Cache a pointer the last mbuf in the sockbuf for faster insertion.
[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) 2002, 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
36  *
37  * License terms: all terms for the DragonFly license above plus the following:
38  *
39  * 4. All advertising materials mentioning features or use of this software
40  *    must display the following acknowledgement:
41  *
42  *      This product includes software developed by Jeffrey M. Hsu
43  *      for the DragonFly Project.
44  *
45  *    This requirement may be waived with permission from Jeffrey Hsu.
46  *    This requirement will sunset and may be removed on July 8 2005,
47  *    after which the standard DragonFly license (as shown above) will
48  *    apply.
49  */
50
51 /*
52  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
53  *      The Regents of the University of California.  All rights reserved.
54  *
55  * Redistribution and use in source and binary forms, with or without
56  * modification, are permitted provided that the following conditions
57  * are met:
58  * 1. Redistributions of source code must retain the above copyright
59  *    notice, this list of conditions and the following disclaimer.
60  * 2. Redistributions in binary form must reproduce the above copyright
61  *    notice, this list of conditions and the following disclaimer in the
62  *    documentation and/or other materials provided with the distribution.
63  * 3. All advertising materials mentioning features or use of this software
64  *    must display the following acknowledgement:
65  *      This product includes software developed by the University of
66  *      California, Berkeley and its contributors.
67  * 4. Neither the name of the University nor the names of its contributors
68  *    may be used to endorse or promote products derived from this software
69  *    without specific prior written permission.
70  *
71  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
81  * SUCH DAMAGE.
82  *
83  *      @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
84  * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
85  * $DragonFly: src/sys/netinet/tcp_input.c,v 1.45 2004/12/08 23:59:01 hsu Exp $
86  */
87
88 #include "opt_ipfw.h"           /* for ipfw_fwd         */
89 #include "opt_inet6.h"
90 #include "opt_ipsec.h"
91 #include "opt_tcpdebug.h"
92 #include "opt_tcp_input.h"
93
94 #include <sys/param.h>
95 #include <sys/systm.h>
96 #include <sys/kernel.h>
97 #include <sys/sysctl.h>
98 #include <sys/malloc.h>
99 #include <sys/mbuf.h>
100 #include <sys/proc.h>           /* for proc0 declaration */
101 #include <sys/protosw.h>
102 #include <sys/socket.h>
103 #include <sys/socketvar.h>
104 #include <sys/syslog.h>
105 #include <sys/in_cksum.h>
106
107 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
108 #include <machine/stdarg.h>
109
110 #include <net/if.h>
111 #include <net/route.h>
112
113 #include <netinet/in.h>
114 #include <netinet/in_systm.h>
115 #include <netinet/ip.h>
116 #include <netinet/ip_icmp.h>    /* for ICMP_BANDLIM */
117 #include <netinet/in_var.h>
118 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
119 #include <netinet/in_pcb.h>
120 #include <netinet/ip_var.h>
121 #include <netinet/ip6.h>
122 #include <netinet/icmp6.h>
123 #include <netinet6/nd6.h>
124 #include <netinet6/ip6_var.h>
125 #include <netinet6/in6_pcb.h>
126 #include <netinet/tcp.h>
127 #include <netinet/tcp_fsm.h>
128 #include <netinet/tcp_seq.h>
129 #include <netinet/tcp_timer.h>
130 #include <netinet/tcp_var.h>
131 #include <netinet6/tcp6_var.h>
132 #include <netinet/tcpip.h>
133
134 #ifdef TCPDEBUG
135 #include <netinet/tcp_debug.h>
136
137 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
138 struct tcphdr tcp_savetcp;
139 #endif
140
141 #ifdef FAST_IPSEC
142 #include <netproto/ipsec/ipsec.h>
143 #include <netproto/ipsec/ipsec6.h>
144 #endif
145
146 #ifdef IPSEC
147 #include <netinet6/ipsec.h>
148 #include <netinet6/ipsec6.h>
149 #include <netproto/key/key.h>
150 #endif
151
152 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
153
154 tcp_cc  tcp_ccgen;
155 static int log_in_vain = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
157     &log_in_vain, 0, "Log all incoming TCP connections");
158
159 static int blackhole = 0;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
161     &blackhole, 0, "Do not send RST when dropping refused connections");
162
163 int tcp_delack_enabled = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
165     &tcp_delack_enabled, 0,
166     "Delay ACK to try and piggyback it onto a data packet");
167
168 #ifdef TCP_DROP_SYNFIN
169 static int drop_synfin = 0;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
171     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
172 #endif
173
174 static int tcp_do_limitedtransmit = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
176     &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
177
178 static int tcp_do_early_retransmit = 1;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
180     &tcp_do_early_retransmit, 0, "Early retransmit");
181
182 static int tcp_do_rfc3390 = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
184     &tcp_do_rfc3390, 0,
185     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
186
187 static int tcp_do_eifel_detect = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
189     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
190
191 /*
192  * Define as tunable for easy testing with SACK on and off.
193  * Warning:  do not change setting in the middle of an existing active TCP flow,
194  *   else strange things might happen to that flow.
195  */
196 int tcp_do_sack = 1;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
198     &tcp_do_sack, 0, "Enable SACK Algorithms");
199
200 int tcp_do_smartsack = 1;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
202     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
203
204 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
205     "TCP Segment Reassembly Queue");
206
207 int tcp_reass_maxseg = 0;
208 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
209     &tcp_reass_maxseg, 0,
210     "Global maximum number of TCP Segments in Reassembly Queue");
211
212 int tcp_reass_qsize = 0;
213 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
214     &tcp_reass_qsize, 0,
215     "Global number of TCP Segments currently in Reassembly Queue");
216
217 static int tcp_reass_overflows = 0;
218 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
219     &tcp_reass_overflows, 0,
220     "Global number of TCP Segment Reassembly Queue Overflows");
221
222 static void      tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
223 static void      tcp_pulloutofband(struct socket *,
224                      struct tcphdr *, struct mbuf *, int);
225 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
226                      struct mbuf *);
227 static void      tcp_xmit_timer(struct tcpcb *, int);
228 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
229 static void      tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
230
231 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
232 #ifdef INET6
233 #define ND6_HINT(tp) \
234 do { \
235         if ((tp) && (tp)->t_inpcb && \
236             ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
237             (tp)->t_inpcb->in6p_route.ro_rt) \
238                 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
239 } while (0)
240 #else
241 #define ND6_HINT(tp)
242 #endif
243
244 /*
245  * Indicate whether this ack should be delayed.  We can delay the ack if
246  *      - delayed acks are enabled and
247  *      - there is no delayed ack timer in progress and
248  *      - our last ack wasn't a 0-sized window.  We never want to delay
249  *        the ack that opens up a 0-sized window.
250  */
251 #define DELAY_ACK(tp) \
252         (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
253         !(tp->t_flags & TF_RXWIN0SENT))
254
255 #define acceptable_window_update(tp, th, tiwin)                         \
256     (SEQ_LT(tp->snd_wl1, th->th_seq) ||                                 \
257      (tp->snd_wl1 == th->th_seq &&                                      \
258       (SEQ_LT(tp->snd_wl2, th->th_ack) ||                               \
259        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
260
261 static int
262 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
263 {
264         struct tseg_qent *q;
265         struct tseg_qent *p = NULL;
266         struct tseg_qent *te;
267         struct socket *so = tp->t_inpcb->inp_socket;
268         int flags;
269
270         /*
271          * Call with th == NULL after become established to
272          * force pre-ESTABLISHED data up to user socket.
273          */
274         if (th == NULL)
275                 goto present;
276
277         /*
278          * Limit the number of segments in the reassembly queue to prevent
279          * holding on to too many segments (and thus running out of mbufs).
280          * Make sure to let the missing segment through which caused this
281          * queue.  Always keep one global queue entry spare to be able to
282          * process the missing segment.
283          */
284         if (th->th_seq != tp->rcv_nxt &&
285             tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
286                 tcp_reass_overflows++;
287                 tcpstat.tcps_rcvmemdrop++;
288                 m_freem(m);
289                 /* no SACK block to report */
290                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
291                 return (0);
292         }
293
294         /* Allocate a new queue entry. */
295         MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
296                M_INTWAIT | M_NULLOK);
297         if (te == NULL) {
298                 tcpstat.tcps_rcvmemdrop++;
299                 m_freem(m);
300                 /* no SACK block to report */
301                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
302                 return (0);
303         }
304         tcp_reass_qsize++;
305
306         /*
307          * Find a segment which begins after this one does.
308          */
309         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
310                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
311                         break;
312                 p = q;
313         }
314
315         /*
316          * If there is a preceding segment, it may provide some of
317          * our data already.  If so, drop the data from the incoming
318          * segment.  If it provides all of our data, drop us.
319          */
320         if (p != NULL) {
321                 int i;
322
323                 /* conversion to int (in i) handles seq wraparound */
324                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
325                 if (i > 0) {            /* overlaps preceding segment */
326                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
327                         /* enclosing block starts w/ preceding segment */
328                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
329                         if (i >= *tlenp) {
330                                 /* preceding encloses incoming segment */
331                                 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
332                                     p->tqe_len;
333                                 tcpstat.tcps_rcvduppack++;
334                                 tcpstat.tcps_rcvdupbyte += *tlenp;
335                                 m_freem(m);
336                                 free(te, M_TSEGQ);
337                                 tcp_reass_qsize--;
338                                 /*
339                                  * Try to present any queued data
340                                  * at the left window edge to the user.
341                                  * This is needed after the 3-WHS
342                                  * completes.
343                                  */
344                                 goto present;   /* ??? */
345                         }
346                         m_adj(m, i);
347                         *tlenp -= i;
348                         th->th_seq += i;
349                         /* incoming segment end is enclosing block end */
350                         tp->encloseblk.rblk_end = th->th_seq + *tlenp +
351                             ((th->th_flags & TH_FIN) != 0);
352                         /* trim end of reported D-SACK block */
353                         tp->reportblk.rblk_end = th->th_seq;
354                 }
355         }
356         tcpstat.tcps_rcvoopack++;
357         tcpstat.tcps_rcvoobyte += *tlenp;
358
359         /*
360          * While we overlap succeeding segments trim them or,
361          * if they are completely covered, dequeue them.
362          */
363         while (q) {
364                 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
365                 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
366                 struct tseg_qent *nq;
367
368                 if (i <= 0)
369                         break;
370                 if (!(tp->t_flags & TF_DUPSEG)) {    /* first time through */
371                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
372                         tp->encloseblk = tp->reportblk;
373                         /* report trailing duplicate D-SACK segment */
374                         tp->reportblk.rblk_start = q->tqe_th->th_seq;
375                 }
376                 if ((tp->t_flags & TF_ENCLOSESEG) &&
377                     SEQ_GT(qend, tp->encloseblk.rblk_end)) {
378                         /* extend enclosing block if one exists */
379                         tp->encloseblk.rblk_end = qend;
380                 }
381                 if (i < q->tqe_len) {
382                         q->tqe_th->th_seq += i;
383                         q->tqe_len -= i;
384                         m_adj(q->tqe_m, i);
385                         break;
386                 }
387
388                 nq = LIST_NEXT(q, tqe_q);
389                 LIST_REMOVE(q, tqe_q);
390                 m_freem(q->tqe_m);
391                 free(q, M_TSEGQ);
392                 tcp_reass_qsize--;
393                 q = nq;
394         }
395
396         /* Insert the new segment queue entry into place. */
397         te->tqe_m = m;
398         te->tqe_th = th;
399         te->tqe_len = *tlenp;
400
401         /* check if can coalesce with following segment */
402         if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
403                 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
404
405                 te->tqe_len += q->tqe_len;
406                 if (q->tqe_th->th_flags & TH_FIN)
407                         te->tqe_th->th_flags |= TH_FIN;
408                 m_cat(te->tqe_m, q->tqe_m);
409                 tp->encloseblk.rblk_end = tend;
410                 /*
411                  * When not reporting a duplicate segment, use
412                  * the larger enclosing block as the SACK block.
413                  */
414                 if (!(tp->t_flags & TF_DUPSEG))
415                         tp->reportblk.rblk_end = tend;
416                 LIST_REMOVE(q, tqe_q);
417                 free(q, M_TSEGQ);
418                 tcp_reass_qsize--;
419         }
420
421         if (p == NULL) {
422                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
423         } else {
424                 /* check if can coalesce with preceding segment */
425                 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
426                         p->tqe_len += te->tqe_len;
427                         m_cat(p->tqe_m, te->tqe_m);
428                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
429                         /*
430                          * When not reporting a duplicate segment, use
431                          * the larger enclosing block as the SACK block.
432                          */
433                         if (!(tp->t_flags & TF_DUPSEG))
434                                 tp->reportblk.rblk_start = p->tqe_th->th_seq;
435                         free(te, M_TSEGQ);
436                         tcp_reass_qsize--;
437                 } else
438                         LIST_INSERT_AFTER(p, te, tqe_q);
439         }
440
441 present:
442         /*
443          * Present data to user, advancing rcv_nxt through
444          * completed sequence space.
445          */
446         if (!TCPS_HAVEESTABLISHED(tp->t_state))
447                 return (0);
448         q = LIST_FIRST(&tp->t_segq);
449         if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
450                 return (0);
451         tp->rcv_nxt += q->tqe_len;
452         if (!(tp->t_flags & TF_DUPSEG)) {
453                 /* no SACK block to report since ACK advanced */
454                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
455         }
456         /* no enclosing block to report since ACK advanced */
457         tp->t_flags &= ~TF_ENCLOSESEG;
458         flags = q->tqe_th->th_flags & TH_FIN;
459         LIST_REMOVE(q, tqe_q);
460         KASSERT(LIST_EMPTY(&tp->t_segq) ||
461                 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
462                 ("segment not coalesced"));
463         if (so->so_state & SS_CANTRCVMORE)
464                 m_freem(q->tqe_m);
465         else
466                 sbappendstream(&so->so_rcv, q->tqe_m);
467         free(q, M_TSEGQ);
468         tcp_reass_qsize--;
469         ND6_HINT(tp);
470         sorwakeup(so);
471         return (flags);
472 }
473
474 /*
475  * TCP input routine, follows pages 65-76 of the
476  * protocol specification dated September, 1981 very closely.
477  */
478 #ifdef INET6
479 int
480 tcp6_input(struct mbuf **mp, int *offp, int proto)
481 {
482         struct mbuf *m = *mp;
483         struct in6_ifaddr *ia6;
484
485         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
486
487         /*
488          * draft-itojun-ipv6-tcp-to-anycast
489          * better place to put this in?
490          */
491         ia6 = ip6_getdstifaddr(m);
492         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
493                 struct ip6_hdr *ip6;
494
495                 ip6 = mtod(m, struct ip6_hdr *);
496                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
497                             (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
498                 return (IPPROTO_DONE);
499         }
500
501         tcp_input(m, *offp, proto);
502         return (IPPROTO_DONE);
503 }
504 #endif
505
506 void
507 tcp_input(struct mbuf *m, ...)
508 {
509         __va_list ap;
510         int off0, proto;
511         struct tcphdr *th;
512         struct ip *ip = NULL;
513         struct ipovly *ipov;
514         struct inpcb *inp = NULL;
515         u_char *optp = NULL;
516         int optlen = 0;
517         int len, tlen, off;
518         int drop_hdrlen;
519         struct tcpcb *tp = NULL;
520         int thflags;
521         struct socket *so = 0;
522         int todrop, acked;
523         boolean_t ourfinisacked, needoutput = FALSE;
524         u_long tiwin;
525         int recvwin;
526         struct tcpopt to;               /* options in this segment */
527         struct rmxp_tao *taop;          /* pointer to our TAO cache entry */
528         struct rmxp_tao tao_noncached;  /* in case there's no cached entry */
529         struct sockaddr_in *next_hop = NULL;
530         int rstreason; /* For badport_bandlim accounting purposes */
531         int cpu;
532         struct ip6_hdr *ip6 = NULL;
533 #ifdef INET6
534         boolean_t isipv6;
535 #else
536         const boolean_t isipv6 = FALSE;
537 #endif
538 #ifdef TCPDEBUG
539         short ostate = 0;
540 #endif
541
542         __va_start(ap, m);
543         off0 = __va_arg(ap, int);
544         proto = __va_arg(ap, int);
545         __va_end(ap);
546
547         tcpstat.tcps_rcvtotal++;
548
549         /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
550         while  (m->m_type == MT_TAG) {
551                 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
552                         next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
553                 m = m->m_next;
554         }
555
556 #ifdef INET6
557         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
558 #endif
559
560         if (isipv6) {
561                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
562                 ip6 = mtod(m, struct ip6_hdr *);
563                 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
564                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
565                         tcpstat.tcps_rcvbadsum++;
566                         goto drop;
567                 }
568                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
569
570                 /*
571                  * Be proactive about unspecified IPv6 address in source.
572                  * As we use all-zero to indicate unbounded/unconnected pcb,
573                  * unspecified IPv6 address can be used to confuse us.
574                  *
575                  * Note that packets with unspecified IPv6 destination is
576                  * already dropped in ip6_input.
577                  */
578                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
579                         /* XXX stat */
580                         goto drop;
581                 }
582         } else {
583                 /*
584                  * Get IP and TCP header together in first mbuf.
585                  * Note: IP leaves IP header in first mbuf.
586                  */
587                 if (off0 > sizeof(struct ip)) {
588                         ip_stripoptions(m);
589                         off0 = sizeof(struct ip);
590                 }
591                 /* already checked and pulled up in ip_demux() */
592                 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
593                     ("TCP header not in one mbuf: m->m_len %d", m->m_len));
594                 ip = mtod(m, struct ip *);
595                 ipov = (struct ipovly *)ip;
596                 th = (struct tcphdr *)((caddr_t)ip + off0);
597                 tlen = ip->ip_len;
598
599                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
600                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
601                                 th->th_sum = m->m_pkthdr.csum_data;
602                         else
603                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
604                                                 ip->ip_dst.s_addr,
605                                                 htonl(m->m_pkthdr.csum_data +
606                                                         ip->ip_len +
607                                                         IPPROTO_TCP));
608                         th->th_sum ^= 0xffff;
609                 } else {
610                         /*
611                          * Checksum extended TCP header and data.
612                          */
613                         len = sizeof(struct ip) + tlen;
614                         bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
615                         ipov->ih_len = (u_short)tlen;
616                         ipov->ih_len = htons(ipov->ih_len);
617                         th->th_sum = in_cksum(m, len);
618                 }
619                 if (th->th_sum) {
620                         tcpstat.tcps_rcvbadsum++;
621                         goto drop;
622                 }
623 #ifdef INET6
624                 /* Re-initialization for later version check */
625                 ip->ip_v = IPVERSION;
626 #endif
627         }
628
629         /*
630          * Check that TCP offset makes sense,
631          * pull out TCP options and adjust length.              XXX
632          */
633         off = th->th_off << 2;
634         /* already checked and pulled up in ip_demux() */
635         KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
636             ("bad TCP data offset %d (tlen %d)", off, tlen));
637         tlen -= off;    /* tlen is used instead of ti->ti_len */
638         if (off > sizeof(struct tcphdr)) {
639                 if (isipv6) {
640                         IP6_EXTHDR_CHECK(m, off0, off, );
641                         ip6 = mtod(m, struct ip6_hdr *);
642                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
643                 } else {
644                         /* already pulled up in ip_demux() */
645                         KASSERT(m->m_len >= sizeof(struct ip) + off,
646                             ("TCP header and options not in one mbuf: "
647                              "m_len %d, off %d", m->m_len, off));
648                 }
649                 optlen = off - sizeof(struct tcphdr);
650                 optp = (u_char *)(th + 1);
651         }
652         thflags = th->th_flags;
653
654 #ifdef TCP_DROP_SYNFIN
655         /*
656          * If the drop_synfin option is enabled, drop all packets with
657          * both the SYN and FIN bits set. This prevents e.g. nmap from
658          * identifying the TCP/IP stack.
659          *
660          * This is a violation of the TCP specification.
661          */
662         if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
663                 goto drop;
664 #endif
665
666         /*
667          * Convert TCP protocol specific fields to host format.
668          */
669         th->th_seq = ntohl(th->th_seq);
670         th->th_ack = ntohl(th->th_ack);
671         th->th_win = ntohs(th->th_win);
672         th->th_urp = ntohs(th->th_urp);
673
674         /*
675          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
676          * until after ip6_savecontrol() is called and before other functions
677          * which don't want those proto headers.
678          * Because ip6_savecontrol() is going to parse the mbuf to
679          * search for data to be passed up to user-land, it wants mbuf
680          * parameters to be unchanged.
681          * XXX: the call of ip6_savecontrol() has been obsoleted based on
682          * latest version of the advanced API (20020110).
683          */
684         drop_hdrlen = off0 + off;
685
686         /*
687          * Locate pcb for segment.
688          */
689 findpcb:
690         /* IPFIREWALL_FORWARD section */
691         if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
692                 /*
693                  * Transparently forwarded. Pretend to be the destination.
694                  * already got one like this?
695                  */
696                 cpu = mycpu->gd_cpuid;
697                 inp = in_pcblookup_hash(&tcbinfo[cpu],
698                                         ip->ip_src, th->th_sport,
699                                         ip->ip_dst, th->th_dport,
700                                         0, m->m_pkthdr.rcvif);
701                 if (!inp) {
702                         /*
703                          * It's new.  Try to find the ambushing socket.
704                          */
705
706                         /*
707                          * The rest of the ipfw code stores the port in
708                          * host order.  XXX
709                          * (The IP address is still in network order.)
710                          */
711                         in_port_t dport = next_hop->sin_port ?
712                                                 htons(next_hop->sin_port) :
713                                                 th->th_dport;
714
715                         cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
716                                           next_hop->sin_addr.s_addr, dport);
717                         inp = in_pcblookup_hash(&tcbinfo[cpu],
718                                                 ip->ip_src, th->th_sport,
719                                                 next_hop->sin_addr, dport,
720                                                 1, m->m_pkthdr.rcvif);
721                 }
722         } else {
723                 if (isipv6) {
724                         inp = in6_pcblookup_hash(&tcbinfo[0],
725                                                  &ip6->ip6_src, th->th_sport,
726                                                  &ip6->ip6_dst, th->th_dport,
727                                                  1, m->m_pkthdr.rcvif);
728                 } else {
729                         cpu = mycpu->gd_cpuid;
730                         inp = in_pcblookup_hash(&tcbinfo[cpu],
731                                                 ip->ip_src, th->th_sport,
732                                                 ip->ip_dst, th->th_dport,
733                                                 1, m->m_pkthdr.rcvif);
734                 }
735         }
736
737         /*
738          * If the state is CLOSED (i.e., TCB does not exist) then
739          * all data in the incoming segment is discarded.
740          * If the TCB exists but is in CLOSED state, it is embryonic,
741          * but should either do a listen or a connect soon.
742          */
743         if (inp == NULL) {
744                 if (log_in_vain) {
745 #ifdef INET6
746                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
747 #else
748                         char dbuf[4 * sizeof "123"], sbuf[4 * sizeof "123"];
749 #endif
750                         if (isipv6) {
751                                 strcpy(dbuf, "[");
752                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
753                                 strcat(dbuf, "]");
754                                 strcpy(sbuf, "[");
755                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
756                                 strcat(sbuf, "]");
757                         } else {
758                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
759                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
760                         }
761                         switch (log_in_vain) {
762                         case 1:
763                                 if (!(thflags & TH_SYN))
764                                         break;
765                         case 2:
766                                 log(LOG_INFO,
767                                     "Connection attempt to TCP %s:%d "
768                                     "from %s:%d flags:0x%02x\n",
769                                     dbuf, ntohs(th->th_dport), sbuf,
770                                     ntohs(th->th_sport), thflags);
771                                 break;
772                         default:
773                                 break;
774                         }
775                 }
776                 if (blackhole) {
777                         switch (blackhole) {
778                         case 1:
779                                 if (thflags & TH_SYN)
780                                         goto drop;
781                                 break;
782                         case 2:
783                                 goto drop;
784                         default:
785                                 goto drop;
786                         }
787                 }
788                 rstreason = BANDLIM_RST_CLOSEDPORT;
789                 goto dropwithreset;
790         }
791
792 #ifdef IPSEC
793         if (isipv6) {
794                 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
795                         ipsec6stat.in_polvio++;
796                         goto drop;
797                 }
798         } else {
799                 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
800                         ipsecstat.in_polvio++;
801                         goto drop;
802                 }
803         }
804 #endif
805 #ifdef FAST_IPSEC
806         if (isipv6) {
807                 if (ipsec6_in_reject(m, inp))
808                         goto drop;
809         } else {
810                 if (ipsec4_in_reject(m, inp))
811                         goto drop;
812         }
813 #endif
814
815         tp = intotcpcb(inp);
816         if (tp == NULL) {
817                 rstreason = BANDLIM_RST_CLOSEDPORT;
818                 goto dropwithreset;
819         }
820         if (tp->t_state <= TCPS_CLOSED)
821                 goto drop;
822
823         /* Unscale the window into a 32-bit value. */
824         if (!(thflags & TH_SYN))
825                 tiwin = th->th_win << tp->snd_scale;
826         else
827                 tiwin = th->th_win;
828
829         so = inp->inp_socket;
830
831 #ifdef TCPDEBUG
832         if (so->so_options & SO_DEBUG) {
833                 ostate = tp->t_state;
834                 if (isipv6)
835                         bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
836                 else
837                         bcopy(ip, tcp_saveipgen, sizeof(*ip));
838                 tcp_savetcp = *th;
839         }
840 #endif
841
842         bzero(&to, sizeof(to));
843
844         if (so->so_options & SO_ACCEPTCONN) {
845                 struct in_conninfo inc;
846
847 #ifdef INET6
848                 inc.inc_isipv6 = (isipv6 == TRUE);
849 #endif
850                 if (isipv6) {
851                         inc.inc6_faddr = ip6->ip6_src;
852                         inc.inc6_laddr = ip6->ip6_dst;
853                         inc.inc6_route.ro_rt = NULL;            /* XXX */
854                 } else {
855                         inc.inc_faddr = ip->ip_src;
856                         inc.inc_laddr = ip->ip_dst;
857                         inc.inc_route.ro_rt = NULL;             /* XXX */
858                 }
859                 inc.inc_fport = th->th_sport;
860                 inc.inc_lport = th->th_dport;
861
862                 /*
863                  * If the state is LISTEN then ignore segment if it contains
864                  * a RST.  If the segment contains an ACK then it is bad and
865                  * send a RST.  If it does not contain a SYN then it is not
866                  * interesting; drop it.
867                  *
868                  * If the state is SYN_RECEIVED (syncache) and seg contains
869                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
870                  * contains a RST, check the sequence number to see if it
871                  * is a valid reset segment.
872                  */
873                 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
874                         if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
875                                 if (!syncache_expand(&inc, th, &so, m)) {
876                                         /*
877                                          * No syncache entry, or ACK was not
878                                          * for our SYN/ACK.  Send a RST.
879                                          */
880                                         tcpstat.tcps_badsyn++;
881                                         rstreason = BANDLIM_RST_OPENPORT;
882                                         goto dropwithreset;
883                                 }
884                                 if (so == NULL)
885                                         /*
886                                          * Could not complete 3-way handshake,
887                                          * connection is being closed down, and
888                                          * syncache will free mbuf.
889                                          */
890                                         return;
891                                 /*
892                                  * Socket is created in state SYN_RECEIVED.
893                                  * Continue processing segment.
894                                  */
895                                 inp = sotoinpcb(so);
896                                 tp = intotcpcb(inp);
897                                 /*
898                                  * This is what would have happened in
899                                  * tcp_output() when the SYN,ACK was sent.
900                                  */
901                                 tp->snd_up = tp->snd_una;
902                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
903                                 tp->last_ack_sent = tp->rcv_nxt;
904 /*
905  * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
906  * until the _second_ ACK is received:
907  *    rcv SYN (set wscale opts)  --> send SYN/ACK, set snd_wnd = window.
908  *    rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
909  *        move to ESTAB, set snd_wnd to tiwin.
910  */
911                                 tp->snd_wnd = tiwin;    /* unscaled */
912                                 goto after_listen;
913                         }
914                         if (thflags & TH_RST) {
915                                 syncache_chkrst(&inc, th);
916                                 goto drop;
917                         }
918                         if (thflags & TH_ACK) {
919                                 syncache_badack(&inc);
920                                 tcpstat.tcps_badsyn++;
921                                 rstreason = BANDLIM_RST_OPENPORT;
922                                 goto dropwithreset;
923                         }
924                         goto drop;
925                 }
926
927                 /*
928                  * Segment's flags are (SYN) or (SYN | FIN).
929                  */
930 #ifdef INET6
931                 /*
932                  * If deprecated address is forbidden,
933                  * we do not accept SYN to deprecated interface
934                  * address to prevent any new inbound connection from
935                  * getting established.
936                  * When we do not accept SYN, we send a TCP RST,
937                  * with deprecated source address (instead of dropping
938                  * it).  We compromise it as it is much better for peer
939                  * to send a RST, and RST will be the final packet
940                  * for the exchange.
941                  *
942                  * If we do not forbid deprecated addresses, we accept
943                  * the SYN packet.  RFC2462 does not suggest dropping
944                  * SYN in this case.
945                  * If we decipher RFC2462 5.5.4, it says like this:
946                  * 1. use of deprecated addr with existing
947                  *    communication is okay - "SHOULD continue to be
948                  *    used"
949                  * 2. use of it with new communication:
950                  *   (2a) "SHOULD NOT be used if alternate address
951                  *        with sufficient scope is available"
952                  *   (2b) nothing mentioned otherwise.
953                  * Here we fall into (2b) case as we have no choice in
954                  * our source address selection - we must obey the peer.
955                  *
956                  * The wording in RFC2462 is confusing, and there are
957                  * multiple description text for deprecated address
958                  * handling - worse, they are not exactly the same.
959                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
960                  */
961                 if (isipv6 && !ip6_use_deprecated) {
962                         struct in6_ifaddr *ia6;
963
964                         if ((ia6 = ip6_getdstifaddr(m)) &&
965                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
966                                 tp = NULL;
967                                 rstreason = BANDLIM_RST_OPENPORT;
968                                 goto dropwithreset;
969                         }
970                 }
971 #endif
972                 /*
973                  * If it is from this socket, drop it, it must be forged.
974                  * Don't bother responding if the destination was a broadcast.
975                  */
976                 if (th->th_dport == th->th_sport) {
977                         if (isipv6) {
978                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
979                                                        &ip6->ip6_src))
980                                         goto drop;
981                         } else {
982                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
983                                         goto drop;
984                         }
985                 }
986                 /*
987                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
988                  *
989                  * Note that it is quite possible to receive unicast
990                  * link-layer packets with a broadcast IP address. Use
991                  * in_broadcast() to find them.
992                  */
993                 if (m->m_flags & (M_BCAST | M_MCAST))
994                         goto drop;
995                 if (isipv6) {
996                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
997                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
998                                 goto drop;
999                 } else {
1000                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1001                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1002                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1003                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1004                                 goto drop;
1005                 }
1006                 /*
1007                  * SYN appears to be valid; create compressed TCP state
1008                  * for syncache, or perform t/tcp connection.
1009                  */
1010                 if (so->so_qlen <= so->so_qlimit) {
1011                         tcp_dooptions(&to, optp, optlen, TRUE);
1012                         if (!syncache_add(&inc, &to, th, &so, m))
1013                                 goto drop;
1014                         if (so == NULL)
1015                                 /*
1016                                  * Entry added to syncache, mbuf used to
1017                                  * send SYN,ACK packet.
1018                                  */
1019                                 return;
1020                         /*
1021                          * Segment passed TAO tests.
1022                          */
1023                         inp = sotoinpcb(so);
1024                         tp = intotcpcb(inp);
1025                         tp->snd_wnd = tiwin;
1026                         tp->t_starttime = ticks;
1027                         tp->t_state = TCPS_ESTABLISHED;
1028
1029                         /*
1030                          * If there is a FIN, or if there is data and the
1031                          * connection is local, then delay SYN,ACK(SYN) in
1032                          * the hope of piggy-backing it on a response
1033                          * segment.  Otherwise must send ACK now in case
1034                          * the other side is slow starting.
1035                          */
1036                         if (DELAY_ACK(tp) &&
1037                             ((thflags & TH_FIN) ||
1038                              (tlen != 0 &&
1039                               ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1040                                (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1041                                 callout_reset(tp->tt_delack, tcp_delacktime,
1042                                                 tcp_timer_delack, tp);
1043                                 tp->t_flags |= TF_NEEDSYN;
1044                         } else
1045                                 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1046
1047                         tcpstat.tcps_connects++;
1048                         soisconnected(so);
1049                         goto trimthenstep6;
1050                 }
1051                 goto drop;
1052         }
1053 after_listen:
1054
1055         /* should not happen - syncache should pick up these connections */
1056         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1057
1058         /*
1059          * Segment received on connection.
1060          * Reset idle time and keep-alive timer.
1061          */
1062         tp->t_rcvtime = ticks;
1063         if (TCPS_HAVEESTABLISHED(tp->t_state))
1064                 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1065
1066         /*
1067          * Process options.
1068          * XXX this is tradtitional behavior, may need to be cleaned up.
1069          */
1070         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1071         if (thflags & TH_SYN) {
1072                 if (to.to_flags & TOF_SCALE) {
1073                         tp->t_flags |= TF_RCVD_SCALE;
1074                         tp->requested_s_scale = to.to_requested_s_scale;
1075                 }
1076                 if (to.to_flags & TOF_TS) {
1077                         tp->t_flags |= TF_RCVD_TSTMP;
1078                         tp->ts_recent = to.to_tsval;
1079                         tp->ts_recent_age = ticks;
1080                 }
1081                 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1082                         tp->t_flags |= TF_RCVD_CC;
1083                 if (to.to_flags & TOF_MSS)
1084                         tcp_mss(tp, to.to_mss);
1085                 /*
1086                  * Only set the TF_SACK_PERMITTED per-connection flag
1087                  * if we got a SACK_PERMITTED option from the other side
1088                  * and the global tcp_do_sack variable is true.
1089                  */
1090                 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1091                         tp->t_flags |= TF_SACK_PERMITTED;
1092         }
1093
1094         /*
1095          * Header prediction: check for the two common cases
1096          * of a uni-directional data xfer.  If the packet has
1097          * no control flags, is in-sequence, the window didn't
1098          * change and we're not retransmitting, it's a
1099          * candidate.  If the length is zero and the ack moved
1100          * forward, we're the sender side of the xfer.  Just
1101          * free the data acked & wake any higher level process
1102          * that was blocked waiting for space.  If the length
1103          * is non-zero and the ack didn't move, we're the
1104          * receiver side.  If we're getting packets in-order
1105          * (the reassembly queue is empty), add the data to
1106          * the socket buffer and note that we need a delayed ack.
1107          * Make sure that the hidden state-flags are also off.
1108          * Since we check for TCPS_ESTABLISHED above, it can only
1109          * be TH_NEEDSYN.
1110          */
1111         if (tp->t_state == TCPS_ESTABLISHED &&
1112             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1113             !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1114             (!(to.to_flags & TOF_TS) ||
1115              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1116             /*
1117              * Using the CC option is compulsory if once started:
1118              *   the segment is OK if no T/TCP was negotiated or
1119              *   if the segment has a CC option equal to CCrecv
1120              */
1121             ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1122              ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1123             th->th_seq == tp->rcv_nxt &&
1124             tp->snd_nxt == tp->snd_max) {
1125
1126                 /*
1127                  * If last ACK falls within this segment's sequence numbers,
1128                  * record the timestamp.
1129                  * NOTE that the test is modified according to the latest
1130                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1131                  */
1132                 if ((to.to_flags & TOF_TS) &&
1133                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1134                         tp->ts_recent_age = ticks;
1135                         tp->ts_recent = to.to_tsval;
1136                 }
1137
1138                 if (tlen == 0) {
1139                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1140                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1141                             tp->snd_cwnd >= tp->snd_wnd &&
1142                             !IN_FASTRECOVERY(tp)) {
1143                                 /*
1144                                  * this is a pure ack for outstanding data.
1145                                  */
1146                                 ++tcpstat.tcps_predack;
1147                                 /*
1148                                  * "bad retransmit" recovery
1149                                  *
1150                                  * If Eifel detection applies, then
1151                                  * it is deterministic, so use it
1152                                  * unconditionally over the old heuristic.
1153                                  * Otherwise, fall back to the old heuristic.
1154                                  */
1155                                 if (tcp_do_eifel_detect &&
1156                                     (to.to_flags & TOF_TS) && to.to_tsecr &&
1157                                     (tp->t_flags & TF_FIRSTACCACK)) {
1158                                         /* Eifel detection applicable. */
1159                                         if (to.to_tsecr < tp->t_rexmtTS) {
1160                                                 tcp_revert_congestion_state(tp);
1161                                                 ++tcpstat.tcps_eifeldetected;
1162                                         }
1163                                 } else if (tp->t_rxtshift == 1 &&
1164                                            ticks < tp->t_badrxtwin) {
1165                                         tcp_revert_congestion_state(tp);
1166                                         ++tcpstat.tcps_rttdetected;
1167                                 }
1168                                 tp->t_flags &= ~(TF_FIRSTACCACK |
1169                                                  TF_FASTREXMT | TF_EARLYREXMT);
1170                                 /*
1171                                  * Recalculate the retransmit timer / rtt.
1172                                  *
1173                                  * Some machines (certain windows boxes)
1174                                  * send broken timestamp replies during the
1175                                  * SYN+ACK phase, ignore timestamps of 0.
1176                                  */
1177                                 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1178                                         tcp_xmit_timer(tp,
1179                                                        ticks - to.to_tsecr + 1);
1180                                 } else if (tp->t_rtttime &&
1181                                            SEQ_GT(th->th_ack, tp->t_rtseq)) {
1182                                         tcp_xmit_timer(tp,
1183                                                        ticks - tp->t_rtttime);
1184                                 }
1185                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1186                                 acked = th->th_ack - tp->snd_una;
1187                                 tcpstat.tcps_rcvackpack++;
1188                                 tcpstat.tcps_rcvackbyte += acked;
1189                                 sbdrop(&so->so_snd, acked);
1190                                 tp->snd_recover = th->th_ack - 1;
1191                                 tp->snd_una = th->th_ack;
1192                                 tp->t_dupacks = 0;
1193                                 /*
1194                                  * Update window information.
1195                                  */
1196                                 if (tiwin != tp->snd_wnd &&
1197                                     acceptable_window_update(tp, th, tiwin)) {
1198                                         /* keep track of pure window updates */
1199                                         if (tp->snd_wl2 == th->th_ack &&
1200                                             tiwin > tp->snd_wnd)
1201                                                 tcpstat.tcps_rcvwinupd++;
1202                                         tp->snd_wnd = tiwin;
1203                                         tp->snd_wl1 = th->th_seq;
1204                                         tp->snd_wl2 = th->th_ack;
1205                                         if (tp->snd_wnd > tp->max_sndwnd)
1206                                                 tp->max_sndwnd = tp->snd_wnd;
1207                                 }
1208                                 m_freem(m);
1209                                 ND6_HINT(tp); /* some progress has been done */
1210                                 /*
1211                                  * If all outstanding data are acked, stop
1212                                  * retransmit timer, otherwise restart timer
1213                                  * using current (possibly backed-off) value.
1214                                  * If process is waiting for space,
1215                                  * wakeup/selwakeup/signal.  If data
1216                                  * are ready to send, let tcp_output
1217                                  * decide between more output or persist.
1218                                  */
1219                                 if (tp->snd_una == tp->snd_max)
1220                                         callout_stop(tp->tt_rexmt);
1221                                 else if (!callout_active(tp->tt_persist))
1222                                         callout_reset(tp->tt_rexmt,
1223                                                       tp->t_rxtcur,
1224                                                       tcp_timer_rexmt, tp);
1225
1226                                 sowwakeup(so);
1227                                 if (so->so_snd.sb_cc)
1228                                         (void) tcp_output(tp);
1229                                 return;
1230                         }
1231                 } else if (tiwin == tp->snd_wnd &&
1232                     th->th_ack == tp->snd_una &&
1233                     LIST_EMPTY(&tp->t_segq) &&
1234                     tlen <= sbspace(&so->so_rcv)) {
1235                         /*
1236                          * this is a pure, in-sequence data packet
1237                          * with nothing on the reassembly queue and
1238                          * we have enough buffer space to take it.
1239                          */
1240                         ++tcpstat.tcps_preddat;
1241                         tp->rcv_nxt += tlen;
1242                         tcpstat.tcps_rcvpack++;
1243                         tcpstat.tcps_rcvbyte += tlen;
1244                         ND6_HINT(tp);   /* some progress has been done */
1245                         /*
1246                          * Add data to socket buffer.
1247                          */
1248                         if (so->so_state & SS_CANTRCVMORE) {
1249                                 m_freem(m);
1250                         } else {
1251                                 m_adj(m, drop_hdrlen); /* delayed header drop */
1252                                 sbappendstream(&so->so_rcv, m);
1253                         }
1254                         sorwakeup(so);
1255
1256                         /*
1257                          * This code is responsible for most of the ACKs
1258                          * the TCP stack sends back after receiving a data
1259                          * packet.  Note that the DELAY_ACK check fails if
1260                          * the delack timer is already running, which results
1261                          * in an ack being sent every other packet (which is
1262                          * what we want).
1263                          */
1264                         if (DELAY_ACK(tp)) {
1265                                 callout_reset(tp->tt_delack, tcp_delacktime,
1266                                     tcp_timer_delack, tp);
1267                         } else {
1268                                 tp->t_flags |= TF_ACKNOW;
1269                                 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1270                                         tp->t_flags |= TF_ONOUTPUTQ;
1271                                         tp->tt_cpu = mycpu->gd_cpuid;
1272                                         TAILQ_INSERT_TAIL(
1273                                             &tcpcbackq[tp->tt_cpu],
1274                                             tp, t_outputq);
1275                                 }
1276                         }
1277                         return;
1278                 }
1279         }
1280
1281         /*
1282          * Calculate amount of space in receive window,
1283          * and then do TCP input processing.
1284          * Receive window is amount of space in rcv queue,
1285          * but not less than advertised window.
1286          */
1287         recvwin = sbspace(&so->so_rcv);
1288         if (recvwin < 0)
1289                 recvwin = 0;
1290         tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1291
1292         switch (tp->t_state) {
1293         /*
1294          * If the state is SYN_RECEIVED:
1295          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1296          */
1297         case TCPS_SYN_RECEIVED:
1298                 if ((thflags & TH_ACK) &&
1299                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1300                      SEQ_GT(th->th_ack, tp->snd_max))) {
1301                                 rstreason = BANDLIM_RST_OPENPORT;
1302                                 goto dropwithreset;
1303                 }
1304                 break;
1305
1306         /*
1307          * If the state is SYN_SENT:
1308          *      if seg contains an ACK, but not for our SYN, drop the input.
1309          *      if seg contains a RST, then drop the connection.
1310          *      if seg does not contain SYN, then drop it.
1311          * Otherwise this is an acceptable SYN segment
1312          *      initialize tp->rcv_nxt and tp->irs
1313          *      if seg contains ack then advance tp->snd_una
1314          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1315          *      arrange for segment to be acked (eventually)
1316          *      continue processing rest of data/controls, beginning with URG
1317          */
1318         case TCPS_SYN_SENT:
1319                 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1320                         taop = &tao_noncached;
1321                         bzero(taop, sizeof(*taop));
1322                 }
1323
1324                 if ((thflags & TH_ACK) &&
1325                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1326                      SEQ_GT(th->th_ack, tp->snd_max))) {
1327                         /*
1328                          * If we have a cached CCsent for the remote host,
1329                          * hence we haven't just crashed and restarted,
1330                          * do not send a RST.  This may be a retransmission
1331                          * from the other side after our earlier ACK was lost.
1332                          * Our new SYN, when it arrives, will serve as the
1333                          * needed ACK.
1334                          */
1335                         if (taop->tao_ccsent != 0)
1336                                 goto drop;
1337                         else {
1338                                 rstreason = BANDLIM_UNLIMITED;
1339                                 goto dropwithreset;
1340                         }
1341                 }
1342                 if (thflags & TH_RST) {
1343                         if (thflags & TH_ACK)
1344                                 tp = tcp_drop(tp, ECONNREFUSED);
1345                         goto drop;
1346                 }
1347                 if (!(thflags & TH_SYN))
1348                         goto drop;
1349                 tp->snd_wnd = th->th_win;       /* initial send window */
1350                 tp->cc_recv = to.to_cc;         /* foreign CC */
1351
1352                 tp->irs = th->th_seq;
1353                 tcp_rcvseqinit(tp);
1354                 if (thflags & TH_ACK) {
1355                         /*
1356                          * Our SYN was acked.  If segment contains CC.ECHO
1357                          * option, check it to make sure this segment really
1358                          * matches our SYN.  If not, just drop it as old
1359                          * duplicate, but send an RST if we're still playing
1360                          * by the old rules.  If no CC.ECHO option, make sure
1361                          * we don't get fooled into using T/TCP.
1362                          */
1363                         if (to.to_flags & TOF_CCECHO) {
1364                                 if (tp->cc_send != to.to_ccecho) {
1365                                         if (taop->tao_ccsent != 0)
1366                                                 goto drop;
1367                                         else {
1368                                                 rstreason = BANDLIM_UNLIMITED;
1369                                                 goto dropwithreset;
1370                                         }
1371                                 }
1372                         } else
1373                                 tp->t_flags &= ~TF_RCVD_CC;
1374                         tcpstat.tcps_connects++;
1375                         soisconnected(so);
1376                         /* Do window scaling on this connection? */
1377                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1378                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1379                                 tp->snd_scale = tp->requested_s_scale;
1380                                 tp->rcv_scale = tp->request_r_scale;
1381                         }
1382                         /* Segment is acceptable, update cache if undefined. */
1383                         if (taop->tao_ccsent == 0)
1384                                 taop->tao_ccsent = to.to_ccecho;
1385
1386                         tp->rcv_adv += tp->rcv_wnd;
1387                         tp->snd_una++;          /* SYN is acked */
1388                         callout_stop(tp->tt_rexmt);
1389                         /*
1390                          * If there's data, delay ACK; if there's also a FIN
1391                          * ACKNOW will be turned on later.
1392                          */
1393                         if (DELAY_ACK(tp) && tlen != 0)
1394                                 callout_reset(tp->tt_delack, tcp_delacktime,
1395                                     tcp_timer_delack, tp);
1396                         else
1397                                 tp->t_flags |= TF_ACKNOW;
1398                         /*
1399                          * Received <SYN,ACK> in SYN_SENT[*] state.
1400                          * Transitions:
1401                          *      SYN_SENT  --> ESTABLISHED
1402                          *      SYN_SENT* --> FIN_WAIT_1
1403                          */
1404                         tp->t_starttime = ticks;
1405                         if (tp->t_flags & TF_NEEDFIN) {
1406                                 tp->t_state = TCPS_FIN_WAIT_1;
1407                                 tp->t_flags &= ~TF_NEEDFIN;
1408                                 thflags &= ~TH_SYN;
1409                         } else {
1410                                 tp->t_state = TCPS_ESTABLISHED;
1411                                 callout_reset(tp->tt_keep, tcp_keepidle,
1412                                               tcp_timer_keep, tp);
1413                         }
1414                 } else {
1415                         /*
1416                          * Received initial SYN in SYN-SENT[*] state =>
1417                          * simultaneous open.  If segment contains CC option
1418                          * and there is a cached CC, apply TAO test.
1419                          * If it succeeds, connection is * half-synchronized.
1420                          * Otherwise, do 3-way handshake:
1421                          *        SYN-SENT -> SYN-RECEIVED
1422                          *        SYN-SENT* -> SYN-RECEIVED*
1423                          * If there was no CC option, clear cached CC value.
1424                          */
1425                         tp->t_flags |= TF_ACKNOW;
1426                         callout_stop(tp->tt_rexmt);
1427                         if (to.to_flags & TOF_CC) {
1428                                 if (taop->tao_cc != 0 &&
1429                                     CC_GT(to.to_cc, taop->tao_cc)) {
1430                                         /*
1431                                          * update cache and make transition:
1432                                          *        SYN-SENT -> ESTABLISHED*
1433                                          *        SYN-SENT* -> FIN-WAIT-1*
1434                                          */
1435                                         taop->tao_cc = to.to_cc;
1436                                         tp->t_starttime = ticks;
1437                                         if (tp->t_flags & TF_NEEDFIN) {
1438                                                 tp->t_state = TCPS_FIN_WAIT_1;
1439                                                 tp->t_flags &= ~TF_NEEDFIN;
1440                                         } else {
1441                                                 tp->t_state = TCPS_ESTABLISHED;
1442                                                 callout_reset(tp->tt_keep,
1443                                                               tcp_keepidle,
1444                                                               tcp_timer_keep,
1445                                                               tp);
1446                                         }
1447                                         tp->t_flags |= TF_NEEDSYN;
1448                                 } else
1449                                         tp->t_state = TCPS_SYN_RECEIVED;
1450                         } else {
1451                                 /* CC.NEW or no option => invalidate cache */
1452                                 taop->tao_cc = 0;
1453                                 tp->t_state = TCPS_SYN_RECEIVED;
1454                         }
1455                 }
1456
1457 trimthenstep6:
1458                 /*
1459                  * Advance th->th_seq to correspond to first data byte.
1460                  * If data, trim to stay within window,
1461                  * dropping FIN if necessary.
1462                  */
1463                 th->th_seq++;
1464                 if (tlen > tp->rcv_wnd) {
1465                         todrop = tlen - tp->rcv_wnd;
1466                         m_adj(m, -todrop);
1467                         tlen = tp->rcv_wnd;
1468                         thflags &= ~TH_FIN;
1469                         tcpstat.tcps_rcvpackafterwin++;
1470                         tcpstat.tcps_rcvbyteafterwin += todrop;
1471                 }
1472                 tp->snd_wl1 = th->th_seq - 1;
1473                 tp->rcv_up = th->th_seq;
1474                 /*
1475                  * Client side of transaction: already sent SYN and data.
1476                  * If the remote host used T/TCP to validate the SYN,
1477                  * our data will be ACK'd; if so, enter normal data segment
1478                  * processing in the middle of step 5, ack processing.
1479                  * Otherwise, goto step 6.
1480                  */
1481                 if (thflags & TH_ACK)
1482                         goto process_ACK;
1483
1484                 goto step6;
1485
1486         /*
1487          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1488          *      if segment contains a SYN and CC [not CC.NEW] option:
1489          *              if state == TIME_WAIT and connection duration > MSL,
1490          *                  drop packet and send RST;
1491          *
1492          *              if SEG.CC > CCrecv then is new SYN, and can implicitly
1493          *                  ack the FIN (and data) in retransmission queue.
1494          *                  Complete close and delete TCPCB.  Then reprocess
1495          *                  segment, hoping to find new TCPCB in LISTEN state;
1496          *
1497          *              else must be old SYN; drop it.
1498          *      else do normal processing.
1499          */
1500         case TCPS_LAST_ACK:
1501         case TCPS_CLOSING:
1502         case TCPS_TIME_WAIT:
1503                 if ((thflags & TH_SYN) &&
1504                     (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1505                         if (tp->t_state == TCPS_TIME_WAIT &&
1506                                         (ticks - tp->t_starttime) > tcp_msl) {
1507                                 rstreason = BANDLIM_UNLIMITED;
1508                                 goto dropwithreset;
1509                         }
1510                         if (CC_GT(to.to_cc, tp->cc_recv)) {
1511                                 tp = tcp_close(tp);
1512                                 goto findpcb;
1513                         }
1514                         else
1515                                 goto drop;
1516                 }
1517                 break;  /* continue normal processing */
1518         }
1519
1520         /*
1521          * States other than LISTEN or SYN_SENT.
1522          * First check the RST flag and sequence number since reset segments
1523          * are exempt from the timestamp and connection count tests.  This
1524          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1525          * below which allowed reset segments in half the sequence space
1526          * to fall though and be processed (which gives forged reset
1527          * segments with a random sequence number a 50 percent chance of
1528          * killing a connection).
1529          * Then check timestamp, if present.
1530          * Then check the connection count, if present.
1531          * Then check that at least some bytes of segment are within
1532          * receive window.  If segment begins before rcv_nxt,
1533          * drop leading data (and SYN); if nothing left, just ack.
1534          *
1535          *
1536          * If the RST bit is set, check the sequence number to see
1537          * if this is a valid reset segment.
1538          * RFC 793 page 37:
1539          *   In all states except SYN-SENT, all reset (RST) segments
1540          *   are validated by checking their SEQ-fields.  A reset is
1541          *   valid if its sequence number is in the window.
1542          * Note: this does not take into account delayed ACKs, so
1543          *   we should test against last_ack_sent instead of rcv_nxt.
1544          *   The sequence number in the reset segment is normally an
1545          *   echo of our outgoing acknowledgement numbers, but some hosts
1546          *   send a reset with the sequence number at the rightmost edge
1547          *   of our receive window, and we have to handle this case.
1548          * If we have multiple segments in flight, the intial reset
1549          * segment sequence numbers will be to the left of last_ack_sent,
1550          * but they will eventually catch up.
1551          * In any case, it never made sense to trim reset segments to
1552          * fit the receive window since RFC 1122 says:
1553          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1554          *
1555          *    A TCP SHOULD allow a received RST segment to include data.
1556          *
1557          *    DISCUSSION
1558          *         It has been suggested that a RST segment could contain
1559          *         ASCII text that encoded and explained the cause of the
1560          *         RST.  No standard has yet been established for such
1561          *         data.
1562          *
1563          * If the reset segment passes the sequence number test examine
1564          * the state:
1565          *    SYN_RECEIVED STATE:
1566          *      If passive open, return to LISTEN state.
1567          *      If active open, inform user that connection was refused.
1568          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1569          *      Inform user that connection was reset, and close tcb.
1570          *    CLOSING, LAST_ACK STATES:
1571          *      Close the tcb.
1572          *    TIME_WAIT STATE:
1573          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1574          *      RFC 1337.
1575          */
1576         if (thflags & TH_RST) {
1577                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1578                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1579                         switch (tp->t_state) {
1580
1581                         case TCPS_SYN_RECEIVED:
1582                                 so->so_error = ECONNREFUSED;
1583                                 goto close;
1584
1585                         case TCPS_ESTABLISHED:
1586                         case TCPS_FIN_WAIT_1:
1587                         case TCPS_FIN_WAIT_2:
1588                         case TCPS_CLOSE_WAIT:
1589                                 so->so_error = ECONNRESET;
1590                         close:
1591                                 tp->t_state = TCPS_CLOSED;
1592                                 tcpstat.tcps_drops++;
1593                                 tp = tcp_close(tp);
1594                                 break;
1595
1596                         case TCPS_CLOSING:
1597                         case TCPS_LAST_ACK:
1598                                 tp = tcp_close(tp);
1599                                 break;
1600
1601                         case TCPS_TIME_WAIT:
1602                                 break;
1603                         }
1604                 }
1605                 goto drop;
1606         }
1607
1608         /*
1609          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1610          * and it's less than ts_recent, drop it.
1611          */
1612         if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1613             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1614
1615                 /* Check to see if ts_recent is over 24 days old.  */
1616                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1617                         /*
1618                          * Invalidate ts_recent.  If this segment updates
1619                          * ts_recent, the age will be reset later and ts_recent
1620                          * will get a valid value.  If it does not, setting
1621                          * ts_recent to zero will at least satisfy the
1622                          * requirement that zero be placed in the timestamp
1623                          * echo reply when ts_recent isn't valid.  The
1624                          * age isn't reset until we get a valid ts_recent
1625                          * because we don't want out-of-order segments to be
1626                          * dropped when ts_recent is old.
1627                          */
1628                         tp->ts_recent = 0;
1629                 } else {
1630                         tcpstat.tcps_rcvduppack++;
1631                         tcpstat.tcps_rcvdupbyte += tlen;
1632                         tcpstat.tcps_pawsdrop++;
1633                         if (tlen)
1634                                 goto dropafterack;
1635                         goto drop;
1636                 }
1637         }
1638
1639         /*
1640          * T/TCP mechanism
1641          *   If T/TCP was negotiated and the segment doesn't have CC,
1642          *   or if its CC is wrong then drop the segment.
1643          *   RST segments do not have to comply with this.
1644          */
1645         if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1646             (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1647                 goto dropafterack;
1648
1649         /*
1650          * In the SYN-RECEIVED state, validate that the packet belongs to
1651          * this connection before trimming the data to fit the receive
1652          * window.  Check the sequence number versus IRS since we know
1653          * the sequence numbers haven't wrapped.  This is a partial fix
1654          * for the "LAND" DoS attack.
1655          */
1656         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1657                 rstreason = BANDLIM_RST_OPENPORT;
1658                 goto dropwithreset;
1659         }
1660
1661         todrop = tp->rcv_nxt - th->th_seq;
1662         if (todrop > 0) {
1663                 if (TCP_DO_SACK(tp)) {
1664                         /* Report duplicate segment at head of packet. */
1665                         tp->reportblk.rblk_start = th->th_seq;
1666                         tp->reportblk.rblk_end = th->th_seq + tlen;
1667                         if (thflags & TH_FIN)
1668                                 ++tp->reportblk.rblk_end;
1669                         if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1670                                 tp->reportblk.rblk_end = tp->rcv_nxt;
1671                         tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1672                 }
1673                 if (thflags & TH_SYN) {
1674                         thflags &= ~TH_SYN;
1675                         th->th_seq++;
1676                         if (th->th_urp > 1)
1677                                 th->th_urp--;
1678                         else
1679                                 thflags &= ~TH_URG;
1680                         todrop--;
1681                 }
1682                 /*
1683                  * Following if statement from Stevens, vol. 2, p. 960.
1684                  */
1685                 if (todrop > tlen ||
1686                     (todrop == tlen && !(thflags & TH_FIN))) {
1687                         /*
1688                          * Any valid FIN must be to the left of the window.
1689                          * At this point the FIN must be a duplicate or out
1690                          * of sequence; drop it.
1691                          */
1692                         thflags &= ~TH_FIN;
1693
1694                         /*
1695                          * Send an ACK to resynchronize and drop any data.
1696                          * But keep on processing for RST or ACK.
1697                          */
1698                         tp->t_flags |= TF_ACKNOW;
1699                         todrop = tlen;
1700                         tcpstat.tcps_rcvduppack++;
1701                         tcpstat.tcps_rcvdupbyte += todrop;
1702                 } else {
1703                         tcpstat.tcps_rcvpartduppack++;
1704                         tcpstat.tcps_rcvpartdupbyte += todrop;
1705                 }
1706                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1707                 th->th_seq += todrop;
1708                 tlen -= todrop;
1709                 if (th->th_urp > todrop)
1710                         th->th_urp -= todrop;
1711                 else {
1712                         thflags &= ~TH_URG;
1713                         th->th_urp = 0;
1714                 }
1715         }
1716
1717         /*
1718          * If new data are received on a connection after the
1719          * user processes are gone, then RST the other end.
1720          */
1721         if ((so->so_state & SS_NOFDREF) &&
1722             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1723                 tp = tcp_close(tp);
1724                 tcpstat.tcps_rcvafterclose++;
1725                 rstreason = BANDLIM_UNLIMITED;
1726                 goto dropwithreset;
1727         }
1728
1729         /*
1730          * If segment ends after window, drop trailing data
1731          * (and PUSH and FIN); if nothing left, just ACK.
1732          */
1733         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1734         if (todrop > 0) {
1735                 tcpstat.tcps_rcvpackafterwin++;
1736                 if (todrop >= tlen) {
1737                         tcpstat.tcps_rcvbyteafterwin += tlen;
1738                         /*
1739                          * If a new connection request is received
1740                          * while in TIME_WAIT, drop the old connection
1741                          * and start over if the sequence numbers
1742                          * are above the previous ones.
1743                          */
1744                         if (thflags & TH_SYN &&
1745                             tp->t_state == TCPS_TIME_WAIT &&
1746                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1747                                 tp = tcp_close(tp);
1748                                 goto findpcb;
1749                         }
1750                         /*
1751                          * If window is closed can only take segments at
1752                          * window edge, and have to drop data and PUSH from
1753                          * incoming segments.  Continue processing, but
1754                          * remember to ack.  Otherwise, drop segment
1755                          * and ack.
1756                          */
1757                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1758                                 tp->t_flags |= TF_ACKNOW;
1759                                 tcpstat.tcps_rcvwinprobe++;
1760                         } else
1761                                 goto dropafterack;
1762                 } else
1763                         tcpstat.tcps_rcvbyteafterwin += todrop;
1764                 m_adj(m, -todrop);
1765                 tlen -= todrop;
1766                 thflags &= ~(TH_PUSH | TH_FIN);
1767         }
1768
1769         /*
1770          * If last ACK falls within this segment's sequence numbers,
1771          * record its timestamp.
1772          * NOTE that the test is modified according to the latest
1773          * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1774          */
1775         if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1776                 tp->ts_recent_age = ticks;
1777                 tp->ts_recent = to.to_tsval;
1778         }
1779
1780         /*
1781          * If a SYN is in the window, then this is an
1782          * error and we send an RST and drop the connection.
1783          */
1784         if (thflags & TH_SYN) {
1785                 tp = tcp_drop(tp, ECONNRESET);
1786                 rstreason = BANDLIM_UNLIMITED;
1787                 goto dropwithreset;
1788         }
1789
1790         /*
1791          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1792          * flag is on (half-synchronized state), then queue data for
1793          * later processing; else drop segment and return.
1794          */
1795         if (!(thflags & TH_ACK)) {
1796                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1797                     (tp->t_flags & TF_NEEDSYN))
1798                         goto step6;
1799                 else
1800                         goto drop;
1801         }
1802
1803         /*
1804          * Ack processing.
1805          */
1806         switch (tp->t_state) {
1807         /*
1808          * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1809          * ESTABLISHED state and continue processing.
1810          * The ACK was checked above.
1811          */
1812         case TCPS_SYN_RECEIVED:
1813
1814                 tcpstat.tcps_connects++;
1815                 soisconnected(so);
1816                 /* Do window scaling? */
1817                 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1818                     (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1819                         tp->snd_scale = tp->requested_s_scale;
1820                         tp->rcv_scale = tp->request_r_scale;
1821                 }
1822                 /*
1823                  * Upon successful completion of 3-way handshake,
1824                  * update cache.CC if it was undefined, pass any queued
1825                  * data to the user, and advance state appropriately.
1826                  */
1827                 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1828                     taop->tao_cc == 0)
1829                         taop->tao_cc = tp->cc_recv;
1830
1831                 /*
1832                  * Make transitions:
1833                  *      SYN-RECEIVED  -> ESTABLISHED
1834                  *      SYN-RECEIVED* -> FIN-WAIT-1
1835                  */
1836                 tp->t_starttime = ticks;
1837                 if (tp->t_flags & TF_NEEDFIN) {
1838                         tp->t_state = TCPS_FIN_WAIT_1;
1839                         tp->t_flags &= ~TF_NEEDFIN;
1840                 } else {
1841                         tp->t_state = TCPS_ESTABLISHED;
1842                         callout_reset(tp->tt_keep, tcp_keepidle,
1843                                       tcp_timer_keep, tp);
1844                 }
1845                 /*
1846                  * If segment contains data or ACK, will call tcp_reass()
1847                  * later; if not, do so now to pass queued data to user.
1848                  */
1849                 if (tlen == 0 && !(thflags & TH_FIN))
1850                         (void) tcp_reass(tp, NULL, NULL, NULL);
1851                 /* fall into ... */
1852
1853         /*
1854          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1855          * ACKs.  If the ack is in the range
1856          *      tp->snd_una < th->th_ack <= tp->snd_max
1857          * then advance tp->snd_una to th->th_ack and drop
1858          * data from the retransmission queue.  If this ACK reflects
1859          * more up to date window information we update our window information.
1860          */
1861         case TCPS_ESTABLISHED:
1862         case TCPS_FIN_WAIT_1:
1863         case TCPS_FIN_WAIT_2:
1864         case TCPS_CLOSE_WAIT:
1865         case TCPS_CLOSING:
1866         case TCPS_LAST_ACK:
1867         case TCPS_TIME_WAIT:
1868
1869                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1870                         if (TCP_DO_SACK(tp))
1871                                 tcp_sack_update_scoreboard(tp, &to);
1872                         if (tlen != 0 || tiwin != tp->snd_wnd) {
1873                                 tp->t_dupacks = 0;
1874                                 break;
1875                         }
1876                         tcpstat.tcps_rcvdupack++;
1877                         if (!callout_active(tp->tt_rexmt) ||
1878                             th->th_ack != tp->snd_una) {
1879                                 tp->t_dupacks = 0;
1880                                 break;
1881                         }
1882                         /*
1883                          * We have outstanding data (other than
1884                          * a window probe), this is a completely
1885                          * duplicate ack (ie, window info didn't
1886                          * change), the ack is the biggest we've
1887                          * seen and we've seen exactly our rexmt
1888                          * threshhold of them, so assume a packet
1889                          * has been dropped and retransmit it.
1890                          * Kludge snd_nxt & the congestion
1891                          * window so we send only this one
1892                          * packet.
1893                          */
1894                         if (IN_FASTRECOVERY(tp)) {
1895                                 if (TCP_DO_SACK(tp)) {
1896                                         /* No artifical cwnd inflation. */
1897                                         tcp_sack_rexmt(tp, th);
1898                                 } else {
1899                                         /*
1900                                          * Dup acks mean that packets
1901                                          * have left the network
1902                                          * (they're now cached at the
1903                                          * receiver) so bump cwnd by
1904                                          * the amount in the receiver
1905                                          * to keep a constant cwnd
1906                                          * packets in the network.
1907                                          */
1908                                         tp->snd_cwnd += tp->t_maxseg;
1909                                         (void) tcp_output(tp);
1910                                 }
1911                         } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1912                                 tp->t_dupacks = 0;
1913                                 break;
1914                         } else if (++tp->t_dupacks == tcprexmtthresh) {
1915                                 tcp_seq old_snd_nxt;
1916                                 u_int win;
1917
1918 fastretransmit:
1919                                 if (tcp_do_eifel_detect &&
1920                                     (tp->t_flags & TF_RCVD_TSTMP)) {
1921                                         tcp_save_congestion_state(tp);
1922                                         tp->t_flags |= TF_FASTREXMT;
1923                                 }
1924                                 /*
1925                                  * We know we're losing at the current
1926                                  * window size, so do congestion avoidance:
1927                                  * set ssthresh to half the current window
1928                                  * and pull our congestion window back to the
1929                                  * new ssthresh.
1930                                  */
1931                                 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1932                                     tp->t_maxseg;
1933                                 if (win < 2)
1934                                         win = 2;
1935                                 tp->snd_ssthresh = win * tp->t_maxseg;
1936                                 ENTER_FASTRECOVERY(tp);
1937                                 tp->snd_recover = tp->snd_max;
1938                                 callout_stop(tp->tt_rexmt);
1939                                 tp->t_rtttime = 0;
1940                                 old_snd_nxt = tp->snd_nxt;
1941                                 tp->snd_nxt = th->th_ack;
1942                                 tp->snd_cwnd = tp->t_maxseg;
1943                                 (void) tcp_output(tp);
1944                                 ++tcpstat.tcps_sndfastrexmit;
1945                                 tp->snd_cwnd = tp->snd_ssthresh;
1946                                 tp->rexmt_high = tp->snd_nxt;
1947                                 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1948                                         tp->snd_nxt = old_snd_nxt;
1949                                 KASSERT(tp->snd_limited <= 2,
1950                                     ("tp->snd_limited too big"));
1951                                 if (TCP_DO_SACK(tp))
1952                                         tcp_sack_rexmt(tp, th);
1953                                 else
1954                                         tp->snd_cwnd += tp->t_maxseg *
1955                                             (tp->t_dupacks - tp->snd_limited);
1956                         } else if (tcp_do_limitedtransmit) {
1957                                 u_long oldcwnd = tp->snd_cwnd;
1958                                 tcp_seq oldsndmax = tp->snd_max;
1959                                 /* outstanding data */
1960                                 uint32_t ownd = tp->snd_max - tp->snd_una;
1961                                 u_int sent;
1962
1963 #define iceildiv(n, d)          (((n)+(d)-1) / (d))
1964
1965                                 KASSERT(tp->t_dupacks == 1 ||
1966                                         tp->t_dupacks == 2,
1967                                     ("dupacks not 1 or 2"));
1968                                 if (tp->t_dupacks == 1)
1969                                         tp->snd_limited = 0;
1970                                 tp->snd_cwnd = ownd +
1971                                     (tp->t_dupacks - tp->snd_limited) *
1972                                     tp->t_maxseg;
1973                                 (void) tcp_output(tp);
1974                                 tp->snd_cwnd = oldcwnd;
1975                                 sent = tp->snd_max - oldsndmax;
1976                                 if (sent > tp->t_maxseg) {
1977                                         KASSERT((tp->t_dupacks == 2 &&
1978                                                  tp->snd_limited == 0) ||
1979                                                 (sent == tp->t_maxseg + 1 &&
1980                                                  tp->t_flags & TF_SENTFIN),
1981                                             ("sent too much"));
1982                                         KASSERT(sent <= tp->t_maxseg * 2,
1983                                             ("sent too many segments"));
1984                                         tp->snd_limited = 2;
1985                                         tcpstat.tcps_sndlimited += 2;
1986                                 } else if (sent > 0) {
1987                                         ++tp->snd_limited;
1988                                         ++tcpstat.tcps_sndlimited;
1989                                 } else if (tcp_do_early_retransmit &&
1990                                     (tcp_do_eifel_detect &&
1991                                      (tp->t_flags & TF_RCVD_TSTMP)) &&
1992                                     ownd < 4 * tp->t_maxseg &&
1993                                     tp->t_dupacks + 1 >=
1994                                       iceildiv(ownd, tp->t_maxseg) &&
1995                                     (!TCP_DO_SACK(tp) ||
1996                                      ownd <= tp->t_maxseg || 
1997                                      tcp_sack_has_sacked(&tp->scb,
1998                                          ownd - tp->t_maxseg))) {
1999                                         ++tcpstat.tcps_sndearlyrexmit;
2000                                         tp->t_flags |= TF_EARLYREXMT;
2001                                         goto fastretransmit;
2002                                 }
2003                         }
2004                         goto drop;
2005                 }
2006
2007                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2008                 tp->t_dupacks = 0;
2009                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2010                         /*
2011                          * Detected optimistic ACK attack.
2012                          * Force slow-start to de-synchronize attack.
2013                          */
2014                         tp->snd_cwnd = tp->t_maxseg;
2015
2016                         tcpstat.tcps_rcvacktoomuch++;
2017                         goto dropafterack;
2018                 }
2019                 /*
2020                  * If we reach this point, ACK is not a duplicate,
2021                  *     i.e., it ACKs something we sent.
2022                  */
2023                 if (tp->t_flags & TF_NEEDSYN) {
2024                         /*
2025                          * T/TCP: Connection was half-synchronized, and our
2026                          * SYN has been ACK'd (so connection is now fully
2027                          * synchronized).  Go to non-starred state,
2028                          * increment snd_una for ACK of SYN, and check if
2029                          * we can do window scaling.
2030                          */
2031                         tp->t_flags &= ~TF_NEEDSYN;
2032                         tp->snd_una++;
2033                         /* Do window scaling? */
2034                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2035                             (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2036                                 tp->snd_scale = tp->requested_s_scale;
2037                                 tp->rcv_scale = tp->request_r_scale;
2038                         }
2039                 }
2040
2041 process_ACK:
2042                 acked = th->th_ack - tp->snd_una;
2043                 tcpstat.tcps_rcvackpack++;
2044                 tcpstat.tcps_rcvackbyte += acked;
2045
2046                 if (tcp_do_eifel_detect && acked > 0 &&
2047                     (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2048                     (tp->t_flags & TF_FIRSTACCACK)) {
2049                         /* Eifel detection applicable. */
2050                         if (to.to_tsecr < tp->t_rexmtTS) {
2051                                 ++tcpstat.tcps_eifeldetected;
2052                                 tcp_revert_congestion_state(tp);
2053                                 if (tp->t_rxtshift == 1 &&
2054                                     ticks >= tp->t_badrxtwin)
2055                                         ++tcpstat.tcps_rttcantdetect;
2056                         }
2057                 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2058                         /*
2059                          * If we just performed our first retransmit,
2060                          * and the ACK arrives within our recovery window,
2061                          * then it was a mistake to do the retransmit
2062                          * in the first place.  Recover our original cwnd
2063                          * and ssthresh, and proceed to transmit where we
2064                          * left off.
2065                          */
2066                         tcp_revert_congestion_state(tp);
2067                         ++tcpstat.tcps_rttdetected;
2068                 }
2069
2070                 /*
2071                  * If we have a timestamp reply, update smoothed
2072                  * round trip time.  If no timestamp is present but
2073                  * transmit timer is running and timed sequence
2074                  * number was acked, update smoothed round trip time.
2075                  * Since we now have an rtt measurement, cancel the
2076                  * timer backoff (cf., Phil Karn's retransmit alg.).
2077                  * Recompute the initial retransmit timer.
2078                  *
2079                  * Some machines (certain windows boxes) send broken
2080                  * timestamp replies during the SYN+ACK phase, ignore
2081                  * timestamps of 0.
2082                  */
2083                 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2084                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2085                 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2086                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2087                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2088
2089                 /*
2090                  * If no data (only SYN) was ACK'd,
2091                  *    skip rest of ACK processing.
2092                  */
2093                 if (acked == 0)
2094                         goto step6;
2095
2096                 /* Stop looking for an acceptable ACK since one was received. */
2097                 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2098
2099                 if (acked > so->so_snd.sb_cc) {
2100                         tp->snd_wnd -= so->so_snd.sb_cc;
2101                         sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2102                         ourfinisacked = TRUE;
2103                 } else {
2104                         sbdrop(&so->so_snd, acked);
2105                         tp->snd_wnd -= acked;
2106                         ourfinisacked = FALSE;
2107                 }
2108                 sowwakeup(so);
2109
2110                 /*
2111                  * Update window information.
2112                  * Don't look at window if no ACK:
2113                  * TAC's send garbage on first SYN.
2114                  */
2115                 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2116                     (tp->snd_wl1 == th->th_seq &&
2117                      (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2118                       (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2119                         /* keep track of pure window updates */
2120                         if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2121                             tiwin > tp->snd_wnd)
2122                                 tcpstat.tcps_rcvwinupd++;
2123                         tp->snd_wnd = tiwin;
2124                         tp->snd_wl1 = th->th_seq;
2125                         tp->snd_wl2 = th->th_ack;
2126                         if (tp->snd_wnd > tp->max_sndwnd)
2127                                 tp->max_sndwnd = tp->snd_wnd;
2128                         needoutput = TRUE;
2129                 }
2130
2131                 tp->snd_una = th->th_ack;
2132                 if (TCP_DO_SACK(tp))
2133                         tcp_sack_update_scoreboard(tp, &to);
2134                 if (IN_FASTRECOVERY(tp)) {
2135                         if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2136                                 EXIT_FASTRECOVERY(tp);
2137                                 needoutput = TRUE;
2138                                 /*
2139                                  * If the congestion window was inflated
2140                                  * to account for the other side's
2141                                  * cached packets, retract it.
2142                                  *
2143                                  * Window inflation should have left us
2144                                  * with approximately snd_ssthresh outstanding
2145                                  * data.  But, in case we would be inclined
2146                                  * to send a burst, better do it using
2147                                  * slow start.
2148                                  */
2149                                 if (!TCP_DO_SACK(tp))
2150                                         tp->snd_cwnd = tp->snd_ssthresh;
2151
2152                                 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2153                                            tp->snd_max + 2 * tp->t_maxseg))
2154                                         tp->snd_cwnd =
2155                                             (tp->snd_max - tp->snd_una) +
2156                                             2 * tp->t_maxseg;
2157                         } else {
2158                                 if (TCP_DO_SACK(tp)) {
2159                                         tp->snd_max_rexmt = tp->snd_max;
2160                                         tcp_sack_rexmt(tp, th);
2161                                 } else {
2162                                         tcp_newreno_partial_ack(tp, th, acked);
2163                                 }
2164                                 needoutput = FALSE;
2165                         }
2166                 } else {
2167                         /*
2168                          * When new data is acked, open the congestion window.
2169                          * If the window gives us less than ssthresh packets
2170                          * in flight, open exponentially (maxseg per packet).
2171                          * Otherwise open linearly: maxseg per window
2172                          * (maxseg^2 / cwnd per packet).
2173                          */
2174                         u_int cw = tp->snd_cwnd;
2175                         u_int incr;
2176
2177                         if (cw > tp->snd_ssthresh)
2178                                 incr = tp->t_maxseg * tp->t_maxseg / cw;
2179                         else
2180                                 incr = tp->t_maxseg;
2181                         tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2182                         tp->snd_recover = th->th_ack - 1;
2183                 }
2184                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2185                         tp->snd_nxt = tp->snd_una;
2186
2187                 /*
2188                  * If all outstanding data is acked, stop retransmit
2189                  * timer and remember to restart (more output or persist).
2190                  * If there is more data to be acked, restart retransmit
2191                  * timer, using current (possibly backed-off) value.
2192                  */
2193                 if (th->th_ack == tp->snd_max) {
2194                         callout_stop(tp->tt_rexmt);
2195                         needoutput = TRUE;
2196                 } else if (!callout_active(tp->tt_persist))
2197                         callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2198                                       tcp_timer_rexmt, tp);
2199
2200                 switch (tp->t_state) {
2201                 /*
2202                  * In FIN_WAIT_1 STATE in addition to the processing
2203                  * for the ESTABLISHED state if our FIN is now acknowledged
2204                  * then enter FIN_WAIT_2.
2205                  */
2206                 case TCPS_FIN_WAIT_1:
2207                         if (ourfinisacked) {
2208                                 /*
2209                                  * If we can't receive any more
2210                                  * data, then closing user can proceed.
2211                                  * Starting the timer is contrary to the
2212                                  * specification, but if we don't get a FIN
2213                                  * we'll hang forever.
2214                                  */
2215                                 if (so->so_state & SS_CANTRCVMORE) {
2216                                         soisdisconnected(so);
2217                                         callout_reset(tp->tt_2msl, tcp_maxidle,
2218                                                       tcp_timer_2msl, tp);
2219                                 }
2220                                 tp->t_state = TCPS_FIN_WAIT_2;
2221                         }
2222                         break;
2223
2224                 /*
2225                  * In CLOSING STATE in addition to the processing for
2226                  * the ESTABLISHED state if the ACK acknowledges our FIN
2227                  * then enter the TIME-WAIT state, otherwise ignore
2228                  * the segment.
2229                  */
2230                 case TCPS_CLOSING:
2231                         if (ourfinisacked) {
2232                                 tp->t_state = TCPS_TIME_WAIT;
2233                                 tcp_canceltimers(tp);
2234                                 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2235                                 if (tp->cc_recv != 0 &&
2236                                     (ticks - tp->t_starttime) < tcp_msl)
2237                                         callout_reset(tp->tt_2msl,
2238                                             tp->t_rxtcur * TCPTV_TWTRUNC,
2239                                             tcp_timer_2msl, tp);
2240                                 else
2241                                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2242                                             tcp_timer_2msl, tp);
2243                                 soisdisconnected(so);
2244                         }
2245                         break;
2246
2247                 /*
2248                  * In LAST_ACK, we may still be waiting for data to drain
2249                  * and/or to be acked, as well as for the ack of our FIN.
2250                  * If our FIN is now acknowledged, delete the TCB,
2251                  * enter the closed state and return.
2252                  */
2253                 case TCPS_LAST_ACK:
2254                         if (ourfinisacked) {
2255                                 tp = tcp_close(tp);
2256                                 goto drop;
2257                         }
2258                         break;
2259
2260                 /*
2261                  * In TIME_WAIT state the only thing that should arrive
2262                  * is a retransmission of the remote FIN.  Acknowledge
2263                  * it and restart the finack timer.
2264                  */
2265                 case TCPS_TIME_WAIT:
2266                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2267                                       tcp_timer_2msl, tp);
2268                         goto dropafterack;
2269                 }
2270         }
2271
2272 step6:
2273         /*
2274          * Update window information.
2275          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2276          */
2277         if ((thflags & TH_ACK) &&
2278             acceptable_window_update(tp, th, tiwin)) {
2279                 /* keep track of pure window updates */
2280                 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2281                     tiwin > tp->snd_wnd)
2282                         tcpstat.tcps_rcvwinupd++;
2283                 tp->snd_wnd = tiwin;
2284                 tp->snd_wl1 = th->th_seq;
2285                 tp->snd_wl2 = th->th_ack;
2286                 if (tp->snd_wnd > tp->max_sndwnd)
2287                         tp->max_sndwnd = tp->snd_wnd;
2288                 needoutput = TRUE;
2289         }
2290
2291         /*
2292          * Process segments with URG.
2293          */
2294         if ((thflags & TH_URG) && th->th_urp &&
2295             !TCPS_HAVERCVDFIN(tp->t_state)) {
2296                 /*
2297                  * This is a kludge, but if we receive and accept
2298                  * random urgent pointers, we'll crash in
2299                  * soreceive.  It's hard to imagine someone
2300                  * actually wanting to send this much urgent data.
2301                  */
2302                 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2303                         th->th_urp = 0;                 /* XXX */
2304                         thflags &= ~TH_URG;             /* XXX */
2305                         goto dodata;                    /* XXX */
2306                 }
2307                 /*
2308                  * If this segment advances the known urgent pointer,
2309                  * then mark the data stream.  This should not happen
2310                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2311                  * a FIN has been received from the remote side.
2312                  * In these states we ignore the URG.
2313                  *
2314                  * According to RFC961 (Assigned Protocols),
2315                  * the urgent pointer points to the last octet
2316                  * of urgent data.  We continue, however,
2317                  * to consider it to indicate the first octet
2318                  * of data past the urgent section as the original
2319                  * spec states (in one of two places).
2320                  */
2321                 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2322                         tp->rcv_up = th->th_seq + th->th_urp;
2323                         so->so_oobmark = so->so_rcv.sb_cc +
2324                             (tp->rcv_up - tp->rcv_nxt) - 1;
2325                         if (so->so_oobmark == 0)
2326                                 so->so_state |= SS_RCVATMARK;
2327                         sohasoutofband(so);
2328                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2329                 }
2330                 /*
2331                  * Remove out of band data so doesn't get presented to user.
2332                  * This can happen independent of advancing the URG pointer,
2333                  * but if two URG's are pending at once, some out-of-band
2334                  * data may creep in... ick.
2335                  */
2336                 if (th->th_urp <= (u_long)tlen &&
2337                     !(so->so_options & SO_OOBINLINE)) {
2338                         /* hdr drop is delayed */
2339                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2340                 }
2341         } else {
2342                 /*
2343                  * If no out of band data is expected,
2344                  * pull receive urgent pointer along
2345                  * with the receive window.
2346                  */
2347                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2348                         tp->rcv_up = tp->rcv_nxt;
2349         }
2350
2351 dodata:                                                 /* XXX */
2352         /*
2353          * Process the segment text, merging it into the TCP sequencing queue,
2354          * and arranging for acknowledgment of receipt if necessary.
2355          * This process logically involves adjusting tp->rcv_wnd as data
2356          * is presented to the user (this happens in tcp_usrreq.c,
2357          * case PRU_RCVD).  If a FIN has already been received on this
2358          * connection then we just ignore the text.
2359          */
2360         if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2361                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2362                 /*
2363                  * Insert segment which includes th into TCP reassembly queue
2364                  * with control block tp.  Set thflags to whether reassembly now
2365                  * includes a segment with FIN.  This handles the common case
2366                  * inline (segment is the next to be received on an established
2367                  * connection, and the queue is empty), avoiding linkage into
2368                  * and removal from the queue and repetition of various
2369                  * conversions.
2370                  * Set DELACK for segments received in order, but ack
2371                  * immediately when segments are out of order (so
2372                  * fast retransmit can work).
2373                  */
2374                 if (th->th_seq == tp->rcv_nxt &&
2375                     LIST_EMPTY(&tp->t_segq) &&
2376                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2377                         if (DELAY_ACK(tp))
2378                                 callout_reset(tp->tt_delack, tcp_delacktime,
2379                                               tcp_timer_delack, tp);
2380                         else
2381                                 tp->t_flags |= TF_ACKNOW;
2382                         tp->rcv_nxt += tlen;
2383                         thflags = th->th_flags & TH_FIN;
2384                         tcpstat.tcps_rcvpack++;
2385                         tcpstat.tcps_rcvbyte += tlen;
2386                         ND6_HINT(tp);
2387                         if (so->so_state & SS_CANTRCVMORE)
2388                                 m_freem(m);
2389                         else
2390                                 sbappendstream(&so->so_rcv, m);
2391                         sorwakeup(so);
2392                 } else {
2393                         if (!(tp->t_flags & TF_DUPSEG)) {
2394                                 /* Initialize SACK report block. */
2395                                 tp->reportblk.rblk_start = th->th_seq;
2396                                 tp->reportblk.rblk_end = th->th_seq + tlen +
2397                                     ((thflags & TH_FIN) != 0);
2398                         }
2399                         thflags = tcp_reass(tp, th, &tlen, m);
2400                         tp->t_flags |= TF_ACKNOW;
2401                 }
2402
2403                 /*
2404                  * Note the amount of data that peer has sent into
2405                  * our window, in order to estimate the sender's
2406                  * buffer size.
2407                  */
2408                 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2409         } else {
2410                 m_freem(m);
2411                 thflags &= ~TH_FIN;
2412         }
2413
2414         /*
2415          * If FIN is received ACK the FIN and let the user know
2416          * that the connection is closing.
2417          */
2418         if (thflags & TH_FIN) {
2419                 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2420                         socantrcvmore(so);
2421                         /*
2422                          * If connection is half-synchronized
2423                          * (ie NEEDSYN flag on) then delay ACK,
2424                          * so it may be piggybacked when SYN is sent.
2425                          * Otherwise, since we received a FIN then no
2426                          * more input can be expected, send ACK now.
2427                          */
2428                         if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2429                                 callout_reset(tp->tt_delack, tcp_delacktime,
2430                                     tcp_timer_delack, tp);
2431                         else
2432                                 tp->t_flags |= TF_ACKNOW;
2433                         tp->rcv_nxt++;
2434                 }
2435
2436                 switch (tp->t_state) {
2437                 /*
2438                  * In SYN_RECEIVED and ESTABLISHED STATES
2439                  * enter the CLOSE_WAIT state.
2440                  */
2441                 case TCPS_SYN_RECEIVED:
2442                         tp->t_starttime = ticks;
2443                         /*FALLTHROUGH*/
2444                 case TCPS_ESTABLISHED:
2445                         tp->t_state = TCPS_CLOSE_WAIT;
2446                         break;
2447
2448                 /*
2449                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2450                  * enter the CLOSING state.
2451                  */
2452                 case TCPS_FIN_WAIT_1:
2453                         tp->t_state = TCPS_CLOSING;
2454                         break;
2455
2456                 /*
2457                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2458                  * starting the time-wait timer, turning off the other
2459                  * standard timers.
2460                  */
2461                 case TCPS_FIN_WAIT_2:
2462                         tp->t_state = TCPS_TIME_WAIT;
2463                         tcp_canceltimers(tp);
2464                         /* Shorten TIME_WAIT [RFC-1644, p.28] */
2465                         if (tp->cc_recv != 0 &&
2466                             (ticks - tp->t_starttime) < tcp_msl) {
2467                                 callout_reset(tp->tt_2msl,
2468                                               tp->t_rxtcur * TCPTV_TWTRUNC,
2469                                               tcp_timer_2msl, tp);
2470                                 /* For transaction client, force ACK now. */
2471                                 tp->t_flags |= TF_ACKNOW;
2472                         }
2473                         else
2474                                 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2475                                               tcp_timer_2msl, tp);
2476                         soisdisconnected(so);
2477                         break;
2478
2479                 /*
2480                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2481                  */
2482                 case TCPS_TIME_WAIT:
2483                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2484                                       tcp_timer_2msl, tp);
2485                         break;
2486                 }
2487         }
2488
2489 #ifdef TCPDEBUG
2490         if (so->so_options & SO_DEBUG)
2491                 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2492                           &tcp_savetcp, 0);
2493 #endif
2494
2495         /*
2496          * Return any desired output.
2497          */
2498         if (needoutput || (tp->t_flags & TF_ACKNOW))
2499                 (void) tcp_output(tp);
2500         return;
2501
2502 dropafterack:
2503         /*
2504          * Generate an ACK dropping incoming segment if it occupies
2505          * sequence space, where the ACK reflects our state.
2506          *
2507          * We can now skip the test for the RST flag since all
2508          * paths to this code happen after packets containing
2509          * RST have been dropped.
2510          *
2511          * In the SYN-RECEIVED state, don't send an ACK unless the
2512          * segment we received passes the SYN-RECEIVED ACK test.
2513          * If it fails send a RST.  This breaks the loop in the
2514          * "LAND" DoS attack, and also prevents an ACK storm
2515          * between two listening ports that have been sent forged
2516          * SYN segments, each with the source address of the other.
2517          */
2518         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2519             (SEQ_GT(tp->snd_una, th->th_ack) ||
2520              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2521                 rstreason = BANDLIM_RST_OPENPORT;
2522                 goto dropwithreset;
2523         }
2524 #ifdef TCPDEBUG
2525         if (so->so_options & SO_DEBUG)
2526                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2527                           &tcp_savetcp, 0);
2528 #endif
2529         m_freem(m);
2530         tp->t_flags |= TF_ACKNOW;
2531         (void) tcp_output(tp);
2532         return;
2533
2534 dropwithreset:
2535         /*
2536          * Generate a RST, dropping incoming segment.
2537          * Make ACK acceptable to originator of segment.
2538          * Don't bother to respond if destination was broadcast/multicast.
2539          */
2540         if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2541                 goto drop;
2542         if (isipv6) {
2543                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2544                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2545                         goto drop;
2546         } else {
2547                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2548                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2549                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2550                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2551                         goto drop;
2552         }
2553         /* IPv6 anycast check is done at tcp6_input() */
2554
2555         /*
2556          * Perform bandwidth limiting.
2557          */
2558 #ifdef ICMP_BANDLIM
2559         if (badport_bandlim(rstreason) < 0)
2560                 goto drop;
2561 #endif
2562
2563 #ifdef TCPDEBUG
2564         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2565                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2566                           &tcp_savetcp, 0);
2567 #endif
2568         if (thflags & TH_ACK)
2569                 /* mtod() below is safe as long as hdr dropping is delayed */
2570                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2571                             TH_RST);
2572         else {
2573                 if (thflags & TH_SYN)
2574                         tlen++;
2575                 /* mtod() below is safe as long as hdr dropping is delayed */
2576                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2577                             (tcp_seq)0, TH_RST | TH_ACK);
2578         }
2579         return;
2580
2581 drop:
2582         /*
2583          * Drop space held by incoming segment and return.
2584          */
2585 #ifdef TCPDEBUG
2586         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2587                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2588                           &tcp_savetcp, 0);
2589 #endif
2590         m_freem(m);
2591         return;
2592 }
2593
2594 /*
2595  * Parse TCP options and place in tcpopt.
2596  */
2597 static void
2598 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2599 {
2600         int opt, optlen, i;
2601
2602         to->to_flags = 0;
2603         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2604                 opt = cp[0];
2605                 if (opt == TCPOPT_EOL)
2606                         break;
2607                 if (opt == TCPOPT_NOP)
2608                         optlen = 1;
2609                 else {
2610                         if (cnt < 2)
2611                                 break;
2612                         optlen = cp[1];
2613                         if (optlen < 2 || optlen > cnt)
2614                                 break;
2615                 }
2616                 switch (opt) {
2617                 case TCPOPT_MAXSEG:
2618                         if (optlen != TCPOLEN_MAXSEG)
2619                                 continue;
2620                         if (!is_syn)
2621                                 continue;
2622                         to->to_flags |= TOF_MSS;
2623                         bcopy(cp + 2, &to->to_mss, sizeof(to->to_mss));
2624                         to->to_mss = ntohs(to->to_mss);
2625                         break;
2626                 case TCPOPT_WINDOW:
2627                         if (optlen != TCPOLEN_WINDOW)
2628                                 continue;
2629                         if (!is_syn)
2630                                 continue;
2631                         to->to_flags |= TOF_SCALE;
2632                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2633                         break;
2634                 case TCPOPT_TIMESTAMP:
2635                         if (optlen != TCPOLEN_TIMESTAMP)
2636                                 continue;
2637                         to->to_flags |= TOF_TS;
2638                         bcopy(cp + 2, &to->to_tsval, sizeof(to->to_tsval));
2639                         to->to_tsval = ntohl(to->to_tsval);
2640                         bcopy(cp + 6, &to->to_tsecr, sizeof(to->to_tsecr));
2641                         to->to_tsecr = ntohl(to->to_tsecr);
2642                         break;
2643                 case TCPOPT_CC:
2644                         if (optlen != TCPOLEN_CC)
2645                                 continue;
2646                         to->to_flags |= TOF_CC;
2647                         bcopy(cp + 2, &to->to_cc, sizeof(to->to_cc));
2648                         to->to_cc = ntohl(to->to_cc);
2649                         break;
2650                 case TCPOPT_CCNEW:
2651                         if (optlen != TCPOLEN_CC)
2652                                 continue;
2653                         if (!is_syn)
2654                                 continue;
2655                         to->to_flags |= TOF_CCNEW;
2656                         bcopy(cp + 2, &to->to_cc, sizeof(to->to_cc));
2657                         to->to_cc = ntohl(to->to_cc);
2658                         break;
2659                 case TCPOPT_CCECHO:
2660                         if (optlen != TCPOLEN_CC)
2661                                 continue;
2662                         if (!is_syn)
2663                                 continue;
2664                         to->to_flags |= TOF_CCECHO;
2665                         bcopy(cp + 2, &to->to_ccecho, sizeof(to->to_ccecho));
2666                         to->to_ccecho = ntohl(to->to_ccecho);
2667                         break;
2668                 case TCPOPT_SACK_PERMITTED:
2669                         if (optlen != TCPOLEN_SACK_PERMITTED)
2670                                 continue;
2671                         if (!is_syn)
2672                                 continue;
2673                         to->to_flags |= TOF_SACK_PERMITTED;
2674                         break;
2675                 case TCPOPT_SACK:
2676                         if ((optlen - 2) & 0x07)        /* not multiple of 8 */
2677                                 continue;
2678                         to->to_nsackblocks = (optlen - 2) / 8;
2679                         to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2680                         to->to_flags |= TOF_SACK;
2681                         for (i = 0; i < to->to_nsackblocks; i++) {
2682                                 struct raw_sackblock *r = &to->to_sackblocks[i];
2683
2684                                 r->rblk_start = ntohl(r->rblk_start);
2685                                 r->rblk_end = ntohl(r->rblk_end);
2686                         }
2687                         break;
2688                 default:
2689                         continue;
2690                 }
2691         }
2692 }
2693
2694 /*
2695  * Pull out of band byte out of a segment so
2696  * it doesn't appear in the user's data queue.
2697  * It is still reflected in the segment length for
2698  * sequencing purposes.
2699  * "off" is the delayed to be dropped hdrlen.
2700  */
2701 static void
2702 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2703 {
2704         int cnt = off + th->th_urp - 1;
2705
2706         while (cnt >= 0) {
2707                 if (m->m_len > cnt) {
2708                         char *cp = mtod(m, caddr_t) + cnt;
2709                         struct tcpcb *tp = sototcpcb(so);
2710
2711                         tp->t_iobc = *cp;
2712                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2713                         bcopy(cp + 1, cp, m->m_len - cnt - 1);
2714                         m->m_len--;
2715                         if (m->m_flags & M_PKTHDR)
2716                                 m->m_pkthdr.len--;
2717                         return;
2718                 }
2719                 cnt -= m->m_len;
2720                 m = m->m_next;
2721                 if (m == 0)
2722                         break;
2723         }
2724         panic("tcp_pulloutofband");
2725 }
2726
2727 /*
2728  * Collect new round-trip time estimate
2729  * and update averages and current timeout.
2730  */
2731 static void
2732 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2733 {
2734         int delta;
2735
2736         tcpstat.tcps_rttupdated++;
2737         tp->t_rttupdated++;
2738         if (tp->t_srtt != 0) {
2739                 /*
2740                  * srtt is stored as fixed point with 5 bits after the
2741                  * binary point (i.e., scaled by 8).  The following magic
2742                  * is equivalent to the smoothing algorithm in rfc793 with
2743                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2744                  * point).  Adjust rtt to origin 0.
2745                  */
2746                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2747                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2748
2749                 if ((tp->t_srtt += delta) <= 0)
2750                         tp->t_srtt = 1;
2751
2752                 /*
2753                  * We accumulate a smoothed rtt variance (actually, a
2754                  * smoothed mean difference), then set the retransmit
2755                  * timer to smoothed rtt + 4 times the smoothed variance.
2756                  * rttvar is stored as fixed point with 4 bits after the
2757                  * binary point (scaled by 16).  The following is
2758                  * equivalent to rfc793 smoothing with an alpha of .75
2759                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2760                  * rfc793's wired-in beta.
2761                  */
2762                 if (delta < 0)
2763                         delta = -delta;
2764                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2765                 if ((tp->t_rttvar += delta) <= 0)
2766                         tp->t_rttvar = 1;
2767                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2768                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2769         } else {
2770                 /*
2771                  * No rtt measurement yet - use the unsmoothed rtt.
2772                  * Set the variance to half the rtt (so our first
2773                  * retransmit happens at 3*rtt).
2774                  */
2775                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2776                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2777                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2778         }
2779         tp->t_rtttime = 0;
2780         tp->t_rxtshift = 0;
2781
2782         /*
2783          * the retransmit should happen at rtt + 4 * rttvar.
2784          * Because of the way we do the smoothing, srtt and rttvar
2785          * will each average +1/2 tick of bias.  When we compute
2786          * the retransmit timer, we want 1/2 tick of rounding and
2787          * 1 extra tick because of +-1/2 tick uncertainty in the
2788          * firing of the timer.  The bias will give us exactly the
2789          * 1.5 tick we need.  But, because the bias is
2790          * statistical, we have to test that we don't drop below
2791          * the minimum feasible timer (which is 2 ticks).
2792          */
2793         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2794                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2795
2796         /*
2797          * We received an ack for a packet that wasn't retransmitted;
2798          * it is probably safe to discard any error indications we've
2799          * received recently.  This isn't quite right, but close enough
2800          * for now (a route might have failed after we sent a segment,
2801          * and the return path might not be symmetrical).
2802          */
2803         tp->t_softerror = 0;
2804 }
2805
2806 /*
2807  * Determine a reasonable value for maxseg size.
2808  * If the route is known, check route for mtu.
2809  * If none, use an mss that can be handled on the outgoing
2810  * interface without forcing IP to fragment; if bigger than
2811  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2812  * to utilize large mbufs.  If no route is found, route has no mtu,
2813  * or the destination isn't local, use a default, hopefully conservative
2814  * size (usually 512 or the default IP max size, but no more than the mtu
2815  * of the interface), as we can't discover anything about intervening
2816  * gateways or networks.  We also initialize the congestion/slow start
2817  * window to be a single segment if the destination isn't local.
2818  * While looking at the routing entry, we also initialize other path-dependent
2819  * parameters from pre-set or cached values in the routing entry.
2820  *
2821  * Also take into account the space needed for options that we
2822  * send regularly.  Make maxseg shorter by that amount to assure
2823  * that we can send maxseg amount of data even when the options
2824  * are present.  Store the upper limit of the length of options plus
2825  * data in maxopd.
2826  *
2827  * NOTE that this routine is only called when we process an incoming
2828  * segment, for outgoing segments only tcp_mssopt is called.
2829  *
2830  * In case of T/TCP, we call this routine during implicit connection
2831  * setup as well (offer = -1), to initialize maxseg from the cached
2832  * MSS of our peer.
2833  */
2834 void
2835 tcp_mss(struct tcpcb *tp, int offer)
2836 {
2837         struct rtentry *rt;
2838         struct ifnet *ifp;
2839         int rtt, mss;
2840         u_long bufsize;
2841         struct inpcb *inp = tp->t_inpcb;
2842         struct socket *so;
2843         struct rmxp_tao *taop;
2844         int origoffer = offer;
2845 #ifdef INET6
2846         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2847         size_t min_protoh = isipv6 ?
2848                             sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2849                             sizeof(struct tcpiphdr);
2850 #else
2851         const boolean_t isipv6 = FALSE;
2852         const size_t min_protoh = sizeof(struct tcpiphdr);
2853 #endif
2854
2855         if (isipv6)
2856                 rt = tcp_rtlookup6(&inp->inp_inc);
2857         else
2858                 rt = tcp_rtlookup(&inp->inp_inc);
2859         if (rt == NULL) {
2860                 tp->t_maxopd = tp->t_maxseg =
2861                     (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2862                 return;
2863         }
2864         ifp = rt->rt_ifp;
2865         so = inp->inp_socket;
2866
2867         taop = rmx_taop(rt->rt_rmx);
2868         /*
2869          * Offer == -1 means that we didn't receive SYN yet,
2870          * use cached value in that case;
2871          */
2872         if (offer == -1)
2873                 offer = taop->tao_mssopt;
2874         /*
2875          * Offer == 0 means that there was no MSS on the SYN segment,
2876          * in this case we use tcp_mssdflt.
2877          */
2878         if (offer == 0)
2879                 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2880         else
2881                 /*
2882                  * Sanity check: make sure that maxopd will be large
2883                  * enough to allow some data on segments even is the
2884                  * all the option space is used (40bytes).  Otherwise
2885                  * funny things may happen in tcp_output.
2886                  */
2887                 offer = max(offer, 64);
2888         taop->tao_mssopt = offer;
2889
2890         /*
2891          * While we're here, check if there's an initial rtt
2892          * or rttvar.  Convert from the route-table units
2893          * to scaled multiples of the slow timeout timer.
2894          */
2895         if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2896                 /*
2897                  * XXX the lock bit for RTT indicates that the value
2898                  * is also a minimum value; this is subject to time.
2899                  */
2900                 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2901                         tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2902                 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2903                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2904                 tcpstat.tcps_usedrtt++;
2905                 if (rt->rt_rmx.rmx_rttvar) {
2906                         tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2907                             (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2908                         tcpstat.tcps_usedrttvar++;
2909                 } else {
2910                         /* default variation is +- 1 rtt */
2911                         tp->t_rttvar =
2912                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2913                 }
2914                 TCPT_RANGESET(tp->t_rxtcur,
2915                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2916                               tp->t_rttmin, TCPTV_REXMTMAX);
2917         }
2918         /*
2919          * if there's an mtu associated with the route, use it
2920          * else, use the link mtu.
2921          */
2922         if (rt->rt_rmx.rmx_mtu)
2923                 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2924         else {
2925                 if (isipv6) {
2926                         mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2927                                 min_protoh;
2928                         if (!in6_localaddr(&inp->in6p_faddr))
2929                                 mss = min(mss, tcp_v6mssdflt);
2930                 } else {
2931                         mss = ifp->if_mtu - min_protoh;
2932                         if (!in_localaddr(inp->inp_faddr))
2933                                 mss = min(mss, tcp_mssdflt);
2934                 }
2935         }
2936         mss = min(mss, offer);
2937         /*
2938          * maxopd stores the maximum length of data AND options
2939          * in a segment; maxseg is the amount of data in a normal
2940          * segment.  We need to store this value (maxopd) apart
2941          * from maxseg, because now every segment carries options
2942          * and thus we normally have somewhat less data in segments.
2943          */
2944         tp->t_maxopd = mss;
2945
2946         /*
2947          * In case of T/TCP, origoffer==-1 indicates, that no segments
2948          * were received yet.  In this case we just guess, otherwise
2949          * we do the same as before T/TCP.
2950          */
2951         if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2952             (origoffer == -1 ||
2953              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2954                 mss -= TCPOLEN_TSTAMP_APPA;
2955         if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
2956             (origoffer == -1 ||
2957              (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2958                 mss -= TCPOLEN_CC_APPA;
2959
2960 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
2961                 if (mss > MCLBYTES)
2962                         mss &= ~(MCLBYTES-1);
2963 #else
2964                 if (mss > MCLBYTES)
2965                         mss = mss / MCLBYTES * MCLBYTES;
2966 #endif
2967         /*
2968          * If there's a pipesize, change the socket buffer
2969          * to that size.  Make the socket buffers an integral
2970          * number of mss units; if the mss is larger than
2971          * the socket buffer, decrease the mss.
2972          */
2973 #ifdef RTV_SPIPE
2974         if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2975 #endif
2976                 bufsize = so->so_snd.sb_hiwat;
2977         if (bufsize < mss)
2978                 mss = bufsize;
2979         else {
2980                 bufsize = roundup(bufsize, mss);
2981                 if (bufsize > sb_max)
2982                         bufsize = sb_max;
2983                 if (bufsize > so->so_snd.sb_hiwat)
2984                         (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2985         }
2986         tp->t_maxseg = mss;
2987
2988 #ifdef RTV_RPIPE
2989         if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2990 #endif
2991                 bufsize = so->so_rcv.sb_hiwat;
2992         if (bufsize > mss) {
2993                 bufsize = roundup(bufsize, mss);
2994                 if (bufsize > sb_max)
2995                         bufsize = sb_max;
2996                 if (bufsize > so->so_rcv.sb_hiwat)
2997                         (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2998         }
2999
3000         /*
3001          * Set the slow-start flight size depending on whether this
3002          * is a local network or not.
3003          */
3004         if (tcp_do_rfc3390)
3005                 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3006         else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3007                  (!isipv6 && in_localaddr(inp->inp_faddr)))
3008                 tp->snd_cwnd = mss * ss_fltsz_local;
3009         else
3010                 tp->snd_cwnd = mss * ss_fltsz;
3011
3012         if (rt->rt_rmx.rmx_ssthresh) {
3013                 /*
3014                  * There's some sort of gateway or interface
3015                  * buffer limit on the path.  Use this to set
3016                  * the slow start threshhold, but set the
3017                  * threshold to no less than 2*mss.
3018                  */
3019                 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3020                 tcpstat.tcps_usedssthresh++;
3021         }
3022 }
3023
3024 /*
3025  * Determine the MSS option to send on an outgoing SYN.
3026  */
3027 int
3028 tcp_mssopt(struct tcpcb *tp)
3029 {
3030         struct rtentry *rt;
3031 #ifdef INET6
3032         boolean_t isipv6 =
3033             ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3034         int min_protoh = isipv6 ?
3035                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3036                              sizeof(struct tcpiphdr);
3037 #else
3038         const boolean_t isipv6 = FALSE;
3039         const size_t min_protoh = sizeof(struct tcpiphdr);
3040 #endif
3041
3042         if (isipv6)
3043                 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3044         else
3045                 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3046         if (rt == NULL)
3047                 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3048
3049         return (rt->rt_ifp->if_mtu - min_protoh);
3050 }
3051
3052 /*
3053  * When a partial ack arrives, force the retransmission of the
3054  * next unacknowledged segment.  Do not exit Fast Recovery.
3055  *
3056  * Implement the Slow-but-Steady variant of NewReno by restarting the
3057  * the retransmission timer.  Turn it off here so it can be restarted
3058  * later in tcp_output().
3059  */
3060 static void
3061 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3062 {
3063         tcp_seq old_snd_nxt = tp->snd_nxt;
3064         u_long ocwnd = tp->snd_cwnd;
3065
3066         callout_stop(tp->tt_rexmt);
3067         tp->t_rtttime = 0;
3068         tp->snd_nxt = th->th_ack;
3069         /* Set snd_cwnd to one segment beyond acknowledged offset. */
3070         tp->snd_cwnd = tp->t_maxseg;
3071         tp->t_flags |= TF_ACKNOW;
3072         (void) tcp_output(tp);
3073         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3074                 tp->snd_nxt = old_snd_nxt;
3075         /* partial window deflation */
3076         tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3077 }
3078
3079 /*
3080  * In contrast to the Slow-but-Steady NewReno variant,
3081  * we do not reset the retransmission timer for SACK retransmissions,
3082  * except when retransmitting snd_una.
3083  */
3084 static void
3085 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3086 {
3087         uint32_t pipe, seglen;
3088         tcp_seq nextrexmt;
3089         boolean_t lostdup;
3090         tcp_seq old_snd_nxt = tp->snd_nxt;
3091         u_long ocwnd = tp->snd_cwnd;
3092         int nseg = 0;           /* consecutive new segments */
3093 #define MAXBURST 4              /* limit burst of new packets on partial ack */
3094
3095         tp->t_rtttime = 0;
3096         pipe = tcp_sack_compute_pipe(tp);
3097         while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3098             (!tcp_do_smartsack || nseg < MAXBURST) &&
3099             tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3100                 uint32_t sent;
3101                 tcp_seq old_snd_max;
3102                 int error;
3103
3104                 if (nextrexmt == tp->snd_max) ++nseg;
3105                 tp->snd_nxt = nextrexmt;
3106                 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3107                 old_snd_max = tp->snd_max;
3108                 if (nextrexmt == tp->snd_una)
3109                         callout_stop(tp->tt_rexmt);
3110                 error = tcp_output(tp);
3111                 if (error != 0)
3112                         break;
3113                 sent = tp->snd_nxt - nextrexmt;
3114                 if (sent <= 0)
3115                         break;
3116                 if (!lostdup)
3117                         pipe += sent;
3118                 tcpstat.tcps_sndsackpack++;
3119                 tcpstat.tcps_sndsackbyte += sent;
3120                 if (SEQ_LT(nextrexmt, old_snd_max) &&
3121                     SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3122                         tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3123         }
3124         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3125                 tp->snd_nxt = old_snd_nxt;
3126         tp->snd_cwnd = ocwnd;
3127 }