Merge from vendor branch NETGRAPH:
[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 1.67 2007/04/22 01:13:14 dillon 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_var.h>
114 #include <netinet6/tcp6_var.h>
115 #include <netinet/tcpip.h>
116
117 #ifdef TCPDEBUG
118 #include <netinet/tcp_debug.h>
119
120 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
121 struct tcphdr tcp_savetcp;
122 #endif
123
124 #ifdef FAST_IPSEC
125 #include <netproto/ipsec/ipsec.h>
126 #include <netproto/ipsec/ipsec6.h>
127 #endif
128
129 #ifdef IPSEC
130 #include <netinet6/ipsec.h>
131 #include <netinet6/ipsec6.h>
132 #include <netproto/key/key.h>
133 #endif
134
135 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
136
137 tcp_cc  tcp_ccgen;
138 static int log_in_vain = 0;
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
140     &log_in_vain, 0, "Log all incoming TCP connections");
141
142 static int blackhole = 0;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
144     &blackhole, 0, "Do not send RST when dropping refused connections");
145
146 int tcp_delack_enabled = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
148     &tcp_delack_enabled, 0,
149     "Delay ACK to try and piggyback it onto a data packet");
150
151 #ifdef TCP_DROP_SYNFIN
152 static int drop_synfin = 0;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
154     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
155 #endif
156
157 static int tcp_do_limitedtransmit = 1;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
159     &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
160
161 static int tcp_do_early_retransmit = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
163     &tcp_do_early_retransmit, 0, "Early retransmit");
164
165 int tcp_aggregate_acks = 1;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
167     &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
168
169 int tcp_do_rfc3390 = 1;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
171     &tcp_do_rfc3390, 0,
172     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
173
174 static int tcp_do_eifel_detect = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
176     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
177
178 static int tcp_do_abc = 1;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
180     &tcp_do_abc, 0,
181     "TCP Appropriate Byte Counting (RFC 3465)");
182
183 /*
184  * Define as tunable for easy testing with SACK on and off.
185  * Warning:  do not change setting in the middle of an existing active TCP flow,
186  *   else strange things might happen to that flow.
187  */
188 int tcp_do_sack = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
190     &tcp_do_sack, 0, "Enable SACK Algorithms");
191
192 int tcp_do_smartsack = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
194     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
195
196 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
197     "TCP Segment Reassembly Queue");
198
199 int tcp_reass_maxseg = 0;
200 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
201     &tcp_reass_maxseg, 0,
202     "Global maximum number of TCP Segments in Reassembly Queue");
203
204 int tcp_reass_qsize = 0;
205 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
206     &tcp_reass_qsize, 0,
207     "Global number of TCP Segments currently in Reassembly Queue");
208
209 static int tcp_reass_overflows = 0;
210 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
211     &tcp_reass_overflows, 0,
212     "Global number of TCP Segment Reassembly Queue Overflows");
213
214 static void      tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
215 static void      tcp_pulloutofband(struct socket *,
216                      struct tcphdr *, struct mbuf *, int);
217 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
218                      struct mbuf *);
219 static void      tcp_xmit_timer(struct tcpcb *, int);
220 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
221 static void      tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
222
223 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
224 #ifdef INET6
225 #define ND6_HINT(tp) \
226 do { \
227         if ((tp) && (tp)->t_inpcb && \
228             ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
229             (tp)->t_inpcb->in6p_route.ro_rt) \
230                 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
231 } while (0)
232 #else
233 #define ND6_HINT(tp)
234 #endif
235
236 /*
237  * Indicate whether this ack should be delayed.  We can delay the ack if
238  *      - delayed acks are enabled and
239  *      - there is no delayed ack timer in progress and
240  *      - our last ack wasn't a 0-sized window.  We never want to delay
241  *        the ack that opens up a 0-sized window.
242  */
243 #define DELAY_ACK(tp) \
244         (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
245         !(tp->t_flags & TF_RXWIN0SENT))
246
247 #define acceptable_window_update(tp, th, tiwin)                         \
248     (SEQ_LT(tp->snd_wl1, th->th_seq) ||                                 \
249      (tp->snd_wl1 == th->th_seq &&                                      \
250       (SEQ_LT(tp->snd_wl2, th->th_ack) ||                               \
251        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
252
253 static int
254 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
255 {
256         struct tseg_qent *q;
257         struct tseg_qent *p = NULL;
258         struct tseg_qent *te;
259         struct socket *so = tp->t_inpcb->inp_socket;
260         int flags;
261
262         /*
263          * Call with th == NULL after become established to
264          * force pre-ESTABLISHED data up to user socket.
265          */
266         if (th == NULL)
267                 goto present;
268
269         /*
270          * Limit the number of segments in the reassembly queue to prevent
271          * holding on to too many segments (and thus running out of mbufs).
272          * Make sure to let the missing segment through which caused this
273          * queue.  Always keep one global queue entry spare to be able to
274          * process the missing segment.
275          */
276         if (th->th_seq != tp->rcv_nxt &&
277             tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
278                 tcp_reass_overflows++;
279                 tcpstat.tcps_rcvmemdrop++;
280                 m_freem(m);
281                 /* no SACK block to report */
282                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
283                 return (0);
284         }
285
286         /* Allocate a new queue entry. */
287         MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
288                M_INTWAIT | M_NULLOK);
289         if (te == NULL) {
290                 tcpstat.tcps_rcvmemdrop++;
291                 m_freem(m);
292                 /* no SACK block to report */
293                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
294                 return (0);
295         }
296         tcp_reass_qsize++;
297
298         /*
299          * Find a segment which begins after this one does.
300          */
301         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
302                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
303                         break;
304                 p = q;
305         }
306
307         /*
308          * If there is a preceding segment, it may provide some of
309          * our data already.  If so, drop the data from the incoming
310          * segment.  If it provides all of our data, drop us.
311          */
312         if (p != NULL) {
313                 tcp_seq_diff_t i;
314
315                 /* conversion to int (in i) handles seq wraparound */
316                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
317                 if (i > 0) {            /* overlaps preceding segment */
318                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
319                         /* enclosing block starts w/ preceding segment */
320                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
321                         if (i >= *tlenp) {
322                                 /* preceding encloses incoming segment */
323                                 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
324                                     p->tqe_len;
325                                 tcpstat.tcps_rcvduppack++;
326                                 tcpstat.tcps_rcvdupbyte += *tlenp;
327                                 m_freem(m);
328                                 kfree(te, M_TSEGQ);
329                                 tcp_reass_qsize--;
330                                 /*
331                                  * Try to present any queued data
332                                  * at the left window edge to the user.
333                                  * This is needed after the 3-WHS
334                                  * completes.
335                                  */
336                                 goto present;   /* ??? */
337                         }
338                         m_adj(m, i);
339                         *tlenp -= i;
340                         th->th_seq += i;
341                         /* incoming segment end is enclosing block end */
342                         tp->encloseblk.rblk_end = th->th_seq + *tlenp +
343                             ((th->th_flags & TH_FIN) != 0);
344                         /* trim end of reported D-SACK block */
345                         tp->reportblk.rblk_end = th->th_seq;
346                 }
347         }
348         tcpstat.tcps_rcvoopack++;
349         tcpstat.tcps_rcvoobyte += *tlenp;
350
351         /*
352          * While we overlap succeeding segments trim them or,
353          * if they are completely covered, dequeue them.
354          */
355         while (q) {
356                 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
357                 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
358                 struct tseg_qent *nq;
359
360                 if (i <= 0)
361                         break;
362                 if (!(tp->t_flags & TF_DUPSEG)) {    /* first time through */
363                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
364                         tp->encloseblk = tp->reportblk;
365                         /* report trailing duplicate D-SACK segment */
366                         tp->reportblk.rblk_start = q->tqe_th->th_seq;
367                 }
368                 if ((tp->t_flags & TF_ENCLOSESEG) &&
369                     SEQ_GT(qend, tp->encloseblk.rblk_end)) {
370                         /* extend enclosing block if one exists */
371                         tp->encloseblk.rblk_end = qend;
372                 }
373                 if (i < q->tqe_len) {
374                         q->tqe_th->th_seq += i;
375                         q->tqe_len -= i;
376                         m_adj(q->tqe_m, i);
377                         break;
378                 }
379
380                 nq = LIST_NEXT(q, tqe_q);
381                 LIST_REMOVE(q, tqe_q);
382                 m_freem(q->tqe_m);
383                 kfree(q, M_TSEGQ);
384                 tcp_reass_qsize--;
385                 q = nq;
386         }
387
388         /* Insert the new segment queue entry into place. */
389         te->tqe_m = m;
390         te->tqe_th = th;
391         te->tqe_len = *tlenp;
392
393         /* check if can coalesce with following segment */
394         if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
395                 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
396
397                 te->tqe_len += q->tqe_len;
398                 if (q->tqe_th->th_flags & TH_FIN)
399                         te->tqe_th->th_flags |= TH_FIN;
400                 m_cat(te->tqe_m, q->tqe_m);
401                 tp->encloseblk.rblk_end = tend;
402                 /*
403                  * When not reporting a duplicate segment, use
404                  * the larger enclosing block as the SACK block.
405                  */
406                 if (!(tp->t_flags & TF_DUPSEG))
407                         tp->reportblk.rblk_end = tend;
408                 LIST_REMOVE(q, tqe_q);
409                 kfree(q, M_TSEGQ);
410                 tcp_reass_qsize--;
411         }
412
413         if (p == NULL) {
414                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
415         } else {
416                 /* check if can coalesce with preceding segment */
417                 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
418                         p->tqe_len += te->tqe_len;
419                         m_cat(p->tqe_m, te->tqe_m);
420                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
421                         /*
422                          * When not reporting a duplicate segment, use
423                          * the larger enclosing block as the SACK block.
424                          */
425                         if (!(tp->t_flags & TF_DUPSEG))
426                                 tp->reportblk.rblk_start = p->tqe_th->th_seq;
427                         kfree(te, M_TSEGQ);
428                         tcp_reass_qsize--;
429                 } else
430                         LIST_INSERT_AFTER(p, te, tqe_q);
431         }
432
433 present:
434         /*
435          * Present data to user, advancing rcv_nxt through
436          * completed sequence space.
437          */
438         if (!TCPS_HAVEESTABLISHED(tp->t_state))
439                 return (0);
440         q = LIST_FIRST(&tp->t_segq);
441         if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
442                 return (0);
443         tp->rcv_nxt += q->tqe_len;
444         if (!(tp->t_flags & TF_DUPSEG)) {
445                 /* no SACK block to report since ACK advanced */
446                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
447         }
448         /* no enclosing block to report since ACK advanced */
449         tp->t_flags &= ~TF_ENCLOSESEG;
450         flags = q->tqe_th->th_flags & TH_FIN;
451         LIST_REMOVE(q, tqe_q);
452         KASSERT(LIST_EMPTY(&tp->t_segq) ||
453                 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
454                 ("segment not coalesced"));
455         if (so->so_state & SS_CANTRCVMORE)
456                 m_freem(q->tqe_m);
457         else
458                 ssb_appendstream(&so->so_rcv, q->tqe_m);
459         kfree(q, M_TSEGQ);
460         tcp_reass_qsize--;
461         ND6_HINT(tp);
462         sorwakeup(so);
463         return (flags);
464 }
465
466 /*
467  * TCP input routine, follows pages 65-76 of the
468  * protocol specification dated September, 1981 very closely.
469  */
470 #ifdef INET6
471 int
472 tcp6_input(struct mbuf **mp, int *offp, int proto)
473 {
474         struct mbuf *m = *mp;
475         struct in6_ifaddr *ia6;
476
477         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
478
479         /*
480          * draft-itojun-ipv6-tcp-to-anycast
481          * better place to put this in?
482          */
483         ia6 = ip6_getdstifaddr(m);
484         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
485                 struct ip6_hdr *ip6;
486
487                 ip6 = mtod(m, struct ip6_hdr *);
488                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
489                             offsetof(struct ip6_hdr, ip6_dst));
490                 return (IPPROTO_DONE);
491         }
492
493         tcp_input(m, *offp, proto);
494         return (IPPROTO_DONE);
495 }
496 #endif
497
498 void
499 tcp_input(struct mbuf *m, ...)
500 {
501         __va_list ap;
502         int off0, proto;
503         struct tcphdr *th;
504         struct ip *ip = NULL;
505         struct ipovly *ipov;
506         struct inpcb *inp = NULL;
507         u_char *optp = NULL;
508         int optlen = 0;
509         int len, tlen, off;
510         int drop_hdrlen;
511         struct tcpcb *tp = NULL;
512         int thflags;
513         struct socket *so = 0;
514         int todrop, acked;
515         boolean_t ourfinisacked, needoutput = FALSE;
516         u_long tiwin;
517         int recvwin;
518         struct tcpopt to;               /* options in this segment */
519         struct rmxp_tao *taop;          /* pointer to our TAO cache entry */
520         struct rmxp_tao tao_noncached;  /* in case there's no cached entry */
521         struct sockaddr_in *next_hop = NULL;
522         int rstreason; /* For badport_bandlim accounting purposes */
523         int cpu;
524         struct ip6_hdr *ip6 = NULL;
525 #ifdef INET6
526         boolean_t isipv6;
527 #else
528         const boolean_t isipv6 = FALSE;
529 #endif
530 #ifdef TCPDEBUG
531         short ostate = 0;
532 #endif
533
534         __va_start(ap, m);
535         off0 = __va_arg(ap, int);
536         proto = __va_arg(ap, int);
537         __va_end(ap);
538
539         tcpstat.tcps_rcvtotal++;
540
541         /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
542         while  (m->m_type == MT_TAG) {
543                 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
544                         next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
545                 m = m->m_next;
546         }
547
548 #ifdef INET6
549         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
550 #endif
551
552         if (isipv6) {
553                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
554                 ip6 = mtod(m, struct ip6_hdr *);
555                 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
556                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
557                         tcpstat.tcps_rcvbadsum++;
558                         goto drop;
559                 }
560                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
561
562                 /*
563                  * Be proactive about unspecified IPv6 address in source.
564                  * As we use all-zero to indicate unbounded/unconnected pcb,
565                  * unspecified IPv6 address can be used to confuse us.
566                  *
567                  * Note that packets with unspecified IPv6 destination is
568                  * already dropped in ip6_input.
569                  */
570                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
571                         /* XXX stat */
572                         goto drop;
573                 }
574         } else {
575                 /*
576                  * Get IP and TCP header together in first mbuf.
577                  * Note: IP leaves IP header in first mbuf.
578                  */
579                 if (off0 > sizeof(struct ip)) {
580                         ip_stripoptions(m);
581                         off0 = sizeof(struct ip);
582                 }
583                 /* already checked and pulled up in ip_demux() */
584                 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
585                     ("TCP header not in one mbuf: m->m_len %d", m->m_len));
586                 ip = mtod(m, struct ip *);
587                 ipov = (struct ipovly *)ip;
588                 th = (struct tcphdr *)((caddr_t)ip + off0);
589                 tlen = ip->ip_len;
590
591                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
592                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
593                                 th->th_sum = m->m_pkthdr.csum_data;
594                         else
595                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
596                                                 ip->ip_dst.s_addr,
597                                                 htonl(m->m_pkthdr.csum_data +
598                                                         ip->ip_len +
599                                                         IPPROTO_TCP));
600                         th->th_sum ^= 0xffff;
601                 } else {
602                         /*
603                          * Checksum extended TCP header and data.
604                          */
605                         len = sizeof(struct ip) + tlen;
606                         bzero(ipov->ih_x1, sizeof ipov->ih_x1);
607                         ipov->ih_len = (u_short)tlen;
608                         ipov->ih_len = htons(ipov->ih_len);
609                         th->th_sum = in_cksum(m, len);
610                 }
611                 if (th->th_sum) {
612                         tcpstat.tcps_rcvbadsum++;
613                         goto drop;
614                 }
615 #ifdef INET6
616                 /* Re-initialization for later version check */
617                 ip->ip_v = IPVERSION;
618 #endif
619         }
620
621         /*
622          * Check that TCP offset makes sense,
623          * pull out TCP options and adjust length.              XXX
624          */
625         off = th->th_off << 2;
626         /* already checked and pulled up in ip_demux() */
627         KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
628             ("bad TCP data offset %d (tlen %d)", off, tlen));
629         tlen -= off;    /* tlen is used instead of ti->ti_len */
630         if (off > sizeof(struct tcphdr)) {
631                 if (isipv6) {
632                         IP6_EXTHDR_CHECK(m, off0, off, );
633                         ip6 = mtod(m, struct ip6_hdr *);
634                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
635                 } else {
636                         /* already pulled up in ip_demux() */
637                         KASSERT(m->m_len >= sizeof(struct ip) + off,
638                             ("TCP header and options not in one mbuf: "
639                              "m_len %d, off %d", m->m_len, off));
640                 }
641                 optlen = off - sizeof(struct tcphdr);
642                 optp = (u_char *)(th + 1);
643         }
644         thflags = th->th_flags;
645
646 #ifdef TCP_DROP_SYNFIN
647         /*
648          * If the drop_synfin option is enabled, drop all packets with
649          * both the SYN and FIN bits set. This prevents e.g. nmap from
650          * identifying the TCP/IP stack.
651          *
652          * This is a violation of the TCP specification.
653          */
654         if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
655                 goto drop;
656 #endif
657
658         /*
659          * Convert TCP protocol specific fields to host format.
660          */
661         th->th_seq = ntohl(th->th_seq);
662         th->th_ack = ntohl(th->th_ack);
663         th->th_win = ntohs(th->th_win);
664         th->th_urp = ntohs(th->th_urp);
665
666         /*
667          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
668          * until after ip6_savecontrol() is called and before other functions
669          * which don't want those proto headers.
670          * Because ip6_savecontrol() is going to parse the mbuf to
671          * search for data to be passed up to user-land, it wants mbuf
672          * parameters to be unchanged.
673          * XXX: the call of ip6_savecontrol() has been obsoleted based on
674          * latest version of the advanced API (20020110).
675          */
676         drop_hdrlen = off0 + off;
677
678         /*
679          * Locate pcb for segment.
680          */
681 findpcb:
682         /* IPFIREWALL_FORWARD section */
683         if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
684                 /*
685                  * Transparently forwarded. Pretend to be the destination.
686                  * already got one like this?
687                  */
688                 cpu = mycpu->gd_cpuid;
689                 inp = in_pcblookup_hash(&tcbinfo[cpu],
690                                         ip->ip_src, th->th_sport,
691                                         ip->ip_dst, th->th_dport,
692                                         0, m->m_pkthdr.rcvif);
693                 if (!inp) {
694                         /*
695                          * It's new.  Try to find the ambushing socket.
696                          */
697
698                         /*
699                          * The rest of the ipfw code stores the port in
700                          * host order.  XXX
701                          * (The IP address is still in network order.)
702                          */
703                         in_port_t dport = next_hop->sin_port ?
704                                                 htons(next_hop->sin_port) :
705                                                 th->th_dport;
706
707                         cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
708                                           next_hop->sin_addr.s_addr, dport);
709                         inp = in_pcblookup_hash(&tcbinfo[cpu],
710                                                 ip->ip_src, th->th_sport,
711                                                 next_hop->sin_addr, dport,
712                                                 1, m->m_pkthdr.rcvif);
713                 }
714         } else {
715                 if (isipv6) {
716                         inp = in6_pcblookup_hash(&tcbinfo[0],
717                                                  &ip6->ip6_src, th->th_sport,
718                                                  &ip6->ip6_dst, th->th_dport,
719                                                  1, m->m_pkthdr.rcvif);
720                 } else {
721                         cpu = mycpu->gd_cpuid;
722                         inp = in_pcblookup_hash(&tcbinfo[cpu],
723                                                 ip->ip_src, th->th_sport,
724                                                 ip->ip_dst, th->th_dport,
725                                                 1, m->m_pkthdr.rcvif);
726                 }
727         }
728
729         /*
730          * If the state is CLOSED (i.e., TCB does not exist) then
731          * all data in the incoming segment is discarded.
732          * If the TCB exists but is in CLOSED state, it is embryonic,
733          * but should either do a listen or a connect soon.
734          */
735         if (inp == NULL) {
736                 if (log_in_vain) {
737 #ifdef INET6
738                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
739 #else
740                         char dbuf[sizeof "aaa.bbb.ccc.ddd"];
741                         char sbuf[sizeof "aaa.bbb.ccc.ddd"];
742 #endif
743                         if (isipv6) {
744                                 strcpy(dbuf, "[");
745                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
746                                 strcat(dbuf, "]");
747                                 strcpy(sbuf, "[");
748                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
749                                 strcat(sbuf, "]");
750                         } else {
751                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
752                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
753                         }
754                         switch (log_in_vain) {
755                         case 1:
756                                 if (!(thflags & TH_SYN))
757                                         break;
758                         case 2:
759                                 log(LOG_INFO,
760                                     "Connection attempt to TCP %s:%d "
761                                     "from %s:%d flags:0x%02x\n",
762                                     dbuf, ntohs(th->th_dport), sbuf,
763                                     ntohs(th->th_sport), thflags);
764                                 break;
765                         default:
766                                 break;
767                         }
768                 }
769                 if (blackhole) {
770                         switch (blackhole) {
771                         case 1:
772                                 if (thflags & TH_SYN)
773                                         goto drop;
774                                 break;
775                         case 2:
776                                 goto drop;
777                         default:
778                                 goto drop;
779                         }
780                 }
781                 rstreason = BANDLIM_RST_CLOSEDPORT;
782                 goto dropwithreset;
783         }
784
785 #ifdef IPSEC
786         if (isipv6) {
787                 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
788                         ipsec6stat.in_polvio++;
789                         goto drop;
790                 }
791         } else {
792                 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
793                         ipsecstat.in_polvio++;
794                         goto drop;
795                 }
796         }
797 #endif
798 #ifdef FAST_IPSEC
799         if (isipv6) {
800                 if (ipsec6_in_reject(m, inp))
801                         goto drop;
802         } else {
803                 if (ipsec4_in_reject(m, inp))
804                         goto drop;
805         }
806 #endif
807         /* Check the minimum TTL for socket. */
808 #ifdef INET6
809         if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
810                 goto drop;
811 #endif
812
813         tp = intotcpcb(inp);
814         if (tp == NULL) {
815                 rstreason = BANDLIM_RST_CLOSEDPORT;
816                 goto dropwithreset;
817         }
818         if (tp->t_state <= TCPS_CLOSED)
819                 goto drop;
820
821         /* Unscale the window into a 32-bit value. */
822         if (!(thflags & TH_SYN))
823                 tiwin = th->th_win << tp->snd_scale;
824         else
825                 tiwin = th->th_win;
826
827         so = inp->inp_socket;
828
829 #ifdef TCPDEBUG
830         if (so->so_options & SO_DEBUG) {
831                 ostate = tp->t_state;
832                 if (isipv6)
833                         bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
834                 else
835                         bcopy(ip, tcp_saveipgen, sizeof(*ip));
836                 tcp_savetcp = *th;
837         }
838 #endif
839
840         bzero(&to, sizeof to);
841
842         if (so->so_options & SO_ACCEPTCONN) {
843                 struct in_conninfo inc;
844
845 #ifdef INET6
846                 inc.inc_isipv6 = (isipv6 == TRUE);
847 #endif
848                 if (isipv6) {
849                         inc.inc6_faddr = ip6->ip6_src;
850                         inc.inc6_laddr = ip6->ip6_dst;
851                         inc.inc6_route.ro_rt = NULL;            /* XXX */
852                 } else {
853                         inc.inc_faddr = ip->ip_src;
854                         inc.inc_laddr = ip->ip_dst;
855                         inc.inc_route.ro_rt = NULL;             /* XXX */
856                 }
857                 inc.inc_fport = th->th_sport;
858                 inc.inc_lport = th->th_dport;
859
860                 /*
861                  * If the state is LISTEN then ignore segment if it contains
862                  * a RST.  If the segment contains an ACK then it is bad and
863                  * send a RST.  If it does not contain a SYN then it is not
864                  * interesting; drop it.
865                  *
866                  * If the state is SYN_RECEIVED (syncache) and seg contains
867                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
868                  * contains a RST, check the sequence number to see if it
869                  * is a valid reset segment.
870                  */
871                 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
872                         if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
873                                 if (!syncache_expand(&inc, th, &so, m)) {
874                                         /*
875                                          * No syncache entry, or ACK was not
876                                          * for our SYN/ACK.  Send a RST.
877                                          */
878                                         tcpstat.tcps_badsyn++;
879                                         rstreason = BANDLIM_RST_OPENPORT;
880                                         goto dropwithreset;
881                                 }
882                                 if (so == NULL)
883                                         /*
884                                          * Could not complete 3-way handshake,
885                                          * connection is being closed down, and
886                                          * syncache will free mbuf.
887                                          */
888                                         return;
889                                 /*
890                                  * Socket is created in state SYN_RECEIVED.
891                                  * Continue processing segment.
892                                  */
893                                 inp = so->so_pcb;
894                                 tp = intotcpcb(inp);
895                                 /*
896                                  * This is what would have happened in
897                                  * tcp_output() when the SYN,ACK was sent.
898                                  */
899                                 tp->snd_up = tp->snd_una;
900                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
901                                 tp->last_ack_sent = tp->rcv_nxt;
902 /*
903  * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
904  * until the _second_ ACK is received:
905  *    rcv SYN (set wscale opts)  --> send SYN/ACK, set snd_wnd = window.
906  *    rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
907  *        move to ESTAB, set snd_wnd to tiwin.
908  */
909                                 tp->snd_wnd = tiwin;    /* unscaled */
910                                 goto after_listen;
911                         }
912                         if (thflags & TH_RST) {
913                                 syncache_chkrst(&inc, th);
914                                 goto drop;
915                         }
916                         if (thflags & TH_ACK) {
917                                 syncache_badack(&inc);
918                                 tcpstat.tcps_badsyn++;
919                                 rstreason = BANDLIM_RST_OPENPORT;
920                                 goto dropwithreset;
921                         }
922                         goto drop;
923                 }
924
925                 /*
926                  * Segment's flags are (SYN) or (SYN | FIN).
927                  */
928 #ifdef INET6
929                 /*
930                  * If deprecated address is forbidden,
931                  * we do not accept SYN to deprecated interface
932                  * address to prevent any new inbound connection from
933                  * getting established.
934                  * When we do not accept SYN, we send a TCP RST,
935                  * with deprecated source address (instead of dropping
936                  * it).  We compromise it as it is much better for peer
937                  * to send a RST, and RST will be the final packet
938                  * for the exchange.
939                  *
940                  * If we do not forbid deprecated addresses, we accept
941                  * the SYN packet.  RFC2462 does not suggest dropping
942                  * SYN in this case.
943                  * If we decipher RFC2462 5.5.4, it says like this:
944                  * 1. use of deprecated addr with existing
945                  *    communication is okay - "SHOULD continue to be
946                  *    used"
947                  * 2. use of it with new communication:
948                  *   (2a) "SHOULD NOT be used if alternate address
949                  *        with sufficient scope is available"
950                  *   (2b) nothing mentioned otherwise.
951                  * Here we fall into (2b) case as we have no choice in
952                  * our source address selection - we must obey the peer.
953                  *
954                  * The wording in RFC2462 is confusing, and there are
955                  * multiple description text for deprecated address
956                  * handling - worse, they are not exactly the same.
957                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
958                  */
959                 if (isipv6 && !ip6_use_deprecated) {
960                         struct in6_ifaddr *ia6;
961
962                         if ((ia6 = ip6_getdstifaddr(m)) &&
963                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
964                                 tp = NULL;
965                                 rstreason = BANDLIM_RST_OPENPORT;
966                                 goto dropwithreset;
967                         }
968                 }
969 #endif
970                 /*
971                  * If it is from this socket, drop it, it must be forged.
972                  * Don't bother responding if the destination was a broadcast.
973                  */
974                 if (th->th_dport == th->th_sport) {
975                         if (isipv6) {
976                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
977                                                        &ip6->ip6_src))
978                                         goto drop;
979                         } else {
980                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
981                                         goto drop;
982                         }
983                 }
984                 /*
985                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
986                  *
987                  * Note that it is quite possible to receive unicast
988                  * link-layer packets with a broadcast IP address. Use
989                  * in_broadcast() to find them.
990                  */
991                 if (m->m_flags & (M_BCAST | M_MCAST))
992                         goto drop;
993                 if (isipv6) {
994                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
995                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
996                                 goto drop;
997                 } else {
998                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
999                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1000                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1001                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1002                                 goto drop;
1003                 }
1004                 /*
1005                  * SYN appears to be valid; create compressed TCP state
1006                  * for syncache, or perform t/tcp connection.
1007                  */
1008                 if (so->so_qlen <= so->so_qlimit) {
1009                         tcp_dooptions(&to, optp, optlen, TRUE);
1010                         if (!syncache_add(&inc, &to, th, &so, m))
1011                                 goto drop;
1012                         if (so == NULL)
1013                                 /*
1014                                  * Entry added to syncache, mbuf used to
1015                                  * send SYN,ACK packet.
1016                                  */
1017                                 return;
1018                         /*
1019                          * Segment passed TAO tests.
1020                          */
1021                         inp = so->so_pcb;
1022                         tp = intotcpcb(inp);
1023                         tp->snd_wnd = tiwin;
1024                         tp->t_starttime = ticks;
1025                         tp->t_state = TCPS_ESTABLISHED;
1026
1027                         /*
1028                          * If there is a FIN, or if there is data and the
1029                          * connection is local, then delay SYN,ACK(SYN) in
1030                          * the hope of piggy-backing it on a response
1031                          * segment.  Otherwise must send ACK now in case
1032                          * the other side is slow starting.
1033                          */
1034                         if (DELAY_ACK(tp) &&
1035                             ((thflags & TH_FIN) ||
1036                              (tlen != 0 &&
1037                               ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1038                                (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1039                                 callout_reset(tp->tt_delack, tcp_delacktime,
1040                                                 tcp_timer_delack, tp);
1041                                 tp->t_flags |= TF_NEEDSYN;
1042                         } else
1043                                 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1044
1045                         tcpstat.tcps_connects++;
1046                         soisconnected(so);
1047                         goto trimthenstep6;
1048                 }
1049                 goto drop;
1050         }
1051 after_listen:
1052
1053         /* should not happen - syncache should pick up these connections */
1054         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1055
1056         /*
1057          * Segment received on connection.
1058          * Reset idle time and keep-alive timer.
1059          */
1060         tp->t_rcvtime = ticks;
1061         if (TCPS_HAVEESTABLISHED(tp->t_state))
1062                 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1063
1064         /*
1065          * Process options.
1066          * XXX this is tradtitional behavior, may need to be cleaned up.
1067          */
1068         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1069         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1070                 if (to.to_flags & TOF_SCALE) {
1071                         tp->t_flags |= TF_RCVD_SCALE;
1072                         tp->requested_s_scale = to.to_requested_s_scale;
1073                 }
1074                 if (to.to_flags & TOF_TS) {
1075                         tp->t_flags |= TF_RCVD_TSTMP;
1076                         tp->ts_recent = to.to_tsval;
1077                         tp->ts_recent_age = ticks;
1078                 }
1079                 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1080                         tp->t_flags |= TF_RCVD_CC;
1081                 if (to.to_flags & TOF_MSS)
1082                         tcp_mss(tp, to.to_mss);
1083                 /*
1084                  * Only set the TF_SACK_PERMITTED per-connection flag
1085                  * if we got a SACK_PERMITTED option from the other side
1086                  * and the global tcp_do_sack variable is true.
1087                  */
1088                 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1089                         tp->t_flags |= TF_SACK_PERMITTED;
1090         }
1091
1092         /*
1093          * Header prediction: check for the two common cases
1094          * of a uni-directional data xfer.  If the packet has
1095          * no control flags, is in-sequence, the window didn't
1096          * change and we're not retransmitting, it's a
1097          * candidate.  If the length is zero and the ack moved
1098          * forward, we're the sender side of the xfer.  Just
1099          * free the data acked & wake any higher level process
1100          * that was blocked waiting for space.  If the length
1101          * is non-zero and the ack didn't move, we're the
1102          * receiver side.  If we're getting packets in-order
1103          * (the reassembly queue is empty), add the data to
1104          * the socket buffer and note that we need a delayed ack.
1105          * Make sure that the hidden state-flags are also off.
1106          * Since we check for TCPS_ESTABLISHED above, it can only
1107          * be TH_NEEDSYN.
1108          */
1109         if (tp->t_state == TCPS_ESTABLISHED &&
1110             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1111             !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1112             (!(to.to_flags & TOF_TS) ||
1113              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1114             /*
1115              * Using the CC option is compulsory if once started:
1116              *   the segment is OK if no T/TCP was negotiated or
1117              *   if the segment has a CC option equal to CCrecv
1118              */
1119             ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1120              ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1121             th->th_seq == tp->rcv_nxt &&
1122             tp->snd_nxt == tp->snd_max) {
1123
1124                 /*
1125                  * If last ACK falls within this segment's sequence numbers,
1126                  * record the timestamp.
1127                  * NOTE that the test is modified according to the latest
1128                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1129                  */
1130                 if ((to.to_flags & TOF_TS) &&
1131                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1132                         tp->ts_recent_age = ticks;
1133                         tp->ts_recent = to.to_tsval;
1134                 }
1135
1136                 if (tlen == 0) {
1137                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1138                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1139                             tp->snd_cwnd >= tp->snd_wnd &&
1140                             !IN_FASTRECOVERY(tp)) {
1141                                 /*
1142                                  * This is a pure ack for outstanding data.
1143                                  */
1144                                 ++tcpstat.tcps_predack;
1145                                 /*
1146                                  * "bad retransmit" recovery
1147                                  *
1148                                  * If Eifel detection applies, then
1149                                  * it is deterministic, so use it
1150                                  * unconditionally over the old heuristic.
1151                                  * Otherwise, fall back to the old heuristic.
1152                                  */
1153                                 if (tcp_do_eifel_detect &&
1154                                     (to.to_flags & TOF_TS) && to.to_tsecr &&
1155                                     (tp->t_flags & TF_FIRSTACCACK)) {
1156                                         /* Eifel detection applicable. */
1157                                         if (to.to_tsecr < tp->t_rexmtTS) {
1158                                                 tcp_revert_congestion_state(tp);
1159                                                 ++tcpstat.tcps_eifeldetected;
1160                                         }
1161                                 } else if (tp->t_rxtshift == 1 &&
1162                                            ticks < tp->t_badrxtwin) {
1163                                         tcp_revert_congestion_state(tp);
1164                                         ++tcpstat.tcps_rttdetected;
1165                                 }
1166                                 tp->t_flags &= ~(TF_FIRSTACCACK |
1167                                                  TF_FASTREXMT | TF_EARLYREXMT);
1168                                 /*
1169                                  * Recalculate the retransmit timer / rtt.
1170                                  *
1171                                  * Some machines (certain windows boxes)
1172                                  * send broken timestamp replies during the
1173                                  * SYN+ACK phase, ignore timestamps of 0.
1174                                  */
1175                                 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1176                                         tcp_xmit_timer(tp,
1177                                                        ticks - to.to_tsecr + 1);
1178                                 } else if (tp->t_rtttime &&
1179                                            SEQ_GT(th->th_ack, tp->t_rtseq)) {
1180                                         tcp_xmit_timer(tp,
1181                                                        ticks - tp->t_rtttime);
1182                                 }
1183                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1184                                 acked = th->th_ack - tp->snd_una;
1185                                 tcpstat.tcps_rcvackpack++;
1186                                 tcpstat.tcps_rcvackbyte += acked;
1187                                 sbdrop(&so->so_snd.sb, acked);
1188                                 tp->snd_recover = th->th_ack - 1;
1189                                 tp->snd_una = th->th_ack;
1190                                 tp->t_dupacks = 0;
1191                                 /*
1192                                  * Update window information.
1193                                  */
1194                                 if (tiwin != tp->snd_wnd &&
1195                                     acceptable_window_update(tp, th, tiwin)) {
1196                                         /* keep track of pure window updates */
1197                                         if (tp->snd_wl2 == th->th_ack &&
1198                                             tiwin > tp->snd_wnd)
1199                                                 tcpstat.tcps_rcvwinupd++;
1200                                         tp->snd_wnd = tiwin;
1201                                         tp->snd_wl1 = th->th_seq;
1202                                         tp->snd_wl2 = th->th_ack;
1203                                         if (tp->snd_wnd > tp->max_sndwnd)
1204                                                 tp->max_sndwnd = tp->snd_wnd;
1205                                 }
1206                                 m_freem(m);
1207                                 ND6_HINT(tp); /* some progress has been done */
1208                                 /*
1209                                  * If all outstanding data are acked, stop
1210                                  * retransmit timer, otherwise restart timer
1211                                  * using current (possibly backed-off) value.
1212                                  * If process is waiting for space,
1213                                  * wakeup/selwakeup/signal.  If data
1214                                  * are ready to send, let tcp_output
1215                                  * decide between more output or persist.
1216                                  */
1217                                 if (tp->snd_una == tp->snd_max)
1218                                         callout_stop(tp->tt_rexmt);
1219                                 else if (!callout_active(tp->tt_persist))
1220                                         callout_reset(tp->tt_rexmt,
1221                                                       tp->t_rxtcur,
1222                                                       tcp_timer_rexmt, tp);
1223                                 sowwakeup(so);
1224                                 if (so->so_snd.ssb_cc > 0)
1225                                         tcp_output(tp);
1226                                 return;
1227                         }
1228                 } else if (tiwin == tp->snd_wnd &&
1229                     th->th_ack == tp->snd_una &&
1230                     LIST_EMPTY(&tp->t_segq) &&
1231                     tlen <= ssb_space(&so->so_rcv)) {
1232                         /*
1233                          * This is a pure, in-sequence data packet
1234                          * with nothing on the reassembly queue and
1235                          * we have enough buffer space to take it.
1236                          */
1237                         ++tcpstat.tcps_preddat;
1238                         tp->rcv_nxt += tlen;
1239                         tcpstat.tcps_rcvpack++;
1240                         tcpstat.tcps_rcvbyte += tlen;
1241                         ND6_HINT(tp);   /* some progress has been done */
1242                         /*
1243                          * Add data to socket buffer.
1244                          */
1245                         if (so->so_state & SS_CANTRCVMORE) {
1246                                 m_freem(m);
1247                         } else {
1248                                 m_adj(m, drop_hdrlen); /* delayed header drop */
1249                                 ssb_appendstream(&so->so_rcv, m);
1250                         }
1251                         sorwakeup(so);
1252                         /*
1253                          * This code is responsible for most of the ACKs
1254                          * the TCP stack sends back after receiving a data
1255                          * packet.  Note that the DELAY_ACK check fails if
1256                          * the delack timer is already running, which results
1257                          * in an ack being sent every other packet (which is
1258                          * what we want).
1259                          *
1260                          * We then further aggregate acks by not actually
1261                          * sending one until the protocol thread has completed
1262                          * processing the current backlog of packets.  This
1263                          * does not delay the ack any further, but allows us
1264                          * to take advantage of the packet aggregation that
1265                          * high speed NICs do (usually blocks of 8-10 packets)
1266                          * to send a single ack rather then four or five acks,
1267                          * greatly reducing the ack rate, the return channel
1268                          * bandwidth, and the protocol overhead on both ends.
1269                          *
1270                          * Since this also has the effect of slowing down
1271                          * the exponential slow-start ramp-up, systems with 
1272                          * very large bandwidth-delay products might want
1273                          * to turn the feature off.
1274                          */
1275                         if (DELAY_ACK(tp)) {
1276                                 callout_reset(tp->tt_delack, tcp_delacktime,
1277                                     tcp_timer_delack, tp);
1278                         } else if (tcp_aggregate_acks) {
1279                                 tp->t_flags |= TF_ACKNOW;
1280                                 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1281                                         tp->t_flags |= TF_ONOUTPUTQ;
1282                                         tp->tt_cpu = mycpu->gd_cpuid;
1283                                         TAILQ_INSERT_TAIL(
1284                                             &tcpcbackq[tp->tt_cpu],
1285                                             tp, t_outputq);
1286                                 }
1287                         } else {
1288                                 tp->t_flags |= TF_ACKNOW;
1289                                 tcp_output(tp);
1290                         }
1291                         return;
1292                 }
1293         }
1294
1295         /*
1296          * Calculate amount of space in receive window,
1297          * and then do TCP input processing.
1298          * Receive window is amount of space in rcv queue,
1299          * but not less than advertised window.
1300          */
1301         recvwin = ssb_space(&so->so_rcv);
1302         if (recvwin < 0)
1303                 recvwin = 0;
1304         tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1305
1306         switch (tp->t_state) {
1307         /*
1308          * If the state is SYN_RECEIVED:
1309          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1310          */
1311         case TCPS_SYN_RECEIVED:
1312                 if ((thflags & TH_ACK) &&
1313                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1314                      SEQ_GT(th->th_ack, tp->snd_max))) {
1315                                 rstreason = BANDLIM_RST_OPENPORT;
1316                                 goto dropwithreset;
1317                 }
1318                 break;
1319
1320         /*
1321          * If the state is SYN_SENT:
1322          *      if seg contains an ACK, but not for our SYN, drop the input.
1323          *      if seg contains a RST, then drop the connection.
1324          *      if seg does not contain SYN, then drop it.
1325          * Otherwise this is an acceptable SYN segment
1326          *      initialize tp->rcv_nxt and tp->irs
1327          *      if seg contains ack then advance tp->snd_una
1328          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1329          *      arrange for segment to be acked (eventually)
1330          *      continue processing rest of data/controls, beginning with URG
1331          */
1332         case TCPS_SYN_SENT:
1333                 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1334                         taop = &tao_noncached;
1335                         bzero(taop, sizeof *taop);
1336                 }
1337
1338                 if ((thflags & TH_ACK) &&
1339                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1340                      SEQ_GT(th->th_ack, tp->snd_max))) {
1341                         /*
1342                          * If we have a cached CCsent for the remote host,
1343                          * hence we haven't just crashed and restarted,
1344                          * do not send a RST.  This may be a retransmission
1345                          * from the other side after our earlier ACK was lost.
1346                          * Our new SYN, when it arrives, will serve as the
1347                          * needed ACK.
1348                          */
1349                         if (taop->tao_ccsent != 0)
1350                                 goto drop;
1351                         else {
1352                                 rstreason = BANDLIM_UNLIMITED;
1353                                 goto dropwithreset;
1354                         }
1355                 }
1356                 if (thflags & TH_RST) {
1357                         if (thflags & TH_ACK)
1358                                 tp = tcp_drop(tp, ECONNREFUSED);
1359                         goto drop;
1360                 }
1361                 if (!(thflags & TH_SYN))
1362                         goto drop;
1363                 tp->snd_wnd = th->th_win;       /* initial send window */
1364                 tp->cc_recv = to.to_cc;         /* foreign CC */
1365
1366                 tp->irs = th->th_seq;
1367                 tcp_rcvseqinit(tp);
1368                 if (thflags & TH_ACK) {
1369                         /*
1370                          * Our SYN was acked.  If segment contains CC.ECHO
1371                          * option, check it to make sure this segment really
1372                          * matches our SYN.  If not, just drop it as old
1373                          * duplicate, but send an RST if we're still playing
1374                          * by the old rules.  If no CC.ECHO option, make sure
1375                          * we don't get fooled into using T/TCP.
1376                          */
1377                         if (to.to_flags & TOF_CCECHO) {
1378                                 if (tp->cc_send != to.to_ccecho) {
1379                                         if (taop->tao_ccsent != 0)
1380                                                 goto drop;
1381                                         else {
1382                                                 rstreason = BANDLIM_UNLIMITED;
1383                                                 goto dropwithreset;
1384                                         }
1385                                 }
1386                         } else
1387                                 tp->t_flags &= ~TF_RCVD_CC;
1388                         tcpstat.tcps_connects++;
1389                         soisconnected(so);
1390                         /* Do window scaling on this connection? */
1391                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1392                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1393                                 tp->snd_scale = tp->requested_s_scale;
1394                                 tp->rcv_scale = tp->request_r_scale;
1395                         }
1396                         /* Segment is acceptable, update cache if undefined. */
1397                         if (taop->tao_ccsent == 0)
1398                                 taop->tao_ccsent = to.to_ccecho;
1399
1400                         tp->rcv_adv += tp->rcv_wnd;
1401                         tp->snd_una++;          /* SYN is acked */
1402                         callout_stop(tp->tt_rexmt);
1403                         /*
1404                          * If there's data, delay ACK; if there's also a FIN
1405                          * ACKNOW will be turned on later.
1406                          */
1407                         if (DELAY_ACK(tp) && tlen != 0)
1408                                 callout_reset(tp->tt_delack, tcp_delacktime,
1409                                     tcp_timer_delack, tp);
1410                         else
1411                                 tp->t_flags |= TF_ACKNOW;
1412                         /*
1413                          * Received <SYN,ACK> in SYN_SENT[*] state.
1414                          * Transitions:
1415                          *      SYN_SENT  --> ESTABLISHED
1416                          *      SYN_SENT* --> FIN_WAIT_1
1417                          */
1418                         tp->t_starttime = ticks;
1419                         if (tp->t_flags & TF_NEEDFIN) {
1420                                 tp->t_state = TCPS_FIN_WAIT_1;
1421                                 tp->t_flags &= ~TF_NEEDFIN;
1422                                 thflags &= ~TH_SYN;
1423                         } else {
1424                                 tp->t_state = TCPS_ESTABLISHED;
1425                                 callout_reset(tp->tt_keep, tcp_keepidle,
1426                                               tcp_timer_keep, tp);
1427                         }
1428                 } else {
1429                         /*
1430                          * Received initial SYN in SYN-SENT[*] state =>
1431                          * simultaneous open.  If segment contains CC option
1432                          * and there is a cached CC, apply TAO test.
1433                          * If it succeeds, connection is * half-synchronized.
1434                          * Otherwise, do 3-way handshake:
1435                          *        SYN-SENT -> SYN-RECEIVED
1436                          *        SYN-SENT* -> SYN-RECEIVED*
1437                          * If there was no CC option, clear cached CC value.
1438                          */
1439                         tp->t_flags |= TF_ACKNOW;
1440                         callout_stop(tp->tt_rexmt);
1441                         if (to.to_flags & TOF_CC) {
1442                                 if (taop->tao_cc != 0 &&
1443                                     CC_GT(to.to_cc, taop->tao_cc)) {
1444                                         /*
1445                                          * update cache and make transition:
1446                                          *        SYN-SENT -> ESTABLISHED*
1447                                          *        SYN-SENT* -> FIN-WAIT-1*
1448                                          */
1449                                         taop->tao_cc = to.to_cc;
1450                                         tp->t_starttime = ticks;
1451                                         if (tp->t_flags & TF_NEEDFIN) {
1452                                                 tp->t_state = TCPS_FIN_WAIT_1;
1453                                                 tp->t_flags &= ~TF_NEEDFIN;
1454                                         } else {
1455                                                 tp->t_state = TCPS_ESTABLISHED;
1456                                                 callout_reset(tp->tt_keep,
1457                                                               tcp_keepidle,
1458                                                               tcp_timer_keep,
1459                                                               tp);
1460                                         }
1461                                         tp->t_flags |= TF_NEEDSYN;
1462                                 } else
1463                                         tp->t_state = TCPS_SYN_RECEIVED;
1464                         } else {
1465                                 /* CC.NEW or no option => invalidate cache */
1466                                 taop->tao_cc = 0;
1467                                 tp->t_state = TCPS_SYN_RECEIVED;
1468                         }
1469                 }
1470
1471 trimthenstep6:
1472                 /*
1473                  * Advance th->th_seq to correspond to first data byte.
1474                  * If data, trim to stay within window,
1475                  * dropping FIN if necessary.
1476                  */
1477                 th->th_seq++;
1478                 if (tlen > tp->rcv_wnd) {
1479                         todrop = tlen - tp->rcv_wnd;
1480                         m_adj(m, -todrop);
1481                         tlen = tp->rcv_wnd;
1482                         thflags &= ~TH_FIN;
1483                         tcpstat.tcps_rcvpackafterwin++;
1484                         tcpstat.tcps_rcvbyteafterwin += todrop;
1485                 }
1486                 tp->snd_wl1 = th->th_seq - 1;
1487                 tp->rcv_up = th->th_seq;
1488                 /*
1489                  * Client side of transaction: already sent SYN and data.
1490                  * If the remote host used T/TCP to validate the SYN,
1491                  * our data will be ACK'd; if so, enter normal data segment
1492                  * processing in the middle of step 5, ack processing.
1493                  * Otherwise, goto step 6.
1494                  */
1495                 if (thflags & TH_ACK)
1496                         goto process_ACK;
1497
1498                 goto step6;
1499
1500         /*
1501          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1502          *      if segment contains a SYN and CC [not CC.NEW] option:
1503          *              if state == TIME_WAIT and connection duration > MSL,
1504          *                  drop packet and send RST;
1505          *
1506          *              if SEG.CC > CCrecv then is new SYN, and can implicitly
1507          *                  ack the FIN (and data) in retransmission queue.
1508          *                  Complete close and delete TCPCB.  Then reprocess
1509          *                  segment, hoping to find new TCPCB in LISTEN state;
1510          *
1511          *              else must be old SYN; drop it.
1512          *      else do normal processing.
1513          */
1514         case TCPS_LAST_ACK:
1515         case TCPS_CLOSING:
1516         case TCPS_TIME_WAIT:
1517                 if ((thflags & TH_SYN) &&
1518                     (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1519                         if (tp->t_state == TCPS_TIME_WAIT &&
1520                                         (ticks - tp->t_starttime) > tcp_msl) {
1521                                 rstreason = BANDLIM_UNLIMITED;
1522                                 goto dropwithreset;
1523                         }
1524                         if (CC_GT(to.to_cc, tp->cc_recv)) {
1525                                 tp = tcp_close(tp);
1526                                 goto findpcb;
1527                         }
1528                         else
1529                                 goto drop;
1530                 }
1531                 break;  /* continue normal processing */
1532         }
1533
1534         /*
1535          * States other than LISTEN or SYN_SENT.
1536          * First check the RST flag and sequence number since reset segments
1537          * are exempt from the timestamp and connection count tests.  This
1538          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1539          * below which allowed reset segments in half the sequence space
1540          * to fall though and be processed (which gives forged reset
1541          * segments with a random sequence number a 50 percent chance of
1542          * killing a connection).
1543          * Then check timestamp, if present.
1544          * Then check the connection count, if present.
1545          * Then check that at least some bytes of segment are within
1546          * receive window.  If segment begins before rcv_nxt,
1547          * drop leading data (and SYN); if nothing left, just ack.
1548          *
1549          *
1550          * If the RST bit is set, check the sequence number to see
1551          * if this is a valid reset segment.
1552          * RFC 793 page 37:
1553          *   In all states except SYN-SENT, all reset (RST) segments
1554          *   are validated by checking their SEQ-fields.  A reset is
1555          *   valid if its sequence number is in the window.
1556          * Note: this does not take into account delayed ACKs, so
1557          *   we should test against last_ack_sent instead of rcv_nxt.
1558          *   The sequence number in the reset segment is normally an
1559          *   echo of our outgoing acknowledgement numbers, but some hosts
1560          *   send a reset with the sequence number at the rightmost edge
1561          *   of our receive window, and we have to handle this case.
1562          * If we have multiple segments in flight, the intial reset
1563          * segment sequence numbers will be to the left of last_ack_sent,
1564          * but they will eventually catch up.
1565          * In any case, it never made sense to trim reset segments to
1566          * fit the receive window since RFC 1122 says:
1567          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1568          *
1569          *    A TCP SHOULD allow a received RST segment to include data.
1570          *
1571          *    DISCUSSION
1572          *         It has been suggested that a RST segment could contain
1573          *         ASCII text that encoded and explained the cause of the
1574          *         RST.  No standard has yet been established for such
1575          *         data.
1576          *
1577          * If the reset segment passes the sequence number test examine
1578          * the state:
1579          *    SYN_RECEIVED STATE:
1580          *      If passive open, return to LISTEN state.
1581          *      If active open, inform user that connection was refused.
1582          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1583          *      Inform user that connection was reset, and close tcb.
1584          *    CLOSING, LAST_ACK STATES:
1585          *      Close the tcb.
1586          *    TIME_WAIT STATE:
1587          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1588          *      RFC 1337.
1589          */
1590         if (thflags & TH_RST) {
1591                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1592                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1593                         switch (tp->t_state) {
1594
1595                         case TCPS_SYN_RECEIVED:
1596                                 so->so_error = ECONNREFUSED;
1597                                 goto close;
1598
1599                         case TCPS_ESTABLISHED:
1600                         case TCPS_FIN_WAIT_1:
1601                         case TCPS_FIN_WAIT_2:
1602                         case TCPS_CLOSE_WAIT:
1603                                 so->so_error = ECONNRESET;
1604                         close:
1605                                 tp->t_state = TCPS_CLOSED;
1606                                 tcpstat.tcps_drops++;
1607                                 tp = tcp_close(tp);
1608                                 break;
1609
1610                         case TCPS_CLOSING:
1611                         case TCPS_LAST_ACK:
1612                                 tp = tcp_close(tp);
1613                                 break;
1614
1615                         case TCPS_TIME_WAIT:
1616                                 break;
1617                         }
1618                 }
1619                 goto drop;
1620         }
1621
1622         /*
1623          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1624          * and it's less than ts_recent, drop it.
1625          */
1626         if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1627             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1628
1629                 /* Check to see if ts_recent is over 24 days old.  */
1630                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1631                         /*
1632                          * Invalidate ts_recent.  If this segment updates
1633                          * ts_recent, the age will be reset later and ts_recent
1634                          * will get a valid value.  If it does not, setting
1635                          * ts_recent to zero will at least satisfy the
1636                          * requirement that zero be placed in the timestamp
1637                          * echo reply when ts_recent isn't valid.  The
1638                          * age isn't reset until we get a valid ts_recent
1639                          * because we don't want out-of-order segments to be
1640                          * dropped when ts_recent is old.
1641                          */
1642                         tp->ts_recent = 0;
1643                 } else {
1644                         tcpstat.tcps_rcvduppack++;
1645                         tcpstat.tcps_rcvdupbyte += tlen;
1646                         tcpstat.tcps_pawsdrop++;
1647                         if (tlen)
1648                                 goto dropafterack;
1649                         goto drop;
1650                 }
1651         }
1652
1653         /*
1654          * T/TCP mechanism
1655          *   If T/TCP was negotiated and the segment doesn't have CC,
1656          *   or if its CC is wrong then drop the segment.
1657          *   RST segments do not have to comply with this.
1658          */
1659         if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1660             (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1661                 goto dropafterack;
1662
1663         /*
1664          * In the SYN-RECEIVED state, validate that the packet belongs to
1665          * this connection before trimming the data to fit the receive
1666          * window.  Check the sequence number versus IRS since we know
1667          * the sequence numbers haven't wrapped.  This is a partial fix
1668          * for the "LAND" DoS attack.
1669          */
1670         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1671                 rstreason = BANDLIM_RST_OPENPORT;
1672                 goto dropwithreset;
1673         }
1674
1675         todrop = tp->rcv_nxt - th->th_seq;
1676         if (todrop > 0) {
1677                 if (TCP_DO_SACK(tp)) {
1678                         /* Report duplicate segment at head of packet. */
1679                         tp->reportblk.rblk_start = th->th_seq;
1680                         tp->reportblk.rblk_end = th->th_seq + tlen;
1681                         if (thflags & TH_FIN)
1682                                 ++tp->reportblk.rblk_end;
1683                         if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1684                                 tp->reportblk.rblk_end = tp->rcv_nxt;
1685                         tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1686                 }
1687                 if (thflags & TH_SYN) {
1688                         thflags &= ~TH_SYN;
1689                         th->th_seq++;
1690                         if (th->th_urp > 1)
1691                                 th->th_urp--;
1692                         else
1693                                 thflags &= ~TH_URG;
1694                         todrop--;
1695                 }
1696                 /*
1697                  * Following if statement from Stevens, vol. 2, p. 960.
1698                  */
1699                 if (todrop > tlen ||
1700                     (todrop == tlen && !(thflags & TH_FIN))) {
1701                         /*
1702                          * Any valid FIN must be to the left of the window.
1703                          * At this point the FIN must be a duplicate or out
1704                          * of sequence; drop it.
1705                          */
1706                         thflags &= ~TH_FIN;
1707
1708                         /*
1709                          * Send an ACK to resynchronize and drop any data.
1710                          * But keep on processing for RST or ACK.
1711                          */
1712                         tp->t_flags |= TF_ACKNOW;
1713                         todrop = tlen;
1714                         tcpstat.tcps_rcvduppack++;
1715                         tcpstat.tcps_rcvdupbyte += todrop;
1716                 } else {
1717                         tcpstat.tcps_rcvpartduppack++;
1718                         tcpstat.tcps_rcvpartdupbyte += todrop;
1719                 }
1720                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1721                 th->th_seq += todrop;
1722                 tlen -= todrop;
1723                 if (th->th_urp > todrop)
1724                         th->th_urp -= todrop;
1725                 else {
1726                         thflags &= ~TH_URG;
1727                         th->th_urp = 0;
1728                 }
1729         }
1730
1731         /*
1732          * If new data are received on a connection after the
1733          * user processes are gone, then RST the other end.
1734          */
1735         if ((so->so_state & SS_NOFDREF) &&
1736             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1737                 tp = tcp_close(tp);
1738                 tcpstat.tcps_rcvafterclose++;
1739                 rstreason = BANDLIM_UNLIMITED;
1740                 goto dropwithreset;
1741         }
1742
1743         /*
1744          * If segment ends after window, drop trailing data
1745          * (and PUSH and FIN); if nothing left, just ACK.
1746          */
1747         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1748         if (todrop > 0) {
1749                 tcpstat.tcps_rcvpackafterwin++;
1750                 if (todrop >= tlen) {
1751                         tcpstat.tcps_rcvbyteafterwin += tlen;
1752                         /*
1753                          * If a new connection request is received
1754                          * while in TIME_WAIT, drop the old connection
1755                          * and start over if the sequence numbers
1756                          * are above the previous ones.
1757                          */
1758                         if (thflags & TH_SYN &&
1759                             tp->t_state == TCPS_TIME_WAIT &&
1760                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1761                                 tp = tcp_close(tp);
1762                                 goto findpcb;
1763                         }
1764                         /*
1765                          * If window is closed can only take segments at
1766                          * window edge, and have to drop data and PUSH from
1767                          * incoming segments.  Continue processing, but
1768                          * remember to ack.  Otherwise, drop segment
1769                          * and ack.
1770                          */
1771                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1772                                 tp->t_flags |= TF_ACKNOW;
1773                                 tcpstat.tcps_rcvwinprobe++;
1774                         } else
1775                                 goto dropafterack;
1776                 } else
1777                         tcpstat.tcps_rcvbyteafterwin += todrop;
1778                 m_adj(m, -todrop);
1779                 tlen -= todrop;
1780                 thflags &= ~(TH_PUSH | TH_FIN);
1781         }
1782
1783         /*
1784          * If last ACK falls within this segment's sequence numbers,
1785          * record its timestamp.
1786          * NOTE:
1787          * 1) That the test incorporates suggestions from the latest
1788          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1789          * 2) That updating only on newer timestamps interferes with
1790          *    our earlier PAWS tests, so this check should be solely
1791          *    predicated on the sequence space of this segment.
1792          * 3) That we modify the segment boundary check to be
1793          *        Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1794          *    instead of RFC1323's
1795          *        Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1796          *    This modified check allows us to overcome RFC1323's
1797          *    limitations as described in Stevens TCP/IP Illustrated
1798          *    Vol. 2 p.869. In such cases, we can still calculate the
1799          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1800          */
1801         if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1802             SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1803                                         + ((thflags & TH_SYN) != 0)
1804                                         + ((thflags & TH_FIN) != 0)))) {
1805                 tp->ts_recent_age = ticks;
1806                 tp->ts_recent = to.to_tsval;
1807         }
1808
1809         /*
1810          * If a SYN is in the window, then this is an
1811          * error and we send an RST and drop the connection.
1812          */
1813         if (thflags & TH_SYN) {
1814                 tp = tcp_drop(tp, ECONNRESET);
1815                 rstreason = BANDLIM_UNLIMITED;
1816                 goto dropwithreset;
1817         }
1818
1819         /*
1820          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1821          * flag is on (half-synchronized state), then queue data for
1822          * later processing; else drop segment and return.
1823          */
1824         if (!(thflags & TH_ACK)) {
1825                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1826                     (tp->t_flags & TF_NEEDSYN))
1827                         goto step6;
1828                 else
1829                         goto drop;
1830         }
1831
1832         /*
1833          * Ack processing.
1834          */
1835         switch (tp->t_state) {
1836         /*
1837          * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1838          * ESTABLISHED state and continue processing.
1839          * The ACK was checked above.
1840          */
1841         case TCPS_SYN_RECEIVED:
1842
1843                 tcpstat.tcps_connects++;
1844                 soisconnected(so);
1845                 /* Do window scaling? */
1846                 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1847                     (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1848                         tp->snd_scale = tp->requested_s_scale;
1849                         tp->rcv_scale = tp->request_r_scale;
1850                 }
1851                 /*
1852                  * Upon successful completion of 3-way handshake,
1853                  * update cache.CC if it was undefined, pass any queued
1854                  * data to the user, and advance state appropriately.
1855                  */
1856                 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1857                     taop->tao_cc == 0)
1858                         taop->tao_cc = tp->cc_recv;
1859
1860                 /*
1861                  * Make transitions:
1862                  *      SYN-RECEIVED  -> ESTABLISHED
1863                  *      SYN-RECEIVED* -> FIN-WAIT-1
1864                  */
1865                 tp->t_starttime = ticks;
1866                 if (tp->t_flags & TF_NEEDFIN) {
1867                         tp->t_state = TCPS_FIN_WAIT_1;
1868                         tp->t_flags &= ~TF_NEEDFIN;
1869                 } else {
1870                         tp->t_state = TCPS_ESTABLISHED;
1871                         callout_reset(tp->tt_keep, tcp_keepidle,
1872                                       tcp_timer_keep, tp);
1873                 }
1874                 /*
1875                  * If segment contains data or ACK, will call tcp_reass()
1876                  * later; if not, do so now to pass queued data to user.
1877                  */
1878                 if (tlen == 0 && !(thflags & TH_FIN))
1879                         tcp_reass(tp, NULL, NULL, NULL);
1880                 /* fall into ... */
1881
1882         /*
1883          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1884          * ACKs.  If the ack is in the range
1885          *      tp->snd_una < th->th_ack <= tp->snd_max
1886          * then advance tp->snd_una to th->th_ack and drop
1887          * data from the retransmission queue.  If this ACK reflects
1888          * more up to date window information we update our window information.
1889          */
1890         case TCPS_ESTABLISHED:
1891         case TCPS_FIN_WAIT_1:
1892         case TCPS_FIN_WAIT_2:
1893         case TCPS_CLOSE_WAIT:
1894         case TCPS_CLOSING:
1895         case TCPS_LAST_ACK:
1896         case TCPS_TIME_WAIT:
1897
1898                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1899                         if (TCP_DO_SACK(tp))
1900                                 tcp_sack_update_scoreboard(tp, &to);
1901                         if (tlen != 0 || tiwin != tp->snd_wnd) {
1902                                 tp->t_dupacks = 0;
1903                                 break;
1904                         }
1905                         tcpstat.tcps_rcvdupack++;
1906                         if (!callout_active(tp->tt_rexmt) ||
1907                             th->th_ack != tp->snd_una) {
1908                                 tp->t_dupacks = 0;
1909                                 break;
1910                         }
1911                         /*
1912                          * We have outstanding data (other than
1913                          * a window probe), this is a completely
1914                          * duplicate ack (ie, window info didn't
1915                          * change), the ack is the biggest we've
1916                          * seen and we've seen exactly our rexmt
1917                          * threshhold of them, so assume a packet
1918                          * has been dropped and retransmit it.
1919                          * Kludge snd_nxt & the congestion
1920                          * window so we send only this one
1921                          * packet.
1922                          */
1923                         if (IN_FASTRECOVERY(tp)) {
1924                                 if (TCP_DO_SACK(tp)) {
1925                                         /* No artifical cwnd inflation. */
1926                                         tcp_sack_rexmt(tp, th);
1927                                 } else {
1928                                         /*
1929                                          * Dup acks mean that packets
1930                                          * have left the network
1931                                          * (they're now cached at the
1932                                          * receiver) so bump cwnd by
1933                                          * the amount in the receiver
1934                                          * to keep a constant cwnd
1935                                          * packets in the network.
1936                                          */
1937                                         tp->snd_cwnd += tp->t_maxseg;
1938                                         tcp_output(tp);
1939                                 }
1940                         } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1941                                 tp->t_dupacks = 0;
1942                                 break;
1943                         } else if (++tp->t_dupacks == tcprexmtthresh) {
1944                                 tcp_seq old_snd_nxt;
1945                                 u_int win;
1946
1947 fastretransmit:
1948                                 if (tcp_do_eifel_detect &&
1949                                     (tp->t_flags & TF_RCVD_TSTMP)) {
1950                                         tcp_save_congestion_state(tp);
1951                                         tp->t_flags |= TF_FASTREXMT;
1952                                 }
1953                                 /*
1954                                  * We know we're losing at the current
1955                                  * window size, so do congestion avoidance:
1956                                  * set ssthresh to half the current window
1957                                  * and pull our congestion window back to the
1958                                  * new ssthresh.
1959                                  */
1960                                 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1961                                     tp->t_maxseg;
1962                                 if (win < 2)
1963                                         win = 2;
1964                                 tp->snd_ssthresh = win * tp->t_maxseg;
1965                                 ENTER_FASTRECOVERY(tp);
1966                                 tp->snd_recover = tp->snd_max;
1967                                 callout_stop(tp->tt_rexmt);
1968                                 tp->t_rtttime = 0;
1969                                 old_snd_nxt = tp->snd_nxt;
1970                                 tp->snd_nxt = th->th_ack;
1971                                 tp->snd_cwnd = tp->t_maxseg;
1972                                 tcp_output(tp);
1973                                 ++tcpstat.tcps_sndfastrexmit;
1974                                 tp->snd_cwnd = tp->snd_ssthresh;
1975                                 tp->rexmt_high = tp->snd_nxt;
1976                                 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1977                                         tp->snd_nxt = old_snd_nxt;
1978                                 KASSERT(tp->snd_limited <= 2,
1979                                     ("tp->snd_limited too big"));
1980                                 if (TCP_DO_SACK(tp))
1981                                         tcp_sack_rexmt(tp, th);
1982                                 else
1983                                         tp->snd_cwnd += tp->t_maxseg *
1984                                             (tp->t_dupacks - tp->snd_limited);
1985                         } else if (tcp_do_limitedtransmit) {
1986                                 u_long oldcwnd = tp->snd_cwnd;
1987                                 tcp_seq oldsndmax = tp->snd_max;
1988                                 tcp_seq oldsndnxt = tp->snd_nxt;
1989                                 /* outstanding data */
1990                                 uint32_t ownd = tp->snd_max - tp->snd_una;
1991                                 u_int sent;
1992
1993 #define iceildiv(n, d)          (((n)+(d)-1) / (d))
1994
1995                                 KASSERT(tp->t_dupacks == 1 ||
1996                                         tp->t_dupacks == 2,
1997                                     ("dupacks not 1 or 2"));
1998                                 if (tp->t_dupacks == 1)
1999                                         tp->snd_limited = 0;
2000                                 tp->snd_nxt = tp->snd_max;
2001                                 tp->snd_cwnd = ownd +
2002                                     (tp->t_dupacks - tp->snd_limited) *
2003                                     tp->t_maxseg;
2004                                 tcp_output(tp);
2005
2006                                 /*
2007                                  * Other acks may have been processed,
2008                                  * snd_nxt cannot be reset to a value less
2009                                  * then snd_una.
2010                                  */
2011                                 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2012                                     if (SEQ_GT(oldsndnxt, tp->snd_una))
2013                                         tp->snd_nxt = oldsndnxt;
2014                                     else
2015                                         tp->snd_nxt = tp->snd_una;
2016                                 }
2017                                 tp->snd_cwnd = oldcwnd;
2018                                 sent = tp->snd_max - oldsndmax;
2019                                 if (sent > tp->t_maxseg) {
2020                                         KASSERT((tp->t_dupacks == 2 &&
2021                                                  tp->snd_limited == 0) ||
2022                                                 (sent == tp->t_maxseg + 1 &&
2023                                                  tp->t_flags & TF_SENTFIN),
2024                                             ("sent too much"));
2025                                         KASSERT(sent <= tp->t_maxseg * 2,
2026                                             ("sent too many segments"));
2027                                         tp->snd_limited = 2;
2028                                         tcpstat.tcps_sndlimited += 2;
2029                                 } else if (sent > 0) {
2030                                         ++tp->snd_limited;
2031                                         ++tcpstat.tcps_sndlimited;
2032                                 } else if (tcp_do_early_retransmit &&
2033                                     (tcp_do_eifel_detect &&
2034                                      (tp->t_flags & TF_RCVD_TSTMP)) &&
2035                                     ownd < 4 * tp->t_maxseg &&
2036                                     tp->t_dupacks + 1 >=
2037                                       iceildiv(ownd, tp->t_maxseg) &&
2038                                     (!TCP_DO_SACK(tp) ||
2039                                      ownd <= tp->t_maxseg ||
2040                                      tcp_sack_has_sacked(&tp->scb,
2041                                                         ownd - tp->t_maxseg))) {
2042                                         ++tcpstat.tcps_sndearlyrexmit;
2043                                         tp->t_flags |= TF_EARLYREXMT;
2044                                         goto fastretransmit;
2045                                 }
2046                         }
2047                         goto drop;
2048                 }
2049
2050                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2051                 tp->t_dupacks = 0;
2052                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2053                         /*
2054                          * Detected optimistic ACK attack.
2055                          * Force slow-start to de-synchronize attack.
2056                          */
2057                         tp->snd_cwnd = tp->t_maxseg;
2058                         tp->snd_wacked = 0;
2059
2060                         tcpstat.tcps_rcvacktoomuch++;
2061                         goto dropafterack;
2062                 }
2063                 /*
2064                  * If we reach this point, ACK is not a duplicate,
2065                  *     i.e., it ACKs something we sent.
2066                  */
2067                 if (tp->t_flags & TF_NEEDSYN) {
2068                         /*
2069                          * T/TCP: Connection was half-synchronized, and our
2070                          * SYN has been ACK'd (so connection is now fully
2071                          * synchronized).  Go to non-starred state,
2072                          * increment snd_una for ACK of SYN, and check if
2073                          * we can do window scaling.
2074                          */
2075                         tp->t_flags &= ~TF_NEEDSYN;
2076                         tp->snd_una++;
2077                         /* Do window scaling? */
2078                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2079                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2080                                 tp->snd_scale = tp->requested_s_scale;
2081                                 tp->rcv_scale = tp->request_r_scale;
2082                         }
2083                 }
2084
2085 process_ACK:
2086                 acked = th->th_ack - tp->snd_una;
2087                 tcpstat.tcps_rcvackpack++;
2088                 tcpstat.tcps_rcvackbyte += acked;
2089
2090                 if (tcp_do_eifel_detect && acked > 0 &&
2091                     (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2092                     (tp->t_flags & TF_FIRSTACCACK)) {
2093                         /* Eifel detection applicable. */
2094                         if (to.to_tsecr < tp->t_rexmtTS) {
2095                                 ++tcpstat.tcps_eifeldetected;
2096                                 tcp_revert_congestion_state(tp);
2097                                 if (tp->t_rxtshift == 1 &&
2098                                     ticks >= tp->t_badrxtwin)
2099                                         ++tcpstat.tcps_rttcantdetect;
2100                         }
2101                 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2102                         /*
2103                          * If we just performed our first retransmit,
2104                          * and the ACK arrives within our recovery window,
2105                          * then it was a mistake to do the retransmit
2106                          * in the first place.  Recover our original cwnd
2107                          * and ssthresh, and proceed to transmit where we
2108                          * left off.
2109                          */
2110                         tcp_revert_congestion_state(tp);
2111                         ++tcpstat.tcps_rttdetected;
2112                 }
2113
2114                 /*
2115                  * If we have a timestamp reply, update smoothed
2116                  * round trip time.  If no timestamp is present but
2117                  * transmit timer is running and timed sequence
2118                  * number was acked, update smoothed round trip time.
2119                  * Since we now have an rtt measurement, cancel the
2120                  * timer backoff (cf., Phil Karn's retransmit alg.).
2121                  * Recompute the initial retransmit timer.
2122                  *
2123                  * Some machines (certain windows boxes) send broken
2124                  * timestamp replies during the SYN+ACK phase, ignore
2125                  * timestamps of 0.
2126                  */
2127                 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2128                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2129                 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2130                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2131                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2132
2133                 /*
2134                  * If no data (only SYN) was ACK'd,
2135                  *    skip rest of ACK processing.
2136                  */
2137                 if (acked == 0)
2138                         goto step6;
2139
2140                 /* Stop looking for an acceptable ACK since one was received. */
2141                 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2142
2143                 if (acked > so->so_snd.ssb_cc) {
2144                         tp->snd_wnd -= so->so_snd.ssb_cc;
2145                         sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2146                         ourfinisacked = TRUE;
2147                 } else {
2148                         sbdrop(&so->so_snd.sb, acked);
2149                         tp->snd_wnd -= acked;
2150                         ourfinisacked = FALSE;
2151                 }
2152                 sowwakeup(so);
2153
2154                 /*
2155                  * Update window information.
2156                  * Don't look at window if no ACK:
2157                  * TAC's send garbage on first SYN.
2158                  */
2159                 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2160                     (tp->snd_wl1 == th->th_seq &&
2161                      (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2162                       (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2163                         /* keep track of pure window updates */
2164                         if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2165                             tiwin > tp->snd_wnd)
2166                                 tcpstat.tcps_rcvwinupd++;
2167                         tp->snd_wnd = tiwin;
2168                         tp->snd_wl1 = th->th_seq;
2169                         tp->snd_wl2 = th->th_ack;
2170                         if (tp->snd_wnd > tp->max_sndwnd)
2171                                 tp->max_sndwnd = tp->snd_wnd;
2172                         needoutput = TRUE;
2173                 }
2174
2175                 tp->snd_una = th->th_ack;
2176                 if (TCP_DO_SACK(tp))
2177                         tcp_sack_update_scoreboard(tp, &to);
2178                 if (IN_FASTRECOVERY(tp)) {
2179                         if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2180                                 EXIT_FASTRECOVERY(tp);
2181                                 needoutput = TRUE;
2182                                 /*
2183                                  * If the congestion window was inflated
2184                                  * to account for the other side's
2185                                  * cached packets, retract it.
2186                                  */
2187                                 if (!TCP_DO_SACK(tp))
2188                                         tp->snd_cwnd = tp->snd_ssthresh;
2189
2190                                 /*
2191                                  * Window inflation should have left us
2192                                  * with approximately snd_ssthresh outstanding
2193                                  * data.  But, in case we would be inclined
2194                                  * to send a burst, better do it using
2195                                  * slow start.
2196                                  */
2197                                 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2198                                            tp->snd_max + 2 * tp->t_maxseg))
2199                                         tp->snd_cwnd =
2200                                             (tp->snd_max - tp->snd_una) +
2201                                             2 * tp->t_maxseg;
2202
2203                                 tp->snd_wacked = 0;
2204                         } else {
2205                                 if (TCP_DO_SACK(tp)) {
2206                                         tp->snd_max_rexmt = tp->snd_max;
2207                                         tcp_sack_rexmt(tp, th);
2208                                 } else {
2209                                         tcp_newreno_partial_ack(tp, th, acked);
2210                                 }
2211                                 needoutput = FALSE;
2212                         }
2213                 } else {
2214                         /*
2215                          * Open the congestion window.  When in slow-start,
2216                          * open exponentially: maxseg per packet.  Otherwise,
2217                          * open linearly: maxseg per window.
2218                          */
2219                         if (tp->snd_cwnd <= tp->snd_ssthresh) {
2220                                 u_int abc_sslimit =
2221                                     (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2222                                      tp->t_maxseg : 2 * tp->t_maxseg);
2223
2224                                 /* slow-start */
2225                                 tp->snd_cwnd += tcp_do_abc ?
2226                                     min(acked, abc_sslimit) : tp->t_maxseg;
2227                         } else {
2228                                 /* linear increase */
2229                                 tp->snd_wacked += tcp_do_abc ? acked :
2230                                     tp->t_maxseg;
2231                                 if (tp->snd_wacked >= tp->snd_cwnd) {
2232                                         tp->snd_wacked -= tp->snd_cwnd;
2233                                         tp->snd_cwnd += tp->t_maxseg;
2234                                 }
2235                         }
2236                         tp->snd_cwnd = min(tp->snd_cwnd,
2237                                            TCP_MAXWIN << tp->snd_scale);
2238                         tp->snd_recover = th->th_ack - 1;
2239                 }
2240                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2241                         tp->snd_nxt = tp->snd_una;
2242
2243                 /*
2244                  * If all outstanding data is acked, stop retransmit
2245                  * timer and remember to restart (more output or persist).
2246                  * If there is more data to be acked, restart retransmit
2247                  * timer, using current (possibly backed-off) value.
2248                  */
2249                 if (th->th_ack == tp->snd_max) {
2250                         callout_stop(tp->tt_rexmt);
2251                         needoutput = TRUE;
2252                 } else if (!callout_active(tp->tt_persist))
2253                         callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2254                                       tcp_timer_rexmt, tp);
2255
2256                 switch (tp->t_state) {
2257                 /*
2258                  * In FIN_WAIT_1 STATE in addition to the processing
2259                  * for the ESTABLISHED state if our FIN is now acknowledged
2260                  * then enter FIN_WAIT_2.
2261                  */
2262                 case TCPS_FIN_WAIT_1:
2263                         if (ourfinisacked) {
2264                                 /*
2265                                  * If we can't receive any more
2266                                  * data, then closing user can proceed.
2267                                  * Starting the timer is contrary to the
2268                                  * specification, but if we don't get a FIN
2269                                  * we'll hang forever.
2270                                  */
2271                                 if (so->so_state & SS_CANTRCVMORE) {
2272                                         soisdisconnected(so);
2273                                         callout_reset(tp->tt_2msl, tcp_maxidle,
2274                                                       tcp_timer_2msl, tp);
2275                                 }
2276                                 tp->t_state = TCPS_FIN_WAIT_2;
2277                         }
2278                         break;
2279
2280                 /*
2281                  * In CLOSING STATE in addition to the processing for
2282                  * the ESTABLISHED state if the ACK acknowledges our FIN
2283                  * then enter the TIME-WAIT state, otherwise ignore
2284                  * the segment.
2285                  */
2286                 case TCPS_CLOSING:
2287                         if (ourfinisacked) {
2288                                 tp->t_state = TCPS_TIME_WAIT;
2289                                 tcp_canceltimers(tp);
2290                                 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2291                                 if (tp->cc_recv != 0 &&
2292                                     (ticks - tp->t_starttime) < tcp_msl)
2293                                         callout_reset(tp->tt_2msl,
2294                                             tp->t_rxtcur * TCPTV_TWTRUNC,
2295                                             tcp_timer_2msl, tp);
2296                                 else
2297                                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2298                                             tcp_timer_2msl, tp);
2299                                 soisdisconnected(so);
2300                         }
2301                         break;
2302
2303                 /*
2304                  * In LAST_ACK, we may still be waiting for data to drain
2305                  * and/or to be acked, as well as for the ack of our FIN.
2306                  * If our FIN is now acknowledged, delete the TCB,
2307                  * enter the closed state and return.
2308                  */
2309                 case TCPS_LAST_ACK:
2310                         if (ourfinisacked) {
2311                                 tp = tcp_close(tp);
2312                                 goto drop;
2313                         }
2314                         break;
2315
2316                 /*
2317                  * In TIME_WAIT state the only thing that should arrive
2318                  * is a retransmission of the remote FIN.  Acknowledge
2319                  * it and restart the finack timer.
2320                  */
2321                 case TCPS_TIME_WAIT:
2322                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2323                                       tcp_timer_2msl, tp);
2324                         goto dropafterack;
2325                 }
2326         }
2327
2328 step6:
2329         /*
2330          * Update window information.
2331          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2332          */
2333         if ((thflags & TH_ACK) &&
2334             acceptable_window_update(tp, th, tiwin)) {
2335                 /* keep track of pure window updates */
2336                 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2337                     tiwin > tp->snd_wnd)
2338                         tcpstat.tcps_rcvwinupd++;
2339                 tp->snd_wnd = tiwin;
2340                 tp->snd_wl1 = th->th_seq;
2341                 tp->snd_wl2 = th->th_ack;
2342                 if (tp->snd_wnd > tp->max_sndwnd)
2343                         tp->max_sndwnd = tp->snd_wnd;
2344                 needoutput = TRUE;
2345         }
2346
2347         /*
2348          * Process segments with URG.
2349          */
2350         if ((thflags & TH_URG) && th->th_urp &&
2351             !TCPS_HAVERCVDFIN(tp->t_state)) {
2352                 /*
2353                  * This is a kludge, but if we receive and accept
2354                  * random urgent pointers, we'll crash in
2355                  * soreceive.  It's hard to imagine someone
2356                  * actually wanting to send this much urgent data.
2357                  */
2358                 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2359                         th->th_urp = 0;                 /* XXX */
2360                         thflags &= ~TH_URG;             /* XXX */
2361                         goto dodata;                    /* XXX */
2362                 }
2363                 /*
2364                  * If this segment advances the known urgent pointer,
2365                  * then mark the data stream.  This should not happen
2366                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2367                  * a FIN has been received from the remote side.
2368                  * In these states we ignore the URG.
2369                  *
2370                  * According to RFC961 (Assigned Protocols),
2371                  * the urgent pointer points to the last octet
2372                  * of urgent data.  We continue, however,
2373                  * to consider it to indicate the first octet
2374                  * of data past the urgent section as the original
2375                  * spec states (in one of two places).
2376                  */
2377                 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2378                         tp->rcv_up = th->th_seq + th->th_urp;
2379                         so->so_oobmark = so->so_rcv.ssb_cc +
2380                             (tp->rcv_up - tp->rcv_nxt) - 1;
2381                         if (so->so_oobmark == 0)
2382                                 so->so_state |= SS_RCVATMARK;
2383                         sohasoutofband(so);
2384                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2385                 }
2386                 /*
2387                  * Remove out of band data so doesn't get presented to user.
2388                  * This can happen independent of advancing the URG pointer,
2389                  * but if two URG's are pending at once, some out-of-band
2390                  * data may creep in... ick.
2391                  */
2392                 if (th->th_urp <= (u_long)tlen &&
2393                     !(so->so_options & SO_OOBINLINE)) {
2394                         /* hdr drop is delayed */
2395                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2396                 }
2397         } else {
2398                 /*
2399                  * If no out of band data is expected,
2400                  * pull receive urgent pointer along
2401                  * with the receive window.
2402                  */
2403                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2404                         tp->rcv_up = tp->rcv_nxt;
2405         }
2406
2407 dodata:                                                 /* XXX */
2408         /*
2409          * Process the segment text, merging it into the TCP sequencing queue,
2410          * and arranging for acknowledgment of receipt if necessary.
2411          * This process logically involves adjusting tp->rcv_wnd as data
2412          * is presented to the user (this happens in tcp_usrreq.c,
2413          * case PRU_RCVD).  If a FIN has already been received on this
2414          * connection then we just ignore the text.
2415          */
2416         if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2417                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2418                 /*
2419                  * Insert segment which includes th into TCP reassembly queue
2420                  * with control block tp.  Set thflags to whether reassembly now
2421                  * includes a segment with FIN.  This handles the common case
2422                  * inline (segment is the next to be received on an established
2423                  * connection, and the queue is empty), avoiding linkage into
2424                  * and removal from the queue and repetition of various
2425                  * conversions.
2426                  * Set DELACK for segments received in order, but ack
2427                  * immediately when segments are out of order (so
2428                  * fast retransmit can work).
2429                  */
2430                 if (th->th_seq == tp->rcv_nxt &&
2431                     LIST_EMPTY(&tp->t_segq) &&
2432                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2433                         if (DELAY_ACK(tp))
2434                                 callout_reset(tp->tt_delack, tcp_delacktime,
2435                                               tcp_timer_delack, tp);
2436                         else
2437                                 tp->t_flags |= TF_ACKNOW;
2438                         tp->rcv_nxt += tlen;
2439                         thflags = th->th_flags & TH_FIN;
2440                         tcpstat.tcps_rcvpack++;
2441                         tcpstat.tcps_rcvbyte += tlen;
2442                         ND6_HINT(tp);
2443                         if (so->so_state & SS_CANTRCVMORE)
2444                                 m_freem(m);
2445                         else
2446                                 ssb_appendstream(&so->so_rcv, m);
2447                         sorwakeup(so);
2448                 } else {
2449                         if (!(tp->t_flags & TF_DUPSEG)) {
2450                                 /* Initialize SACK report block. */
2451                                 tp->reportblk.rblk_start = th->th_seq;
2452                                 tp->reportblk.rblk_end = th->th_seq + tlen +
2453                                     ((thflags & TH_FIN) != 0);
2454                         }
2455                         thflags = tcp_reass(tp, th, &tlen, m);
2456                         tp->t_flags |= TF_ACKNOW;
2457                 }
2458
2459                 /*
2460                  * Note the amount of data that peer has sent into
2461                  * our window, in order to estimate the sender's
2462                  * buffer size.
2463                  */
2464                 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2465         } else {
2466                 m_freem(m);
2467                 thflags &= ~TH_FIN;
2468         }
2469
2470         /*
2471          * If FIN is received ACK the FIN and let the user know
2472          * that the connection is closing.
2473          */
2474         if (thflags & TH_FIN) {
2475                 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2476                         socantrcvmore(so);
2477                         /*
2478                          * If connection is half-synchronized
2479                          * (ie NEEDSYN flag on) then delay ACK,
2480                          * so it may be piggybacked when SYN is sent.
2481                          * Otherwise, since we received a FIN then no
2482                          * more input can be expected, send ACK now.
2483                          */
2484                         if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2485                                 callout_reset(tp->tt_delack, tcp_delacktime,
2486                                     tcp_timer_delack, tp);
2487                         else
2488                                 tp->t_flags |= TF_ACKNOW;
2489                         tp->rcv_nxt++;
2490                 }
2491
2492                 switch (tp->t_state) {
2493                 /*
2494                  * In SYN_RECEIVED and ESTABLISHED STATES
2495                  * enter the CLOSE_WAIT state.
2496                  */
2497                 case TCPS_SYN_RECEIVED:
2498                         tp->t_starttime = ticks;
2499                         /*FALLTHROUGH*/
2500                 case TCPS_ESTABLISHED:
2501                         tp->t_state = TCPS_CLOSE_WAIT;
2502                         break;
2503
2504                 /*
2505                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2506                  * enter the CLOSING state.
2507                  */
2508                 case TCPS_FIN_WAIT_1:
2509                         tp->t_state = TCPS_CLOSING;
2510                         break;
2511
2512                 /*
2513                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2514                  * starting the time-wait timer, turning off the other
2515                  * standard timers.
2516                  */
2517                 case TCPS_FIN_WAIT_2:
2518                         tp->t_state = TCPS_TIME_WAIT;
2519                         tcp_canceltimers(tp);
2520                         /* Shorten TIME_WAIT [RFC-1644, p.28] */
2521                         if (tp->cc_recv != 0 &&
2522                             (ticks - tp->t_starttime) < tcp_msl) {
2523                                 callout_reset(tp->tt_2msl,
2524                                               tp->t_rxtcur * TCPTV_TWTRUNC,
2525                                               tcp_timer_2msl, tp);
2526                                 /* For transaction client, force ACK now. */
2527                                 tp->t_flags |= TF_ACKNOW;
2528                         }
2529                         else
2530                                 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2531                                               tcp_timer_2msl, tp);
2532                         soisdisconnected(so);
2533                         break;
2534
2535                 /*
2536                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2537                  */
2538                 case TCPS_TIME_WAIT:
2539                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2540                                       tcp_timer_2msl, tp);
2541                         break;
2542                 }
2543         }
2544
2545 #ifdef TCPDEBUG
2546         if (so->so_options & SO_DEBUG)
2547                 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2548 #endif
2549
2550         /*
2551          * Return any desired output.
2552          */
2553         if (needoutput || (tp->t_flags & TF_ACKNOW))
2554                 tcp_output(tp);
2555         return;
2556
2557 dropafterack:
2558         /*
2559          * Generate an ACK dropping incoming segment if it occupies
2560          * sequence space, where the ACK reflects our state.
2561          *
2562          * We can now skip the test for the RST flag since all
2563          * paths to this code happen after packets containing
2564          * RST have been dropped.
2565          *
2566          * In the SYN-RECEIVED state, don't send an ACK unless the
2567          * segment we received passes the SYN-RECEIVED ACK test.
2568          * If it fails send a RST.  This breaks the loop in the
2569          * "LAND" DoS attack, and also prevents an ACK storm
2570          * between two listening ports that have been sent forged
2571          * SYN segments, each with the source address of the other.
2572          */
2573         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2574             (SEQ_GT(tp->snd_una, th->th_ack) ||
2575              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2576                 rstreason = BANDLIM_RST_OPENPORT;
2577                 goto dropwithreset;
2578         }
2579 #ifdef TCPDEBUG
2580         if (so->so_options & SO_DEBUG)
2581                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2582 #endif
2583         m_freem(m);
2584         tp->t_flags |= TF_ACKNOW;
2585         tcp_output(tp);
2586         return;
2587
2588 dropwithreset:
2589         /*
2590          * Generate a RST, dropping incoming segment.
2591          * Make ACK acceptable to originator of segment.
2592          * Don't bother to respond if destination was broadcast/multicast.
2593          */
2594         if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2595                 goto drop;
2596         if (isipv6) {
2597                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2598                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2599                         goto drop;
2600         } else {
2601                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2602                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2603                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2604                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2605                         goto drop;
2606         }
2607         /* IPv6 anycast check is done at tcp6_input() */
2608
2609         /*
2610          * Perform bandwidth limiting.
2611          */
2612 #ifdef ICMP_BANDLIM
2613         if (badport_bandlim(rstreason) < 0)
2614                 goto drop;
2615 #endif
2616
2617 #ifdef TCPDEBUG
2618         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2619                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2620 #endif
2621         if (thflags & TH_ACK)
2622                 /* mtod() below is safe as long as hdr dropping is delayed */
2623                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2624                             TH_RST);
2625         else {
2626                 if (thflags & TH_SYN)
2627                         tlen++;
2628                 /* mtod() below is safe as long as hdr dropping is delayed */
2629                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2630                             (tcp_seq)0, TH_RST | TH_ACK);
2631         }
2632         return;
2633
2634 drop:
2635         /*
2636          * Drop space held by incoming segment and return.
2637          */
2638 #ifdef TCPDEBUG
2639         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2640                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2641 #endif
2642         m_freem(m);
2643         return;
2644 }
2645
2646 /*
2647  * Parse TCP options and place in tcpopt.
2648  */
2649 static void
2650 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2651 {
2652         int opt, optlen, i;
2653
2654         to->to_flags = 0;
2655         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2656                 opt = cp[0];
2657                 if (opt == TCPOPT_EOL)
2658                         break;
2659                 if (opt == TCPOPT_NOP)
2660                         optlen = 1;
2661                 else {
2662                         if (cnt < 2)
2663                                 break;
2664                         optlen = cp[1];
2665                         if (optlen < 2 || optlen > cnt)
2666                                 break;
2667                 }
2668                 switch (opt) {
2669                 case TCPOPT_MAXSEG:
2670                         if (optlen != TCPOLEN_MAXSEG)
2671                                 continue;
2672                         if (!is_syn)
2673                                 continue;
2674                         to->to_flags |= TOF_MSS;
2675                         bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2676                         to->to_mss = ntohs(to->to_mss);
2677                         break;
2678                 case TCPOPT_WINDOW:
2679                         if (optlen != TCPOLEN_WINDOW)
2680                                 continue;
2681                         if (!is_syn)
2682                                 continue;
2683                         to->to_flags |= TOF_SCALE;
2684                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2685                         break;
2686                 case TCPOPT_TIMESTAMP:
2687                         if (optlen != TCPOLEN_TIMESTAMP)
2688                                 continue;
2689                         to->to_flags |= TOF_TS;
2690                         bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2691                         to->to_tsval = ntohl(to->to_tsval);
2692                         bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2693                         to->to_tsecr = ntohl(to->to_tsecr);
2694                         /*
2695                          * If echoed timestamp is later than the current time,
2696                          * fall back to non RFC1323 RTT calculation.
2697                          */
2698                         if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2699                                 to->to_tsecr = 0;
2700                         break;
2701                 case TCPOPT_CC:
2702                         if (optlen != TCPOLEN_CC)
2703                                 continue;
2704                         to->to_flags |= TOF_CC;
2705                         bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2706                         to->to_cc = ntohl(to->to_cc);
2707                         break;
2708                 case TCPOPT_CCNEW:
2709                         if (optlen != TCPOLEN_CC)
2710                                 continue;
2711                         if (!is_syn)
2712                                 continue;
2713                         to->to_flags |= TOF_CCNEW;
2714                         bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2715                         to->to_cc = ntohl(to->to_cc);
2716                         break;
2717                 case TCPOPT_CCECHO:
2718                         if (optlen != TCPOLEN_CC)
2719                                 continue;
2720                         if (!is_syn)
2721                                 continue;
2722                         to->to_flags |= TOF_CCECHO;
2723                         bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho);
2724                         to->to_ccecho = ntohl(to->to_ccecho);
2725                         break;
2726                 case TCPOPT_SACK_PERMITTED:
2727                         if (optlen != TCPOLEN_SACK_PERMITTED)
2728                                 continue;
2729                         if (!is_syn)
2730                                 continue;
2731                         to->to_flags |= TOF_SACK_PERMITTED;
2732                         break;
2733                 case TCPOPT_SACK:
2734                         if ((optlen - 2) & 0x07)        /* not multiple of 8 */
2735                                 continue;
2736                         to->to_nsackblocks = (optlen - 2) / 8;
2737                         to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2738                         to->to_flags |= TOF_SACK;
2739                         for (i = 0; i < to->to_nsackblocks; i++) {
2740                                 struct raw_sackblock *r = &to->to_sackblocks[i];
2741
2742                                 r->rblk_start = ntohl(r->rblk_start);
2743                                 r->rblk_end = ntohl(r->rblk_end);
2744                         }
2745                         break;
2746                 default:
2747                         continue;
2748                 }
2749         }
2750 }
2751
2752 /*
2753  * Pull out of band byte out of a segment so
2754  * it doesn't appear in the user's data queue.
2755  * It is still reflected in the segment length for
2756  * sequencing purposes.
2757  * "off" is the delayed to be dropped hdrlen.
2758  */
2759 static void
2760 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2761 {
2762         int cnt = off + th->th_urp - 1;
2763
2764         while (cnt >= 0) {
2765                 if (m->m_len > cnt) {
2766                         char *cp = mtod(m, caddr_t) + cnt;
2767                         struct tcpcb *tp = sototcpcb(so);
2768
2769                         tp->t_iobc = *cp;
2770                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2771                         bcopy(cp + 1, cp, m->m_len - cnt - 1);
2772                         m->m_len--;
2773                         if (m->m_flags & M_PKTHDR)
2774                                 m->m_pkthdr.len--;
2775                         return;
2776                 }
2777                 cnt -= m->m_len;
2778                 m = m->m_next;
2779                 if (m == 0)
2780                         break;
2781         }
2782         panic("tcp_pulloutofband");
2783 }
2784
2785 /*
2786  * Collect new round-trip time estimate
2787  * and update averages and current timeout.
2788  */
2789 static void
2790 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2791 {
2792         int delta;
2793
2794         tcpstat.tcps_rttupdated++;
2795         tp->t_rttupdated++;
2796         if (tp->t_srtt != 0) {
2797                 /*
2798                  * srtt is stored as fixed point with 5 bits after the
2799                  * binary point (i.e., scaled by 8).  The following magic
2800                  * is equivalent to the smoothing algorithm in rfc793 with
2801                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2802                  * point).  Adjust rtt to origin 0.
2803                  */
2804                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2805                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2806
2807                 if ((tp->t_srtt += delta) <= 0)
2808                         tp->t_srtt = 1;
2809
2810                 /*
2811                  * We accumulate a smoothed rtt variance (actually, a
2812                  * smoothed mean difference), then set the retransmit
2813                  * timer to smoothed rtt + 4 times the smoothed variance.
2814                  * rttvar is stored as fixed point with 4 bits after the
2815                  * binary point (scaled by 16).  The following is
2816                  * equivalent to rfc793 smoothing with an alpha of .75
2817                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2818                  * rfc793's wired-in beta.
2819                  */
2820                 if (delta < 0)
2821                         delta = -delta;
2822                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2823                 if ((tp->t_rttvar += delta) <= 0)
2824                         tp->t_rttvar = 1;
2825                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2826                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2827         } else {
2828                 /*
2829                  * No rtt measurement yet - use the unsmoothed rtt.
2830                  * Set the variance to half the rtt (so our first
2831                  * retransmit happens at 3*rtt).
2832                  */
2833                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2834                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2835                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2836         }
2837         tp->t_rtttime = 0;
2838         tp->t_rxtshift = 0;
2839
2840         /*
2841          * the retransmit should happen at rtt + 4 * rttvar.
2842          * Because of the way we do the smoothing, srtt and rttvar
2843          * will each average +1/2 tick of bias.  When we compute
2844          * the retransmit timer, we want 1/2 tick of rounding and
2845          * 1 extra tick because of +-1/2 tick uncertainty in the
2846          * firing of the timer.  The bias will give us exactly the
2847          * 1.5 tick we need.  But, because the bias is
2848          * statistical, we have to test that we don't drop below
2849          * the minimum feasible timer (which is 2 ticks).
2850          */
2851         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2852                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2853
2854         /*
2855          * We received an ack for a packet that wasn't retransmitted;
2856          * it is probably safe to discard any error indications we've
2857          * received recently.  This isn't quite right, but close enough
2858          * for now (a route might have failed after we sent a segment,
2859          * and the return path might not be symmetrical).
2860          */
2861         tp->t_softerror = 0;
2862 }
2863
2864 /*
2865  * Determine a reasonable value for maxseg size.
2866  * If the route is known, check route for mtu.
2867  * If none, use an mss that can be handled on the outgoing
2868  * interface without forcing IP to fragment; if bigger than
2869  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2870  * to utilize large mbufs.  If no route is found, route has no mtu,
2871  * or the destination isn't local, use a default, hopefully conservative
2872  * size (usually 512 or the default IP max size, but no more than the mtu
2873  * of the interface), as we can't discover anything about intervening
2874  * gateways or networks.  We also initialize the congestion/slow start
2875  * window to be a single segment if the destination isn't local.
2876  * While looking at the routing entry, we also initialize other path-dependent
2877  * parameters from pre-set or cached values in the routing entry.
2878  *
2879  * Also take into account the space needed for options that we
2880  * send regularly.  Make maxseg shorter by that amount to assure
2881  * that we can send maxseg amount of data even when the options
2882  * are present.  Store the upper limit of the length of options plus
2883  * data in maxopd.
2884  *
2885  * NOTE that this routine is only called when we process an incoming
2886  * segment, for outgoing segments only tcp_mssopt is called.
2887  *
2888  * In case of T/TCP, we call this routine during implicit connection
2889  * setup as well (offer = -1), to initialize maxseg from the cached
2890  * MSS of our peer.
2891  */
2892 void
2893 tcp_mss(struct tcpcb *tp, int offer)
2894 {
2895         struct rtentry *rt;
2896         struct ifnet *ifp;
2897         int rtt, mss;
2898         u_long bufsize;
2899         struct inpcb *inp = tp->t_inpcb;
2900         struct socket *so;
2901         struct rmxp_tao *taop;
2902         int origoffer = offer;
2903 #ifdef INET6
2904         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2905         size_t min_protoh = isipv6 ?
2906                             sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2907                             sizeof(struct tcpiphdr);
2908 #else
2909         const boolean_t isipv6 = FALSE;
2910         const size_t min_protoh = sizeof(struct tcpiphdr);
2911 #endif
2912
2913         if (isipv6)
2914                 rt = tcp_rtlookup6(&inp->inp_inc);
2915         else
2916                 rt = tcp_rtlookup(&inp->inp_inc);
2917         if (rt == NULL) {
2918                 tp->t_maxopd = tp->t_maxseg =
2919                     (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2920                 return;
2921         }
2922         ifp = rt->rt_ifp;
2923         so = inp->inp_socket;
2924
2925         taop = rmx_taop(rt->rt_rmx);
2926         /*
2927          * Offer == -1 means that we didn't receive SYN yet,
2928          * use cached value in that case;
2929          */
2930         if (offer == -1)
2931                 offer = taop->tao_mssopt;
2932         /*
2933          * Offer == 0 means that there was no MSS on the SYN segment,
2934          * in this case we use tcp_mssdflt.
2935          */
2936         if (offer == 0)
2937                 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2938         else
2939                 /*
2940                  * Sanity check: make sure that maxopd will be large
2941                  * enough to allow some data on segments even is the
2942                  * all the option space is used (40bytes).  Otherwise
2943                  * funny things may happen in tcp_output.
2944                  */
2945                 offer = max(offer, 64);
2946         taop->tao_mssopt = offer;
2947
2948         /*
2949          * While we're here, check if there's an initial rtt
2950          * or rttvar.  Convert from the route-table units
2951          * to scaled multiples of the slow timeout timer.
2952          */
2953         if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2954                 /*
2955                  * XXX the lock bit for RTT indicates that the value
2956                  * is also a minimum value; this is subject to time.
2957                  */
2958                 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2959                         tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2960                 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2961                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2962                 tcpstat.tcps_usedrtt++;
2963                 if (rt->rt_rmx.rmx_rttvar) {
2964                         tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2965                             (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2966                         tcpstat.tcps_usedrttvar++;
2967                 } else {
2968                         /* default variation is +- 1 rtt */
2969                         tp->t_rttvar =
2970                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2971                 }
2972                 TCPT_RANGESET(tp->t_rxtcur,
2973                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2974                               tp->t_rttmin, TCPTV_REXMTMAX);
2975         }
2976         /*
2977          * if there's an mtu associated with the route, use it
2978          * else, use the link mtu.
2979          */
2980         if (rt->rt_rmx.rmx_mtu)
2981                 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2982         else {
2983                 if (isipv6) {
2984                         mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2985                         if (!in6_localaddr(&inp->in6p_faddr))
2986                                 mss = min(mss, tcp_v6mssdflt);
2987                 } else {
2988                         mss = ifp->if_mtu - min_protoh;
2989                         if (!in_localaddr(inp->inp_faddr))
2990                                 mss = min(mss, tcp_mssdflt);
2991                 }
2992         }
2993         mss = min(mss, offer);
2994         /*
2995          * maxopd stores the maximum length of data AND options
2996          * in a segment; maxseg is the amount of data in a normal
2997          * segment.  We need to store this value (maxopd) apart
2998          * from maxseg, because now every segment carries options
2999          * and thus we normally have somewhat less data in segments.
3000          */
3001         tp->t_maxopd = mss;
3002
3003         /*
3004          * In case of T/TCP, origoffer==-1 indicates, that no segments
3005          * were received yet.  In this case we just guess, otherwise
3006          * we do the same as before T/TCP.
3007          */
3008         if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3009             (origoffer == -1 ||
3010              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3011                 mss -= TCPOLEN_TSTAMP_APPA;
3012         if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
3013             (origoffer == -1 ||
3014              (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3015                 mss -= TCPOLEN_CC_APPA;
3016
3017 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
3018                 if (mss > MCLBYTES)
3019                         mss &= ~(MCLBYTES-1);
3020 #else
3021                 if (mss > MCLBYTES)
3022                         mss = mss / MCLBYTES * MCLBYTES;
3023 #endif
3024         /*
3025          * If there's a pipesize, change the socket buffer
3026          * to that size.  Make the socket buffers an integral
3027          * number of mss units; if the mss is larger than
3028          * the socket buffer, decrease the mss.
3029          */
3030 #ifdef RTV_SPIPE
3031         if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3032 #endif
3033                 bufsize = so->so_snd.ssb_hiwat;
3034         if (bufsize < mss)
3035                 mss = bufsize;
3036         else {
3037                 bufsize = roundup(bufsize, mss);
3038                 if (bufsize > sb_max)
3039                         bufsize = sb_max;
3040                 if (bufsize > so->so_snd.ssb_hiwat)
3041                         ssb_reserve(&so->so_snd, bufsize, so, NULL);
3042         }
3043         tp->t_maxseg = mss;
3044
3045 #ifdef RTV_RPIPE
3046         if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3047 #endif
3048                 bufsize = so->so_rcv.ssb_hiwat;
3049         if (bufsize > mss) {
3050                 bufsize = roundup(bufsize, mss);
3051                 if (bufsize > sb_max)
3052                         bufsize = sb_max;
3053                 if (bufsize > so->so_rcv.ssb_hiwat)
3054                         ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3055         }
3056
3057         /*
3058          * Set the slow-start flight size depending on whether this
3059          * is a local network or not.
3060          */
3061         if (tcp_do_rfc3390)
3062                 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3063         else
3064                 tp->snd_cwnd = mss;
3065
3066         if (rt->rt_rmx.rmx_ssthresh) {
3067                 /*
3068                  * There's some sort of gateway or interface
3069                  * buffer limit on the path.  Use this to set
3070                  * the slow start threshhold, but set the
3071                  * threshold to no less than 2*mss.
3072                  */
3073                 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3074                 tcpstat.tcps_usedssthresh++;
3075         }
3076 }
3077
3078 /*
3079  * Determine the MSS option to send on an outgoing SYN.
3080  */
3081 int
3082 tcp_mssopt(struct tcpcb *tp)
3083 {
3084         struct rtentry *rt;
3085 #ifdef INET6
3086         boolean_t isipv6 =
3087             ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3088         int min_protoh = isipv6 ?
3089                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3090                              sizeof(struct tcpiphdr);
3091 #else
3092         const boolean_t isipv6 = FALSE;
3093         const size_t min_protoh = sizeof(struct tcpiphdr);
3094 #endif
3095
3096         if (isipv6)
3097                 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3098         else
3099                 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3100         if (rt == NULL)
3101                 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3102
3103         return (rt->rt_ifp->if_mtu - min_protoh);
3104 }
3105
3106 /*
3107  * When a partial ack arrives, force the retransmission of the
3108  * next unacknowledged segment.  Do not exit Fast Recovery.
3109  *
3110  * Implement the Slow-but-Steady variant of NewReno by restarting the
3111  * the retransmission timer.  Turn it off here so it can be restarted
3112  * later in tcp_output().
3113  */
3114 static void
3115 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3116 {
3117         tcp_seq old_snd_nxt = tp->snd_nxt;
3118         u_long ocwnd = tp->snd_cwnd;
3119
3120         callout_stop(tp->tt_rexmt);
3121         tp->t_rtttime = 0;
3122         tp->snd_nxt = th->th_ack;
3123         /* Set snd_cwnd to one segment beyond acknowledged offset. */
3124         tp->snd_cwnd = tp->t_maxseg;
3125         tp->t_flags |= TF_ACKNOW;
3126         tcp_output(tp);
3127         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3128                 tp->snd_nxt = old_snd_nxt;
3129         /* partial window deflation */
3130         if (ocwnd > acked)
3131                 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3132         else
3133                 tp->snd_cwnd = tp->t_maxseg;
3134 }
3135
3136 /*
3137  * In contrast to the Slow-but-Steady NewReno variant,
3138  * we do not reset the retransmission timer for SACK retransmissions,
3139  * except when retransmitting snd_una.
3140  */
3141 static void
3142 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3143 {
3144         uint32_t pipe, seglen;
3145         tcp_seq nextrexmt;
3146         boolean_t lostdup;
3147         tcp_seq old_snd_nxt = tp->snd_nxt;
3148         u_long ocwnd = tp->snd_cwnd;
3149         int nseg = 0;           /* consecutive new segments */
3150 #define MAXBURST 4              /* limit burst of new packets on partial ack */
3151
3152         tp->t_rtttime = 0;
3153         pipe = tcp_sack_compute_pipe(tp);
3154         while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3155             (!tcp_do_smartsack || nseg < MAXBURST) &&
3156             tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3157                 uint32_t sent;
3158                 tcp_seq old_snd_max;
3159                 int error;
3160
3161                 if (nextrexmt == tp->snd_max)
3162                         ++nseg;
3163                 tp->snd_nxt = nextrexmt;
3164                 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3165                 old_snd_max = tp->snd_max;
3166                 if (nextrexmt == tp->snd_una)
3167                         callout_stop(tp->tt_rexmt);
3168                 error = tcp_output(tp);
3169                 if (error != 0)
3170                         break;
3171                 sent = tp->snd_nxt - nextrexmt;
3172                 if (sent <= 0)
3173                         break;
3174                 if (!lostdup)
3175                         pipe += sent;
3176                 tcpstat.tcps_sndsackpack++;
3177                 tcpstat.tcps_sndsackbyte += sent;
3178                 if (SEQ_LT(nextrexmt, old_snd_max) &&
3179                     SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3180                         tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3181         }
3182         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3183                 tp->snd_nxt = old_snd_nxt;
3184         tp->snd_cwnd = ocwnd;
3185 }