missing segment through which caused this
394          * queue.  Always keep one global queue entry spare to be able to
395          * process the missing segment.
396          */
397         if (th->th_seq != tp->rcv_nxt &&
398             tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
399                 tcp_reass_overflows++;
400                 tcpstat.tcps_rcvmemdrop++;
401                 m_freem(m);
402                 /* no SACK block to report */
403                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
404                 return (0);
405         }
406
407         /* Allocate a new queue entry. */
408         te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
409         if (te == NULL) {
410                 tcpstat.tcps_rcvmemdrop++;
411                 m_freem(m);
412                 /* no SACK block to report */
413                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
414                 return (0);
415         }
416         atomic_add_int(&tcp_reass_qsize, 1);
417
418         if (th->th_flags & TH_FIN)
419                 tp->t_flags |= TF_QUEDFIN;
420
421         /*
422          * Find a segment which begins after this one does.
423          */
424         TAILQ_FOREACH(q, &tp->t_segq, tqe_q) {
425                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
426                         break;
427                 p = q;
428         }
429
430         /*
431          * If there is a preceding segment, it may provide some of
432          * our data already.  If so, drop the data from the incoming
433          * segment.  If it provides all of our data, drop us.
434          */
435         if (p != NULL) {
436                 tcp_seq_diff_t i;
437
438                 /* conversion to int (in i) handles seq wraparound */
439                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
440                 if (i > 0) {            /* overlaps preceding segment */
441                         tp->sack_flags |=
442                             (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
443                         /* enclosing block starts w/ preceding segment */
444                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
445                         if (i >= *tlenp) {
446                                 if (th->th_flags & TH_FIN)
447                                         p->tqe_th->th_flags |= TH_FIN;
448
449                                 /* preceding encloses incoming segment */
450                                 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
451                                     p->tqe_th->th_seq + p->tqe_len,
452                                     p->tqe_th->th_flags);
453                                 tcpstat.tcps_rcvduppack++;
454                                 tcpstat.tcps_rcvdupbyte += *tlenp;
455                                 m_freem(m);
456                                 kfree(te, M_TSEGQ);
457                                 atomic_add_int(&tcp_reass_qsize, -1);
458                                 /*
459                                  * Try to present any queued data
460                                  * at the left window edge to the user.
461                                  * This is needed after the 3-WHS
462                                  * completes.
463                                  */
464                                 goto present;   /* ??? */
465                         }
466                         m_adj(m, i);
467                         *tlenp -= i;
468                         th->th_seq += i;
469                         /* incoming segment end is enclosing block end */
470                         tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
471                             th->th_seq + *tlenp, th->th_flags);
472                         /* trim end of reported D-SACK block */
473                         tp->reportblk.rblk_end = th->th_seq;
474                 }
475         }
476         tcpstat.tcps_rcvoopack++;
477         tcpstat.tcps_rcvoobyte += *tlenp;
478
479         /*
480          * While we overlap succeeding segments trim them or,
481          * if they are completely covered, dequeue them.
482          */
483         while (q) {
484                 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
485                 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
486                 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
487                 struct tseg_qent *nq;
488
489                 if (i <= 0)
490                         break;
491                 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
492                         /* first time through */
493                         tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
494                         tp->encloseblk = tp->reportblk;
495                         /* report trailing duplicate D-SACK segment */
496                         tp->reportblk.rblk_start = q->tqe_th->th_seq;
497                 }
498                 if ((tp->sack_flags & TSACK_F_ENCLOSESEG) &&
499                     SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
500                         /* extend enclosing block if one exists */
501                         tp->encloseblk.rblk_end = qend_sack;
502                 }
503                 if (i < q->tqe_len) {
504                         q->tqe_th->th_seq += i;
505                         q->tqe_len -= i;
506                         m_adj(q->tqe_m, i);
507                         break;
508                 }
509
510                 if (q->tqe_th->th_flags & TH_FIN)
511                         th->th_flags |= TH_FIN;
512
513                 nq = TAILQ_NEXT(q, tqe_q);
514                 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
515                 m_freem(q->tqe_m);
516                 kfree(q, M_TSEGQ);
517                 atomic_add_int(&tcp_reass_qsize, -1);
518                 q = nq;
519         }
520
521         /* Insert the new segment queue entry into place. */
522         te->tqe_m = m;
523         te->tqe_th = th;
524         te->tqe_len = *tlenp;
525
526         /* check if can coalesce with following segment */
527         if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
528                 tcp_seq tend_sack;
529
530                 te->tqe_len += q->tqe_len;
531                 if (q->tqe_th->th_flags & TH_FIN)
532                         te->tqe_th->th_flags |= TH_FIN;
533                 tend_sack = TCP_SACK_BLKEND(te->tqe_th->th_seq + te->tqe_len,
534                     te->tqe_th->th_flags);
535
536                 m_cat(te->tqe_m, q->tqe_m);
537                 tp->encloseblk.rblk_end = tend_sack;
538                 /*
539                  * When not reporting a duplicate segment, use
540                  * the larger enclosing block as the SACK block.
541                  */
542                 if (!(tp->sack_flags & TSACK_F_DUPSEG))
543                         tp->reportblk.rblk_end = tend_sack;
544                 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
545                 kfree(q, M_TSEGQ);
546                 atomic_add_int(&tcp_reass_qsize, -1);
547         }
548
549         if (p == NULL) {
550                 TAILQ_INSERT_HEAD(&tp->t_segq, te, tqe_q);
551         } else {
552                 /* check if can coalesce with preceding segment */
553                 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
554                         if (te->tqe_th->th_flags & TH_FIN)
555                                 p->tqe_th->th_flags |= TH_FIN;
556                         p->tqe_len += te->tqe_len;
557                         m_cat(p->tqe_m, te->tqe_m);
558                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
559                         /*
560                          * When not reporting a duplicate segment, use
561                          * the larger enclosing block as the SACK block.
562                          */
563                         if (!(tp->sack_flags & TSACK_F_DUPSEG))
564                                 tp->reportblk.rblk_start = p->tqe_th->th_seq;
565                         kfree(te, M_TSEGQ);
566                         atomic_add_int(&tcp_reass_qsize, -1);
567                 } else {
568                         TAILQ_INSERT_AFTER(&tp->t_segq, p, te, tqe_q);
569                 }
570         }
571
572 present:
573         /*
574          * Present data to user, advancing rcv_nxt through
575          * completed sequence space.
576          */
577         if (!TCPS_HAVEESTABLISHED(tp->t_state))
578                 return (0);
579         q = TAILQ_FIRST(&tp->t_segq);
580         if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
581                 return (0);
582         tp->rcv_nxt += q->tqe_len;
583         if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
584                 /* no SACK block to report since ACK advanced */
585                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
586         }
587         /* no enclosing block to report since ACK advanced */
588         tp->sack_flags &= ~TSACK_F_ENCLOSESEG;
589         flags = q->tqe_th->th_flags & TH_FIN;
590         TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
591         KASSERT(TAILQ_EMPTY(&tp->t_segq) ||
592                 TAILQ_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
593                 ("segment not coalesced"));
594         if (so->so_state & SS_CANTRCVMORE) {
595                 m_freem(q->tqe_m);
596         } else {
597                 lwkt_gettoken(&so->so_rcv.ssb_token);
598                 ssb_appendstream(&so->so_rcv, q->tqe_m);
599                 lwkt_reltoken(&so->so_rcv.ssb_token);
600         }
601         kfree(q, M_TSEGQ);
602         atomic_add_int(&tcp_reass_qsize, -1);
603         ND6_HINT(tp);
604         sorwakeup(so);
605         return (flags);
606 }
607
608 /*
609  * TCP input routine, follows pages 65-76 of the
610  * protocol specification dated September, 1981 very closely.
611  */
612 #ifdef INET6
613 int
614 tcp6_input(struct mbuf **mp, int *offp, int proto)
615 {
616         struct mbuf *m = *mp;
617         struct in6_ifaddr *ia6;
618
619         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
620
621         /*
622          * draft-itojun-ipv6-tcp-to-anycast
623          * better place to put this in?
624          */
625         ia6 = ip6_getdstifaddr(m);
626         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
627                 struct ip6_hdr *ip6;
628
629                 ip6 = mtod(m, struct ip6_hdr *);
630                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
631                             offsetof(struct ip6_hdr, ip6_dst));
632                 return (IPPROTO_DONE);
633         }
634
635         tcp_input(mp, offp, proto);
636         return (IPPROTO_DONE);
637 }
638 #endif
639
640 int
641 tcp_input(struct mbuf **mp, int *offp, int proto)
642 {
643         int off0;
644         struct tcphdr *th;
645         struct ip *ip = NULL;
646         struct ipovly *ipov;
647         struct inpcb *inp = NULL;
648         u_char *optp = NULL;
649         int optlen = 0;
650         int tlen, off;
651         int len = 0;
652         int drop_hdrlen;
653         struct tcpcb *tp = NULL;
654         int thflags;
655         struct socket *so = NULL;
656         int todrop, acked;
657         boolean_t ourfinisacked, needoutput = FALSE, delayed_dupack = FALSE;
658         tcp_seq th_dupack = 0; /* XXX gcc warning */
659         u_int to_flags = 0; /* XXX gcc warning */
660         u_long tiwin;
661         int recvwin;
662         struct tcpopt to;               /* options in this segment */
663         struct sockaddr_in *next_hop = NULL;
664         int rstreason; /* For badport_bandlim accounting purposes */
665         int cpu;
666         struct ip6_hdr *ip6 = NULL;
667         struct mbuf *m;
668 #ifdef INET6
669         boolean_t isipv6;
670 #else
671         const boolean_t isipv6 = FALSE;
672 #endif
673 #ifdef TCPDEBUG
674         short ostate = 0;
675 #endif
676
677         off0 = *offp;
678         m = *mp;
679         *mp = NULL;
680
681         tcpstat.tcps_rcvtotal++;
682
683         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
684                 struct m_tag *mtag;
685
686                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
687                 KKASSERT(mtag != NULL);
688                 next_hop = m_tag_data(mtag);
689         }
690
691 #ifdef INET6
692         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
693 #endif
694
695         if (isipv6) {
696                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
697                 ip6 = mtod(m, struct ip6_hdr *);
698                 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
699                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
700                         tcpstat.tcps_rcvbadsum++;
701                         goto drop;
702                 }
703                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
704
705                 /*
706                  * Be proactive about unspecified IPv6 address in source.
707                  * As we use all-zero to indicate unbounded/unconnected pcb,
708                  * unspecified IPv6 address can be used to confuse us.
709                  *
710                  * Note that packets with unspecified IPv6 destination is
711                  * already dropped in ip6_input.
712                  */
713                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
714                         /* XXX stat */
715                         goto drop;
716                 }
717         } else {
718                 /*
719                  * Get IP and TCP header together in first mbuf.
720                  * Note: IP leaves IP header in first mbuf.
721                  */
722                 if (off0 > sizeof(struct ip)) {
723                         ip_stripoptions(m);
724                         off0 = sizeof(struct ip);
725                 }
726                 /* already checked and pulled up in ip_demux() */
727                 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
728                     ("TCP header not in one mbuf: m->m_len %d", m->m_len));
729                 ip = mtod(m, struct ip *);
730                 ipov = (struct ipovly *)ip;
731                 th = (struct tcphdr *)((caddr_t)ip + off0);
732                 tlen = ip->ip_len;
733
734                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
735                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
736                                 th->th_sum = m->m_pkthdr.csum_data;
737                         else
738                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
739                                                 ip->ip_dst.s_addr,
740                                                 htonl(m->m_pkthdr.csum_data +
741                                                         ip->ip_len +
742                                                         IPPROTO_TCP));
743                         th->th_sum ^= 0xffff;
744                 } else {
745                         /*
746                          * Checksum extended TCP header and data.
747                          */
748                         len = sizeof(struct ip) + tlen;
749                         bzero(ipov->ih_x1, sizeof ipov->ih_x1);
750                         ipov->ih_len = (u_short)tlen;
751                         ipov->ih_len = htons(ipov->ih_len);
752                         th->th_sum = in_cksum(m, len);
753                 }
754                 if (th->th_sum) {
755                         tcpstat.tcps_rcvbadsum++;
756                         goto drop;
757                 }
758 #ifdef INET6
759                 /* Re-initialization for later version check */
760                 ip->ip_v = IPVERSION;
761 #endif
762         }
763
764         /*
765          * Check that TCP offset makes sense,
766          * pull out TCP options and adjust length.              XXX
767          */
768         off = th->th_off << 2;
769         /* already checked and pulled up in ip_demux() */
770         KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
771             ("bad TCP data offset %d (tlen %d)", off, tlen));
772         tlen -= off;    /* tlen is used instead of ti->ti_len */
773         if (off > sizeof(struct tcphdr)) {
774                 if (isipv6) {
775                         IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
776                         ip6 = mtod(m, struct ip6_hdr *);
777                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
778                 } else {
779                         /* already pulled up in ip_demux() */
780                         KASSERT(m->m_len >= sizeof(struct ip) + off,
781                             ("TCP header and options not in one mbuf: "
782                              "m_len %d, off %d", m->m_len, off));
783                 }
784                 optlen = off - sizeof(struct tcphdr);
785                 optp = (u_char *)(th + 1);
786         }
787         thflags = th->th_flags;
788
789 #ifdef TCP_DROP_SYNFIN
790         /*
791          * If the drop_synfin option is enabled, drop all packets with
792          * both the SYN and FIN bits set. This prevents e.g. nmap from
793          * identifying the TCP/IP stack.
794          *
795          * This is a violation of the TCP specification.
796          */
797         if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
798                 goto drop;
799 #endif
800
801         /*
802          * Convert TCP protocol specific fields to host format.
803          */
804         th->th_seq = ntohl(th->th_seq);
805         th->th_ack = ntohl(th->th_ack);
806         th->th_win = ntohs(th->th_win);
807         th->th_urp = ntohs(th->th_urp);
808
809         /*
810          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
811          * until after ip6_savecontrol() is called and before other functions
812          * which don't want those proto headers.
813          * Because ip6_savecontrol() is going to parse the mbuf to
814          * search for data to be passed up to user-land, it wants mbuf
815          * parameters to be unchanged.
816          * XXX: the call of ip6_savecontrol() has been obsoleted based on
817          * latest version of the advanced API (20020110).
818          */
819         drop_hdrlen = off0 + off;
820
821         /*
822          * Locate pcb for segment.
823          */
824 findpcb:
825         /* IPFIREWALL_FORWARD section */
826         if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
827                 /*
828                  * Transparently forwarded. Pretend to be the destination.
829                  * already got one like this?
830                  */
831                 cpu = mycpu->gd_cpuid;
832                 inp = in_pcblookup_hash(&tcbinfo[cpu],
833                                         ip->ip_src, th->th_sport,
834                                         ip->ip_dst, th->th_dport,
835                                         0, m->m_pkthdr.rcvif);
836                 if (!inp) {
837                         /*
838                          * It's new.  Try to find the ambushing socket.
839                          */
840
841                         /*
842                          * The rest of the ipfw code stores the port in
843                          * host order.  XXX
844                          * (The IP address is still in network order.)
845                          */
846                         in_port_t dport = next_hop->sin_port ?
847                                                 htons(next_hop->sin_port) :
848                                                 th->th_dport;
849
850                         cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
851                                           next_hop->sin_addr.s_addr, dport);
852                         inp = in_pcblookup_hash(&tcbinfo[cpu],
853                                                 ip->ip_src, th->th_sport,
854                                                 next_hop->sin_addr, dport,
855                                                 1, m->m_pkthdr.rcvif);
856                 }
857         } else {
858                 if (isipv6) {
859                         inp = in6_pcblookup_hash(&tcbinfo[0],
860                                                  &ip6->ip6_src, th->th_sport,
861                                                  &ip6->ip6_dst, th->th_dport,
862                                                  1, m->m_pkthdr.rcvif);
863                 } else {
864                         cpu = mycpu->gd_cpuid;
865                         inp = in_pcblookup_hash(&tcbinfo[cpu],
866                                                 ip->ip_src, th->th_sport,
867                                                 ip->ip_dst, th->th_dport,
868                                                 1, m->m_pkthdr.rcvif);
869                 }
870         }
871
872         /*
873          * If the state is CLOSED (i.e., TCB does not exist) then
874          * all data in the incoming segment is discarded.
875          * If the TCB exists but is in CLOSED state, it is embryonic,
876          * but should either do a listen or a connect soon.
877          */
878         if (inp == NULL) {
879                 if (log_in_vain) {
880 #ifdef INET6
881                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
882 #else
883                         char dbuf[sizeof "aaa.bbb.ccc.ddd"];
884                         char sbuf[sizeof "aaa.bbb.ccc.ddd"];
885 #endif
886                         if (isipv6) {
887                                 strcpy(dbuf, "[");
888                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
889                                 strcat(dbuf, "]");
890                                 strcpy(sbuf, "[");
891                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
892                                 strcat(sbuf, "]");
893                         } else {
894                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
895                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
896                         }
897                         switch (log_in_vain) {
898                         case 1:
899                                 if (!(thflags & TH_SYN))
900                                         break;
901                         case 2:
902                                 log(LOG_INFO,
903                                     "Connection attempt to TCP %s:%d "
904                                     "from %s:%d flags:0x%02x\n",
905                                     dbuf, ntohs(th->th_dport), sbuf,
906                                     ntohs(th->th_sport), thflags);
907                                 break;
908                         default:
909                                 break;
910                         }
911                 }
912                 if (blackhole) {
913                         switch (blackhole) {
914                         case 1:
915                                 if (thflags & TH_SYN)
916                                         goto drop;
917                                 break;
918                         case 2:
919                                 goto drop;
920                         default:
921                                 goto drop;
922                         }
923                 }
924                 rstreason = BANDLIM_RST_CLOSEDPORT;
925                 goto dropwithreset;
926         }
927
928 #ifdef IPSEC
929         if (isipv6) {
930                 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
931                         ipsec6stat.in_polvio++;
932                         goto drop;
933                 }
934         } else {
935                 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
936                         ipsecstat.in_polvio++;
937                         goto drop;
938                 }
939         }
940 #endif
941 #ifdef FAST_IPSEC
942         if (isipv6) {
943                 if (ipsec6_in_reject(m, inp))
944                         goto drop;
945         } else {
946                 if (ipsec4_in_reject(m, inp))
947                         goto drop;
948         }
949 #endif
950         /* Check the minimum TTL for socket. */
951 #ifdef INET6
952         if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
953                 goto drop;
954 #endif
955
956         tp = intotcpcb(inp);
957         if (tp == NULL) {
958                 rstreason = BANDLIM_RST_CLOSEDPORT;
959                 goto dropwithreset;
960         }
961         if (tp->t_state <= TCPS_CLOSED)
962                 goto drop;
963
964         so = inp->inp_socket;
965
966 #ifdef TCPDEBUG
967         if (so->so_options & SO_DEBUG) {
968                 ostate = tp->t_state;
969                 if (isipv6)
970                         bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
971                 else
972                         bcopy(ip, tcp_saveipgen, sizeof(*ip));
973                 tcp_savetcp = *th;
974         }
975 #endif
976
977         bzero(&to, sizeof to);
978
979         if (so->so_options & SO_ACCEPTCONN) {
980                 struct in_conninfo inc;
981
982 #ifdef INET6
983                 inc.inc_isipv6 = (isipv6 == TRUE);
984 #endif
985                 if (isipv6) {
986                         inc.inc6_faddr = ip6->ip6_src;
987                         inc.inc6_laddr = ip6->ip6_dst;
988                         inc.inc6_route.ro_rt = NULL;            /* XXX */
989                 } else {
990                         inc.inc_faddr = ip->ip_src;
991                         inc.inc_laddr = ip->ip_dst;
992                         inc.inc_route.ro_rt = NULL;             /* XXX */
993                 }
994                 inc.inc_fport = th->th_sport;
995                 inc.inc_lport = th->th_dport;
996
997                 /*
998                  * If the state is LISTEN then ignore segment if it contains
999                  * a RST.  If the segment contains an ACK then it is bad and
1000                  * send a RST.  If it does not contain a SYN then it is not
1001                  * interesting; drop it.
1002                  *
1003                  * If the state is SYN_RECEIVED (syncache) and seg contains
1004                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
1005                  * contains a RST, check the sequence number to see if it
1006                  * is a valid reset segment.
1007                  */
1008                 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
1009                         if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
1010                                 if (!syncache_expand(&inc, th, &so, m)) {
1011                                         /*
1012                                          * No syncache entry, or ACK was not
1013                                          * for our SYN/ACK.  Send a RST.
1014                                          */
1015                                         tcpstat.tcps_badsyn++;
1016                                         rstreason = BANDLIM_RST_OPENPORT;
1017                                         goto dropwithreset;
1018                                 }
1019
1020                                 /*
1021                                  * Could not complete 3-way handshake,
1022                                  * connection is being closed down, and
1023                                  * syncache will free mbuf.
1024                                  */
1025                                 if (so == NULL)
1026                                         return(IPPROTO_DONE);
1027
1028                                 /*
1029                                  * We must be in the correct protocol thread
1030                                  * for this connection.
1031                                  */
1032                                 KKASSERT(so->so_port == &curthread->td_msgport);
1033
1034                                 /*
1035                                  * Socket is created in state SYN_RECEIVED.
1036                                  * Continue processing segment.
1037                                  */
1038                                 inp = so->so_pcb;
1039                                 tp = intotcpcb(inp);
1040                                 /*
1041                                  * This is what would have happened in
1042                                  * tcp_output() when the SYN,ACK was sent.
1043                                  */
1044                                 tp->snd_up = tp->snd_una;
1045                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
1046                                 tp->last_ack_sent = tp->rcv_nxt;
1047
1048                                 goto after_listen;
1049                         }
1050                         if (thflags & TH_RST) {
1051                                 syncache_chkrst(&inc, th);
1052                                 goto drop;
1053                         }
1054                         if (thflags & TH_ACK) {
1055                                 syncache_badack(&inc);
1056                                 tcpstat.tcps_badsyn++;
1057                                 rstreason = BANDLIM_RST_OPENPORT;
1058                                 goto dropwithreset;
1059                         }
1060                         goto drop;
1061                 }
1062
1063                 /*
1064                  * Segment's flags are (SYN) or (SYN | FIN).
1065                  */
1066 #ifdef INET6
1067                 /*
1068                  * If deprecated address is forbidden,
1069                  * we do not accept SYN to deprecated interface
1070                  * address to prevent any new inbound connection from
1071                  * getting established.
1072                  * When we do not accept SYN, we send a TCP RST,
1073                  * with deprecated source address (instead of dropping
1074                  * it).  We compromise it as it is much better for peer
1075                  * to send a RST, and RST will be the final packet
1076                  * for the exchange.
1077                  *
1078                  * If we do not forbid deprecated addresses, we accept
1079                  * the SYN packet.  RFC2462 does not suggest dropping
1080                  * SYN in this case.
1081                  * If we decipher RFC2462 5.5.4, it says like this:
1082                  * 1. use of deprecated addr with existing
1083                  *    communication is okay - "SHOULD continue to be
1084                  *    used"
1085                  * 2. use of it with new communication:
1086                  *   (2a) "SHOULD NOT be used if alternate address
1087                  *        with sufficient scope is available"
1088                  *   (2b) nothing mentioned otherwise.
1089                  * Here we fall into (2b) case as we have no choice in
1090                  * our source address selection - we must obey the peer.
1091                  *
1092                  * The wording in RFC2462 is confusing, and there are
1093                  * multiple description text for deprecated address
1094                  * handling - worse, they are not exactly the same.
1095                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
1096                  */
1097                 if (isipv6 && !ip6_use_deprecated) {
1098                         struct in6_ifaddr *ia6;
1099
1100                         if ((ia6 = ip6_getdstifaddr(m)) &&
1101                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1102                                 tp = NULL;
1103                                 rstreason = BANDLIM_RST_OPENPORT;
1104                                 goto dropwithreset;
1105                         }
1106                 }
1107 #endif
1108                 /*
1109                  * If it is from this socket, drop it, it must be forged.
1110                  * Don't bother responding if the destination was a broadcast.
1111                  */
1112                 if (th->th_dport == th->th_sport) {
1113                         if (isipv6) {
1114                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1115                                                        &ip6->ip6_src))
1116                                         goto drop;
1117                         } else {
1118                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1119                                         goto drop;
1120                         }
1121                 }
1122                 /*
1123                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1124                  *
1125                  * Note that it is quite possible to receive unicast
1126                  * link-layer packets with a broadcast IP address. Use
1127                  * in_broadcast() to find them.
1128                  */
1129                 if (m->m_flags & (M_BCAST | M_MCAST))
1130                         goto drop;
1131                 if (isipv6) {
1132                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1133                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1134                                 goto drop;
1135                 } else {
1136                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1137                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1138                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1139                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1140                                 goto drop;
1141                 }
1142                 /*
1143                  * SYN appears to be valid; create compressed TCP state
1144                  * for syncache, or perform t/tcp connection.
1145                  */
1146                 if (so->so_qlen <= so->so_qlimit) {
1147                         tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack);
1148                         if (!syncache_add(&inc, &to, th, so, m))
1149                                 goto drop;
1150
1151                         /*
1152                          * Entry added to syncache, mbuf used to
1153                          * send SYN,ACK packet.
1154                          */
1155                         return(IPPROTO_DONE);
1156                 }
1157                 goto drop;
1158         }
1159
1160 after_listen:
1161         /*
1162          * Should not happen - syncache should pick up these connections.
1163          *
1164          * Once we are past handling listen sockets we must be in the
1165          * correct protocol processing thread.
1166          */
1167         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1168         KKASSERT(so->so_port == &curthread->td_msgport);
1169
1170         /* Unscale the window into a 32-bit value. */
1171         if (!(thflags & TH_SYN))
1172                 tiwin = th->th_win << tp->snd_scale;
1173         else
1174                 tiwin = th->th_win;
1175
1176         /*
1177          * This is the second part of the MSS DoS prevention code (after
1178          * minmss on the sending side) and it deals with too many too small
1179          * tcp packets in a too short timeframe (1 second).
1180          *
1181          * XXX Removed.  This code was crap.  It does not scale to network
1182          *     speed, and default values break NFS.  Gone.
1183          */
1184         /* REMOVED */
1185
1186         /*
1187          * Segment received on connection.
1188          *
1189          * Reset idle time and keep-alive timer.  Don't waste time if less
1190          * then a second has elapsed.
1191          */
1192         if ((int)(ticks - tp->t_rcvtime) > hz)
1193                 tcp_timer_keep_activity(tp, thflags);
1194
1195         /*
1196          * Process options.
1197          * XXX this is tradtitional behavior, may need to be cleaned up.
1198          */
1199         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack);
1200         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1201                 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
1202                         tp->t_flags |= TF_RCVD_SCALE;
1203                         tp->snd_scale = to.to_requested_s_scale;
1204                 }
1205
1206                 /*
1207                  * Initial send window; will be updated upon next ACK
1208                  */
1209                 tp->snd_wnd = th->th_win;
1210
1211                 if (to.to_flags & TOF_TS) {
1212                         tp->t_flags |= TF_RCVD_TSTMP;
1213                         tp->ts_recent = to.to_tsval;
1214                         tp->ts_recent_age = ticks;
1215                 }
1216                 if (!(to.to_flags & TOF_MSS))
1217                         to.to_mss = 0;
1218                 tcp_mss(tp, to.to_mss);
1219                 /*
1220                  * Only set the TF_SACK_PERMITTED per-connection flag
1221                  * if we got a SACK_PERMITTED option from the other side
1222                  * and the global tcp_do_sack variable is true.
1223                  */
1224                 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1225                         tp->t_flags |= TF_SACK_PERMITTED;
1226         }
1227
1228         /*
1229          * Header prediction: check for the two common cases
1230          * of a uni-directional data xfer.  If the packet has
1231          * no control flags, is in-sequence, the window didn't
1232          * change and we're not retransmitting, it's a
1233          * candidate.  If the length is zero and the ack moved
1234          * forward, we're the sender side of the xfer.  Just
1235          * free the data acked & wake any higher level process
1236          * that was blocked waiting for space.  If the length
1237          * is non-zero and the ack didn't move, we're the
1238          * receiver side.  If we're getting packets in-order
1239          * (the reassembly queue is empty), add the data to
1240          * the socket buffer and note that we need a delayed ack.
1241          * Make sure that the hidden state-flags are also off.
1242          * Since we check for TCPS_ESTABLISHED above, it can only
1243          * be TH_NEEDSYN.
1244          */
1245         if (tp->t_state == TCPS_ESTABLISHED &&
1246             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1247             !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1248             (!(to.to_flags & TOF_TS) ||
1249              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1250             th->th_seq == tp->rcv_nxt &&
1251             tp->snd_nxt == tp->snd_max) {
1252
1253                 /*
1254                  * If last ACK falls within this segment's sequence numbers,
1255                  * record the timestamp.
1256                  * NOTE that the test is modified according to the latest
1257                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1258                  */
1259                 if ((to.to_flags & TOF_TS) &&
1260                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1261                         tp->ts_recent_age = ticks;
1262                         tp->ts_recent = to.to_tsval;
1263                 }
1264
1265                 if (tlen == 0) {
1266                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1267                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1268                             tp->snd_cwnd >= tp->snd_wnd &&
1269                             !IN_FASTRECOVERY(tp)) {
1270                                 /*
1271                                  * This is a pure ack for outstanding data.
1272                                  */
1273                                 ++tcpstat.tcps_predack;
1274                                 /*
1275                                  * "bad retransmit" recovery
1276                                  *
1277                                  * If Eifel detection applies, then
1278                                  * it is deterministic, so use it
1279                                  * unconditionally over the old heuristic.
1280                                  * Otherwise, fall back to the old heuristic.
1281                                  */
1282                                 if (tcp_do_eifel_detect &&
1283                                     (to.to_flags & TOF_TS) && to.to_tsecr &&
1284                                     (tp->rxt_flags & TRXT_F_FIRSTACCACK)) {
1285                                         /* Eifel detection applicable. */
1286                                         if (to.to_tsecr < tp->t_rexmtTS) {
1287                                                 tcp_revert_congestion_state(tp);
1288                                                 ++tcpstat.tcps_eifeldetected;
1289                                                 if (tp->t_rxtshift != 1 ||
1290                                                     ticks >= tp->t_badrxtwin)
1291                                                         ++tcpstat.tcps_rttcantdetect;
1292                                         }
1293                                 } else if (tp->t_rxtshift == 1 &&
1294                                            ticks < tp->t_badrxtwin) {
1295                                         tcp_revert_congestion_state(tp);
1296                                         ++tcpstat.tcps_rttdetected;
1297                                 }
1298                                 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK |
1299                                     TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT);
1300                                 /*
1301                                  * Recalculate the retransmit timer / rtt.
1302                                  *
1303                                  * Some machines (certain windows boxes)
1304                                  * send broken timestamp replies during the
1305                                  * SYN+ACK phase, ignore timestamps of 0.
1306                                  */
1307                                 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1308                                         tcp_xmit_timer(tp,
1309                                             ticks - to.to_tsecr + 1,
1310                                             th->th_ack);
1311                                 } else if (tp->t_rtttime &&
1312                                            SEQ_GT(th->th_ack, tp->t_rtseq)) {
1313                                         tcp_xmit_timer(tp,
1314                                             ticks - tp->t_rtttime,
1315                                             th->th_ack);
1316                                 }
1317                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1318                                 acked = th->th_ack - tp->snd_una;
1319                                 tcpstat.tcps_rcvackpack++;
1320                                 tcpstat.tcps_rcvackbyte += acked;
1321                                 sbdrop(&so->so_snd.sb, acked);
1322                                 tp->snd_recover = th->th_ack - 1;
1323                                 tp->snd_una = th->th_ack;
1324                                 tp->t_dupacks = 0;
1325                                 /*
1326                                  * Update window information.
1327                                  */
1328                                 if (tiwin != tp->snd_wnd &&
1329                                     acceptable_window_update(tp, th, tiwin)) {
1330                                         /* keep track of pure window updates */
1331                                         if (tp->snd_wl2 == th->th_ack &&
1332                                             tiwin > tp->snd_wnd)
1333                                                 tcpstat.tcps_rcvwinupd++;
1334                                         tp->snd_wnd = tiwin;
1335                                         tp->snd_wl1 = th->th_seq;
1336                                         tp->snd_wl2 = th->th_ack;
1337                                         if (tp->snd_wnd > tp->max_sndwnd)
1338                                                 tp->max_sndwnd = tp->snd_wnd;
1339                                 }
1340                                 m_freem(m);
1341                                 ND6_HINT(tp); /* some progress has been done */
1342                                 /*
1343                                  * If all outstanding data are acked, stop
1344                                  * retransmit timer, otherwise restart timer
1345                                  * using current (possibly backed-off) value.
1346                                  * If process is waiting for space,
1347                                  * wakeup/selwakeup/signal.  If data
1348                                  * are ready to send, let tcp_output
1349                                  * decide between more output or persist.
1350                                  */
1351                                 if (tp->snd_una == tp->snd_max) {
1352                                         tcp_callout_stop(tp, tp->tt_rexmt);
1353                                 } else if (!tcp_callout_active(tp,
1354                                             tp->tt_persist)) {
1355                                         tcp_callout_reset(tp, tp->tt_rexmt,
1356                                             tp->t_rxtcur, tcp_timer_rexmt);
1357                                 }
1358                                 sowwakeup(so);
1359                                 if (so->so_snd.ssb_cc > 0)
1360                                         tcp_output(tp);
1361                                 return(IPPROTO_DONE);
1362                         }
1363                 } else if (tiwin == tp->snd_wnd &&
1364                     th->th_ack == tp->snd_una &&
1365                     TAILQ_EMPTY(&tp->t_segq) &&
1366                     tlen <= ssb_space(&so->so_rcv)) {
1367                         u_long newsize = 0;     /* automatic sockbuf scaling */
1368                         /*
1369                          * This is a pure, in-sequence data packet
1370                          * with nothing on the reassembly queue and
1371                          * we have enough buffer space to take it.
1372                          */
1373                         ++tcpstat.tcps_preddat;
1374                         tp->rcv_nxt += tlen;
1375                         tcpstat.tcps_rcvpack++;
1376                         tcpstat.tcps_rcvbyte += tlen;
1377                         ND6_HINT(tp);   /* some progress has been done */
1378                 /*
1379                  * Automatic sizing of receive socket buffer.  Often the send
1380                  * buffer size is not optimally adjusted to the actual network
1381                  * conditions at hand (delay bandwidth product).  Setting the
1382                  * buffer size too small limits throughput on links with high
1383                  * bandwidth and high delay (eg. trans-continental/oceanic links).
1384                  *
1385                  * On the receive side the socket buffer memory is only rarely
1386                  * used to any significant extent.  This allows us to be much
1387                  * more aggressive in scaling the receive socket buffer.  For
1388                  * the case that the buffer space is actually used to a large
1389                  * extent and we run out of kernel memory we can simply drop
1390                  * the new segments; TCP on the sender will just retransmit it
1391                  * later.  Setting the buffer size too big may only consume too
1392                  * much kernel memory if the application doesn't read() from
1393                  * the socket or packet loss or reordering makes use of the
1394                  * reassembly queue.
1395                  *
1396                  * The criteria to step up the receive buffer one notch are:
1397                  *  1. the number of bytes received during the time it takes
1398                  *     one timestamp to be reflected back to us (the RTT);
1399                  *  2. received bytes per RTT is within seven eighth of the
1400                  *     current socket buffer size;
1401                  *  3. receive buffer size has not hit maximal automatic size;
1402                  *
1403                  * This algorithm does one step per RTT at most and only if
1404                  * we receive a bulk stream w/o packet losses or reorderings.
1405                  * Shrinking the buffer during idle times is not necessary as
1406                  * it doesn't consume any memory when idle.
1407                  *
1408                  * TODO: Only step up if the application is actually serving
1409                  * the buffer to better manage the socket buffer resources.
1410                  */
1411                         if (tcp_do_autorcvbuf &&
1412                             to.to_tsecr &&
1413                             (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1414                                 if (to.to_tsecr > tp->rfbuf_ts &&
1415                                     to.to_tsecr - tp->rfbuf_ts < hz) {
1416                                         if (tp->rfbuf_cnt >
1417                                             (so->so_rcv.ssb_hiwat / 8 * 7) &&
1418                                             so->so_rcv.ssb_hiwat <
1419                                             tcp_autorcvbuf_max) {
1420                                                 newsize =
1421                                                     ulmin(so->so_rcv.ssb_hiwat +
1422                                                           tcp_autorcvbuf_inc,
1423                                                           tcp_autorcvbuf_max);
1424                                         }
1425                                         /* Start over with next RTT. */
1426                                         tp->rfbuf_ts = 0;
1427                                         tp->rfbuf_cnt = 0;
1428                                 } else
1429                                         tp->rfbuf_cnt += tlen;  /* add up */
1430                         }
1431                         /*
1432                          * Add data to socket buffer.
1433                          */
1434                         if (so->so_state & SS_CANTRCVMORE) {
1435                                 m_freem(m);
1436                         } else {
1437                                 /*
1438                                  * Set new socket buffer size, give up when
1439                                  * limit is reached.
1440                                  *
1441                                  * Adjusting the size can mess up ACK
1442                                  * sequencing when pure window updates are
1443                                  * being avoided (which is the default),
1444                                  * so force an ack.
1445                                  */
1446                                 lwkt_gettoken(&so->so_rcv.ssb_token);
1447                                 if (newsize) {
1448                                         tp->t_flags |= TF_RXRESIZED;
1449                                         if (!ssb_reserve(&so->so_rcv, newsize,
1450                                                          so, NULL)) {
1451                                                 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1452                                         }
1453                                         if (newsize >=
1454                                             (TCP_MAXWIN << tp->rcv_scale)) {
1455                                                 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1456                                         }
1457                                 }
1458                                 m_adj(m, drop_hdrlen); /* delayed header drop */
1459                                 ssb_appendstream(&so->so_rcv, m);
1460                                 lwkt_reltoken(&so->so_rcv.ssb_token);
1461                         }
1462                         sorwakeup(so);
1463                         /*
1464                          * This code is responsible for most of the ACKs
1465                          * the TCP stack sends back after receiving a data
1466                          * packet.  Note that the DELAY_ACK check fails if
1467                          * the delack timer is already running, which results
1468                          * in an ack being sent every other packet (which is
1469                          * what we want).
1470                          *
1471                          * We then further aggregate acks by not actually
1472                          * sending one until the protocol thread has completed
1473                          * processing the current backlog of packets.  This
1474                          * does not delay the ack any further, but allows us
1475                          * to take advantage of the packet aggregation that
1476                          * high speed NICs do (usually blocks of 8-10 packets)
1477                          * to send a single ack rather then four or five acks,
1478                          * greatly reducing the ack rate, the return channel
1479                          * bandwidth, and the protocol overhead on both ends.
1480                          *
1481                          * Since this also has the effect of slowing down
1482                          * the exponential slow-start ramp-up, systems with 
1483                          * very large bandwidth-delay products might want
1484                          * to turn the feature off.
1485                          */
1486                         if (DELAY_ACK(tp)) {
1487                                 tcp_callout_reset(tp, tp->tt_delack,
1488                                     tcp_delacktime, tcp_timer_delack);
1489                         } else if (tcp_aggregate_acks) {
1490                                 tp->t_flags |= TF_ACKNOW;
1491                                 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1492                                         tp->t_flags |= TF_ONOUTPUTQ;
1493                                         tp->tt_cpu = mycpu->gd_cpuid;
1494                                         TAILQ_INSERT_TAIL(
1495                                             &tcpcbackq[tp->tt_cpu],
1496                                             tp, t_outputq);
1497                                 }
1498                         } else {
1499                                 tp->t_flags |= TF_ACKNOW;
1500                                 tcp_output(tp);
1501                         }
1502                         return(IPPROTO_DONE);
1503                 }
1504         }
1505
1506         /*
1507          * Calculate amount of space in receive window,
1508          * and then do TCP input processing.
1509          * Receive window is amount of space in rcv queue,
1510          * but not less than advertised window.
1511          */
1512         recvwin = ssb_space(&so->so_rcv);
1513         if (recvwin < 0)
1514                 recvwin = 0;
1515         tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1516
1517         /* Reset receive buffer auto scaling when not in bulk receive mode. */
1518         tp->rfbuf_ts = 0;
1519         tp->rfbuf_cnt = 0;
1520
1521         switch (tp->t_state) {
1522         /*
1523          * If the state is SYN_RECEIVED:
1524          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1525          */
1526         case TCPS_SYN_RECEIVED:
1527                 if ((thflags & TH_ACK) &&
1528                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1529                      SEQ_GT(th->th_ack, tp->snd_max))) {
1530                                 rstreason = BANDLIM_RST_OPENPORT;
1531                                 goto dropwithreset;
1532                 }
1533                 break;
1534
1535         /*
1536          * If the state is SYN_SENT:
1537          *      if seg contains an ACK, but not for our SYN, drop the input.
1538          *      if seg contains a RST, then drop the connection.
1539          *      if seg does not contain SYN, then drop it.
1540          * Otherwise this is an acceptable SYN segment
1541          *      initialize tp->rcv_nxt and tp->irs
1542          *      if seg contains ack then advance tp->snd_una
1543          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1544          *      arrange for segment to be acked (eventually)
1545          *      continue processing rest of data/controls, beginning with URG
1546          */
1547         case TCPS_SYN_SENT:
1548                 if ((thflags & TH_ACK) &&
1549                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1550                      SEQ_GT(th->th_ack, tp->snd_max))) {
1551                         rstreason = BANDLIM_UNLIMITED;
1552                         goto dropwithreset;
1553                 }
1554                 if (thflags & TH_RST) {
1555                         if (thflags & TH_ACK)
1556                                 tp = tcp_drop(tp, ECONNREFUSED);
1557                         goto drop;
1558                 }
1559                 if (!(thflags & TH_SYN))
1560                         goto drop;
1561
1562                 tp->irs = th->th_seq;
1563                 tcp_rcvseqinit(tp);
1564                 if (thflags & TH_ACK) {
1565                         /* Our SYN was acked. */
1566                         tcpstat.tcps_connects++;
1567                         soisconnected(so);
1568                         /* Do window scaling on this connection? */
1569                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1570                             (TF_RCVD_SCALE | TF_REQ_SCALE))
1571                                 tp->rcv_scale = tp->request_r_scale;
1572                         tp->rcv_adv += tp->rcv_wnd;
1573                         tp->snd_una++;          /* SYN is acked */
1574                         tcp_callout_stop(tp, tp->tt_rexmt);
1575                         /*
1576                          * If there's data, delay ACK; if there's also a FIN
1577                          * ACKNOW will be turned on later.
1578                          */
1579                         if (DELAY_ACK(tp) && tlen != 0) {
1580                                 tcp_callout_reset(tp, tp->tt_delack,
1581                                     tcp_delacktime, tcp_timer_delack);
1582                         } else {
1583                                 tp->t_flags |= TF_ACKNOW;
1584                         }
1585                         /*
1586                          * Received <SYN,ACK> in SYN_SENT[*] state.
1587                          * Transitions:
1588                          *      SYN_SENT  --> ESTABLISHED
1589                          *      SYN_SENT* --> FIN_WAIT_1
1590                          */
1591                         tp->t_starttime = ticks;
1592                         if (tp->t_flags & TF_NEEDFIN) {
1593                                 tp->t_state = TCPS_FIN_WAIT_1;
1594                                 tp->t_flags &= ~TF_NEEDFIN;
1595                                 thflags &= ~TH_SYN;
1596                         } else {
1597                                 tcp_established(tp);
1598                         }
1599                 } else {
1600                         /*
1601                          * Received initial SYN in SYN-SENT[*] state =>
1602                          * simultaneous open.
1603                          * Do 3-way handshake:
1604                          *        SYN-SENT -> SYN-RECEIVED
1605                          *        SYN-SENT* -> SYN-RECEIVED*
1606                          */
1607                         tp->t_flags |= TF_ACKNOW;
1608                         tcp_callout_stop(tp, tp->tt_rexmt);
1609                         tp->t_state = TCPS_SYN_RECEIVED;
1610                 }
1611
1612                 /*
1613                  * Advance th->th_seq to correspond to first data byte.
1614                  * If data, trim to stay within window,
1615                  * dropping FIN if necessary.
1616                  */
1617                 th->th_seq++;
1618                 if (tlen > tp->rcv_wnd) {
1619                         todrop = tlen - tp->rcv_wnd;
1620                         m_adj(m, -todrop);
1621                         tlen = tp->rcv_wnd;
1622                         thflags &= ~TH_FIN;
1623                         tcpstat.tcps_rcvpackafterwin++;
1624                         tcpstat.tcps_rcvbyteafterwin += todrop;
1625                 }
1626                 tp->snd_wl1 = th->th_seq - 1;
1627                 tp->rcv_up = th->th_seq;
1628                 /*
1629                  * Client side of transaction: already sent SYN and data.
1630                  * If the remote host used T/TCP to validate the SYN,
1631                  * our data will be ACK'd; if so, enter normal data segment
1632                  * processing in the middle of step 5, ack processing.
1633                  * Otherwise, goto step 6.
1634                  */
1635                 if (thflags & TH_ACK)
1636                         goto process_ACK;
1637
1638                 goto step6;
1639
1640         /*
1641          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1642          *      do normal processing (we no longer bother with T/TCP).
1643          */
1644         case TCPS_LAST_ACK:
1645         case TCPS_CLOSING:
1646         case TCPS_TIME_WAIT:
1647                 break;  /* continue normal processing */
1648         }
1649
1650         /*
1651          * States other than LISTEN or SYN_SENT.
1652          * First check the RST flag and sequence number since reset segments
1653          * are exempt from the timestamp and connection count tests.  This
1654          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1655          * below which allowed reset segments in half the sequence space
1656          * to fall though and be processed (which gives forged reset
1657          * segments with a random sequence number a 50 percent chance of
1658          * killing a connection).
1659          * Then check timestamp, if present.
1660          * Then check the connection count, if present.
1661          * Then check that at least some bytes of segment are within
1662          * receive window.  If segment begins before rcv_nxt,
1663          * drop leading data (and SYN); if nothing left, just ack.
1664          *
1665          *
1666          * If the RST bit is set, check the sequence number to see
1667          * if this is a valid reset segment.
1668          * RFC 793 page 37:
1669          *   In all states except SYN-SENT, all reset (RST) segments
1670          *   are validated by checking their SEQ-fields.  A reset is
1671          *   valid if its sequence number is in the window.
1672          * Note: this does not take into account delayed ACKs, so
1673          *   we should test against last_ack_sent instead of rcv_nxt.
1674          *   The sequence number in the reset segment is normally an
1675          *   echo of our outgoing acknowledgement numbers, but some hosts
1676          *   send a reset with the sequence number at the rightmost edge
1677          *   of our receive window, and we have to handle this case.
1678          * If we have multiple segments in flight, the intial reset
1679          * segment sequence numbers will be to the left of last_ack_sent,
1680          * but they will eventually catch up.
1681          * In any case, it never made sense to trim reset segments to
1682          * fit the receive window since RFC 1122 says:
1683          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1684          *
1685          *    A TCP SHOULD allow a received RST segment to include data.
1686          *
1687          *    DISCUSSION
1688          *         It has been suggested that a RST segment could contain
1689          *         ASCII text that encoded and explained the cause of the
1690          *         RST.  No standard has yet been established for such
1691          *         data.
1692          *
1693          * If the reset segment passes the sequence number test examine
1694          * the state:
1695          *    SYN_RECEIVED STATE:
1696          *      If passive open, return to LISTEN state.
1697          *      If active open, inform user that connection was refused.
1698          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1699          *      Inform user that connection was reset, and close tcb.
1700          *    CLOSING, LAST_ACK STATES:
1701          *      Close the tcb.
1702          *    TIME_WAIT STATE:
1703          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1704          *      RFC 1337.
1705          */
1706         if (thflags & TH_RST) {
1707                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1708                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1709                         switch (tp->t_state) {
1710
1711                         case TCPS_SYN_RECEIVED:
1712                                 so->so_error = ECONNREFUSED;
1713                                 goto close;
1714
1715                         case TCPS_ESTABLISHED:
1716                         case TCPS_FIN_WAIT_1:
1717                         case TCPS_FIN_WAIT_2:
1718                         case TCPS_CLOSE_WAIT:
1719                                 so->so_error = ECONNRESET;
1720                         close:
1721                                 tp->t_state = TCPS_CLOSED;
1722                                 tcpstat.tcps_drops++;
1723                                 tp = tcp_close(tp);
1724                                 break;
1725
1726                         case TCPS_CLOSING:
1727                         case TCPS_LAST_ACK:
1728                                 tp = tcp_close(tp);
1729                                 break;
1730
1731                         case TCPS_TIME_WAIT:
1732                                 break;
1733                         }
1734                 }
1735                 goto drop;
1736         }
1737
1738         /*
1739          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1740          * and it's less than ts_recent, drop it.
1741          */
1742         if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1743             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1744                 /* Check to see if ts_recent is over 24 days old.  */
1745                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1746                         /*
1747                          * Invalidate ts_recent.  If this segment updates
1748                          * ts_recent, the age will be reset later and ts_recent
1749                          * will get a valid value.  If it does not, setting
1750                          * ts_recent to zero will at least satisfy the
1751                          * requirement that zero be placed in the timestamp
1752                          * echo reply when ts_recent isn't valid.  The
1753                          * age isn't reset until we get a valid ts_recent
1754                          * because we don't want out-of-order segments to be
1755                          * dropped when ts_recent is old.
1756                          */
1757                         tp->ts_recent = 0;
1758                 } else if (tcp_paws_tolerance && tlen != 0 &&
1759                     tp->t_state == TCPS_ESTABLISHED &&
1760                     (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK&&
1761                     !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1762                     th->th_ack == tp->snd_una &&
1763                     tiwin == tp->snd_wnd &&
1764                     TSTMP_GEQ(to.to_tsval + tcp_paws_tolerance, tp->ts_recent)&&
1765                     (th->th_seq == tp->rcv_nxt ||
1766                      (SEQ_GT(th->th_seq, tp->rcv_nxt) && 
1767                       tcp_paws_canreasslast(tp, th, tlen)))) {
1768                         /*
1769                          * This tends to prevent valid new segments from being
1770                          * dropped by the reordered segments sent by the fast
1771                          * retransmission algorithm on the sending side, i.e.
1772                          * the fast retransmitted segment w/ larger timestamp
1773                          * arrives earlier than the previously sent new segments
1774                          * w/ smaller timestamp.
1775                          *
1776                          * If following conditions are met, the segment is
1777                          * accepted:
1778                          * - The segment contains data
1779                          * - The connection is established
1780                          * - The header does not contain important flags
1781                          * - SYN or FIN is not needed
1782                          * - It does not acknowledge new data
1783                          * - Receive window is not changed
1784                          * - The timestamp is within "acceptable" range
1785                          * - The new segment is what we are expecting or
1786                          *   the new segment could be merged w/ the last
1787                          *   pending segment on the reassemble queue
1788                          */
1789                         tcpstat.tcps_pawsaccept++;
1790                         tcpstat.tcps_pawsdrop++;
1791                 } else {
1792                         tcpstat.tcps_rcvduppack++;
1793                         tcpstat.tcps_rcvdupbyte += tlen;
1794                         tcpstat.tcps_pawsdrop++;
1795                         if (tlen)
1796                                 goto dropafterack;
1797                         goto drop;
1798                 }
1799         }
1800
1801         /*
1802          * In the SYN-RECEIVED state, validate that the packet belongs to
1803          * this connection before trimming the data to fit the receive
1804          * window.  Check the sequence number versus IRS since we know
1805          * the sequence numbers haven't wrapped.  This is a partial fix
1806          * for the "LAND" DoS attack.
1807          */
1808         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1809                 rstreason = BANDLIM_RST_OPENPORT;
1810                 goto dropwithreset;
1811         }
1812
1813         todrop = tp->rcv_nxt - th->th_seq;
1814         if (todrop > 0) {
1815                 if (TCP_DO_SACK(tp)) {
1816                         /* Report duplicate segment at head of packet. */
1817                         tp->reportblk.rblk_start = th->th_seq;
1818                         tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1819                             th->th_seq + tlen, thflags);
1820                         if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1821                                 tp->reportblk.rblk_end = tp->rcv_nxt;
1822                         tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_SACKLEFT);
1823                         tp->t_flags |= TF_ACKNOW;
1824                 }
1825                 if (thflags & TH_SYN) {
1826                         thflags &= ~TH_SYN;
1827                         th->th_seq++;
1828                         if (th->th_urp > 1)
1829                                 th->th_urp--;
1830                         else
1831                                 thflags &= ~TH_URG;
1832                         todrop--;
1833                 }
1834                 /*
1835                  * Following if statement from Stevens, vol. 2, p. 960.
1836                  */
1837                 if (todrop > tlen ||
1838                     (todrop == tlen && !(thflags & TH_FIN))) {
1839                         /*
1840                          * Any valid FIN must be to the left of the window.
1841                          * At this point the FIN must be a duplicate or out
1842                          * of sequence; drop it.
1843                          */
1844                         thflags &= ~TH_FIN;
1845
1846                         /*
1847                          * Send an ACK to resynchronize and drop any data.
1848                          * But keep on processing for RST or ACK.
1849                          */
1850                         tp->t_flags |= TF_ACKNOW;
1851                         todrop = tlen;
1852                         tcpstat.tcps_rcvduppack++;
1853                         tcpstat.tcps_rcvdupbyte += todrop;
1854                 } else {
1855                         tcpstat.tcps_rcvpartduppack++;
1856                         tcpstat.tcps_rcvpartdupbyte += todrop;
1857                 }
1858                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1859                 th->th_seq += todrop;
1860                 tlen -= todrop;
1861                 if (th->th_urp > todrop)
1862                         th->th_urp -= todrop;
1863                 else {
1864                         thflags &= ~TH_URG;
1865                         th->th_urp = 0;
1866                 }
1867         }
1868
1869         /*
1870          * If new data are received on a connection after the
1871          * user processes are gone, then RST the other end.
1872          */
1873         if ((so->so_state & SS_NOFDREF) &&
1874             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1875                 tp = tcp_close(tp);
1876                 tcpstat.tcps_rcvafterclose++;
1877                 rstreason = BANDLIM_UNLIMITED;
1878                 goto dropwithreset;
1879         }
1880
1881         /*
1882          * If segment ends after window, drop trailing data
1883          * (and PUSH and FIN); if nothing left, just ACK.
1884          */
1885         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1886         if (todrop > 0) {
1887                 tcpstat.tcps_rcvpackafterwin++;
1888                 if (todrop >= tlen) {
1889                         tcpstat.tcps_rcvbyteafterwin += tlen;
1890                         /*
1891                          * If a new connection request is received
1892                          * while in TIME_WAIT, drop the old connection
1893                          * and start over if the sequence numbers
1894                          * are above the previous ones.
1895                          */
1896                         if (thflags & TH_SYN &&
1897                             tp->t_state == TCPS_TIME_WAIT &&
1898                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1899                                 tp = tcp_close(tp);
1900                                 goto findpcb;
1901                         }
1902                         /*
1903                          * If window is closed can only take segments at
1904                          * window edge, and have to drop data and PUSH from
1905                          * incoming segments.  Continue processing, but
1906                          * remember to ack.  Otherwise, drop segment
1907                          * and ack.
1908                          */
1909                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1910                                 tp->t_flags |= TF_ACKNOW;
1911                                 tcpstat.tcps_rcvwinprobe++;
1912                         } else
1913                                 goto dropafterack;
1914                 } else
1915                         tcpstat.tcps_rcvbyteafterwin += todrop;
1916                 m_adj(m, -todrop);
1917                 tlen -= todrop;
1918                 thflags &= ~(TH_PUSH | TH_FIN);
1919         }
1920
1921         /*
1922          * If last ACK falls within this segment's sequence numbers,
1923          * record its timestamp.
1924          * NOTE:
1925          * 1) That the test incorporates suggestions from the latest
1926          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1927          * 2) That updating only on newer timestamps interferes with
1928          *    our earlier PAWS tests, so this check should be solely
1929          *    predicated on the sequence space of this segment.
1930          * 3) That we modify the segment boundary check to be
1931          *        Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1932          *    instead of RFC1323's
1933          *        Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1934          *    This modified check allows us to overcome RFC1323's
1935          *    limitations as described in Stevens TCP/IP Illustrated
1936          *    Vol. 2 p.869. In such cases, we can still calculate the
1937          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1938          */
1939         if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1940             SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1941                                         + ((thflags & TH_SYN) != 0)
1942                                         + ((thflags & TH_FIN) != 0)))) {
1943                 tp->ts_recent_age = ticks;
1944                 tp->ts_recent = to.to_tsval;
1945         }
1946
1947         /*
1948          * If a SYN is in the window, then this is an
1949          * error and we send an RST and drop the connection.
1950          */
1951         if (thflags & TH_SYN) {
1952                 tp = tcp_drop(tp, ECONNRESET);
1953                 rstreason = BANDLIM_UNLIMITED;
1954                 goto dropwithreset;
1955         }
1956
1957         /*
1958          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1959          * flag is on (half-synchronized state), then queue data for
1960          * later processing; else drop segment and return.
1961          */
1962         if (!(thflags & TH_ACK)) {
1963                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1964                     (tp->t_flags & TF_NEEDSYN))
1965                         goto step6;
1966                 else
1967                         goto drop;
1968         }
1969
1970         /*
1971          * Ack processing.
1972          */
1973         switch (tp->t_state) {
1974         /*
1975          * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1976          * ESTABLISHED state and continue processing.
1977          * The ACK was checked above.
1978          */
1979         case TCPS_SYN_RECEIVED:
1980
1981                 tcpstat.tcps_connects++;
1982                 soisconnected(so);
1983                 /* Do window scaling? */
1984                 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1985                     (TF_RCVD_SCALE | TF_REQ_SCALE))
1986                         tp->rcv_scale = tp->request_r_scale;
1987                 /*
1988                  * Make transitions:
1989                  *      SYN-RECEIVED  -> ESTABLISHED
1990                  *      SYN-RECEIVED* -> FIN-WAIT-1
1991                  */
1992                 tp->t_starttime = ticks;
1993                 if (tp->t_flags & TF_NEEDFIN) {
1994                         tp->t_state = TCPS_FIN_WAIT_1;
1995                         tp->t_flags &= ~TF_NEEDFIN;
1996                 } else {
1997                         tcp_established(tp);
1998                 }
1999                 /*
2000                  * If segment contains data or ACK, will call tcp_reass()
2001                  * later; if not, do so now to pass queued data to user.
2002                  */
2003                 if (tlen == 0 && !(thflags & TH_FIN))
2004                         tcp_reass(tp, NULL, NULL, NULL);
2005                 /* fall into ... */
2006
2007         /*
2008          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2009          * ACKs.  If the ack is in the range
2010          *      tp->snd_una < th->th_ack <= tp->snd_max
2011          * then advance tp->snd_una to th->th_ack and drop
2012          * data from the retransmission queue.  If this ACK reflects
2013          * more up to date window information we update our window information.
2014          */
2015         case TCPS_ESTABLISHED:
2016         case TCPS_FIN_WAIT_1:
2017         case TCPS_FIN_WAIT_2:
2018         case TCPS_CLOSE_WAIT:
2019         case TCPS_CLOSING:
2020         case TCPS_LAST_ACK:
2021         case TCPS_TIME_WAIT:
2022
2023                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2024                         boolean_t maynotdup = FALSE;
2025
2026                         if (TCP_DO_SACK(tp))
2027                                 tcp_sack_update_scoreboard(tp, &to);
2028
2029                         if (tlen != 0 || tiwin != tp->snd_wnd ||
2030                             ((thflags & TH_FIN) && !(tp->t_flags & TF_SAWFIN)))
2031                                 maynotdup = TRUE;
2032
2033                         if (!tcp_callout_active(tp, tp->tt_rexmt) ||
2034                             th->th_ack != tp->snd_una) {
2035                                 if (!maynotdup)
2036                                         tcpstat.tcps_rcvdupack++;
2037                                 tp->t_dupacks = 0;
2038                                 break;
2039                         }
2040
2041 #define DELAY_DUPACK \
2042 do { \
2043         delayed_dupack = TRUE; \
2044         th_dupack = th->th_ack; \
2045         to_flags = to.to_flags; \
2046 } while (0)
2047                         if (maynotdup) {
2048                                 if (!tcp_do_rfc6675 ||
2049                                     !TCP_DO_SACK(tp) ||
2050                                     (to.to_flags &
2051                                      (TOF_SACK | TOF_SACK_REDUNDANT))
2052                                      != TOF_SACK) {
2053                                         tp->t_dupacks = 0;
2054                                 } else {
2055                                         DELAY_DUPACK;
2056                                 }
2057                                 break;
2058                         }
2059                         if ((thflags & TH_FIN) && !(tp->t_flags & TF_QUEDFIN)) {
2060                                 /*
2061                                  * This could happen, if the reassemable
2062                                  * queue overflew or was drained.  Don't
2063                                  * drop this FIN here; defer the duplicated
2064                                  * ACK processing until this FIN gets queued.
2065                                  */
2066                                 DELAY_DUPACK;
2067                                 break;
2068                         }
2069 #undef DELAY_DUPACK
2070
2071                         if (tcp_recv_dupack(tp, th->th_ack, to.to_flags))
2072                                 goto drop;
2073                         else
2074                                 break;
2075                 }
2076
2077                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2078                 tp->t_dupacks = 0;
2079                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2080                         /*
2081                          * Detected optimistic ACK attack.
2082                          * Force slow-start to de-synchronize attack.
2083                          */
2084                         tp->snd_cwnd = tp->t_maxseg;
2085                         tp->snd_wacked = 0;
2086
2087                         tcpstat.tcps_rcvacktoomuch++;
2088                         goto dropafterack;
2089                 }
2090                 /*
2091                  * If we reach this point, ACK is not a duplicate,
2092                  *     i.e., it ACKs something we sent.
2093                  */
2094                 if (tp->t_flags & TF_NEEDSYN) {
2095                         /*
2096                          * T/TCP: Connection was half-synchronized, and our
2097                          * SYN has been ACK'd (so connection is now fully
2098                          * synchronized).  Go to non-starred state,
2099                          * increment snd_una for ACK of SYN, and check if
2100                          * we can do window scaling.
2101                          */
2102                         tp->t_flags &= ~TF_NEEDSYN;
2103                         tp->snd_una++;
2104                         /* Do window scaling? */
2105                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2106                             (TF_RCVD_SCALE | TF_REQ_SCALE))
2107                                 tp->rcv_scale = tp->request_r_scale;
2108                 }
2109
2110 process_ACK:
2111                 acked = th->th_ack - tp->snd_una;
2112                 tcpstat.tcps_rcvackpack++;
2113                 tcpstat.tcps_rcvackbyte += acked;
2114
2115                 if (tcp_do_eifel_detect && acked > 0 &&
2116                     (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2117                     (tp->rxt_flags & TRXT_F_FIRSTACCACK)) {
2118                         /* Eifel detection applicable. */
2119                         if (to.to_tsecr < tp->t_rexmtTS) {
2120                                 ++tcpstat.tcps_eifeldetected;
2121                                 tcp_revert_congestion_state(tp);
2122                                 if (tp->t_rxtshift != 1 ||
2123                                     ticks >= tp->t_badrxtwin)
2124                                         ++tcpstat.tcps_rttcantdetect;
2125                         }
2126                 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2127                         /*
2128                          * If we just performed our first retransmit,
2129                          * and the ACK arrives within our recovery window,
2130                          * then it was a mistake to do the retransmit
2131                          * in the first place.  Recover our original cwnd
2132                          * and ssthresh, and proceed to transmit where we
2133                          * left off.
2134                          */
2135                         tcp_revert_congestion_state(tp);
2136                         ++tcpstat.tcps_rttdetected;
2137                 }
2138
2139                 /*
2140                  * If we have a timestamp reply, update smoothed
2141                  * round trip time.  If no timestamp is present but
2142                  * transmit timer is running and timed sequence
2143                  * number was acked, update smoothed round trip time.
2144                  * Since we now have an rtt measurement, cancel the
2145                  * timer backoff (cf., Phil Karn's retransmit alg.).
2146                  * Recompute the initial retransmit timer.
2147                  *
2148                  * Some machines (certain windows boxes) send broken
2149                  * timestamp replies during the SYN+ACK phase, ignore
2150                  * timestamps of 0.
2151                  */
2152                 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2153                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, th->th_ack);
2154                 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2155                         tcp_xmit_timer(tp, ticks - tp->t_rtttime, th->th_ack);
2156                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2157
2158                 /*
2159                  * If no data (only SYN) was ACK'd,
2160                  *    skip rest of ACK processing.
2161                  */
2162                 if (acked == 0)
2163                         goto step6;
2164
2165                 /* Stop looking for an acceptable ACK since one was received. */
2166                 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK |
2167                     TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT);
2168
2169                 if (acked > so->so_snd.ssb_cc) {
2170                         tp->snd_wnd -= so->so_snd.ssb_cc;
2171                         sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2172                         ourfinisacked = TRUE;
2173                 } else {
2174                         sbdrop(&so->so_snd.sb, acked);
2175                         tp->snd_wnd -= acked;
2176                         ourfinisacked = FALSE;
2177                 }
2178                 sowwakeup(so);
2179
2180                 /*
2181                  * Update window information.
2182                  */
2183                 if (acceptable_window_update(tp, th, tiwin)) {
2184                         /* keep track of pure window updates */
2185                         if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2186                             tiwin > tp->snd_wnd)
2187                                 tcpstat.tcps_rcvwinupd++;
2188                         tp->snd_wnd = tiwin;
2189                         tp->snd_wl1 = th->th_seq;
2190                         tp->snd_wl2 = th->th_ack;
2191                         if (tp->snd_wnd > tp->max_sndwnd)
2192                                 tp->max_sndwnd = tp->snd_wnd;
2193                         needoutput = TRUE;
2194                 }
2195
2196                 tp->snd_una = th->th_ack;
2197                 if (TCP_DO_SACK(tp))
2198                         tcp_sack_update_scoreboard(tp, &to);
2199                 if (IN_FASTRECOVERY(tp)) {
2200                         if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2201                                 EXIT_FASTRECOVERY(tp);
2202                                 needoutput = TRUE;
2203                                 /*
2204                                  * If the congestion window was inflated
2205                                  * to account for the other side's
2206                                  * cached packets, retract it.
2207                                  */
2208                                 if (!TCP_DO_SACK(tp))
2209                                         tp->snd_cwnd = tp->snd_ssthresh;
2210
2211                                 /*
2212                                  * Window inflation should have left us
2213                                  * with approximately snd_ssthresh outstanding
2214                                  * data.  But, in case we would be inclined
2215                                  * to send a burst, better do it using
2216                                  * slow start.
2217                                  */
2218                                 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2219                                            tp->snd_max + 2 * tp->t_maxseg))
2220                                         tp->snd_cwnd =
2221                                             (tp->snd_max - tp->snd_una) +
2222                                             2 * tp->t_maxseg;
2223
2224                                 tp->snd_wacked = 0;
2225                         } else {
2226                                 if (TCP_DO_SACK(tp)) {
2227                                         tp->snd_max_rexmt = tp->snd_max;
2228                                         tcp_sack_rexmt(tp,
2229                                             tp->snd_una == tp->rexmt_high);
2230                                 } else {
2231                                         tcp_newreno_partial_ack(tp, th, acked);
2232                                 }
2233                                 needoutput = FALSE;
2234                         }
2235                 } else {
2236                         /*
2237                          * Open the congestion window.  When in slow-start,
2238                          * open exponentially: maxseg per packet.  Otherwise,
2239                          * open linearly: maxseg per window.
2240                          */
2241                         if (tp->snd_cwnd <= tp->snd_ssthresh) {
2242                                 u_int abc_sslimit =
2243                                     (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2244                                      tp->t_maxseg : 2 * tp->t_maxseg);
2245
2246                                 /* slow-start */
2247                                 tp->snd_cwnd += tcp_do_abc ?
2248                                     min(acked, abc_sslimit) : tp->t_maxseg;
2249                         } else {
2250                                 /* linear increase */
2251                                 tp->snd_wacked += tcp_do_abc ? acked :
2252                                     tp->t_maxseg;
2253                                 if (tp->snd_wacked >= tp->snd_cwnd) {
2254                                         tp->snd_wacked -= tp->snd_cwnd;
2255                                         tp->snd_cwnd += tp->t_maxseg;
2256                                 }
2257                         }
2258                         tp->snd_cwnd = min(tp->snd_cwnd,
2259                                            TCP_MAXWIN << tp->snd_scale);
2260                         tp->snd_recover = th->th_ack - 1;
2261                 }
2262                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2263                         tp->snd_nxt = tp->snd_una;
2264
2265                 /*
2266                  * If all outstanding data is acked, stop retransmit
2267                  * timer and remember to restart (more output or persist).
2268                  * If there is more data to be acked, restart retransmit
2269                  * timer, using current (possibly backed-off) value.
2270                  */
2271                 if (th->th_ack == tp->snd_max) {
2272                         tcp_callout_stop(tp, tp->tt_rexmt);
2273                         needoutput = TRUE;
2274                 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2275                         tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2276                             tcp_timer_rexmt);
2277                 }
2278
2279                 switch (tp->t_state) {
2280                 /*
2281                  * In FIN_WAIT_1 STATE in addition to the processing
2282                  * for the ESTABLISHED state if our FIN is now acknowledged
2283                  * then enter FIN_WAIT_2.
2284                  */
2285                 case TCPS_FIN_WAIT_1:
2286                         if (ourfinisacked) {
2287                                 /*
2288                                  * If we can't receive any more
2289                                  * data, then closing user can proceed.
2290                                  * Starting the timer is contrary to the
2291                                  * specification, but if we don't get a FIN
2292                                  * we'll hang forever.
2293                                  */
2294                                 if (so->so_state & SS_CANTRCVMORE) {
2295                                         soisdisconnected(so);
2296                                         tcp_callout_reset(tp, tp->tt_2msl,
2297                                             tp->t_maxidle, tcp_timer_2msl);
2298                                 }
2299                                 tp->t_state = TCPS_FIN_WAIT_2;
2300                         }
2301                         break;
2302
2303                 /*
2304                  * In CLOSING STATE in addition to the processing for
2305                  * the ESTABLISHED state if the ACK acknowledges our FIN
2306                  * then enter the TIME-WAIT state, otherwise ignore
2307                  * the segment.
2308                  */
2309                 case TCPS_CLOSING:
2310                         if (ourfinisacked) {
2311                                 tp->t_state = TCPS_TIME_WAIT;
2312                                 tcp_canceltimers(tp);
2313                                 tcp_callout_reset(tp, tp->tt_2msl,
2314                                             2 * tcp_rmx_msl(tp),
2315                                             tcp_timer_2msl);
2316                                 soisdisconnected(so);
2317                         }
2318                         break;
2319
2320                 /*
2321                  * In LAST_ACK, we may still be waiting for data to drain
2322                  * and/or to be acked, as well as for the ack of our FIN.
2323                  * If our FIN is now acknowledged, delete the TCB,
2324                  * enter the closed state and return.
2325                  */
2326                 case TCPS_LAST_ACK:
2327                         if (ourfinisacked) {
2328                                 tp = tcp_close(tp);
2329                                 goto drop;
2330                         }
2331                         break;
2332
2333                 /*
2334                  * In TIME_WAIT state the only thing that should arrive
2335                  * is a retransmission of the remote FIN.  Acknowledge
2336                  * it and restart the finack timer.
2337                  */
2338                 case TCPS_TIME_WAIT:
2339                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2340                             tcp_timer_2msl);
2341                         goto dropafterack;
2342                 }
2343         }
2344
2345 step6:
2346         /*
2347          * Update window information.
2348          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2349          */
2350         if ((thflags & TH_ACK) &&
2351             acceptable_window_update(tp, th, tiwin)) {
2352                 /* keep track of pure window updates */
2353                 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2354                     tiwin > tp->snd_wnd)
2355                         tcpstat.tcps_rcvwinupd++;
2356                 tp->snd_wnd = tiwin;
2357                 tp->snd_wl1 = th->th_seq;
2358                 tp->snd_wl2 = th->th_ack;
2359                 if (tp->snd_wnd > tp->max_sndwnd)
2360                         tp->max_sndwnd = tp->snd_wnd;
2361                 needoutput = TRUE;
2362         }
2363
2364         /*
2365          * Process segments with URG.
2366          */
2367         if ((thflags & TH_URG) && th->th_urp &&
2368             !TCPS_HAVERCVDFIN(tp->t_state)) {
2369                 /*
2370                  * This is a kludge, but if we receive and accept
2371                  * random urgent pointers, we'll crash in
2372                  * soreceive.  It's hard to imagine someone
2373                  * actually wanting to send this much urgent data.
2374                  */
2375                 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2376                         th->th_urp = 0;                 /* XXX */
2377                         thflags &= ~TH_URG;             /* XXX */
2378                         goto dodata;                    /* XXX */
2379                 }
2380                 /*
2381                  * If this segment advances the known urgent pointer,
2382                  * then mark the data stream.  This should not happen
2383                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2384                  * a FIN has been received from the remote side.
2385                  * In these states we ignore the URG.
2386                  *
2387                  * According to RFC961 (Assigned Protocols),
2388                  * the urgent pointer points to the last octet
2389                  * of urgent data.  We continue, however,
2390                  * to consider it to indicate the first octet
2391                  * of data past the urgent section as the original
2392                  * spec states (in one of two places).
2393                  */
2394                 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2395                         tp->rcv_up = th->th_seq + th->th_urp;
2396                         so->so_oobmark = so->so_rcv.ssb_cc +
2397                             (tp->rcv_up - tp->rcv_nxt) - 1;
2398                         if (so->so_oobmark == 0)
2399                                 sosetstate(so, SS_RCVATMARK);
2400                         sohasoutofband(so);
2401                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2402                 }
2403                 /*
2404                  * Remove out of band data so doesn't get presented to user.
2405                  * This can happen independent of advancing the URG pointer,
2406                  * but if two URG's are pending at once, some out-of-band
2407                  * data may creep in... ick.
2408                  */
2409                 if (th->th_urp <= (u_long)tlen &&
2410                     !(so->so_options & SO_OOBINLINE)) {
2411                         /* hdr drop is delayed */
2412                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2413                 }
2414         } else {
2415                 /*
2416                  * If no out of band data is expected,
2417                  * pull receive urgent pointer along
2418                  * with the receive window.
2419                  */
2420                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2421                         tp->rcv_up = tp->rcv_nxt;
2422         }
2423
2424 dodata:                                                 /* XXX */
2425         /*
2426          * Process the segment text, merging it into the TCP sequencing queue,
2427          * and arranging for acknowledgment of receipt if necessary.
2428          * This process logically involves adjusting tp->rcv_wnd as data
2429          * is presented to the user (this happens in tcp_usrreq.c,
2430          * case PRU_RCVD).  If a FIN has already been received on this
2431          * connection then we just ignore the text.
2432          */
2433         if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2434                 if (thflags & TH_FIN)
2435                         tp->t_flags |= TF_SAWFIN;
2436                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2437                 /*
2438                  * Insert segment which includes th into TCP reassembly queue
2439                  * with control block tp.  Set thflags to whether reassembly now
2440                  * includes a segment with FIN.  This handles the common case
2441                  * inline (segment is the next to be received on an established
2442                  * connection, and the queue is empty), avoiding linkage into
2443                  * and removal from the queue and repetition of various
2444                  * conversions.
2445                  * Set DELACK for segments received in order, but ack
2446                  * immediately when segments are out of order (so
2447                  * fast retransmit can work).
2448                  */
2449                 if (th->th_seq == tp->rcv_nxt &&
2450                     TAILQ_EMPTY(&tp->t_segq) &&
2451                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2452                         if (thflags & TH_FIN)
2453                                 tp->t_flags |= TF_QUEDFIN;
2454                         if (DELAY_ACK(tp)) {
2455                                 tcp_callout_reset(tp, tp->tt_delack,
2456                                     tcp_delacktime, tcp_timer_delack);
2457                         } else {
2458                                 tp->t_flags |= TF_ACKNOW;
2459                         }
2460                         tp->rcv_nxt += tlen;
2461                         thflags = th->th_flags & TH_FIN;
2462                         tcpstat.tcps_rcvpack++;
2463                         tcpstat.tcps_rcvbyte += tlen;
2464                         ND6_HINT(tp);
2465                         if (so->so_state & SS_CANTRCVMORE) {
2466                                 m_freem(m);
2467                         } else {
2468                                 lwkt_gettoken(&so->so_rcv.ssb_token);
2469                                 ssb_appendstream(&so->so_rcv, m);
2470                                 lwkt_reltoken(&so->so_rcv.ssb_token);
2471                         }
2472                         sorwakeup(so);
2473                 } else {
2474                         if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
2475                                 /* Initialize SACK report block. */
2476                                 tp->reportblk.rblk_start = th->th_seq;
2477                                 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2478                                     th->th_seq + tlen, thflags);
2479                         }
2480                         thflags = tcp_reass(tp, th, &tlen, m);
2481                         tp->t_flags |= TF_ACKNOW;
2482                 }
2483
2484                 /*
2485                  * Note the amount of data that peer has sent into
2486                  * our window, in order to estimate the sender's
2487                  * buffer size.
2488                  */
2489                 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2490         } else {
2491                 m_freem(m);
2492                 thflags &= ~TH_FIN;
2493         }
2494
2495         /*
2496          * If FIN is received ACK the FIN and let the user know
2497          * that the connection is closing.
2498          */
2499         if (thflags & TH_FIN) {
2500                 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2501                         socantrcvmore(so);
2502                         /*
2503                          * If connection is half-synchronized
2504                          * (ie NEEDSYN flag on) then delay ACK,
2505                          * so it may be piggybacked when SYN is sent.
2506                          * Otherwise, since we received a FIN then no
2507                          * more input can be expected, send ACK now.
2508                          */
2509                         if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2510                                 tcp_callout_reset(tp, tp->tt_delack,
2511                                     tcp_delacktime, tcp_timer_delack);
2512                         } else {
2513                                 tp->t_flags |= TF_ACKNOW;
2514                         }
2515                         tp->rcv_nxt++;
2516                 }
2517
2518                 switch (tp->t_state) {
2519                 /*
2520                  * In SYN_RECEIVED and ESTABLISHED STATES
2521                  * enter the CLOSE_WAIT state.
2522                  */
2523                 case TCPS_SYN_RECEIVED:
2524                         tp->t_starttime = ticks;
2525                         /*FALLTHROUGH*/
2526                 case TCPS_ESTABLISHED:
2527                         tp->t_state = TCPS_CLOSE_WAIT;
2528                         break;
2529
2530                 /*
2531                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2532                  * enter the CLOSING state.
2533                  */
2534                 case TCPS_FIN_WAIT_1:
2535                         tp->t_state = TCPS_CLOSING;
2536                         break;
2537
2538                 /*
2539                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2540                  * starting the time-wait timer, turning off the other
2541                  * standard timers.
2542                  */
2543                 case TCPS_FIN_WAIT_2:
2544                         tp->t_state = TCPS_TIME_WAIT;
2545                         tcp_canceltimers(tp);
2546                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2547                                     tcp_timer_2msl);
2548                         soisdisconnected(so);
2549                         break;
2550
2551                 /*
2552                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2553                  */
2554                 case TCPS_TIME_WAIT:
2555                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2556                             tcp_timer_2msl);
2557                         break;
2558                 }
2559         }
2560
2561 #ifdef TCPDEBUG
2562         if (so->so_options & SO_DEBUG)
2563                 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2564 #endif
2565
2566         /*
2567          * Delayed duplicated ACK processing
2568          */
2569         if (delayed_dupack && tcp_recv_dupack(tp, th_dupack, to_flags))
2570                 needoutput = FALSE;
2571
2572         /*
2573          * Return any desired output.
2574          */
2575         if (needoutput || (tp->t_flags & TF_ACKNOW))
2576                 tcp_output(tp);
2577         tcp_sack_report_cleanup(tp);
2578         return(IPPROTO_DONE);
2579
2580 dropafterack:
2581         /*
2582          * Generate an ACK dropping incoming segment if it occupies
2583          * sequence space, where the ACK reflects our state.
2584          *
2585          * We can now skip the test for the RST flag since all
2586          * paths to this code happen after packets containing
2587          * RST have been dropped.
2588          *
2589          * In the SYN-RECEIVED state, don't send an ACK unless the
2590          * segment we received passes the SYN-RECEIVED ACK test.
2591          * If it fails send a RST.  This breaks the loop in the
2592          * "LAND" DoS attack, and also prevents an ACK storm
2593          * between two listening ports that have been sent forged
2594          * SYN segments, each with the source address of the other.
2595          */
2596         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2597             (SEQ_GT(tp->snd_una, th->th_ack) ||
2598              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2599                 rstreason = BANDLIM_RST_OPENPORT;
2600                 goto dropwithreset;
2601         }
2602 #ifdef TCPDEBUG
2603         if (so->so_options & SO_DEBUG)
2604                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2605 #endif
2606         m_freem(m);
2607         tp->t_flags |= TF_ACKNOW;
2608         tcp_output(tp);
2609         tcp_sack_report_cleanup(tp);
2610         return(IPPROTO_DONE);
2611
2612 dropwithreset:
2613         /*
2614          * Generate a RST, dropping incoming segment.
2615          * Make ACK acceptable to originator of segment.
2616          * Don't bother to respond if destination was broadcast/multicast.
2617          */
2618         if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2619                 goto drop;
2620         if (isipv6) {
2621                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2622                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2623                         goto drop;
2624         } else {
2625                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2626                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2627                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2628                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2629                         goto drop;
2630         }
2631         /* IPv6 anycast check is done at tcp6_input() */
2632
2633         /*
2634          * Perform bandwidth limiting.
2635          */
2636 #ifdef ICMP_BANDLIM
2637         if (badport_bandlim(rstreason) < 0)
2638                 goto drop;
2639 #endif
2640
2641 #ifdef TCPDEBUG
2642         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2643                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2644 #endif
2645         if (thflags & TH_ACK)
2646                 /* mtod() below is safe as long as hdr dropping is delayed */
2647                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2648                             TH_RST);
2649         else {
2650                 if (thflags & TH_SYN)
2651                         tlen++;
2652                 /* mtod() below is safe as long as hdr dropping is delayed */
2653                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2654                             (tcp_seq)0, TH_RST | TH_ACK);
2655         }
2656         if (tp != NULL)
2657                 tcp_sack_report_cleanup(tp);
2658         return(IPPROTO_DONE);
2659
2660 drop:
2661         /*
2662          * Drop space held by incoming segment and return.
2663          */
2664 #ifdef TCPDEBUG
2665         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2666                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2667 #endif
2668         m_freem(m);
2669         if (tp != NULL)
2670                 tcp_sack_report_cleanup(tp);
2671         return(IPPROTO_DONE);
2672 }
2673
2674 /*
2675  * Parse TCP options and place in tcpopt.
2676  */
2677 static void
2678 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn,
2679     tcp_seq ack)
2680 {
2681         int opt, optlen, i;
2682
2683         to->to_flags = 0;
2684         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2685                 opt = cp[0];
2686                 if (opt == TCPOPT_EOL)
2687                         break;
2688                 if (opt == TCPOPT_NOP)
2689                         optlen = 1;
2690                 else {
2691                         if (cnt < 2)
2692                                 break;
2693                         optlen = cp[1];
2694                         if (optlen < 2 || optlen > cnt)
2695                                 break;
2696                 }
2697                 switch (opt) {
2698                 case TCPOPT_MAXSEG:
2699                         if (optlen != TCPOLEN_MAXSEG)
2700                                 continue;
2701                         if (!is_syn)
2702                                 continue;
2703                         to->to_flags |= TOF_MSS;
2704                         bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2705                         to->to_mss = ntohs(to->to_mss);
2706                         break;
2707                 case TCPOPT_WINDOW:
2708                         if (optlen != TCPOLEN_WINDOW)
2709                                 continue;
2710                         if (!is_syn)
2711                                 continue;
2712                         to->to_flags |= TOF_SCALE;
2713                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2714                         break;
2715                 case TCPOPT_TIMESTAMP:
2716                         if (optlen != TCPOLEN_TIMESTAMP)
2717                                 continue;
2718                         to->to_flags |= TOF_TS;
2719                         bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2720                         to->to_tsval = ntohl(to->to_tsval);
2721                         bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2722                         to->to_tsecr = ntohl(to->to_tsecr);
2723                         /*
2724                          * If echoed timestamp is later than the current time,
2725                          * fall back to non RFC1323 RTT calculation.
2726                          */
2727                         if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2728                                 to->to_tsecr = 0;
2729                         break;
2730                 case TCPOPT_SACK_PERMITTED:
2731                         if (optlen != TCPOLEN_SACK_PERMITTED)
2732                                 continue;
2733                         if (!is_syn)
2734                                 continue;
2735                         to->to_flags |= TOF_SACK_PERMITTED;
2736                         break;
2737                 case TCPOPT_SACK:
2738                         if ((optlen - 2) & 0x07)        /* not multiple of 8 */
2739                                 continue;
2740                         to->to_nsackblocks = (optlen - 2) / 8;
2741                         to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2742                         to->to_flags |= TOF_SACK;
2743                         for (i = 0; i < to->to_nsackblocks; i++) {
2744                                 struct raw_sackblock *r = &to->to_sackblocks[i];
2745
2746                                 r->rblk_start = ntohl(r->rblk_start);
2747                                 r->rblk_end = ntohl(r->rblk_end);
2748
2749                                 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2750                                         /*
2751                                          * Invalid SACK block; discard all
2752                                          * SACK blocks
2753                                          */
2754                                         tcpstat.tcps_rcvbadsackopt++;
2755                                         to->to_nsackblocks = 0;
2756                                         to->to_sackblocks = NULL;
2757                                         to->to_flags &= ~TOF_SACK;
2758                                         break;
2759                                 }
2760                         }
2761                         if ((to->to_flags & TOF_SACK) &&
2762                             tcp_sack_ndsack_blocks(to->to_sackblocks,
2763                             to->to_nsackblocks, ack))
2764                                 to->to_flags |= TOF_DSACK;
2765                         break;
2766 #ifdef TCP_SIGNATURE
2767                 /*
2768                  * XXX In order to reply to a host which has set the
2769                  * TCP_SIGNATURE option in its initial SYN, we have to
2770                  * record the fact that the option was observed here
2771                  * for the syncache code to perform the correct response.
2772                  */
2773                 case TCPOPT_SIGNATURE:
2774                         if (optlen != TCPOLEN_SIGNATURE)
2775                                 continue;
2776                         to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2777                         break;
2778 #endif /* TCP_SIGNATURE */
2779                 default:
2780                         continue;
2781                 }
2782         }
2783 }
2784
2785 /*
2786  * Pull out of band byte out of a segment so
2787  * it doesn't appear in the user's data queue.
2788  * It is still reflected in the segment length for
2789  * sequencing purposes.
2790  * "off" is the delayed to be dropped hdrlen.
2791  */
2792 static void
2793 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2794 {
2795         int cnt = off + th->th_urp - 1;
2796
2797         while (cnt >= 0) {
2798                 if (m->m_len > cnt) {
2799                         char *cp = mtod(m, caddr_t) + cnt;
2800                         struct tcpcb *tp = sototcpcb(so);
2801
2802                         tp->t_iobc = *cp;
2803                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2804                         bcopy(cp + 1, cp, m->m_len - cnt - 1);
2805                         m->m_len--;
2806                         if (m->m_flags & M_PKTHDR)
2807                                 m->m_pkthdr.len--;
2808                         return;
2809                 }
2810                 cnt -= m->m_len;
2811                 m = m->m_next;
2812                 if (m == NULL)
2813                         break;
2814         }
2815         panic("tcp_pulloutofband");
2816 }
2817
2818 /*
2819  * Collect new round-trip time estimate
2820  * and update averages and current timeout.
2821  */
2822 static void
2823 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack)
2824 {
2825         int rebaserto = 0;
2826
2827         tcpstat.tcps_rttupdated++;
2828         tp->t_rttupdated++;
2829         if ((tp->rxt_flags & TRXT_F_REBASERTO) &&
2830             SEQ_GT(ack, tp->snd_max_prev)) {
2831 #ifdef DEBUG_EIFEL_RESPONSE
2832                 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ",
2833                     tp->t_srtt_prev, tp->t_rttvar_prev,
2834                     tp->t_srtt, tp->t_rttvar);
2835 #endif
2836
2837                 tcpstat.tcps_eifelresponse++;
2838                 rebaserto = 1;
2839                 tp->rxt_flags &= ~TRXT_F_REBASERTO;
2840                 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT));
2841                 tp->t_rttvar = max(tp->t_rttvar_prev,
2842                     (rtt << (TCP_RTTVAR_SHIFT - 1)));
2843                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2844                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2845
2846 #ifdef DEBUG_EIFEL_RESPONSE
2847                 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar);
2848 #endif
2849         } else if (tp->t_srtt != 0) {
2850                 int delta;
2851
2852                 /*
2853                  * srtt is stored as fixed point with 5 bits after the
2854                  * binary point (i.e., scaled by 8).  The following magic
2855                  * is equivalent to the smoothing algorithm in rfc793 with
2856                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2857                  * point).  Adjust rtt to origin 0.
2858                  */
2859                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2860                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2861
2862                 if ((tp->t_srtt += delta) <= 0)
2863                         tp->t_srtt = 1;
2864
2865                 /*
2866                  * We accumulate a smoothed rtt variance (actually, a
2867                  * smoothed mean difference), then set the retransmit
2868                  * timer to smoothed rtt + 4 times the smoothed variance.
2869                  * rttvar is stored as fixed point with 4 bits after the
2870                  * binary point (scaled by 16).  The following is
2871                  * equivalent to rfc793 smoothing with an alpha of .75
2872                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2873                  * rfc793's wired-in beta.
2874                  */
2875                 if (delta < 0)
2876                         delta = -delta;
2877                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2878                 if ((tp->t_rttvar += delta) <= 0)
2879                         tp->t_rttvar = 1;
2880                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2881                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2882         } else {
2883                 /*
2884                  * No rtt measurement yet - use the unsmoothed rtt.
2885                  * Set the variance to half the rtt (so our first
2886                  * retransmit happens at 3*rtt).
2887                  */
2888                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2889                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2890                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2891         }
2892         tp->t_rtttime = 0;
2893         tp->t_rxtshift = 0;
2894
2895 #ifdef DEBUG_EIFEL_RESPONSE
2896         if (rebaserto) {
2897                 kprintf("| rxtcur prev %d, old %d, ",
2898                     tp->t_rxtcur_prev, tp->t_rxtcur);
2899         }
2900 #endif
2901
2902         /*
2903          * the retransmit should happen at rtt + 4 * rttvar.
2904          * Because of the way we do the smoothing, srtt and rttvar
2905          * will each average +1/2 tick of bias.  When we compute
2906          * the retransmit timer, we want 1/2 tick of rounding and
2907          * 1 extra tick because of +-1/2 tick uncertainty in the
2908          * firing of the timer.  The bias will give us exactly the
2909          * 1.5 tick we need.  But, because the bias is
2910          * statistical, we have to test that we don't drop below
2911          * the minimum feasible timer (which is 2 ticks).
2912          */
2913         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2914                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2915
2916         if (rebaserto) {
2917                 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) {
2918                         /*
2919                          * RFC4015 requires that the new RTO is at least
2920                          * 2*G (tcp_eifel_rtoinc) greater then the RTO
2921                          * (t_rxtcur_prev) when the spurious retransmit
2922                          * timeout happens.
2923                          *
2924                          * The above condition could be true, if the SRTT
2925                          * and RTTVAR used to calculate t_rxtcur_prev
2926                          * resulted in a value less than t_rttmin.  So
2927                          * simply increasing SRTT by tcp_eifel_rtoinc when
2928                          * preparing for the Eifel response could not ensure
2929                          * that the new RTO will be tcp_eifel_rtoinc greater
2930                          * t_rxtcur_prev.
2931                          */
2932                         tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc;
2933                 }
2934 #ifdef DEBUG_EIFEL_RESPONSE
2935                 kprintf("new %d\n", tp->t_rxtcur);
2936 #endif
2937         }
2938
2939         /*
2940          * We received an ack for a packet that wasn't retransmitted;
2941          * it is probably safe to discard any error indications we've
2942          * received recently.  This isn't quite right, but close enough
2943          * for now (a route might have failed after we sent a segment,
2944          * and the return path might not be symmetrical).
2945          */
2946         tp->t_softerror = 0;
2947 }
2948
2949 /*
2950  * Determine a reasonable value for maxseg size.
2951  * If the route is known, check route for mtu.
2952  * If none, use an mss that can be handled on the outgoing
2953  * interface without forcing IP to fragment; if bigger than
2954  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2955  * to utilize large mbufs.  If no route is found, route has no mtu,
2956  * or the destination isn't local, use a default, hopefully conservative
2957  * size (usually 512 or the default IP max size, but no more than the mtu
2958  * of the interface), as we can't discover anything about intervening
2959  * gateways or networks.  We also initialize the congestion/slow start
2960  * window to be a single segment if the destination isn't local.
2961  * While looking at the routing entry, we also initialize other path-dependent
2962  * parameters from pre-set or cached values in the routing entry.
2963  *
2964  * Also take into account the space needed for options that we
2965  * send regularly.  Make maxseg shorter by that amount to assure
2966  * that we can send maxseg amount of data even when the options
2967  * are present.  Store the upper limit of the length of options plus
2968  * data in maxopd.
2969  *
2970  * NOTE that this routine is only called when we process an incoming
2971  * segment, for outgoing segments only tcp_mssopt is called.
2972  */
2973 void
2974 tcp_mss(struct tcpcb *tp, int offer)
2975 {
2976         struct rtentry *rt;
2977         struct ifnet *ifp;
2978         int rtt, mss;
2979         u_long bufsize;
2980         struct inpcb *inp = tp->t_inpcb;
2981         struct socket *so;
2982 #ifdef INET6
2983         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2984         size_t min_protoh = isipv6 ?
2985                             sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2986                             sizeof(struct tcpiphdr);
2987 #else
2988         const boolean_t isipv6 = FALSE;
2989         const size_t min_protoh = sizeof(struct tcpiphdr);
2990 #endif
2991
2992         if (isipv6)
2993                 rt = tcp_rtlookup6(&inp->inp_inc);
2994         else
2995                 rt = tcp_rtlookup(&inp->inp_inc);
2996         if (rt == NULL) {
2997                 tp->t_maxopd = tp->t_maxseg =
2998                     (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2999                 return;
3000         }
3001         ifp = rt->rt_ifp;
3002         so = inp->inp_socket;
3003
3004         /*
3005          * Offer == 0 means that there was no MSS on the SYN segment,
3006          * in this case we use either the interface mtu or tcp_mssdflt.
3007          *
3008          * An offer which is too large will be cut down later.
3009          */
3010         if (offer == 0) {
3011                 if (isipv6) {
3012                         if (in6_localaddr(&inp->in6p_faddr)) {
3013                                 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
3014                                         min_protoh;
3015                         } else {
3016                                 offer = tcp_v6mssdflt;
3017                         }
3018                 } else {
3019                         if (in_localaddr(inp->inp_faddr))
3020                                 offer = ifp->if_mtu - min_protoh;
3021                         else
3022                                 offer = tcp_mssdflt;
3023                 }
3024         }
3025
3026         /*
3027          * Prevent DoS attack with too small MSS. Round up
3028          * to at least minmss.
3029          *
3030          * Sanity check: make sure that maxopd will be large
3031          * enough to allow some data on segments even is the
3032          * all the option space is used (40bytes).  Otherwise
3033          * funny things may happen in tcp_output.
3034          */
3035         offer = max(offer, tcp_minmss);
3036         offer = max(offer, 64);
3037
3038         rt->rt_rmx.rmx_mssopt = offer;
3039
3040         /*
3041          * While we're here, check if there's an initial rtt
3042          * or rttvar.  Convert from the route-table units
3043          * to scaled multiples of the slow timeout timer.
3044          */
3045         if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3046                 /*
3047                  * XXX the lock bit for RTT indicates that the value
3048                  * is also a minimum value; this is subject to time.
3049                  */
3050                 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3051                         tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3052                 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3053                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3054                 tcpstat.tcps_usedrtt++;
3055                 if (rt->rt_rmx.rmx_rttvar) {
3056                         tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3057                             (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3058                         tcpstat.tcps_usedrttvar++;
3059                 } else {
3060                         /* default variation is +- 1 rtt */
3061                         tp->t_rttvar =
3062                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3063                 }
3064                 TCPT_RANGESET(tp->t_rxtcur,
3065                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3066                               tp->t_rttmin, TCPTV_REXMTMAX);
3067         }
3068
3069         /*
3070          * if there's an mtu associated with the route, use it
3071          * else, use the link mtu.  Take the smaller of mss or offer
3072          * as our final mss.
3073          */
3074         if (rt->rt_rmx.rmx_mtu) {
3075                 mss = rt->rt_rmx.rmx_mtu - min_protoh;
3076         } else {
3077                 if (isipv6)
3078                         mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
3079                 else
3080                         mss = ifp->if_mtu - min_protoh;
3081         }
3082         mss = min(mss, offer);
3083
3084         /*
3085          * maxopd stores the maximum length of data AND options
3086          * in a segment; maxseg is the amount of data in a normal
3087          * segment.  We need to store this value (maxopd) apart
3088          * from maxseg, because now every segment carries options
3089          * and thus we normally have somewhat less data in segments.
3090          */
3091         tp->t_maxopd = mss;
3092
3093         if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3094             ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3095                 mss -= TCPOLEN_TSTAMP_APPA;
3096
3097 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
3098         if (mss > MCLBYTES)
3099                 mss &= ~(MCLBYTES-1);
3100 #else
3101         if (mss > MCLBYTES)
3102                 mss = mss / MCLBYTES * MCLBYTES;
3103 #endif
3104         /*
3105          * If there's a pipesize, change the socket buffer
3106          * to that size.  Make the socket buffers an integral
3107          * number of mss units; if the mss is larger than
3108          * the socket buffer, decrease the mss.
3109          */
3110 #ifdef RTV_SPIPE
3111         if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3112 #endif
3113                 bufsize = so->so_snd.ssb_hiwat;
3114         if (bufsize < mss)
3115                 mss = bufsize;
3116         else {
3117                 bufsize = roundup(bufsize, mss);
3118                 if (bufsize > sb_max)
3119                         bufsize = sb_max;
3120                 if (bufsize > so->so_snd.ssb_hiwat)
3121                         ssb_reserve(&so->so_snd, bufsize, so, NULL);
3122         }
3123         tp->t_maxseg = mss;
3124
3125 #ifdef RTV_RPIPE
3126         if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3127 #endif
3128                 bufsize = so->so_rcv.ssb_hiwat;
3129         if (bufsize > mss) {
3130                 bufsize = roundup(bufsize, mss);
3131                 if (bufsize > sb_max)
3132                         bufsize = sb_max;
3133                 if (bufsize > so->so_rcv.ssb_hiwat) {
3134                         lwkt_gettoken(&so->so_rcv.ssb_token);
3135                         ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3136                         lwkt_reltoken(&so->so_rcv.ssb_token);
3137                 }
3138         }
3139
3140         /*
3141          * Set the slow-start flight size
3142          *
3143          * NOTE: t_maxseg must have been configured!
3144          */
3145         tp->snd_cwnd = tcp_initial_window(tp);
3146
3147         if (rt->rt_rmx.rmx_ssthresh) {
3148                 /*
3149                  * There's some sort of gateway or interface
3150                  * buffer limit on the path.  Use this to set
3151                  * the slow start threshhold, but set the
3152                  * threshold to no less than 2*mss.
3153                  */
3154                 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3155                 tcpstat.tcps_usedssthresh++;
3156         }
3157 }
3158
3159 /*
3160  * Determine the MSS option to send on an outgoing SYN.
3161  */
3162 int
3163 tcp_mssopt(struct tcpcb *tp)
3164 {
3165         struct rtentry *rt;
3166 #ifdef INET6
3167         boolean_t isipv6 =
3168             ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3169         int min_protoh = isipv6 ?
3170                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3171                              sizeof(struct tcpiphdr);
3172 #else
3173         const boolean_t isipv6 = FALSE;
3174         const size_t min_protoh = sizeof(struct tcpiphdr);
3175 #endif
3176
3177         if (isipv6)
3178                 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3179         else
3180                 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3181         if (rt == NULL)
3182                 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3183
3184         return (rt->rt_ifp->if_mtu - min_protoh);
3185 }
3186
3187 /*
3188  * When a partial ack arrives, force the retransmission of the
3189  * next unacknowledged segment.  Do not exit Fast Recovery.
3190  *
3191  * Implement the Slow-but-Steady variant of NewReno by restarting the
3192  * the retransmission timer.  Turn it off here so it can be restarted
3193  * later in tcp_output().
3194  */
3195 static void
3196 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3197 {
3198         tcp_seq old_snd_nxt = tp->snd_nxt;
3199         u_long ocwnd = tp->snd_cwnd;
3200
3201         tcp_callout_stop(tp, tp->tt_rexmt);
3202         tp->t_rtttime = 0;
3203         tp->snd_nxt = th->th_ack;
3204         /* Set snd_cwnd to one segment beyond acknowledged offset. */
3205         tp->snd_cwnd = tp->t_maxseg;
3206         tp->t_flags |= TF_ACKNOW;
3207         tcp_output(tp);
3208         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3209                 tp->snd_nxt = old_snd_nxt;
3210         /* partial window deflation */
3211         if (ocwnd > acked)
3212                 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3213         else
3214                 tp->snd_cwnd = tp->t_maxseg;
3215 }
3216
3217 /*
3218  * In contrast to the Slow-but-Steady NewReno variant,
3219  * we do not reset the retransmission timer for SACK retransmissions,
3220  * except when retransmitting snd_una.
3221  */
3222 static void
3223 tcp_sack_rexmt(struct tcpcb *tp, boolean_t force)
3224 {
3225         tcp_seq old_snd_nxt = tp->snd_nxt;
3226         u_long ocwnd = tp->snd_cwnd;
3227         uint32_t pipe;
3228         int nseg = 0;           /* consecutive new segments */
3229         int nseg_rexmt = 0;     /* retransmitted segments */
3230         int maxrexmt = 0;
3231 #define MAXBURST 4              /* limit burst of new packets on partial ack */
3232
3233         if (force) {
3234                 uint32_t unsacked = tcp_sack_first_unsacked_len(tp);
3235
3236                 /*
3237                  * Try to fill the first hole in the receiver's
3238                  * reassemble queue.
3239                  */
3240                 maxrexmt = howmany(unsacked, tp->t_maxseg);
3241                 if (maxrexmt > tcp_force_sackrxt)
3242                         maxrexmt = tcp_force_sackrxt;
3243         }
3244
3245         tp->t_rtttime = 0;
3246         pipe = tcp_sack_compute_pipe(tp);
3247         while (((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg
3248                 || (force && nseg_rexmt < maxrexmt && nseg == 0)) &&
3249             (!tcp_do_smartsack || nseg < MAXBURST)) {
3250                 tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3251                 uint32_t sent, seglen;
3252                 boolean_t rescue;
3253                 int error;
3254
3255                 old_rexmt_high = tp->rexmt_high;
3256                 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
3257                         tp->rexmt_high = old_rexmt_high;
3258                         break;
3259                 }
3260
3261                 /*
3262                  * If the next tranmission is a rescue retranmission,
3263                  * we check whether we have already sent some data
3264                  * (either new segments or retransmitted segments)
3265                  * into the the network or not.  Since the idea of rescue
3266                  * retransmission is to sustain ACK clock, as long as
3267                  * some segments are in the network, ACK clock will be
3268                  * kept ticking.
3269                  */
3270                 if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3271                         tp->rexmt_high = old_rexmt_high;
3272                         break;
3273                 }
3274
3275                 if (nextrexmt == tp->snd_max)
3276                         ++nseg;
3277                 else
3278                         ++nseg_rexmt;
3279                 tp->snd_nxt = nextrexmt;
3280                 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3281                 old_snd_max = tp->snd_max;
3282                 if (nextrexmt == tp->snd_una)
3283                         tcp_callout_stop(tp, tp->tt_rexmt);
3284                 tp->t_flags |= TF_XMITNOW;
3285                 error = tcp_output(tp);
3286                 if (error != 0) {
3287                         tp->rexmt_high = old_rexmt_high;
3288                         break;
3289                 }
3290                 sent = tp->snd_nxt - nextrexmt;
3291                 if (sent <= 0) {
3292                         tp->rexmt_high = old_rexmt_high;
3293                         break;
3294                 }
3295                 pipe += sent;
3296                 tcpstat.tcps_sndsackpack++;
3297                 tcpstat.tcps_sndsackbyte += sent;
3298
3299                 if (rescue) {
3300                         tcpstat.tcps_sackrescue++;
3301                         tp->rexmt_rescue = tp->snd_nxt;
3302                         tp->sack_flags |= TSACK_F_SACKRESCUED;
3303                         break;
3304                 }
3305                 if (SEQ_LT(nextrexmt, old_snd_max) &&
3306                     SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
3307                         tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3308                         if (tcp_aggressive_rescuesack &&
3309                             (tp->sack_flags & TSACK_F_SACKRESCUED) &&
3310                             SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3311                                 /* Drag RescueRxt along with HighRxt */
3312                                 tp->rexmt_rescue = tp->rexmt_high;
3313                         }
3314                 }
3315         }
3316         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3317                 tp->snd_nxt = old_snd_nxt;
3318         tp->snd_cwnd = ocwnd;
3319 }
3320
3321 /*
3322  * Return TRUE, if some new segments are sent
3323  */
3324 static boolean_t
3325 tcp_sack_limitedxmit(struct tcpcb *tp)
3326 {
3327         tcp_seq oldsndnxt = tp->snd_nxt;
3328         tcp_seq oldsndmax = tp->snd_max;
3329         u_long ocwnd = tp->snd_cwnd;
3330         uint32_t pipe, sent;
3331         boolean_t ret = FALSE;
3332         tcp_seq_diff_t cwnd_left;
3333         tcp_seq next;
3334
3335         tp->rexmt_high = tp->snd_una - 1;
3336         pipe = tcp_sack_compute_pipe(tp);
3337         cwnd_left = (tcp_seq_diff_t)(ocwnd - pipe);
3338         if (cwnd_left < (tcp_seq_diff_t)tp->t_maxseg)
3339                 return FALSE;
3340
3341         next = tp->snd_nxt = tp->snd_max;
3342         tp->snd_cwnd = tp->snd_nxt - tp->snd_una +
3343             rounddown(cwnd_left, tp->t_maxseg);
3344
3345         tp->t_flags |= TF_XMITNOW;
3346         tcp_output(tp);
3347
3348         sent = tp->snd_nxt - next;
3349         if (sent > 0) {
3350                 tcpstat.tcps_sndlimited += howmany(sent, tp->t_maxseg);
3351                 ret = TRUE;
3352         }
3353
3354         if (SEQ_LT(oldsndnxt, oldsndmax)) {
3355                 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3356                     ("snd_una moved in other threads"));
3357                 tp->snd_nxt = oldsndnxt;
3358         }
3359         tp->snd_cwnd = ocwnd;
3360
3361         if (ret && TCP_DO_NCR(tp))
3362                 tcp_ncr_update_rxtthresh(tp);
3363
3364         return ret;
3365 }
3366
3367 /*
3368  * Reset idle time and keep-alive timer, typically called when a valid
3369  * tcp packet is received but may also be called when FASTKEEP is set
3370  * to prevent the previous long-timeout from calculating to a drop.
3371  *
3372  * Only update t_rcvtime for non-SYN packets.
3373  *
3374  * Handle the case where one side thinks the connection is established
3375  * but the other side has, say, rebooted without cleaning out the
3376  * connection.   The SYNs could be construed as an attack and wind
3377  * up ignored, but in case it isn't an attack we can validate the
3378  * connection by forcing a keepalive.
3379  */
3380 void
3381 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3382 {
3383         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3384                 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3385                         tp->t_flags |= TF_KEEPALIVE;
3386                         tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3387                                           tcp_timer_keep);
3388                 } else {
3389                         tp->t_rcvtime = ticks;
3390                         tp->t_flags &= ~TF_KEEPALIVE;
3391                         tcp_callout_reset(tp, tp->tt_keep,
3392                                           tp->t_keepidle,
3393                                           tcp_timer_keep);
3394                 }
3395         }
3396 }
3397
3398 static int
3399 tcp_rmx_msl(const struct tcpcb *tp)
3400 {
3401         struct rtentry *rt;
3402         struct inpcb *inp = tp->t_inpcb;
3403         int msl;
3404 #ifdef INET6
3405         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3406 #else
3407         const boolean_t isipv6 = FALSE;
3408 #endif
3409
3410         if (isipv6)
3411                 rt = tcp_rtlookup6(&inp->inp_inc);
3412         else
3413                 rt = tcp_rtlookup(&inp->inp_inc);
3414         if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3415                 return tcp_msl;
3416
3417         msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3418         if (msl == 0)
3419                 msl = 1;
3420
3421         return msl;
3422 }
3423
3424 static void
3425 tcp_established(struct tcpcb *tp)
3426 {
3427         tp->t_state = TCPS_ESTABLISHED;
3428         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
3429
3430         if (tp->t_rxtsyn > 0) {
3431                 /*
3432                  * RFC6298:
3433                  * "If the timer expires awaiting the ACK of a SYN segment
3434                  *  and the TCP implementation is using an RTO less than 3
3435                  *  seconds, the RTO MUST be re-initialized to 3 seconds
3436                  *  when data transmission begins"
3437                  */
3438                 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3439                         tp->t_rxtcur = TCPTV_RTOBASE3;
3440         }
3441 }
3442
3443 /*
3444  * Returns TRUE, if the ACK should be dropped
3445  */
3446 static boolean_t
3447 tcp_recv_dupack(struct tcpcb *tp, tcp_seq th_ack, u_int to_flags)
3448 {
3449         boolean_t fast_sack_rexmt = TRUE;
3450
3451         tcpstat.tcps_rcvdupack++;
3452
3453         /*
3454          * We have outstanding data (other than a window probe),
3455          * this is a completely duplicate ack (ie, window info
3456          * didn't change), the ack is the biggest we've seen and
3457          * we've seen exactly our rexmt threshhold of them, so
3458          * assume a packet has been dropped and retransmit it.
3459          * Kludge snd_nxt & the congestion window so we send only
3460          * this one packet.
3461          */
3462         if (IN_FASTRECOVERY(tp)) {
3463                 if (TCP_DO_SACK(tp)) {
3464                         boolean_t force = FALSE;
3465
3466                         if (tp->snd_una == tp->rexmt_high &&
3467                             (to_flags & (TOF_SACK | TOF_SACK_REDUNDANT)) ==
3468                             TOF_SACK) {
3469                                 /*
3470                                  * New segments got SACKed and
3471                                  * no retransmit yet.
3472                                  */
3473                                 force = TRUE;
3474                         }
3475
3476                         /* No artifical cwnd inflation. */
3477                         tcp_sack_rexmt(tp, force);
3478                 } else {
3479                         /*
3480                          * Dup acks mean that packets have left
3481                          * the network (they're now cached at the
3482                          * receiver) so bump cwnd by the amount in
3483                          * the receiver to keep a constant cwnd
3484                          * packets in the network.
3485                          */
3486                         tp->snd_cwnd += tp->t_maxseg;
3487                         tcp_output(tp);
3488                 }
3489                 return TRUE;
3490         } else if (SEQ_LT(th_ack, tp->snd_recover)) {
3491                 tp->t_dupacks = 0;
3492                 return FALSE;
3493         } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) &&
3494             (to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) ==
3495             (TOF_DSACK | TOF_SACK_REDUNDANT)) {
3496                 /*
3497                  * If the ACK carries DSACK and other SACK blocks
3498                  * carry information that we have already known,
3499                  * don't count this ACK as duplicate ACK.  This
3500                  * prevents spurious early retransmit and fast
3501                  * retransmit.  This also meets the requirement of
3502                  * RFC3042 that new segments should not be sent if
3503                  * the SACK blocks do not contain new information
3504                  * (XXX we actually loosen the requirment that only
3505                  * DSACK is checked here).
3506                  *
3507                  * This kind of ACKs are usually sent after spurious
3508                  * retransmit.
3509                  */
3510                 /* Do nothing; don't change t_dupacks */
3511                 return TRUE;
3512         } else if (tp->t_dupacks == 0 && TCP_DO_NCR(tp)) {
3513