ccacc01e65b71b5e564164eb861f6068e34afbdd
[dragonfly.git] / sys / netinet / tcp_input.c
1 /*
2  * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2002, 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36  *      The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *      This product includes software developed by the University of
49  *      California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *      @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67  * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68  * $DragonFly: src/sys/netinet/tcp_input.c,v 1.68 2008/08/22 09:14:17 sephe Exp $
69  */
70
71 #include "opt_ipfw.h"           /* for ipfw_fwd         */
72 #include "opt_inet6.h"
73 #include "opt_ipsec.h"
74 #include "opt_tcpdebug.h"
75 #include "opt_tcp_input.h"
76
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/malloc.h>
82 #include <sys/mbuf.h>
83 #include <sys/proc.h>           /* for proc0 declaration */
84 #include <sys/protosw.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/syslog.h>
88 #include <sys/in_cksum.h>
89
90 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
91 #include <machine/stdarg.h>
92
93 #include <net/if.h>
94 #include <net/route.h>
95
96 #include <netinet/in.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_icmp.h>    /* for ICMP_BANDLIM */
100 #include <netinet/in_var.h>
101 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
102 #include <netinet/in_pcb.h>
103 #include <netinet/ip_var.h>
104 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet/tcp.h>
110 #include <netinet/tcp_fsm.h>
111 #include <netinet/tcp_seq.h>
112 #include <netinet/tcp_timer.h>
113 #include <netinet/tcp_timer2.h>
114 #include <netinet/tcp_var.h>
115 #include <netinet6/tcp6_var.h>
116 #include <netinet/tcpip.h>
117
118 #ifdef TCPDEBUG
119 #include <netinet/tcp_debug.h>
120
121 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
122 struct tcphdr tcp_savetcp;
123 #endif
124
125 #ifdef FAST_IPSEC
126 #include <netproto/ipsec/ipsec.h>
127 #include <netproto/ipsec/ipsec6.h>
128 #endif
129
130 #ifdef IPSEC
131 #include <netinet6/ipsec.h>
132 #include <netinet6/ipsec6.h>
133 #include <netproto/key/key.h>
134 #endif
135
136 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
137
138 tcp_cc  tcp_ccgen;
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141     &log_in_vain, 0, "Log all incoming TCP connections");
142
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145     &blackhole, 0, "Do not send RST when dropping refused connections");
146
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149     &tcp_delack_enabled, 0,
150     "Delay ACK to try and piggyback it onto a data packet");
151
152 #ifdef TCP_DROP_SYNFIN
153 static int drop_synfin = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
156 #endif
157
158 static int tcp_do_limitedtransmit = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160     &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
161
162 static int tcp_do_early_retransmit = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164     &tcp_do_early_retransmit, 0, "Early retransmit");
165
166 int tcp_aggregate_acks = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168     &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
169
170 int tcp_do_rfc3390 = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
172     &tcp_do_rfc3390, 0,
173     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
174
175 static int tcp_do_eifel_detect = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
177     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
178
179 static int tcp_do_abc = 1;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
181     &tcp_do_abc, 0,
182     "TCP Appropriate Byte Counting (RFC 3465)");
183
184 /*
185  * Define as tunable for easy testing with SACK on and off.
186  * Warning:  do not change setting in the middle of an existing active TCP flow,
187  *   else strange things might happen to that flow.
188  */
189 int tcp_do_sack = 1;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
191     &tcp_do_sack, 0, "Enable SACK Algorithms");
192
193 int tcp_do_smartsack = 1;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
195     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
196
197 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
198     "TCP Segment Reassembly Queue");
199
200 int tcp_reass_maxseg = 0;
201 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
202     &tcp_reass_maxseg, 0,
203     "Global maximum number of TCP Segments in Reassembly Queue");
204
205 int tcp_reass_qsize = 0;
206 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
207     &tcp_reass_qsize, 0,
208     "Global number of TCP Segments currently in Reassembly Queue");
209
210 static int tcp_reass_overflows = 0;
211 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
212     &tcp_reass_overflows, 0,
213     "Global number of TCP Segment Reassembly Queue Overflows");
214
215 int tcp_do_autorcvbuf = 1;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
217     &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
218
219 int tcp_autorcvbuf_inc = 16*1024;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
221     &tcp_autorcvbuf_inc, 0,
222     "Incrementor step size of automatic receive buffer");
223
224 int tcp_autorcvbuf_max = 2*1024*1024;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
226     &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
227
228
229 static void      tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
230 static void      tcp_pulloutofband(struct socket *,
231                      struct tcphdr *, struct mbuf *, int);
232 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
233                      struct mbuf *);
234 static void      tcp_xmit_timer(struct tcpcb *, int);
235 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
236 static void      tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
237
238 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
239 #ifdef INET6
240 #define ND6_HINT(tp) \
241 do { \
242         if ((tp) && (tp)->t_inpcb && \
243             ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
244             (tp)->t_inpcb->in6p_route.ro_rt) \
245                 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
246 } while (0)
247 #else
248 #define ND6_HINT(tp)
249 #endif
250
251 /*
252  * Indicate whether this ack should be delayed.  We can delay the ack if
253  *      - delayed acks are enabled and
254  *      - there is no delayed ack timer in progress and
255  *      - our last ack wasn't a 0-sized window.  We never want to delay
256  *        the ack that opens up a 0-sized window.
257  */
258 #define DELAY_ACK(tp) \
259         (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
260         !(tp->t_flags & TF_RXWIN0SENT))
261
262 #define acceptable_window_update(tp, th, tiwin)                         \
263     (SEQ_LT(tp->snd_wl1, th->th_seq) ||                                 \
264      (tp->snd_wl1 == th->th_seq &&                                      \
265       (SEQ_LT(tp->snd_wl2, th->th_ack) ||                               \
266        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
267
268 static int
269 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
270 {
271         struct tseg_qent *q;
272         struct tseg_qent *p = NULL;
273         struct tseg_qent *te;
274         struct socket *so = tp->t_inpcb->inp_socket;
275         int flags;
276
277         /*
278          * Call with th == NULL after become established to
279          * force pre-ESTABLISHED data up to user socket.
280          */
281         if (th == NULL)
282                 goto present;
283
284         /*
285          * Limit the number of segments in the reassembly queue to prevent
286          * holding on to too many segments (and thus running out of mbufs).
287          * Make sure to let the missing segment through which caused this
288          * queue.  Always keep one global queue entry spare to be able to
289          * process the missing segment.
290          */
291         if (th->th_seq != tp->rcv_nxt &&
292             tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
293                 tcp_reass_overflows++;
294                 tcpstat.tcps_rcvmemdrop++;
295                 m_freem(m);
296                 /* no SACK block to report */
297                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
298                 return (0);
299         }
300
301         /* Allocate a new queue entry. */
302         MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
303                M_INTWAIT | M_NULLOK);
304         if (te == NULL) {
305                 tcpstat.tcps_rcvmemdrop++;
306                 m_freem(m);
307                 /* no SACK block to report */
308                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
309                 return (0);
310         }
311         tcp_reass_qsize++;
312
313         /*
314          * Find a segment which begins after this one does.
315          */
316         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
317                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
318                         break;
319                 p = q;
320         }
321
322         /*
323          * If there is a preceding segment, it may provide some of
324          * our data already.  If so, drop the data from the incoming
325          * segment.  If it provides all of our data, drop us.
326          */
327         if (p != NULL) {
328                 tcp_seq_diff_t i;
329
330                 /* conversion to int (in i) handles seq wraparound */
331                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
332                 if (i > 0) {            /* overlaps preceding segment */
333                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
334                         /* enclosing block starts w/ preceding segment */
335                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
336                         if (i >= *tlenp) {
337                                 /* preceding encloses incoming segment */
338                                 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
339                                     p->tqe_len;
340                                 tcpstat.tcps_rcvduppack++;
341                                 tcpstat.tcps_rcvdupbyte += *tlenp;
342                                 m_freem(m);
343                                 kfree(te, M_TSEGQ);
344                                 tcp_reass_qsize--;
345                                 /*
346                                  * Try to present any queued data
347                                  * at the left window edge to the user.
348                                  * This is needed after the 3-WHS
349                                  * completes.
350                                  */
351                                 goto present;   /* ??? */
352                         }
353                         m_adj(m, i);
354                         *tlenp -= i;
355                         th->th_seq += i;
356                         /* incoming segment end is enclosing block end */
357                         tp->encloseblk.rblk_end = th->th_seq + *tlenp +
358                             ((th->th_flags & TH_FIN) != 0);
359                         /* trim end of reported D-SACK block */
360                         tp->reportblk.rblk_end = th->th_seq;
361                 }
362         }
363         tcpstat.tcps_rcvoopack++;
364         tcpstat.tcps_rcvoobyte += *tlenp;
365
366         /*
367          * While we overlap succeeding segments trim them or,
368          * if they are completely covered, dequeue them.
369          */
370         while (q) {
371                 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
372                 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
373                 struct tseg_qent *nq;
374
375                 if (i <= 0)
376                         break;
377                 if (!(tp->t_flags & TF_DUPSEG)) {    /* first time through */
378                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
379                         tp->encloseblk = tp->reportblk;
380                         /* report trailing duplicate D-SACK segment */
381                         tp->reportblk.rblk_start = q->tqe_th->th_seq;
382                 }
383                 if ((tp->t_flags & TF_ENCLOSESEG) &&
384                     SEQ_GT(qend, tp->encloseblk.rblk_end)) {
385                         /* extend enclosing block if one exists */
386                         tp->encloseblk.rblk_end = qend;
387                 }
388                 if (i < q->tqe_len) {
389                         q->tqe_th->th_seq += i;
390                         q->tqe_len -= i;
391                         m_adj(q->tqe_m, i);
392                         break;
393                 }
394
395                 nq = LIST_NEXT(q, tqe_q);
396                 LIST_REMOVE(q, tqe_q);
397                 m_freem(q->tqe_m);
398                 kfree(q, M_TSEGQ);
399                 tcp_reass_qsize--;
400                 q = nq;
401         }
402
403         /* Insert the new segment queue entry into place. */
404         te->tqe_m = m;
405         te->tqe_th = th;
406         te->tqe_len = *tlenp;
407
408         /* check if can coalesce with following segment */
409         if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
410                 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
411
412                 te->tqe_len += q->tqe_len;
413                 if (q->tqe_th->th_flags & TH_FIN)
414                         te->tqe_th->th_flags |= TH_FIN;
415                 m_cat(te->tqe_m, q->tqe_m);
416                 tp->encloseblk.rblk_end = tend;
417                 /*
418                  * When not reporting a duplicate segment, use
419                  * the larger enclosing block as the SACK block.
420                  */
421                 if (!(tp->t_flags & TF_DUPSEG))
422                         tp->reportblk.rblk_end = tend;
423                 LIST_REMOVE(q, tqe_q);
424                 kfree(q, M_TSEGQ);
425                 tcp_reass_qsize--;
426         }
427
428         if (p == NULL) {
429                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
430         } else {
431                 /* check if can coalesce with preceding segment */
432                 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
433                         p->tqe_len += te->tqe_len;
434                         m_cat(p->tqe_m, te->tqe_m);
435                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
436                         /*
437                          * When not reporting a duplicate segment, use
438                          * the larger enclosing block as the SACK block.
439                          */
440                         if (!(tp->t_flags & TF_DUPSEG))
441                                 tp->reportblk.rblk_start = p->tqe_th->th_seq;
442                         kfree(te, M_TSEGQ);
443                         tcp_reass_qsize--;
444                 } else
445                         LIST_INSERT_AFTER(p, te, tqe_q);
446         }
447
448 present:
449         /*
450          * Present data to user, advancing rcv_nxt through
451          * completed sequence space.
452          */
453         if (!TCPS_HAVEESTABLISHED(tp->t_state))
454                 return (0);
455         q = LIST_FIRST(&tp->t_segq);
456         if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
457                 return (0);
458         tp->rcv_nxt += q->tqe_len;
459         if (!(tp->t_flags & TF_DUPSEG)) {
460                 /* no SACK block to report since ACK advanced */
461                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
462         }
463         /* no enclosing block to report since ACK advanced */
464         tp->t_flags &= ~TF_ENCLOSESEG;
465         flags = q->tqe_th->th_flags & TH_FIN;
466         LIST_REMOVE(q, tqe_q);
467         KASSERT(LIST_EMPTY(&tp->t_segq) ||
468                 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
469                 ("segment not coalesced"));
470         if (so->so_state & SS_CANTRCVMORE)
471                 m_freem(q->tqe_m);
472         else
473                 ssb_appendstream(&so->so_rcv, q->tqe_m);
474         kfree(q, M_TSEGQ);
475         tcp_reass_qsize--;
476         ND6_HINT(tp);
477         sorwakeup(so);
478         return (flags);
479 }
480
481 /*
482  * TCP input routine, follows pages 65-76 of the
483  * protocol specification dated September, 1981 very closely.
484  */
485 #ifdef INET6
486 int
487 tcp6_input(struct mbuf **mp, int *offp, int proto)
488 {
489         struct mbuf *m = *mp;
490         struct in6_ifaddr *ia6;
491
492         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
493
494         /*
495          * draft-itojun-ipv6-tcp-to-anycast
496          * better place to put this in?
497          */
498         ia6 = ip6_getdstifaddr(m);
499         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
500                 struct ip6_hdr *ip6;
501
502                 ip6 = mtod(m, struct ip6_hdr *);
503                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
504                             offsetof(struct ip6_hdr, ip6_dst));
505                 return (IPPROTO_DONE);
506         }
507
508         tcp_input(m, *offp, proto);
509         return (IPPROTO_DONE);
510 }
511 #endif
512
513 void
514 tcp_input(struct mbuf *m, ...)
515 {
516         __va_list ap;
517         int off0, proto;
518         struct tcphdr *th;
519         struct ip *ip = NULL;
520         struct ipovly *ipov;
521         struct inpcb *inp = NULL;
522         u_char *optp = NULL;
523         int optlen = 0;
524         int len, tlen, off;
525         int drop_hdrlen;
526         struct tcpcb *tp = NULL;
527         int thflags;
528         struct socket *so = 0;
529         int todrop, acked;
530         boolean_t ourfinisacked, needoutput = FALSE;
531         u_long tiwin;
532         int recvwin;
533         struct tcpopt to;               /* options in this segment */
534         struct rmxp_tao *taop;          /* pointer to our TAO cache entry */
535         struct rmxp_tao tao_noncached;  /* in case there's no cached entry */
536         struct sockaddr_in *next_hop = NULL;
537         int rstreason; /* For badport_bandlim accounting purposes */
538         int cpu;
539         struct ip6_hdr *ip6 = NULL;
540 #ifdef INET6
541         boolean_t isipv6;
542 #else
543         const boolean_t isipv6 = FALSE;
544 #endif
545 #ifdef TCPDEBUG
546         short ostate = 0;
547 #endif
548
549         __va_start(ap, m);
550         off0 = __va_arg(ap, int);
551         proto = __va_arg(ap, int);
552         __va_end(ap);
553
554         tcpstat.tcps_rcvtotal++;
555
556         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
557                 struct m_tag *mtag;
558
559                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
560                 KKASSERT(mtag != NULL);
561                 next_hop = m_tag_data(mtag);
562         }
563
564 #ifdef INET6
565         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
566 #endif
567
568         if (isipv6) {
569                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
570                 ip6 = mtod(m, struct ip6_hdr *);
571                 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
572                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
573                         tcpstat.tcps_rcvbadsum++;
574                         goto drop;
575                 }
576                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
577
578                 /*
579                  * Be proactive about unspecified IPv6 address in source.
580                  * As we use all-zero to indicate unbounded/unconnected pcb,
581                  * unspecified IPv6 address can be used to confuse us.
582                  *
583                  * Note that packets with unspecified IPv6 destination is
584                  * already dropped in ip6_input.
585                  */
586                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
587                         /* XXX stat */
588                         goto drop;
589                 }
590         } else {
591                 /*
592                  * Get IP and TCP header together in first mbuf.
593                  * Note: IP leaves IP header in first mbuf.
594                  */
595                 if (off0 > sizeof(struct ip)) {
596                         ip_stripoptions(m);
597                         off0 = sizeof(struct ip);
598                 }
599                 /* already checked and pulled up in ip_demux() */
600                 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
601                     ("TCP header not in one mbuf: m->m_len %d", m->m_len));
602                 ip = mtod(m, struct ip *);
603                 ipov = (struct ipovly *)ip;
604                 th = (struct tcphdr *)((caddr_t)ip + off0);
605                 tlen = ip->ip_len;
606
607                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
608                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
609                                 th->th_sum = m->m_pkthdr.csum_data;
610                         else
611                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
612                                                 ip->ip_dst.s_addr,
613                                                 htonl(m->m_pkthdr.csum_data +
614                                                         ip->ip_len +
615                                                         IPPROTO_TCP));
616                         th->th_sum ^= 0xffff;
617                 } else {
618                         /*
619                          * Checksum extended TCP header and data.
620                          */
621                         len = sizeof(struct ip) + tlen;
622                         bzero(ipov->ih_x1, sizeof ipov->ih_x1);
623                         ipov->ih_len = (u_short)tlen;
624                         ipov->ih_len = htons(ipov->ih_len);
625                         th->th_sum = in_cksum(m, len);
626                 }
627                 if (th->th_sum) {
628                         tcpstat.tcps_rcvbadsum++;
629                         goto drop;
630                 }
631 #ifdef INET6
632                 /* Re-initialization for later version check */
633                 ip->ip_v = IPVERSION;
634 #endif
635         }
636
637         /*
638          * Check that TCP offset makes sense,
639          * pull out TCP options and adjust length.              XXX
640          */
641         off = th->th_off << 2;
642         /* already checked and pulled up in ip_demux() */
643         KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
644             ("bad TCP data offset %d (tlen %d)", off, tlen));
645         tlen -= off;    /* tlen is used instead of ti->ti_len */
646         if (off > sizeof(struct tcphdr)) {
647                 if (isipv6) {
648                         IP6_EXTHDR_CHECK(m, off0, off, );
649                         ip6 = mtod(m, struct ip6_hdr *);
650                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
651                 } else {
652                         /* already pulled up in ip_demux() */
653                         KASSERT(m->m_len >= sizeof(struct ip) + off,
654                             ("TCP header and options not in one mbuf: "
655                              "m_len %d, off %d", m->m_len, off));
656                 }
657                 optlen = off - sizeof(struct tcphdr);
658                 optp = (u_char *)(th + 1);
659         }
660         thflags = th->th_flags;
661
662 #ifdef TCP_DROP_SYNFIN
663         /*
664          * If the drop_synfin option is enabled, drop all packets with
665          * both the SYN and FIN bits set. This prevents e.g. nmap from
666          * identifying the TCP/IP stack.
667          *
668          * This is a violation of the TCP specification.
669          */
670         if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
671                 goto drop;
672 #endif
673
674         /*
675          * Convert TCP protocol specific fields to host format.
676          */
677         th->th_seq = ntohl(th->th_seq);
678         th->th_ack = ntohl(th->th_ack);
679         th->th_win = ntohs(th->th_win);
680         th->th_urp = ntohs(th->th_urp);
681
682         /*
683          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
684          * until after ip6_savecontrol() is called and before other functions
685          * which don't want those proto headers.
686          * Because ip6_savecontrol() is going to parse the mbuf to
687          * search for data to be passed up to user-land, it wants mbuf
688          * parameters to be unchanged.
689          * XXX: the call of ip6_savecontrol() has been obsoleted based on
690          * latest version of the advanced API (20020110).
691          */
692         drop_hdrlen = off0 + off;
693
694         /*
695          * Locate pcb for segment.
696          */
697 findpcb:
698         /* IPFIREWALL_FORWARD section */
699         if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
700                 /*
701                  * Transparently forwarded. Pretend to be the destination.
702                  * already got one like this?
703                  */
704                 cpu = mycpu->gd_cpuid;
705                 inp = in_pcblookup_hash(&tcbinfo[cpu],
706                                         ip->ip_src, th->th_sport,
707                                         ip->ip_dst, th->th_dport,
708                                         0, m->m_pkthdr.rcvif);
709                 if (!inp) {
710                         /*
711                          * It's new.  Try to find the ambushing socket.
712                          */
713
714                         /*
715                          * The rest of the ipfw code stores the port in
716                          * host order.  XXX
717                          * (The IP address is still in network order.)
718                          */
719                         in_port_t dport = next_hop->sin_port ?
720                                                 htons(next_hop->sin_port) :
721                                                 th->th_dport;
722
723                         cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
724                                           next_hop->sin_addr.s_addr, dport);
725                         inp = in_pcblookup_hash(&tcbinfo[cpu],
726                                                 ip->ip_src, th->th_sport,
727                                                 next_hop->sin_addr, dport,
728                                                 1, m->m_pkthdr.rcvif);
729                 }
730         } else {
731                 if (isipv6) {
732                         inp = in6_pcblookup_hash(&tcbinfo[0],
733                                                  &ip6->ip6_src, th->th_sport,
734                                                  &ip6->ip6_dst, th->th_dport,
735                                                  1, m->m_pkthdr.rcvif);
736                 } else {
737                         cpu = mycpu->gd_cpuid;
738                         inp = in_pcblookup_hash(&tcbinfo[cpu],
739                                                 ip->ip_src, th->th_sport,
740                                                 ip->ip_dst, th->th_dport,
741                                                 1, m->m_pkthdr.rcvif);
742                 }
743         }
744
745         /*
746          * If the state is CLOSED (i.e., TCB does not exist) then
747          * all data in the incoming segment is discarded.
748          * If the TCB exists but is in CLOSED state, it is embryonic,
749          * but should either do a listen or a connect soon.
750          */
751         if (inp == NULL) {
752                 if (log_in_vain) {
753 #ifdef INET6
754                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
755 #else
756                         char dbuf[sizeof "aaa.bbb.ccc.ddd"];
757                         char sbuf[sizeof "aaa.bbb.ccc.ddd"];
758 #endif
759                         if (isipv6) {
760                                 strcpy(dbuf, "[");
761                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
762                                 strcat(dbuf, "]");
763                                 strcpy(sbuf, "[");
764                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
765                                 strcat(sbuf, "]");
766                         } else {
767                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
768                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
769                         }
770                         switch (log_in_vain) {
771                         case 1:
772                                 if (!(thflags & TH_SYN))
773                                         break;
774                         case 2:
775                                 log(LOG_INFO,
776                                     "Connection attempt to TCP %s:%d "
777                                     "from %s:%d flags:0x%02x\n",
778                                     dbuf, ntohs(th->th_dport), sbuf,
779                                     ntohs(th->th_sport), thflags);
780                                 break;
781                         default:
782                                 break;
783                         }
784                 }
785                 if (blackhole) {
786                         switch (blackhole) {
787                         case 1:
788                                 if (thflags & TH_SYN)
789                                         goto drop;
790                                 break;
791                         case 2:
792                                 goto drop;
793                         default:
794                                 goto drop;
795                         }
796                 }
797                 rstreason = BANDLIM_RST_CLOSEDPORT;
798                 goto dropwithreset;
799         }
800
801 #ifdef IPSEC
802         if (isipv6) {
803                 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
804                         ipsec6stat.in_polvio++;
805                         goto drop;
806                 }
807         } else {
808                 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
809                         ipsecstat.in_polvio++;
810                         goto drop;
811                 }
812         }
813 #endif
814 #ifdef FAST_IPSEC
815         if (isipv6) {
816                 if (ipsec6_in_reject(m, inp))
817                         goto drop;
818         } else {
819                 if (ipsec4_in_reject(m, inp))
820                         goto drop;
821         }
822 #endif
823         /* Check the minimum TTL for socket. */
824 #ifdef INET6
825         if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
826                 goto drop;
827 #endif
828
829         tp = intotcpcb(inp);
830         if (tp == NULL) {
831                 rstreason = BANDLIM_RST_CLOSEDPORT;
832                 goto dropwithreset;
833         }
834         if (tp->t_state <= TCPS_CLOSED)
835                 goto drop;
836
837         /* Unscale the window into a 32-bit value. */
838         if (!(thflags & TH_SYN))
839                 tiwin = th->th_win << tp->snd_scale;
840         else
841                 tiwin = th->th_win;
842
843         so = inp->inp_socket;
844
845 #ifdef TCPDEBUG
846         if (so->so_options & SO_DEBUG) {
847                 ostate = tp->t_state;
848                 if (isipv6)
849                         bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
850                 else
851                         bcopy(ip, tcp_saveipgen, sizeof(*ip));
852                 tcp_savetcp = *th;
853         }
854 #endif
855
856         bzero(&to, sizeof to);
857
858         if (so->so_options & SO_ACCEPTCONN) {
859                 struct in_conninfo inc;
860
861 #ifdef INET6
862                 inc.inc_isipv6 = (isipv6 == TRUE);
863 #endif
864                 if (isipv6) {
865                         inc.inc6_faddr = ip6->ip6_src;
866                         inc.inc6_laddr = ip6->ip6_dst;
867                         inc.inc6_route.ro_rt = NULL;            /* XXX */
868                 } else {
869                         inc.inc_faddr = ip->ip_src;
870                         inc.inc_laddr = ip->ip_dst;
871                         inc.inc_route.ro_rt = NULL;             /* XXX */
872                 }
873                 inc.inc_fport = th->th_sport;
874                 inc.inc_lport = th->th_dport;
875
876                 /*
877                  * If the state is LISTEN then ignore segment if it contains
878                  * a RST.  If the segment contains an ACK then it is bad and
879                  * send a RST.  If it does not contain a SYN then it is not
880                  * interesting; drop it.
881                  *
882                  * If the state is SYN_RECEIVED (syncache) and seg contains
883                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
884                  * contains a RST, check the sequence number to see if it
885                  * is a valid reset segment.
886                  */
887                 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
888                         if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
889                                 if (!syncache_expand(&inc, th, &so, m)) {
890                                         /*
891                                          * No syncache entry, or ACK was not
892                                          * for our SYN/ACK.  Send a RST.
893                                          */
894                                         tcpstat.tcps_badsyn++;
895                                         rstreason = BANDLIM_RST_OPENPORT;
896                                         goto dropwithreset;
897                                 }
898                                 if (so == NULL)
899                                         /*
900                                          * Could not complete 3-way handshake,
901                                          * connection is being closed down, and
902                                          * syncache will free mbuf.
903                                          */
904                                         return;
905                                 /*
906                                  * Socket is created in state SYN_RECEIVED.
907                                  * Continue processing segment.
908                                  */
909                                 inp = so->so_pcb;
910                                 tp = intotcpcb(inp);
911                                 /*
912                                  * This is what would have happened in
913                                  * tcp_output() when the SYN,ACK was sent.
914                                  */
915                                 tp->snd_up = tp->snd_una;
916                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
917                                 tp->last_ack_sent = tp->rcv_nxt;
918 /*
919  * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
920  * until the _second_ ACK is received:
921  *    rcv SYN (set wscale opts)  --> send SYN/ACK, set snd_wnd = window.
922  *    rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
923  *        move to ESTAB, set snd_wnd to tiwin.
924  */
925                                 tp->snd_wnd = tiwin;    /* unscaled */
926                                 goto after_listen;
927                         }
928                         if (thflags & TH_RST) {
929                                 syncache_chkrst(&inc, th);
930                                 goto drop;
931                         }
932                         if (thflags & TH_ACK) {
933                                 syncache_badack(&inc);
934                                 tcpstat.tcps_badsyn++;
935                                 rstreason = BANDLIM_RST_OPENPORT;
936                                 goto dropwithreset;
937                         }
938                         goto drop;
939                 }
940
941                 /*
942                  * Segment's flags are (SYN) or (SYN | FIN).
943                  */
944 #ifdef INET6
945                 /*
946                  * If deprecated address is forbidden,
947                  * we do not accept SYN to deprecated interface
948                  * address to prevent any new inbound connection from
949                  * getting established.
950                  * When we do not accept SYN, we send a TCP RST,
951                  * with deprecated source address (instead of dropping
952                  * it).  We compromise it as it is much better for peer
953                  * to send a RST, and RST will be the final packet
954                  * for the exchange.
955                  *
956                  * If we do not forbid deprecated addresses, we accept
957                  * the SYN packet.  RFC2462 does not suggest dropping
958                  * SYN in this case.
959                  * If we decipher RFC2462 5.5.4, it says like this:
960                  * 1. use of deprecated addr with existing
961                  *    communication is okay - "SHOULD continue to be
962                  *    used"
963                  * 2. use of it with new communication:
964                  *   (2a) "SHOULD NOT be used if alternate address
965                  *        with sufficient scope is available"
966                  *   (2b) nothing mentioned otherwise.
967                  * Here we fall into (2b) case as we have no choice in
968                  * our source address selection - we must obey the peer.
969                  *
970                  * The wording in RFC2462 is confusing, and there are
971                  * multiple description text for deprecated address
972                  * handling - worse, they are not exactly the same.
973                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
974                  */
975                 if (isipv6 && !ip6_use_deprecated) {
976                         struct in6_ifaddr *ia6;
977
978                         if ((ia6 = ip6_getdstifaddr(m)) &&
979                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
980                                 tp = NULL;
981                                 rstreason = BANDLIM_RST_OPENPORT;
982                                 goto dropwithreset;
983                         }
984                 }
985 #endif
986                 /*
987                  * If it is from this socket, drop it, it must be forged.
988                  * Don't bother responding if the destination was a broadcast.
989                  */
990                 if (th->th_dport == th->th_sport) {
991                         if (isipv6) {
992                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
993                                                        &ip6->ip6_src))
994                                         goto drop;
995                         } else {
996                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
997                                         goto drop;
998                         }
999                 }
1000                 /*
1001                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1002                  *
1003                  * Note that it is quite possible to receive unicast
1004                  * link-layer packets with a broadcast IP address. Use
1005                  * in_broadcast() to find them.
1006                  */
1007                 if (m->m_flags & (M_BCAST | M_MCAST))
1008                         goto drop;
1009                 if (isipv6) {
1010                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1011                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1012                                 goto drop;
1013                 } else {
1014                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1015                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1016                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1017                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1018                                 goto drop;
1019                 }
1020                 /*
1021                  * SYN appears to be valid; create compressed TCP state
1022                  * for syncache, or perform t/tcp connection.
1023                  */
1024                 if (so->so_qlen <= so->so_qlimit) {
1025                         tcp_dooptions(&to, optp, optlen, TRUE);
1026                         if (!syncache_add(&inc, &to, th, &so, m))
1027                                 goto drop;
1028                         if (so == NULL)
1029                                 /*
1030                                  * Entry added to syncache, mbuf used to
1031                                  * send SYN,ACK packet.
1032                                  */
1033                                 return;
1034                         /*
1035                          * Segment passed TAO tests.
1036                          */
1037                         inp = so->so_pcb;
1038                         tp = intotcpcb(inp);
1039                         tp->snd_wnd = tiwin;
1040                         tp->t_starttime = ticks;
1041                         tp->t_state = TCPS_ESTABLISHED;
1042
1043                         /*
1044                          * If there is a FIN, or if there is data and the
1045                          * connection is local, then delay SYN,ACK(SYN) in
1046                          * the hope of piggy-backing it on a response
1047                          * segment.  Otherwise must send ACK now in case
1048                          * the other side is slow starting.
1049                          */
1050                         if (DELAY_ACK(tp) &&
1051                             ((thflags & TH_FIN) ||
1052                              (tlen != 0 &&
1053                               ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1054                                (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1055                                 tcp_callout_reset(tp, tp->tt_delack,
1056                                     tcp_delacktime, tcp_timer_delack);
1057                                 tp->t_flags |= TF_NEEDSYN;
1058                         } else {
1059                                 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1060                         }
1061
1062                         tcpstat.tcps_connects++;
1063                         soisconnected(so);
1064                         goto trimthenstep6;
1065                 }
1066                 goto drop;
1067         }
1068 after_listen:
1069
1070         /* should not happen - syncache should pick up these connections */
1071         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1072
1073         /*
1074          * This is the second part of the MSS DoS prevention code (after
1075          * minmss on the sending side) and it deals with too many too small
1076          * tcp packets in a too short timeframe (1 second).
1077          *
1078          * XXX Removed.  This code was crap.  It does not scale to network
1079          *     speed, and default values break NFS.  Gone.
1080          */
1081         /* REMOVED */
1082
1083         /*
1084          * Segment received on connection.
1085          *
1086          * Reset idle time and keep-alive timer.  Don't waste time if less
1087          * then a second has elapsed.  Only update t_rcvtime for non-SYN
1088          * packets.
1089          *
1090          * Handle the case where one side thinks the connection is established
1091          * but the other side has, say, rebooted without cleaning out the
1092          * connection.   The SYNs could be construed as an attack and wind
1093          * up ignored, but in case it isn't an attack we can validate the
1094          * connection by forcing a keepalive.
1095          */
1096         if (TCPS_HAVEESTABLISHED(tp->t_state) && (ticks - tp->t_rcvtime) > hz) {
1097                 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
1098                         tp->t_flags |= TF_KEEPALIVE;
1099                         tcp_callout_reset(tp, tp->tt_keep, hz / 2,
1100                                           tcp_timer_keep);
1101                 } else {
1102                         tp->t_rcvtime = ticks;
1103                         tp->t_flags &= ~TF_KEEPALIVE;
1104                         tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1105                                           tcp_timer_keep);
1106                 }
1107         }
1108
1109         /*
1110          * Process options.
1111          * XXX this is tradtitional behavior, may need to be cleaned up.
1112          */
1113         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1114         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1115                 if (to.to_flags & TOF_SCALE) {
1116                         tp->t_flags |= TF_RCVD_SCALE;
1117                         tp->requested_s_scale = to.to_requested_s_scale;
1118                 }
1119                 if (to.to_flags & TOF_TS) {
1120                         tp->t_flags |= TF_RCVD_TSTMP;
1121                         tp->ts_recent = to.to_tsval;
1122                         tp->ts_recent_age = ticks;
1123                 }
1124                 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1125                         tp->t_flags |= TF_RCVD_CC;
1126                 if (to.to_flags & TOF_MSS)
1127                         tcp_mss(tp, to.to_mss);
1128                 /*
1129                  * Only set the TF_SACK_PERMITTED per-connection flag
1130                  * if we got a SACK_PERMITTED option from the other side
1131                  * and the global tcp_do_sack variable is true.
1132                  */
1133                 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1134                         tp->t_flags |= TF_SACK_PERMITTED;
1135         }
1136
1137         /*
1138          * Header prediction: check for the two common cases
1139          * of a uni-directional data xfer.  If the packet has
1140          * no control flags, is in-sequence, the window didn't
1141          * change and we're not retransmitting, it's a
1142          * candidate.  If the length is zero and the ack moved
1143          * forward, we're the sender side of the xfer.  Just
1144          * free the data acked & wake any higher level process
1145          * that was blocked waiting for space.  If the length
1146          * is non-zero and the ack didn't move, we're the
1147          * receiver side.  If we're getting packets in-order
1148          * (the reassembly queue is empty), add the data to
1149          * the socket buffer and note that we need a delayed ack.
1150          * Make sure that the hidden state-flags are also off.
1151          * Since we check for TCPS_ESTABLISHED above, it can only
1152          * be TH_NEEDSYN.
1153          */
1154         if (tp->t_state == TCPS_ESTABLISHED &&
1155             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1156             !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1157             (!(to.to_flags & TOF_TS) ||
1158              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1159             /*
1160              * Using the CC option is compulsory if once started:
1161              *   the segment is OK if no T/TCP was negotiated or
1162              *   if the segment has a CC option equal to CCrecv
1163              */
1164             ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1165              ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1166             th->th_seq == tp->rcv_nxt &&
1167             tp->snd_nxt == tp->snd_max) {
1168
1169                 /*
1170                  * If last ACK falls within this segment's sequence numbers,
1171                  * record the timestamp.
1172                  * NOTE that the test is modified according to the latest
1173                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1174                  */
1175                 if ((to.to_flags & TOF_TS) &&
1176                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1177                         tp->ts_recent_age = ticks;
1178                         tp->ts_recent = to.to_tsval;
1179                 }
1180
1181                 if (tlen == 0) {
1182                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1183                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1184                             tp->snd_cwnd >= tp->snd_wnd &&
1185                             !IN_FASTRECOVERY(tp)) {
1186                                 /*
1187                                  * This is a pure ack for outstanding data.
1188                                  */
1189                                 ++tcpstat.tcps_predack;
1190                                 /*
1191                                  * "bad retransmit" recovery
1192                                  *
1193                                  * If Eifel detection applies, then
1194                                  * it is deterministic, so use it
1195                                  * unconditionally over the old heuristic.
1196                                  * Otherwise, fall back to the old heuristic.
1197                                  */
1198                                 if (tcp_do_eifel_detect &&
1199                                     (to.to_flags & TOF_TS) && to.to_tsecr &&
1200                                     (tp->t_flags & TF_FIRSTACCACK)) {
1201                                         /* Eifel detection applicable. */
1202                                         if (to.to_tsecr < tp->t_rexmtTS) {
1203                                                 tcp_revert_congestion_state(tp);
1204                                                 ++tcpstat.tcps_eifeldetected;
1205                                         }
1206                                 } else if (tp->t_rxtshift == 1 &&
1207                                            ticks < tp->t_badrxtwin) {
1208                                         tcp_revert_congestion_state(tp);
1209                                         ++tcpstat.tcps_rttdetected;
1210                                 }
1211                                 tp->t_flags &= ~(TF_FIRSTACCACK |
1212                                                  TF_FASTREXMT | TF_EARLYREXMT);
1213                                 /*
1214                                  * Recalculate the retransmit timer / rtt.
1215                                  *
1216                                  * Some machines (certain windows boxes)
1217                                  * send broken timestamp replies during the
1218                                  * SYN+ACK phase, ignore timestamps of 0.
1219                                  */
1220                                 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1221                                         tcp_xmit_timer(tp,
1222                                                        ticks - to.to_tsecr + 1);
1223                                 } else if (tp->t_rtttime &&
1224                                            SEQ_GT(th->th_ack, tp->t_rtseq)) {
1225                                         tcp_xmit_timer(tp,
1226                                                        ticks - tp->t_rtttime);
1227                                 }
1228                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1229                                 acked = th->th_ack - tp->snd_una;
1230                                 tcpstat.tcps_rcvackpack++;
1231                                 tcpstat.tcps_rcvackbyte += acked;
1232                                 sbdrop(&so->so_snd.sb, acked);
1233                                 tp->snd_recover = th->th_ack - 1;
1234                                 tp->snd_una = th->th_ack;
1235                                 tp->t_dupacks = 0;
1236                                 /*
1237                                  * Update window information.
1238                                  */
1239                                 if (tiwin != tp->snd_wnd &&
1240                                     acceptable_window_update(tp, th, tiwin)) {
1241                                         /* keep track of pure window updates */
1242                                         if (tp->snd_wl2 == th->th_ack &&
1243                                             tiwin > tp->snd_wnd)
1244                                                 tcpstat.tcps_rcvwinupd++;
1245                                         tp->snd_wnd = tiwin;
1246                                         tp->snd_wl1 = th->th_seq;
1247                                         tp->snd_wl2 = th->th_ack;
1248                                         if (tp->snd_wnd > tp->max_sndwnd)
1249                                                 tp->max_sndwnd = tp->snd_wnd;
1250                                 }
1251                                 m_freem(m);
1252                                 ND6_HINT(tp); /* some progress has been done */
1253                                 /*
1254                                  * If all outstanding data are acked, stop
1255                                  * retransmit timer, otherwise restart timer
1256                                  * using current (possibly backed-off) value.
1257                                  * If process is waiting for space,
1258                                  * wakeup/selwakeup/signal.  If data
1259                                  * are ready to send, let tcp_output
1260                                  * decide between more output or persist.
1261                                  */
1262                                 if (tp->snd_una == tp->snd_max) {
1263                                         tcp_callout_stop(tp, tp->tt_rexmt);
1264                                 } else if (!tcp_callout_active(tp,
1265                                             tp->tt_persist)) {
1266                                         tcp_callout_reset(tp, tp->tt_rexmt,
1267                                             tp->t_rxtcur, tcp_timer_rexmt);
1268                                 }
1269                                 sowwakeup(so);
1270                                 if (so->so_snd.ssb_cc > 0)
1271                                         tcp_output(tp);
1272                                 return;
1273                         }
1274                 } else if (tiwin == tp->snd_wnd &&
1275                     th->th_ack == tp->snd_una &&
1276                     LIST_EMPTY(&tp->t_segq) &&
1277                     tlen <= ssb_space(&so->so_rcv)) {
1278                         u_long newsize = 0;     /* automatic sockbuf scaling */
1279                         /*
1280                          * This is a pure, in-sequence data packet
1281                          * with nothing on the reassembly queue and
1282                          * we have enough buffer space to take it.
1283                          */
1284                         ++tcpstat.tcps_preddat;
1285                         tp->rcv_nxt += tlen;
1286                         tcpstat.tcps_rcvpack++;
1287                         tcpstat.tcps_rcvbyte += tlen;
1288                         ND6_HINT(tp);   /* some progress has been done */
1289                 /*
1290                  * Automatic sizing of receive socket buffer.  Often the send
1291                  * buffer size is not optimally adjusted to the actual network
1292                  * conditions at hand (delay bandwidth product).  Setting the
1293                  * buffer size too small limits throughput on links with high
1294                  * bandwidth and high delay (eg. trans-continental/oceanic links).
1295                  *
1296                  * On the receive side the socket buffer memory is only rarely
1297                  * used to any significant extent.  This allows us to be much
1298                  * more aggressive in scaling the receive socket buffer.  For
1299                  * the case that the buffer space is actually used to a large
1300                  * extent and we run out of kernel memory we can simply drop
1301                  * the new segments; TCP on the sender will just retransmit it
1302                  * later.  Setting the buffer size too big may only consume too
1303                  * much kernel memory if the application doesn't read() from
1304                  * the socket or packet loss or reordering makes use of the
1305                  * reassembly queue.
1306                  *
1307                  * The criteria to step up the receive buffer one notch are:
1308                  *  1. the number of bytes received during the time it takes
1309                  *     one timestamp to be reflected back to us (the RTT);
1310                  *  2. received bytes per RTT is within seven eighth of the
1311                  *     current socket buffer size;
1312                  *  3. receive buffer size has not hit maximal automatic size;
1313                  *
1314                  * This algorithm does one step per RTT at most and only if
1315                  * we receive a bulk stream w/o packet losses or reorderings.
1316                  * Shrinking the buffer during idle times is not necessary as
1317                  * it doesn't consume any memory when idle.
1318                  *
1319                  * TODO: Only step up if the application is actually serving
1320                  * the buffer to better manage the socket buffer resources.
1321                  */
1322                         if (tcp_do_autorcvbuf &&
1323                             to.to_tsecr &&
1324                             (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1325                                 if (to.to_tsecr > tp->rfbuf_ts &&
1326                                     to.to_tsecr - tp->rfbuf_ts < hz) {
1327                                         if (tp->rfbuf_cnt >
1328                                             (so->so_rcv.ssb_hiwat / 8 * 7) &&
1329                                             so->so_rcv.ssb_hiwat <
1330                                             tcp_autorcvbuf_max) {
1331                                                 newsize =
1332                                                     ulmin(so->so_rcv.ssb_hiwat +
1333                                                           tcp_autorcvbuf_inc,
1334                                                           tcp_autorcvbuf_max);
1335                                         }
1336                                         /* Start over with next RTT. */
1337                                         tp->rfbuf_ts = 0;
1338                                         tp->rfbuf_cnt = 0;
1339                                 } else
1340                                         tp->rfbuf_cnt += tlen;  /* add up */
1341                         }
1342                         /*
1343                          * Add data to socket buffer.
1344                          */
1345                         if (so->so_state & SS_CANTRCVMORE) {
1346                                 m_freem(m);
1347                         } else {
1348                                 /*
1349                                  * Set new socket buffer size, give up when
1350                                  * limit is reached.
1351                                  *
1352                                  * Adjusting the size can mess up ACK
1353                                  * sequencing when pure window updates are
1354                                  * being avoided (which is the default),
1355                                  * so force an ack.
1356                                  */
1357                                 if (newsize) {
1358                                         tp->t_flags |= TF_RXRESIZED;
1359                                         if (!ssb_reserve(&so->so_rcv, newsize,
1360                                                          so, NULL)) {
1361                                                 so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE;
1362                                         }
1363                                         if (newsize >=
1364                                             (TCP_MAXWIN << tp->rcv_scale)) {
1365                                                 so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE;
1366                                         }
1367                                 }
1368                                 m_adj(m, drop_hdrlen); /* delayed header drop */
1369                                 ssb_appendstream(&so->so_rcv, m);
1370                         }
1371                         sorwakeup(so);
1372                         /*
1373                          * This code is responsible for most of the ACKs
1374                          * the TCP stack sends back after receiving a data
1375                          * packet.  Note that the DELAY_ACK check fails if
1376                          * the delack timer is already running, which results
1377                          * in an ack being sent every other packet (which is
1378                          * what we want).
1379                          *
1380                          * We then further aggregate acks by not actually
1381                          * sending one until the protocol thread has completed
1382                          * processing the current backlog of packets.  This
1383                          * does not delay the ack any further, but allows us
1384                          * to take advantage of the packet aggregation that
1385                          * high speed NICs do (usually blocks of 8-10 packets)
1386                          * to send a single ack rather then four or five acks,
1387                          * greatly reducing the ack rate, the return channel
1388                          * bandwidth, and the protocol overhead on both ends.
1389                          *
1390                          * Since this also has the effect of slowing down
1391                          * the exponential slow-start ramp-up, systems with 
1392                          * very large bandwidth-delay products might want
1393                          * to turn the feature off.
1394                          */
1395                         if (DELAY_ACK(tp)) {
1396                                 tcp_callout_reset(tp, tp->tt_delack,
1397                                     tcp_delacktime, tcp_timer_delack);
1398                         } else if (tcp_aggregate_acks) {
1399                                 tp->t_flags |= TF_ACKNOW;
1400                                 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1401                                         tp->t_flags |= TF_ONOUTPUTQ;
1402                                         tp->tt_cpu = mycpu->gd_cpuid;
1403                                         TAILQ_INSERT_TAIL(
1404                                             &tcpcbackq[tp->tt_cpu],
1405                                             tp, t_outputq);
1406                                 }
1407                         } else {
1408                                 tp->t_flags |= TF_ACKNOW;
1409                                 tcp_output(tp);
1410                         }
1411                         return;
1412                 }
1413         }
1414
1415         /*
1416          * Calculate amount of space in receive window,
1417          * and then do TCP input processing.
1418          * Receive window is amount of space in rcv queue,
1419          * but not less than advertised window.
1420          */
1421         recvwin = ssb_space(&so->so_rcv);
1422         if (recvwin < 0)
1423                 recvwin = 0;
1424         tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1425
1426         /* Reset receive buffer auto scaling when not in bulk receive mode. */
1427         tp->rfbuf_ts = 0;
1428         tp->rfbuf_cnt = 0;
1429
1430         switch (tp->t_state) {
1431         /*
1432          * If the state is SYN_RECEIVED:
1433          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1434          */
1435         case TCPS_SYN_RECEIVED:
1436                 if ((thflags & TH_ACK) &&
1437                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1438                      SEQ_GT(th->th_ack, tp->snd_max))) {
1439                                 rstreason = BANDLIM_RST_OPENPORT;
1440                                 goto dropwithreset;
1441                 }
1442                 break;
1443
1444         /*
1445          * If the state is SYN_SENT:
1446          *      if seg contains an ACK, but not for our SYN, drop the input.
1447          *      if seg contains a RST, then drop the connection.
1448          *      if seg does not contain SYN, then drop it.
1449          * Otherwise this is an acceptable SYN segment
1450          *      initialize tp->rcv_nxt and tp->irs
1451          *      if seg contains ack then advance tp->snd_una
1452          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1453          *      arrange for segment to be acked (eventually)
1454          *      continue processing rest of data/controls, beginning with URG
1455          */
1456         case TCPS_SYN_SENT:
1457                 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1458                         taop = &tao_noncached;
1459                         bzero(taop, sizeof *taop);
1460                 }
1461
1462                 if ((thflags & TH_ACK) &&
1463                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1464                      SEQ_GT(th->th_ack, tp->snd_max))) {
1465                         /*
1466                          * If we have a cached CCsent for the remote host,
1467                          * hence we haven't just crashed and restarted,
1468                          * do not send a RST.  This may be a retransmission
1469                          * from the other side after our earlier ACK was lost.
1470                          * Our new SYN, when it arrives, will serve as the
1471                          * needed ACK.
1472                          */
1473                         if (taop->tao_ccsent != 0)
1474                                 goto drop;
1475                         else {
1476                                 rstreason = BANDLIM_UNLIMITED;
1477                                 goto dropwithreset;
1478                         }
1479                 }
1480                 if (thflags & TH_RST) {
1481                         if (thflags & TH_ACK)
1482                                 tp = tcp_drop(tp, ECONNREFUSED);
1483                         goto drop;
1484                 }
1485                 if (!(thflags & TH_SYN))
1486                         goto drop;
1487                 tp->snd_wnd = th->th_win;       /* initial send window */
1488                 tp->cc_recv = to.to_cc;         /* foreign CC */
1489
1490                 tp->irs = th->th_seq;
1491                 tcp_rcvseqinit(tp);
1492                 if (thflags & TH_ACK) {
1493                         /*
1494                          * Our SYN was acked.  If segment contains CC.ECHO
1495                          * option, check it to make sure this segment really
1496                          * matches our SYN.  If not, just drop it as old
1497                          * duplicate, but send an RST if we're still playing
1498                          * by the old rules.  If no CC.ECHO option, make sure
1499                          * we don't get fooled into using T/TCP.
1500                          */
1501                         if (to.to_flags & TOF_CCECHO) {
1502                                 if (tp->cc_send != to.to_ccecho) {
1503                                         if (taop->tao_ccsent != 0)
1504                                                 goto drop;
1505                                         else {
1506                                                 rstreason = BANDLIM_UNLIMITED;
1507                                                 goto dropwithreset;
1508                                         }
1509                                 }
1510                         } else
1511                                 tp->t_flags &= ~TF_RCVD_CC;
1512                         tcpstat.tcps_connects++;
1513                         soisconnected(so);
1514                         /* Do window scaling on this connection? */
1515                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1516                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1517                                 tp->snd_scale = tp->requested_s_scale;
1518                                 tp->rcv_scale = tp->request_r_scale;
1519                         }
1520                         /* Segment is acceptable, update cache if undefined. */
1521                         if (taop->tao_ccsent == 0)
1522                                 taop->tao_ccsent = to.to_ccecho;
1523
1524                         tp->rcv_adv += tp->rcv_wnd;
1525                         tp->snd_una++;          /* SYN is acked */
1526                         tcp_callout_stop(tp, tp->tt_rexmt);
1527                         /*
1528                          * If there's data, delay ACK; if there's also a FIN
1529                          * ACKNOW will be turned on later.
1530                          */
1531                         if (DELAY_ACK(tp) && tlen != 0) {
1532                                 tcp_callout_reset(tp, tp->tt_delack,
1533                                     tcp_delacktime, tcp_timer_delack);
1534                         } else {
1535                                 tp->t_flags |= TF_ACKNOW;
1536                         }
1537                         /*
1538                          * Received <SYN,ACK> in SYN_SENT[*] state.
1539                          * Transitions:
1540                          *      SYN_SENT  --> ESTABLISHED
1541                          *      SYN_SENT* --> FIN_WAIT_1
1542                          */
1543                         tp->t_starttime = ticks;
1544                         if (tp->t_flags & TF_NEEDFIN) {
1545                                 tp->t_state = TCPS_FIN_WAIT_1;
1546                                 tp->t_flags &= ~TF_NEEDFIN;
1547                                 thflags &= ~TH_SYN;
1548                         } else {
1549                                 tp->t_state = TCPS_ESTABLISHED;
1550                                 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1551                                     tcp_timer_keep);
1552                         }
1553                 } else {
1554                         /*
1555                          * Received initial SYN in SYN-SENT[*] state =>
1556                          * simultaneous open.  If segment contains CC option
1557                          * and there is a cached CC, apply TAO test.
1558                          * If it succeeds, connection is * half-synchronized.
1559                          * Otherwise, do 3-way handshake:
1560                          *        SYN-SENT -> SYN-RECEIVED
1561                          *        SYN-SENT* -> SYN-RECEIVED*
1562                          * If there was no CC option, clear cached CC value.
1563                          */
1564                         tp->t_flags |= TF_ACKNOW;
1565                         tcp_callout_stop(tp, tp->tt_rexmt);
1566                         if (to.to_flags & TOF_CC) {
1567                                 if (taop->tao_cc != 0 &&
1568                                     CC_GT(to.to_cc, taop->tao_cc)) {
1569                                         /*
1570                                          * update cache and make transition:
1571                                          *        SYN-SENT -> ESTABLISHED*
1572                                          *        SYN-SENT* -> FIN-WAIT-1*
1573                                          */
1574                                         taop->tao_cc = to.to_cc;
1575                                         tp->t_starttime = ticks;
1576                                         if (tp->t_flags & TF_NEEDFIN) {
1577                                                 tp->t_state = TCPS_FIN_WAIT_1;
1578                                                 tp->t_flags &= ~TF_NEEDFIN;
1579                                         } else {
1580                                                 tp->t_state = TCPS_ESTABLISHED;
1581                                                 tcp_callout_reset(tp,
1582                                                     tp->tt_keep, tcp_keepidle,
1583                                                     tcp_timer_keep);
1584                                         }
1585                                         tp->t_flags |= TF_NEEDSYN;
1586                                 } else
1587                                         tp->t_state = TCPS_SYN_RECEIVED;
1588                         } else {
1589                                 /* CC.NEW or no option => invalidate cache */
1590                                 taop->tao_cc = 0;
1591                                 tp->t_state = TCPS_SYN_RECEIVED;
1592                         }
1593                 }
1594
1595 trimthenstep6:
1596                 /*
1597                  * Advance th->th_seq to correspond to first data byte.
1598                  * If data, trim to stay within window,
1599                  * dropping FIN if necessary.
1600                  */
1601                 th->th_seq++;
1602                 if (tlen > tp->rcv_wnd) {
1603                         todrop = tlen - tp->rcv_wnd;
1604                         m_adj(m, -todrop);
1605                         tlen = tp->rcv_wnd;
1606                         thflags &= ~TH_FIN;
1607                         tcpstat.tcps_rcvpackafterwin++;
1608                         tcpstat.tcps_rcvbyteafterwin += todrop;
1609                 }
1610                 tp->snd_wl1 = th->th_seq - 1;
1611                 tp->rcv_up = th->th_seq;
1612                 /*
1613                  * Client side of transaction: already sent SYN and data.
1614                  * If the remote host used T/TCP to validate the SYN,
1615                  * our data will be ACK'd; if so, enter normal data segment
1616                  * processing in the middle of step 5, ack processing.
1617                  * Otherwise, goto step 6.
1618                  */
1619                 if (thflags & TH_ACK)
1620                         goto process_ACK;
1621
1622                 goto step6;
1623
1624         /*
1625          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1626          *      if segment contains a SYN and CC [not CC.NEW] option:
1627          *              if state == TIME_WAIT and connection duration > MSL,
1628          *                  drop packet and send RST;
1629          *
1630          *              if SEG.CC > CCrecv then is new SYN, and can implicitly
1631          *                  ack the FIN (and data) in retransmission queue.
1632          *                  Complete close and delete TCPCB.  Then reprocess
1633          *                  segment, hoping to find new TCPCB in LISTEN state;
1634          *
1635          *              else must be old SYN; drop it.
1636          *      else do normal processing.
1637          */
1638         case TCPS_LAST_ACK:
1639         case TCPS_CLOSING:
1640         case TCPS_TIME_WAIT:
1641                 if ((thflags & TH_SYN) &&
1642                     (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1643                         if (tp->t_state == TCPS_TIME_WAIT &&
1644                                         (ticks - tp->t_starttime) > tcp_msl) {
1645                                 rstreason = BANDLIM_UNLIMITED;
1646                                 goto dropwithreset;
1647                         }
1648                         if (CC_GT(to.to_cc, tp->cc_recv)) {
1649                                 tp = tcp_close(tp);
1650                                 goto findpcb;
1651                         }
1652                         else
1653                                 goto drop;
1654                 }
1655                 break;  /* continue normal processing */
1656         }
1657
1658         /*
1659          * States other than LISTEN or SYN_SENT.
1660          * First check the RST flag and sequence number since reset segments
1661          * are exempt from the timestamp and connection count tests.  This
1662          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1663          * below which allowed reset segments in half the sequence space
1664          * to fall though and be processed (which gives forged reset
1665          * segments with a random sequence number a 50 percent chance of
1666          * killing a connection).
1667          * Then check timestamp, if present.
1668          * Then check the connection count, if present.
1669          * Then check that at least some bytes of segment are within
1670          * receive window.  If segment begins before rcv_nxt,
1671          * drop leading data (and SYN); if nothing left, just ack.
1672          *
1673          *
1674          * If the RST bit is set, check the sequence number to see
1675          * if this is a valid reset segment.
1676          * RFC 793 page 37:
1677          *   In all states except SYN-SENT, all reset (RST) segments
1678          *   are validated by checking their SEQ-fields.  A reset is
1679          *   valid if its sequence number is in the window.
1680          * Note: this does not take into account delayed ACKs, so
1681          *   we should test against last_ack_sent instead of rcv_nxt.
1682          *   The sequence number in the reset segment is normally an
1683          *   echo of our outgoing acknowledgement numbers, but some hosts
1684          *   send a reset with the sequence number at the rightmost edge
1685          *   of our receive window, and we have to handle this case.
1686          * If we have multiple segments in flight, the intial reset
1687          * segment sequence numbers will be to the left of last_ack_sent,
1688          * but they will eventually catch up.
1689          * In any case, it never made sense to trim reset segments to
1690          * fit the receive window since RFC 1122 says:
1691          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1692          *
1693          *    A TCP SHOULD allow a received RST segment to include data.
1694          *
1695          *    DISCUSSION
1696          *         It has been suggested that a RST segment could contain
1697          *         ASCII text that encoded and explained the cause of the
1698          *         RST.  No standard has yet been established for such
1699          *         data.
1700          *
1701          * If the reset segment passes the sequence number test examine
1702          * the state:
1703          *    SYN_RECEIVED STATE:
1704          *      If passive open, return to LISTEN state.
1705          *      If active open, inform user that connection was refused.
1706          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1707          *      Inform user that connection was reset, and close tcb.
1708          *    CLOSING, LAST_ACK STATES:
1709          *      Close the tcb.
1710          *    TIME_WAIT STATE:
1711          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1712          *      RFC 1337.
1713          */
1714         if (thflags & TH_RST) {
1715                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1716                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1717                         switch (tp->t_state) {
1718
1719                         case TCPS_SYN_RECEIVED:
1720                                 so->so_error = ECONNREFUSED;
1721                                 goto close;
1722
1723                         case TCPS_ESTABLISHED:
1724                         case TCPS_FIN_WAIT_1:
1725                         case TCPS_FIN_WAIT_2:
1726                         case TCPS_CLOSE_WAIT:
1727                                 so->so_error = ECONNRESET;
1728                         close:
1729                                 tp->t_state = TCPS_CLOSED;
1730                                 tcpstat.tcps_drops++;
1731                                 tp = tcp_close(tp);
1732                                 break;
1733
1734                         case TCPS_CLOSING:
1735                         case TCPS_LAST_ACK:
1736                                 tp = tcp_close(tp);
1737                                 break;
1738
1739                         case TCPS_TIME_WAIT:
1740                                 break;
1741                         }
1742                 }
1743                 goto drop;
1744         }
1745
1746         /*
1747          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1748          * and it's less than ts_recent, drop it.
1749          */
1750         if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1751             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1752
1753                 /* Check to see if ts_recent is over 24 days old.  */
1754                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1755                         /*
1756                          * Invalidate ts_recent.  If this segment updates
1757                          * ts_recent, the age will be reset later and ts_recent
1758                          * will get a valid value.  If it does not, setting
1759                          * ts_recent to zero will at least satisfy the
1760                          * requirement that zero be placed in the timestamp
1761                          * echo reply when ts_recent isn't valid.  The
1762                          * age isn't reset until we get a valid ts_recent
1763                          * because we don't want out-of-order segments to be
1764                          * dropped when ts_recent is old.
1765                          */
1766                         tp->ts_recent = 0;
1767                 } else {
1768                         tcpstat.tcps_rcvduppack++;
1769                         tcpstat.tcps_rcvdupbyte += tlen;
1770                         tcpstat.tcps_pawsdrop++;
1771                         if (tlen)
1772                                 goto dropafterack;
1773                         goto drop;
1774                 }
1775         }
1776
1777         /*
1778          * T/TCP mechanism
1779          *   If T/TCP was negotiated and the segment doesn't have CC,
1780          *   or if its CC is wrong then drop the segment.
1781          *   RST segments do not have to comply with this.
1782          */
1783         if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1784             (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1785                 goto dropafterack;
1786
1787         /*
1788          * In the SYN-RECEIVED state, validate that the packet belongs to
1789          * this connection before trimming the data to fit the receive
1790          * window.  Check the sequence number versus IRS since we know
1791          * the sequence numbers haven't wrapped.  This is a partial fix
1792          * for the "LAND" DoS attack.
1793          */
1794         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1795                 rstreason = BANDLIM_RST_OPENPORT;
1796                 goto dropwithreset;
1797         }
1798
1799         todrop = tp->rcv_nxt - th->th_seq;
1800         if (todrop > 0) {
1801                 if (TCP_DO_SACK(tp)) {
1802                         /* Report duplicate segment at head of packet. */
1803                         tp->reportblk.rblk_start = th->th_seq;
1804                         tp->reportblk.rblk_end = th->th_seq + tlen;
1805                         if (thflags & TH_FIN)
1806                                 ++tp->reportblk.rblk_end;
1807                         if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1808                                 tp->reportblk.rblk_end = tp->rcv_nxt;
1809                         tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1810                 }
1811                 if (thflags & TH_SYN) {
1812                         thflags &= ~TH_SYN;
1813                         th->th_seq++;
1814                         if (th->th_urp > 1)
1815                                 th->th_urp--;
1816                         else
1817                                 thflags &= ~TH_URG;
1818                         todrop--;
1819                 }
1820                 /*
1821                  * Following if statement from Stevens, vol. 2, p. 960.
1822                  */
1823                 if (todrop > tlen ||
1824                     (todrop == tlen && !(thflags & TH_FIN))) {
1825                         /*
1826                          * Any valid FIN must be to the left of the window.
1827                          * At this point the FIN must be a duplicate or out
1828                          * of sequence; drop it.
1829                          */
1830                         thflags &= ~TH_FIN;
1831
1832                         /*
1833                          * Send an ACK to resynchronize and drop any data.
1834                          * But keep on processing for RST or ACK.
1835                          */
1836                         tp->t_flags |= TF_ACKNOW;
1837                         todrop = tlen;
1838                         tcpstat.tcps_rcvduppack++;
1839                         tcpstat.tcps_rcvdupbyte += todrop;
1840                 } else {
1841                         tcpstat.tcps_rcvpartduppack++;
1842                         tcpstat.tcps_rcvpartdupbyte += todrop;
1843                 }
1844                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1845                 th->th_seq += todrop;
1846                 tlen -= todrop;
1847                 if (th->th_urp > todrop)
1848                         th->th_urp -= todrop;
1849                 else {
1850                         thflags &= ~TH_URG;
1851                         th->th_urp = 0;
1852                 }
1853         }
1854
1855         /*
1856          * If new data are received on a connection after the
1857          * user processes are gone, then RST the other end.
1858          */
1859         if ((so->so_state & SS_NOFDREF) &&
1860             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1861                 tp = tcp_close(tp);
1862                 tcpstat.tcps_rcvafterclose++;
1863                 rstreason = BANDLIM_UNLIMITED;
1864                 goto dropwithreset;
1865         }
1866
1867         /*
1868          * If segment ends after window, drop trailing data
1869          * (and PUSH and FIN); if nothing left, just ACK.
1870          */
1871         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1872         if (todrop > 0) {
1873                 tcpstat.tcps_rcvpackafterwin++;
1874                 if (todrop >= tlen) {
1875                         tcpstat.tcps_rcvbyteafterwin += tlen;
1876                         /*
1877                          * If a new connection request is received
1878                          * while in TIME_WAIT, drop the old connection
1879                          * and start over if the sequence numbers
1880                          * are above the previous ones.
1881                          */
1882                         if (thflags & TH_SYN &&
1883                             tp->t_state == TCPS_TIME_WAIT &&
1884                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1885                                 tp = tcp_close(tp);
1886                                 goto findpcb;
1887                         }
1888                         /*
1889                          * If window is closed can only take segments at
1890                          * window edge, and have to drop data and PUSH from
1891                          * incoming segments.  Continue processing, but
1892                          * remember to ack.  Otherwise, drop segment
1893                          * and ack.
1894                          */
1895                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1896                                 tp->t_flags |= TF_ACKNOW;
1897                                 tcpstat.tcps_rcvwinprobe++;
1898                         } else
1899                                 goto dropafterack;
1900                 } else
1901                         tcpstat.tcps_rcvbyteafterwin += todrop;
1902                 m_adj(m, -todrop);
1903                 tlen -= todrop;
1904                 thflags &= ~(TH_PUSH | TH_FIN);
1905         }
1906
1907         /*
1908          * If last ACK falls within this segment's sequence numbers,
1909          * record its timestamp.
1910          * NOTE:
1911          * 1) That the test incorporates suggestions from the latest
1912          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1913          * 2) That updating only on newer timestamps interferes with
1914          *    our earlier PAWS tests, so this check should be solely
1915          *    predicated on the sequence space of this segment.
1916          * 3) That we modify the segment boundary check to be
1917          *        Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1918          *    instead of RFC1323's
1919          *        Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1920          *    This modified check allows us to overcome RFC1323's
1921          *    limitations as described in Stevens TCP/IP Illustrated
1922          *    Vol. 2 p.869. In such cases, we can still calculate the
1923          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1924          */
1925         if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1926             SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1927                                         + ((thflags & TH_SYN) != 0)
1928                                         + ((thflags & TH_FIN) != 0)))) {
1929                 tp->ts_recent_age = ticks;
1930                 tp->ts_recent = to.to_tsval;
1931         }
1932
1933         /*
1934          * If a SYN is in the window, then this is an
1935          * error and we send an RST and drop the connection.
1936          */
1937         if (thflags & TH_SYN) {
1938                 tp = tcp_drop(tp, ECONNRESET);
1939                 rstreason = BANDLIM_UNLIMITED;
1940                 goto dropwithreset;
1941         }
1942
1943         /*
1944          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1945          * flag is on (half-synchronized state), then queue data for
1946          * later processing; else drop segment and return.
1947          */
1948         if (!(thflags & TH_ACK)) {
1949                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1950                     (tp->t_flags & TF_NEEDSYN))
1951                         goto step6;
1952                 else
1953                         goto drop;
1954         }
1955
1956         /*
1957          * Ack processing.
1958          */
1959         switch (tp->t_state) {
1960         /*
1961          * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1962          * ESTABLISHED state and continue processing.
1963          * The ACK was checked above.
1964          */
1965         case TCPS_SYN_RECEIVED:
1966
1967                 tcpstat.tcps_connects++;
1968                 soisconnected(so);
1969                 /* Do window scaling? */
1970                 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1971                     (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1972                         tp->snd_scale = tp->requested_s_scale;
1973                         tp->rcv_scale = tp->request_r_scale;
1974                 }
1975                 /*
1976                  * Upon successful completion of 3-way handshake,
1977                  * update cache.CC if it was undefined, pass any queued
1978                  * data to the user, and advance state appropriately.
1979                  */
1980                 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1981                     taop->tao_cc == 0)
1982                         taop->tao_cc = tp->cc_recv;
1983
1984                 /*
1985                  * Make transitions:
1986                  *      SYN-RECEIVED  -> ESTABLISHED
1987                  *      SYN-RECEIVED* -> FIN-WAIT-1
1988                  */
1989                 tp->t_starttime = ticks;
1990                 if (tp->t_flags & TF_NEEDFIN) {
1991                         tp->t_state = TCPS_FIN_WAIT_1;
1992                         tp->t_flags &= ~TF_NEEDFIN;
1993                 } else {
1994                         tp->t_state = TCPS_ESTABLISHED;
1995                         tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1996                             tcp_timer_keep);
1997                 }
1998                 /*
1999                  * If segment contains data or ACK, will call tcp_reass()
2000                  * later; if not, do so now to pass queued data to user.
2001                  */
2002                 if (tlen == 0 && !(thflags & TH_FIN))
2003                         tcp_reass(tp, NULL, NULL, NULL);
2004                 /* fall into ... */
2005
2006         /*
2007          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2008          * ACKs.  If the ack is in the range
2009          *      tp->snd_una < th->th_ack <= tp->snd_max
2010          * then advance tp->snd_una to th->th_ack and drop
2011          * data from the retransmission queue.  If this ACK reflects
2012          * more up to date window information we update our window information.
2013          */
2014         case TCPS_ESTABLISHED:
2015         case TCPS_FIN_WAIT_1:
2016         case TCPS_FIN_WAIT_2:
2017         case TCPS_CLOSE_WAIT:
2018         case TCPS_CLOSING:
2019         case TCPS_LAST_ACK:
2020         case TCPS_TIME_WAIT:
2021
2022                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2023                         if (TCP_DO_SACK(tp))
2024                                 tcp_sack_update_scoreboard(tp, &to);
2025                         if (tlen != 0 || tiwin != tp->snd_wnd) {
2026                                 tp->t_dupacks = 0;
2027                                 break;
2028                         }
2029                         tcpstat.tcps_rcvdupack++;
2030                         if (!tcp_callout_active(tp, tp->tt_rexmt) ||
2031                             th->th_ack != tp->snd_una) {
2032                                 tp->t_dupacks = 0;
2033                                 break;
2034                         }
2035                         /*
2036                          * We have outstanding data (other than
2037                          * a window probe), this is a completely
2038                          * duplicate ack (ie, window info didn't
2039                          * change), the ack is the biggest we've
2040                          * seen and we've seen exactly our rexmt
2041                          * threshhold of them, so assume a packet
2042                          * has been dropped and retransmit it.
2043                          * Kludge snd_nxt & the congestion
2044                          * window so we send only this one
2045                          * packet.
2046                          */
2047                         if (IN_FASTRECOVERY(tp)) {
2048                                 if (TCP_DO_SACK(tp)) {
2049                                         /* No artifical cwnd inflation. */
2050                                         tcp_sack_rexmt(tp, th);
2051                                 } else {
2052                                         /*
2053                                          * Dup acks mean that packets
2054                                          * have left the network
2055                                          * (they're now cached at the
2056                                          * receiver) so bump cwnd by
2057                                          * the amount in the receiver
2058                                          * to keep a constant cwnd
2059                                          * packets in the network.
2060                                          */
2061                                         tp->snd_cwnd += tp->t_maxseg;
2062                                         tcp_output(tp);
2063                                 }
2064                         } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2065                                 tp->t_dupacks = 0;
2066                                 break;
2067                         } else if (++tp->t_dupacks == tcprexmtthresh) {
2068                                 tcp_seq old_snd_nxt;
2069                                 u_int win;
2070
2071 fastretransmit:
2072                                 if (tcp_do_eifel_detect &&
2073                                     (tp->t_flags & TF_RCVD_TSTMP)) {
2074                                         tcp_save_congestion_state(tp);
2075                                         tp->t_flags |= TF_FASTREXMT;
2076                                 }
2077                                 /*
2078                                  * We know we're losing at the current
2079                                  * window size, so do congestion avoidance:
2080                                  * set ssthresh to half the current window
2081                                  * and pull our congestion window back to the
2082                                  * new ssthresh.
2083                                  */
2084                                 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
2085                                     tp->t_maxseg;
2086                                 if (win < 2)
2087                                         win = 2;
2088                                 tp->snd_ssthresh = win * tp->t_maxseg;
2089                                 ENTER_FASTRECOVERY(tp);
2090                                 tp->snd_recover = tp->snd_max;
2091                                 tcp_callout_stop(tp, tp->tt_rexmt);
2092                                 tp->t_rtttime = 0;
2093                                 old_snd_nxt = tp->snd_nxt;
2094                                 tp->snd_nxt = th->th_ack;
2095                                 tp->snd_cwnd = tp->t_maxseg;
2096                                 tcp_output(tp);
2097                                 ++tcpstat.tcps_sndfastrexmit;
2098                                 tp->snd_cwnd = tp->snd_ssthresh;
2099                                 tp->rexmt_high = tp->snd_nxt;
2100                                 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
2101                                         tp->snd_nxt = old_snd_nxt;
2102                                 KASSERT(tp->snd_limited <= 2,
2103                                     ("tp->snd_limited too big"));
2104                                 if (TCP_DO_SACK(tp))
2105                                         tcp_sack_rexmt(tp, th);
2106                                 else
2107                                         tp->snd_cwnd += tp->t_maxseg *
2108                                             (tp->t_dupacks - tp->snd_limited);
2109                         } else if (tcp_do_limitedtransmit) {
2110                                 u_long oldcwnd = tp->snd_cwnd;
2111                                 tcp_seq oldsndmax = tp->snd_max;
2112                                 tcp_seq oldsndnxt = tp->snd_nxt;
2113                                 /* outstanding data */
2114                                 uint32_t ownd = tp->snd_max - tp->snd_una;
2115                                 u_int sent;
2116
2117 #define iceildiv(n, d)          (((n)+(d)-1) / (d))
2118
2119                                 KASSERT(tp->t_dupacks == 1 ||
2120                                         tp->t_dupacks == 2,
2121                                     ("dupacks not 1 or 2"));
2122                                 if (tp->t_dupacks == 1)
2123                                         tp->snd_limited = 0;
2124                                 tp->snd_nxt = tp->snd_max;
2125                                 tp->snd_cwnd = ownd +
2126                                     (tp->t_dupacks - tp->snd_limited) *
2127                                     tp->t_maxseg;
2128                                 tcp_output(tp);
2129
2130                                 /*
2131                                  * Other acks may have been processed,
2132                                  * snd_nxt cannot be reset to a value less
2133                                  * then snd_una.
2134                                  */
2135                                 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2136                                     if (SEQ_GT(oldsndnxt, tp->snd_una))
2137                                         tp->snd_nxt = oldsndnxt;
2138                                     else
2139                                         tp->snd_nxt = tp->snd_una;
2140                                 }
2141                                 tp->snd_cwnd = oldcwnd;
2142                                 sent = tp->snd_max - oldsndmax;
2143                                 if (sent > tp->t_maxseg) {
2144                                         KASSERT((tp->t_dupacks == 2 &&
2145                                                  tp->snd_limited == 0) ||
2146                                                 (sent == tp->t_maxseg + 1 &&
2147                                                  tp->t_flags & TF_SENTFIN),
2148                                             ("sent too much"));
2149                                         KASSERT(sent <= tp->t_maxseg * 2,
2150                                             ("sent too many segments"));
2151                                         tp->snd_limited = 2;
2152                                         tcpstat.tcps_sndlimited += 2;
2153                                 } else if (sent > 0) {
2154                                         ++tp->snd_limited;
2155                                         ++tcpstat.tcps_sndlimited;
2156                                 } else if (tcp_do_early_retransmit &&
2157                                     (tcp_do_eifel_detect &&
2158                                      (tp->t_flags & TF_RCVD_TSTMP)) &&
2159                                     ownd < 4 * tp->t_maxseg &&
2160                                     tp->t_dupacks + 1 >=
2161                                       iceildiv(ownd, tp->t_maxseg) &&
2162                                     (!TCP_DO_SACK(tp) ||
2163                                      ownd <= tp->t_maxseg ||
2164                                      tcp_sack_has_sacked(&tp->scb,
2165                                                         ownd - tp->t_maxseg))) {
2166                                         ++tcpstat.tcps_sndearlyrexmit;
2167                                         tp->t_flags |= TF_EARLYREXMT;
2168                                         goto fastretransmit;
2169                                 }
2170                         }
2171                         goto drop;
2172                 }
2173
2174                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2175                 tp->t_dupacks = 0;
2176                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2177                         /*
2178                          * Detected optimistic ACK attack.
2179                          * Force slow-start to de-synchronize attack.
2180                          */
2181                         tp->snd_cwnd = tp->t_maxseg;
2182                         tp->snd_wacked = 0;
2183
2184                         tcpstat.tcps_rcvacktoomuch++;
2185                         goto dropafterack;
2186                 }
2187                 /*
2188                  * If we reach this point, ACK is not a duplicate,
2189                  *     i.e., it ACKs something we sent.
2190                  */
2191                 if (tp->t_flags & TF_NEEDSYN) {
2192                         /*
2193                          * T/TCP: Connection was half-synchronized, and our
2194                          * SYN has been ACK'd (so connection is now fully
2195                          * synchronized).  Go to non-starred state,
2196                          * increment snd_una for ACK of SYN, and check if
2197                          * we can do window scaling.
2198                          */
2199                         tp->t_flags &= ~TF_NEEDSYN;
2200                         tp->snd_una++;
2201                         /* Do window scaling? */
2202                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2203                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2204                                 tp->snd_scale = tp->requested_s_scale;
2205                                 tp->rcv_scale = tp->request_r_scale;
2206                         }
2207                 }
2208
2209 process_ACK:
2210                 acked = th->th_ack - tp->snd_una;
2211                 tcpstat.tcps_rcvackpack++;
2212                 tcpstat.tcps_rcvackbyte += acked;
2213
2214                 if (tcp_do_eifel_detect && acked > 0 &&
2215                     (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2216                     (tp->t_flags & TF_FIRSTACCACK)) {
2217                         /* Eifel detection applicable. */
2218                         if (to.to_tsecr < tp->t_rexmtTS) {
2219                                 ++tcpstat.tcps_eifeldetected;
2220                                 tcp_revert_congestion_state(tp);
2221                                 if (tp->t_rxtshift == 1 &&
2222                                     ticks >= tp->t_badrxtwin)
2223                                         ++tcpstat.tcps_rttcantdetect;
2224                         }
2225                 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2226                         /*
2227                          * If we just performed our first retransmit,
2228                          * and the ACK arrives within our recovery window,
2229                          * then it was a mistake to do the retransmit
2230                          * in the first place.  Recover our original cwnd
2231                          * and ssthresh, and proceed to transmit where we
2232                          * left off.
2233                          */
2234                         tcp_revert_congestion_state(tp);
2235                         ++tcpstat.tcps_rttdetected;
2236                 }
2237
2238                 /*
2239                  * If we have a timestamp reply, update smoothed
2240                  * round trip time.  If no timestamp is present but
2241                  * transmit timer is running and timed sequence
2242                  * number was acked, update smoothed round trip time.
2243                  * Since we now have an rtt measurement, cancel the
2244                  * timer backoff (cf., Phil Karn's retransmit alg.).
2245                  * Recompute the initial retransmit timer.
2246                  *
2247                  * Some machines (certain windows boxes) send broken
2248                  * timestamp replies during the SYN+ACK phase, ignore
2249                  * timestamps of 0.
2250                  */
2251                 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2252                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2253                 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2254                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2255                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2256
2257                 /*
2258                  * If no data (only SYN) was ACK'd,
2259                  *    skip rest of ACK processing.
2260                  */
2261                 if (acked == 0)
2262                         goto step6;
2263
2264                 /* Stop looking for an acceptable ACK since one was received. */
2265                 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2266
2267                 if (acked > so->so_snd.ssb_cc) {
2268                         tp->snd_wnd -= so->so_snd.ssb_cc;
2269                         sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2270                         ourfinisacked = TRUE;
2271                 } else {
2272                         sbdrop(&so->so_snd.sb, acked);
2273                         tp->snd_wnd -= acked;
2274                         ourfinisacked = FALSE;
2275                 }
2276                 sowwakeup(so);
2277
2278                 /*
2279                  * Update window information.
2280                  * Don't look at window if no ACK:
2281                  * TAC's send garbage on first SYN.
2282                  */
2283                 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2284                     (tp->snd_wl1 == th->th_seq &&
2285                      (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2286                       (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2287                         /* keep track of pure window updates */
2288                         if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2289                             tiwin > tp->snd_wnd)
2290                                 tcpstat.tcps_rcvwinupd++;
2291                         tp->snd_wnd = tiwin;
2292                         tp->snd_wl1 = th->th_seq;
2293                         tp->snd_wl2 = th->th_ack;
2294                         if (tp->snd_wnd > tp->max_sndwnd)
2295                                 tp->max_sndwnd = tp->snd_wnd;
2296                         needoutput = TRUE;
2297                 }
2298
2299                 tp->snd_una = th->th_ack;
2300                 if (TCP_DO_SACK(tp))
2301                         tcp_sack_update_scoreboard(tp, &to);
2302                 if (IN_FASTRECOVERY(tp)) {
2303                         if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2304                                 EXIT_FASTRECOVERY(tp);
2305                                 needoutput = TRUE;
2306                                 /*
2307                                  * If the congestion window was inflated
2308                                  * to account for the other side's
2309                                  * cached packets, retract it.
2310                                  */
2311                                 if (!TCP_DO_SACK(tp))
2312                                         tp->snd_cwnd = tp->snd_ssthresh;
2313
2314                                 /*
2315                                  * Window inflation should have left us
2316                                  * with approximately snd_ssthresh outstanding
2317                                  * data.  But, in case we would be inclined
2318                                  * to send a burst, better do it using
2319                                  * slow start.
2320                                  */
2321                                 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2322                                            tp->snd_max + 2 * tp->t_maxseg))
2323                                         tp->snd_cwnd =
2324                                             (tp->snd_max - tp->snd_una) +
2325                                             2 * tp->t_maxseg;
2326
2327                                 tp->snd_wacked = 0;
2328                         } else {
2329                                 if (TCP_DO_SACK(tp)) {
2330                                         tp->snd_max_rexmt = tp->snd_max;
2331                                         tcp_sack_rexmt(tp, th);
2332                                 } else {
2333                                         tcp_newreno_partial_ack(tp, th, acked);
2334                                 }
2335                                 needoutput = FALSE;
2336                         }
2337                 } else {
2338                         /*
2339                          * Open the congestion window.  When in slow-start,
2340                          * open exponentially: maxseg per packet.  Otherwise,
2341                          * open linearly: maxseg per window.
2342                          */
2343                         if (tp->snd_cwnd <= tp->snd_ssthresh) {
2344                                 u_int abc_sslimit =
2345                                     (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2346                                      tp->t_maxseg : 2 * tp->t_maxseg);
2347
2348                                 /* slow-start */
2349                                 tp->snd_cwnd += tcp_do_abc ?
2350                                     min(acked, abc_sslimit) : tp->t_maxseg;
2351                         } else {
2352                                 /* linear increase */
2353                                 tp->snd_wacked += tcp_do_abc ? acked :
2354                                     tp->t_maxseg;
2355                                 if (tp->snd_wacked >= tp->snd_cwnd) {
2356                                         tp->snd_wacked -= tp->snd_cwnd;
2357                                         tp->snd_cwnd += tp->t_maxseg;
2358                                 }
2359                         }
2360                         tp->snd_cwnd = min(tp->snd_cwnd,
2361                                            TCP_MAXWIN << tp->snd_scale);
2362                         tp->snd_recover = th->th_ack - 1;
2363                 }
2364                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2365                         tp->snd_nxt = tp->snd_una;
2366
2367                 /*
2368                  * If all outstanding data is acked, stop retransmit
2369                  * timer and remember to restart (more output or persist).
2370                  * If there is more data to be acked, restart retransmit
2371                  * timer, using current (possibly backed-off) value.
2372                  */
2373                 if (th->th_ack == tp->snd_max) {
2374                         tcp_callout_stop(tp, tp->tt_rexmt);
2375                         needoutput = TRUE;
2376                 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2377                         tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2378                             tcp_timer_rexmt);
2379                 }
2380
2381                 switch (tp->t_state) {
2382                 /*
2383                  * In FIN_WAIT_1 STATE in addition to the processing
2384                  * for the ESTABLISHED state if our FIN is now acknowledged
2385                  * then enter FIN_WAIT_2.
2386                  */
2387                 case TCPS_FIN_WAIT_1:
2388                         if (ourfinisacked) {
2389                                 /*
2390                                  * If we can't receive any more
2391                                  * data, then closing user can proceed.
2392                                  * Starting the timer is contrary to the
2393                                  * specification, but if we don't get a FIN
2394                                  * we'll hang forever.
2395                                  */
2396                                 if (so->so_state & SS_CANTRCVMORE) {
2397                                         soisdisconnected(so);
2398                                         tcp_callout_reset(tp, tp->tt_2msl,
2399                                             tcp_maxidle, tcp_timer_2msl);
2400                                 }
2401                                 tp->t_state = TCPS_FIN_WAIT_2;
2402                         }
2403                         break;
2404
2405                 /*
2406                  * In CLOSING STATE in addition to the processing for
2407                  * the ESTABLISHED state if the ACK acknowledges our FIN
2408                  * then enter the TIME-WAIT state, otherwise ignore
2409                  * the segment.
2410                  */
2411                 case TCPS_CLOSING:
2412                         if (ourfinisacked) {
2413                                 tp->t_state = TCPS_TIME_WAIT;
2414                                 tcp_canceltimers(tp);
2415                                 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2416                                 if (tp->cc_recv != 0 &&
2417                                     (ticks - tp->t_starttime) < tcp_msl) {
2418                                         tcp_callout_reset(tp, tp->tt_2msl,
2419                                             tp->t_rxtcur * TCPTV_TWTRUNC,
2420                                             tcp_timer_2msl);
2421                                 } else {
2422                                         tcp_callout_reset(tp, tp->tt_2msl,
2423                                             2 * tcp_msl, tcp_timer_2msl);
2424                                 }
2425                                 soisdisconnected(so);
2426                         }
2427                         break;
2428
2429                 /*
2430                  * In LAST_ACK, we may still be waiting for data to drain
2431                  * and/or to be acked, as well as for the ack of our FIN.
2432                  * If our FIN is now acknowledged, delete the TCB,
2433                  * enter the closed state and return.
2434                  */
2435                 case TCPS_LAST_ACK:
2436                         if (ourfinisacked) {
2437                                 tp = tcp_close(tp);
2438                                 goto drop;
2439                         }
2440                         break;
2441
2442                 /*
2443                  * In TIME_WAIT state the only thing that should arrive
2444                  * is a retransmission of the remote FIN.  Acknowledge
2445                  * it and restart the finack timer.
2446                  */
2447                 case TCPS_TIME_WAIT:
2448                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2449                             tcp_timer_2msl);
2450                         goto dropafterack;
2451                 }
2452         }
2453
2454 step6:
2455         /*
2456          * Update window information.
2457          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2458          */
2459         if ((thflags & TH_ACK) &&
2460             acceptable_window_update(tp, th, tiwin)) {
2461                 /* keep track of pure window updates */
2462                 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2463                     tiwin > tp->snd_wnd)
2464                         tcpstat.tcps_rcvwinupd++;
2465                 tp->snd_wnd = tiwin;
2466                 tp->snd_wl1 = th->th_seq;
2467                 tp->snd_wl2 = th->th_ack;
2468                 if (tp->snd_wnd > tp->max_sndwnd)
2469                         tp->max_sndwnd = tp->snd_wnd;
2470                 needoutput = TRUE;
2471         }
2472
2473         /*
2474          * Process segments with URG.
2475          */
2476         if ((thflags & TH_URG) && th->th_urp &&
2477             !TCPS_HAVERCVDFIN(tp->t_state)) {
2478                 /*
2479                  * This is a kludge, but if we receive and accept
2480                  * random urgent pointers, we'll crash in
2481                  * soreceive.  It's hard to imagine someone
2482                  * actually wanting to send this much urgent data.
2483                  */
2484                 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2485                         th->th_urp = 0;                 /* XXX */
2486                         thflags &= ~TH_URG;             /* XXX */
2487                         goto dodata;                    /* XXX */
2488                 }
2489                 /*
2490                  * If this segment advances the known urgent pointer,
2491                  * then mark the data stream.  This should not happen
2492                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2493                  * a FIN has been received from the remote side.
2494                  * In these states we ignore the URG.
2495                  *
2496                  * According to RFC961 (Assigned Protocols),
2497                  * the urgent pointer points to the last octet
2498                  * of urgent data.  We continue, however,
2499                  * to consider it to indicate the first octet
2500                  * of data past the urgent section as the original
2501                  * spec states (in one of two places).
2502                  */
2503                 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2504                         tp->rcv_up = th->th_seq + th->th_urp;
2505                         so->so_oobmark = so->so_rcv.ssb_cc +
2506                             (tp->rcv_up - tp->rcv_nxt) - 1;
2507                         if (so->so_oobmark == 0)
2508                                 so->so_state |= SS_RCVATMARK;
2509                         sohasoutofband(so);
2510                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2511                 }
2512                 /*
2513                  * Remove out of band data so doesn't get presented to user.
2514                  * This can happen independent of advancing the URG pointer,
2515                  * but if two URG's are pending at once, some out-of-band
2516                  * data may creep in... ick.
2517                  */
2518                 if (th->th_urp <= (u_long)tlen &&
2519                     !(so->so_options & SO_OOBINLINE)) {
2520                         /* hdr drop is delayed */
2521                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2522                 }
2523         } else {
2524                 /*
2525                  * If no out of band data is expected,
2526                  * pull receive urgent pointer along
2527                  * with the receive window.
2528                  */
2529                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2530                         tp->rcv_up = tp->rcv_nxt;
2531         }
2532
2533 dodata:                                                 /* XXX */
2534         /*
2535          * Process the segment text, merging it into the TCP sequencing queue,
2536          * and arranging for acknowledgment of receipt if necessary.
2537          * This process logically involves adjusting tp->rcv_wnd as data
2538          * is presented to the user (this happens in tcp_usrreq.c,
2539          * case PRU_RCVD).  If a FIN has already been received on this
2540          * connection then we just ignore the text.
2541          */
2542         if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2543                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2544                 /*
2545                  * Insert segment which includes th into TCP reassembly queue
2546                  * with control block tp.  Set thflags to whether reassembly now
2547                  * includes a segment with FIN.  This handles the common case
2548                  * inline (segment is the next to be received on an established
2549                  * connection, and the queue is empty), avoiding linkage into
2550                  * and removal from the queue and repetition of various
2551                  * conversions.
2552                  * Set DELACK for segments received in order, but ack
2553                  * immediately when segments are out of order (so
2554                  * fast retransmit can work).
2555                  */
2556                 if (th->th_seq == tp->rcv_nxt &&
2557                     LIST_EMPTY(&tp->t_segq) &&
2558                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2559                         if (DELAY_ACK(tp)) {
2560                                 tcp_callout_reset(tp, tp->tt_delack,
2561                                     tcp_delacktime, tcp_timer_delack);
2562                         } else {
2563                                 tp->t_flags |= TF_ACKNOW;
2564                         }
2565                         tp->rcv_nxt += tlen;
2566                         thflags = th->th_flags & TH_FIN;
2567                         tcpstat.tcps_rcvpack++;
2568                         tcpstat.tcps_rcvbyte += tlen;
2569                         ND6_HINT(tp);
2570                         if (so->so_state & SS_CANTRCVMORE)
2571                                 m_freem(m);
2572                         else
2573                                 ssb_appendstream(&so->so_rcv, m);
2574                         sorwakeup(so);
2575                 } else {
2576                         if (!(tp->t_flags & TF_DUPSEG)) {
2577                                 /* Initialize SACK report block. */
2578                                 tp->reportblk.rblk_start = th->th_seq;
2579                                 tp->reportblk.rblk_end = th->th_seq + tlen +
2580                                     ((thflags & TH_FIN) != 0);
2581                         }
2582                         thflags = tcp_reass(tp, th, &tlen, m);
2583                         tp->t_flags |= TF_ACKNOW;
2584                 }
2585
2586                 /*
2587                  * Note the amount of data that peer has sent into
2588                  * our window, in order to estimate the sender's
2589                  * buffer size.
2590                  */
2591                 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2592         } else {
2593                 m_freem(m);
2594                 thflags &= ~TH_FIN;
2595         }
2596
2597         /*
2598          * If FIN is received ACK the FIN and let the user know
2599          * that the connection is closing.
2600          */
2601         if (thflags & TH_FIN) {
2602                 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2603                         socantrcvmore(so);
2604                         /*
2605                          * If connection is half-synchronized
2606                          * (ie NEEDSYN flag on) then delay ACK,
2607                          * so it may be piggybacked when SYN is sent.
2608                          * Otherwise, since we received a FIN then no
2609                          * more input can be expected, send ACK now.
2610                          */
2611                         if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2612                                 tcp_callout_reset(tp, tp->tt_delack,
2613                                     tcp_delacktime, tcp_timer_delack);
2614                         } else {
2615                                 tp->t_flags |= TF_ACKNOW;
2616                         }
2617                         tp->rcv_nxt++;
2618                 }
2619
2620                 switch (tp->t_state) {
2621                 /*
2622                  * In SYN_RECEIVED and ESTABLISHED STATES
2623                  * enter the CLOSE_WAIT state.
2624                  */
2625                 case TCPS_SYN_RECEIVED:
2626                         tp->t_starttime = ticks;
2627                         /*FALLTHROUGH*/
2628                 case TCPS_ESTABLISHED:
2629                         tp->t_state = TCPS_CLOSE_WAIT;
2630                         break;
2631
2632                 /*
2633                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2634                  * enter the CLOSING state.
2635                  */
2636                 case TCPS_FIN_WAIT_1:
2637                         tp->t_state = TCPS_CLOSING;
2638                         break;
2639
2640                 /*
2641                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2642                  * starting the time-wait timer, turning off the other
2643                  * standard timers.
2644                  */
2645                 case TCPS_FIN_WAIT_2:
2646                         tp->t_state = TCPS_TIME_WAIT;
2647                         tcp_canceltimers(tp);
2648                         /* Shorten TIME_WAIT [RFC-1644, p.28] */
2649                         if (tp->cc_recv != 0 &&
2650                             (ticks - tp->t_starttime) < tcp_msl) {
2651                                 tcp_callout_reset(tp, tp->tt_2msl,
2652                                     tp->t_rxtcur * TCPTV_TWTRUNC,
2653                                     tcp_timer_2msl);
2654                                 /* For transaction client, force ACK now. */
2655                                 tp->t_flags |= TF_ACKNOW;
2656                         } else {
2657                                 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2658                                     tcp_timer_2msl);
2659                         }
2660                         soisdisconnected(so);
2661                         break;
2662
2663                 /*
2664                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2665                  */
2666                 case TCPS_TIME_WAIT:
2667                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2668                             tcp_timer_2msl);
2669                         break;
2670                 }
2671         }
2672
2673 #ifdef TCPDEBUG
2674         if (so->so_options & SO_DEBUG)
2675                 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2676 #endif
2677
2678         /*
2679          * Return any desired output.
2680          */
2681         if (needoutput || (tp->t_flags & TF_ACKNOW))
2682                 tcp_output(tp);
2683         return;
2684
2685 dropafterack:
2686         /*
2687          * Generate an ACK dropping incoming segment if it occupies
2688          * sequence space, where the ACK reflects our state.
2689          *
2690          * We can now skip the test for the RST flag since all
2691          * paths to this code happen after packets containing
2692          * RST have been dropped.
2693          *
2694          * In the SYN-RECEIVED state, don't send an ACK unless the
2695          * segment we received passes the SYN-RECEIVED ACK test.
2696          * If it fails send a RST.  This breaks the loop in the
2697          * "LAND" DoS attack, and also prevents an ACK storm
2698          * between two listening ports that have been sent forged
2699          * SYN segments, each with the source address of the other.
2700          */
2701         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2702             (SEQ_GT(tp->snd_una, th->th_ack) ||
2703              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2704                 rstreason = BANDLIM_RST_OPENPORT;
2705                 goto dropwithreset;
2706         }
2707 #ifdef TCPDEBUG
2708         if (so->so_options & SO_DEBUG)
2709                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2710 #endif
2711         m_freem(m);
2712         tp->t_flags |= TF_ACKNOW;
2713         tcp_output(tp);
2714         return;
2715
2716 dropwithreset:
2717         /*
2718          * Generate a RST, dropping incoming segment.
2719          * Make ACK acceptable to originator of segment.
2720          * Don't bother to respond if destination was broadcast/multicast.
2721          */
2722         if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2723                 goto drop;
2724         if (isipv6) {
2725                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2726                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2727                         goto drop;
2728         } else {
2729                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2730                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2731                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2732                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2733                         goto drop;
2734         }
2735         /* IPv6 anycast check is done at tcp6_input() */
2736
2737         /*
2738          * Perform bandwidth limiting.
2739          */
2740 #ifdef ICMP_BANDLIM
2741         if (badport_bandlim(rstreason) < 0)
2742                 goto drop;
2743 #endif
2744
2745 #ifdef TCPDEBUG
2746         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2747                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2748 #endif
2749         if (thflags & TH_ACK)
2750                 /* mtod() below is safe as long as hdr dropping is delayed */
2751                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2752                             TH_RST);
2753         else {
2754                 if (thflags & TH_SYN)
2755                         tlen++;
2756                 /* mtod() below is safe as long as hdr dropping is delayed */
2757                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2758                             (tcp_seq)0, TH_RST | TH_ACK);
2759         }
2760         return;
2761
2762 drop:
2763         /*
2764          * Drop space held by incoming segment and return.
2765          */
2766 #ifdef TCPDEBUG
2767         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2768                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2769 #endif
2770         m_freem(m);
2771         return;
2772 }
2773
2774 /*
2775  * Parse TCP options and place in tcpopt.
2776  */
2777 static void
2778 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2779 {
2780         int opt, optlen, i;
2781
2782         to->to_flags = 0;
2783         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2784                 opt = cp[0];
2785                 if (opt == TCPOPT_EOL)
2786                         break;
2787                 if (opt == TCPOPT_NOP)
2788                         optlen = 1;
2789                 else {
2790                         if (cnt < 2)
2791                                 break;
2792                         optlen = cp[1];
2793                         if (optlen < 2 || optlen > cnt)
2794                                 break;
2795                 }
2796                 switch (opt) {
2797                 case TCPOPT_MAXSEG:
2798                         if (optlen != TCPOLEN_MAXSEG)
2799                                 continue;
2800                         if (!is_syn)
2801                                 continue;
2802                         to->to_flags |= TOF_MSS;
2803                         bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2804                         to->to_mss = ntohs(to->to_mss);
2805                         break;
2806                 case TCPOPT_WINDOW:
2807                         if (optlen != TCPOLEN_WINDOW)
2808                                 continue;
2809                         if (!is_syn)
2810                                 continue;
2811                         to->to_flags |= TOF_SCALE;
2812                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2813                         break;
2814                 case TCPOPT_TIMESTAMP:
2815                         if (optlen != TCPOLEN_TIMESTAMP)
2816                                 continue;
2817                         to->to_flags |= TOF_TS;
2818                         bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2819                         to->to_tsval = ntohl(to->to_tsval);
2820                         bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2821                         to->to_tsecr = ntohl(to->to_tsecr);
2822                         /*
2823                          * If echoed timestamp is later than the current time,
2824                          * fall back to non RFC1323 RTT calculation.
2825                          */
2826                         if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2827                                 to->to_tsecr = 0;
2828                         break;
2829                 case TCPOPT_CC:
2830                         if (optlen != TCPOLEN_CC)
2831                                 continue;
2832                         to->to_flags |= TOF_CC;
2833                         bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2834                         to->to_cc = ntohl(to->to_cc);
2835                         break;
2836                 case TCPOPT_CCNEW:
2837                         if (optlen != TCPOLEN_CC)
2838                                 continue;
2839                         if (!is_syn)
2840                                 continue;
2841                         to->to_flags |= TOF_CCNEW;
2842                         bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2843                         to->to_cc = ntohl(to->to_cc);
2844                         break;
2845                 case TCPOPT_CCECHO:
2846                         if (optlen != TCPOLEN_CC)
2847                                 continue;
2848                         if (!is_syn)
2849                                 continue;
2850                         to->to_flags |= TOF_CCECHO;
2851                         bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho);
2852                         to->to_ccecho = ntohl(to->to_ccecho);
2853                         break;
2854                 case TCPOPT_SACK_PERMITTED:
2855                         if (optlen != TCPOLEN_SACK_PERMITTED)
2856                                 continue;
2857                         if (!is_syn)
2858                                 continue;
2859                         to->to_flags |= TOF_SACK_PERMITTED;
2860                         break;
2861                 case TCPOPT_SACK:
2862                         if ((optlen - 2) & 0x07)        /* not multiple of 8 */
2863                                 continue;
2864                         to->to_nsackblocks = (optlen - 2) / 8;
2865                         to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2866                         to->to_flags |= TOF_SACK;
2867                         for (i = 0; i < to->to_nsackblocks; i++) {
2868                                 struct raw_sackblock *r = &to->to_sackblocks[i];
2869
2870                                 r->rblk_start = ntohl(r->rblk_start);
2871                                 r->rblk_end = ntohl(r->rblk_end);
2872                         }
2873                         break;
2874                 default:
2875                         continue;
2876                 }
2877         }
2878 }
2879
2880 /*
2881  * Pull out of band byte out of a segment so
2882  * it doesn't appear in the user's data queue.
2883  * It is still reflected in the segment length for
2884  * sequencing purposes.
2885  * "off" is the delayed to be dropped hdrlen.
2886  */
2887 static void
2888 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2889 {
2890         int cnt = off + th->th_urp - 1;
2891
2892         while (cnt >= 0) {
2893                 if (m->m_len > cnt) {
2894                         char *cp = mtod(m, caddr_t) + cnt;
2895                         struct tcpcb *tp = sototcpcb(so);
2896
2897                         tp->t_iobc = *cp;
2898                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2899                         bcopy(cp + 1, cp, m->m_len - cnt - 1);
2900                         m->m_len--;
2901                         if (m->m_flags & M_PKTHDR)
2902                                 m->m_pkthdr.len--;
2903                         return;
2904                 }
2905                 cnt -= m->m_len;
2906                 m = m->m_next;
2907                 if (m == 0)
2908                         break;
2909         }
2910         panic("tcp_pulloutofband");
2911 }
2912
2913 /*
2914  * Collect new round-trip time estimate
2915  * and update averages and current timeout.
2916  */
2917 static void
2918 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2919 {
2920         int delta;
2921
2922         tcpstat.tcps_rttupdated++;
2923         tp->t_rttupdated++;
2924         if (tp->t_srtt != 0) {
2925                 /*
2926                  * srtt is stored as fixed point with 5 bits after the
2927                  * binary point (i.e., scaled by 8).  The following magic
2928                  * is equivalent to the smoothing algorithm in rfc793 with
2929                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2930                  * point).  Adjust rtt to origin 0.
2931                  */
2932                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2933                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2934
2935                 if ((tp->t_srtt += delta) <= 0)
2936                         tp->t_srtt = 1;
2937
2938                 /*
2939                  * We accumulate a smoothed rtt variance (actually, a
2940                  * smoothed mean difference), then set the retransmit
2941                  * timer to smoothed rtt + 4 times the smoothed variance.
2942                  * rttvar is stored as fixed point with 4 bits after the
2943                  * binary point (scaled by 16).  The following is
2944                  * equivalent to rfc793 smoothing with an alpha of .75
2945                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2946                  * rfc793's wired-in beta.
2947                  */
2948                 if (delta < 0)
2949                         delta = -delta;
2950                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2951                 if ((tp->t_rttvar += delta) <= 0)
2952                         tp->t_rttvar = 1;
2953                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2954                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2955         } else {
2956                 /*
2957                  * No rtt measurement yet - use the unsmoothed rtt.
2958                  * Set the variance to half the rtt (so our first
2959                  * retransmit happens at 3*rtt).
2960                  */
2961                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2962                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2963                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2964         }
2965         tp->t_rtttime = 0;
2966         tp->t_rxtshift = 0;
2967
2968         /*
2969          * the retransmit should happen at rtt + 4 * rttvar.
2970          * Because of the way we do the smoothing, srtt and rttvar
2971          * will each average +1/2 tick of bias.  When we compute
2972          * the retransmit timer, we want 1/2 tick of rounding and
2973          * 1 extra tick because of +-1/2 tick uncertainty in the
2974          * firing of the timer.  The bias will give us exactly the
2975          * 1.5 tick we need.  But, because the bias is
2976          * statistical, we have to test that we don't drop below
2977          * the minimum feasible timer (which is 2 ticks).
2978          */
2979         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2980                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2981
2982         /*
2983          * We received an ack for a packet that wasn't retransmitted;
2984          * it is probably safe to discard any error indications we've
2985          * received recently.  This isn't quite right, but close enough
2986          * for now (a route might have failed after we sent a segment,
2987          * and the return path might not be symmetrical).
2988          */
2989         tp->t_softerror = 0;
2990 }
2991
2992 /*
2993  * Determine a reasonable value for maxseg size.
2994  * If the route is known, check route for mtu.
2995  * If none, use an mss that can be handled on the outgoing
2996  * interface without forcing IP to fragment; if bigger than
2997  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2998  * to utilize large mbufs.  If no route is found, route has no mtu,
2999  * or the destination isn't local, use a default, hopefully conservative
3000  * size (usually 512 or the default IP max size, but no more than the mtu
3001  * of the interface), as we can't discover anything about intervening
3002  * gateways or networks.  We also initialize the congestion/slow start
3003  * window to be a single segment if the destination isn't local.
3004  * While looking at the routing entry, we also initialize other path-dependent
3005  * parameters from pre-set or cached values in the routing entry.
3006  *
3007  * Also take into account the space needed for options that we
3008  * send regularly.  Make maxseg shorter by that amount to assure
3009  * that we can send maxseg amount of data even when the options
3010  * are present.  Store the upper limit of the length of options plus
3011  * data in maxopd.
3012  *
3013  * NOTE that this routine is only called when we process an incoming
3014  * segment, for outgoing segments only tcp_mssopt is called.
3015  *
3016  * In case of T/TCP, we call this routine during implicit connection
3017  * setup as well (offer = -1), to initialize maxseg from the cached
3018  * MSS of our peer.
3019  */
3020 void
3021 tcp_mss(struct tcpcb *tp, int offer)
3022 {
3023         struct rtentry *rt;
3024         struct ifnet *ifp;
3025         int rtt, mss;
3026         u_long bufsize;
3027         struct inpcb *inp = tp->t_inpcb;
3028         struct socket *so;
3029         struct rmxp_tao *taop;
3030         int origoffer = offer;
3031 #ifdef INET6
3032         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3033         size_t min_protoh = isipv6 ?
3034                             sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3035                             sizeof(struct tcpiphdr);
3036 #else
3037         const boolean_t isipv6 = FALSE;
3038         const size_t min_protoh = sizeof(struct tcpiphdr);
3039 #endif
3040
3041         if (isipv6)
3042                 rt = tcp_rtlookup6(&inp->inp_inc);
3043         else
3044                 rt = tcp_rtlookup(&inp->inp_inc);
3045         if (rt == NULL) {
3046                 tp->t_maxopd = tp->t_maxseg =
3047                     (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3048                 return;
3049         }
3050         ifp = rt->rt_ifp;
3051         so = inp->inp_socket;
3052
3053         taop = rmx_taop(rt->rt_rmx);
3054         /*
3055          * Offer == -1 means that we didn't receive SYN yet,
3056          * use cached value in that case;
3057          */
3058         if (offer == -1)
3059                 offer = taop->tao_mssopt;
3060         /*
3061          * Offer == 0 means that there was no MSS on the SYN segment,
3062          * in this case we use tcp_mssdflt.
3063          */
3064         if (offer == 0) {
3065                 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3066         } else {
3067                 /*
3068                  * Prevent DoS attack with too small MSS. Round up
3069                  * to at least minmss.
3070                  */
3071                 offer = max(offer, tcp_minmss);
3072                 /*
3073                  * Sanity check: make sure that maxopd will be large
3074                  * enough to allow some data on segments even is the
3075                  * all the option space is used (40bytes).  Otherwise
3076                  * funny things may happen in tcp_output.
3077                  */
3078                 offer = max(offer, 64);
3079         }
3080         taop->tao_mssopt = offer;
3081
3082         /*
3083          * While we're here, check if there's an initial rtt
3084          * or rttvar.  Convert from the route-table units
3085          * to scaled multiples of the slow timeout timer.
3086          */
3087         if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3088                 /*
3089                  * XXX the lock bit for RTT indicates that the value
3090                  * is also a minimum value; this is subject to time.
3091                  */
3092                 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3093                         tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3094                 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3095                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3096                 tcpstat.tcps_usedrtt++;
3097                 if (rt->rt_rmx.rmx_rttvar) {
3098                         tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3099                             (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3100                         tcpstat.tcps_usedrttvar++;
3101                 } else {
3102                         /* default variation is +- 1 rtt */
3103                         tp->t_rttvar =
3104                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3105                 }
3106                 TCPT_RANGESET(tp->t_rxtcur,
3107                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3108                               tp->t_rttmin, TCPTV_REXMTMAX);
3109         }
3110         /*
3111          * if there's an mtu associated with the route, use it
3112          * else, use the link mtu.
3113          */
3114         if (rt->rt_rmx.rmx_mtu)
3115                 mss = rt->rt_rmx.rmx_mtu - min_protoh;
3116         else {
3117                 if (isipv6) {
3118                         mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
3119                         if (!in6_localaddr(&inp->in6p_faddr))
3120                                 mss = min(mss, tcp_v6mssdflt);
3121                 } else {
3122                         mss = ifp->if_mtu - min_protoh;
3123                         if (!in_localaddr(inp->inp_faddr))
3124                                 mss = min(mss, tcp_mssdflt);
3125                 }
3126         }
3127         mss = min(mss, offer);
3128         /*
3129          * maxopd stores the maximum length of data AND options
3130          * in a segment; maxseg is the amount of data in a normal
3131          * segment.  We need to store this value (maxopd) apart
3132          * from maxseg, because now every segment carries options
3133          * and thus we normally have somewhat less data in segments.
3134          */
3135         tp->t_maxopd = mss;
3136
3137         /*
3138          * In case of T/TCP, origoffer==-1 indicates, that no segments
3139          * were received yet.  In this case we just guess, otherwise
3140          * we do the same as before T/TCP.
3141          */
3142         if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3143             (origoffer == -1 ||
3144              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3145                 mss -= TCPOLEN_TSTAMP_APPA;
3146         if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
3147             (origoffer == -1 ||
3148              (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3149                 mss -= TCPOLEN_CC_APPA;
3150
3151 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
3152                 if (mss > MCLBYTES)
3153                         mss &= ~(MCLBYTES-1);
3154 #else
3155                 if (mss > MCLBYTES)
3156                         mss = mss / MCLBYTES * MCLBYTES;
3157 #endif
3158         /*
3159          * If there's a pipesize, change the socket buffer
3160          * to that size.  Make the socket buffers an integral
3161          * number of mss units; if the mss is larger than
3162          * the socket buffer, decrease the mss.
3163          */
3164 #ifdef RTV_SPIPE
3165         if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3166 #endif
3167                 bufsize = so->so_snd.ssb_hiwat;
3168         if (bufsize < mss)
3169                 mss = bufsize;
3170         else {
3171                 bufsize = roundup(bufsize, mss);
3172                 if (bufsize > sb_max)
3173                         bufsize = sb_max;
3174                 if (bufsize > so->so_snd.ssb_hiwat)
3175                         ssb_reserve(&so->so_snd, bufsize, so, NULL);
3176         }
3177         tp->t_maxseg = mss;
3178
3179 #ifdef RTV_RPIPE
3180         if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3181 #endif
3182                 bufsize = so->so_rcv.ssb_hiwat;
3183         if (bufsize > mss) {
3184                 bufsize = roundup(bufsize, mss);
3185                 if (bufsize > sb_max)
3186                         bufsize = sb_max;
3187                 if (bufsize > so->so_rcv.ssb_hiwat)
3188                         ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3189         }
3190
3191         /*
3192          * Set the slow-start flight size depending on whether this
3193          * is a local network or not.
3194          */
3195         if (tcp_do_rfc3390)
3196                 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3197         else
3198                 tp->snd_cwnd = mss;
3199
3200         if (rt->rt_rmx.rmx_ssthresh) {
3201                 /*
3202                  * There's some sort of gateway or interface
3203                  * buffer limit on the path.  Use this to set
3204                  * the slow start threshhold, but set the
3205                  * threshold to no less than 2*mss.
3206                  */
3207                 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3208                 tcpstat.tcps_usedssthresh++;
3209         }
3210 }
3211
3212 /*
3213  * Determine the MSS option to send on an outgoing SYN.
3214  */
3215 int
3216 tcp_mssopt(struct tcpcb *tp)
3217 {
3218         struct rtentry *rt;
3219 #ifdef INET6
3220         boolean_t isipv6 =
3221             ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3222         int min_protoh = isipv6 ?
3223                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3224                              sizeof(struct tcpiphdr);
3225 #else
3226         const boolean_t isipv6 = FALSE;
3227         const size_t min_protoh = sizeof(struct tcpiphdr);
3228 #endif
3229
3230         if (isipv6)
3231                 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3232         else
3233                 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3234         if (rt == NULL)
3235                 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3236
3237         return (rt->rt_ifp->if_mtu - min_protoh);
3238 }
3239
3240 /*
3241  * When a partial ack arrives, force the retransmission of the
3242  * next unacknowledged segment.  Do not exit Fast Recovery.
3243  *
3244  * Implement the Slow-but-Steady variant of NewReno by restarting the
3245  * the retransmission timer.  Turn it off here so it can be restarted
3246  * later in tcp_output().
3247  */
3248 static void
3249 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3250 {
3251         tcp_seq old_snd_nxt = tp->snd_nxt;
3252         u_long ocwnd = tp->snd_cwnd;
3253
3254         tcp_callout_stop(tp, tp->tt_rexmt);
3255         tp->t_rtttime = 0;
3256         tp->snd_nxt = th->th_ack;
3257         /* Set snd_cwnd to one segment beyond acknowledged offset. */
3258         tp->snd_cwnd = tp->t_maxseg;
3259         tp->t_flags |= TF_ACKNOW;
3260         tcp_output(tp);
3261         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3262                 tp->snd_nxt = old_snd_nxt;
3263         /* partial window deflation */
3264         if (ocwnd > acked)
3265                 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3266         else
3267                 tp->snd_cwnd = tp->t_maxseg;
3268 }
3269
3270 /*
3271  * In contrast to the Slow-but-Steady NewReno variant,
3272  * we do not reset the retransmission timer for SACK retransmissions,
3273  * except when retransmitting snd_una.
3274  */
3275 static void
3276 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3277 {
3278         uint32_t pipe, seglen;
3279         tcp_seq nextrexmt;
3280         boolean_t lostdup;
3281         tcp_seq old_snd_nxt = tp->snd_nxt;
3282         u_long ocwnd = tp->snd_cwnd;
3283         int nseg = 0;           /* consecutive new segments */
3284 #define MAXBURST 4              /* limit burst of new packets on partial ack */
3285
3286         tp->t_rtttime = 0;
3287         pipe = tcp_sack_compute_pipe(tp);
3288         while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3289             (!tcp_do_smartsack || nseg < MAXBURST) &&
3290             tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3291                 uint32_t sent;
3292                 tcp_seq old_snd_max;
3293                 int error;
3294
3295                 if (nextrexmt == tp->snd_max)
3296                         ++nseg;
3297                 tp->snd_nxt = nextrexmt;
3298                 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3299                 old_snd_max = tp->snd_max;
3300                 if (nextrexmt == tp->snd_una)
3301                         tcp_callout_stop(tp, tp->tt_rexmt);
3302                 error = tcp_output(tp);
3303                 if (error != 0)
3304                         break;
3305                 sent = tp->snd_nxt - nextrexmt;
3306                 if (sent <= 0)
3307                         break;
3308                 if (!lostdup)
3309                         pipe += sent;
3310                 tcpstat.tcps_sndsackpack++;
3311                 tcpstat.tcps_sndsackbyte += sent;
3312                 if (SEQ_LT(nextrexmt, old_snd_max) &&
3313                     SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3314                         tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3315         }
3316         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3317                 tp->snd_nxt = old_snd_nxt;
3318         tp->snd_cwnd = ocwnd;
3319 }