tcp/sack: Implement RFC3517bis
[dragonfly.git] / sys / netinet / tcp_input.c
1 /*
2  * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2002, 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36  *      The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *      This product includes software developed by the University of
49  *      California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *      @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67  * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68  */
69
70 #include "opt_inet.h"
71 #include "opt_inet6.h"
72 #include "opt_ipsec.h"
73 #include "opt_tcpdebug.h"
74 #include "opt_tcp_input.h"
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 #include <sys/malloc.h>
81 #include <sys/mbuf.h>
82 #include <sys/proc.h>           /* for proc0 declaration */
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/syslog.h>
87 #include <sys/in_cksum.h>
88
89 #include <sys/socketvar2.h>
90
91 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
92 #include <machine/stdarg.h>
93
94 #include <net/if.h>
95 #include <net/route.h>
96
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip_icmp.h>    /* for ICMP_BANDLIM */
101 #include <netinet/in_var.h>
102 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
103 #include <netinet/in_pcb.h>
104 #include <netinet/ip_var.h>
105 #include <netinet/ip6.h>
106 #include <netinet/icmp6.h>
107 #include <netinet6/nd6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/in6_pcb.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
117 #include <netinet/tcpip.h>
118
119 #ifdef TCPDEBUG
120 #include <netinet/tcp_debug.h>
121
122 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
123 struct tcphdr tcp_savetcp;
124 #endif
125
126 #ifdef FAST_IPSEC
127 #include <netproto/ipsec/ipsec.h>
128 #include <netproto/ipsec/ipsec6.h>
129 #endif
130
131 #ifdef IPSEC
132 #include <netinet6/ipsec.h>
133 #include <netinet6/ipsec6.h>
134 #include <netproto/key/key.h>
135 #endif
136
137 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
138
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141     &log_in_vain, 0, "Log all incoming TCP connections");
142
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145     &blackhole, 0, "Do not send RST when dropping refused connections");
146
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149     &tcp_delack_enabled, 0,
150     "Delay ACK to try and piggyback it onto a data packet");
151
152 #ifdef TCP_DROP_SYNFIN
153 static int drop_synfin = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
156 #endif
157
158 static int tcp_do_limitedtransmit = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160     &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
161
162 static int tcp_do_early_retransmit = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164     &tcp_do_early_retransmit, 0, "Early retransmit");
165
166 int tcp_aggregate_acks = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168     &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
169
170 static int tcp_do_eifel_detect = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
172     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
173
174 static int tcp_do_abc = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
176     &tcp_do_abc, 0,
177     "TCP Appropriate Byte Counting (RFC 3465)");
178
179 /*
180  * Define as tunable for easy testing with SACK on and off.
181  * Warning:  do not change setting in the middle of an existing active TCP flow,
182  *   else strange things might happen to that flow.
183  */
184 int tcp_do_sack = 1;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
186     &tcp_do_sack, 0, "Enable SACK Algorithms");
187
188 int tcp_do_smartsack = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
190     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
191
192 int tcp_do_rescuesack = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW,
194     &tcp_do_rescuesack, 0, "Rescue retransmission for SACK");
195
196 int tcp_aggressive_rescuesack = 0;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW,
198     &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK");
199
200 int tcp_do_rfc3517bis = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis, CTLFLAG_RW,
202     &tcp_do_rfc3517bis, 0, "Enable RFC3517 update");
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 int tcp_do_autorcvbuf = 1;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
224     &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
225
226 int tcp_autorcvbuf_inc = 16*1024;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
228     &tcp_autorcvbuf_inc, 0,
229     "Incrementor step size of automatic receive buffer");
230
231 int tcp_autorcvbuf_max = 2*1024*1024;
232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
233     &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
234
235 int tcp_sosend_agglim = 2;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW,
237     &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit");
238
239 int tcp_sosend_async = 1;
240 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW,
241     &tcp_sosend_async, 0, "TCP asynchronized pru_send");
242
243 static int tcp_ignore_redun_dsack = 1;
244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW,
245     &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK");
246
247 static void      tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t,
248                     tcp_seq);
249 static void      tcp_pulloutofband(struct socket *,
250                      struct tcphdr *, struct mbuf *, int);
251 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
252                      struct mbuf *);
253 static void      tcp_xmit_timer(struct tcpcb *, int, tcp_seq);
254 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
255 static void      tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
256 static boolean_t tcp_sack_limitedxmit(struct tcpcb *);
257 static int       tcp_rmx_msl(const struct tcpcb *);
258 static void      tcp_established(struct tcpcb *);
259
260 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
261 #ifdef INET6
262 #define ND6_HINT(tp) \
263 do { \
264         if ((tp) && (tp)->t_inpcb && \
265             ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
266             (tp)->t_inpcb->in6p_route.ro_rt) \
267                 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
268 } while (0)
269 #else
270 #define ND6_HINT(tp)
271 #endif
272
273 /*
274  * Indicate whether this ack should be delayed.  We can delay the ack if
275  *      - delayed acks are enabled and
276  *      - there is no delayed ack timer in progress and
277  *      - our last ack wasn't a 0-sized window.  We never want to delay
278  *        the ack that opens up a 0-sized window.
279  */
280 #define DELAY_ACK(tp) \
281         (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
282         !(tp->t_flags & TF_RXWIN0SENT))
283
284 #define acceptable_window_update(tp, th, tiwin)                         \
285     (SEQ_LT(tp->snd_wl1, th->th_seq) ||                                 \
286      (tp->snd_wl1 == th->th_seq &&                                      \
287       (SEQ_LT(tp->snd_wl2, th->th_ack) ||                               \
288        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
289
290 #define iceildiv(n, d)          (((n)+(d)-1) / (d))
291 #define need_early_retransmit(tp, ownd) \
292     (tcp_do_early_retransmit && \
293      (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \
294      ownd < (4 * tp->t_maxseg) && \
295      tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \
296      (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \
297       tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg)))
298
299 static int
300 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
301 {
302         struct tseg_qent *q;
303         struct tseg_qent *p = NULL;
304         struct tseg_qent *te;
305         struct socket *so = tp->t_inpcb->inp_socket;
306         int flags;
307
308         /*
309          * Call with th == NULL after become established to
310          * force pre-ESTABLISHED data up to user socket.
311          */
312         if (th == NULL)
313                 goto present;
314
315         /*
316          * Limit the number of segments in the reassembly queue to prevent
317          * holding on to too many segments (and thus running out of mbufs).
318          * Make sure to let the missing segment through which caused this
319          * queue.  Always keep one global queue entry spare to be able to
320          * process the missing segment.
321          */
322         if (th->th_seq != tp->rcv_nxt &&
323             tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
324                 tcp_reass_overflows++;
325                 tcpstat.tcps_rcvmemdrop++;
326                 m_freem(m);
327                 /* no SACK block to report */
328                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
329                 return (0);
330         }
331
332         /* Allocate a new queue entry. */
333         te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
334         if (te == NULL) {
335                 tcpstat.tcps_rcvmemdrop++;
336                 m_freem(m);
337                 /* no SACK block to report */
338                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
339                 return (0);
340         }
341         atomic_add_int(&tcp_reass_qsize, 1);
342
343         /*
344          * Find a segment which begins after this one does.
345          */
346         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
347                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
348                         break;
349                 p = q;
350         }
351
352         /*
353          * If there is a preceding segment, it may provide some of
354          * our data already.  If so, drop the data from the incoming
355          * segment.  If it provides all of our data, drop us.
356          */
357         if (p != NULL) {
358                 tcp_seq_diff_t i;
359
360                 /* conversion to int (in i) handles seq wraparound */
361                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
362                 if (i > 0) {            /* overlaps preceding segment */
363                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
364                         /* enclosing block starts w/ preceding segment */
365                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
366                         if (i >= *tlenp) {
367                                 /* preceding encloses incoming segment */
368                                 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
369                                     p->tqe_th->th_seq + p->tqe_len,
370                                     p->tqe_th->th_flags);
371                                 tcpstat.tcps_rcvduppack++;
372                                 tcpstat.tcps_rcvdupbyte += *tlenp;
373                                 m_freem(m);
374                                 kfree(te, M_TSEGQ);
375                                 atomic_add_int(&tcp_reass_qsize, -1);
376                                 /*
377                                  * Try to present any queued data
378                                  * at the left window edge to the user.
379                                  * This is needed after the 3-WHS
380                                  * completes.
381                                  */
382                                 goto present;   /* ??? */
383                         }
384                         m_adj(m, i);
385                         *tlenp -= i;
386                         th->th_seq += i;
387                         /* incoming segment end is enclosing block end */
388                         tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
389                             th->th_seq + *tlenp, th->th_flags);
390                         /* trim end of reported D-SACK block */
391                         tp->reportblk.rblk_end = th->th_seq;
392                 }
393         }
394         tcpstat.tcps_rcvoopack++;
395         tcpstat.tcps_rcvoobyte += *tlenp;
396
397         /*
398          * While we overlap succeeding segments trim them or,
399          * if they are completely covered, dequeue them.
400          */
401         while (q) {
402                 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
403                 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
404                 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
405                 struct tseg_qent *nq;
406
407                 if (i <= 0)
408                         break;
409                 if (!(tp->t_flags & TF_DUPSEG)) {    /* first time through */
410                         tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
411                         tp->encloseblk = tp->reportblk;
412                         /* report trailing duplicate D-SACK segment */
413                         tp->reportblk.rblk_start = q->tqe_th->th_seq;
414                 }
415                 if ((tp->t_flags & TF_ENCLOSESEG) &&
416                     SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
417                         /* extend enclosing block if one exists */
418                         tp->encloseblk.rblk_end = qend_sack;
419                 }
420                 if (i < q->tqe_len) {
421                         q->tqe_th->th_seq += i;
422                         q->tqe_len -= i;
423                         m_adj(q->tqe_m, i);
424                         break;
425                 }
426
427                 nq = LIST_NEXT(q, tqe_q);
428                 LIST_REMOVE(q, tqe_q);
429                 m_freem(q->tqe_m);
430                 kfree(q, M_TSEGQ);
431                 atomic_add_int(&tcp_reass_qsize, -1);
432                 q = nq;
433         }
434
435         /* Insert the new segment queue entry into place. */
436         te->tqe_m = m;
437         te->tqe_th = th;
438         te->tqe_len = *tlenp;
439
440         /* check if can coalesce with following segment */
441         if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
442                 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
443                 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
444
445                 te->tqe_len += q->tqe_len;
446                 if (q->tqe_th->th_flags & TH_FIN)
447                         te->tqe_th->th_flags |= TH_FIN;
448                 m_cat(te->tqe_m, q->tqe_m);
449                 tp->encloseblk.rblk_end = tend_sack;
450                 /*
451                  * When not reporting a duplicate segment, use
452                  * the larger enclosing block as the SACK block.
453                  */
454                 if (!(tp->t_flags & TF_DUPSEG))
455                         tp->reportblk.rblk_end = tend_sack;
456                 LIST_REMOVE(q, tqe_q);
457                 kfree(q, M_TSEGQ);
458                 atomic_add_int(&tcp_reass_qsize, -1);
459         }
460
461         if (p == NULL) {
462                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
463         } else {
464                 /* check if can coalesce with preceding segment */
465                 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
466                         p->tqe_len += te->tqe_len;
467                         m_cat(p->tqe_m, te->tqe_m);
468                         tp->encloseblk.rblk_start = p->tqe_th->th_seq;
469                         /*
470                          * When not reporting a duplicate segment, use
471                          * the larger enclosing block as the SACK block.
472                          */
473                         if (!(tp->t_flags & TF_DUPSEG))
474                                 tp->reportblk.rblk_start = p->tqe_th->th_seq;
475                         kfree(te, M_TSEGQ);
476                         atomic_add_int(&tcp_reass_qsize, -1);
477                 } else {
478                         LIST_INSERT_AFTER(p, te, tqe_q);
479                 }
480         }
481
482 present:
483         /*
484          * Present data to user, advancing rcv_nxt through
485          * completed sequence space.
486          */
487         if (!TCPS_HAVEESTABLISHED(tp->t_state))
488                 return (0);
489         q = LIST_FIRST(&tp->t_segq);
490         if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
491                 return (0);
492         tp->rcv_nxt += q->tqe_len;
493         if (!(tp->t_flags & TF_DUPSEG)) {
494                 /* no SACK block to report since ACK advanced */
495                 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
496         }
497         /* no enclosing block to report since ACK advanced */
498         tp->t_flags &= ~TF_ENCLOSESEG;
499         flags = q->tqe_th->th_flags & TH_FIN;
500         LIST_REMOVE(q, tqe_q);
501         KASSERT(LIST_EMPTY(&tp->t_segq) ||
502                 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
503                 ("segment not coalesced"));
504         if (so->so_state & SS_CANTRCVMORE) {
505                 m_freem(q->tqe_m);
506         } else {
507                 lwkt_gettoken(&so->so_rcv.ssb_token);
508                 ssb_appendstream(&so->so_rcv, q->tqe_m);
509                 lwkt_reltoken(&so->so_rcv.ssb_token);
510         }
511         kfree(q, M_TSEGQ);
512         atomic_add_int(&tcp_reass_qsize, -1);
513         ND6_HINT(tp);
514         sorwakeup(so);
515         return (flags);
516 }
517
518 /*
519  * TCP input routine, follows pages 65-76 of the
520  * protocol specification dated September, 1981 very closely.
521  */
522 #ifdef INET6
523 int
524 tcp6_input(struct mbuf **mp, int *offp, int proto)
525 {
526         struct mbuf *m = *mp;
527         struct in6_ifaddr *ia6;
528
529         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
530
531         /*
532          * draft-itojun-ipv6-tcp-to-anycast
533          * better place to put this in?
534          */
535         ia6 = ip6_getdstifaddr(m);
536         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
537                 struct ip6_hdr *ip6;
538
539                 ip6 = mtod(m, struct ip6_hdr *);
540                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
541                             offsetof(struct ip6_hdr, ip6_dst));
542                 return (IPPROTO_DONE);
543         }
544
545         tcp_input(mp, offp, proto);
546         return (IPPROTO_DONE);
547 }
548 #endif
549
550 int
551 tcp_input(struct mbuf **mp, int *offp, int proto)
552 {
553         int off0;
554         struct tcphdr *th;
555         struct ip *ip = NULL;
556         struct ipovly *ipov;
557         struct inpcb *inp = NULL;
558         u_char *optp = NULL;
559         int optlen = 0;
560         int tlen, off;
561         int len = 0;
562         int drop_hdrlen;
563         struct tcpcb *tp = NULL;
564         int thflags;
565         struct socket *so = NULL;
566         int todrop, acked;
567         boolean_t ourfinisacked, needoutput = FALSE;
568         u_long tiwin;
569         int recvwin;
570         struct tcpopt to;               /* options in this segment */
571         struct sockaddr_in *next_hop = NULL;
572         int rstreason; /* For badport_bandlim accounting purposes */
573         int cpu;
574         struct ip6_hdr *ip6 = NULL;
575         struct mbuf *m;
576 #ifdef INET6
577         boolean_t isipv6;
578 #else
579         const boolean_t isipv6 = FALSE;
580 #endif
581 #ifdef TCPDEBUG
582         short ostate = 0;
583 #endif
584
585         off0 = *offp;
586         m = *mp;
587         *mp = NULL;
588
589         tcpstat.tcps_rcvtotal++;
590
591         if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
592                 struct m_tag *mtag;
593
594                 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
595                 KKASSERT(mtag != NULL);
596                 next_hop = m_tag_data(mtag);
597         }
598
599 #ifdef INET6
600         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
601 #endif
602
603         if (isipv6) {
604                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
605                 ip6 = mtod(m, struct ip6_hdr *);
606                 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
607                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
608                         tcpstat.tcps_rcvbadsum++;
609                         goto drop;
610                 }
611                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
612
613                 /*
614                  * Be proactive about unspecified IPv6 address in source.
615                  * As we use all-zero to indicate unbounded/unconnected pcb,
616                  * unspecified IPv6 address can be used to confuse us.
617                  *
618                  * Note that packets with unspecified IPv6 destination is
619                  * already dropped in ip6_input.
620                  */
621                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
622                         /* XXX stat */
623                         goto drop;
624                 }
625         } else {
626                 /*
627                  * Get IP and TCP header together in first mbuf.
628                  * Note: IP leaves IP header in first mbuf.
629                  */
630                 if (off0 > sizeof(struct ip)) {
631                         ip_stripoptions(m);
632                         off0 = sizeof(struct ip);
633                 }
634                 /* already checked and pulled up in ip_demux() */
635                 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
636                     ("TCP header not in one mbuf: m->m_len %d", m->m_len));
637                 ip = mtod(m, struct ip *);
638                 ipov = (struct ipovly *)ip;
639                 th = (struct tcphdr *)((caddr_t)ip + off0);
640                 tlen = ip->ip_len;
641
642                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
643                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
644                                 th->th_sum = m->m_pkthdr.csum_data;
645                         else
646                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
647                                                 ip->ip_dst.s_addr,
648                                                 htonl(m->m_pkthdr.csum_data +
649                                                         ip->ip_len +
650                                                         IPPROTO_TCP));
651                         th->th_sum ^= 0xffff;
652                 } else {
653                         /*
654                          * Checksum extended TCP header and data.
655                          */
656                         len = sizeof(struct ip) + tlen;
657                         bzero(ipov->ih_x1, sizeof ipov->ih_x1);
658                         ipov->ih_len = (u_short)tlen;
659                         ipov->ih_len = htons(ipov->ih_len);
660                         th->th_sum = in_cksum(m, len);
661                 }
662                 if (th->th_sum) {
663                         tcpstat.tcps_rcvbadsum++;
664                         goto drop;
665                 }
666 #ifdef INET6
667                 /* Re-initialization for later version check */
668                 ip->ip_v = IPVERSION;
669 #endif
670         }
671
672         /*
673          * Check that TCP offset makes sense,
674          * pull out TCP options and adjust length.              XXX
675          */
676         off = th->th_off << 2;
677         /* already checked and pulled up in ip_demux() */
678         KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
679             ("bad TCP data offset %d (tlen %d)", off, tlen));
680         tlen -= off;    /* tlen is used instead of ti->ti_len */
681         if (off > sizeof(struct tcphdr)) {
682                 if (isipv6) {
683                         IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
684                         ip6 = mtod(m, struct ip6_hdr *);
685                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
686                 } else {
687                         /* already pulled up in ip_demux() */
688                         KASSERT(m->m_len >= sizeof(struct ip) + off,
689                             ("TCP header and options not in one mbuf: "
690                              "m_len %d, off %d", m->m_len, off));
691                 }
692                 optlen = off - sizeof(struct tcphdr);
693                 optp = (u_char *)(th + 1);
694         }
695         thflags = th->th_flags;
696
697 #ifdef TCP_DROP_SYNFIN
698         /*
699          * If the drop_synfin option is enabled, drop all packets with
700          * both the SYN and FIN bits set. This prevents e.g. nmap from
701          * identifying the TCP/IP stack.
702          *
703          * This is a violation of the TCP specification.
704          */
705         if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
706                 goto drop;
707 #endif
708
709         /*
710          * Convert TCP protocol specific fields to host format.
711          */
712         th->th_seq = ntohl(th->th_seq);
713         th->th_ack = ntohl(th->th_ack);
714         th->th_win = ntohs(th->th_win);
715         th->th_urp = ntohs(th->th_urp);
716
717         /*
718          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
719          * until after ip6_savecontrol() is called and before other functions
720          * which don't want those proto headers.
721          * Because ip6_savecontrol() is going to parse the mbuf to
722          * search for data to be passed up to user-land, it wants mbuf
723          * parameters to be unchanged.
724          * XXX: the call of ip6_savecontrol() has been obsoleted based on
725          * latest version of the advanced API (20020110).
726          */
727         drop_hdrlen = off0 + off;
728
729         /*
730          * Locate pcb for segment.
731          */
732 findpcb:
733         /* IPFIREWALL_FORWARD section */
734         if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
735                 /*
736                  * Transparently forwarded. Pretend to be the destination.
737                  * already got one like this?
738                  */
739                 cpu = mycpu->gd_cpuid;
740                 inp = in_pcblookup_hash(&tcbinfo[cpu],
741                                         ip->ip_src, th->th_sport,
742                                         ip->ip_dst, th->th_dport,
743                                         0, m->m_pkthdr.rcvif);
744                 if (!inp) {
745                         /*
746                          * It's new.  Try to find the ambushing socket.
747                          */
748
749                         /*
750                          * The rest of the ipfw code stores the port in
751                          * host order.  XXX
752                          * (The IP address is still in network order.)
753                          */
754                         in_port_t dport = next_hop->sin_port ?
755                                                 htons(next_hop->sin_port) :
756                                                 th->th_dport;
757
758                         cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
759                                           next_hop->sin_addr.s_addr, dport);
760                         inp = in_pcblookup_hash(&tcbinfo[cpu],
761                                                 ip->ip_src, th->th_sport,
762                                                 next_hop->sin_addr, dport,
763                                                 1, m->m_pkthdr.rcvif);
764                 }
765         } else {
766                 if (isipv6) {
767                         inp = in6_pcblookup_hash(&tcbinfo[0],
768                                                  &ip6->ip6_src, th->th_sport,
769                                                  &ip6->ip6_dst, th->th_dport,
770                                                  1, m->m_pkthdr.rcvif);
771                 } else {
772                         cpu = mycpu->gd_cpuid;
773                         inp = in_pcblookup_hash(&tcbinfo[cpu],
774                                                 ip->ip_src, th->th_sport,
775                                                 ip->ip_dst, th->th_dport,
776                                                 1, m->m_pkthdr.rcvif);
777                 }
778         }
779
780         /*
781          * If the state is CLOSED (i.e., TCB does not exist) then
782          * all data in the incoming segment is discarded.
783          * If the TCB exists but is in CLOSED state, it is embryonic,
784          * but should either do a listen or a connect soon.
785          */
786         if (inp == NULL) {
787                 if (log_in_vain) {
788 #ifdef INET6
789                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
790 #else
791                         char dbuf[sizeof "aaa.bbb.ccc.ddd"];
792                         char sbuf[sizeof "aaa.bbb.ccc.ddd"];
793 #endif
794                         if (isipv6) {
795                                 strcpy(dbuf, "[");
796                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
797                                 strcat(dbuf, "]");
798                                 strcpy(sbuf, "[");
799                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
800                                 strcat(sbuf, "]");
801                         } else {
802                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
803                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
804                         }
805                         switch (log_in_vain) {
806                         case 1:
807                                 if (!(thflags & TH_SYN))
808                                         break;
809                         case 2:
810                                 log(LOG_INFO,
811                                     "Connection attempt to TCP %s:%d "
812                                     "from %s:%d flags:0x%02x\n",
813                                     dbuf, ntohs(th->th_dport), sbuf,
814                                     ntohs(th->th_sport), thflags);
815                                 break;
816                         default:
817                                 break;
818                         }
819                 }
820                 if (blackhole) {
821                         switch (blackhole) {
822                         case 1:
823                                 if (thflags & TH_SYN)
824                                         goto drop;
825                                 break;
826                         case 2:
827                                 goto drop;
828                         default:
829                                 goto drop;
830                         }
831                 }
832                 rstreason = BANDLIM_RST_CLOSEDPORT;
833                 goto dropwithreset;
834         }
835
836 #ifdef IPSEC
837         if (isipv6) {
838                 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
839                         ipsec6stat.in_polvio++;
840                         goto drop;
841                 }
842         } else {
843                 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
844                         ipsecstat.in_polvio++;
845                         goto drop;
846                 }
847         }
848 #endif
849 #ifdef FAST_IPSEC
850         if (isipv6) {
851                 if (ipsec6_in_reject(m, inp))
852                         goto drop;
853         } else {
854                 if (ipsec4_in_reject(m, inp))
855                         goto drop;
856         }
857 #endif
858         /* Check the minimum TTL for socket. */
859 #ifdef INET6
860         if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
861                 goto drop;
862 #endif
863
864         tp = intotcpcb(inp);
865         if (tp == NULL) {
866                 rstreason = BANDLIM_RST_CLOSEDPORT;
867                 goto dropwithreset;
868         }
869         if (tp->t_state <= TCPS_CLOSED)
870                 goto drop;
871
872         so = inp->inp_socket;
873
874 #ifdef TCPDEBUG
875         if (so->so_options & SO_DEBUG) {
876                 ostate = tp->t_state;
877                 if (isipv6)
878                         bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
879                 else
880                         bcopy(ip, tcp_saveipgen, sizeof(*ip));
881                 tcp_savetcp = *th;
882         }
883 #endif
884
885         bzero(&to, sizeof to);
886
887         if (so->so_options & SO_ACCEPTCONN) {
888                 struct in_conninfo inc;
889
890 #ifdef INET6
891                 inc.inc_isipv6 = (isipv6 == TRUE);
892 #endif
893                 if (isipv6) {
894                         inc.inc6_faddr = ip6->ip6_src;
895                         inc.inc6_laddr = ip6->ip6_dst;
896                         inc.inc6_route.ro_rt = NULL;            /* XXX */
897                 } else {
898                         inc.inc_faddr = ip->ip_src;
899                         inc.inc_laddr = ip->ip_dst;
900                         inc.inc_route.ro_rt = NULL;             /* XXX */
901                 }
902                 inc.inc_fport = th->th_sport;
903                 inc.inc_lport = th->th_dport;
904
905                 /*
906                  * If the state is LISTEN then ignore segment if it contains
907                  * a RST.  If the segment contains an ACK then it is bad and
908                  * send a RST.  If it does not contain a SYN then it is not
909                  * interesting; drop it.
910                  *
911                  * If the state is SYN_RECEIVED (syncache) and seg contains
912                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
913                  * contains a RST, check the sequence number to see if it
914                  * is a valid reset segment.
915                  */
916                 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
917                         if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
918                                 if (!syncache_expand(&inc, th, &so, m)) {
919                                         /*
920                                          * No syncache entry, or ACK was not
921                                          * for our SYN/ACK.  Send a RST.
922                                          */
923                                         tcpstat.tcps_badsyn++;
924                                         rstreason = BANDLIM_RST_OPENPORT;
925                                         goto dropwithreset;
926                                 }
927
928                                 /*
929                                  * Could not complete 3-way handshake,
930                                  * connection is being closed down, and
931                                  * syncache will free mbuf.
932                                  */
933                                 if (so == NULL)
934                                         return(IPPROTO_DONE);
935
936                                 /*
937                                  * We must be in the correct protocol thread
938                                  * for this connection.
939                                  */
940                                 KKASSERT(so->so_port == &curthread->td_msgport);
941
942                                 /*
943                                  * Socket is created in state SYN_RECEIVED.
944                                  * Continue processing segment.
945                                  */
946                                 inp = so->so_pcb;
947                                 tp = intotcpcb(inp);
948                                 /*
949                                  * This is what would have happened in
950                                  * tcp_output() when the SYN,ACK was sent.
951                                  */
952                                 tp->snd_up = tp->snd_una;
953                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
954                                 tp->last_ack_sent = tp->rcv_nxt;
955
956                                 goto after_listen;
957                         }
958                         if (thflags & TH_RST) {
959                                 syncache_chkrst(&inc, th);
960                                 goto drop;
961                         }
962                         if (thflags & TH_ACK) {
963                                 syncache_badack(&inc);
964                                 tcpstat.tcps_badsyn++;
965                                 rstreason = BANDLIM_RST_OPENPORT;
966                                 goto dropwithreset;
967                         }
968                         goto drop;
969                 }
970
971                 /*
972                  * Segment's flags are (SYN) or (SYN | FIN).
973                  */
974 #ifdef INET6
975                 /*
976                  * If deprecated address is forbidden,
977                  * we do not accept SYN to deprecated interface
978                  * address to prevent any new inbound connection from
979                  * getting established.
980                  * When we do not accept SYN, we send a TCP RST,
981                  * with deprecated source address (instead of dropping
982                  * it).  We compromise it as it is much better for peer
983                  * to send a RST, and RST will be the final packet
984                  * for the exchange.
985                  *
986                  * If we do not forbid deprecated addresses, we accept
987                  * the SYN packet.  RFC2462 does not suggest dropping
988                  * SYN in this case.
989                  * If we decipher RFC2462 5.5.4, it says like this:
990                  * 1. use of deprecated addr with existing
991                  *    communication is okay - "SHOULD continue to be
992                  *    used"
993                  * 2. use of it with new communication:
994                  *   (2a) "SHOULD NOT be used if alternate address
995                  *        with sufficient scope is available"
996                  *   (2b) nothing mentioned otherwise.
997                  * Here we fall into (2b) case as we have no choice in
998                  * our source address selection - we must obey the peer.
999                  *
1000                  * The wording in RFC2462 is confusing, and there are
1001                  * multiple description text for deprecated address
1002                  * handling - worse, they are not exactly the same.
1003                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
1004                  */
1005                 if (isipv6 && !ip6_use_deprecated) {
1006                         struct in6_ifaddr *ia6;
1007
1008                         if ((ia6 = ip6_getdstifaddr(m)) &&
1009                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1010                                 tp = NULL;
1011                                 rstreason = BANDLIM_RST_OPENPORT;
1012                                 goto dropwithreset;
1013                         }
1014                 }
1015 #endif
1016                 /*
1017                  * If it is from this socket, drop it, it must be forged.
1018                  * Don't bother responding if the destination was a broadcast.
1019                  */
1020                 if (th->th_dport == th->th_sport) {
1021                         if (isipv6) {
1022                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1023                                                        &ip6->ip6_src))
1024                                         goto drop;
1025                         } else {
1026                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1027                                         goto drop;
1028                         }
1029                 }
1030                 /*
1031                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1032                  *
1033                  * Note that it is quite possible to receive unicast
1034                  * link-layer packets with a broadcast IP address. Use
1035                  * in_broadcast() to find them.
1036                  */
1037                 if (m->m_flags & (M_BCAST | M_MCAST))
1038                         goto drop;
1039                 if (isipv6) {
1040                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1041                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1042                                 goto drop;
1043                 } else {
1044                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1045                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1046                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1047                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1048                                 goto drop;
1049                 }
1050                 /*
1051                  * SYN appears to be valid; create compressed TCP state
1052                  * for syncache, or perform t/tcp connection.
1053                  */
1054                 if (so->so_qlen <= so->so_qlimit) {
1055                         tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack);
1056                         if (!syncache_add(&inc, &to, th, so, m))
1057                                 goto drop;
1058
1059                         /*
1060                          * Entry added to syncache, mbuf used to
1061                          * send SYN,ACK packet.
1062                          */
1063                         return(IPPROTO_DONE);
1064                 }
1065                 goto drop;
1066         }
1067
1068 after_listen:
1069         /*
1070          * Should not happen - syncache should pick up these connections.
1071          *
1072          * Once we are past handling listen sockets we must be in the
1073          * correct protocol processing thread.
1074          */
1075         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1076         KKASSERT(so->so_port == &curthread->td_msgport);
1077
1078         /* Unscale the window into a 32-bit value. */
1079         if (!(thflags & TH_SYN))
1080                 tiwin = th->th_win << tp->snd_scale;
1081         else
1082                 tiwin = th->th_win;
1083
1084         /*
1085          * This is the second part of the MSS DoS prevention code (after
1086          * minmss on the sending side) and it deals with too many too small
1087          * tcp packets in a too short timeframe (1 second).
1088          *
1089          * XXX Removed.  This code was crap.  It does not scale to network
1090          *     speed, and default values break NFS.  Gone.
1091          */
1092         /* REMOVED */
1093
1094         /*
1095          * Segment received on connection.
1096          *
1097          * Reset idle time and keep-alive timer.  Don't waste time if less
1098          * then a second has elapsed.
1099          */
1100         if ((int)(ticks - tp->t_rcvtime) > hz)
1101                 tcp_timer_keep_activity(tp, thflags);
1102
1103         /*
1104          * Process options.
1105          * XXX this is tradtitional behavior, may need to be cleaned up.
1106          */
1107         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack);
1108         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1109                 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
1110                         tp->t_flags |= TF_RCVD_SCALE;
1111                         tp->snd_scale = to.to_requested_s_scale;
1112                 }
1113
1114                 /*
1115                  * Initial send window; will be updated upon next ACK
1116                  */
1117                 tp->snd_wnd = th->th_win;
1118
1119                 if (to.to_flags & TOF_TS) {
1120                         tp->t_flags |= TF_RCVD_TSTMP;
1121                         tp->ts_recent = to.to_tsval;
1122                         tp->ts_recent_age = ticks;
1123                 }
1124                 if (!(to.to_flags & TOF_MSS))
1125                         to.to_mss = 0;
1126                 tcp_mss(tp, to.to_mss);
1127                 /*
1128                  * Only set the TF_SACK_PERMITTED per-connection flag
1129                  * if we got a SACK_PERMITTED option from the other side
1130                  * and the global tcp_do_sack variable is true.
1131                  */
1132                 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1133                         tp->t_flags |= TF_SACK_PERMITTED;
1134         }
1135
1136         /*
1137          * Header prediction: check for the two common cases
1138          * of a uni-directional data xfer.  If the packet has
1139          * no control flags, is in-sequence, the window didn't
1140          * change and we're not retransmitting, it's a
1141          * candidate.  If the length is zero and the ack moved
1142          * forward, we're the sender side of the xfer.  Just
1143          * free the data acked & wake any higher level process
1144          * that was blocked waiting for space.  If the length
1145          * is non-zero and the ack didn't move, we're the
1146          * receiver side.  If we're getting packets in-order
1147          * (the reassembly queue is empty), add the data to
1148          * the socket buffer and note that we need a delayed ack.
1149          * Make sure that the hidden state-flags are also off.
1150          * Since we check for TCPS_ESTABLISHED above, it can only
1151          * be TH_NEEDSYN.
1152          */
1153         if (tp->t_state == TCPS_ESTABLISHED &&
1154             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1155             !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1156             (!(to.to_flags & TOF_TS) ||
1157              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1158             th->th_seq == tp->rcv_nxt &&
1159             tp->snd_nxt == tp->snd_max) {
1160
1161                 /*
1162                  * If last ACK falls within this segment's sequence numbers,
1163                  * record the timestamp.
1164                  * NOTE that the test is modified according to the latest
1165                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1166                  */
1167                 if ((to.to_flags & TOF_TS) &&
1168                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1169                         tp->ts_recent_age = ticks;
1170                         tp->ts_recent = to.to_tsval;
1171                 }
1172
1173                 if (tlen == 0) {
1174                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1175                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1176                             tp->snd_cwnd >= tp->snd_wnd &&
1177                             !IN_FASTRECOVERY(tp)) {
1178                                 /*
1179                                  * This is a pure ack for outstanding data.
1180                                  */
1181                                 ++tcpstat.tcps_predack;
1182                                 /*
1183                                  * "bad retransmit" recovery
1184                                  *
1185                                  * If Eifel detection applies, then
1186                                  * it is deterministic, so use it
1187                                  * unconditionally over the old heuristic.
1188                                  * Otherwise, fall back to the old heuristic.
1189                                  */
1190                                 if (tcp_do_eifel_detect &&
1191                                     (to.to_flags & TOF_TS) && to.to_tsecr &&
1192                                     (tp->t_flags & TF_FIRSTACCACK)) {
1193                                         /* Eifel detection applicable. */
1194                                         if (to.to_tsecr < tp->t_rexmtTS) {
1195                                                 tcp_revert_congestion_state(tp);
1196                                                 ++tcpstat.tcps_eifeldetected;
1197                                                 if (tp->t_rxtshift != 1 ||
1198                                                     ticks >= tp->t_badrxtwin)
1199                                                         ++tcpstat.tcps_rttcantdetect;
1200                                         }
1201                                 } else if (tp->t_rxtshift == 1 &&
1202                                            ticks < tp->t_badrxtwin) {
1203                                         tcp_revert_congestion_state(tp);
1204                                         ++tcpstat.tcps_rttdetected;
1205                                 }
1206                                 tp->t_flags &= ~(TF_FIRSTACCACK |
1207                                                  TF_FASTREXMT | TF_EARLYREXMT);
1208                                 /*
1209                                  * Recalculate the retransmit timer / rtt.
1210                                  *
1211                                  * Some machines (certain windows boxes)
1212                                  * send broken timestamp replies during the
1213                                  * SYN+ACK phase, ignore timestamps of 0.
1214                                  */
1215                                 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1216                                         tcp_xmit_timer(tp,
1217                                             ticks - to.to_tsecr + 1,
1218                                             th->th_ack);
1219                                 } else if (tp->t_rtttime &&
1220                                            SEQ_GT(th->th_ack, tp->t_rtseq)) {
1221                                         tcp_xmit_timer(tp,
1222                                             ticks - tp->t_rtttime,
1223                                             th->th_ack);
1224                                 }
1225                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1226                                 acked = th->th_ack - tp->snd_una;
1227                                 tcpstat.tcps_rcvackpack++;
1228                                 tcpstat.tcps_rcvackbyte += acked;
1229                                 sbdrop(&so->so_snd.sb, acked);
1230                                 tp->snd_recover = th->th_ack - 1;
1231                                 tp->snd_una = th->th_ack;
1232                                 tp->t_dupacks = 0;
1233                                 /*
1234                                  * Update window information.
1235                                  */
1236                                 if (tiwin != tp->snd_wnd &&
1237                                     acceptable_window_update(tp, th, tiwin)) {
1238                                         /* keep track of pure window updates */
1239                                         if (tp->snd_wl2 == th->th_ack &&
1240                                             tiwin > tp->snd_wnd)
1241                                                 tcpstat.tcps_rcvwinupd++;
1242                                         tp->snd_wnd = tiwin;
1243                                         tp->snd_wl1 = th->th_seq;
1244                                         tp->snd_wl2 = th->th_ack;
1245                                         if (tp->snd_wnd > tp->max_sndwnd)
1246                                                 tp->max_sndwnd = tp->snd_wnd;
1247                                 }
1248                                 m_freem(m);
1249                                 ND6_HINT(tp); /* some progress has been done */
1250                                 /*
1251                                  * If all outstanding data are acked, stop
1252                                  * retransmit timer, otherwise restart timer
1253                                  * using current (possibly backed-off) value.
1254                                  * If process is waiting for space,
1255                                  * wakeup/selwakeup/signal.  If data
1256                                  * are ready to send, let tcp_output
1257                                  * decide between more output or persist.
1258                                  */
1259                                 if (tp->snd_una == tp->snd_max) {
1260                                         tcp_callout_stop(tp, tp->tt_rexmt);
1261                                 } else if (!tcp_callout_active(tp,
1262                                             tp->tt_persist)) {
1263                                         tcp_callout_reset(tp, tp->tt_rexmt,
1264                                             tp->t_rxtcur, tcp_timer_rexmt);
1265                                 }
1266                                 sowwakeup(so);
1267                                 if (so->so_snd.ssb_cc > 0)
1268                                         tcp_output(tp);
1269                                 return(IPPROTO_DONE);
1270                         }
1271                 } else if (tiwin == tp->snd_wnd &&
1272                     th->th_ack == tp->snd_una &&
1273                     LIST_EMPTY(&tp->t_segq) &&
1274                     tlen <= ssb_space(&so->so_rcv)) {
1275                         u_long newsize = 0;     /* automatic sockbuf scaling */
1276                         /*
1277                          * This is a pure, in-sequence data packet
1278                          * with nothing on the reassembly queue and
1279                          * we have enough buffer space to take it.
1280                          */
1281                         ++tcpstat.tcps_preddat;
1282                         tp->rcv_nxt += tlen;
1283                         tcpstat.tcps_rcvpack++;
1284                         tcpstat.tcps_rcvbyte += tlen;
1285                         ND6_HINT(tp);   /* some progress has been done */
1286                 /*
1287                  * Automatic sizing of receive socket buffer.  Often the send
1288                  * buffer size is not optimally adjusted to the actual network
1289                  * conditions at hand (delay bandwidth product).  Setting the
1290                  * buffer size too small limits throughput on links with high
1291                  * bandwidth and high delay (eg. trans-continental/oceanic links).
1292                  *
1293                  * On the receive side the socket buffer memory is only rarely
1294                  * used to any significant extent.  This allows us to be much
1295                  * more aggressive in scaling the receive socket buffer.  For
1296                  * the case that the buffer space is actually used to a large
1297                  * extent and we run out of kernel memory we can simply drop
1298                  * the new segments; TCP on the sender will just retransmit it
1299                  * later.  Setting the buffer size too big may only consume too
1300                  * much kernel memory if the application doesn't read() from
1301                  * the socket or packet loss or reordering makes use of the
1302                  * reassembly queue.
1303                  *
1304                  * The criteria to step up the receive buffer one notch are:
1305                  *  1. the number of bytes received during the time it takes
1306                  *     one timestamp to be reflected back to us (the RTT);
1307                  *  2. received bytes per RTT is within seven eighth of the
1308                  *     current socket buffer size;
1309                  *  3. receive buffer size has not hit maximal automatic size;
1310                  *
1311                  * This algorithm does one step per RTT at most and only if
1312                  * we receive a bulk stream w/o packet losses or reorderings.
1313                  * Shrinking the buffer during idle times is not necessary as
1314                  * it doesn't consume any memory when idle.
1315                  *
1316                  * TODO: Only step up if the application is actually serving
1317                  * the buffer to better manage the socket buffer resources.
1318                  */
1319                         if (tcp_do_autorcvbuf &&
1320                             to.to_tsecr &&
1321                             (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1322                                 if (to.to_tsecr > tp->rfbuf_ts &&
1323                                     to.to_tsecr - tp->rfbuf_ts < hz) {
1324                                         if (tp->rfbuf_cnt >
1325                                             (so->so_rcv.ssb_hiwat / 8 * 7) &&
1326                                             so->so_rcv.ssb_hiwat <
1327                                             tcp_autorcvbuf_max) {
1328                                                 newsize =
1329                                                     ulmin(so->so_rcv.ssb_hiwat +
1330                                                           tcp_autorcvbuf_inc,
1331                                                           tcp_autorcvbuf_max);
1332                                         }
1333                                         /* Start over with next RTT. */
1334                                         tp->rfbuf_ts = 0;
1335                                         tp->rfbuf_cnt = 0;
1336                                 } else
1337                                         tp->rfbuf_cnt += tlen;  /* add up */
1338                         }
1339                         /*
1340                          * Add data to socket buffer.
1341                          */
1342                         if (so->so_state & SS_CANTRCVMORE) {
1343                                 m_freem(m);
1344                         } else {
1345                                 /*
1346                                  * Set new socket buffer size, give up when
1347                                  * limit is reached.
1348                                  *
1349                                  * Adjusting the size can mess up ACK
1350                                  * sequencing when pure window updates are
1351                                  * being avoided (which is the default),
1352                                  * so force an ack.
1353                                  */
1354                                 lwkt_gettoken(&so->so_rcv.ssb_token);
1355                                 if (newsize) {
1356                                         tp->t_flags |= TF_RXRESIZED;
1357                                         if (!ssb_reserve(&so->so_rcv, newsize,
1358                                                          so, NULL)) {
1359                                                 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1360                                         }
1361                                         if (newsize >=
1362                                             (TCP_MAXWIN << tp->rcv_scale)) {
1363                                                 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1364                                         }
1365                                 }
1366                                 m_adj(m, drop_hdrlen); /* delayed header drop */
1367                                 ssb_appendstream(&so->so_rcv, m);
1368                                 lwkt_reltoken(&so->so_rcv.ssb_token);
1369                         }
1370                         sorwakeup(so);
1371                         /*
1372                          * This code is responsible for most of the ACKs
1373                          * the TCP stack sends back after receiving a data
1374                          * packet.  Note that the DELAY_ACK check fails if
1375                          * the delack timer is already running, which results
1376                          * in an ack being sent every other packet (which is
1377                          * what we want).
1378                          *
1379                          * We then further aggregate acks by not actually
1380                          * sending one until the protocol thread has completed
1381                          * processing the current backlog of packets.  This
1382                          * does not delay the ack any further, but allows us
1383                          * to take advantage of the packet aggregation that
1384                          * high speed NICs do (usually blocks of 8-10 packets)
1385                          * to send a single ack rather then four or five acks,
1386                          * greatly reducing the ack rate, the return channel
1387                          * bandwidth, and the protocol overhead on both ends.
1388                          *
1389                          * Since this also has the effect of slowing down
1390                          * the exponential slow-start ramp-up, systems with 
1391                          * very large bandwidth-delay products might want
1392                          * to turn the feature off.
1393                          */
1394                         if (DELAY_ACK(tp)) {
1395                                 tcp_callout_reset(tp, tp->tt_delack,
1396                                     tcp_delacktime, tcp_timer_delack);
1397                         } else if (tcp_aggregate_acks) {
1398                                 tp->t_flags |= TF_ACKNOW;
1399                                 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1400                                         tp->t_flags |= TF_ONOUTPUTQ;
1401                                         tp->tt_cpu = mycpu->gd_cpuid;
1402                                         TAILQ_INSERT_TAIL(
1403                                             &tcpcbackq[tp->tt_cpu],
1404                                             tp, t_outputq);
1405                                 }
1406                         } else {
1407                                 tp->t_flags |= TF_ACKNOW;
1408                                 tcp_output(tp);
1409                         }
1410                         return(IPPROTO_DONE);
1411                 }
1412         }
1413
1414         /*
1415          * Calculate amount of space in receive window,
1416          * and then do TCP input processing.
1417          * Receive window is amount of space in rcv queue,
1418          * but not less than advertised window.
1419          */
1420         recvwin = ssb_space(&so->so_rcv);
1421         if (recvwin < 0)
1422                 recvwin = 0;
1423         tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1424
1425         /* Reset receive buffer auto scaling when not in bulk receive mode. */
1426         tp->rfbuf_ts = 0;
1427         tp->rfbuf_cnt = 0;
1428
1429         switch (tp->t_state) {
1430         /*
1431          * If the state is SYN_RECEIVED:
1432          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1433          */
1434         case TCPS_SYN_RECEIVED:
1435                 if ((thflags & TH_ACK) &&
1436                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1437                      SEQ_GT(th->th_ack, tp->snd_max))) {
1438                                 rstreason = BANDLIM_RST_OPENPORT;
1439                                 goto dropwithreset;
1440                 }
1441                 break;
1442
1443         /*
1444          * If the state is SYN_SENT:
1445          *      if seg contains an ACK, but not for our SYN, drop the input.
1446          *      if seg contains a RST, then drop the connection.
1447          *      if seg does not contain SYN, then drop it.
1448          * Otherwise this is an acceptable SYN segment
1449          *      initialize tp->rcv_nxt and tp->irs
1450          *      if seg contains ack then advance tp->snd_una
1451          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1452          *      arrange for segment to be acked (eventually)
1453          *      continue processing rest of data/controls, beginning with URG
1454          */
1455         case TCPS_SYN_SENT:
1456                 if ((thflags & TH_ACK) &&
1457                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1458                      SEQ_GT(th->th_ack, tp->snd_max))) {
1459                         rstreason = BANDLIM_UNLIMITED;
1460                         goto dropwithreset;
1461                 }
1462                 if (thflags & TH_RST) {
1463                         if (thflags & TH_ACK)
1464                                 tp = tcp_drop(tp, ECONNREFUSED);
1465                         goto drop;
1466                 }
1467                 if (!(thflags & TH_SYN))
1468                         goto drop;
1469
1470                 tp->irs = th->th_seq;
1471                 tcp_rcvseqinit(tp);
1472                 if (thflags & TH_ACK) {
1473                         /* Our SYN was acked. */
1474                         tcpstat.tcps_connects++;
1475                         soisconnected(so);
1476                         /* Do window scaling on this connection? */
1477                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1478                             (TF_RCVD_SCALE | TF_REQ_SCALE))
1479                                 tp->rcv_scale = tp->request_r_scale;
1480                         tp->rcv_adv += tp->rcv_wnd;
1481                         tp->snd_una++;          /* SYN is acked */
1482                         tcp_callout_stop(tp, tp->tt_rexmt);
1483                         /*
1484                          * If there's data, delay ACK; if there's also a FIN
1485                          * ACKNOW will be turned on later.
1486                          */
1487                         if (DELAY_ACK(tp) && tlen != 0) {
1488                                 tcp_callout_reset(tp, tp->tt_delack,
1489                                     tcp_delacktime, tcp_timer_delack);
1490                         } else {
1491                                 tp->t_flags |= TF_ACKNOW;
1492                         }
1493                         /*
1494                          * Received <SYN,ACK> in SYN_SENT[*] state.
1495                          * Transitions:
1496                          *      SYN_SENT  --> ESTABLISHED
1497                          *      SYN_SENT* --> FIN_WAIT_1
1498                          */
1499                         tp->t_starttime = ticks;
1500                         if (tp->t_flags & TF_NEEDFIN) {
1501                                 tp->t_state = TCPS_FIN_WAIT_1;
1502                                 tp->t_flags &= ~TF_NEEDFIN;
1503                                 thflags &= ~TH_SYN;
1504                         } else {
1505                                 tcp_established(tp);
1506                         }
1507                 } else {
1508                         /*
1509                          * Received initial SYN in SYN-SENT[*] state =>
1510                          * simultaneous open.
1511                          * Do 3-way handshake:
1512                          *        SYN-SENT -> SYN-RECEIVED
1513                          *        SYN-SENT* -> SYN-RECEIVED*
1514                          */
1515                         tp->t_flags |= TF_ACKNOW;
1516                         tcp_callout_stop(tp, tp->tt_rexmt);
1517                         tp->t_state = TCPS_SYN_RECEIVED;
1518                 }
1519
1520                 /*
1521                  * Advance th->th_seq to correspond to first data byte.
1522                  * If data, trim to stay within window,
1523                  * dropping FIN if necessary.
1524                  */
1525                 th->th_seq++;
1526                 if (tlen > tp->rcv_wnd) {
1527                         todrop = tlen - tp->rcv_wnd;
1528                         m_adj(m, -todrop);
1529                         tlen = tp->rcv_wnd;
1530                         thflags &= ~TH_FIN;
1531                         tcpstat.tcps_rcvpackafterwin++;
1532                         tcpstat.tcps_rcvbyteafterwin += todrop;
1533                 }
1534                 tp->snd_wl1 = th->th_seq - 1;
1535                 tp->rcv_up = th->th_seq;
1536                 /*
1537                  * Client side of transaction: already sent SYN and data.
1538                  * If the remote host used T/TCP to validate the SYN,
1539                  * our data will be ACK'd; if so, enter normal data segment
1540                  * processing in the middle of step 5, ack processing.
1541                  * Otherwise, goto step 6.
1542                  */
1543                 if (thflags & TH_ACK)
1544                         goto process_ACK;
1545
1546                 goto step6;
1547
1548         /*
1549          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1550          *      do normal processing (we no longer bother with T/TCP).
1551          */
1552         case TCPS_LAST_ACK:
1553         case TCPS_CLOSING:
1554         case TCPS_TIME_WAIT:
1555                 break;  /* continue normal processing */
1556         }
1557
1558         /*
1559          * States other than LISTEN or SYN_SENT.
1560          * First check the RST flag and sequence number since reset segments
1561          * are exempt from the timestamp and connection count tests.  This
1562          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1563          * below which allowed reset segments in half the sequence space
1564          * to fall though and be processed (which gives forged reset
1565          * segments with a random sequence number a 50 percent chance of
1566          * killing a connection).
1567          * Then check timestamp, if present.
1568          * Then check the connection count, if present.
1569          * Then check that at least some bytes of segment are within
1570          * receive window.  If segment begins before rcv_nxt,
1571          * drop leading data (and SYN); if nothing left, just ack.
1572          *
1573          *
1574          * If the RST bit is set, check the sequence number to see
1575          * if this is a valid reset segment.
1576          * RFC 793 page 37:
1577          *   In all states except SYN-SENT, all reset (RST) segments
1578          *   are validated by checking their SEQ-fields.  A reset is
1579          *   valid if its sequence number is in the window.
1580          * Note: this does not take into account delayed ACKs, so
1581          *   we should test against last_ack_sent instead of rcv_nxt.
1582          *   The sequence number in the reset segment is normally an
1583          *   echo of our outgoing acknowledgement numbers, but some hosts
1584          *   send a reset with the sequence number at the rightmost edge
1585          *   of our receive window, and we have to handle this case.
1586          * If we have multiple segments in flight, the intial reset
1587          * segment sequence numbers will be to the left of last_ack_sent,
1588          * but they will eventually catch up.
1589          * In any case, it never made sense to trim reset segments to
1590          * fit the receive window since RFC 1122 says:
1591          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1592          *
1593          *    A TCP SHOULD allow a received RST segment to include data.
1594          *
1595          *    DISCUSSION
1596          *         It has been suggested that a RST segment could contain
1597          *         ASCII text that encoded and explained the cause of the
1598          *         RST.  No standard has yet been established for such
1599          *         data.
1600          *
1601          * If the reset segment passes the sequence number test examine
1602          * the state:
1603          *    SYN_RECEIVED STATE:
1604          *      If passive open, return to LISTEN state.
1605          *      If active open, inform user that connection was refused.
1606          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1607          *      Inform user that connection was reset, and close tcb.
1608          *    CLOSING, LAST_ACK STATES:
1609          *      Close the tcb.
1610          *    TIME_WAIT STATE:
1611          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1612          *      RFC 1337.
1613          */
1614         if (thflags & TH_RST) {
1615                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1616                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1617                         switch (tp->t_state) {
1618
1619                         case TCPS_SYN_RECEIVED:
1620                                 so->so_error = ECONNREFUSED;
1621                                 goto close;
1622
1623                         case TCPS_ESTABLISHED:
1624                         case TCPS_FIN_WAIT_1:
1625                         case TCPS_FIN_WAIT_2:
1626                         case TCPS_CLOSE_WAIT:
1627                                 so->so_error = ECONNRESET;
1628                         close:
1629                                 tp->t_state = TCPS_CLOSED;
1630                                 tcpstat.tcps_drops++;
1631                                 tp = tcp_close(tp);
1632                                 break;
1633
1634                         case TCPS_CLOSING:
1635                         case TCPS_LAST_ACK:
1636                                 tp = tcp_close(tp);
1637                                 break;
1638
1639                         case TCPS_TIME_WAIT:
1640                                 break;
1641                         }
1642                 }
1643                 goto drop;
1644         }
1645
1646         /*
1647          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1648          * and it's less than ts_recent, drop it.
1649          */
1650         if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1651             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1652
1653                 /* Check to see if ts_recent is over 24 days old.  */
1654                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1655                         /*
1656                          * Invalidate ts_recent.  If this segment updates
1657                          * ts_recent, the age will be reset later and ts_recent
1658                          * will get a valid value.  If it does not, setting
1659                          * ts_recent to zero will at least satisfy the
1660                          * requirement that zero be placed in the timestamp
1661                          * echo reply when ts_recent isn't valid.  The
1662                          * age isn't reset until we get a valid ts_recent
1663                          * because we don't want out-of-order segments to be
1664                          * dropped when ts_recent is old.
1665                          */
1666                         tp->ts_recent = 0;
1667                 } else {
1668                         tcpstat.tcps_rcvduppack++;
1669                         tcpstat.tcps_rcvdupbyte += tlen;
1670                         tcpstat.tcps_pawsdrop++;
1671                         if (tlen)
1672                                 goto dropafterack;
1673                         goto drop;
1674                 }
1675         }
1676
1677         /*
1678          * In the SYN-RECEIVED state, validate that the packet belongs to
1679          * this connection before trimming the data to fit the receive
1680          * window.  Check the sequence number versus IRS since we know
1681          * the sequence numbers haven't wrapped.  This is a partial fix
1682          * for the "LAND" DoS attack.
1683          */
1684         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1685                 rstreason = BANDLIM_RST_OPENPORT;
1686                 goto dropwithreset;
1687         }
1688
1689         todrop = tp->rcv_nxt - th->th_seq;
1690         if (todrop > 0) {
1691                 if (TCP_DO_SACK(tp)) {
1692                         /* Report duplicate segment at head of packet. */
1693                         tp->reportblk.rblk_start = th->th_seq;
1694                         tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1695                             th->th_seq + tlen, thflags);
1696                         if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1697                                 tp->reportblk.rblk_end = tp->rcv_nxt;
1698                         tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1699                 }
1700                 if (thflags & TH_SYN) {
1701                         thflags &= ~TH_SYN;
1702                         th->th_seq++;
1703                         if (th->th_urp > 1)
1704                                 th->th_urp--;
1705                         else
1706                                 thflags &= ~TH_URG;
1707                         todrop--;
1708                 }
1709                 /*
1710                  * Following if statement from Stevens, vol. 2, p. 960.
1711                  */
1712                 if (todrop > tlen ||
1713                     (todrop == tlen && !(thflags & TH_FIN))) {
1714                         /*
1715                          * Any valid FIN must be to the left of the window.
1716                          * At this point the FIN must be a duplicate or out
1717                          * of sequence; drop it.
1718                          */
1719                         thflags &= ~TH_FIN;
1720
1721                         /*
1722                          * Send an ACK to resynchronize and drop any data.
1723                          * But keep on processing for RST or ACK.
1724                          */
1725                         tp->t_flags |= TF_ACKNOW;
1726                         todrop = tlen;
1727                         tcpstat.tcps_rcvduppack++;
1728                         tcpstat.tcps_rcvdupbyte += todrop;
1729                 } else {
1730                         tcpstat.tcps_rcvpartduppack++;
1731                         tcpstat.tcps_rcvpartdupbyte += todrop;
1732                 }
1733                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1734                 th->th_seq += todrop;
1735                 tlen -= todrop;
1736                 if (th->th_urp > todrop)
1737                         th->th_urp -= todrop;
1738                 else {
1739                         thflags &= ~TH_URG;
1740                         th->th_urp = 0;
1741                 }
1742         }
1743
1744         /*
1745          * If new data are received on a connection after the
1746          * user processes are gone, then RST the other end.
1747          */
1748         if ((so->so_state & SS_NOFDREF) &&
1749             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1750                 tp = tcp_close(tp);
1751                 tcpstat.tcps_rcvafterclose++;
1752                 rstreason = BANDLIM_UNLIMITED;
1753                 goto dropwithreset;
1754         }
1755
1756         /*
1757          * If segment ends after window, drop trailing data
1758          * (and PUSH and FIN); if nothing left, just ACK.
1759          */
1760         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1761         if (todrop > 0) {
1762                 tcpstat.tcps_rcvpackafterwin++;
1763                 if (todrop >= tlen) {
1764                         tcpstat.tcps_rcvbyteafterwin += tlen;
1765                         /*
1766                          * If a new connection request is received
1767                          * while in TIME_WAIT, drop the old connection
1768                          * and start over if the sequence numbers
1769                          * are above the previous ones.
1770                          */
1771                         if (thflags & TH_SYN &&
1772                             tp->t_state == TCPS_TIME_WAIT &&
1773                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1774                                 tp = tcp_close(tp);
1775                                 goto findpcb;
1776                         }
1777                         /*
1778                          * If window is closed can only take segments at
1779                          * window edge, and have to drop data and PUSH from
1780                          * incoming segments.  Continue processing, but
1781                          * remember to ack.  Otherwise, drop segment
1782                          * and ack.
1783                          */
1784                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1785                                 tp->t_flags |= TF_ACKNOW;
1786                                 tcpstat.tcps_rcvwinprobe++;
1787                         } else
1788                                 goto dropafterack;
1789                 } else
1790                         tcpstat.tcps_rcvbyteafterwin += todrop;
1791                 m_adj(m, -todrop);
1792                 tlen -= todrop;
1793                 thflags &= ~(TH_PUSH | TH_FIN);
1794         }
1795
1796         /*
1797          * If last ACK falls within this segment's sequence numbers,
1798          * record its timestamp.
1799          * NOTE:
1800          * 1) That the test incorporates suggestions from the latest
1801          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1802          * 2) That updating only on newer timestamps interferes with
1803          *    our earlier PAWS tests, so this check should be solely
1804          *    predicated on the sequence space of this segment.
1805          * 3) That we modify the segment boundary check to be
1806          *        Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1807          *    instead of RFC1323's
1808          *        Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1809          *    This modified check allows us to overcome RFC1323's
1810          *    limitations as described in Stevens TCP/IP Illustrated
1811          *    Vol. 2 p.869. In such cases, we can still calculate the
1812          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1813          */
1814         if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1815             SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1816                                         + ((thflags & TH_SYN) != 0)
1817                                         + ((thflags & TH_FIN) != 0)))) {
1818                 tp->ts_recent_age = ticks;
1819                 tp->ts_recent = to.to_tsval;
1820         }
1821
1822         /*
1823          * If a SYN is in the window, then this is an
1824          * error and we send an RST and drop the connection.
1825          */
1826         if (thflags & TH_SYN) {
1827                 tp = tcp_drop(tp, ECONNRESET);
1828                 rstreason = BANDLIM_UNLIMITED;
1829                 goto dropwithreset;
1830         }
1831
1832         /*
1833          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1834          * flag is on (half-synchronized state), then queue data for
1835          * later processing; else drop segment and return.
1836          */
1837         if (!(thflags & TH_ACK)) {
1838                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1839                     (tp->t_flags & TF_NEEDSYN))
1840                         goto step6;
1841                 else
1842                         goto drop;
1843         }
1844
1845         /*
1846          * Ack processing.
1847          */
1848         switch (tp->t_state) {
1849         /*
1850          * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1851          * ESTABLISHED state and continue processing.
1852          * The ACK was checked above.
1853          */
1854         case TCPS_SYN_RECEIVED:
1855
1856                 tcpstat.tcps_connects++;
1857                 soisconnected(so);
1858                 /* Do window scaling? */
1859                 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1860                     (TF_RCVD_SCALE | TF_REQ_SCALE))
1861                         tp->rcv_scale = tp->request_r_scale;
1862                 /*
1863                  * Make transitions:
1864                  *      SYN-RECEIVED  -> ESTABLISHED
1865                  *      SYN-RECEIVED* -> FIN-WAIT-1
1866                  */
1867                 tp->t_starttime = ticks;
1868                 if (tp->t_flags & TF_NEEDFIN) {
1869                         tp->t_state = TCPS_FIN_WAIT_1;
1870                         tp->t_flags &= ~TF_NEEDFIN;
1871                 } else {
1872                         tcp_established(tp);
1873                 }
1874                 /*
1875                  * If segment contains data or ACK, will call tcp_reass()
1876                  * later; if not, do so now to pass queued data to user.
1877                  */
1878                 if (tlen == 0 && !(thflags & TH_FIN))
1879                         tcp_reass(tp, NULL, NULL, NULL);
1880                 /* fall into ... */
1881
1882         /*
1883          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1884          * ACKs.  If the ack is in the range
1885          *      tp->snd_una < th->th_ack <= tp->snd_max
1886          * then advance tp->snd_una to th->th_ack and drop
1887          * data from the retransmission queue.  If this ACK reflects
1888          * more up to date window information we update our window information.
1889          */
1890         case TCPS_ESTABLISHED:
1891         case TCPS_FIN_WAIT_1:
1892         case TCPS_FIN_WAIT_2:
1893         case TCPS_CLOSE_WAIT:
1894         case TCPS_CLOSING:
1895         case TCPS_LAST_ACK:
1896         case TCPS_TIME_WAIT:
1897
1898                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1899                         if (TCP_DO_SACK(tp))
1900                                 tcp_sack_update_scoreboard(tp, &to);
1901                         if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1902                             th->th_ack != tp->snd_una) {
1903                                 tcpstat.tcps_rcvdupack++;
1904                                 tp->t_dupacks = 0;
1905                                 break;
1906                         }
1907                         if (tlen != 0 || tiwin != tp->snd_wnd) {
1908                                 if (!tcp_do_rfc3517bis ||
1909                                     !TCP_DO_SACK(tp) ||
1910                                     (to.to_flags &
1911                                      (TOF_SACK | TOF_SACK_REDUNDANT))
1912                                      != TOF_SACK) {
1913                                         tp->t_dupacks = 0;
1914                                         break;
1915                                 }
1916                                 /*
1917                                  * Update window information.
1918                                  */
1919                                 if (tiwin != tp->snd_wnd &&
1920                                     acceptable_window_update(tp, th, tiwin)) {
1921                                         /* keep track of pure window updates */
1922                                         if (tlen == 0 &&
1923                                             tp->snd_wl2 == th->th_ack &&
1924                                             tiwin > tp->snd_wnd)
1925                                                 tcpstat.tcps_rcvwinupd++;
1926                                         tp->snd_wnd = tiwin;
1927                                         tp->snd_wl1 = th->th_seq;
1928                                         tp->snd_wl2 = th->th_ack;
1929                                         if (tp->snd_wnd > tp->max_sndwnd)
1930                                                 tp->max_sndwnd = tp->snd_wnd;
1931                                 }
1932                         }
1933                         tcpstat.tcps_rcvdupack++;
1934
1935                         /*
1936                          * We have outstanding data (other than
1937                          * a window probe), this is a completely
1938                          * duplicate ack (ie, window info didn't
1939                          * change), the ack is the biggest we've
1940                          * seen and we've seen exactly our rexmt
1941                          * threshhold of them, so assume a packet
1942                          * has been dropped and retransmit it.
1943                          * Kludge snd_nxt & the congestion
1944                          * window so we send only this one
1945                          * packet.
1946                          */
1947                         if (IN_FASTRECOVERY(tp)) {
1948                                 if (TCP_DO_SACK(tp)) {
1949                                         /* No artifical cwnd inflation. */
1950                                         tcp_sack_rexmt(tp, th);
1951                                 } else {
1952                                         /*
1953                                          * Dup acks mean that packets
1954                                          * have left the network
1955                                          * (they're now cached at the
1956                                          * receiver) so bump cwnd by
1957                                          * the amount in the receiver
1958                                          * to keep a constant cwnd
1959                                          * packets in the network.
1960                                          */
1961                                         tp->snd_cwnd += tp->t_maxseg;
1962                                         tcp_output(tp);
1963                                 }
1964                         } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1965                                 tp->t_dupacks = 0;
1966                                 break;
1967                         } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) &&
1968                             (to.to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) ==
1969                             (TOF_DSACK | TOF_SACK_REDUNDANT)) {
1970                                 /*
1971                                  * If the ACK carries DSACK and other
1972                                  * SACK blocks carry information that
1973                                  * we have already known, don't count
1974                                  * this ACK as duplicate ACK.  This
1975                                  * prevents spurious early retransmit
1976                                  * and fast retransmit.  This also
1977                                  * meets the requirement of RFC3042
1978                                  * that new segments should not be sent
1979                                  * if the SACK blocks do not contain
1980                                  * new information (XXX we actually
1981                                  * loosen the requirment that only DSACK
1982                                  * is checked here).
1983                                  *
1984                                  * This kind of ACKs are usually sent
1985                                  * after spurious retransmit.
1986                                  */
1987                                 /* Do nothing; don't change t_dupacks */
1988                         } else if (++tp->t_dupacks == tp->t_rxtthresh) {
1989                                 tcp_seq old_snd_nxt;
1990                                 u_int win;
1991
1992 fastretransmit:
1993                                 if (tcp_do_eifel_detect &&
1994                                     (tp->t_flags & TF_RCVD_TSTMP)) {
1995                                         tcp_save_congestion_state(tp);
1996                                         tp->t_flags |= TF_FASTREXMT;
1997                                 }
1998                                 /*
1999                                  * We know we're losing at the current
2000                                  * window size, so do congestion avoidance:
2001                                  * set ssthresh to half the current window
2002                                  * and pull our congestion window back to the
2003                                  * new ssthresh.
2004                                  */
2005                                 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
2006                                     tp->t_maxseg;
2007                                 if (win < 2)
2008                                         win = 2;
2009                                 tp->snd_ssthresh = win * tp->t_maxseg;
2010                                 ENTER_FASTRECOVERY(tp);
2011                                 tp->snd_recover = tp->snd_max;
2012                                 tcp_callout_stop(tp, tp->tt_rexmt);
2013                                 tp->t_rtttime = 0;
2014                                 old_snd_nxt = tp->snd_nxt;
2015                                 tp->snd_nxt = th->th_ack;
2016                                 tp->snd_cwnd = tp->t_maxseg;
2017                                 tcp_output(tp);
2018                                 ++tcpstat.tcps_sndfastrexmit;
2019                                 tp->snd_cwnd = tp->snd_ssthresh;
2020                                 tp->rexmt_high = tp->snd_nxt;
2021                                 tp->t_flags &= ~TF_SACKRESCUED;
2022                                 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
2023                                         tp->snd_nxt = old_snd_nxt;
2024                                 KASSERT(tp->snd_limited <= 2,
2025                                     ("tp->snd_limited too big"));
2026                                 if (TCP_DO_SACK(tp))
2027                                         tcp_sack_rexmt(tp, th);
2028                                 else
2029                                         tp->snd_cwnd += tp->t_maxseg *
2030                                             (tp->t_dupacks - tp->snd_limited);
2031                         } else if (tcp_do_rfc3517bis && TCP_DO_SACK(tp)) {
2032                                 if (tcp_sack_islost(&tp->scb, tp->snd_una))
2033                                         goto fastretransmit;
2034                                 if (tcp_do_limitedtransmit) {
2035                                         /* outstanding data */
2036                                         uint32_t ownd =
2037                                             tp->snd_max - tp->snd_una;
2038
2039                                         if (!tcp_sack_limitedxmit(tp) &&
2040                                             need_early_retransmit(tp, ownd)) {
2041                                                 ++tcpstat.tcps_sndearlyrexmit;
2042                                                 tp->t_flags |= TF_EARLYREXMT;
2043                                                 goto fastretransmit;
2044                                         }
2045                                 }
2046                         } else if (tcp_do_limitedtransmit) {
2047                                 u_long oldcwnd = tp->snd_cwnd;
2048                                 tcp_seq oldsndmax = tp->snd_max;
2049                                 tcp_seq oldsndnxt = tp->snd_nxt;
2050                                 /* outstanding data */
2051                                 uint32_t ownd = tp->snd_max - tp->snd_una;
2052                                 u_int sent;
2053
2054                                 KASSERT(tp->t_dupacks == 1 ||
2055                                         tp->t_dupacks == 2,
2056                                     ("dupacks not 1 or 2"));
2057                                 if (tp->t_dupacks == 1)
2058                                         tp->snd_limited = 0;
2059                                 tp->snd_nxt = tp->snd_max;
2060                                 tp->snd_cwnd = ownd +
2061                                     (tp->t_dupacks - tp->snd_limited) *
2062                                     tp->t_maxseg;
2063                                 tcp_output(tp);
2064
2065                                 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2066                                         KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
2067                                             ("snd_una moved in other threads"));
2068                                         tp->snd_nxt = oldsndnxt;
2069                                 }
2070                                 tp->snd_cwnd = oldcwnd;
2071                                 sent = tp->snd_max - oldsndmax;
2072                                 if (sent > tp->t_maxseg) {
2073                                         KASSERT((tp->t_dupacks == 2 &&
2074                                                  tp->snd_limited == 0) ||
2075                                                 (sent == tp->t_maxseg + 1 &&
2076                                                  tp->t_flags & TF_SENTFIN),
2077                                             ("sent too much"));
2078                                         KASSERT(sent <= tp->t_maxseg * 2,
2079                                             ("sent too many segments"));
2080                                         tp->snd_limited = 2;
2081                                         tcpstat.tcps_sndlimited += 2;
2082                                 } else if (sent > 0) {
2083                                         ++tp->snd_limited;
2084                                         ++tcpstat.tcps_sndlimited;
2085                                 } else if (need_early_retransmit(tp, ownd)) {
2086                                         ++tcpstat.tcps_sndearlyrexmit;
2087                                         tp->t_flags |= TF_EARLYREXMT;
2088                                         goto fastretransmit;
2089                                 }
2090                         }
2091                         if (tlen != 0)
2092                                 break;
2093                         else
2094                                 goto drop;
2095                 }
2096
2097                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2098                 tp->t_dupacks = 0;
2099                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2100                         /*
2101                          * Detected optimistic ACK attack.
2102                          * Force slow-start to de-synchronize attack.
2103                          */
2104                         tp->snd_cwnd = tp->t_maxseg;
2105                         tp->snd_wacked = 0;
2106
2107                         tcpstat.tcps_rcvacktoomuch++;
2108                         goto dropafterack;
2109                 }
2110                 /*
2111                  * If we reach this point, ACK is not a duplicate,
2112                  *     i.e., it ACKs something we sent.
2113                  */
2114                 if (tp->t_flags & TF_NEEDSYN) {
2115                         /*
2116                          * T/TCP: Connection was half-synchronized, and our
2117                          * SYN has been ACK'd (so connection is now fully
2118                          * synchronized).  Go to non-starred state,
2119                          * increment snd_una for ACK of SYN, and check if
2120                          * we can do window scaling.
2121                          */
2122                         tp->t_flags &= ~TF_NEEDSYN;
2123                         tp->snd_una++;
2124                         /* Do window scaling? */
2125                         if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2126                             (TF_RCVD_SCALE | TF_REQ_SCALE))
2127                                 tp->rcv_scale = tp->request_r_scale;
2128                 }
2129
2130 process_ACK:
2131                 acked = th->th_ack - tp->snd_una;
2132                 tcpstat.tcps_rcvackpack++;
2133                 tcpstat.tcps_rcvackbyte += acked;
2134
2135                 if (tcp_do_eifel_detect && acked > 0 &&
2136                     (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2137                     (tp->t_flags & TF_FIRSTACCACK)) {
2138                         /* Eifel detection applicable. */
2139                         if (to.to_tsecr < tp->t_rexmtTS) {
2140                                 ++tcpstat.tcps_eifeldetected;
2141                                 tcp_revert_congestion_state(tp);
2142                                 if (tp->t_rxtshift != 1 ||
2143                                     ticks >= tp->t_badrxtwin)
2144                                         ++tcpstat.tcps_rttcantdetect;
2145                         }
2146                 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2147                         /*
2148                          * If we just performed our first retransmit,
2149                          * and the ACK arrives within our recovery window,
2150                          * then it was a mistake to do the retransmit
2151                          * in the first place.  Recover our original cwnd
2152                          * and ssthresh, and proceed to transmit where we
2153                          * left off.
2154                          */
2155                         tcp_revert_congestion_state(tp);
2156                         ++tcpstat.tcps_rttdetected;
2157                 }
2158
2159                 /*
2160                  * If we have a timestamp reply, update smoothed
2161                  * round trip time.  If no timestamp is present but
2162                  * transmit timer is running and timed sequence
2163                  * number was acked, update smoothed round trip time.
2164                  * Since we now have an rtt measurement, cancel the
2165                  * timer backoff (cf., Phil Karn's retransmit alg.).
2166                  * Recompute the initial retransmit timer.
2167                  *
2168                  * Some machines (certain windows boxes) send broken
2169                  * timestamp replies during the SYN+ACK phase, ignore
2170                  * timestamps of 0.
2171                  */
2172                 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2173                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, th->th_ack);
2174                 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2175                         tcp_xmit_timer(tp, ticks - tp->t_rtttime, th->th_ack);
2176                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2177
2178                 /*
2179                  * If no data (only SYN) was ACK'd,
2180                  *    skip rest of ACK processing.
2181                  */
2182                 if (acked == 0)
2183                         goto step6;
2184
2185                 /* Stop looking for an acceptable ACK since one was received. */
2186                 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2187
2188                 if (acked > so->so_snd.ssb_cc) {
2189                         tp->snd_wnd -= so->so_snd.ssb_cc;
2190                         sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2191                         ourfinisacked = TRUE;
2192                 } else {
2193                         sbdrop(&so->so_snd.sb, acked);
2194                         tp->snd_wnd -= acked;
2195                         ourfinisacked = FALSE;
2196                 }
2197                 sowwakeup(so);
2198
2199                 /*
2200                  * Update window information.
2201                  */
2202                 if (acceptable_window_update(tp, th, tiwin)) {
2203                         /* keep track of pure window updates */
2204                         if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2205                             tiwin > tp->snd_wnd)
2206                                 tcpstat.tcps_rcvwinupd++;
2207                         tp->snd_wnd = tiwin;
2208                         tp->snd_wl1 = th->th_seq;
2209                         tp->snd_wl2 = th->th_ack;
2210                         if (tp->snd_wnd > tp->max_sndwnd)
2211                                 tp->max_sndwnd = tp->snd_wnd;
2212                         needoutput = TRUE;
2213                 }
2214
2215                 tp->snd_una = th->th_ack;
2216                 if (TCP_DO_SACK(tp))
2217                         tcp_sack_update_scoreboard(tp, &to);
2218                 if (IN_FASTRECOVERY(tp)) {
2219                         if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2220                                 EXIT_FASTRECOVERY(tp);
2221                                 needoutput = TRUE;
2222                                 /*
2223                                  * If the congestion window was inflated
2224                                  * to account for the other side's
2225                                  * cached packets, retract it.
2226                                  */
2227                                 if (!TCP_DO_SACK(tp))
2228                                         tp->snd_cwnd = tp->snd_ssthresh;
2229
2230                                 /*
2231                                  * Window inflation should have left us
2232                                  * with approximately snd_ssthresh outstanding
2233                                  * data.  But, in case we would be inclined
2234                                  * to send a burst, better do it using
2235                                  * slow start.
2236                                  */
2237                                 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2238                                            tp->snd_max + 2 * tp->t_maxseg))
2239                                         tp->snd_cwnd =
2240                                             (tp->snd_max - tp->snd_una) +
2241                                             2 * tp->t_maxseg;
2242
2243                                 tp->snd_wacked = 0;
2244                         } else {
2245                                 if (TCP_DO_SACK(tp)) {
2246                                         tp->snd_max_rexmt = tp->snd_max;
2247                                         tcp_sack_rexmt(tp, th);
2248                                 } else {
2249                                         tcp_newreno_partial_ack(tp, th, acked);
2250                                 }
2251                                 needoutput = FALSE;
2252                         }
2253                 } else {
2254                         /*
2255                          * Open the congestion window.  When in slow-start,
2256                          * open exponentially: maxseg per packet.  Otherwise,
2257                          * open linearly: maxseg per window.
2258                          */
2259                         if (tp->snd_cwnd <= tp->snd_ssthresh) {
2260                                 u_int abc_sslimit =
2261                                     (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2262                                      tp->t_maxseg : 2 * tp->t_maxseg);
2263
2264                                 /* slow-start */
2265                                 tp->snd_cwnd += tcp_do_abc ?
2266                                     min(acked, abc_sslimit) : tp->t_maxseg;
2267                         } else {
2268                                 /* linear increase */
2269                                 tp->snd_wacked += tcp_do_abc ? acked :
2270                                     tp->t_maxseg;
2271                                 if (tp->snd_wacked >= tp->snd_cwnd) {
2272                                         tp->snd_wacked -= tp->snd_cwnd;
2273                                         tp->snd_cwnd += tp->t_maxseg;
2274                                 }
2275                         }
2276                         tp->snd_cwnd = min(tp->snd_cwnd,
2277                                            TCP_MAXWIN << tp->snd_scale);
2278                         tp->snd_recover = th->th_ack - 1;
2279                 }
2280                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2281                         tp->snd_nxt = tp->snd_una;
2282
2283                 /*
2284                  * If all outstanding data is acked, stop retransmit
2285                  * timer and remember to restart (more output or persist).
2286                  * If there is more data to be acked, restart retransmit
2287                  * timer, using current (possibly backed-off) value.
2288                  */
2289                 if (th->th_ack == tp->snd_max) {
2290                         tcp_callout_stop(tp, tp->tt_rexmt);
2291                         needoutput = TRUE;
2292                 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2293                         tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2294                             tcp_timer_rexmt);
2295                 }
2296
2297                 switch (tp->t_state) {
2298                 /*
2299                  * In FIN_WAIT_1 STATE in addition to the processing
2300                  * for the ESTABLISHED state if our FIN is now acknowledged
2301                  * then enter FIN_WAIT_2.
2302                  */
2303                 case TCPS_FIN_WAIT_1:
2304                         if (ourfinisacked) {
2305                                 /*
2306                                  * If we can't receive any more
2307                                  * data, then closing user can proceed.
2308                                  * Starting the timer is contrary to the
2309                                  * specification, but if we don't get a FIN
2310                                  * we'll hang forever.
2311                                  */
2312                                 if (so->so_state & SS_CANTRCVMORE) {
2313                                         soisdisconnected(so);
2314                                         tcp_callout_reset(tp, tp->tt_2msl,
2315                                             tp->t_maxidle, tcp_timer_2msl);
2316                                 }
2317                                 tp->t_state = TCPS_FIN_WAIT_2;
2318                         }
2319                         break;
2320
2321                 /*
2322                  * In CLOSING STATE in addition to the processing for
2323                  * the ESTABLISHED state if the ACK acknowledges our FIN
2324                  * then enter the TIME-WAIT state, otherwise ignore
2325                  * the segment.
2326                  */
2327                 case TCPS_CLOSING:
2328                         if (ourfinisacked) {
2329                                 tp->t_state = TCPS_TIME_WAIT;
2330                                 tcp_canceltimers(tp);
2331                                 tcp_callout_reset(tp, tp->tt_2msl,
2332                                             2 * tcp_rmx_msl(tp),
2333                                             tcp_timer_2msl);
2334                                 soisdisconnected(so);
2335                         }
2336                         break;
2337
2338                 /*
2339                  * In LAST_ACK, we may still be waiting for data to drain
2340                  * and/or to be acked, as well as for the ack of our FIN.
2341                  * If our FIN is now acknowledged, delete the TCB,
2342                  * enter the closed state and return.
2343                  */
2344                 case TCPS_LAST_ACK:
2345                         if (ourfinisacked) {
2346                                 tp = tcp_close(tp);
2347                                 goto drop;
2348                         }
2349                         break;
2350
2351                 /*
2352                  * In TIME_WAIT state the only thing that should arrive
2353                  * is a retransmission of the remote FIN.  Acknowledge
2354                  * it and restart the finack timer.
2355                  */
2356                 case TCPS_TIME_WAIT:
2357                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2358                             tcp_timer_2msl);
2359                         goto dropafterack;
2360                 }
2361         }
2362
2363 step6:
2364         /*
2365          * Update window information.
2366          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2367          */
2368         if ((thflags & TH_ACK) &&
2369             acceptable_window_update(tp, th, tiwin)) {
2370                 /* keep track of pure window updates */
2371                 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2372                     tiwin > tp->snd_wnd)
2373                         tcpstat.tcps_rcvwinupd++;
2374                 tp->snd_wnd = tiwin;
2375                 tp->snd_wl1 = th->th_seq;
2376                 tp->snd_wl2 = th->th_ack;
2377                 if (tp->snd_wnd > tp->max_sndwnd)
2378                         tp->max_sndwnd = tp->snd_wnd;
2379                 needoutput = TRUE;
2380         }
2381
2382         /*
2383          * Process segments with URG.
2384          */
2385         if ((thflags & TH_URG) && th->th_urp &&
2386             !TCPS_HAVERCVDFIN(tp->t_state)) {
2387                 /*
2388                  * This is a kludge, but if we receive and accept
2389                  * random urgent pointers, we'll crash in
2390                  * soreceive.  It's hard to imagine someone
2391                  * actually wanting to send this much urgent data.
2392                  */
2393                 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2394                         th->th_urp = 0;                 /* XXX */
2395                         thflags &= ~TH_URG;             /* XXX */
2396                         goto dodata;                    /* XXX */
2397                 }
2398                 /*
2399                  * If this segment advances the known urgent pointer,
2400                  * then mark the data stream.  This should not happen
2401                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2402                  * a FIN has been received from the remote side.
2403                  * In these states we ignore the URG.
2404                  *
2405                  * According to RFC961 (Assigned Protocols),
2406                  * the urgent pointer points to the last octet
2407                  * of urgent data.  We continue, however,
2408                  * to consider it to indicate the first octet
2409                  * of data past the urgent section as the original
2410                  * spec states (in one of two places).
2411                  */
2412                 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2413                         tp->rcv_up = th->th_seq + th->th_urp;
2414                         so->so_oobmark = so->so_rcv.ssb_cc +
2415                             (tp->rcv_up - tp->rcv_nxt) - 1;
2416                         if (so->so_oobmark == 0)
2417                                 sosetstate(so, SS_RCVATMARK);
2418                         sohasoutofband(so);
2419                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2420                 }
2421                 /*
2422                  * Remove out of band data so doesn't get presented to user.
2423                  * This can happen independent of advancing the URG pointer,
2424                  * but if two URG's are pending at once, some out-of-band
2425                  * data may creep in... ick.
2426                  */
2427                 if (th->th_urp <= (u_long)tlen &&
2428                     !(so->so_options & SO_OOBINLINE)) {
2429                         /* hdr drop is delayed */
2430                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2431                 }
2432         } else {
2433                 /*
2434                  * If no out of band data is expected,
2435                  * pull receive urgent pointer along
2436                  * with the receive window.
2437                  */
2438                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2439                         tp->rcv_up = tp->rcv_nxt;
2440         }
2441
2442 dodata:                                                 /* XXX */
2443         /*
2444          * Process the segment text, merging it into the TCP sequencing queue,
2445          * and arranging for acknowledgment of receipt if necessary.
2446          * This process logically involves adjusting tp->rcv_wnd as data
2447          * is presented to the user (this happens in tcp_usrreq.c,
2448          * case PRU_RCVD).  If a FIN has already been received on this
2449          * connection then we just ignore the text.
2450          */
2451         if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2452                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2453                 /*
2454                  * Insert segment which includes th into TCP reassembly queue
2455                  * with control block tp.  Set thflags to whether reassembly now
2456                  * includes a segment with FIN.  This handles the common case
2457                  * inline (segment is the next to be received on an established
2458                  * connection, and the queue is empty), avoiding linkage into
2459                  * and removal from the queue and repetition of various
2460                  * conversions.
2461                  * Set DELACK for segments received in order, but ack
2462                  * immediately when segments are out of order (so
2463                  * fast retransmit can work).
2464                  */
2465                 if (th->th_seq == tp->rcv_nxt &&
2466                     LIST_EMPTY(&tp->t_segq) &&
2467                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2468                         if (DELAY_ACK(tp)) {
2469                                 tcp_callout_reset(tp, tp->tt_delack,
2470                                     tcp_delacktime, tcp_timer_delack);
2471                         } else {
2472                                 tp->t_flags |= TF_ACKNOW;
2473                         }
2474                         tp->rcv_nxt += tlen;
2475                         thflags = th->th_flags & TH_FIN;
2476                         tcpstat.tcps_rcvpack++;
2477                         tcpstat.tcps_rcvbyte += tlen;
2478                         ND6_HINT(tp);
2479                         if (so->so_state & SS_CANTRCVMORE) {
2480                                 m_freem(m);
2481                         } else {
2482                                 lwkt_gettoken(&so->so_rcv.ssb_token);
2483                                 ssb_appendstream(&so->so_rcv, m);
2484                                 lwkt_reltoken(&so->so_rcv.ssb_token);
2485                         }
2486                         sorwakeup(so);
2487                 } else {
2488                         if (!(tp->t_flags & TF_DUPSEG)) {
2489                                 /* Initialize SACK report block. */
2490                                 tp->reportblk.rblk_start = th->th_seq;
2491                                 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2492                                     th->th_seq + tlen, thflags);
2493                         }
2494                         thflags = tcp_reass(tp, th, &tlen, m);
2495                         tp->t_flags |= TF_ACKNOW;
2496                 }
2497
2498                 /*
2499                  * Note the amount of data that peer has sent into
2500                  * our window, in order to estimate the sender's
2501                  * buffer size.
2502                  */
2503                 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2504         } else {
2505                 m_freem(m);
2506                 thflags &= ~TH_FIN;
2507         }
2508
2509         /*
2510          * If FIN is received ACK the FIN and let the user know
2511          * that the connection is closing.
2512          */
2513         if (thflags & TH_FIN) {
2514                 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2515                         socantrcvmore(so);
2516                         /*
2517                          * If connection is half-synchronized
2518                          * (ie NEEDSYN flag on) then delay ACK,
2519                          * so it may be piggybacked when SYN is sent.
2520                          * Otherwise, since we received a FIN then no
2521                          * more input can be expected, send ACK now.
2522                          */
2523                         if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2524                                 tcp_callout_reset(tp, tp->tt_delack,
2525                                     tcp_delacktime, tcp_timer_delack);
2526                         } else {
2527                                 tp->t_flags |= TF_ACKNOW;
2528                         }
2529                         tp->rcv_nxt++;
2530                 }
2531
2532                 switch (tp->t_state) {
2533                 /*
2534                  * In SYN_RECEIVED and ESTABLISHED STATES
2535                  * enter the CLOSE_WAIT state.
2536                  */
2537                 case TCPS_SYN_RECEIVED:
2538                         tp->t_starttime = ticks;
2539                         /*FALLTHROUGH*/
2540                 case TCPS_ESTABLISHED:
2541                         tp->t_state = TCPS_CLOSE_WAIT;
2542                         break;
2543
2544                 /*
2545                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2546                  * enter the CLOSING state.
2547                  */
2548                 case TCPS_FIN_WAIT_1:
2549                         tp->t_state = TCPS_CLOSING;
2550                         break;
2551
2552                 /*
2553                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2554                  * starting the time-wait timer, turning off the other
2555                  * standard timers.
2556                  */
2557                 case TCPS_FIN_WAIT_2:
2558                         tp->t_state = TCPS_TIME_WAIT;
2559                         tcp_canceltimers(tp);
2560                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2561                                     tcp_timer_2msl);
2562                         soisdisconnected(so);
2563                         break;
2564
2565                 /*
2566                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2567                  */
2568                 case TCPS_TIME_WAIT:
2569                         tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2570                             tcp_timer_2msl);
2571                         break;
2572                 }
2573         }
2574
2575 #ifdef TCPDEBUG
2576         if (so->so_options & SO_DEBUG)
2577                 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2578 #endif
2579
2580         /*
2581          * Return any desired output.
2582          */
2583         if (needoutput || (tp->t_flags & TF_ACKNOW))
2584                 tcp_output(tp);
2585         tcp_sack_report_cleanup(tp);
2586         return(IPPROTO_DONE);
2587
2588 dropafterack:
2589         /*
2590          * Generate an ACK dropping incoming segment if it occupies
2591          * sequence space, where the ACK reflects our state.
2592          *
2593          * We can now skip the test for the RST flag since all
2594          * paths to this code happen after packets containing
2595          * RST have been dropped.
2596          *
2597          * In the SYN-RECEIVED state, don't send an ACK unless the
2598          * segment we received passes the SYN-RECEIVED ACK test.
2599          * If it fails send a RST.  This breaks the loop in the
2600          * "LAND" DoS attack, and also prevents an ACK storm
2601          * between two listening ports that have been sent forged
2602          * SYN segments, each with the source address of the other.
2603          */
2604         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2605             (SEQ_GT(tp->snd_una, th->th_ack) ||
2606              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2607                 rstreason = BANDLIM_RST_OPENPORT;
2608                 goto dropwithreset;
2609         }
2610 #ifdef TCPDEBUG
2611         if (so->so_options & SO_DEBUG)
2612                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2613 #endif
2614         m_freem(m);
2615         tp->t_flags |= TF_ACKNOW;
2616         tcp_output(tp);
2617         tcp_sack_report_cleanup(tp);
2618         return(IPPROTO_DONE);
2619
2620 dropwithreset:
2621         /*
2622          * Generate a RST, dropping incoming segment.
2623          * Make ACK acceptable to originator of segment.
2624          * Don't bother to respond if destination was broadcast/multicast.
2625          */
2626         if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2627                 goto drop;
2628         if (isipv6) {
2629                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2630                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2631                         goto drop;
2632         } else {
2633                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2634                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2635                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2636                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2637                         goto drop;
2638         }
2639         /* IPv6 anycast check is done at tcp6_input() */
2640
2641         /*
2642          * Perform bandwidth limiting.
2643          */
2644 #ifdef ICMP_BANDLIM
2645         if (badport_bandlim(rstreason) < 0)
2646                 goto drop;
2647 #endif
2648
2649 #ifdef TCPDEBUG
2650         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2651                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2652 #endif
2653         if (thflags & TH_ACK)
2654                 /* mtod() below is safe as long as hdr dropping is delayed */
2655                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2656                             TH_RST);
2657         else {
2658                 if (thflags & TH_SYN)
2659                         tlen++;
2660                 /* mtod() below is safe as long as hdr dropping is delayed */
2661                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2662                             (tcp_seq)0, TH_RST | TH_ACK);
2663         }
2664         if (tp != NULL)
2665                 tcp_sack_report_cleanup(tp);
2666         return(IPPROTO_DONE);
2667
2668 drop:
2669         /*
2670          * Drop space held by incoming segment and return.
2671          */
2672 #ifdef TCPDEBUG
2673         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2674                 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2675 #endif
2676         m_freem(m);
2677         if (tp != NULL)
2678                 tcp_sack_report_cleanup(tp);
2679         return(IPPROTO_DONE);
2680 }
2681
2682 /*
2683  * Parse TCP options and place in tcpopt.
2684  */
2685 static void
2686 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn,
2687     tcp_seq ack)
2688 {
2689         int opt, optlen, i;
2690
2691         to->to_flags = 0;
2692         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2693                 opt = cp[0];
2694                 if (opt == TCPOPT_EOL)
2695                         break;
2696                 if (opt == TCPOPT_NOP)
2697                         optlen = 1;
2698                 else {
2699                         if (cnt < 2)
2700                                 break;
2701                         optlen = cp[1];
2702                         if (optlen < 2 || optlen > cnt)
2703                                 break;
2704                 }
2705                 switch (opt) {
2706                 case TCPOPT_MAXSEG:
2707                         if (optlen != TCPOLEN_MAXSEG)
2708                                 continue;
2709                         if (!is_syn)
2710                                 continue;
2711                         to->to_flags |= TOF_MSS;
2712                         bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2713                         to->to_mss = ntohs(to->to_mss);
2714                         break;
2715                 case TCPOPT_WINDOW:
2716                         if (optlen != TCPOLEN_WINDOW)
2717                                 continue;
2718                         if (!is_syn)
2719                                 continue;
2720                         to->to_flags |= TOF_SCALE;
2721                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2722                         break;
2723                 case TCPOPT_TIMESTAMP:
2724                         if (optlen != TCPOLEN_TIMESTAMP)
2725                                 continue;
2726                         to->to_flags |= TOF_TS;
2727                         bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2728                         to->to_tsval = ntohl(to->to_tsval);
2729                         bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2730                         to->to_tsecr = ntohl(to->to_tsecr);
2731                         /*
2732                          * If echoed timestamp is later than the current time,
2733                          * fall back to non RFC1323 RTT calculation.
2734                          */
2735                         if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2736                                 to->to_tsecr = 0;
2737                         break;
2738                 case TCPOPT_SACK_PERMITTED:
2739                         if (optlen != TCPOLEN_SACK_PERMITTED)
2740                                 continue;
2741                         if (!is_syn)
2742                                 continue;
2743                         to->to_flags |= TOF_SACK_PERMITTED;
2744                         break;
2745                 case TCPOPT_SACK:
2746                         if ((optlen - 2) & 0x07)        /* not multiple of 8 */
2747                                 continue;
2748                         to->to_nsackblocks = (optlen - 2) / 8;
2749                         to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2750                         to->to_flags |= TOF_SACK;
2751                         for (i = 0; i < to->to_nsackblocks; i++) {
2752                                 struct raw_sackblock *r = &to->to_sackblocks[i];
2753
2754                                 r->rblk_start = ntohl(r->rblk_start);
2755                                 r->rblk_end = ntohl(r->rblk_end);
2756
2757                                 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2758                                         /*
2759                                          * Invalid SACK block; discard all
2760                                          * SACK blocks
2761                                          */
2762                                         tcpstat.tcps_rcvbadsackopt++;
2763                                         to->to_nsackblocks = 0;
2764                                         to->to_sackblocks = NULL;
2765                                         to->to_flags &= ~TOF_SACK;
2766                                         break;
2767                                 }
2768                         }
2769                         if ((to->to_flags & TOF_SACK) &&
2770                             tcp_sack_ndsack_blocks(to->to_sackblocks,
2771                             to->to_nsackblocks, ack))
2772                                 to->to_flags |= TOF_DSACK;
2773                         break;
2774 #ifdef TCP_SIGNATURE
2775                 /*
2776                  * XXX In order to reply to a host which has set the
2777                  * TCP_SIGNATURE option in its initial SYN, we have to
2778                  * record the fact that the option was observed here
2779                  * for the syncache code to perform the correct response.
2780                  */
2781                 case TCPOPT_SIGNATURE:
2782                         if (optlen != TCPOLEN_SIGNATURE)
2783                                 continue;
2784                         to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2785                         break;
2786 #endif /* TCP_SIGNATURE */
2787                 default:
2788                         continue;
2789                 }
2790         }
2791 }
2792
2793 /*
2794  * Pull out of band byte out of a segment so
2795  * it doesn't appear in the user's data queue.
2796  * It is still reflected in the segment length for
2797  * sequencing purposes.
2798  * "off" is the delayed to be dropped hdrlen.
2799  */
2800 static void
2801 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2802 {
2803         int cnt = off + th->th_urp - 1;
2804
2805         while (cnt >= 0) {
2806                 if (m->m_len > cnt) {
2807                         char *cp = mtod(m, caddr_t) + cnt;
2808                         struct tcpcb *tp = sototcpcb(so);
2809
2810                         tp->t_iobc = *cp;
2811                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2812                         bcopy(cp + 1, cp, m->m_len - cnt - 1);
2813                         m->m_len--;
2814                         if (m->m_flags & M_PKTHDR)
2815                                 m->m_pkthdr.len--;
2816                         return;
2817                 }
2818                 cnt -= m->m_len;
2819                 m = m->m_next;
2820                 if (m == NULL)
2821                         break;
2822         }
2823         panic("tcp_pulloutofband");
2824 }
2825
2826 /*
2827  * Collect new round-trip time estimate
2828  * and update averages and current timeout.
2829  */
2830 static void
2831 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack)
2832 {
2833         int rebaserto = 0;
2834
2835         tcpstat.tcps_rttupdated++;
2836         tp->t_rttupdated++;
2837         if ((tp->t_flags & TF_REBASERTO) && SEQ_GT(ack, tp->snd_max_prev)) {
2838 #ifdef DEBUG_EIFEL_RESPONSE
2839                 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ",
2840                     tp->t_srtt_prev, tp->t_rttvar_prev,
2841                     tp->t_srtt, tp->t_rttvar);
2842 #endif
2843
2844                 tcpstat.tcps_eifelresponse++;
2845                 rebaserto = 1;
2846                 tp->t_flags &= ~TF_REBASERTO;
2847                 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT));
2848                 tp->t_rttvar = max(tp->t_rttvar_prev,
2849                     (rtt << (TCP_RTTVAR_SHIFT - 1)));
2850                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2851                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2852
2853 #ifdef DEBUG_EIFEL_RESPONSE
2854                 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar);
2855 #endif
2856         } else if (tp->t_srtt != 0) {
2857                 int delta;
2858
2859                 /*
2860                  * srtt is stored as fixed point with 5 bits after the
2861                  * binary point (i.e., scaled by 8).  The following magic
2862                  * is equivalent to the smoothing algorithm in rfc793 with
2863                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2864                  * point).  Adjust rtt to origin 0.
2865                  */
2866                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2867                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2868
2869                 if ((tp->t_srtt += delta) <= 0)
2870                         tp->t_srtt = 1;
2871
2872                 /*
2873                  * We accumulate a smoothed rtt variance (actually, a
2874                  * smoothed mean difference), then set the retransmit
2875                  * timer to smoothed rtt + 4 times the smoothed variance.
2876                  * rttvar is stored as fixed point with 4 bits after the
2877                  * binary point (scaled by 16).  The following is
2878                  * equivalent to rfc793 smoothing with an alpha of .75
2879                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2880                  * rfc793's wired-in beta.
2881                  */
2882                 if (delta < 0)
2883                         delta = -delta;
2884                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2885                 if ((tp->t_rttvar += delta) <= 0)
2886                         tp->t_rttvar = 1;
2887                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2888                         tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2889         } else {
2890                 /*
2891                  * No rtt measurement yet - use the unsmoothed rtt.
2892                  * Set the variance to half the rtt (so our first
2893                  * retransmit happens at 3*rtt).
2894                  */
2895                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2896                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2897                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2898         }
2899         tp->t_rtttime = 0;
2900         tp->t_rxtshift = 0;
2901
2902 #ifdef DEBUG_EIFEL_RESPONSE
2903         if (rebaserto) {
2904                 kprintf("| rxtcur prev %d, old %d, ",
2905                     tp->t_rxtcur_prev, tp->t_rxtcur);
2906         }
2907 #endif
2908
2909         /*
2910          * the retransmit should happen at rtt + 4 * rttvar.
2911          * Because of the way we do the smoothing, srtt and rttvar
2912          * will each average +1/2 tick of bias.  When we compute
2913          * the retransmit timer, we want 1/2 tick of rounding and
2914          * 1 extra tick because of +-1/2 tick uncertainty in the
2915          * firing of the timer.  The bias will give us exactly the
2916          * 1.5 tick we need.  But, because the bias is
2917          * statistical, we have to test that we don't drop below
2918          * the minimum feasible timer (which is 2 ticks).
2919          */
2920         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2921                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2922
2923         if (rebaserto) {
2924                 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) {
2925                         /*
2926                          * RFC4015 requires that the new RTO is at least
2927                          * 2*G (tcp_eifel_rtoinc) greater then the RTO
2928                          * (t_rxtcur_prev) when the spurious retransmit
2929                          * timeout happens.
2930                          *
2931                          * The above condition could be true, if the SRTT
2932                          * and RTTVAR used to calculate t_rxtcur_prev
2933                          * resulted in a value less than t_rttmin.  So
2934                          * simply increasing SRTT by tcp_eifel_rtoinc when
2935                          * preparing for the Eifel response in
2936                          * tcp_save_congestion_state() could not ensure
2937                          * that the new RTO will be tcp_eifel_rtoinc greater
2938                          * t_rxtcur_prev.
2939                          */
2940                         tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc;
2941                 }
2942 #ifdef DEBUG_EIFEL_RESPONSE
2943                 kprintf("new %d\n", tp->t_rxtcur);
2944 #endif
2945         }
2946
2947         /*
2948          * We received an ack for a packet that wasn't retransmitted;
2949          * it is probably safe to discard any error indications we've
2950          * received recently.  This isn't quite right, but close enough
2951          * for now (a route might have failed after we sent a segment,
2952          * and the return path might not be symmetrical).
2953          */
2954         tp->t_softerror = 0;
2955 }
2956
2957 /*
2958  * Determine a reasonable value for maxseg size.
2959  * If the route is known, check route for mtu.
2960  * If none, use an mss that can be handled on the outgoing
2961  * interface without forcing IP to fragment; if bigger than
2962  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2963  * to utilize large mbufs.  If no route is found, route has no mtu,
2964  * or the destination isn't local, use a default, hopefully conservative
2965  * size (usually 512 or the default IP max size, but no more than the mtu
2966  * of the interface), as we can't discover anything about intervening
2967  * gateways or networks.  We also initialize the congestion/slow start
2968  * window to be a single segment if the destination isn't local.
2969  * While looking at the routing entry, we also initialize other path-dependent
2970  * parameters from pre-set or cached values in the routing entry.
2971  *
2972  * Also take into account the space needed for options that we
2973  * send regularly.  Make maxseg shorter by that amount to assure
2974  * that we can send maxseg amount of data even when the options
2975  * are present.  Store the upper limit of the length of options plus
2976  * data in maxopd.
2977  *
2978  * NOTE that this routine is only called when we process an incoming
2979  * segment, for outgoing segments only tcp_mssopt is called.
2980  */
2981 void
2982 tcp_mss(struct tcpcb *tp, int offer)
2983 {
2984         struct rtentry *rt;
2985         struct ifnet *ifp;
2986         int rtt, mss;
2987         u_long bufsize;
2988         struct inpcb *inp = tp->t_inpcb;
2989         struct socket *so;
2990 #ifdef INET6
2991         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2992         size_t min_protoh = isipv6 ?
2993                             sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2994                             sizeof(struct tcpiphdr);
2995 #else
2996         const boolean_t isipv6 = FALSE;
2997         const size_t min_protoh = sizeof(struct tcpiphdr);
2998 #endif
2999
3000         if (isipv6)
3001                 rt = tcp_rtlookup6(&inp->inp_inc);
3002         else
3003                 rt = tcp_rtlookup(&inp->inp_inc);
3004         if (rt == NULL) {
3005                 tp->t_maxopd = tp->t_maxseg =
3006                     (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3007                 return;
3008         }
3009         ifp = rt->rt_ifp;
3010         so = inp->inp_socket;
3011
3012         /*
3013          * Offer == 0 means that there was no MSS on the SYN segment,
3014          * in this case we use either the interface mtu or tcp_mssdflt.
3015          *
3016          * An offer which is too large will be cut down later.
3017          */
3018         if (offer == 0) {
3019                 if (isipv6) {
3020                         if (in6_localaddr(&inp->in6p_faddr)) {
3021                                 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
3022                                         min_protoh;
3023                         } else {
3024                                 offer = tcp_v6mssdflt;
3025                         }
3026                 } else {
3027                         if (in_localaddr(inp->inp_faddr))
3028                                 offer = ifp->if_mtu - min_protoh;
3029                         else
3030                                 offer = tcp_mssdflt;
3031                 }
3032         }
3033
3034         /*
3035          * Prevent DoS attack with too small MSS. Round up
3036          * to at least minmss.
3037          *
3038          * Sanity check: make sure that maxopd will be large
3039          * enough to allow some data on segments even is the
3040          * all the option space is used (40bytes).  Otherwise
3041          * funny things may happen in tcp_output.
3042          */
3043         offer = max(offer, tcp_minmss);
3044         offer = max(offer, 64);
3045
3046         rt->rt_rmx.rmx_mssopt = offer;
3047
3048         /*
3049          * While we're here, check if there's an initial rtt
3050          * or rttvar.  Convert from the route-table units
3051          * to scaled multiples of the slow timeout timer.
3052          */
3053         if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3054                 /*
3055                  * XXX the lock bit for RTT indicates that the value
3056                  * is also a minimum value; this is subject to time.
3057                  */
3058                 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3059                         tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3060                 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3061                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3062                 tcpstat.tcps_usedrtt++;
3063                 if (rt->rt_rmx.rmx_rttvar) {
3064                         tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3065                             (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3066                         tcpstat.tcps_usedrttvar++;
3067                 } else {
3068                         /* default variation is +- 1 rtt */
3069                         tp->t_rttvar =
3070                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3071                 }
3072                 TCPT_RANGESET(tp->t_rxtcur,
3073                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3074                               tp->t_rttmin, TCPTV_REXMTMAX);
3075         }
3076
3077         /*
3078          * if there's an mtu associated with the route, use it
3079          * else, use the link mtu.  Take the smaller of mss or offer
3080          * as our final mss.
3081          */
3082         if (rt->rt_rmx.rmx_mtu) {
3083                 mss = rt->rt_rmx.rmx_mtu - min_protoh;
3084         } else {
3085                 if (isipv6)
3086                         mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
3087                 else
3088                         mss = ifp->if_mtu - min_protoh;
3089         }
3090         mss = min(mss, offer);
3091
3092         /*
3093          * maxopd stores the maximum length of data AND options
3094          * in a segment; maxseg is the amount of data in a normal
3095          * segment.  We need to store this value (maxopd) apart
3096          * from maxseg, because now every segment carries options
3097          * and thus we normally have somewhat less data in segments.
3098          */
3099         tp->t_maxopd = mss;
3100
3101         if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3102             ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3103                 mss -= TCPOLEN_TSTAMP_APPA;
3104
3105 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
3106         if (mss > MCLBYTES)
3107                 mss &= ~(MCLBYTES-1);
3108 #else
3109         if (mss > MCLBYTES)
3110                 mss = mss / MCLBYTES * MCLBYTES;
3111 #endif
3112         /*
3113          * If there's a pipesize, change the socket buffer
3114          * to that size.  Make the socket buffers an integral
3115          * number of mss units; if the mss is larger than
3116          * the socket buffer, decrease the mss.
3117          */
3118 #ifdef RTV_SPIPE
3119         if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3120 #endif
3121                 bufsize = so->so_snd.ssb_hiwat;
3122         if (bufsize < mss)
3123                 mss = bufsize;
3124         else {
3125                 bufsize = roundup(bufsize, mss);
3126                 if (bufsize > sb_max)
3127                         bufsize = sb_max;
3128                 if (bufsize > so->so_snd.ssb_hiwat)
3129                         ssb_reserve(&so->so_snd, bufsize, so, NULL);
3130         }
3131         tp->t_maxseg = mss;
3132
3133 #ifdef RTV_RPIPE
3134         if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3135 #endif
3136                 bufsize = so->so_rcv.ssb_hiwat;
3137         if (bufsize > mss) {
3138                 bufsize = roundup(bufsize, mss);
3139                 if (bufsize > sb_max)
3140                         bufsize = sb_max;
3141                 if (bufsize > so->so_rcv.ssb_hiwat) {
3142                         lwkt_gettoken(&so->so_rcv.ssb_token);
3143                         ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3144                         lwkt_reltoken(&so->so_rcv.ssb_token);
3145                 }
3146         }
3147
3148         /*
3149          * Set the slow-start flight size
3150          *
3151          * NOTE: t_maxseg must have been configured!
3152          */
3153         tp->snd_cwnd = tcp_initial_window(tp);
3154
3155         if (rt->rt_rmx.rmx_ssthresh) {
3156                 /*
3157                  * There's some sort of gateway or interface
3158                  * buffer limit on the path.  Use this to set
3159                  * the slow start threshhold, but set the
3160                  * threshold to no less than 2*mss.
3161                  */
3162                 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3163                 tcpstat.tcps_usedssthresh++;
3164         }
3165 }
3166
3167 /*
3168  * Determine the MSS option to send on an outgoing SYN.
3169  */
3170 int
3171 tcp_mssopt(struct tcpcb *tp)
3172 {
3173         struct rtentry *rt;
3174 #ifdef INET6
3175         boolean_t isipv6 =
3176             ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3177         int min_protoh = isipv6 ?
3178                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3179                              sizeof(struct tcpiphdr);
3180 #else
3181         const boolean_t isipv6 = FALSE;
3182         const size_t min_protoh = sizeof(struct tcpiphdr);
3183 #endif
3184
3185         if (isipv6)
3186                 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3187         else
3188                 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3189         if (rt == NULL)
3190                 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3191
3192         return (rt->rt_ifp->if_mtu - min_protoh);
3193 }
3194
3195 /*
3196  * When a partial ack arrives, force the retransmission of the
3197  * next unacknowledged segment.  Do not exit Fast Recovery.
3198  *
3199  * Implement the Slow-but-Steady variant of NewReno by restarting the
3200  * the retransmission timer.  Turn it off here so it can be restarted
3201  * later in tcp_output().
3202  */
3203 static void
3204 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3205 {
3206         tcp_seq old_snd_nxt = tp->snd_nxt;
3207         u_long ocwnd = tp->snd_cwnd;
3208
3209         tcp_callout_stop(tp, tp->tt_rexmt);
3210         tp->t_rtttime = 0;
3211         tp->snd_nxt = th->th_ack;
3212         /* Set snd_cwnd to one segment beyond acknowledged offset. */
3213         tp->snd_cwnd = tp->t_maxseg;
3214         tp->t_flags |= TF_ACKNOW;
3215         tcp_output(tp);
3216         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3217                 tp->snd_nxt = old_snd_nxt;
3218         /* partial window deflation */
3219         if (ocwnd > acked)
3220                 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3221         else
3222                 tp->snd_cwnd = tp->t_maxseg;
3223 }
3224
3225 /*
3226  * In contrast to the Slow-but-Steady NewReno variant,
3227  * we do not reset the retransmission timer for SACK retransmissions,
3228  * except when retransmitting snd_una.
3229  */
3230 static void
3231 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3232 {
3233         tcp_seq old_snd_nxt = tp->snd_nxt;
3234         u_long ocwnd = tp->snd_cwnd;
3235         uint32_t pipe;
3236         int nseg = 0;           /* consecutive new segments */
3237         int nseg_rexmt = 0;     /* retransmitted segments */
3238 #define MAXBURST 4              /* limit burst of new packets on partial ack */
3239
3240         tp->t_rtttime = 0;
3241         pipe = tcp_sack_compute_pipe(tp);
3242         while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3243             (!tcp_do_smartsack || nseg < MAXBURST)) {
3244                 tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3245                 uint32_t sent, seglen;
3246                 boolean_t rescue;
3247                 int error;
3248
3249                 old_rexmt_high = tp->rexmt_high;
3250                 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
3251                         tp->rexmt_high = old_rexmt_high;
3252                         break;
3253                 }
3254
3255                 /*
3256                  * If the next tranmission is a rescue retranmission,
3257                  * we check whether we have already sent some data
3258                  * (either new segments or retransmitted segments)
3259                  * into the the network or not.  Since the idea of rescue
3260                  * retransmission is to sustain ACK clock, as long as
3261                  * some segments are in the network, ACK clock will be
3262                  * kept ticking.
3263                  */
3264                 if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3265                         tp->rexmt_high = old_rexmt_high;
3266                         break;
3267                 }
3268
3269                 if (nextrexmt == tp->snd_max)
3270                         ++nseg;
3271                 else
3272                         ++nseg_rexmt;
3273                 tp->snd_nxt = nextrexmt;
3274                 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3275                 old_snd_max = tp->snd_max;
3276                 if (nextrexmt == tp->snd_una)
3277                         tcp_callout_stop(tp, tp->tt_rexmt);
3278                 error = tcp_output(tp);
3279                 if (error != 0) {
3280                         tp->rexmt_high = old_rexmt_high;
3281                         break;
3282                 }
3283                 sent = tp->snd_nxt - nextrexmt;
3284                 if (sent <= 0) {
3285                         tp->rexmt_high = old_rexmt_high;
3286                         break;
3287                 }
3288                 pipe += sent;
3289                 tcpstat.tcps_sndsackpack++;
3290                 tcpstat.tcps_sndsackbyte += sent;
3291
3292                 if (rescue) {
3293                         tcpstat.tcps_sackrescue++;
3294                         tp->rexmt_rescue = tp->snd_nxt;
3295                         tp->t_flags |= TF_SACKRESCUED;
3296                         break;
3297                 }
3298                 if (SEQ_LT(nextrexmt, old_snd_max) &&
3299                     SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
3300                         tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3301                         if (tcp_aggressive_rescuesack &&
3302                             (tp->t_flags & TF_SACKRESCUED) &&
3303                             SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3304                                 /* Drag RescueRxt along with HighRxt */
3305                                 tp->rexmt_rescue = tp->rexmt_high;
3306                         }
3307                 }
3308         }
3309         if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3310                 tp->snd_nxt = old_snd_nxt;
3311         tp->snd_cwnd = ocwnd;
3312 }
3313
3314 static boolean_t
3315 tcp_sack_limitedxmit(struct tcpcb *tp)
3316 {
3317         tcp_seq oldsndnxt = tp->snd_nxt;
3318         tcp_seq oldsndmax = tp->snd_max;
3319         u_long ocwnd = tp->snd_cwnd;
3320         uint32_t pipe;
3321         boolean_t ret = FALSE;
3322
3323         tp->rexmt_high = tp->snd_una - 1;
3324         pipe = tcp_sack_compute_pipe(tp);
3325         while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg) {
3326                 uint32_t sent;
3327                 tcp_seq next;
3328                 int error;
3329
3330                 next = tp->snd_nxt = tp->snd_max;
3331                 tp->snd_cwnd = tp->snd_nxt - tp->snd_una + tp->t_maxseg;
3332
3333                 error = tcp_output(tp);
3334                 if (error)
3335                         break;
3336
3337                 sent = tp->snd_nxt - next;
3338                 if (sent <= 0)
3339                         break;
3340                 pipe += sent;
3341                 ++tcpstat.tcps_sndlimited;
3342                 ret = TRUE;
3343         }
3344
3345         if (SEQ_LT(oldsndnxt, oldsndmax)) {
3346                 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3347                     ("snd_una moved in other threads"));
3348                 tp->snd_nxt = oldsndnxt;
3349         }
3350         tp->snd_cwnd = ocwnd;
3351
3352         return ret;
3353 }
3354
3355 /*
3356  * Reset idle time and keep-alive timer, typically called when a valid
3357  * tcp packet is received but may also be called when FASTKEEP is set
3358  * to prevent the previous long-timeout from calculating to a drop.
3359  *
3360  * Only update t_rcvtime for non-SYN packets.
3361  *
3362  * Handle the case where one side thinks the connection is established
3363  * but the other side has, say, rebooted without cleaning out the
3364  * connection.   The SYNs could be construed as an attack and wind
3365  * up ignored, but in case it isn't an attack we can validate the
3366  * connection by forcing a keepalive.
3367  */
3368 void
3369 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3370 {
3371         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3372                 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3373                         tp->t_flags |= TF_KEEPALIVE;
3374                         tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3375                                           tcp_timer_keep);
3376                 } else {
3377                         tp->t_rcvtime = ticks;
3378                         tp->t_flags &= ~TF_KEEPALIVE;
3379                         tcp_callout_reset(tp, tp->tt_keep,
3380                                           tp->t_keepidle,
3381                                           tcp_timer_keep);
3382                 }
3383         }
3384 }
3385
3386 static int
3387 tcp_rmx_msl(const struct tcpcb *tp)
3388 {
3389         struct rtentry *rt;
3390         struct inpcb *inp = tp->t_inpcb;
3391         int msl;
3392 #ifdef INET6
3393         boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3394 #else
3395         const boolean_t isipv6 = FALSE;
3396 #endif
3397
3398         if (isipv6)
3399                 rt = tcp_rtlookup6(&inp->inp_inc);
3400         else
3401                 rt = tcp_rtlookup(&inp->inp_inc);
3402         if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3403                 return tcp_msl;
3404
3405         msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3406         if (msl == 0)
3407                 msl = 1;
3408
3409         return msl;
3410 }
3411
3412 static void
3413 tcp_established(struct tcpcb *tp)
3414 {
3415         tp->t_state = TCPS_ESTABLISHED;
3416         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
3417
3418         if (tp->t_rxtsyn > 0) {
3419                 /*
3420                  * RFC6298:
3421                  * "If the timer expires awaiting the ACK of a SYN segment
3422                  *  and the TCP implementation is using an RTO less than 3
3423                  *  seconds, the RTO MUST be re-initialized to 3 seconds
3424                  *  when data transmission begins"
3425                  */
3426                 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3427                         tp->t_rxtcur = TCPTV_RTOBASE3;
3428         }
3429 }