Merge branch 'vendor/FILE'
[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  * $DragonFly: src/sys/netinet/tcp_input.c,v 1.68 2008/08/22 09:14:17 sephe Exp $
69  */
70
71 #include "opt_ipfw.h"           /* for ipfw_fwd         */
72 #include "opt_inet6.h"
73 #include "opt_ipsec.h"
74 #include "opt_tcpdebug.h"
75 #include "opt_tcp_input.h"
76
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/malloc.h>
82 #include <sys/mbuf.h>
83 #include <sys/proc.h>           /* for proc0 declaration */
84 #include <sys/protosw.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/syslog.h>
88 #include <sys/in_cksum.h>
89
90 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
91 #include <machine/stdarg.h>
92
93 #include <net/if.h>
94 #include <net/route.h>
95
96 #include <netinet/in.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_icmp.h>    /* for ICMP_BANDLIM */
100 #include <netinet/in_var.h>
101 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
102 #include <netinet/in_pcb.h>
103 #include <netinet/ip_var.h>
104 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet/tcp.h>
110 #include <netinet/tcp_fsm.h>
111 #include <netinet/tcp_seq.h>
112 #include <netinet/tcp_timer.h>
113 #include <netinet/tcp_timer2.h>
114 #include <netinet/tcp_var.h>
115 #include <netinet6/tcp6_var.h>
116 #include <netinet/tcpip.h>
117
118 #ifdef TCPDEBUG
119 #include <netinet/tcp_debug.h>
120
121 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
122 struct tcphdr tcp_savetcp;
123 #endif
124
125 #ifdef FAST_IPSEC
126 #include <netproto/ipsec/ipsec.h>
127 #include <netproto/ipsec/ipsec6.h>
128 #endif
129
130 #ifdef IPSEC
131 #include <netinet6/ipsec.h>
132 #include <netinet6/ipsec6.h>
133 #include <netproto/key/key.h>
134 #endif
135
136 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
137
138 tcp_cc  tcp_ccgen;
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 int tcp_do_rfc3390 = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
172     &tcp_do_rfc3390, 0,
173     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
174
175 static int tcp_do_eifel_detect = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
177     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
178
179 static int tcp_do_abc = 1;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
181     &tcp_do_abc, 0,
182     "TCP Appropriate Byte Counting (RFC 3465)");
183
184 /*
185  * Define as tunable for easy testing with SACK on and off.
186  * Warning:  do not change setting in the middle of an existing active TCP flow,
187  *   else strange things might happen to that flow.
188  */
189 int tcp_do_sack = 1;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
191     &tcp_do_sack, 0, "Enable SACK Algorithms");
192
193 int tcp_do_smartsack = 1;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
195     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
196
197 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
198     "TCP Segment Reassembly Queue");
199
200 int tcp_reass_maxseg = 0;
201 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
202     &tcp_reass_maxseg, 0,
203     "Global maximum number of TCP Segments in Reassembly Queue");
204
205 int tcp_reass_qsize = 0;
206 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
207     &tcp_reass_qsize, 0,
208     "Global number of TCP Segments currently in Reassembly Queue");
209
210 static int tcp_reass_overflows = 0;
211 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
212     &tcp_reass_overflows, 0,
213     "Global number of TCP Segment Reassembly Queue Overflows");
214
215 static void      tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
216 static void      tcp_pulloutofband(struct socket *,
217                      struct tcphdr *, struct mbuf *, int);
218 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
219                      struct mbuf *);
220 static void      tcp_xmit_timer(struct tcpcb *, int);
221 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
222 static void      tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
223
224 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
225 #ifdef INET6
226 #define ND6_HINT(tp) \
227 do { \
228         if ((tp) && (tp)->t_inpcb && \
229             ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
230             (tp)->t_inpcb->in6p_route.ro_rt) \
231                 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
232 } while (0)
233 #else
234 #define ND6_HINT(tp)
235 #endif
236
237 /*
238  * Indicate whether this ack should be delayed.  We can delay the ack if
239  *      - delayed acks are enabled and
240  *      - there is no delayed ack timer in progress and
241  *      - our last ack wasn't a 0-sized window.  We never want to delay
242  *        the ack that opens up a 0-sized window.
243  */
244 #define DELAY_ACK(tp) \
245         (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
246         !(tp->t_flags & TF_RXWIN0SENT))
247
248 #define acceptable_window_update(tp, th, tiwin)                         \
249     (SEQ_LT(tp->snd_wl1, th->th_seq) ||                                 \
250      (tp->snd_wl1 == th->th_seq &&                                      \
251       (SEQ_LT(tp->snd_wl2, th->th_ack) ||                               \
252        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
253
254 static int
255 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
256 {
257         struct tseg_qent *q;
258         struct tseg_qent *p = NULL;
259         struct tseg_qent *te;
260         struct socket *so = tp->t_inpcb->inp_socket;
261         int flags;
262
263         /*
264          * Call with th == NULL after become established to
265          * force pre-ESTABLISHED data up to user socket.
266          */
267         if (th == NULL)
268                 goto present;
269
270         /*
271          * Limit the number of segments in the reassembly queue to prevent
272          * holding on to too many segments (and thus running out of mbufs).
273          * Make sure to let the missing segment through which caused this
274          * queue.  Always keep one global queue entry spare to be able to
275          * process the missing segment.
276          */
277         if (th->th_seq != tp->rcv_nxt &&
278             tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
279                 tcp_reass_overflows++;
280                 tcpstat.tcps_rcvmemdrop++;
281                 m_freem(m);
282                 /* no SACK block to report */
283                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
284                 return (0);
285         }
286
287         /* Allocate a new queue entry. */
288         MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
289                M_INTWAIT | M_NULLOK);
290         if (te == NULL) {
291                 tcpstat.tcps_rcvmemdrop++;
292                 m_freem(m);
293                 /* no SACK block to report */
294                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
295                 return (0);
296         }
297         tcp_reass_qsize++;
298
299         /*
300          * Find a segment which begins after this one does.
301          */
302         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
303                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
304                         break;
305                 p = q;
306         }
307
308         /*
309          * If there is a preceding segment, it may provide some of
310          * our data already.  If so, drop the data from the incoming
311          * segment.  If it provides all of our data, drop us.
312          */
313         if (p != NULL) {
314                 tcp_seq_diff_t i;
315
316                 /* conversion to int (in i) handles seq wraparound */
317                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
318                 if (i > 0) {            /* overlaps preceding segment */
319                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
320                         /* enclosing block starts w/ preceding segment */
321                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
322                         if (i >= *tlenp) {
323                                 /* preceding encloses incoming segment */
324                                 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
325                                     p->tqe_len;
326                                 tcpstat.tcps_rcvduppack++;
327                                 tcpstat.tcps_rcvdupbyte += *tlenp;
328                                 m_freem(m);
329                                 kfree(te, M_TSEGQ);
330                                 tcp_reass_qsize--;
331                                 /*
332                                  * Try to present any queued data
333                                  * at the left window edge to the user.
334                                  * This is needed after the 3-WHS
335                                  * completes.
336                                  */
337                                 goto present;   /* ??? */
338                         }
339                         m_adj(m, i);
340                         *tlenp -= i;
341                         th->th_seq += i;
342                         /* incoming segment end is enclosing block end */
343                         tp->encloseblk.rblk_end = th->th_seq + *tlenp +
344                             ((th->th_flags & TH_FIN) != 0);
345                         /* trim end of reported D-SACK block */
346                         tp->reportblk.rblk_end = th->th_seq;
347                 }
348         }
349         tcpstat.tcps_rcvoopack++;
350         tcpstat.tcps_rcvoobyte += *tlenp;
351
352         /*
353          * While we overlap succeeding segments trim them or,
354          * if they are completely covered, dequeue them.
355          */
356         while (q) {
357                 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
358                 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
359                 struct tseg_qent *nq;
360
361                 if (i <= 0)
362                         break;
363                 if (!(tp->t_flags & TF_DUPSEG)) {    /* first time through */
364                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
365                         tp->encloseblk = tp->reportblk;
366                         /* report trailing duplicate D-SACK segment */
367                         tp->reportblk.rblk_start = q->tqe_th->th_seq;
368                 }
369                 if ((tp->t_flags & TF_ENCLOSESEG) &&
370                     SEQ_GT(qend, tp->encloseblk.rblk_end)) {
371                         /* extend enclosing block if one exists */
372                         tp->encloseblk.rblk_end = qend;
373                 }
374                 if (i < q->tqe_len) {
375                         q->tqe_th->th_seq += i;
376                         q->tqe_len -= i;
377                         m_adj(q->tqe_m, i);
378                         break;
379                 }
380
381                 nq = LIST_NEXT(q, tqe_q);
382                 LIST_REMOVE(q, tqe_q);
383                 m_freem(q->tqe_m);
384                 kfree(q, M_TSEGQ);
385                 tcp_reass_qsize--;
386                 q = nq;
387         }
388
389         /* Insert the new segment queue entry into place. */
390         te->tqe_m = m;
391         te->tqe_th = th;
392         te->tqe_len = *tlenp;
393
394         /* check if can coalesce with following segment */
395         if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
396                 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
397
398                 te->tqe_len += q->tqe_len;
399                 if (q->tqe_th->th_flags & TH_FIN)
400                         te->tqe_th->th_flags |= TH_FIN;
401                 m_cat(te->tqe_m, q->tqe_m);
402                 tp->encloseblk.rblk_end = tend;
403                 /*
404                  * When not reporting a duplicate segment, use
405                  * the larger enclosing block as the SACK block.
406                  */
407                 if (!(tp->t_flags & TF_DUPSEG))
408                         tp->reportblk.rblk_end = tend;
409                 LIST_REMOVE(q, tqe_q);
410                 kfree(q, M_TSEGQ);
411                 tcp_reass_qsize--;
412         }
413
414         if (p == NULL) {
415                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
416         } else {
417                 /* check if can coalesce with preceding segment */
418                 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
419                         p->tqe_len += te->tqe_len;
420                         m_cat(p->tqe_m, te->tqe_m);
421                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
422                         /*
423                          * When not reporting a duplicate segment, use
424                          * the larger enclosing block as the SACK block.
425                          */
426                         if (!(tp->t_flags & TF_DUPSEG))
427                                 tp->reportblk.rblk_start = p->tqe_th->th_seq;
428                         kfree(te, M_TSEGQ);
429                         tcp_reass_qsize--;
430                 } else
431                         LIST_INSERT_AFTER(p, te, tqe_q);
432         }
433
434 present:
435         /*
436          * Present data to user, advancing rcv_nxt through
437          * completed sequence space.
438          */
439         if (!TCPS_HAVEESTABLISHED(tp->t_state))
440                 return (0);
441         q = LIST_FIRST(&tp->t_segq);
442         if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
443                 return (0);
444         tp->rcv_nxt += q->tqe_len;
445         if (!(tp->t_flags & TF_DUPSEG)) {
446                 /* no SACK block to report since ACK advanced */
447                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
448         }
449         /* no enclosing block to report since ACK advanced */
450         tp->t_flags &= ~TF_ENCLOSESEG;
451         flags = q->tqe_th->th_flags & TH_FIN;
452         LIST_REMOVE(q, tqe_q);
453         KASSERT(LIST_EMPTY(&tp->t_segq) ||
454                 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
455                 ("segment not coalesced"));
456         if (so->so_state & SS_CANTRCVMORE)
457                 m_freem(q->tqe_m);
458         else
459                 ssb_appendstream(&so->so_rcv, q->tqe_m);
460         kfree(q, M_TSEGQ);
461         tcp_reass_qsize--;
462         ND6_HINT(tp);
463         sorwakeup(so);
464         return (flags);
465 }
466
467 /*
468  * TCP input routine, follows pages 65-76 of the
469  * protocol specification dated September, 1981 very closely.
470  */
471 #ifdef INET6
472 int
473 tcp6_input(struct mbuf **mp, int *offp, int proto)
474 {
475         struct mbuf *m = *mp;
476         struct in6_ifaddr *ia6;
477
478         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
479
480         /*
481          * draft-itojun-ipv6-tcp-to-anycast
482          * better place to put this in?
483          */
484         ia6 = ip6_getdstifaddr(m);
485         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
486                 struct ip6_hdr *ip6;
487
488                 ip6 = mtod(m, struct ip6_hdr *);
489                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
490                             offsetof(struct ip6_hdr, ip6_dst));
491                 return (IPPROTO_DONE);
492         }
493
494         tcp_input(m, *offp, proto);
495         return (IPPROTO_DONE);
496 }
497 #endif
498
499 void
500 tcp_input(struct mbuf *m, ...)
501 {
502         __va_list ap;
503         int off0, proto;
504         struct tcphdr *th;
505         struct ip *ip = NULL;
506         struct ipovly *ipov;
507         struct inpcb *inp = NULL;
508         u_char *optp = NULL;
509         int optlen = 0;
510         int len, tlen, off;
511         int drop_hdrlen;
512         struct tcpcb *tp = NULL;
513         int thflags;
514         struct socket *so = 0;
515         int todrop, acked;
516         boolean_t ourfinisacked, needoutput = FALSE;
517         u_long tiwin;
518         int recvwin;
519         struct tcpopt to;               /* options in this segment */
520         struct rmxp_tao *taop;          /* pointer to our TAO cache entry */
521         struct rmxp_tao tao_noncached;  /* in case there's no cached entry */
522         struct sockaddr_in *next_hop = NULL;
523         int rstreason; /* For badport_bandlim accounting purposes */
524         int cpu;
525         struct ip6_hdr *ip6 = NULL;
526 #ifdef INET6
527         boolean_t isipv6;
528 #else
529         const boolean_t isipv6 = FALSE;
530 #endif
531 #ifdef TCPDEBUG
532         short ostate = 0;
533 #endif
534
535         __va_start(ap, m);
536         off0 = __va_arg(ap, int);
537         proto = __va_arg(ap, int);
538         __va_end(ap);
539
540         tcpstat.tcps_rcvtotal++;
541
542         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
543                 struct m_tag *mtag;
544
545                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
546                 KKASSERT(mtag != NULL);
547                 next_hop = m_tag_data(mtag);
548         }
549
550 #ifdef INET6
551         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
552 #endif
553
554         if (isipv6) {
555                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
556                 ip6 = mtod(m, struct ip6_hdr *);
557                 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
558                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
559                         tcpstat.tcps_rcvbadsum++;
560                         goto drop;
561                 }
562                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
563
564                 /*
565                  * Be proactive about unspecified IPv6 address in source.
566                  * As we use all-zero to indicate unbounded/unconnected pcb,
567                  * unspecified IPv6 address can be used to confuse us.
568                  *
569                  * Note that packets with unspecified IPv6 destination is
570                  * already dropped in ip6_input.
571                  */
572                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
573                         /* XXX stat */
574                         goto drop;
575                 }
576         } else {
577                 /*
578                  * Get IP and TCP header together in first mbuf.
579                  * Note: IP leaves IP header in first mbuf.
580                  */
581                 if (off0 > sizeof(struct ip)) {
582                         ip_stripoptions(m);
583                         off0 = sizeof(struct ip);
584                 }
585                 /* already checked and pulled up in ip_demux() */
586                 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
587                     ("TCP header not in one mbuf: m->m_len %d", m->m_len));
588                 ip = mtod(m, struct ip *);
589                 ipov = (struct ipovly *)ip;
590                 th = (struct tcphdr *)((caddr_t)ip + off0);
591                 tlen = ip->ip_len;
592
593                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
594                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
595                                 th->th_sum = m->m_pkthdr.csum_data;
596                         else
597                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
598                                                 ip->ip_dst.s_addr,
599                                                 htonl(m->m_pkthdr.csum_data +
600                                                         ip->ip_len +
601                                                         IPPROTO_TCP));
602                         th->th_sum ^= 0xffff;
603                 } else {
604                         /*
605                          * Checksum extended TCP header and data.
606                          */
607                         len = sizeof(struct ip) + tlen;
608                         bzero(ipov->ih_x1, sizeof ipov->ih_x1);
609                         ipov->ih_len = (u_short)tlen;
610                         ipov->ih_len = htons(ipov->ih_len);
611                         th->th_sum = in_cksum(m, len);
612                 }
613                 if (th->th_sum) {
614                         tcpstat.tcps_rcvbadsum++;
615                         goto drop;
616                 }
617 #ifdef INET6
618                 /* Re-initialization for later version check */
619                 ip->ip_v = IPVERSION;
620 #endif
621         }
622
623         /*
624          * Check that TCP offset makes sense,
625          * pull out TCP options and adjust length.              XXX
626          */
627         off = th->th_off << 2;
628         /* already checked and pulled up in ip_demux() */
629         KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
630             ("bad TCP data offset %d (tlen %d)", off, tlen));
631         tlen -= off;    /* tlen is used instead of ti->ti_len */
632         if (off > sizeof(struct tcphdr)) {
633                 if (isipv6) {
634                         IP6_EXTHDR_CHECK(m, off0, off, );
635                         ip6 = mtod(m, struct ip6_hdr *);
636                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
637                 } else {
638                         /* already pulled up in ip_demux() */
639                         KASSERT(m->m_len >= sizeof(struct ip) + off,
640                             ("TCP header and options not in one mbuf: "
641                              "m_len %d, off %d", m->m_len, off));
642                 }
643                 optlen = off - sizeof(struct tcphdr);
644                 optp = (u_char *)(th + 1);
645         }
646         thflags = th->th_flags;
647
648 #ifdef TCP_DROP_SYNFIN
649         /*
650          * If the drop_synfin option is enabled, drop all packets with
651          * both the SYN and FIN bits set. This prevents e.g. nmap from
652          * identifying the TCP/IP stack.
653          *
654          * This is a violation of the TCP specification.
655          */
656         if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
657                 goto drop;
658 #endif
659
660         /*
661          * Convert TCP protocol specific fields to host format.
662          */
663         th->th_seq = ntohl(th->th_seq);
664         th->th_ack = ntohl(th->th_ack);
665         th->th_win = ntohs(th->th_win);
666         th->th_urp = ntohs(th->th_urp);
667
668         /*
669          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
670          * until after ip6_savecontrol() is called and before other functions
671          * which don't want those proto headers.
672          * Because ip6_savecontrol() is going to parse the mbuf to
673          * search for data to be passed up to user-land, it wants mbuf
674          * parameters to be unchanged.
675          * XXX: the call of ip6_savecontrol() has been obsoleted based on
676          * latest version of the advanced API (20020110).
677          */
678         drop_hdrlen = off0 + off;
679
680         /*
681          * Locate pcb for segment.
682          */
683 findpcb:
684         /* IPFIREWALL_FORWARD section */
685         if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
686                 /*
687                  * Transparently forwarded. Pretend to be the destination.
688                  * already got one like this?
689                  */
690                 cpu = mycpu->gd_cpuid;
691                 inp = in_pcblookup_hash(&tcbinfo[cpu],
692                                         ip->ip_src, th->th_sport,
693                                         ip->ip_dst, th->th_dport,
694                                         0, m->m_pkthdr.rcvif);
695                 if (!inp) {
696                         /*
697                          * It's new.  Try to find the ambushing socket.
698                          */
699
700                         /*
701                          * The rest of the ipfw code stores the port in
702                          * host order.  XXX
703                          * (The IP address is still in network order.)
704                          */
705                         in_port_t dport = next_hop->sin_port ?
706                                                 htons(next_hop->sin_port) :
707                                                 th->th_dport;
708
709                         cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
710                                           next_hop->sin_addr.s_addr, dport);
711                         inp = in_pcblookup_hash(&tcbinfo[cpu],
712                                                 ip->ip_src, th->th_sport,
713                                                 next_hop->sin_addr, dport,
714                                                 1, m->m_pkthdr.rcvif);
715                 }
716         } else {
717                 if (isipv6) {
718                         inp = in6_pcblookup_hash(&tcbinfo[0],
719                                                  &ip6->ip6_src, th->th_sport,
720                                                  &ip6->ip6_dst, th->th_dport,
721                                                  1, m->m_pkthdr.rcvif);
722                 } else {
723                         cpu = mycpu->gd_cpuid;
724                         inp = in_pcblookup_hash(&tcbinfo[cpu],
725                                                 ip->ip_src, th->th_sport,
726                                                 ip->ip_dst, th->th_dport,
727                                                 1, m->m_pkthdr.rcvif);
728                 }
729         }
730
731         /*
732          * If the state is CLOSED (i.e., TCB does not exist) then
733          * all data in the incoming segment is discarded.
734          * If the TCB exists but is in CLOSED state, it is embryonic,
735          * but should either do a listen or a connect soon.
736          */
737         if (inp == NULL) {
738                 if (log_in_vain) {
739 #ifdef INET6
740                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
741 #else
742                         char dbuf[sizeof "aaa.bbb.ccc.ddd"];
743                         char sbuf[sizeof "aaa.bbb.ccc.ddd"];
744 #endif
745                         if (isipv6) {
746                                 strcpy(dbuf, "[");
747                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
748                                 strcat(dbuf, "]");
749                                 strcpy(sbuf, "[");
750                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
751                                 strcat(sbuf, "]");
752                         } else {
753                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
754                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
755                         }
756                         switch (log_in_vain) {
757                         case 1:
758                                 if (!(thflags & TH_SYN))
759                                         break;
760                         case 2:
761                                 log(LOG_INFO,
762                                     "Connection attempt to TCP %s:%d "
763                                     "from %s:%d flags:0x%02x\n",
764                                     dbuf, ntohs(th->th_dport), sbuf,
765                                     ntohs(th->th_sport), thflags);
766                                 break;
767                         default:
768                                 break;
769                         }
770                 }
771                 if (blackhole) {
772                         switch (blackhole) {
773                         case 1:
774                                 if (thflags & TH_SYN)
775                                         goto drop;
776                                 break;
777                         case 2:
778                                 goto drop;
779                         default:
780                                 goto drop;
781                         }
782                 }
783                 rstreason = BANDLIM_RST_CLOSEDPORT;
784                 goto dropwithreset;
785         }
786
787 #ifdef IPSEC
788         if (isipv6) {
789                 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
790                         ipsec6stat.in_polvio++;
791                         goto drop;
792                 }
793         } else {
794                 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
795                         ipsecstat.in_polvio++;
796                         goto drop;
797                 }
798         }
799 #endif
800 #ifdef FAST_IPSEC
801         if (isipv6) {
802                 if (ipsec6_in_reject(m, inp))
803                         goto drop;
804         } else {
805                 if (ipsec4_in_reject(m, inp))
806                         goto drop;
807         }
808 #endif
809         /* Check the minimum TTL for socket. */
810 #ifdef INET6
811         if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
812                 goto drop;
813 #endif
814
815         tp = intotcpcb(inp);
816         if (tp == NULL) {
817                 rstreason = BANDLIM_RST_CLOSEDPORT;
818                 goto dropwithreset;
819         }
820         if (tp->t_state <= TCPS_CLOSED)
821                 goto drop;
822
823         /* Unscale the window into a 32-bit value. */
824         if (!(thflags & TH_SYN))
825                 tiwin = th->th_win << tp->snd_scale;
826         else
827                 tiwin = th->th_win;
828
829         so = inp->inp_socket;
830
831 #ifdef TCPDEBUG
832         if (so->so_options & SO_DEBUG) {
833                 ostate = tp->t_state;
834                 if (isipv6)
835                         bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
836                 else
837                         bcopy(ip, tcp_saveipgen, sizeof(*ip));
838                 tcp_savetcp = *th;
839         }
840 #endif
841
842         bzero(&to, sizeof to);
843
844         if (so->so_options & SO_ACCEPTCONN) {
845                 struct in_conninfo inc;
846
847 #ifdef INET6
848                 inc.inc_isipv6 = (isipv6 == TRUE);
849 #endif
850                 if (isipv6) {
851                         inc.inc6_faddr = ip6->ip6_src;
852                         inc.inc6_laddr = ip6->ip6_dst;
853                         inc.inc6_route.ro_rt = NULL;            /* XXX */
854                 } else {
855                         inc.inc_faddr = ip->ip_src;
856                         inc.inc_laddr = ip->ip_dst;
857                         inc.inc_route.ro_rt = NULL;             /* XXX */
858                 }
859                 inc.inc_fport = th->th_sport;
860                 inc.inc_lport = th->th_dport;
861
862                 /*
863                  * If the state is LISTEN then ignore segment if it contains
864                  * a RST.  If the segment contains an ACK then it is bad and
865                  * send a RST.  If it does not contain a SYN then it is not
866                  * interesting; drop it.
867                  *
868                  * If the state is SYN_RECEIVED (syncache) and seg contains
869                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
870                  * contains a RST, check the sequence number to see if it
871                  * is a valid reset segment.
872                  */
873                 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
874                         if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
875                                 if (!syncache_expand(&inc, th, &so, m)) {
876                                         /*
877                                          * No syncache entry, or ACK was not
878                                          * for our SYN/ACK.  Send a RST.
879                                          */
880                                         tcpstat.tcps_badsyn++;
881                                         rstreason = BANDLIM_RST_OPENPORT;
882                                         goto dropwithreset;
883                                 }
884                                 if (so == NULL)
885                                         /*
886                                          * Could not complete 3-way handshake,
887                                          * connection is being closed down, and
888                                          * syncache will free mbuf.
889                                          */
890                                         return;
891                                 /*
892                                  * Socket is created in state SYN_RECEIVED.
893                                  * Continue processing segment.
894                                  */
895                                 inp = so->so_pcb;
896                                 tp = intotcpcb(inp);
897                                 /*
898                                  * This is what would have happened in
899                                  * tcp_output() when the SYN,ACK was sent.
900                                  */
901                                 tp->snd_up = tp->snd_una;
902                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
903                                 tp->last_ack_sent = tp->rcv_nxt;
904 /*
905  * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
906  * until the _second_ ACK is received:
907  *    rcv SYN (set wscale opts)  --> send SYN/ACK, set snd_wnd = window.
908  *    rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
909  *        move to ESTAB, set snd_wnd to tiwin.
910  */
911                                 tp->snd_wnd = tiwin;    /* unscaled */
912                                 goto after_listen;
913                         }
914                         if (thflags & TH_RST) {
915                                 syncache_chkrst(&inc, th);
916                                 goto drop;
917                         }
918                         if (thflags & TH_ACK) {
919                                 syncache_badack(&inc);
920                                 tcpstat.tcps_badsyn++;
921                                 rstreason = BANDLIM_RST_OPENPORT;
922                                 goto dropwithreset;
923                         }
924                         goto drop;
925                 }
926
927                 /*
928                  * Segment's flags are (SYN) or (SYN | FIN).
929                  */
930 #ifdef INET6
931                 /*
932                  * If deprecated address is forbidden,
933                  * we do not accept SYN to deprecated interface
934                  * address to prevent any new inbound connection from
935                  * getting established.
936                  * When we do not accept SYN, we send a TCP RST,
937                  * with deprecated source address (instead of dropping
938                  * it).  We compromise it as it is much better for peer
939                  * to send a RST, and RST will be the final packet
940                  * for the exchange.
941                  *
942                  * If we do not forbid deprecated addresses, we accept
943                  * the SYN packet.  RFC2462 does not suggest dropping
944                  * SYN in this case.
945                  * If we decipher RFC2462 5.5.4, it says like this:
946                  * 1. use of deprecated addr with existing
947                  *    communication is okay - "SHOULD continue to be
948                  *    used"
949                  * 2. use of it with new communication:
950                  *   (2a) "SHOULD NOT be used if alternate address
951                  *        with sufficient scope is available"
952                  *   (2b) nothing mentioned otherwise.
953                  * Here we fall into (2b) case as we have no choice in
954                  * our source address selection - we must obey the peer.
955                  *
956                  * The wording in RFC2462 is confusing, and there are
957                  * multiple description text for deprecated address
958                  * handling - worse, they are not exactly the same.
959                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
960                  */
961                 if (isipv6 && !ip6_use_deprecated) {
962                         struct in6_ifaddr *ia6;
963
964                         if ((ia6 = ip6_getdstifaddr(m)) &&
965                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
966                                 tp = NULL;
967                                 rstreason = BANDLIM_RST_OPENPORT;
968                                 goto dropwithreset;
969                         }
970                 }
971 #endif
972                 /*
973                  * If it is from this socket, drop it, it must be forged.
974                  * Don't bother responding if the destination was a broadcast.
975                  */
976                 if (th->th_dport == th->th_sport) {
977                         if (isipv6) {
978                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
979                                                        &ip6->ip6_src))
980                                         goto drop;
981                         } else {
982                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
983                                         goto drop;
984                         }
985                 }
986                 /*
987                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
988                  *
989                  * Note that it is quite possible to receive unicast
990                  * link-layer packets with a broadcast IP address. Use
991                  * in_broadcast() to find them.
992                  */
993                 if (m->m_flags & (M_BCAST | M_MCAST))
994                         goto drop;
995                 if (isipv6) {
996                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
997                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
998                                 goto drop;
999                 } else {
1000                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1001                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1002                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1003                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1004                                 goto drop;
1005                 }
1006                 /*
1007                  * SYN appears to be valid; create compressed TCP state
1008                  * for syncache, or perform t/tcp connection.
1009                  */
1010                 if (so->so_qlen <= so->so_qlimit) {
1011                         tcp_dooptions(&to, optp, optlen, TRUE);
1012                         if (!syncache_add(&inc, &to, th, &so, m))
1013                                 goto drop;
1014                         if (so == NULL)
1015                                 /*
1016                                  * Entry added to syncache, mbuf used to
1017                                  * send SYN,ACK packet.
1018                                  */
1019                                 return;
1020                         /*
1021                          * Segment passed TAO tests.
1022                          */
1023                         inp = so->so_pcb;
1024                         tp = intotcpcb(inp);
1025                         tp->snd_wnd = tiwin;
1026                         tp->t_starttime = ticks;
1027                         tp->t_state = TCPS_ESTABLISHED;
1028
1029                         /*
1030                          * If there is a FIN, or if there is data and the
1031                          * connection is local, then delay SYN,ACK(SYN) in
1032                          * the hope of piggy-backing it on a response
1033                          * segment.  Otherwise must send ACK now in case
1034                          * the other side is slow starting.
1035                          */
1036                         if (DELAY_ACK(tp) &&
1037                             ((thflags & TH_FIN) ||
1038                              (tlen != 0 &&
1039                               ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1040                                (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1041                                 tcp_callout_reset(tp, tp->tt_delack,
1042                                     tcp_delacktime, tcp_timer_delack);
1043                                 tp->t_flags |= TF_NEEDSYN;
1044                         } else {
1045                                 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1046                         }
1047
1048                         tcpstat.tcps_connects++;
1049                         soisconnected(so);
1050                         goto trimthenstep6;
1051                 }
1052                 goto drop;
1053         }
1054 after_listen:
1055
1056         /* should not happen - syncache should pick up these connections */
1057         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1058
1059         /*
1060          * Segment received on connection.
1061          * Reset idle time and keep-alive timer.
1062          */
1063         tp->t_rcvtime = ticks;
1064         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
1065                 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1066                     tcp_timer_keep);
1067         }
1068
1069         /*
1070          * Process options.
1071          * XXX this is tradtitional behavior, may need to be cleaned up.
1072          */
1073         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1074         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1075                 if (to.to_flags & TOF_SCALE) {
1076                         tp->t_flags |= TF_RCVD_SCALE;
1077                         tp->requested_s_scale = to.to_requested_s_scale;
1078                 }
1079                 if (to.to_flags & TOF_TS) {
1080                         tp->t_flags |= TF_RCVD_TSTMP;
1081                         tp->ts_recent = to.to_tsval;
1082                         tp->ts_recent_age = ticks;
1083                 }
1084                 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1085                         tp->t_flags |= TF_RCVD_CC;
1086                 if (to.to_flags & TOF_MSS)
1087                         tcp_mss(tp, to.to_mss);
1088                 /*
1089                  * Only set the TF_SACK_PERMITTED per-connection flag
1090                  * if we got a SACK_PERMITTED option from the other side
1091                  * and the global tcp_do_sack variable is true.
1092                  */
1093                 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1094                         tp->t_flags |= TF_SACK_PERMITTED;
1095         }
1096
1097         /*
1098          * Header prediction: check for the two common cases
1099          * of a uni-directional data xfer.  If the packet has
1100          * no control flags, is in-sequence, the window didn't
1101          * change and we're not retransmitting, it's a
1102          * candidate.  If the length is zero and the ack moved
1103          * forward, we're the sender side of the xfer.  Just
1104          * free the data acked & wake any higher level process
1105          * that was blocked waiting for space.  If the length
1106          * is non-zero and the ack didn't move, we're the
1107          * receiver side.  If we're getting packets in-order
1108          * (the reassembly queue is empty), add the data to
1109          * the socket buffer and note that we need a delayed ack.
1110          * Make sure that the hidden state-flags are also off.
1111          * Since we check for TCPS_ESTABLISHED above, it can only
1112          * be TH_NEEDSYN.
1113          */
1114         if (tp->t_state == TCPS_ESTABLISHED &&
1115             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1116             !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1117             (!(to.to_flags & TOF_TS) ||
1118              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1119             /*
1120              * Using the CC option is compulsory if once started:
1121              *   the segment is OK if no T/TCP was negotiated or
1122              *   if the segment has a CC option equal to CCrecv
1123              */
1124             ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1125              ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1126             th->th_seq == tp->rcv_nxt &&
1127             tp->snd_nxt == tp->snd_max) {
1128
1129                 /*
1130                  * If last ACK falls within this segment's sequence numbers,
1131                  * record the timestamp.
1132                  * NOTE that the test is modified according to the latest
1133                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1134                  */
1135                 if ((to.to_flags & TOF_TS) &&
1136                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1137                         tp->ts_recent_age = ticks;
1138                         tp->ts_recent = to.to_tsval;
1139                 }
1140
1141                 if (tlen == 0) {
1142                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1143                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1144                             tp->snd_cwnd >= tp->snd_wnd &&
1145                             !IN_FASTRECOVERY(tp)) {
1146                                 /*
1147                                  * This is a pure ack for outstanding data.
1148                                  */
1149                                 ++tcpstat.tcps_predack;
1150                                 /*
1151                                  * "bad retransmit" recovery
1152                                  *
1153                                  * If Eifel detection applies, then
1154                                  * it is deterministic, so use it
1155                                  * unconditionally over the old heuristic.
1156                                  * Otherwise, fall back to the old heuristic.
1157                                  */
1158                                 if (tcp_do_eifel_detect &&
1159                                     (to.to_flags & TOF_TS) && to.to_tsecr &&
1160                                     (tp->t_flags & TF_FIRSTACCACK)) {
1161                                         /* Eifel detection applicable. */
1162                                         if (to.to_tsecr < tp->t_rexmtTS) {
1163                                                 tcp_revert_congestion_state(tp);
1164                                                 ++tcpstat.tcps_eifeldetected;
1165                                         }
1166                                 } else if (tp->t_rxtshift == 1 &&
1167                                            ticks < tp->t_badrxtwin) {
1168                                         tcp_revert_congestion_state(tp);
1169                                         ++tcpstat.tcps_rttdetected;
1170                                 }
1171                                 tp->t_flags &= ~(TF_FIRSTACCACK |
1172                                                  TF_FASTREXMT | TF_EARLYREXMT);
1173                                 /*
1174                                  * Recalculate the retransmit timer / rtt.
1175                                  *
1176                                  * Some machines (certain windows boxes)
1177                                  * send broken timestamp replies during the
1178                                  * SYN+ACK phase, ignore timestamps of 0.
1179                                  */
1180                                 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1181                                         tcp_xmit_timer(tp,
1182                                                        ticks - to.to_tsecr + 1);
1183                                 } else if (tp->t_rtttime &&
1184                                            SEQ_GT(th->th_ack, tp->t_rtseq)) {
1185                                         tcp_xmit_timer(tp,
1186                                                        ticks - tp->t_rtttime);
1187                                 }
1188                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1189                                 acked = th->th_ack - tp->snd_una;
1190                                 tcpstat.tcps_rcvackpack++;
1191                                 tcpstat.tcps_rcvackbyte += acked;
1192                                 sbdrop(&so->so_snd.sb, acked);
1193                                 tp->snd_recover = th->th_ack - 1;
1194                                 tp->snd_una = th->th_ack;
1195                                 tp->t_dupacks = 0;
1196                                 /*
1197                                  * Update window information.
1198                                  */
1199                                 if (tiwin != tp->snd_wnd &&
1200                                     acceptable_window_update(tp, th, tiwin)) {
1201                                         /* keep track of pure window updates */
1202                                         if (tp->snd_wl2 == th->th_ack &&
1203                                             tiwin > tp->snd_wnd)
1204                                                 tcpstat.tcps_rcvwinupd++;
1205                                         tp->snd_wnd = tiwin;
1206                                         tp->snd_wl1 = th->th_seq;
1207                                         tp->snd_wl2 = th->th_ack;
1208                                         if (tp->snd_wnd > tp->max_sndwnd)
1209                                                 tp->max_sndwnd = tp->snd_wnd;
1210                                 }
1211                                 m_freem(m);
1212                                 ND6_HINT(tp); /* some progress has been done */
1213                                 /*
1214                                  * If all outstanding data are acked, stop
1215                                  * retransmit timer, otherwise restart timer
1216                                  * using current (possibly backed-off) value.
1217                                  * If process is waiting for space,
1218                                  * wakeup/selwakeup/signal.  If data
1219                                  * are ready to send, let tcp_output
1220                                  * decide between more output or persist.
1221                                  */
1222                                 if (tp->snd_una == tp->snd_max) {
1223                                         tcp_callout_stop(tp, tp->tt_rexmt);
1224                                 } else if (!tcp_callout_active(tp,
1225                                             tp->tt_persist)) {
1226                                         tcp_callout_reset(tp, tp->tt_rexmt,
1227                                             tp->t_rxtcur, tcp_timer_rexmt);
1228                                 }
1229                                 sowwakeup(so);
1230                                 if (so->so_snd.ssb_cc > 0)
1231                                         tcp_output(tp);
1232                                 return;
1233                         }
1234                 } else if (tiwin == tp->snd_wnd &&
1235                     th->th_ack == tp->snd_una &&
1236                     LIST_EMPTY(&tp->t_segq) &&
1237                     tlen <= ssb_space(&so->so_rcv)) {
1238                         /*
1239                          * This is a pure, in-sequence data packet
1240                          * with nothing on the reassembly queue and
1241                          * we have enough buffer space to take it.
1242                          */
1243                         ++tcpstat.tcps_preddat;
1244                         tp->rcv_nxt += tlen;
1245                         tcpstat.tcps_rcvpack++;
1246                         tcpstat.tcps_rcvbyte += tlen;
1247                         ND6_HINT(tp);   /* some progress has been done */
1248                         /*
1249                          * Add data to socket buffer.
1250                          */
1251                         if (so->so_state & SS_CANTRCVMORE) {
1252                                 m_freem(m);
1253                         } else {
1254                                 m_adj(m, drop_hdrlen); /* delayed header drop */
1255                                 ssb_appendstream(&so->so_rcv, m);
1256                         }
1257                         sorwakeup(so);
1258                         /*
1259                          * This code is responsible for most of the ACKs
1260                          * the TCP stack sends back after receiving a data
1261                          * packet.  Note that the DELAY_ACK check fails if
1262                          * the delack timer is already running, which results
1263                          * in an ack being sent every other packet (which is
1264                          * what we want).
1265                          *
1266                          * We then further aggregate acks by not actually
1267                          * sending one until the protocol thread has completed
1268                          * processing the current backlog of packets.  This
1269                          * does not delay the ack any further, but allows us
1270                          * to take advantage of the packet aggregation that
1271                          * high speed NICs do (usually blocks of 8-10 packets)
1272                          * to send a single ack rather then four or five acks,
1273                          * greatly reducing the ack rate, the return channel
1274                          * bandwidth, and the protocol overhead on both ends.
1275                          *
1276                          * Since this also has the effect of slowing down
1277                          * the exponential slow-start ramp-up, systems with 
1278                          * very large bandwidth-delay products might want
1279                          * to turn the feature off.
1280                          */
1281                         if (DELAY_ACK(tp)) {
1282                                 tcp_callout_reset(tp, tp->tt_delack,
1283                                     tcp_delacktime, tcp_timer_delack);
1284                         } else if (tcp_aggregate_acks) {
1285                                 tp->t_flags |= TF_ACKNOW;
1286                                 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1287                                         tp->t_flags |= TF_ONOUTPUTQ;
1288                                         tp->tt_cpu = mycpu->gd_cpuid;
1289                                         TAILQ_INSERT_TAIL(
1290                                             &tcpcbackq[tp->tt_cpu],
1291                                             tp, t_outputq);
1292                                 }
1293                         } else {
1294                                 tp->t_flags |= TF_ACKNOW;
1295                                 tcp_output(tp);
1296                         }
1297                         return;
1298                 }
1299         }
1300
1301         /*
1302          * Calculate amount of space in receive window,
1303          * and then do TCP input processing.
1304          * Receive window is amount of space in rcv queue,
1305          * but not less than advertised window.
1306          */
1307         recvwin = ssb_space(&so->so_rcv);
1308         if (recvwin < 0)
1309                 recvwin = 0;
1310         tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1311
1312         switch (tp->t_state) {
1313         /*
1314          * If the state is SYN_RECEIVED:
1315          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1316          */
1317         case TCPS_SYN_RECEIVED:
1318                 if ((thflags & TH_ACK) &&
1319                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1320                      SEQ_GT(th->th_ack, tp->snd_max))) {
1321                                 rstreason = BANDLIM_RST_OPENPORT;
1322                                 goto dropwithreset;
1323                 }
1324                 break;
1325
1326         /*
1327          * If the state is SYN_SENT:
1328          *      if seg contains an ACK, but not for our SYN, drop the input.
1329          *      if seg contains a RST, then drop the connection.
1330          *      if seg does not contain SYN, then drop it.
1331          * Otherwise this is an acceptable SYN segment
1332          *      initialize tp->rcv_nxt and tp->irs
1333          *      if seg contains ack then advance tp->snd_una
1334          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1335          *      arrange for segment to be acked (eventually)
1336          *      continue processing rest of data/controls, beginning with URG
1337          */
1338         case TCPS_SYN_SENT:
1339                 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1340                         taop = &tao_noncached;
1341                         bzero(taop, sizeof *taop);
1342                 }
1343
1344                 if ((thflags & TH_ACK) &&
1345                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1346                      SEQ_GT(th->th_ack, tp->snd_max))) {
1347                         /*
1348                          * If we have a cached CCsent for the remote host,
1349                          * hence we haven't just crashed and restarted,
1350                          * do not send a RST.  This may be a retransmission
1351                          * from the other side after our earlier ACK was lost.
1352                          * Our new SYN, when it arrives, will serve as the
1353                          * needed ACK.
1354                          */
1355                         if (taop->tao_ccsent != 0)
1356                                 goto drop;
1357                         else {
1358                                 rstreason = BANDLIM_UNLIMITED;
1359                                 goto dropwithreset;
1360                         }
1361                 }
1362                 if (thflags & TH_RST) {
1363                         if (thflags & TH_ACK)
1364                                 tp = tcp_drop(tp, ECONNREFUSED);
1365                         goto drop;
1366                 }
1367                 if (!(thflags & TH_SYN))
1368                         goto drop;
1369                 tp->snd_wnd = th->th_win;       /* initial send window */
1370                 tp->cc_recv = to.to_cc;         /* foreign CC */
1371
1372                 tp->irs = th->th_seq;
1373                 tcp_rcvseqinit(tp);
1374                 if (thflags & TH_ACK) {
1375                         /*
1376                          * Our SYN was acked.  If segment contains CC.ECHO
1377                          * option, check it to make sure this segment really
1378                          * matches our SYN.  If not, just drop it as old
1379                          * duplicate, but send an RST if we're still playing
1380                          * by the old rules.  If no CC.ECHO option, make sure
1381                          * we don't get fooled into using T/TCP.
1382                          */
1383                         if (to.to_flags & TOF_CCECHO) {
1384                                 if (tp->cc_send != to.to_ccecho) {
1385                                         if (taop->tao_ccsent != 0)
1386                                                 goto drop;
1387                                         else {
1388                                                 rstreason = BANDLIM_UNLIMITED;
1389                                                 goto dropwithreset;
1390                                         }
1391                                 }
1392                         } else
1393                                 tp->t_flags &= ~TF_RCVD_CC;
1394                         tcpstat.tcps_connects++;
1395                         soisconnected(so);
1396                         /* Do window scaling on this connection? */
1397                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1398                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1399                                 tp->snd_scale = tp->requested_s_scale;
1400                                 tp->rcv_scale = tp->request_r_scale;
1401                         }
1402                         /* Segment is acceptable, update cache if undefined. */
1403                         if (taop->tao_ccsent == 0)
1404                                 taop->tao_ccsent = to.to_ccecho;
1405
1406                         tp->rcv_adv += tp->rcv_wnd;
1407                         tp->snd_una++;          /* SYN is acked */
1408                         tcp_callout_stop(tp, tp->tt_rexmt);
1409                         /*
1410                          * If there's data, delay ACK; if there's also a FIN
1411                          * ACKNOW will be turned on later.
1412                          */
1413                         if (DELAY_ACK(tp) && tlen != 0) {
1414                                 tcp_callout_reset(tp, tp->tt_delack,
1415                                     tcp_delacktime, tcp_timer_delack);
1416                         } else {
1417                                 tp->t_flags |= TF_ACKNOW;
1418                         }
1419                         /*
1420                          * Received <SYN,ACK> in SYN_SENT[*] state.
1421                          * Transitions:
1422                          *      SYN_SENT  --> ESTABLISHED
1423                          *      SYN_SENT* --> FIN_WAIT_1
1424                          */
1425                         tp->t_starttime = ticks;
1426                         if (tp->t_flags & TF_NEEDFIN) {
1427                                 tp->t_state = TCPS_FIN_WAIT_1;
1428                                 tp->t_flags &= ~TF_NEEDFIN;
1429                                 thflags &= ~TH_SYN;
1430                         } else {
1431                                 tp->t_state = TCPS_ESTABLISHED;
1432                                 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1433                                     tcp_timer_keep);
1434                         }
1435                 } else {
1436                         /*
1437                          * Received initial SYN in SYN-SENT[*] state =>
1438                          * simultaneous open.  If segment contains CC option
1439                          * and there is a cached CC, apply TAO test.
1440                          * If it succeeds, connection is * half-synchronized.
1441                          * Otherwise, do 3-way handshake:
1442                          *        SYN-SENT -> SYN-RECEIVED
1443                          *        SYN-SENT* -> SYN-RECEIVED*
1444                          * If there was no CC option, clear cached CC value.
1445                          */
1446                         tp->t_flags |= TF_ACKNOW;
1447                         tcp_callout_stop(tp, tp->tt_rexmt);
1448                         if (to.to_flags & TOF_CC) {
1449                                 if (taop->tao_cc != 0 &&
1450                                     CC_GT(to.to_cc, taop->tao_cc)) {
1451                                         /*
1452                                          * update cache and make transition:
1453                                          *        SYN-SENT -> ESTABLISHED*
1454                                          *        SYN-SENT* -> FIN-WAIT-1*
1455                                          */
1456                                         taop->tao_cc = to.to_cc;
1457                                         tp->t_starttime = ticks;
1458                                         if (tp->t_flags & TF_NEEDFIN) {
1459                                                 tp->t_state = TCPS_FIN_WAIT_1;
1460                                                 tp->t_flags &= ~TF_NEEDFIN;
1461                                         } else {
1462                                                 tp->t_state = TCPS_ESTABLISHED;
1463                                                 tcp_callout_reset(tp,
1464                                                     tp->tt_keep, tcp_keepidle,
1465                                                     tcp_timer_keep);
1466                                         }
1467                                         tp->t_flags |= TF_NEEDSYN;
1468                                 } else
1469                                         tp->t_state = TCPS_SYN_RECEIVED;
1470                         } else {
1471                                 /* CC.NEW or no option => invalidate cache */
1472                                 taop->tao_cc = 0;
1473                                 tp->t_state = TCPS_SYN_RECEIVED;
1474                         }
1475                 }
1476
1477 trimthenstep6:
1478                 /*
1479                  * Advance th->th_seq to correspond to first data byte.
1480                  * If data, trim to stay within window,
1481                  * dropping FIN if necessary.
1482                  */
1483                 th->th_seq++;
1484                 if (tlen > tp->rcv_wnd) {
1485                         todrop = tlen - tp->rcv_wnd;
1486                         m_adj(m, -todrop);
1487                         tlen = tp->rcv_wnd;
1488                         thflags &= ~TH_FIN;
1489                         tcpstat.tcps_rcvpackafterwin++;
1490                         tcpstat.tcps_rcvbyteafterwin += todrop;
1491                 }
1492                 tp->snd_wl1 = th->th_seq - 1;
1493                 tp->rcv_up = th->th_seq;
1494                 /*
1495                  * Client side of transaction: already sent SYN and data.
1496                  * If the remote host used T/TCP to validate the SYN,
1497                  * our data will be ACK'd; if so, enter normal data segment
1498                  * processing in the middle of step 5, ack processing.
1499                  * Otherwise, goto step 6.
1500                  */
1501                 if (thflags & TH_ACK)
1502                         goto process_ACK;
1503
1504                 goto step6;
1505
1506         /*
1507          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1508          *      if segment contains a SYN and CC [not CC.NEW] option:
1509          *              if state == TIME_WAIT and connection duration > MSL,
1510          *                  drop packet and send RST;
1511          *
1512          *              if SEG.CC > CCrecv then is new SYN, and can implicitly
1513          *                  ack the FIN (and data) in retransmission queue.
1514          *                  Complete close and delete TCPCB.  Then reprocess
1515          *                  segment, hoping to find new TCPCB in LISTEN state;
1516          *
1517          *              else must be old SYN; drop it.
1518          *      else do normal processing.
1519          */
1520         case TCPS_LAST_ACK:
1521         case TCPS_CLOSING:
1522         case TCPS_TIME_WAIT:
1523                 if ((thflags & TH_SYN) &&
1524                     (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1525                         if (tp->t_state == TCPS_TIME_WAIT &&
1526                                         (ticks - tp->t_starttime) > tcp_msl) {
1527                                 rstreason = BANDLIM_UNLIMITED;
1528                                 goto dropwithreset;
1529                         }
1530                         if (CC_GT(to.to_cc, tp->cc_recv)) {
1531                                 tp = tcp_close(tp);
1532                                 goto findpcb;
1533                         }
1534                         else
1535                                 goto drop;
1536                 }
1537                 break;  /* continue normal processing */
1538         }
1539
1540         /*
1541          * States other than LISTEN or SYN_SENT.
1542          * First check the RST flag and sequence number since reset segments
1543          * are exempt from the timestamp and connection count tests.  This
1544          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1545          * below which allowed reset segments in half the sequence space
1546          * to fall though and be processed (which gives forged reset
1547          * segments with a random sequence number a 50 percent chance of
1548          * killing a connection).
1549          * Then check timestamp, if present.
1550          * Then check the connection count, if present.
1551          * Then check that at least some bytes of segment are within
1552          * receive window.  If segment begins before rcv_nxt,
1553          * drop leading data (and SYN); if nothing left, just ack.
1554          *
1555          *
1556          * If the RST bit is set, check the sequence number to see
1557          * if this is a valid reset segment.
1558          * RFC 793 page 37:
1559          *   In all states except SYN-SENT, all reset (RST) segments
1560          *   are validated by checking their SEQ-fields.  A reset is
1561          *   valid if its sequence number is in the window.
1562          * Note: this does not take into account delayed ACKs, so
1563          *   we should test against last_ack_sent instead of rcv_nxt.
1564          *   The sequence number in the reset segment is normally an
1565          *   echo of our outgoing acknowledgement numbers, but some hosts
1566          *   send a reset with the sequence number at the rightmost edge
1567          *   of our receive window, and we have to handle this case.
1568          * If we have multiple segments in flight, the intial reset
1569          * segment sequence numbers will be to the left of last_ack_sent,
1570          * but they will eventually catch up.
1571          * In any case, it never made sense to trim reset segments to
1572          * fit the receive window since RFC 1122 says:
1573          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1574          *
1575          *    A TCP SHOULD allow a received RST segment to include data.
1576          *
1577          *    DISCUSSION
1578          *         It has been suggested that a RST segment could contain
1579          *         ASCII text that encoded and explained the cause of the
1580          *         RST.  No standard has yet been established for such
1581          *         data.
1582          *
1583          * If the reset segment passes the sequence number test examine
1584          * the state:
1585          *    SYN_RECEIVED STATE:
1586          *      If passive open, return to LISTEN state.
1587          *      If active open, inform user that connection was refused.
1588          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1589          *      Inform user that connection was reset, and close tcb.
1590          *    CLOSING, LAST_ACK STATES:
1591          *      Close the tcb.
1592          *    TIME_WAIT STATE:
1593          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1594          *      RFC 1337.
1595          */
1596         if (thflags & TH_RST) {
1597                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1598                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1599                         switch (tp->t_state) {
1600
1601                         case TCPS_SYN_RECEIVED:
1602                                 so->so_error = ECONNREFUSED;
1603                                 goto close;
1604
1605                         case TCPS_ESTABLISHED:
1606                         case TCPS_FIN_WAIT_1:
1607                         case TCPS_FIN_WAIT_2:
1608                         case TCPS_CLOSE_WAIT:
1609                                 so->so_error = ECONNRESET;
1610                         close:
1611                                 tp->t_state = TCPS_CLOSED;
1612                                 tcpstat.tcps_drops++;
1613                                 tp = tcp_close(tp);
1614                                 break;
1615
1616                         case TCPS_CLOSING:
1617                         case TCPS_LAST_ACK:
1618                                 tp = tcp_close(tp);
1619                                 break;
1620
1621                         case TCPS_TIME_WAIT:
1622                                 break;
1623                         }
1624                 }
1625                 goto drop;
1626         }
1627
1628         /*
1629          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1630          * and it's less than ts_recent, drop it.
1631          */
1632         if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1633             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1634
1635                 /* Check to see if ts_recent is over 24 days old.  */
1636                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1637                         /*
1638                          * Invalidate ts_recent.  If this segment updates
1639                          * ts_recent, the age will be reset later and ts_recent
1640                          * will get a valid value.  If it does not, setting
1641                          * ts_recent to zero will at least satisfy the
1642                          * requirement that zero be placed in the timestamp
1643                          * echo reply when ts_recent isn't valid.  The
1644                          * age isn't reset until we get a valid ts_recent
1645                          * because we don't want out-of-order segments to be
1646                          * dropped when ts_recent is old.
1647                          */
1648                         tp->ts_recent = 0;
1649                 } else {
1650                         tcpstat.tcps_rcvduppack++;
1651                         tcpstat.tcps_rcvdupbyte += tlen;
1652                         tcpstat.tcps_pawsdrop++;
1653                         if (tlen)
1654                                 goto dropafterack;
1655                         goto drop;
1656                 }
1657         }
1658
1659         /*
1660          * T/TCP mechanism
1661          *   If T/TCP was negotiated and the segment doesn't have CC,
1662          *   or if its CC is wrong then drop the segment.
1663          *   RST segments do not have to comply with this.
1664          */
1665         if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1666             (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1667                 goto dropafterack;
1668
1669         /*
1670          * In the SYN-RECEIVED state, validate that the packet belongs to
1671          * this connection before trimming the data to fit the receive
1672          * window.  Check the sequence number versus IRS since we know
1673          * the sequence numbers haven't wrapped.  This is a partial fix
1674          * for the "LAND" DoS attack.
1675          */
1676         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1677                 rstreason = BANDLIM_RST_OPENPORT;
1678                 goto dropwithreset;
1679         }
1680
1681         todrop = tp->rcv_nxt - th->th_seq;
1682         if (todrop > 0) {
1683                 if (TCP_DO_SACK(tp)) {
1684                         /* Report duplicate segment at head of packet. */
1685                         tp->reportblk.rblk_start = th->th_seq;
1686                         tp->reportblk.rblk_end = th->th_seq + tlen;
1687                         if (thflags & TH_FIN)
1688                                 ++tp->reportblk.rblk_end;
1689                         if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1690                                 tp->reportblk.rblk_end = tp->rcv_nxt;
1691                         tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1692                 }
1693                 if (thflags & TH_SYN) {
1694                         thflags &= ~TH_SYN;
1695                         th->th_seq++;
1696                         if (th->th_urp > 1)
1697                                 th->th_urp--;
1698                         else
1699                                 thflags &= ~TH_URG;
1700                         todrop--;
1701                 }
1702                 /*
1703                  * Following if statement from Stevens, vol. 2, p. 960.
1704                  */
1705                 if (todrop > tlen ||
1706                     (todrop == tlen && !(thflags & TH_FIN))) {
1707                         /*
1708                          * Any valid FIN must be to the left of the window.
1709                          * At this point the FIN must be a duplicate or out
1710                          * of sequence; drop it.
1711                          */
1712                         thflags &= ~TH_FIN;
1713
1714                         /*
1715                          * Send an ACK to resynchronize and drop any data.
1716                          * But keep on processing for RST or ACK.
1717                          */
1718                         tp->t_flags |= TF_ACKNOW;
1719                         todrop = tlen;
1720                         tcpstat.tcps_rcvduppack++;
1721                         tcpstat.tcps_rcvdupbyte += todrop;
1722                 } else {
1723                         tcpstat.tcps_rcvpartduppack++;
1724                         tcpstat.tcps_rcvpartdupbyte += todrop;
1725                 }
1726                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1727                 th->th_seq += todrop;
1728                 tlen -= todrop;
1729                 if (th->th_urp > todrop)
1730                         th->th_urp -= todrop;
1731                 else {
1732                         thflags &= ~TH_URG;
1733                         th->th_urp = 0;
1734                 }
1735         }
1736
1737         /*
1738          * If new data are received on a connection after the
1739          * user processes are gone, then RST the other end.
1740          */
1741         if ((so->so_state & SS_NOFDREF) &&
1742             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1743                 tp = tcp_close(tp);
1744                 tcpstat.tcps_rcvafterclose++;
1745                 rstreason = BANDLIM_UNLIMITED;
1746                 goto dropwithreset;
1747         }
1748
1749         /*
1750          * If segment ends after window, drop trailing data
1751          * (and PUSH and FIN); if nothing left, just ACK.
1752          */
1753         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1754         if (todrop > 0) {
1755                 tcpstat.tcps_rcvpackafterwin++;
1756                 if (todrop >= tlen) {
1757                         tcpstat.tcps_rcvbyteafterwin += tlen;
1758                         /*
1759                          * If a new connection request is received
1760                          * while in TIME_WAIT, drop the old connection
1761                          * and start over if the sequence numbers
1762                          * are above the previous ones.
1763                          */
1764                         if (thflags & TH_SYN &&
1765                             tp->t_state == TCPS_TIME_WAIT &&
1766                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1767                                 tp = tcp_close(tp);
1768                                 goto findpcb;
1769                         }
1770                         /*
1771                          * If window is closed can only take segments at
1772                          * window edge, and have to drop data and PUSH from
1773                          * incoming segments.  Continue processing, but
1774                          * remember to ack.  Otherwise, drop segment
1775                          * and ack.
1776                          */
1777                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1778                                 tp->t_flags |= TF_ACKNOW;
1779                                 tcpstat.tcps_rcvwinprobe++;
1780                         } else
1781                                 goto dropafterack;
1782                 } else
1783                         tcpstat.tcps_rcvbyteafterwin += todrop;
1784                 m_adj(m, -todrop);
1785                 tlen -= todrop;
1786                 thflags &= ~(TH_PUSH | TH_FIN);
1787         }
1788
1789         /*
1790          * If last ACK falls within this segment's sequence numbers,
1791          * record its timestamp.
1792          * NOTE:
1793          * 1) That the test incorporates suggestions from the latest
1794          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1795          * 2) That updating only on newer timestamps interferes with
1796          *    our earlier PAWS tests, so this check should be solely
1797          *    predicated on the sequence space of this segment.
1798          * 3) That we modify the segment boundary check to be
1799          *        Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1800          *    instead of RFC1323's
1801          *        Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1802          *    This modified check allows us to overcome RFC1323's
1803          *    limitations as described in Stevens TCP/IP Illustrated
1804          *    Vol. 2 p.869. In such cases, we can still calculate the
1805          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1806          */
1807         if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1808             SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1809                                         + ((thflags & TH_SYN) != 0)
1810                                         + ((thflags & TH_FIN) != 0)))) {
1811                 tp->ts_recent_age = ticks;
1812                 tp->ts_recent = to.to_tsval;
1813         }
1814
1815         /*
1816          * If a SYN is in the window, then this is an
1817          * error and we send an RST and drop the connection.
1818          */
1819         if (thflags & TH_SYN) {
1820                 tp = tcp_drop(tp, ECONNRESET);
1821                 rstreason = BANDLIM_UNLIMITED;
1822                 goto dropwithreset;
1823         }
1824
1825         /*
1826          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1827          * flag is on (half-synchronized state), then queue data for
1828          * later processing; else drop segment and return.
1829          */
1830         if (!(thflags & TH_ACK)) {
1831                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1832                     (tp->t_flags & TF_NEEDSYN))
1833                         goto step6;
1834                 else
1835                         goto drop;
1836         }
1837
1838         /*
1839          * Ack processing.
1840          */
1841         switch (tp->t_state) {
1842         /*
1843          * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1844          * ESTABLISHED state and continue processing.
1845          * The ACK was checked above.
1846          */
1847         case TCPS_SYN_RECEIVED:
1848
1849                 tcpstat.tcps_connects++;
1850                 soisconnected(so);
1851                 /* Do window scaling? */
1852                 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1853                     (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1854                         tp->snd_scale = tp->requested_s_scale;
1855                         tp->rcv_scale = tp->request_r_scale;
1856                 }
1857                 /*
1858                  * Upon successful completion of 3-way handshake,
1859                  * update cache.CC if it was undefined, pass any queued
1860                  * data to the user, and advance state appropriately.
1861                  */
1862                 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1863                     taop->tao_cc == 0)
1864                         taop->tao_cc = tp->cc_recv;
1865
1866                 /*
1867                  * Make transitions:
1868                  *      SYN-RECEIVED  -> ESTABLISHED
1869                  *      SYN-RECEIVED* -> FIN-WAIT-1
1870                  */
1871                 tp->t_starttime = ticks;
1872                 if (tp->t_flags & TF_NEEDFIN) {
1873                         tp->t_state = TCPS_FIN_WAIT_1;
1874                         tp->t_flags &= ~TF_NEEDFIN;
1875                 } else {
1876                         tp->t_state = TCPS_ESTABLISHED;
1877                         tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1878                             tcp_timer_keep);
1879                 }
1880                 /*
1881                  * If segment contains data or ACK, will call tcp_reass()
1882                  * later; if not, do so now to pass queued data to user.
1883                  */
1884                 if (tlen == 0 && !(thflags & TH_FIN))
1885                         tcp_reass(tp, NULL, NULL, NULL);
1886                 /* fall into ... */
1887
1888         /*
1889          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1890          * ACKs.  If the ack is in the range
1891          *      tp->snd_una < th->th_ack <= tp->snd_max
1892          * then advance tp->snd_una to th->th_ack and drop
1893          * data from the retransmission queue.  If this ACK reflects
1894          * more up to date window information we update our window information.
1895          */
1896         case TCPS_ESTABLISHED:
1897         case TCPS_FIN_WAIT_1:
1898         case TCPS_FIN_WAIT_2:
1899         case TCPS_CLOSE_WAIT:
1900         case TCPS_CLOSING:
1901         case TCPS_LAST_ACK:
1902         case TCPS_TIME_WAIT:
1903
1904                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1905                         if (TCP_DO_SACK(tp))
1906                                 tcp_sack_update_scoreboard(tp, &to);
1907                         if (tlen != 0 || tiwin != tp->snd_wnd) {
1908                                 tp->t_dupacks = 0;
1909                                 break;
1910                         }
1911                         tcpstat.tcps_rcvdupack++;
1912                         if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1913                             th->th_ack != tp->snd_una) {
1914                                 tp->t_dupacks = 0;
1915                                 break;
1916                         }
1917                         /*
1918                          * We have outstanding data (other than
1919                          * a window probe), this is a completely
1920                          * duplicate ack (ie, window info didn't
1921                          * change), the ack is the biggest we've
1922                          * seen and we've seen exactly our rexmt
1923                          * threshhold of them, so assume a packet
1924                          * has been dropped and retransmit it.
1925                          * Kludge snd_nxt & the congestion
1926                          * window so we send only this one
1927                          * packet.
1928                          */
1929                         if (IN_FASTRECOVERY(tp)) {
1930                                 if (TCP_DO_SACK(tp)) {
1931                                         /* No artifical cwnd inflation. */
1932                                         tcp_sack_rexmt(tp, th);
1933                                 } else {
1934                                         /*
1935                                          * Dup acks mean that packets
1936                                          * have left the network
1937                                          * (they're now cached at the
1938                                          * receiver) so bump cwnd by
1939                                          * the amount in the receiver
1940                                          * to keep a constant cwnd
1941                                          * packets in the network.
1942                                          */
1943                                         tp->snd_cwnd += tp->t_maxseg;
1944                                         tcp_output(tp);
1945                                 }
1946                         } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1947                                 tp->t_dupacks = 0;
1948                                 break;
1949                         } else if (++tp->t_dupacks == tcprexmtthresh) {
1950                                 tcp_seq old_snd_nxt;
1951                                 u_int win;
1952
1953 fastretransmit:
1954                                 if (tcp_do_eifel_detect &&
1955                                     (tp->t_flags & TF_RCVD_TSTMP)) {
1956                                         tcp_save_congestion_state(tp);
1957                                         tp->t_flags |= TF_FASTREXMT;
1958                                 }
1959                                 /*
1960                                  * We know we're losing at the current
1961                                  * window size, so do congestion avoidance:
1962                                  * set ssthresh to half the current window
1963                                  * and pull our congestion window back to the
1964                                  * new ssthresh.
1965                                  */
1966                                 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1967                                     tp->t_maxseg;
1968                                 if (win < 2)
1969                                         win = 2;
1970                                 tp->snd_ssthresh = win * tp->t_maxseg;
1971                                 ENTER_FASTRECOVERY(tp);
1972                                 tp->snd_recover = tp->snd_max;
1973                                 tcp_callout_stop(tp, tp->tt_rexmt);
1974                                 tp->t_rtttime = 0;
1975                                 old_snd_nxt = tp->snd_nxt;
1976                                 tp->snd_nxt = th->th_ack;
1977                                 tp->snd_cwnd = tp->t_maxseg;
1978                                 tcp_output(tp);
1979                                 ++tcpstat.tcps_sndfastrexmit;
1980                                 tp->snd_cwnd = tp->snd_ssthresh;
1981                                 tp->rexmt_high = tp->snd_nxt;
1982                                 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1983                                         tp->snd_nxt = old_snd_nxt;
1984                                 KASSERT(tp->snd_limited <= 2,
1985                                     ("tp->snd_limited too big"));
1986                                 if (TCP_DO_SACK(tp))
1987                                         tcp_sack_rexmt(tp, th);
1988                                 else
1989                                         tp->snd_cwnd += tp->t_maxseg *
1990                                             (tp->t_dupacks - tp->snd_limited);
1991                         } else if (tcp_do_limitedtransmit) {
1992                                 u_long oldcwnd = tp->snd_cwnd;
1993                                 tcp_seq oldsndmax = tp->snd_max;
1994                                 tcp_seq oldsndnxt = tp->snd_nxt;
1995                                 /* outstanding data */
1996                                 uint32_t ownd = tp->snd_max - tp->snd_una;
1997                                 u_int sent;
1998
1999 #define iceildiv(n, d)          (((n)+(d)-1) / (d))
2000
2001                                 KASSERT(tp->t_dupacks == 1 ||
2002                                         tp->t_dupacks == 2,
2003                                     ("dupacks not 1 or 2"));
2004                                 if (tp->t_dupacks == 1)
2005                                         tp->snd_limited = 0;
2006                                 tp->snd_nxt = tp->snd_max;
2007                                 tp->snd_cwnd = ownd +
2008                                     (tp->t_dupacks - tp->snd_limited) *
2009                                     tp->t_maxseg;
2010                                 tcp_output(tp);
2011
2012                                 /*
2013                                  * Other acks may have been processed,
2014                                  * snd_nxt cannot be reset to a value less
2015                                  * then snd_una.
2016                                  */
2017                                 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2018                                     if (SEQ_GT(oldsndnxt, tp->snd_una))
2019                                         tp->snd_nxt = oldsndnxt;
2020                                     else
2021                                         tp->snd_nxt = tp->snd_una;
2022                                 }
2023                                 tp->snd_cwnd = oldcwnd;
2024                                 sent = tp->snd_max - oldsndmax;
2025                                 if (sent > tp->t_maxseg) {
2026                                         KASSERT((tp->t_dupacks == 2 &&
2027                                                  tp->snd_limited == 0) ||
2028                                                 (sent == tp->t_maxseg + 1 &&
2029                                                  tp->t_flags & TF_SENTFIN),
2030                                             ("sent too much"));
2031                                         KASSERT(sent <= tp->t_maxseg * 2,
2032                                             ("sent too many segments"));
2033                                         tp->snd_limited = 2;
2034                                         tcpstat.tcps_sndlimited += 2;
2035                                 } else if (sent > 0) {
2036                                         ++tp->snd_limited;
2037                                         ++tcpstat.tcps_sndlimited;
2038                                 } else if (tcp_do_early_retransmit &&
2039                                     (tcp_do_eifel_detect &&
2040                                      (tp->t_flags & TF_RCVD_TSTMP)) &&
2041                                     ownd < 4 * tp->t_maxseg &&
2042                                     tp->t_dupacks + 1 >=
2043                                       iceildiv(ownd, tp->t_maxseg) &&
2044                                     (!TCP_DO_SACK(tp) ||
2045                                      ownd <= tp->t_maxseg ||
2046                                      tcp_sack_has_sacked(&tp->scb,
2047                                                         ownd - tp->t_maxseg))) {
2048                                         ++tcpstat.tcps_sndearlyrexmit;
2049                                         tp->t_flags |= TF_EARLYREXMT;
2050                                         goto fastretransmit;
2051                                 }
2052                         }
2053                         goto drop;
2054                 }
2055
2056                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2057                 tp->t_dupacks = 0;
2058                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2059                         /*
2060                          * Detected optimistic ACK attack.
2061                          * Force slow-start to de-synchronize attack.
2062                          */
2063                         tp->snd_cwnd = tp->t_maxseg;
2064                         tp->snd_wacked = 0;
2065
2066                         tcpstat.tcps_rcvacktoomuch++;
2067                         goto dropafterack;
2068                 }
2069                 /*
2070                  * If we reach this point, ACK is not a duplicate,
2071                  *     i.e., it ACKs something we sent.
2072                  */
2073                 if (tp->t_flags & TF_NEEDSYN) {
2074                         /*
2075                          * T/TCP: Connection was half-synchronized, and our
2076                          * SYN has been ACK'd (so connection is now fully
2077                          * synchronized).  Go to non-starred state,
2078                          * increment snd_una for ACK of SYN, and check if
2079                          * we can do window scaling.
2080                          */
2081                         tp->t_flags &= ~TF_NEEDSYN;
2082                         tp->snd_una++;
2083                         /* Do window scaling? */
2084                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2085                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2086                                 tp->snd_scale = tp->requested_s_scale;
2087                                 tp->rcv_scale = tp->request_r_scale;
2088                         }
2089                 }
2090
2091 process_ACK:
2092                 acked = th->th_ack - tp->snd_una;
2093                 tcpstat.tcps_rcvackpack++;
2094                 tcpstat.tcps_rcvackbyte += acked;
2095
2096                 if (tcp_do_eifel_detect && acked > 0 &&
2097                     (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2098                     (tp->t_flags & TF_FIRSTACCACK)) {
2099                         /* Eifel detection applicable. */
2100                         if (to.to_tsecr < tp->t_rexmtTS) {
2101                                 ++tcpstat.tcps_eifeldetected;
2102                                 tcp_revert_congestion_state(tp);
2103                                 if (tp->t_rxtshift == 1 &&
2104                                     ticks >= tp->t_badrxtwin)
2105                                         ++tcpstat.tcps_rttcantdetect;
2106                         }
2107                 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2108                         /*
2109                          * If we just performed our first retransmit,
2110                          * and the ACK arrives within our recovery window,
2111                          * then it was a mistake to do the retransmit
2112                          * in the first place.  Recover our original cwnd
2113                          * and ssthresh, and proceed to transmit where we
2114                          * left off.
2115                          */
2116                         tcp_revert_congestion_state(tp);
2117                         ++tcpstat.tcps_rttdetected;
2118                 }
2119
2120                 /*
2121                  * If we have a timestamp reply, update smoothed
2122                  * round trip time.  If no timestamp is present but
2123                  * transmit timer is running and timed sequence
2124                  * number was acked, update smoothed round trip time.
2125                  * Since we now have an rtt measurement, cancel the
2126                  * timer backoff (cf., Phil Karn's retransmit alg.).
2127                  * Recompute the initial retransmit timer.
2128                  *
2129                  * Some machines (certain windows boxes) send broken
2130                  * timestamp replies during the SYN+ACK phase, ignore
2131                  * timestamps of 0.
2132                  */
2133                 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2134                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2135                 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2136                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2137                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2138
2139                 /*
2140                  * If no data (only SYN) was ACK'd,
2141                  *    skip rest of ACK processing.
2142                  */
2143                 if (acked == 0)
2144                         goto step6;
2145
2146                 /* Stop looking for an acceptable ACK since one was received. */
2147                 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2148
2149                 if (acked > so->so_snd.ssb_cc) {
2150                         tp->snd_wnd -= so->so_snd.ssb_cc;
2151                         sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2152                         ourfinisacked = TRUE;
2153                 } else {
2154                         sbdrop(&so->so_snd.sb, acked);
2155                         tp->snd_wnd -= acked;
2156                         ourfinisacked = FALSE;
2157                 }
2158                 sowwakeup(so);
2159
2160                 /*
2161                  * Update window information.
2162                  * Don't look at window if no ACK:
2163                  * TAC's send garbage on first SYN.
2164                  */
2165                 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2166                     (tp->snd_wl1 == th->th_seq &&
2167                      (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2168                       (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2169                         /* keep track of pure window updates */
2170                         if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2171                             tiwin > tp->snd_wnd)
2172                                 tcpstat.tcps_rcvwinupd++;
2173                         tp->snd_wnd = tiwin;
2174                         tp->snd_wl1 = th->th_seq;
2175                         tp->snd_wl2 = th->th_ack;
2176                         if (tp->snd_wnd > tp->max_sndwnd)
2177                                 tp->max_sndwnd = tp->snd_wnd;
2178                         needoutput = TRUE;
2179                 }
2180
2181                 tp->snd_una = th->th_ack;
2182                 if (TCP_DO_SACK(tp))
2183                         tcp_sack_update_scoreboard(tp, &to);
2184                 if (IN_FASTRECOVERY(tp)) {
2185                         if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2186                                 EXIT_FASTRECOVERY(tp);
2187                                 needoutput = TRUE;
2188                                 /*
2189                                  * If the congestion window was inflated
2190                                  * to account for the other side's
2191                                  * cached packets, retract it.
2192                                  */
2193                                 if (!TCP_DO_SACK(tp))
2194                                         tp->snd_cwnd = tp->snd_ssthresh;
2195
2196                                 /*
2197                                  * Window inflation should have left us
2198                                  * with approximately snd_ssthresh outstanding
2199                                  * data.  But, in case we would be inclined
2200                                  * to send a burst, better do it using
2201                                  * slow start.
2202                                  */
2203                                 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2204                                            tp->snd_max + 2 * tp->t_maxseg))
2205                                         tp->snd_cwnd =
2206                                             (tp->snd_max - tp->snd_una) +
2207                                             2 * tp->t_maxseg;
2208
2209                                 tp->snd_wacked = 0;
2210                         } else {
2211                                 if (TCP_DO_SACK(tp)) {
2212                                         tp->snd_max_rexmt = tp->snd_max;
2213                                         tcp_sack_rexmt(tp, th);
2214                                 } else {
2215                                         tcp_newreno_partial_ack(tp, th, acked);
2216                                 }
2217                                 needoutput = FALSE;
2218                         }
2219                 } else {
2220                         /*
2221                          * Open the congestion window.  When in slow-start,
2222                          * open exponentially: maxseg per packet.  Otherwise,
2223                          * open linearly: maxseg per window.
2224                          */
2225                         if (tp->snd_cwnd <= tp->snd_ssthresh) {
2226                                 u_int abc_sslimit =
2227                                     (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2228                                      tp->t_maxseg : 2 * tp->t_maxseg);
2229
2230                                 /* slow-start */
2231                                 tp->snd_cwnd += tcp_do_abc ?
2232                                     min(acked, abc_sslimit) : tp->t_maxseg;
2233                         } else {
2234                                 /* linear increase */
2235                                 tp->snd_wacked += tcp_do_abc ? acked :
2236                                     tp->t_maxseg;
2237                                 if (tp->snd_wacked >= tp->snd_cwnd) {
2238                                         tp->snd_wacked -= tp->snd_cwnd;
2239                                         tp->snd_cwnd += tp->t_maxseg;
2240                                 }
2241                         }
2242                         tp->snd_cwnd = min(tp->snd_cwnd,
2243                                            TCP_MAXWIN << tp->snd_scale);
2244                         tp->snd_recover = th->th_ack - 1;
2245                 }
2246                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2247                         tp->snd_nxt = tp->snd_una;
2248
2249                 /*
2250                  * If all outstanding data is acked, stop retransmit
2251                  * timer and remember to restart (more output or persist).
2252                  * If there is more data to be acked, restart retransmit
2253                  * timer, using current (possibly backed-off) value.
2254                  */
2255                 if (th->th_ack == tp->snd_max) {
2256                         tcp_callout_stop(tp, tp->tt_rexmt);
2257                         needoutput = TRUE;
2258                 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2259                         tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2260                             tcp_timer_rexmt);
2261                 }
2262
2263                 switch (tp->t_state) {
2264                 /*
2265                  * In FIN_WAIT_1 STATE in addition to the processing
2266                  * for the ESTABLISHED state if our FIN is now acknowledged
2267                  * then enter FIN_WAIT_2.
2268                  */
2269                 case TCPS_FIN_WAIT_1:
2270                         if (ourfinisacked) {
2271                                 /*
2272                                  * If we can't receive any more
2273                                  * data, then closing user can proceed.
2274                                  * Starting the timer is contrary to the
2275                                  * specification, but if we don't get a FIN
2276                                  * we'll hang forever.
2277                                  */
2278                                 if (so->so_state & SS_CANTRCVMORE) {
2279                                         soisdisconnected(so);
2280                                         tcp_callout_reset(tp, tp->tt_2msl,
2281                                             tcp_maxidle, tcp_timer_2msl);
2282                                 }
2283                                 tp->t_state = TCPS_FIN_WAIT_2;
2284                         }
2285                         break;
2286
2287                 /*
2288                  * In CLOSING STATE in addition to the processing for
2289                  * the ESTABLISHED state if the ACK acknowledges our FIN
2290                  * then enter the TIME-WAIT state, otherwise ignore
2291                  * the segment.
2292                  */
2293                 case TCPS_CLOSING:
2294                         if (ourfinisacked) {
2295                                 tp->t_state = TCPS_TIME_WAIT;
2296                                 tcp_canceltimers(tp);
2297                                 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2298                                 if (tp->cc_recv != 0 &&
2299                                     (ticks - tp->t_starttime) < tcp_msl) {
2300                                         tcp_callout_reset(tp, tp->tt_2msl,
2301                                             tp->t_rxtcur * TCPTV_TWTRUNC,
2302                                             tcp_timer_2msl);
2303                                 } else {
2304                                         tcp_callout_reset(tp, tp->tt_2msl,
2305                                             2 * tcp_msl, tcp_timer_2msl);
2306                                 }
2307                                 soisdisconnected(so);
2308                         }
2309                         break;
2310
2311                 /*
2312                  * In LAST_ACK, we may still be waiting for data to drain
2313                  * and/or to be acked, as well as for the ack of our FIN.
2314                  * If our FIN is now acknowledged, delete the TCB,
2315                  * enter the closed state and return.
2316                  */
2317                 case TCPS_LAST_ACK:
2318                         if (ourfinisacked) {
2319                                 tp = tcp_close(tp);
2320                                 goto drop;
2321                         }
2322                         break;
2323
2324                 /*
2325                  * In TIME_WAIT state the only thing that should arrive
2326                  * is a retransmission of the remote FIN.  Acknowledge
2327                  * it and restart the finack timer.
2328                  */
2329                 case TCPS_TIME_WAIT:
2330                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2331                             tcp_timer_2msl);
2332                         goto dropafterack;
2333                 }
2334         }
2335
2336 step6:
2337         /*
2338          * Update window information.
2339          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2340          */
2341         if ((thflags & TH_ACK) &&
2342             acceptable_window_update(tp, th, tiwin)) {
2343                 /* keep track of pure window updates */
2344                 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2345                     tiwin > tp->snd_wnd)
2346                         tcpstat.tcps_rcvwinupd++;
2347                 tp->snd_wnd = tiwin;
2348                 tp->snd_wl1 = th->th_seq;
2349                 tp->snd_wl2 = th->th_ack;
2350                 if (tp->snd_wnd > tp->max_sndwnd)
2351                         tp->max_sndwnd = tp->snd_wnd;
2352                 needoutput = TRUE;
2353         }
2354
2355         /*
2356          * Process segments with URG.
2357          */
2358         if ((thflags & TH_URG) && th->th_urp &&
2359             !TCPS_HAVERCVDFIN(tp->t_state)) {
2360                 /*
2361                  * This is a kludge, but if we receive and accept
2362                  * random urgent pointers, we'll crash in
2363                  * soreceive.  It's hard to imagine someone
2364                  * actually wanting to send this much urgent data.
2365                  */
2366                 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2367                         th->th_urp = 0;                 /* XXX */
2368                         thflags &= ~TH_URG;             /* XXX */
2369                         goto dodata;                    /* XXX */
2370                 }
2371                 /*
2372                  * If this segment advances the known urgent pointer,
2373                  * then mark the data stream.  This should not happen
2374                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2375                  * a FIN has been received from the remote side.
2376                  * In these states we ignore the URG.
2377                  *
2378                  * According to RFC961 (Assigned Protocols),
2379                  * the urgent pointer points to the last octet
2380                  * of urgent data.  We continue, however,
2381                  * to consider it to indicate the first octet
2382                  * of data past the urgent section as the original
2383                  * spec states (in one of two places).
2384                  */
2385                 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2386                         tp->rcv_up = th->th_seq + th->th_urp;
2387                         so->so_oobmark = so->so_rcv.ssb_cc +
2388                             (tp->rcv_up - tp->rcv_nxt) - 1;
2389                         if (so->so_oobmark == 0)
2390                                 so->so_state |= SS_RCVATMARK;
2391                         sohasoutofband(so);
2392                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2393                 }
2394                 /*
2395                  * Remove out of band data so doesn't get presented to user.
2396                  * This can happen independent of advancing the URG pointer,
2397                  * but if two URG's are pending at once, some out-of-band
2398                  * data may creep in... ick.
2399                  */
2400                 if (th->th_urp <= (u_long)tlen &&
2401                     !(so->so_options & SO_OOBINLINE)) {
2402                         /* hdr drop is delayed */
2403                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2404                 }
2405         } else {
2406                 /*
2407                  * If no out of band data is expected,
2408                  * pull receive urgent pointer along
2409                  * with the receive window.
2410                  */
2411                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2412                         tp->rcv_up = tp->rcv_nxt;
2413         }
2414
2415 dodata:                                                 /* XXX */
2416         /*
2417          * Process the segment text, merging it into the TCP sequencing queue,
2418          * and arranging for acknowledgment of receipt if necessary.
2419          * This process logically involves adjusting tp->rcv_wnd as data
2420          * is presented to the user (this happens in tcp_usrreq.c,
2421          * case PRU_RCVD).  If a FIN has already been received on this
2422          * connection then we just ignore the text.
2423          */
2424         if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2425                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2426                 /*
2427                  * Insert segment which includes th into TCP reassembly queue
2428                  * with control block tp.  Set thflags to whether reassembly now
2429                  * includes a segment with FIN.  This handles the common case
2430                  * inline (segment is the next to be received on an established
2431                  * connection, and the queue is empty), avoiding linkage into
2432                  * and removal from the queue and repetition of various
2433                  * conversions.
2434                  * Set DELACK for segments received in order, but ack
2435                  * immediately when segments are out of order (so
2436                  * fast retransmit can work).
2437                  */
2438                 if (th->th_seq == tp->rcv_nxt &&
2439                     LIST_EMPTY(&tp->t_segq) &&
2440                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2441                         if (DELAY_ACK(tp)) {
2442                                 tcp_callout_reset(tp, tp->tt_delack,
2443                                     tcp_delacktime, tcp_timer_delack);
2444                         } else {
2445                                 tp->t_flags |= TF_ACKNOW;
2446                         }
2447                         tp->rcv_nxt += tlen;
2448                         thflags = th->th_flags & TH_FIN;
2449                         tcpstat.tcps_rcvpack++;
2450                         tcpstat.tcps_rcvbyte += tlen;
2451                         ND6_HINT(tp);
2452                         if (so->so_state & SS_CANTRCVMORE)
2453                                 m_freem(m);
2454                         else
2455                                 ssb_appendstream(&so->so_rcv, m);
2456                         sorwakeup(so);
2457                 } else {
2458                         if (!(tp->t_flags & TF_DUPSEG)) {
2459                                 /* Initialize SACK report block. */
2460                                 tp->reportblk.rblk_start = th->th_seq;
2461                                 tp->reportblk.rblk_end = th->th_seq + tlen +
2462                                     ((thflags & TH_FIN) != 0);
2463                         }
2464                         thflags = tcp_reass(tp, th, &tlen, m);
2465                         tp->t_flags |= TF_ACKNOW;
2466                 }
2467
2468                 /*
2469                  * Note the amount of data that peer has sent into
2470                  * our window, in order to estimate the sender's
2471                  * buffer size.
2472                  */
2473                 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2474         } else {
2475                 m_freem(m);
2476                 thflags &= ~TH_FIN;
2477         }
2478
2479         /*
2480          * If FIN is received ACK the FIN and let the user know
2481          * that the connection is closing.
2482          */
2483         if (thflags & TH_FIN) {
2484                 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2485                         socantrcvmore(so);
2486                         /*
2487                          * If connection is half-synchronized
2488                          * (ie NEEDSYN flag on) then delay ACK,
2489                          * so it may be piggybacked when SYN is sent.
2490                          * Otherwise, since we received a FIN then no
2491                          * more input can be expected, send ACK now.
2492                          */
2493                         if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2494                                 tcp_callout_reset(tp, tp->tt_delack,
2495                                     tcp_delacktime, tcp_timer_delack);
2496                         } else {
2497                                 tp->t_flags |= TF_ACKNOW;
2498                         }
2499                         tp->rcv_nxt++;
2500                 }
2501
2502                 switch (tp->t_state) {
2503                 /*
2504                  * In SYN_RECEIVED and ESTABLISHED STATES
2505                  * enter the CLOSE_WAIT state.
2506                  */
2507                 case TCPS_SYN_RECEIVED:
2508                         tp->t_starttime = ticks;
2509                         /*FALLTHROUGH*/
2510                 case TCPS_ESTABLISHED:
2511                         tp->t_state = TCPS_CLOSE_WAIT;
2512                         break;
2513
2514                 /*
2515                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2516                  * enter the CLOSING state.
2517                  */
2518                 case TCPS_FIN_WAIT_1:
2519                         tp->t_state = TCPS_CLOSING;
2520                         break;
2521
2522                 /*
2523                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2524                  * starting the time-wait timer, turning off the other
2525                  * standard timers.
2526                  */
2527                 case TCPS_FIN_WAIT_2:
2528                         tp->t_state = TCPS_TIME_WAIT;
2529                         tcp_canceltimers(tp);
2530                         /* Shorten TIME_WAIT [RFC-1644, p.28] */
2531                         if (tp->cc_recv != 0 &&
2532                             (ticks - tp->t_starttime) < tcp_msl) {
2533                                 tcp_callout_reset(tp, tp->tt_2msl,
2534                                     tp->t_rxtcur * TCPTV_TWTRUNC,
2535                                     tcp_timer_2msl);
2536                                 /* For transaction client, force ACK now. */
2537                                 tp->t_flags |= TF_ACKNOW;
2538                         } else {
2539                                 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2540                                     tcp_timer_2msl);
2541                         }
2542                         soisdisconnected(so);
2543                         break;
2544
2545                 /*
2546                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2547                  */
2548                 case TCPS_TIME_WAIT:
2549                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2550                             tcp_timer_2msl);
2551                         break;
2552                 }
2553         }
2554
2555 #ifdef TCPDEBUG
2556         if (so->so_options & SO_DEBUG)
2557                 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2558 #endif
2559
2560         /*
2561          * Return any desired output.
2562          */
2563         if (needoutput || (tp->t_flags & TF_ACKNOW))
2564                 tcp_output(tp);
2565         return;
2566
2567 dropafterack:
2568         /*
2569          * Generate an ACK dropping incoming segment if it occupies
2570          * sequence space, where the ACK reflects our state.
2571          *
2572          * We can now skip the test for the RST flag since all
2573          * paths to this code happen after packets containing
2574          * RST have been dropped.
2575          *
2576          * In the SYN-RECEIVED state, don't send an ACK unless the
2577          * segment we received passes the SYN-RECEIVED ACK test.
2578          * If it fails send a RST.  This breaks the loop in the
2579          * "LAND" DoS attack, and also prevents an ACK storm
2580          * between two listening ports that have been sent forged
2581          * SYN segments, each with the source address of the other.
2582          */
2583         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2584             (SEQ_GT(tp->snd_una, th->th_ack) ||
2585              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2586                 rstreason = BANDLIM_RST_OPENPORT;
2587                 goto dropwithreset;
2588         }
2589 #ifdef TCPDEBUG
2590         if (so->so_options & SO_DEBUG)
2591                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2592 #endif
2593         m_freem(m);
2594         tp->t_flags |= TF_ACKNOW;
2595         tcp_output(tp);
2596         return;
2597
2598 dropwithreset:
2599         /*
2600          * Generate a RST, dropping incoming segment.
2601          * Make ACK acceptable to originator of segment.
2602          * Don't bother to respond if destination was broadcast/multicast.
2603          */
2604         if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2605                 goto drop;
2606         if (isipv6) {
2607                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2608                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2609                         goto drop;
2610         } else {
2611                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2612                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2613                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2614                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2615                         goto drop;
2616         }
2617         /* IPv6 anycast check is done at tcp6_input() */
2618
2619         /*
2620          * Perform bandwidth limiting.
2621          */
2622 #ifdef ICMP_BANDLIM
2623         if (badport_bandlim(rstreason) < 0)
2624                 goto drop;
2625 #endif
2626
2627 #ifdef TCPDEBUG
2628         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2629                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2630 #endif
2631         if (thflags & TH_ACK)
2632                 /* mtod() below is safe as long as hdr dropping is delayed */
2633                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2634                             TH_RST);
2635         else {
2636                 if (thflags & TH_SYN)
2637                         tlen++;
2638                 /* mtod() below is safe as long as hdr dropping is delayed */
2639                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2640                             (tcp_seq)0, TH_RST | TH_ACK);
2641         }
2642         return;
2643
2644 drop:
2645         /*
2646          * Drop space held by incoming segment and return.
2647          */
2648 #ifdef TCPDEBUG
2649         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2650                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2651 #endif
2652         m_freem(m);
2653         return;
2654 }
2655
2656 /*
2657  * Parse TCP options and place in tcpopt.
2658  */
2659 static void
2660 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2661 {
2662         int opt, optlen, i;
2663
2664         to->to_flags = 0;
2665         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2666                 opt = cp[0];
2667                 if (opt == TCPOPT_EOL)
2668                         break;
2669                 if (opt == TCPOPT_NOP)
2670                         optlen = 1;
2671                 else {
2672                         if (cnt < 2)
2673                                 break;
2674                         optlen = cp[1];
2675                         if (optlen < 2 || optlen > cnt)
2676                                 break;
2677                 }
2678                 switch (opt) {
2679                 case TCPOPT_MAXSEG:
2680                         if (optlen != TCPOLEN_MAXSEG)
2681                                 continue;
2682                         if (!is_syn)
2683                                 continue;
2684                         to->to_flags |= TOF_MSS;
2685                         bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2686                         to->to_mss = ntohs(to->to_mss);
2687                         break;
2688                 case TCPOPT_WINDOW:
2689                         if (optlen != TCPOLEN_WINDOW)
2690                                 continue;
2691                         if (!is_syn)
2692                                 continue;
2693                         to->to_flags |= TOF_SCALE;
2694                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2695                         break;
2696                 case TCPOPT_TIMESTAMP:
2697                         if (optlen != TCPOLEN_TIMESTAMP)
2698                                 continue;
2699                         to->to_flags |= TOF_TS;
2700                         bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2701                         to->to_tsval = ntohl(to->to_tsval);
2702                         bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2703                         to->to_tsecr = ntohl(to->to_tsecr);
2704                         /*
2705                          * If echoed timestamp is later than the current time,
2706                          * fall back to non RFC1323 RTT calculation.
2707                          */
2708                         if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2709                                 to->to_tsecr = 0;
2710                         break;
2711                 case TCPOPT_CC:
2712                         if (optlen != TCPOLEN_CC)
2713                                 continue;
2714                         to->to_flags |= TOF_CC;
2715                         bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2716                         to->to_cc = ntohl(to->to_cc);
2717                         break;
2718                 case TCPOPT_CCNEW:
2719                         if (optlen != TCPOLEN_CC)
2720                                 continue;
2721                         if (!is_syn)
2722                                 continue;
2723                         to->to_flags |= TOF_CCNEW;
2724                         bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2725                         to->to_cc = ntohl(to->to_cc);
2726                         break;
2727                 case TCPOPT_CCECHO:
2728                         if (optlen != TCPOLEN_CC)
2729                                 continue;
2730                         if (!is_syn)
2731                                 continue;
2732                         to->to_flags |= TOF_CCECHO;
2733                         bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho);
2734                         to->to_ccecho = ntohl(to->to_ccecho);
2735                         break;
2736                 case TCPOPT_SACK_PERMITTED:
2737                         if (optlen != TCPOLEN_SACK_PERMITTED)
2738                                 continue;
2739                         if (!is_syn)
2740                                 continue;
2741                         to->to_flags |= TOF_SACK_PERMITTED;
2742                         break;
2743                 case TCPOPT_SACK:
2744                         if ((optlen - 2) & 0x07)        /* not multiple of 8 */
2745                                 continue;
2746                         to->to_nsackblocks = (optlen - 2) / 8;
2747                         to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2748                         to->to_flags |= TOF_SACK;
2749                         for (i = 0; i < to->to_nsackblocks; i++) {
2750                                 struct raw_sackblock *r = &to->to_sackblocks[i];
2751
2752                                 r->rblk_start = ntohl(r->rblk_start);
2753                                 r->rblk_end = ntohl(r->rblk_end);
2754                         }
2755                         break;
2756                 default:
2757                         continue;
2758                 }
2759         }
2760 }
2761
2762 /*
2763  * Pull out of band byte out of a segment so
2764  * it doesn't appear in the user's data queue.
2765  * It is still reflected in the segment length for
2766  * sequencing purposes.
2767  * "off" is the delayed to be dropped hdrlen.
2768  */
2769 static void
2770 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2771 {
2772         int cnt = off + th->th_urp - 1;
2773
2774         while (cnt >= 0) {
2775                 if (m->m_len > cnt) {
2776                         char *cp = mtod(m, caddr_t) + cnt;
2777                         struct tcpcb *tp = sototcpcb(so);
2778
2779                         tp->t_iobc = *cp;
2780                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2781                         bcopy(cp + 1, cp, m->m_len - cnt - 1);
2782                         m->m_len--;
2783                         if (m->m_flags & M_PKTHDR)
2784                                 m->m_pkthdr.len--;
2785                         return;
2786                 }
2787                 cnt -= m->m_len;
2788                 m = m->m_next;
2789                 if (m == 0)
2790                         break;
2791         }
2792         panic("tcp_pulloutofband");
2793 }
2794
2795 /*
2796  * Collect new round-trip time estimate
2797  * and update averages and current timeout.
2798  */
2799 static void
2800 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2801 {
2802         int delta;
2803
2804         tcpstat.tcps_rttupdated++;
2805         tp->t_rttupdated++;
2806         if (tp->t_srtt != 0) {
2807                 /*
2808                  * srtt is stored as fixed point with 5 bits after the
2809                  * binary point (i.e., scaled by 8).  The following magic
2810                  * is equivalent to the smoothing algorithm in rfc793 with
2811                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2812                  * point).  Adjust rtt to origin 0.
2813                  */
2814                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2815                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2816
2817                 if ((tp->t_srtt += delta) <= 0)
2818                         tp->t_srtt = 1;
2819
2820                 /*
2821                  * We accumulate a smoothed rtt variance (actually, a
2822                  * smoothed mean difference), then set the retransmit
2823                  * timer to smoothed rtt + 4 times the smoothed variance.
2824                  * rttvar is stored as fixed point with 4 bits after the
2825                  * binary point (scaled by 16).  The following is
2826                  * equivalent to rfc793 smoothing with an alpha of .75
2827                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2828                  * rfc793's wired-in beta.
2829                  */
2830                 if (delta < 0)
2831                         delta = -delta;
2832                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2833                 if ((tp->t_rttvar += delta) <= 0)
2834                         tp->t_rttvar = 1;
2835                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2836                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2837         } else {
2838                 /*
2839                  * No rtt measurement yet - use the unsmoothed rtt.
2840                  * Set the variance to half the rtt (so our first
2841                  * retransmit happens at 3*rtt).
2842                  */
2843                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2844                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2845                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2846         }
2847         tp->t_rtttime = 0;
2848         tp->t_rxtshift = 0;
2849
2850         /*
2851          * the retransmit should happen at rtt + 4 * rttvar.
2852          * Because of the way we do the smoothing, srtt and rttvar
2853          * will each average +1/2 tick of bias.  When we compute
2854          * the retransmit timer, we want 1/2 tick of rounding and
2855          * 1 extra tick because of +-1/2 tick uncertainty in the
2856          * firing of the timer.  The bias will give us exactly the
2857          * 1.5 tick we need.  But, because the bias is
2858          * statistical, we have to test that we don't drop below
2859          * the minimum feasible timer (which is 2 ticks).
2860          */
2861         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2862                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2863
2864         /*
2865          * We received an ack for a packet that wasn't retransmitted;
2866          * it is probably safe to discard any error indications we've
2867          * received recently.  This isn't quite right, but close enough
2868          * for now (a route might have failed after we sent a segment,
2869          * and the return path might not be symmetrical).
2870          */
2871         tp->t_softerror = 0;
2872 }
2873
2874 /*
2875  * Determine a reasonable value for maxseg size.
2876  * If the route is known, check route for mtu.
2877  * If none, use an mss that can be handled on the outgoing
2878  * interface without forcing IP to fragment; if bigger than
2879  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2880  * to utilize large mbufs.  If no route is found, route has no mtu,
2881  * or the destination isn't local, use a default, hopefully conservative
2882  * size (usually 512 or the default IP max size, but no more than the mtu
2883  * of the interface), as we can't discover anything about intervening
2884  * gateways or networks.  We also initialize the congestion/slow start
2885  * window to be a single segment if the destination isn't local.
2886  * While looking at the routing entry, we also initialize other path-dependent
2887  * parameters from pre-set or cached values in the routing entry.
2888  *
2889  * Also take into account the space needed for options that we
2890  * send regularly.  Make maxseg shorter by that amount to assure
2891  * that we can send maxseg amount of data even when the options
2892  * are present.  Store the upper limit of the length of options plus
2893  * data in maxopd.
2894  *
2895  * NOTE that this routine is only called when we process an incoming
2896  * segment, for outgoing segments only tcp_mssopt is called.
2897  *
2898  * In case of T/TCP, we call this routine during implicit connection
2899  * setup as well (offer = -1), to initialize maxseg from the cached
2900  * MSS of our peer.
2901  */
2902 void
2903 tcp_mss(struct tcpcb *tp, int offer)
2904 {
2905         struct rtentry *rt;
2906         struct ifnet *ifp;
2907         int rtt, mss;
2908         u_long bufsize;
2909         struct inpcb *inp = tp->t_inpcb;
2910         struct socket *so;
2911         struct rmxp_tao *taop;
2912         int origoffer = offer;
2913 #ifdef INET6
2914         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2915         size_t min_protoh = isipv6 ?
2916                             sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2917                             sizeof(struct tcpiphdr);
2918 #else
2919         const boolean_t isipv6 = FALSE;
2920         const size_t min_protoh = sizeof(struct tcpiphdr);
2921 #endif
2922
2923         if (isipv6)
2924                 rt = tcp_rtlookup6(&inp->inp_inc);
2925         else
2926                 rt = tcp_rtlookup(&inp->inp_inc);
2927         if (rt == NULL) {
2928                 tp->t_maxopd = tp->t_maxseg =
2929                     (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2930                 return;
2931         }
2932         ifp = rt->rt_ifp;
2933         so = inp->inp_socket;
2934
2935         taop = rmx_taop(rt->rt_rmx);
2936         /*
2937          * Offer == -1 means that we didn't receive SYN yet,
2938          * use cached value in that case;
2939          */
2940         if (offer == -1)
2941                 offer = taop->tao_mssopt;
2942         /*
2943          * Offer == 0 means that there was no MSS on the SYN segment,
2944          * in this case we use tcp_mssdflt.
2945          */
2946         if (offer == 0)
2947                 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2948         else
2949                 /*
2950                  * Sanity check: make sure that maxopd will be large
2951                  * enough to allow some data on segments even is the
2952                  * all the option space is used (40bytes).  Otherwise
2953                  * funny things may happen in tcp_output.
2954                  */
2955                 offer = max(offer, 64);
2956         taop->tao_mssopt = offer;
2957
2958         /*
2959          * While we're here, check if there's an initial rtt
2960          * or rttvar.  Convert from the route-table units
2961          * to scaled multiples of the slow timeout timer.
2962          */
2963         if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2964                 /*
2965                  * XXX the lock bit for RTT indicates that the value
2966                  * is also a minimum value; this is subject to time.
2967                  */
2968                 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2969                         tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2970                 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2971                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2972                 tcpstat.tcps_usedrtt++;
2973                 if (rt->rt_rmx.rmx_rttvar) {
2974                         tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2975                             (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2976                         tcpstat.tcps_usedrttvar++;
2977                 } else {
2978                         /* default variation is +- 1 rtt */
2979                         tp->t_rttvar =
2980                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2981                 }
2982                 TCPT_RANGESET(tp->t_rxtcur,
2983                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2984                               tp->t_rttmin, TCPTV_REXMTMAX);
2985         }
2986         /*
2987          * if there's an mtu associated with the route, use it
2988          * else, use the link mtu.
2989          */
2990         if (rt->rt_rmx.rmx_mtu)
2991                 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2992         else {
2993                 if (isipv6) {
2994                         mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2995                         if (!in6_localaddr(&inp->in6p_faddr))
2996                                 mss = min(mss, tcp_v6mssdflt);
2997                 } else {
2998                         mss = ifp->if_mtu - min_protoh;
2999                         if (!in_localaddr(inp->inp_faddr))
3000                                 mss = min(mss, tcp_mssdflt);
3001                 }
3002         }
3003         mss = min(mss, offer);
3004         /*
3005          * maxopd stores the maximum length of data AND options
3006          * in a segment; maxseg is the amount of data in a normal
3007          * segment.  We need to store this value (maxopd) apart
3008          * from maxseg, because now every segment carries options
3009          * and thus we normally have somewhat less data in segments.
3010          */
3011         tp->t_maxopd = mss;
3012
3013         /*
3014          * In case of T/TCP, origoffer==-1 indicates, that no segments
3015          * were received yet.  In this case we just guess, otherwise
3016          * we do the same as before T/TCP.
3017          */
3018         if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3019             (origoffer == -1 ||
3020              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3021                 mss -= TCPOLEN_TSTAMP_APPA;
3022         if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
3023             (origoffer == -1 ||
3024              (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3025                 mss -= TCPOLEN_CC_APPA;
3026
3027 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
3028                 if (mss > MCLBYTES)
3029                         mss &= ~(MCLBYTES-1);
3030 #else
3031                 if (mss > MCLBYTES)
3032                         mss = mss / MCLBYTES * MCLBYTES;
3033 #endif
3034         /*
3035          * If there's a pipesize, change the socket buffer
3036          * to that size.  Make the socket buffers an integral
3037          * number of mss units; if the mss is larger than
3038          * the socket buffer, decrease the mss.
3039          */
3040 #ifdef RTV_SPIPE
3041         if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3042 #endif
3043                 bufsize = so->so_snd.ssb_hiwat;
3044         if (bufsize < mss)
3045                 mss = bufsize;
3046         else {
3047                 bufsize = roundup(bufsize, mss);
3048                 if (bufsize > sb_max)
3049                         bufsize = sb_max;
3050                 if (bufsize > so->so_snd.ssb_hiwat)
3051                         ssb_reserve(&so->so_snd, bufsize, so, NULL);
3052         }
3053         tp->t_maxseg = mss;
3054
3055 #ifdef RTV_RPIPE
3056         if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3057 #endif
3058                 bufsize = so->so_rcv.ssb_hiwat;
3059         if (bufsize > mss) {
3060                 bufsize = roundup(bufsize, mss);
3061                 if (bufsize > sb_max)
3062                         bufsize = sb_max;
3063                 if (bufsize > so->so_rcv.ssb_hiwat)
3064                         ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3065         }
3066
3067         /*
3068          * Set the slow-start flight size depending on whether this
3069          * is a local network or not.
3070          */
3071         if (tcp_do_rfc3390)
3072                 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3073         else
3074                 tp->snd_cwnd = mss;
3075
3076         if (rt->rt_rmx.rmx_ssthresh) {
3077                 /*
3078                  * There's some sort of gateway or interface
3079                  * buffer limit on the path.  Use this to set
3080                  * the slow start threshhold, but set the
3081                  * threshold to no less than 2*mss.
3082                  */
3083                 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3084                 tcpstat.tcps_usedssthresh++;
3085         }
3086 }
3087
3088 /*
3089  * Determine the MSS option to send on an outgoing SYN.
3090  */
3091 int
3092 tcp_mssopt(struct tcpcb *tp)
3093 {
3094         struct rtentry *rt;
3095 #ifdef INET6
3096         boolean_t isipv6 =
3097             ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3098         int min_protoh = isipv6 ?
3099                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3100                              sizeof(struct tcpiphdr);
3101 #else
3102         const boolean_t isipv6 = FALSE;
3103         const size_t min_protoh = sizeof(struct tcpiphdr);
3104 #endif
3105
3106         if (isipv6)
3107                 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3108         else
3109                 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3110         if (rt == NULL)
3111                 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3112
3113         return (rt->rt_ifp->if_mtu - min_protoh);
3114 }
3115
3116 /*
3117  * When a partial ack arrives, force the retransmission of the
3118  * next unacknowledged segment.  Do not exit Fast Recovery.
3119  *
3120  * Implement the Slow-but-Steady variant of NewReno by restarting the
3121  * the retransmission timer.  Turn it off here so it can be restarted
3122  * later in tcp_output().
3123  */
3124 static void
3125 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3126 {
3127         tcp_seq old_snd_nxt = tp->snd_nxt;
3128         u_long ocwnd = tp->snd_cwnd;
3129
3130         tcp_callout_stop(tp, tp->tt_rexmt);
3131         tp->t_rtttime = 0;
3132         tp->snd_nxt = th->th_ack;
3133         /* Set snd_cwnd to one segment beyond acknowledged offset. */
3134         tp->snd_cwnd = tp->t_maxseg;
3135         tp->t_flags |= TF_ACKNOW;
3136         tcp_output(tp);
3137         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3138                 tp->snd_nxt = old_snd_nxt;
3139         /* partial window deflation */
3140         if (ocwnd > acked)
3141                 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3142         else
3143                 tp->snd_cwnd = tp->t_maxseg;
3144 }
3145
3146 /*
3147  * In contrast to the Slow-but-Steady NewReno variant,
3148  * we do not reset the retransmission timer for SACK retransmissions,
3149  * except when retransmitting snd_una.
3150  */
3151 static void
3152 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3153 {
3154         uint32_t pipe, seglen;
3155         tcp_seq nextrexmt;
3156         boolean_t lostdup;
3157         tcp_seq old_snd_nxt = tp->snd_nxt;
3158         u_long ocwnd = tp->snd_cwnd;
3159         int nseg = 0;           /* consecutive new segments */
3160 #define MAXBURST 4              /* limit burst of new packets on partial ack */
3161
3162         tp->t_rtttime = 0;
3163         pipe = tcp_sack_compute_pipe(tp);
3164         while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3165             (!tcp_do_smartsack || nseg < MAXBURST) &&
3166             tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3167                 uint32_t sent;
3168                 tcp_seq old_snd_max;
3169                 int error;
3170
3171                 if (nextrexmt == tp->snd_max)
3172                         ++nseg;
3173                 tp->snd_nxt = nextrexmt;
3174                 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3175                 old_snd_max = tp->snd_max;
3176                 if (nextrexmt == tp->snd_una)
3177                         tcp_callout_stop(tp, tp->tt_rexmt);
3178                 error = tcp_output(tp);
3179                 if (error != 0)
3180                         break;
3181                 sent = tp->snd_nxt - nextrexmt;
3182                 if (sent <= 0)
3183                         break;
3184                 if (!lostdup)
3185                         pipe += sent;
3186                 tcpstat.tcps_sndsackpack++;
3187                 tcpstat.tcps_sndsackbyte += sent;
3188                 if (SEQ_LT(nextrexmt, old_snd_max) &&
3189                     SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3190                         tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3191         }
3192         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3193                 tp->snd_nxt = old_snd_nxt;
3194         tp->snd_cwnd = ocwnd;
3195 }