sys/netinet/tcp_reass.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 4. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
  30  * $FreeBSD$
  31  */
  32
  33 #include "opt_ipfw.h"           /* for ipfw_fwd         */
  34 #include "opt_inet.h"
  35 #include "opt_inet6.h"
  36 #include "opt_ipsec.h"
  37 #include "opt_mac.h"
  38 #include "opt_tcpdebug.h"
  39 #include "opt_tcp_input.h"
  40 #include "opt_tcp_sack.h"
  41
  42 #include <sys/param.h>
  43 #include <sys/kernel.h>
  44 #include <sys/mac.h>
  45 #include <sys/malloc.h>
  46 #include <sys/mbuf.h>
  47 #include <sys/proc.h>           /* for proc0 declaration */
  48 #include <sys/protosw.h>
  49 #include <sys/signalvar.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/sysctl.h>
  53 #include <sys/syslog.h>
  54 #include <sys/systm.h>
  55
  56 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
  57
  58 #include <vm/uma.h>
  59
  60 #include <net/if.h>
  61 #include <net/route.h>
  62
  63 #include <netinet/in.h>
  64 #include <netinet/in_pcb.h>
  65 #include <netinet/in_systm.h>
  66 #include <netinet/in_var.h>
  67 #include <netinet/ip.h>
  68 #include <netinet/ip_icmp.h>    /* required for icmp_var.h */
  69 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
  70 #include <netinet/ip_var.h>
  71 #include <netinet/ip_options.h>
  72 #include <netinet/ip6.h>
  73 #include <netinet/icmp6.h>
  74 #include <netinet6/in6_pcb.h>
  75 #include <netinet6/ip6_var.h>
  76 #include <netinet6/nd6.h>
  77 #include <netinet/tcp.h>
  78 #include <netinet/tcp_fsm.h>
  79 #include <netinet/tcp_seq.h>
  80 #include <netinet/tcp_timer.h>
  81 #include <netinet/tcp_var.h>
  82 #include <netinet6/tcp6_var.h>
  83 #include <netinet/tcpip.h>
  84 #ifdef TCPDEBUG
  85 #include <netinet/tcp_debug.h>
  86 #endif /* TCPDEBUG */
  87
  88 #ifdef FAST_IPSEC
  89 #include <netipsec/ipsec.h>
  90 #include <netipsec/ipsec6.h>
  91 #endif /*FAST_IPSEC*/
  92
  93 #ifdef IPSEC
  94 #include <netinet6/ipsec.h>
  95 #include <netinet6/ipsec6.h>
  96 #include <netkey/key.h>
  97 #endif /*IPSEC*/
  98
  99 #include <machine/in_cksum.h>
 100
 101 static const int tcprexmtthresh = 3;
 102
 103 struct  tcpstat tcpstat;
 104 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
 105     &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
 106
 107 static int log_in_vain = 0;
 108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
 109     &log_in_vain, 0, "Log all incoming TCP connections");
 110
 111 static int blackhole = 0;
 112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
 113         &blackhole, 0, "Do not send RST when dropping refused connections");
 114
 115 int tcp_delack_enabled = 1;
 116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
 117     &tcp_delack_enabled, 0,
 118     "Delay ACK to try and piggyback it onto a data packet");
 119
 120 #ifdef TCP_DROP_SYNFIN
 121 static int drop_synfin = 0;
 122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
 123     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
 124 #endif
 125
 126 static int tcp_do_rfc3042 = 1;
 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
 128     &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
 129
 130 static int tcp_do_rfc3390 = 1;
 131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
 132     &tcp_do_rfc3390, 0,
 133     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
 134
 135 static int tcp_insecure_rst = 0;
 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
 137     &tcp_insecure_rst, 0,
 138     "Follow the old (insecure) criteria for accepting RST packets.");
 139
 140 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
 141             "TCP Segment Reassembly Queue");
 142
 143 static int tcp_reass_maxseg = 0;
 144 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
 145            &tcp_reass_maxseg, 0,
 146            "Global maximum number of TCP Segments in Reassembly Queue");
 147
 148 int tcp_reass_qsize = 0;
 149 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
 150            &tcp_reass_qsize, 0,
 151            "Global number of TCP Segments currently in Reassembly Queue");
 152
 153 static int tcp_reass_maxqlen = 48;
 154 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
 155            &tcp_reass_maxqlen, 0,
 156            "Maximum number of TCP Segments per individual Reassembly Queue");
 157
 158 static int tcp_reass_overflows = 0;
 159 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
 160            &tcp_reass_overflows, 0,
 161            "Global number of TCP Segment Reassembly Queue Overflows");
 162
 163 struct inpcbhead tcb;
 164 #define tcb6    tcb  /* for KAME src sync over BSD*'s */
 165 struct inpcbinfo tcbinfo;
 166 struct mtx      *tcbinfo_mtx;
 167
 168 static void      tcp_dooptions(struct tcpopt *, u_char *, int, int);
 169
 170 static void      tcp_pulloutofband(struct socket *,
 171                      struct tcphdr *, struct mbuf *, int);
 172 static int       tcp_reass(struct tcpcb *, struct tcphdr *, int *,
 173                      struct mbuf *);
 174 static void      tcp_xmit_timer(struct tcpcb *, int);
 175 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
 176 static int       tcp_timewait(struct inpcb *, struct tcpopt *,
 177                      struct tcphdr *, struct mbuf *, int);
 178
 179 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
 180 #ifdef INET6
 181 #define ND6_HINT(tp) \
 182 do { \
 183         if ((tp) && (tp)->t_inpcb && \
 184             ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
 185                 nd6_nud_hint(NULL, NULL, 0); \
 186 } while (0)
 187 #else
 188 #define ND6_HINT(tp)
 189 #endif
 190
 191 /*
 192  * Indicate whether this ack should be delayed.  We can delay the ack if
 193  *      - there is no delayed ack timer in progress and
 194  *      - our last ack wasn't a 0-sized window.  We never want to delay
 195  *        the ack that opens up a 0-sized window and
 196  *              - delayed acks are enabled or
 197  *              - this is a half-synchronized T/TCP connection.
 198  */
 199 #define DELAY_ACK(tp)                                                   \
 200         ((!callout_active(tp->tt_delack) &&                             \
 201             (tp->t_flags & TF_RXWIN0SENT) == 0) &&                      \
 202             (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
 203
 204 /* Initialize TCP reassembly queue */
 205 static void
 206 tcp_reass_zone_change(void *tag)
 207 {
 208
 209         tcp_reass_maxseg = nmbclusters / 16;
 210         uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
 211 }
 212
 213 uma_zone_t      tcp_reass_zone;
 214 void
 215 tcp_reass_init()
 216 {
 217         tcp_reass_maxseg = nmbclusters / 16;
 218         TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
 219             &tcp_reass_maxseg);
 220         tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
 221             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 222         uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
 223         EVENTHANDLER_REGISTER(nmbclusters_change,
 224             tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
 225 }
 226
 227 static int
 228 tcp_reass(tp, th, tlenp, m)
 229         register struct tcpcb *tp;
 230         register struct tcphdr *th;
 231         int *tlenp;
 232         struct mbuf *m;
 233 {
 234         struct tseg_qent *q;
 235         struct tseg_qent *p = NULL;
 236         struct tseg_qent *nq;
 237         struct tseg_qent *te = NULL;
 238         struct socket *so = tp->t_inpcb->inp_socket;
 239         int flags;
 240
 241         INP_LOCK_ASSERT(tp->t_inpcb);
 242
 243         /*
 244          * XXX: tcp_reass() is rather inefficient with its data structures
 245          * and should be rewritten (see NetBSD for optimizations).  While
 246          * doing that it should move to its own file tcp_reass.c.
 247          */
 248
 249         /*
 250          * Call with th==NULL after become established to
 251          * force pre-ESTABLISHED data up to user socket.
 252          */
 253         if (th == NULL)
 254                 goto present;
 255
 256         /*
 257          * Limit the number of segments in the reassembly queue to prevent
 258          * holding on to too many segments (and thus running out of mbufs).
 259          * Make sure to let the missing segment through which caused this
 260          * queue.  Always keep one global queue entry spare to be able to
 261          * process the missing segment.
 262          */
 263         if (th->th_seq != tp->rcv_nxt &&
 264             (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
 265              tp->t_segqlen >= tcp_reass_maxqlen)) {
 266                 tcp_reass_overflows++;
 267                 tcpstat.tcps_rcvmemdrop++;
 268                 m_freem(m);
 269                 *tlenp = 0;
 270                 return (0);
 271         }
 272
 273         /*
 274          * Allocate a new queue entry. If we can't, or hit the zone limit
 275          * just drop the pkt.
 276          */
 277         te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
 278         if (te == NULL) {
 279                 tcpstat.tcps_rcvmemdrop++;
 280                 m_freem(m);
 281                 *tlenp = 0;
 282                 return (0);
 283         }
 284         tp->t_segqlen++;
 285         tcp_reass_qsize++;
 286
 287         /*
 288          * Find a segment which begins after this one does.
 289          */
 290         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
 291                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
 292                         break;
 293                 p = q;
 294         }
 295
 296         /*
 297          * If there is a preceding segment, it may provide some of
 298          * our data already.  If so, drop the data from the incoming
 299          * segment.  If it provides all of our data, drop us.
 300          */
 301         if (p != NULL) {
 302                 register int i;
 303                 /* conversion to int (in i) handles seq wraparound */
 304                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
 305                 if (i > 0) {
 306                         if (i >= *tlenp) {
 307                                 tcpstat.tcps_rcvduppack++;
 308                                 tcpstat.tcps_rcvdupbyte += *tlenp;
 309                                 m_freem(m);
 310                                 uma_zfree(tcp_reass_zone, te);
 311                                 tp->t_segqlen--;
 312                                 tcp_reass_qsize--;
 313                                 /*
 314                                  * Try to present any queued data
 315                                  * at the left window edge to the user.
 316                                  * This is needed after the 3-WHS
 317                                  * completes.
 318                                  */
 319                                 goto present;   /* ??? */
 320                         }
 321                         m_adj(m, i);
 322                         *tlenp -= i;
 323                         th->th_seq += i;
 324                 }
 325         }
 326         tcpstat.tcps_rcvoopack++;
 327         tcpstat.tcps_rcvoobyte += *tlenp;
 328
 329         /*
 330          * While we overlap succeeding segments trim them or,
 331          * if they are completely covered, dequeue them.
 332          */
 333         while (q) {
 334                 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
 335                 if (i <= 0)
 336                         break;
 337                 if (i < q->tqe_len) {
 338                         q->tqe_th->th_seq += i;
 339                         q->tqe_len -= i;
 340                         m_adj(q->tqe_m, i);
 341                         break;
 342                 }
 343
 344                 nq = LIST_NEXT(q, tqe_q);
 345                 LIST_REMOVE(q, tqe_q);
 346                 m_freem(q->tqe_m);
 347                 uma_zfree(tcp_reass_zone, q);
 348                 tp->t_segqlen--;
 349                 tcp_reass_qsize--;
 350                 q = nq;
 351         }
 352
 353         /* Insert the new segment queue entry into place. */
 354         te->tqe_m = m;
 355         te->tqe_th = th;
 356         te->tqe_len = *tlenp;
 357
 358         if (p == NULL) {
 359                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
 360         } else {
 361                 LIST_INSERT_AFTER(p, te, tqe_q);
 362         }
 363
 364 present:
 365         /*
 366          * Present data to user, advancing rcv_nxt through
 367          * completed sequence space.
 368          */
 369         if (!TCPS_HAVEESTABLISHED(tp->t_state))
 370                 return (0);
 371         q = LIST_FIRST(&tp->t_segq);
 372         if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
 373                 return (0);
 374         SOCKBUF_LOCK(&so->so_rcv);
 375         do {
 376                 tp->rcv_nxt += q->tqe_len;
 377                 flags = q->tqe_th->th_flags & TH_FIN;
 378                 nq = LIST_NEXT(q, tqe_q);
 379                 LIST_REMOVE(q, tqe_q);
 380                 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
 381                         m_freem(q->tqe_m);
 382                 else
 383                         sbappendstream_locked(&so->so_rcv, q->tqe_m);
 384                 uma_zfree(tcp_reass_zone, q);
 385                 tp->t_segqlen--;
 386                 tcp_reass_qsize--;
 387                 q = nq;
 388         } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
 389         ND6_HINT(tp);
 390         sorwakeup_locked(so);
 391         return (flags);
 392 }
 393
 394 /*
 395  * TCP input routine, follows pages 65-76 of the
 396  * protocol specification dated September, 1981 very closely.
 397  */
 398 #ifdef INET6
 399 int
 400 tcp6_input(mp, offp, proto)
 401         struct mbuf **mp;
 402         int *offp, proto;
 403 {
 404         register struct mbuf *m = *mp;
 405         struct in6_ifaddr *ia6;
 406
 407         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
 408
 409         /*
 410          * draft-itojun-ipv6-tcp-to-anycast
 411          * better place to put this in?
 412          */
 413         ia6 = ip6_getdstifaddr(m);
 414         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
 415                 struct ip6_hdr *ip6;
 416
 417                 ip6 = mtod(m, struct ip6_hdr *);
 418                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
 419                             (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
 420                 return IPPROTO_DONE;
 421         }
 422
 423         tcp_input(m, *offp);
 424         return IPPROTO_DONE;
 425 }
 426 #endif
 427
 428 void
 429 tcp_input(m, off0)
 430         register struct mbuf *m;
 431         int off0;
 432 {
 433         register struct tcphdr *th;
 434         register struct ip *ip = NULL;
 435         register struct ipovly *ipov;
 436         register struct inpcb *inp = NULL;
 437         u_char *optp = NULL;
 438         int optlen = 0;
 439         int len, tlen, off;
 440         int drop_hdrlen;
 441         register struct tcpcb *tp = 0;
 442         register int thflags;
 443         struct socket *so = 0;
 444         int todrop, acked, ourfinisacked, needoutput = 0;
 445         u_long tiwin;
 446         struct tcpopt to;               /* options in this segment */
 447         int headlocked = 0;
 448 #ifdef IPFIREWALL_FORWARD
 449         struct m_tag *fwd_tag;
 450 #endif
 451         int rstreason; /* For badport_bandlim accounting purposes */
 452
 453         struct ip6_hdr *ip6 = NULL;
 454 #ifdef INET6
 455         int isipv6;
 456 #else
 457         const int isipv6 = 0;
 458 #endif
 459
 460 #ifdef TCPDEBUG
 461         /*
 462          * The size of tcp_saveipgen must be the size of the max ip header,
 463          * now IPv6.
 464          */
 465         u_char tcp_saveipgen[40];
 466         struct tcphdr tcp_savetcp;
 467         short ostate = 0;
 468 #endif
 469
 470 #ifdef INET6
 471         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
 472 #endif
 473         bzero((char *)&to, sizeof(to));
 474
 475         tcpstat.tcps_rcvtotal++;
 476
 477         if (isipv6) {
 478 #ifdef INET6
 479                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
 480                 ip6 = mtod(m, struct ip6_hdr *);
 481                 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
 482                 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
 483                         tcpstat.tcps_rcvbadsum++;
 484                         goto drop;
 485                 }
 486                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
 487
 488                 /*
 489                  * Be proactive about unspecified IPv6 address in source.
 490                  * As we use all-zero to indicate unbounded/unconnected pcb,
 491                  * unspecified IPv6 address can be used to confuse us.
 492                  *
 493                  * Note that packets with unspecified IPv6 destination is
 494                  * already dropped in ip6_input.
 495                  */
 496                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
 497                         /* XXX stat */
 498                         goto drop;
 499                 }
 500 #else
 501                 th = NULL;              /* XXX: avoid compiler warning */
 502 #endif
 503         } else {
 504                 /*
 505                  * Get IP and TCP header together in first mbuf.
 506                  * Note: IP leaves IP header in first mbuf.
 507                  */
 508                 if (off0 > sizeof (struct ip)) {
 509                         ip_stripoptions(m, (struct mbuf *)0);
 510                         off0 = sizeof(struct ip);
 511                 }
 512                 if (m->m_len < sizeof (struct tcpiphdr)) {
 513                         if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
 514                                 tcpstat.tcps_rcvshort++;
 515                                 return;
 516                         }
 517                 }
 518                 ip = mtod(m, struct ip *);
 519                 ipov = (struct ipovly *)ip;
 520                 th = (struct tcphdr *)((caddr_t)ip + off0);
 521                 tlen = ip->ip_len;
 522
 523                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
 524                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 525                                 th->th_sum = m->m_pkthdr.csum_data;
 526                         else
 527                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
 528                                                 ip->ip_dst.s_addr,
 529                                                 htonl(m->m_pkthdr.csum_data +
 530                                                         ip->ip_len +
 531                                                         IPPROTO_TCP));
 532                         th->th_sum ^= 0xffff;
 533 #ifdef TCPDEBUG
 534                         ipov->ih_len = (u_short)tlen;
 535                         ipov->ih_len = htons(ipov->ih_len);
 536 #endif
 537                 } else {
 538                         /*
 539                          * Checksum extended TCP header and data.
 540                          */
 541                         len = sizeof (struct ip) + tlen;
 542                         bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
 543                         ipov->ih_len = (u_short)tlen;
 544                         ipov->ih_len = htons(ipov->ih_len);
 545                         th->th_sum = in_cksum(m, len);
 546                 }
 547                 if (th->th_sum) {
 548                         tcpstat.tcps_rcvbadsum++;
 549                         goto drop;
 550                 }
 551 #ifdef INET6
 552                 /* Re-initialization for later version check */
 553                 ip->ip_v = IPVERSION;
 554 #endif
 555         }
 556
 557         /*
 558          * Check that TCP offset makes sense,
 559          * pull out TCP options and adjust length.              XXX
 560          */
 561         off = th->th_off << 2;
 562         if (off < sizeof (struct tcphdr) || off > tlen) {
 563                 tcpstat.tcps_rcvbadoff++;
 564                 goto drop;
 565         }
 566         tlen -= off;    /* tlen is used instead of ti->ti_len */
 567         if (off > sizeof (struct tcphdr)) {
 568                 if (isipv6) {
 569 #ifdef INET6
 570                         IP6_EXTHDR_CHECK(m, off0, off, );
 571                         ip6 = mtod(m, struct ip6_hdr *);
 572                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
 573 #endif
 574                 } else {
 575                         if (m->m_len < sizeof(struct ip) + off) {
 576                                 if ((m = m_pullup(m, sizeof (struct ip) + off))
 577                                                 == 0) {
 578                                         tcpstat.tcps_rcvshort++;
 579                                         return;
 580                                 }
 581                                 ip = mtod(m, struct ip *);
 582                                 ipov = (struct ipovly *)ip;
 583                                 th = (struct tcphdr *)((caddr_t)ip + off0);
 584                         }
 585                 }
 586                 optlen = off - sizeof (struct tcphdr);
 587                 optp = (u_char *)(th + 1);
 588         }
 589         thflags = th->th_flags;
 590
 591 #ifdef TCP_DROP_SYNFIN
 592         /*
 593          * If the drop_synfin option is enabled, drop all packets with
 594          * both the SYN and FIN bits set. This prevents e.g. nmap from
 595          * identifying the TCP/IP stack.
 596          *
 597          * This is a violation of the TCP specification.
 598          */
 599         if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
 600                 goto drop;
 601 #endif
 602
 603         /*
 604          * Convert TCP protocol specific fields to host format.
 605          */
 606         th->th_seq = ntohl(th->th_seq);
 607         th->th_ack = ntohl(th->th_ack);
 608         th->th_win = ntohs(th->th_win);
 609         th->th_urp = ntohs(th->th_urp);
 610
 611         /*
 612          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
 613          * until after ip6_savecontrol() is called and before other functions
 614          * which don't want those proto headers.
 615          * Because ip6_savecontrol() is going to parse the mbuf to
 616          * search for data to be passed up to user-land, it wants mbuf
 617          * parameters to be unchanged.
 618          * XXX: the call of ip6_savecontrol() has been obsoleted based on
 619          * latest version of the advanced API (20020110).
 620          */
 621         drop_hdrlen = off0 + off;
 622
 623         /*
 624          * Locate pcb for segment.
 625          */
 626         INP_INFO_WLOCK(&tcbinfo);
 627         headlocked = 1;
 628 findpcb:
 629         KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
 630 #ifdef IPFIREWALL_FORWARD
 631         /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
 632         fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 633
 634         if (fwd_tag != NULL && isipv6 == 0) {   /* IPv6 support is not yet */
 635                 struct sockaddr_in *next_hop;
 636
 637                 next_hop = (struct sockaddr_in *)(fwd_tag+1);
 638                 /*
 639                  * Transparently forwarded. Pretend to be the destination.
 640                  * already got one like this?
 641                  */
 642                 inp = in_pcblookup_hash(&tcbinfo,
 643                                         ip->ip_src, th->th_sport,
 644                                         ip->ip_dst, th->th_dport,
 645                                         0, m->m_pkthdr.rcvif);
 646                 if (!inp) {
 647                         /* It's new.  Try to find the ambushing socket. */
 648                         inp = in_pcblookup_hash(&tcbinfo,
 649                                                 ip->ip_src, th->th_sport,
 650                                                 next_hop->sin_addr,
 651                                                 next_hop->sin_port ?
 652                                                     ntohs(next_hop->sin_port) :
 653                                                     th->th_dport,
 654                                                 INPLOOKUP_WILDCARD,
 655                                                 m->m_pkthdr.rcvif);
 656                 }
 657                 /* Remove the tag from the packet.  We don't need it anymore. */
 658                 m_tag_delete(m, fwd_tag);
 659         } else {
 660 #endif /* IPFIREWALL_FORWARD */
 661                 if (isipv6) {
 662 #ifdef INET6
 663                         inp = in6_pcblookup_hash(&tcbinfo,
 664                                                  &ip6->ip6_src, th->th_sport,
 665                                                  &ip6->ip6_dst, th->th_dport,
 666                                                  INPLOOKUP_WILDCARD,
 667                                                  m->m_pkthdr.rcvif);
 668 #endif
 669                 } else
 670                         inp = in_pcblookup_hash(&tcbinfo,
 671                                                 ip->ip_src, th->th_sport,
 672                                                 ip->ip_dst, th->th_dport,
 673                                                 INPLOOKUP_WILDCARD,
 674                                                 m->m_pkthdr.rcvif);
 675 #ifdef IPFIREWALL_FORWARD
 676         }
 677 #endif /* IPFIREWALL_FORWARD */
 678
 679 #if defined(IPSEC) || defined(FAST_IPSEC)
 680 #ifdef INET6
 681         if (isipv6) {
 682                 if (inp != NULL && ipsec6_in_reject(m, inp)) {
 683 #ifdef IPSEC
 684                         ipsec6stat.in_polvio++;
 685 #endif
 686                         goto drop;
 687                 }
 688         } else
 689 #endif /* INET6 */
 690         if (inp != NULL && ipsec4_in_reject(m, inp)) {
 691 #ifdef IPSEC
 692                 ipsecstat.in_polvio++;
 693 #endif
 694                 goto drop;
 695         }
 696 #endif /*IPSEC || FAST_IPSEC*/
 697
 698         /*
 699          * If the state is CLOSED (i.e., TCB does not exist) then
 700          * all data in the incoming segment is discarded.
 701          * If the TCB exists but is in CLOSED state, it is embryonic,
 702          * but should either do a listen or a connect soon.
 703          */
 704         if (inp == NULL) {
 705                 if (log_in_vain) {
 706 #ifdef INET6
 707                         char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
 708 #else
 709                         char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
 710 #endif
 711
 712                         if (isipv6) {
 713 #ifdef INET6
 714                                 strcpy(dbuf, "[");
 715                                 strcpy(sbuf, "[");
 716                                 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
 717                                 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
 718                                 strcat(dbuf, "]");
 719                                 strcat(sbuf, "]");
 720 #endif
 721                         } else {
 722                                 strcpy(dbuf, inet_ntoa(ip->ip_dst));
 723                                 strcpy(sbuf, inet_ntoa(ip->ip_src));
 724                         }
 725                         switch (log_in_vain) {
 726                         case 1:
 727                                 if ((thflags & TH_SYN) == 0)
 728                                         break;
 729                                 /* FALLTHROUGH */
 730                         case 2:
 731                                 log(LOG_INFO,
 732                                     "Connection attempt to TCP %s:%d "
 733                                     "from %s:%d flags:0x%02x\n",
 734                                     dbuf, ntohs(th->th_dport), sbuf,
 735                                     ntohs(th->th_sport), thflags);
 736                                 break;
 737                         default:
 738                                 break;
 739                         }
 740                 }
 741                 if (blackhole) {
 742                         switch (blackhole) {
 743                         case 1:
 744                                 if (thflags & TH_SYN)
 745                                         goto drop;
 746                                 break;
 747                         case 2:
 748                                 goto drop;
 749                         default:
 750                                 goto drop;
 751                         }
 752                 }
 753                 rstreason = BANDLIM_RST_CLOSEDPORT;
 754                 goto dropwithreset;
 755         }
 756         INP_LOCK(inp);
 757
 758         /* Check the minimum TTL for socket. */
 759         if (inp->inp_ip_minttl != 0) {
 760 #ifdef INET6
 761                 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
 762                         goto drop;
 763                 else
 764 #endif
 765                 if (inp->inp_ip_minttl > ip->ip_ttl)
 766                         goto drop;
 767         }
 768
 769         if (inp->inp_vflag & INP_TIMEWAIT) {
 770                 /*
 771                  * The only option of relevance is TOF_CC, and only if
 772                  * present in a SYN segment.  See tcp_timewait().
 773                  */
 774                 if (thflags & TH_SYN)
 775                         tcp_dooptions(&to, optp, optlen, TO_SYN);
 776                 if (tcp_timewait(inp, &to, th, m, tlen))
 777                         goto findpcb;
 778                 /*
 779                  * tcp_timewait unlocks inp.
 780                  */
 781                 INP_INFO_WUNLOCK(&tcbinfo);
 782                 return;
 783         }
 784         tp = intotcpcb(inp);
 785         if (tp == 0) {
 786                 INP_UNLOCK(inp);
 787                 rstreason = BANDLIM_RST_CLOSEDPORT;
 788                 goto dropwithreset;
 789         }
 790         if (tp->t_state == TCPS_CLOSED)
 791                 goto drop;
 792
 793 #ifdef MAC
 794         INP_LOCK_ASSERT(inp);
 795         if (mac_check_inpcb_deliver(inp, m))
 796                 goto drop;
 797 #endif
 798         so = inp->inp_socket;
 799         KASSERT(so != NULL, ("tcp_input: so == NULL"));
 800 #ifdef TCPDEBUG
 801         if (so->so_options & SO_DEBUG) {
 802                 ostate = tp->t_state;
 803                 if (isipv6)
 804                         bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
 805                 else
 806                         bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
 807                 tcp_savetcp = *th;
 808         }
 809 #endif
 810         if (so->so_options & SO_ACCEPTCONN) {
 811                 struct in_conninfo inc;
 812
 813                 bzero(&inc, sizeof(inc));
 814 #ifdef INET6
 815                 inc.inc_isipv6 = isipv6;
 816 #endif
 817                 if (isipv6) {
 818                         inc.inc6_faddr = ip6->ip6_src;
 819                         inc.inc6_laddr = ip6->ip6_dst;
 820                 } else {
 821                         inc.inc_faddr = ip->ip_src;
 822                         inc.inc_laddr = ip->ip_dst;
 823                 }
 824                 inc.inc_fport = th->th_sport;
 825                 inc.inc_lport = th->th_dport;
 826
 827                 /*
 828                  * If the state is LISTEN then ignore segment if it contains
 829                  * a RST.  If the segment contains an ACK then it is bad and
 830                  * send a RST.  If it does not contain a SYN then it is not
 831                  * interesting; drop it.
 832                  *
 833                  * If the state is SYN_RECEIVED (syncache) and seg contains
 834                  * an ACK, but not for our SYN/ACK, send a RST.  If the seg
 835                  * contains a RST, check the sequence number to see if it
 836                  * is a valid reset segment.
 837                  */
 838                 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
 839                         if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
 840                                 if (!syncache_expand(&inc, th, &so, m)) {
 841                                         /*
 842                                          * No syncache entry, or ACK was not
 843                                          * for our SYN/ACK.  Send a RST.
 844                                          */
 845                                         tcpstat.tcps_badsyn++;
 846                                         rstreason = BANDLIM_RST_OPENPORT;
 847                                         goto dropwithreset;
 848                                 }
 849                                 if (so == NULL) {
 850                                         /*
 851                                          * Could not complete 3-way handshake,
 852                                          * connection is being closed down, and
 853                                          * syncache has free'd mbuf.
 854                                          */
 855                                         INP_UNLOCK(inp);
 856                                         INP_INFO_WUNLOCK(&tcbinfo);
 857                                         return;
 858                                 }
 859                                 /*
 860                                  * Socket is created in state SYN_RECEIVED.
 861                                  * Continue processing segment.
 862                                  */
 863                                 INP_UNLOCK(inp);
 864                                 inp = sotoinpcb(so);
 865                                 INP_LOCK(inp);
 866                                 tp = intotcpcb(inp);
 867                                 /*
 868                                  * This is what would have happened in
 869                                  * tcp_output() when the SYN,ACK was sent.
 870                                  */
 871                                 tp->snd_up = tp->snd_una;
 872                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
 873                                 tp->last_ack_sent = tp->rcv_nxt;
 874                                 goto after_listen;
 875                         }
 876                         if (thflags & TH_RST) {
 877                                 syncache_chkrst(&inc, th);
 878                                 goto drop;
 879                         }
 880                         if (thflags & TH_ACK) {
 881                                 syncache_badack(&inc);
 882                                 tcpstat.tcps_badsyn++;
 883                                 rstreason = BANDLIM_RST_OPENPORT;
 884                                 goto dropwithreset;
 885                         }
 886                         goto drop;
 887                 }
 888
 889                 /*
 890                  * Segment's flags are (SYN) or (SYN|FIN).
 891                  */
 892 #ifdef INET6
 893                 /*
 894                  * If deprecated address is forbidden,
 895                  * we do not accept SYN to deprecated interface
 896                  * address to prevent any new inbound connection from
 897                  * getting established.
 898                  * When we do not accept SYN, we send a TCP RST,
 899                  * with deprecated source address (instead of dropping
 900                  * it).  We compromise it as it is much better for peer
 901                  * to send a RST, and RST will be the final packet
 902                  * for the exchange.
 903                  *
 904                  * If we do not forbid deprecated addresses, we accept
 905                  * the SYN packet.  RFC2462 does not suggest dropping
 906                  * SYN in this case.
 907                  * If we decipher RFC2462 5.5.4, it says like this:
 908                  * 1. use of deprecated addr with existing
 909                  *    communication is okay - "SHOULD continue to be
 910                  *    used"
 911                  * 2. use of it with new communication:
 912                  *   (2a) "SHOULD NOT be used if alternate address
 913                  *        with sufficient scope is available"
 914                  *   (2b) nothing mentioned otherwise.
 915                  * Here we fall into (2b) case as we have no choice in
 916                  * our source address selection - we must obey the peer.
 917                  *
 918                  * The wording in RFC2462 is confusing, and there are
 919                  * multiple description text for deprecated address
 920                  * handling - worse, they are not exactly the same.
 921                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
 922                  */
 923                 if (isipv6 && !ip6_use_deprecated) {
 924                         struct in6_ifaddr *ia6;
 925
 926                         if ((ia6 = ip6_getdstifaddr(m)) &&
 927                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
 928                                 INP_UNLOCK(inp);
 929                                 tp = NULL;
 930                                 rstreason = BANDLIM_RST_OPENPORT;
 931                                 goto dropwithreset;
 932                         }
 933                 }
 934 #endif
 935                 /*
 936                  * If it is from this socket, drop it, it must be forged.
 937                  * Don't bother responding if the destination was a broadcast.
 938                  */
 939                 if (th->th_dport == th->th_sport) {
 940                         if (isipv6) {
 941                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
 942                                                        &ip6->ip6_src))
 943                                         goto drop;
 944                         } else {
 945                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
 946                                         goto drop;
 947                         }
 948                 }
 949                 /*
 950                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
 951                  *
 952                  * Note that it is quite possible to receive unicast
 953                  * link-layer packets with a broadcast IP address. Use
 954                  * in_broadcast() to find them.
 955                  */
 956                 if (m->m_flags & (M_BCAST|M_MCAST))
 957                         goto drop;
 958                 if (isipv6) {
 959                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
 960                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
 961                                 goto drop;
 962                 } else {
 963                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 964                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 965                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
 966                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
 967                                 goto drop;
 968                 }
 969                 /*
 970                  * SYN appears to be valid; create compressed TCP state
 971                  * for syncache, or perform t/tcp connection.
 972                  */
 973                 if (so->so_qlen <= so->so_qlimit) {
 974 #ifdef TCPDEBUG
 975                         if (so->so_options & SO_DEBUG)
 976                                 tcp_trace(TA_INPUT, ostate, tp,
 977                                     (void *)tcp_saveipgen, &tcp_savetcp, 0);
 978 #endif
 979                         tcp_dooptions(&to, optp, optlen, TO_SYN);
 980                         if (!syncache_add(&inc, &to, th, inp, &so, m))
 981                                 goto drop;      /* XXX: does not happen */
 982                         if (so == NULL) {
 983                                 /*
 984                                  * Entry added to syncache, mbuf used to
 985                                  * send SYN,ACK packet.  Everything unlocked
 986                                  * already.
 987                                  */
 988                                 return;
 989                         }
 990                         panic("T/TCP not supported at the moment");
 991 #if 0 /* T/TCP */
 992                         /*
 993                          * Segment passed TAO tests.
 994                          * XXX: Can't happen at the moment.
 995                          */
 996                         INP_UNLOCK(inp);
 997                         inp = sotoinpcb(so);
 998                         INP_LOCK(inp);
 999                         tp = intotcpcb(inp);
1000                         tp->t_starttime = ticks;
1001                         tp->t_state = TCPS_ESTABLISHED;
1002
1003                         /*
1004                          * T/TCP logic:
1005                          * If there is a FIN or if there is data, then
1006                          * delay SYN,ACK(SYN) in the hope of piggy-backing
1007                          * it on a response segment.  Otherwise must send
1008                          * ACK now in case the other side is slow starting.
1009                          */
1010                         if (thflags & TH_FIN || tlen != 0)
1011                                 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1012                         else
1013                                 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1014                         tiwin = th->th_win << tp->snd_scale;
1015                         tcpstat.tcps_connects++;
1016                         soisconnected(so);
1017                         goto trimthenstep6;
1018 #endif  /* T/TCP */
1019                 }
1020                 goto drop;
1021         }
1022 after_listen:
1023         KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1024         INP_LOCK_ASSERT(inp);
1025
1026         /* Syncache takes care of sockets in the listen state. */
1027         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1028
1029         /*
1030          * This is the second part of the MSS DoS prevention code (after
1031          * minmss on the sending side) and it deals with too many too small
1032          * tcp packets in a too short timeframe (1 second).
1033          *
1034          * For every full second we count the number of received packets
1035          * and bytes. If we get a lot of packets per second for this connection
1036          * (tcp_minmssoverload) we take a closer look at it and compute the
1037          * average packet size for the past second. If that is less than
1038          * tcp_minmss we get too many packets with very small payload which
1039          * is not good and burdens our system (and every packet generates
1040          * a wakeup to the process connected to our socket). We can reasonable
1041          * expect this to be small packet DoS attack to exhaust our CPU
1042          * cycles.
1043          *
1044          * Care has to be taken for the minimum packet overload value. This
1045          * value defines the minimum number of packets per second before we
1046          * start to worry. This must not be too low to avoid killing for
1047          * example interactive connections with many small packets like
1048          * telnet or SSH.
1049          *
1050          * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1051          * this check.
1052          *
1053          * Account for packet if payload packet, skip over ACK, etc.
1054          */
1055         if (tcp_minmss && tcp_minmssoverload &&
1056             tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1057                 if ((unsigned int)(tp->rcv_second - ticks) < hz) {
1058                         tp->rcv_pps++;
1059                         tp->rcv_byps += tlen + off;
1060                         if (tp->rcv_pps > tcp_minmssoverload) {
1061                                 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1062                                         printf("too many small tcp packets from "
1063                                                "%s:%u, av. %lubyte/packet, "
1064                                                "dropping connection\n",
1065 #ifdef INET6
1066                                                 isipv6 ?
1067                                                 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1068 #endif
1069                                                 inet_ntoa(inp->inp_inc.inc_faddr),
1070                                                 inp->inp_inc.inc_fport,
1071                                                 tp->rcv_byps / tp->rcv_pps);
1072                                         KASSERT(headlocked, ("tcp_input: "
1073                                             "after_listen: tcp_drop: head "
1074                                             "not locked"));
1075                                         tp = tcp_drop(tp, ECONNRESET);
1076                                         tcpstat.tcps_minmssdrops++;
1077                                         goto drop;
1078                                 }
1079                         }
1080                 } else {
1081                         tp->rcv_second = ticks + hz;
1082                         tp->rcv_pps = 1;
1083                         tp->rcv_byps = tlen + off;
1084                 }
1085         }
1086
1087         /*
1088          * Segment received on connection.
1089          * Reset idle time and keep-alive timer.
1090          */
1091         tp->t_rcvtime = ticks;
1092         if (TCPS_HAVEESTABLISHED(tp->t_state))
1093                 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1094
1095         /*
1096          * Unscale the window into a 32-bit value.
1097          * This value is bogus for the TCPS_SYN_SENT state
1098          * and is overwritten later.
1099          */
1100         tiwin = th->th_win << tp->snd_scale;
1101
1102         /*
1103          * Parse options on any incoming segment.
1104          */
1105         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0);
1106
1107         /*
1108          * If echoed timestamp is later than the current time,
1109          * fall back to non RFC1323 RTT calculation.
1110          */
1111         if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
1112             TSTMP_GT(to.to_tsecr, ticks))
1113                 to.to_tsecr = 0;
1114
1115         /*
1116          * Process options only when we get SYN/ACK back. The SYN case
1117          * for incoming connections is handled in tcp_syncache.
1118          * XXX this is traditional behavior, may need to be cleaned up.
1119          */
1120         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1121                 if ((to.to_flags & TOF_SCALE) &&
1122                     (tp->t_flags & TF_REQ_SCALE)) {
1123                         tp->t_flags |= TF_RCVD_SCALE;
1124                         tp->snd_scale = to.to_requested_s_scale;
1125                         tp->snd_wnd = th->th_win << tp->snd_scale;
1126                         tiwin = tp->snd_wnd;
1127                 }
1128                 if (to.to_flags & TOF_TS) {
1129                         tp->t_flags |= TF_RCVD_TSTMP;
1130                         tp->ts_recent = to.to_tsval;
1131                         tp->ts_recent_age = ticks;
1132                 }
1133                 if (to.to_flags & TOF_MSS)
1134                         tcp_mss(tp, to.to_mss);
1135                 if (tp->sack_enable) {
1136                         if (!(to.to_flags & TOF_SACK))
1137                                 tp->sack_enable = 0;
1138                         else
1139                                 tp->t_flags |= TF_SACK_PERMIT;
1140                 }
1141
1142         }
1143
1144         /*
1145          * Header prediction: check for the two common cases
1146          * of a uni-directional data xfer.  If the packet has
1147          * no control flags, is in-sequence, the window didn't
1148          * change and we're not retransmitting, it's a
1149          * candidate.  If the length is zero and the ack moved
1150          * forward, we're the sender side of the xfer.  Just
1151          * free the data acked & wake any higher level process
1152          * that was blocked waiting for space.  If the length
1153          * is non-zero and the ack didn't move, we're the
1154          * receiver side.  If we're getting packets in-order
1155          * (the reassembly queue is empty), add the data to
1156          * the socket buffer and note that we need a delayed ack.
1157          * Make sure that the hidden state-flags are also off.
1158          * Since we check for TCPS_ESTABLISHED above, it can only
1159          * be TH_NEEDSYN.
1160          */
1161         if (tp->t_state == TCPS_ESTABLISHED &&
1162             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1163             ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1164             ((to.to_flags & TOF_TS) == 0 ||
1165              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1166              th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1167              tp->snd_nxt == tp->snd_max) {
1168
1169                 /*
1170                  * If last ACK falls within this segment's sequence numbers,
1171                  * record the timestamp.
1172                  * NOTE that the test is modified according to the latest
1173                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1174                  */
1175                 if ((to.to_flags & TOF_TS) != 0 &&
1176                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1177                         tp->ts_recent_age = ticks;
1178                         tp->ts_recent = to.to_tsval;
1179                 }
1180
1181                 if (tlen == 0) {
1182                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1183                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1184                             tp->snd_cwnd >= tp->snd_wnd &&
1185                             ((!tcp_do_newreno && !tp->sack_enable &&
1186                               tp->t_dupacks < tcprexmtthresh) ||
1187                              ((tcp_do_newreno || tp->sack_enable) &&
1188                               !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1189                               TAILQ_EMPTY(&tp->snd_holes)))) {
1190                                 KASSERT(headlocked, ("headlocked"));
1191                                 INP_INFO_WUNLOCK(&tcbinfo);
1192                                 headlocked = 0;
1193                                 /*
1194                                  * this is a pure ack for outstanding data.
1195                                  */
1196                                 ++tcpstat.tcps_predack;
1197                                 /*
1198                                  * "bad retransmit" recovery
1199                                  */
1200                                 if (tp->t_rxtshift == 1 &&
1201                                     ticks < tp->t_badrxtwin) {
1202                                         ++tcpstat.tcps_sndrexmitbad;
1203                                         tp->snd_cwnd = tp->snd_cwnd_prev;
1204                                         tp->snd_ssthresh =
1205                                             tp->snd_ssthresh_prev;
1206                                         tp->snd_recover = tp->snd_recover_prev;
1207                                         if (tp->t_flags & TF_WASFRECOVERY)
1208                                             ENTER_FASTRECOVERY(tp);
1209                                         tp->snd_nxt = tp->snd_max;
1210                                         tp->t_badrxtwin = 0;
1211                                 }
1212
1213                                 /*
1214                                  * Recalculate the transmit timer / rtt.
1215                                  *
1216                                  * Some boxes send broken timestamp replies
1217                                  * during the SYN+ACK phase, ignore
1218                                  * timestamps of 0 or we could calculate a
1219                                  * huge RTT and blow up the retransmit timer.
1220                                  */
1221                                 if ((to.to_flags & TOF_TS) != 0 &&
1222                                     to.to_tsecr) {
1223                                         if (!tp->t_rttlow ||
1224                                             tp->t_rttlow > ticks - to.to_tsecr)
1225                                                 tp->t_rttlow = ticks - to.to_tsecr;
1226                                         tcp_xmit_timer(tp,
1227                                             ticks - to.to_tsecr + 1);
1228                                 } else if (tp->t_rtttime &&
1229                                             SEQ_GT(th->th_ack, tp->t_rtseq)) {
1230                                         if (!tp->t_rttlow ||
1231                                             tp->t_rttlow > ticks - tp->t_rtttime)
1232                                                 tp->t_rttlow = ticks - tp->t_rtttime;
1233                                         tcp_xmit_timer(tp,
1234                                                         ticks - tp->t_rtttime);
1235                                 }
1236                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1237                                 acked = th->th_ack - tp->snd_una;
1238                                 tcpstat.tcps_rcvackpack++;
1239                                 tcpstat.tcps_rcvackbyte += acked;
1240                                 sbdrop(&so->so_snd, acked);
1241                                 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1242                                     SEQ_LEQ(th->th_ack, tp->snd_recover))
1243                                         tp->snd_recover = th->th_ack - 1;
1244                                 tp->snd_una = th->th_ack;
1245                                 /*
1246                                  * pull snd_wl2 up to prevent seq wrap relative
1247                                  * to th_ack.
1248                                  */
1249                                 tp->snd_wl2 = th->th_ack;
1250                                 tp->t_dupacks = 0;
1251                                 m_freem(m);
1252                                 ND6_HINT(tp); /* some progress has been done */
1253
1254                                 /*
1255                                  * If all outstanding data are acked, stop
1256                                  * retransmit timer, otherwise restart timer
1257                                  * using current (possibly backed-off) value.
1258                                  * If process is waiting for space,
1259                                  * wakeup/selwakeup/signal.  If data
1260                                  * are ready to send, let tcp_output
1261                                  * decide between more output or persist.
1262
1263 #ifdef TCPDEBUG
1264                                 if (so->so_options & SO_DEBUG)
1265                                         tcp_trace(TA_INPUT, ostate, tp,
1266                                             (void *)tcp_saveipgen,
1267                                             &tcp_savetcp, 0);
1268 #endif
1269                                  */
1270                                 if (tp->snd_una == tp->snd_max)
1271                                         callout_stop(tp->tt_rexmt);
1272                                 else if (!callout_active(tp->tt_persist))
1273                                         callout_reset(tp->tt_rexmt,
1274                                                       tp->t_rxtcur,
1275                                                       tcp_timer_rexmt, tp);
1276
1277                                 sowwakeup(so);
1278                                 if (so->so_snd.sb_cc)
1279                                         (void) tcp_output(tp);
1280                                 goto check_delack;
1281                         }
1282                 } else if (th->th_ack == tp->snd_una &&
1283                     LIST_EMPTY(&tp->t_segq) &&
1284                     tlen <= sbspace(&so->so_rcv)) {
1285                         KASSERT(headlocked, ("headlocked"));
1286                         INP_INFO_WUNLOCK(&tcbinfo);
1287                         headlocked = 0;
1288                         /*
1289                          * this is a pure, in-sequence data packet
1290                          * with nothing on the reassembly queue and
1291                          * we have enough buffer space to take it.
1292                          */
1293                         /* Clean receiver SACK report if present */
1294                         if (tp->sack_enable && tp->rcv_numsacks)
1295                                 tcp_clean_sackreport(tp);
1296                         ++tcpstat.tcps_preddat;
1297                         tp->rcv_nxt += tlen;
1298                         /*
1299                          * Pull snd_wl1 up to prevent seq wrap relative to
1300                          * th_seq.
1301                          */
1302                         tp->snd_wl1 = th->th_seq;
1303                         /*
1304                          * Pull rcv_up up to prevent seq wrap relative to
1305                          * rcv_nxt.
1306                          */
1307                         tp->rcv_up = tp->rcv_nxt;
1308                         tcpstat.tcps_rcvpack++;
1309                         tcpstat.tcps_rcvbyte += tlen;
1310                         ND6_HINT(tp);   /* some progress has been done */
1311                         /*
1312 #ifdef TCPDEBUG
1313                         if (so->so_options & SO_DEBUG)
1314                                 tcp_trace(TA_INPUT, ostate, tp,
1315                                     (void *)tcp_saveipgen, &tcp_savetcp, 0);
1316 #endif
1317                          * Add data to socket buffer.
1318                          */
1319                         SOCKBUF_LOCK(&so->so_rcv);
1320                         if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1321                                 m_freem(m);
1322                         } else {
1323                                 m_adj(m, drop_hdrlen);  /* delayed header drop */
1324                                 sbappendstream_locked(&so->so_rcv, m);
1325                         }
1326                         sorwakeup_locked(so);
1327                         if (DELAY_ACK(tp)) {
1328                                 tp->t_flags |= TF_DELACK;
1329                         } else {
1330                                 tp->t_flags |= TF_ACKNOW;
1331                                 tcp_output(tp);
1332                         }
1333                         goto check_delack;
1334                 }
1335         }
1336
1337         /*
1338          * Calculate amount of space in receive window,
1339          * and then do TCP input processing.
1340          * Receive window is amount of space in rcv queue,
1341          * but not less than advertised window.
1342          */
1343         { int win;
1344
1345         win = sbspace(&so->so_rcv);
1346         if (win < 0)
1347                 win = 0;
1348         tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1349         }
1350
1351         switch (tp->t_state) {
1352
1353         /*
1354          * If the state is SYN_RECEIVED:
1355          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1356          */
1357         case TCPS_SYN_RECEIVED:
1358                 if ((thflags & TH_ACK) &&
1359                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1360                      SEQ_GT(th->th_ack, tp->snd_max))) {
1361                                 rstreason = BANDLIM_RST_OPENPORT;
1362                                 goto dropwithreset;
1363                 }
1364                 break;
1365
1366         /*
1367          * If the state is SYN_SENT:
1368          *      if seg contains an ACK, but not for our SYN, drop the input.
1369          *      if seg contains a RST, then drop the connection.
1370          *      if seg does not contain SYN, then drop it.
1371          * Otherwise this is an acceptable SYN segment
1372          *      initialize tp->rcv_nxt and tp->irs
1373          *      if seg contains ack then advance tp->snd_una
1374          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1375          *      arrange for segment to be acked (eventually)
1376          *      continue processing rest of data/controls, beginning with URG
1377          */
1378         case TCPS_SYN_SENT:
1379                 if ((thflags & TH_ACK) &&
1380                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1381                      SEQ_GT(th->th_ack, tp->snd_max))) {
1382                         rstreason = BANDLIM_UNLIMITED;
1383                         goto dropwithreset;
1384                 }
1385                 if (thflags & TH_RST) {
1386                         if (thflags & TH_ACK) {
1387                                 KASSERT(headlocked, ("tcp_input: after_listen"
1388                                     ": tcp_drop.2: head not locked"));
1389                                 tp = tcp_drop(tp, ECONNREFUSED);
1390                         }
1391                         goto drop;
1392                 }
1393                 if ((thflags & TH_SYN) == 0)
1394                         goto drop;
1395
1396                 /* Initial send window, already scaled. */
1397                 tp->snd_wnd = th->th_win;
1398
1399                 tp->irs = th->th_seq;
1400                 tcp_rcvseqinit(tp);
1401                 if (thflags & TH_ACK) {
1402                         tcpstat.tcps_connects++;
1403                         soisconnected(so);
1404 #ifdef MAC
1405                         SOCK_LOCK(so);
1406                         mac_set_socket_peer_from_mbuf(m, so);
1407                         SOCK_UNLOCK(so);
1408 #endif
1409                         /* Do window scaling on this connection? */
1410                         if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1411                                 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1412                                 tp->rcv_scale = tp->request_r_scale;
1413                         }
1414                         tp->rcv_adv += tp->rcv_wnd;
1415                         tp->snd_una++;          /* SYN is acked */
1416                         /*
1417                          * If there's data, delay ACK; if there's also a FIN
1418                          * ACKNOW will be turned on later.
1419                          */
1420                         if (DELAY_ACK(tp) && tlen != 0)
1421                                 callout_reset(tp->tt_delack, tcp_delacktime,
1422                                     tcp_timer_delack, tp);
1423                         else
1424                                 tp->t_flags |= TF_ACKNOW;
1425                         /*
1426                          * Received <SYN,ACK> in SYN_SENT[*] state.
1427                          * Transitions:
1428                          *      SYN_SENT  --> ESTABLISHED
1429                          *      SYN_SENT* --> FIN_WAIT_1
1430                          */
1431                         tp->t_starttime = ticks;
1432                         if (tp->t_flags & TF_NEEDFIN) {
1433                                 tp->t_state = TCPS_FIN_WAIT_1;
1434                                 tp->t_flags &= ~TF_NEEDFIN;
1435                                 thflags &= ~TH_SYN;
1436                         } else {
1437                                 tp->t_state = TCPS_ESTABLISHED;
1438                                 callout_reset(tp->tt_keep, tcp_keepidle,
1439                                               tcp_timer_keep, tp);
1440                         }
1441                 } else {
1442                         /*
1443                          * Received initial SYN in SYN-SENT[*] state =>
1444                          * simultaneous open.  If segment contains CC option
1445                          * and there is a cached CC, apply TAO test.
1446                          * If it succeeds, connection is * half-synchronized.
1447                          * Otherwise, do 3-way handshake:
1448                          *        SYN-SENT -> SYN-RECEIVED
1449                          *        SYN-SENT* -> SYN-RECEIVED*
1450                          * If there was no CC option, clear cached CC value.
1451                          */
1452                         tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1453                         callout_stop(tp->tt_rexmt);
1454                         tp->t_state = TCPS_SYN_RECEIVED;
1455                 }
1456
1457 #if 0 /* T/TCP */
1458 trimthenstep6:
1459 #endif
1460                 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1461                     "locked"));
1462                 INP_LOCK_ASSERT(inp);
1463
1464                 /*
1465                  * Advance th->th_seq to correspond to first data byte.
1466                  * If data, trim to stay within window,
1467                  * dropping FIN if necessary.
1468                  */
1469                 th->th_seq++;
1470                 if (tlen > tp->rcv_wnd) {
1471                         todrop = tlen - tp->rcv_wnd;
1472                         m_adj(m, -todrop);
1473                         tlen = tp->rcv_wnd;
1474                         thflags &= ~TH_FIN;
1475                         tcpstat.tcps_rcvpackafterwin++;
1476                         tcpstat.tcps_rcvbyteafterwin += todrop;
1477                 }
1478                 tp->snd_wl1 = th->th_seq - 1;
1479                 tp->rcv_up = th->th_seq;
1480                 /*
1481                  * Client side of transaction: already sent SYN and data.
1482                  * If the remote host used T/TCP to validate the SYN,
1483                  * our data will be ACK'd; if so, enter normal data segment
1484                  * processing in the middle of step 5, ack processing.
1485                  * Otherwise, goto step 6.
1486                  */
1487                 if (thflags & TH_ACK)
1488                         goto process_ACK;
1489
1490                 goto step6;
1491
1492         /*
1493          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1494          *      do normal processing.
1495          *
1496          * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
1497          */
1498         case TCPS_LAST_ACK:
1499         case TCPS_CLOSING:
1500         case TCPS_TIME_WAIT:
1501                 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1502                 break;  /* continue normal processing */
1503         }
1504
1505         /*
1506          * States other than LISTEN or SYN_SENT.
1507          * First check the RST flag and sequence number since reset segments
1508          * are exempt from the timestamp and connection count tests.  This
1509          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1510          * below which allowed reset segments in half the sequence space
1511          * to fall though and be processed (which gives forged reset
1512          * segments with a random sequence number a 50 percent chance of
1513          * killing a connection).
1514          * Then check timestamp, if present.
1515          * Then check the connection count, if present.
1516          * Then check that at least some bytes of segment are within
1517          * receive window.  If segment begins before rcv_nxt,
1518          * drop leading data (and SYN); if nothing left, just ack.
1519          *
1520          *
1521          * If the RST bit is set, check the sequence number to see
1522          * if this is a valid reset segment.
1523          * RFC 793 page 37:
1524          *   In all states except SYN-SENT, all reset (RST) segments
1525          *   are validated by checking their SEQ-fields.  A reset is
1526          *   valid if its sequence number is in the window.
1527          * Note: this does not take into account delayed ACKs, so
1528          *   we should test against last_ack_sent instead of rcv_nxt.
1529          *   The sequence number in the reset segment is normally an
1530          *   echo of our outgoing acknowlegement numbers, but some hosts
1531          *   send a reset with the sequence number at the rightmost edge
1532          *   of our receive window, and we have to handle this case.
1533          * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1534          *   that brute force RST attacks are possible.  To combat this,
1535          *   we use a much stricter check while in the ESTABLISHED state,
1536          *   only accepting RSTs where the sequence number is equal to
1537          *   last_ack_sent.  In all other states (the states in which a
1538          *   RST is more likely), the more permissive check is used.
1539          * If we have multiple segments in flight, the intial reset
1540          * segment sequence numbers will be to the left of last_ack_sent,
1541          * but they will eventually catch up.
1542          * In any case, it never made sense to trim reset segments to
1543          * fit the receive window since RFC 1122 says:
1544          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1545          *
1546          *    A TCP SHOULD allow a received RST segment to include data.
1547          *
1548          *    DISCUSSION
1549          *         It has been suggested that a RST segment could contain
1550          *         ASCII text that encoded and explained the cause of the
1551          *         RST.  No standard has yet been established for such
1552          *         data.
1553          *
1554          * If the reset segment passes the sequence number test examine
1555          * the state:
1556          *    SYN_RECEIVED STATE:
1557          *      If passive open, return to LISTEN state.
1558          *      If active open, inform user that connection was refused.
1559          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1560          *      Inform user that connection was reset, and close tcb.
1561          *    CLOSING, LAST_ACK STATES:
1562          *      Close the tcb.
1563          *    TIME_WAIT STATE:
1564          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1565          *      RFC 1337.
1566          */
1567         if (thflags & TH_RST) {
1568                 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1569                     SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1570                     (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1571                         switch (tp->t_state) {
1572
1573                         case TCPS_SYN_RECEIVED:
1574                                 so->so_error = ECONNREFUSED;
1575                                 goto close;
1576
1577                         case TCPS_ESTABLISHED:
1578                                 if (tp->last_ack_sent != th->th_seq &&
1579                                     tcp_insecure_rst == 0) {
1580                                         tcpstat.tcps_badrst++;
1581                                         goto drop;
1582                                 }
1583                         case TCPS_FIN_WAIT_1:
1584                         case TCPS_FIN_WAIT_2:
1585                         case TCPS_CLOSE_WAIT:
1586                                 so->so_error = ECONNRESET;
1587                         close:
1588                                 tp->t_state = TCPS_CLOSED;
1589                                 tcpstat.tcps_drops++;
1590                                 KASSERT(headlocked, ("tcp_input: "
1591                                     "trimthenstep6: tcp_close: head not "
1592                                     "locked"));
1593                                 tp = tcp_close(tp);
1594                                 break;
1595
1596                         case TCPS_CLOSING:
1597                         case TCPS_LAST_ACK:
1598                                 KASSERT(headlocked, ("trimthenstep6: "
1599                                     "tcp_close.2: head not locked"));
1600                                 tp = tcp_close(tp);
1601                                 break;
1602
1603                         case TCPS_TIME_WAIT:
1604                                 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1605                                     ("timewait"));
1606                                 break;
1607                         }
1608                 }
1609                 goto drop;
1610         }
1611
1612         /*
1613          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1614          * and it's less than ts_recent, drop it.
1615          */
1616         if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1617             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1618
1619                 /* Check to see if ts_recent is over 24 days old.  */
1620                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1621                         /*
1622                          * Invalidate ts_recent.  If this segment updates
1623                          * ts_recent, the age will be reset later and ts_recent
1624                          * will get a valid value.  If it does not, setting
1625                          * ts_recent to zero will at least satisfy the
1626                          * requirement that zero be placed in the timestamp
1627                          * echo reply when ts_recent isn't valid.  The
1628                          * age isn't reset until we get a valid ts_recent
1629                          * because we don't want out-of-order segments to be
1630                          * dropped when ts_recent is old.
1631                          */
1632                         tp->ts_recent = 0;
1633                 } else {
1634                         tcpstat.tcps_rcvduppack++;
1635                         tcpstat.tcps_rcvdupbyte += tlen;
1636                         tcpstat.tcps_pawsdrop++;
1637                         if (tlen)
1638                                 goto dropafterack;
1639                         goto drop;
1640                 }
1641         }
1642
1643         /*
1644          * In the SYN-RECEIVED state, validate that the packet belongs to
1645          * this connection before trimming the data to fit the receive
1646          * window.  Check the sequence number versus IRS since we know
1647          * the sequence numbers haven't wrapped.  This is a partial fix
1648          * for the "LAND" DoS attack.
1649          */
1650         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1651                 rstreason = BANDLIM_RST_OPENPORT;
1652                 goto dropwithreset;
1653         }
1654
1655         todrop = tp->rcv_nxt - th->th_seq;
1656         if (todrop > 0) {
1657                 if (thflags & TH_SYN) {
1658                         thflags &= ~TH_SYN;
1659                         th->th_seq++;
1660                         if (th->th_urp > 1)
1661                                 th->th_urp--;
1662                         else
1663                                 thflags &= ~TH_URG;
1664                         todrop--;
1665                 }
1666                 /*
1667                  * Following if statement from Stevens, vol. 2, p. 960.
1668                  */
1669                 if (todrop > tlen
1670                     || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1671                         /*
1672                          * Any valid FIN must be to the left of the window.
1673                          * At this point the FIN must be a duplicate or out
1674                          * of sequence; drop it.
1675                          */
1676                         thflags &= ~TH_FIN;
1677
1678                         /*
1679                          * Send an ACK to resynchronize and drop any data.
1680                          * But keep on processing for RST or ACK.
1681                          */
1682                         tp->t_flags |= TF_ACKNOW;
1683                         todrop = tlen;
1684                         tcpstat.tcps_rcvduppack++;
1685                         tcpstat.tcps_rcvdupbyte += todrop;
1686                 } else {
1687                         tcpstat.tcps_rcvpartduppack++;
1688                         tcpstat.tcps_rcvpartdupbyte += todrop;
1689                 }
1690                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1691                 th->th_seq += todrop;
1692                 tlen -= todrop;
1693                 if (th->th_urp > todrop)
1694                         th->th_urp -= todrop;
1695                 else {
1696                         thflags &= ~TH_URG;
1697                         th->th_urp = 0;
1698                 }
1699         }
1700
1701         /*
1702          * If new data are received on a connection after the
1703          * user processes are gone, then RST the other end.
1704          */
1705         if ((so->so_state & SS_NOFDREF) &&
1706             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1707                 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1708                     "locked"));
1709                 tp = tcp_close(tp);
1710                 tcpstat.tcps_rcvafterclose++;
1711                 rstreason = BANDLIM_UNLIMITED;
1712                 goto dropwithreset;
1713         }
1714
1715         /*
1716          * If segment ends after window, drop trailing data
1717          * (and PUSH and FIN); if nothing left, just ACK.
1718          */
1719         todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1720         if (todrop > 0) {
1721                 tcpstat.tcps_rcvpackafterwin++;
1722                 if (todrop >= tlen) {
1723                         tcpstat.tcps_rcvbyteafterwin += tlen;
1724                         /*
1725                          * If a new connection request is received
1726                          * while in TIME_WAIT, drop the old connection
1727                          * and start over if the sequence numbers
1728                          * are above the previous ones.
1729                          */
1730                         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1731                         if (thflags & TH_SYN &&
1732                             tp->t_state == TCPS_TIME_WAIT &&
1733                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1734                                 KASSERT(headlocked, ("trimthenstep6: "
1735                                     "tcp_close.4: head not locked"));
1736                                 tp = tcp_close(tp);
1737                                 goto findpcb;
1738                         }
1739                         /*
1740                          * If window is closed can only take segments at
1741                          * window edge, and have to drop data and PUSH from
1742                          * incoming segments.  Continue processing, but
1743                          * remember to ack.  Otherwise, drop segment
1744                          * and ack.
1745                          */
1746                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1747                                 tp->t_flags |= TF_ACKNOW;
1748                                 tcpstat.tcps_rcvwinprobe++;
1749                         } else
1750                                 goto dropafterack;
1751                 } else
1752                         tcpstat.tcps_rcvbyteafterwin += todrop;
1753                 m_adj(m, -todrop);
1754                 tlen -= todrop;
1755                 thflags &= ~(TH_PUSH|TH_FIN);
1756         }
1757
1758         /*
1759          * If last ACK falls within this segment's sequence numbers,
1760          * record its timestamp.
1761          * NOTE:
1762          * 1) That the test incorporates suggestions from the latest
1763          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1764          * 2) That updating only on newer timestamps interferes with
1765          *    our earlier PAWS tests, so this check should be solely
1766          *    predicated on the sequence space of this segment.
1767          * 3) That we modify the segment boundary check to be
1768          *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
1769          *    instead of RFC1323's
1770          *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
1771          *    This modified check allows us to overcome RFC1323's
1772          *    limitations as described in Stevens TCP/IP Illustrated
1773          *    Vol. 2 p.869. In such cases, we can still calculate the
1774          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1775          */
1776         if ((to.to_flags & TOF_TS) != 0 &&
1777             SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1778             SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1779                 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1780                 tp->ts_recent_age = ticks;
1781                 tp->ts_recent = to.to_tsval;
1782         }
1783
1784         /*
1785          * If a SYN is in the window, then this is an
1786          * error and we send an RST and drop the connection.
1787          */
1788         if (thflags & TH_SYN) {
1789                 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1790                     "head not locked"));
1791                 tp = tcp_drop(tp, ECONNRESET);
1792                 rstreason = BANDLIM_UNLIMITED;
1793                 goto drop;
1794         }
1795
1796         /*
1797          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1798          * flag is on (half-synchronized state), then queue data for
1799          * later processing; else drop segment and return.
1800          */
1801         if ((thflags & TH_ACK) == 0) {
1802                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1803                     (tp->t_flags & TF_NEEDSYN))
1804                         goto step6;
1805                 else if (tp->t_flags & TF_ACKNOW)
1806                         goto dropafterack;
1807                 else
1808                         goto drop;
1809         }
1810
1811         /*
1812          * Ack processing.
1813          */
1814         switch (tp->t_state) {
1815
1816         /*
1817          * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1818          * ESTABLISHED state and continue processing.
1819          * The ACK was checked above.
1820          */
1821         case TCPS_SYN_RECEIVED:
1822
1823                 tcpstat.tcps_connects++;
1824                 soisconnected(so);
1825                 /* Do window scaling? */
1826                 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1827                         (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1828                         tp->rcv_scale = tp->request_r_scale;
1829                         tp->snd_wnd = tiwin;
1830                 }
1831                 /*
1832                  * Make transitions:
1833                  *      SYN-RECEIVED  -> ESTABLISHED
1834                  *      SYN-RECEIVED* -> FIN-WAIT-1
1835                  */
1836                 tp->t_starttime = ticks;
1837                 if (tp->t_flags & TF_NEEDFIN) {
1838                         tp->t_state = TCPS_FIN_WAIT_1;
1839                         tp->t_flags &= ~TF_NEEDFIN;
1840                 } else {
1841                         tp->t_state = TCPS_ESTABLISHED;
1842                         callout_reset(tp->tt_keep, tcp_keepidle,
1843                                       tcp_timer_keep, tp);
1844                 }
1845                 /*
1846                  * If segment contains data or ACK, will call tcp_reass()
1847                  * later; if not, do so now to pass queued data to user.
1848                  */
1849                 if (tlen == 0 && (thflags & TH_FIN) == 0)
1850                         (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1851                             (struct mbuf *)0);
1852                 tp->snd_wl1 = th->th_seq - 1;
1853                 /* FALLTHROUGH */
1854
1855         /*
1856          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1857          * ACKs.  If the ack is in the range
1858          *      tp->snd_una < th->th_ack <= tp->snd_max
1859          * then advance tp->snd_una to th->th_ack and drop
1860          * data from the retransmission queue.  If this ACK reflects
1861          * more up to date window information we update our window information.
1862          */
1863         case TCPS_ESTABLISHED:
1864         case TCPS_FIN_WAIT_1:
1865         case TCPS_FIN_WAIT_2:
1866         case TCPS_CLOSE_WAIT:
1867         case TCPS_CLOSING:
1868         case TCPS_LAST_ACK:
1869         case TCPS_TIME_WAIT:
1870                 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1871                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1872                         tcpstat.tcps_rcvacktoomuch++;
1873                         goto dropafterack;
1874                 }
1875                 if (tp->sack_enable &&
1876                     (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1877                         tcp_sack_doack(tp, &to, th->th_ack);
1878                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1879                         if (tlen == 0 && tiwin == tp->snd_wnd) {
1880                                 tcpstat.tcps_rcvdupack++;
1881                                 /*
1882                                  * If we have outstanding data (other than
1883                                  * a window probe), this is a completely
1884                                  * duplicate ack (ie, window info didn't
1885                                  * change), the ack is the biggest we've
1886                                  * seen and we've seen exactly our rexmt
1887                                  * threshhold of them, assume a packet
1888                                  * has been dropped and retransmit it.
1889                                  * Kludge snd_nxt & the congestion
1890                                  * window so we send only this one
1891                                  * packet.
1892                                  *
1893                                  * We know we're losing at the current
1894                                  * window size so do congestion avoidance
1895                                  * (set ssthresh to half the current window
1896                                  * and pull our congestion window back to
1897                                  * the new ssthresh).
1898                                  *
1899                                  * Dup acks mean that packets have left the
1900                                  * network (they're now cached at the receiver)
1901                                  * so bump cwnd by the amount in the receiver
1902                                  * to keep a constant cwnd packets in the
1903                                  * network.
1904                                  */
1905                                 if (!callout_active(tp->tt_rexmt) ||
1906                                     th->th_ack != tp->snd_una)
1907                                         tp->t_dupacks = 0;
1908                                 else if (++tp->t_dupacks > tcprexmtthresh ||
1909                                          ((tcp_do_newreno || tp->sack_enable) &&
1910                                           IN_FASTRECOVERY(tp))) {
1911                                         if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1912                                                 int awnd;
1913
1914                                                 /*
1915                                                  * Compute the amount of data in flight first.
1916                                                  * We can inject new data into the pipe iff
1917                                                  * we have less than 1/2 the original window's
1918                                                  * worth of data in flight.
1919                                                  */
1920                                                 awnd = (tp->snd_nxt - tp->snd_fack) +
1921                                                         tp->sackhint.sack_bytes_rexmit;
1922                                                 if (awnd < tp->snd_ssthresh) {
1923                                                         tp->snd_cwnd += tp->t_maxseg;
1924                                                         if (tp->snd_cwnd > tp->snd_ssthresh)
1925                                                                 tp->snd_cwnd = tp->snd_ssthresh;
1926                                                 }
1927                                         } else
1928                                                 tp->snd_cwnd += tp->t_maxseg;
1929                                         (void) tcp_output(tp);
1930                                         goto drop;
1931                                 } else if (tp->t_dupacks == tcprexmtthresh) {
1932                                         tcp_seq onxt = tp->snd_nxt;
1933                                         u_int win;
1934
1935                                         /*
1936                                          * If we're doing sack, check to
1937                                          * see if we're already in sack
1938                                          * recovery. If we're not doing sack,
1939                                          * check to see if we're in newreno
1940                                          * recovery.
1941                                          */
1942                                         if (tp->sack_enable) {
1943                                                 if (IN_FASTRECOVERY(tp)) {
1944                                                         tp->t_dupacks = 0;
1945                                                         break;
1946                                                 }
1947                                         } else if (tcp_do_newreno) {
1948                                                 if (SEQ_LEQ(th->th_ack,
1949                                                     tp->snd_recover)) {
1950                                                         tp->t_dupacks = 0;
1951                                                         break;
1952                                                 }
1953                                         }
1954                                         win = min(tp->snd_wnd, tp->snd_cwnd) /
1955                                             2 / tp->t_maxseg;
1956                                         if (win < 2)
1957                                                 win = 2;
1958                                         tp->snd_ssthresh = win * tp->t_maxseg;
1959                                         ENTER_FASTRECOVERY(tp);
1960                                         tp->snd_recover = tp->snd_max;
1961                                         callout_stop(tp->tt_rexmt);
1962                                         tp->t_rtttime = 0;
1963                                         if (tp->sack_enable) {
1964                                                 tcpstat.tcps_sack_recovery_episode++;
1965                                                 tp->sack_newdata = tp->snd_nxt;
1966                                                 tp->snd_cwnd = tp->t_maxseg;
1967                                                 (void) tcp_output(tp);
1968                                                 goto drop;
1969                                         }
1970                                         tp->snd_nxt = th->th_ack;
1971                                         tp->snd_cwnd = tp->t_maxseg;
1972                                         (void) tcp_output(tp);
1973                                         KASSERT(tp->snd_limited <= 2,
1974                                             ("tp->snd_limited too big"));
1975                                         tp->snd_cwnd = tp->snd_ssthresh +
1976                                              tp->t_maxseg *
1977                                              (tp->t_dupacks - tp->snd_limited);
1978                                         if (SEQ_GT(onxt, tp->snd_nxt))
1979                                                 tp->snd_nxt = onxt;
1980                                         goto drop;
1981                                 } else if (tcp_do_rfc3042) {
1982                                         u_long oldcwnd = tp->snd_cwnd;
1983                                         tcp_seq oldsndmax = tp->snd_max;
1984                                         u_int sent;
1985
1986                                         KASSERT(tp->t_dupacks == 1 ||
1987                                             tp->t_dupacks == 2,
1988                                             ("dupacks not 1 or 2"));
1989                                         if (tp->t_dupacks == 1)
1990                                                 tp->snd_limited = 0;
1991                                         tp->snd_cwnd =
1992                                             (tp->snd_nxt - tp->snd_una) +
1993                                             (tp->t_dupacks - tp->snd_limited) *
1994                                             tp->t_maxseg;
1995                                         (void) tcp_output(tp);
1996                                         sent = tp->snd_max - oldsndmax;
1997                                         if (sent > tp->t_maxseg) {
1998                                                 KASSERT((tp->t_dupacks == 2 &&
1999                                                     tp->snd_limited == 0) ||
2000                                                    (sent == tp->t_maxseg + 1 &&
2001                                                     tp->t_flags & TF_SENTFIN),
2002                                                     ("sent too much"));
2003                                                 tp->snd_limited = 2;
2004                                         } else if (sent > 0)
2005                                                 ++tp->snd_limited;
2006                                         tp->snd_cwnd = oldcwnd;
2007                                         goto drop;
2008                                 }
2009                         } else
2010                                 tp->t_dupacks = 0;
2011                         break;
2012                 }
2013
2014                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2015
2016                 /*
2017                  * If the congestion window was inflated to account
2018                  * for the other side's cached packets, retract it.
2019                  */
2020                 if (tcp_do_newreno || tp->sack_enable) {
2021                         if (IN_FASTRECOVERY(tp)) {
2022                                 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2023                                         if (tp->sack_enable)
2024                                                 tcp_sack_partialack(tp, th);
2025                                         else
2026                                                 tcp_newreno_partial_ack(tp, th);
2027                                 } else {
2028                                         /*
2029                                          * Out of fast recovery.
2030                                          * Window inflation should have left us
2031                                          * with approximately snd_ssthresh
2032                                          * outstanding data.
2033                                          * But in case we would be inclined to
2034                                          * send a burst, better to do it via
2035                                          * the slow start mechanism.
2036                                          */
2037                                         if (SEQ_GT(th->th_ack +
2038                                                         tp->snd_ssthresh,
2039                                                    tp->snd_max))
2040                                                 tp->snd_cwnd = tp->snd_max -
2041                                                                 th->th_ack +
2042                                                                 tp->t_maxseg;
2043                                         else
2044                                                 tp->snd_cwnd = tp->snd_ssthresh;
2045                                 }
2046                         }
2047                 } else {
2048                         if (tp->t_dupacks >= tcprexmtthresh &&
2049                             tp->snd_cwnd > tp->snd_ssthresh)
2050                                 tp->snd_cwnd = tp->snd_ssthresh;
2051                 }
2052                 tp->t_dupacks = 0;
2053                 /*
2054                  * If we reach this point, ACK is not a duplicate,
2055                  *     i.e., it ACKs something we sent.
2056                  */
2057                 if (tp->t_flags & TF_NEEDSYN) {
2058                         /*
2059                          * T/TCP: Connection was half-synchronized, and our
2060                          * SYN has been ACK'd (so connection is now fully
2061                          * synchronized).  Go to non-starred state,
2062                          * increment snd_una for ACK of SYN, and check if
2063                          * we can do window scaling.
2064                          */
2065                         tp->t_flags &= ~TF_NEEDSYN;
2066                         tp->snd_una++;
2067                         /* Do window scaling? */
2068                         if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2069                                 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2070                                 tp->rcv_scale = tp->request_r_scale;
2071                                 /* Send window already scaled. */
2072                         }
2073                 }
2074
2075 process_ACK:
2076                 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2077                     "locked"));
2078                 INP_LOCK_ASSERT(inp);
2079
2080                 acked = th->th_ack - tp->snd_una;
2081                 tcpstat.tcps_rcvackpack++;
2082                 tcpstat.tcps_rcvackbyte += acked;
2083
2084                 /*
2085                  * If we just performed our first retransmit, and the ACK
2086                  * arrives within our recovery window, then it was a mistake
2087                  * to do the retransmit in the first place.  Recover our
2088                  * original cwnd and ssthresh, and proceed to transmit where
2089                  * we left off.
2090                  */
2091                 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2092                         ++tcpstat.tcps_sndrexmitbad;
2093                         tp->snd_cwnd = tp->snd_cwnd_prev;
2094                         tp->snd_ssthresh = tp->snd_ssthresh_prev;
2095                         tp->snd_recover = tp->snd_recover_prev;
2096                         if (tp->t_flags & TF_WASFRECOVERY)
2097                                 ENTER_FASTRECOVERY(tp);
2098                         tp->snd_nxt = tp->snd_max;
2099                         tp->t_badrxtwin = 0;    /* XXX probably not required */
2100                 }
2101
2102                 /*
2103                  * If we have a timestamp reply, update smoothed
2104                  * round trip time.  If no timestamp is present but
2105                  * transmit timer is running and timed sequence
2106                  * number was acked, update smoothed round trip time.
2107                  * Since we now have an rtt measurement, cancel the
2108                  * timer backoff (cf., Phil Karn's retransmit alg.).
2109                  * Recompute the initial retransmit timer.
2110                  *
2111                  * Some boxes send broken timestamp replies
2112                  * during the SYN+ACK phase, ignore
2113                  * timestamps of 0 or we could calculate a
2114                  * huge RTT and blow up the retransmit timer.
2115                  */
2116                 if ((to.to_flags & TOF_TS) != 0 &&
2117                     to.to_tsecr) {
2118                         if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2119                                 tp->t_rttlow = ticks - to.to_tsecr;
2120                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2121                 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2122                         if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2123                                 tp->t_rttlow = ticks - tp->t_rtttime;
2124                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2125                 }
2126                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2127
2128                 /*
2129                  * If all outstanding data is acked, stop retransmit
2130                  * timer and remember to restart (more output or persist).
2131                  * If there is more data to be acked, restart retransmit
2132                  * timer, using current (possibly backed-off) value.
2133                  */
2134                 if (th->th_ack == tp->snd_max) {
2135                         callout_stop(tp->tt_rexmt);
2136                         needoutput = 1;
2137                 } else if (!callout_active(tp->tt_persist))
2138                         callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2139                                       tcp_timer_rexmt, tp);
2140
2141                 /*
2142                  * If no data (only SYN) was ACK'd,
2143                  *    skip rest of ACK processing.
2144                  */
2145                 if (acked == 0)
2146                         goto step6;
2147
2148                 /*
2149                  * When new data is acked, open the congestion window.
2150                  * If the window gives us less than ssthresh packets
2151                  * in flight, open exponentially (maxseg per packet).
2152                  * Otherwise open linearly: maxseg per window
2153                  * (maxseg^2 / cwnd per packet).
2154                  */
2155                 if ((!tcp_do_newreno && !tp->sack_enable) ||
2156                     !IN_FASTRECOVERY(tp)) {
2157                         register u_int cw = tp->snd_cwnd;
2158                         register u_int incr = tp->t_maxseg;
2159                         if (cw > tp->snd_ssthresh)
2160                                 incr = incr * incr / cw;
2161                         tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2162                 }
2163                 SOCKBUF_LOCK(&so->so_snd);
2164                 if (acked > so->so_snd.sb_cc) {
2165                         tp->snd_wnd -= so->so_snd.sb_cc;
2166                         sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2167                         ourfinisacked = 1;
2168                 } else {
2169                         sbdrop_locked(&so->so_snd, acked);
2170                         tp->snd_wnd -= acked;
2171                         ourfinisacked = 0;
2172                 }
2173                 sowwakeup_locked(so);
2174                 /* detect una wraparound */
2175                 if ((tcp_do_newreno || tp->sack_enable) &&
2176                     !IN_FASTRECOVERY(tp) &&
2177                     SEQ_GT(tp->snd_una, tp->snd_recover) &&
2178                     SEQ_LEQ(th->th_ack, tp->snd_recover))
2179                         tp->snd_recover = th->th_ack - 1;
2180                 if ((tcp_do_newreno || tp->sack_enable) &&
2181                     IN_FASTRECOVERY(tp) &&
2182                     SEQ_GEQ(th->th_ack, tp->snd_recover))
2183                         EXIT_FASTRECOVERY(tp);
2184                 tp->snd_una = th->th_ack;
2185                 if (tp->sack_enable) {
2186                         if (SEQ_GT(tp->snd_una, tp->snd_recover))
2187                                 tp->snd_recover = tp->snd_una;
2188                 }
2189                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2190                         tp->snd_nxt = tp->snd_una;
2191
2192                 switch (tp->t_state) {
2193
2194                 /*
2195                  * In FIN_WAIT_1 STATE in addition to the processing
2196                  * for the ESTABLISHED state if our FIN is now acknowledged
2197                  * then enter FIN_WAIT_2.
2198                  */
2199                 case TCPS_FIN_WAIT_1:
2200                         if (ourfinisacked) {
2201                                 /*
2202                                  * If we can't receive any more
2203                                  * data, then closing user can proceed.
2204                                  * Starting the timer is contrary to the
2205                                  * specification, but if we don't get a FIN
2206                                  * we'll hang forever.
2207                                  */
2208                 /* XXXjl
2209                  * we should release the tp also, and use a
2210                  * compressed state.
2211                  */
2212                                 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2213                                         soisdisconnected(so);
2214                                         callout_reset(tp->tt_2msl, tcp_maxidle,
2215                                                       tcp_timer_2msl, tp);
2216                                 }
2217                                 tp->t_state = TCPS_FIN_WAIT_2;
2218                         }
2219                         break;
2220
2221                 /*
2222                  * In CLOSING STATE in addition to the processing for
2223                  * the ESTABLISHED state if the ACK acknowledges our FIN
2224                  * then enter the TIME-WAIT state, otherwise ignore
2225                  * the segment.
2226                  */
2227                 case TCPS_CLOSING:
2228                         if (ourfinisacked) {
2229                                 KASSERT(headlocked, ("tcp_input: process_ACK: "
2230                                     "head not locked"));
2231                                 tcp_twstart(tp);
2232                                 INP_INFO_WUNLOCK(&tcbinfo);
2233                                 m_freem(m);
2234                                 return;
2235                         }
2236                         break;
2237
2238                 /*
2239                  * In LAST_ACK, we may still be waiting for data to drain
2240                  * and/or to be acked, as well as for the ack of our FIN.
2241                  * If our FIN is now acknowledged, delete the TCB,
2242                  * enter the closed state and return.
2243                  */
2244                 case TCPS_LAST_ACK:
2245                         if (ourfinisacked) {
2246                                 KASSERT(headlocked, ("tcp_input: process_ACK:"
2247                                     " tcp_close: head not locked"));
2248                                 tp = tcp_close(tp);
2249                                 goto drop;
2250                         }
2251                         break;
2252
2253                 /*
2254                  * In TIME_WAIT state the only thing that should arrive
2255                  * is a retransmission of the remote FIN.  Acknowledge
2256                  * it and restart the finack timer.
2257                  */
2258                 case TCPS_TIME_WAIT:
2259                         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2260                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2261                                       tcp_timer_2msl, tp);
2262                         goto dropafterack;
2263                 }
2264         }
2265
2266 step6:
2267         KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2268         INP_LOCK_ASSERT(inp);
2269
2270         /*
2271          * Update window information.
2272          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2273          */
2274         if ((thflags & TH_ACK) &&
2275             (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2276             (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2277              (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2278                 /* keep track of pure window updates */
2279                 if (tlen == 0 &&
2280                     tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2281                         tcpstat.tcps_rcvwinupd++;
2282                 tp->snd_wnd = tiwin;
2283                 tp->snd_wl1 = th->th_seq;
2284                 tp->snd_wl2 = th->th_ack;
2285                 if (tp->snd_wnd > tp->max_sndwnd)
2286                         tp->max_sndwnd = tp->snd_wnd;
2287                 needoutput = 1;
2288         }
2289
2290         /*
2291          * Process segments with URG.
2292          */
2293         if ((thflags & TH_URG) && th->th_urp &&
2294             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2295                 /*
2296                  * This is a kludge, but if we receive and accept
2297                  * random urgent pointers, we'll crash in
2298                  * soreceive.  It's hard to imagine someone
2299                  * actually wanting to send this much urgent data.
2300                  */
2301                 SOCKBUF_LOCK(&so->so_rcv);
2302                 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2303                         th->th_urp = 0;                 /* XXX */
2304                         thflags &= ~TH_URG;             /* XXX */
2305                         SOCKBUF_UNLOCK(&so->so_rcv);    /* XXX */
2306                         goto dodata;                    /* XXX */
2307                 }
2308                 /*
2309                  * If this segment advances the known urgent pointer,
2310                  * then mark the data stream.  This should not happen
2311                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2312                  * a FIN has been received from the remote side.
2313                  * In these states we ignore the URG.
2314                  *
2315                  * According to RFC961 (Assigned Protocols),
2316                  * the urgent pointer points to the last octet
2317                  * of urgent data.  We continue, however,
2318                  * to consider it to indicate the first octet
2319                  * of data past the urgent section as the original
2320                  * spec states (in one of two places).
2321                  */
2322                 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2323                         tp->rcv_up = th->th_seq + th->th_urp;
2324                         so->so_oobmark = so->so_rcv.sb_cc +
2325                             (tp->rcv_up - tp->rcv_nxt) - 1;
2326                         if (so->so_oobmark == 0)
2327                                 so->so_rcv.sb_state |= SBS_RCVATMARK;
2328                         sohasoutofband(so);
2329                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2330                 }
2331                 SOCKBUF_UNLOCK(&so->so_rcv);
2332                 /*
2333                  * Remove out of band data so doesn't get presented to user.
2334                  * This can happen independent of advancing the URG pointer,
2335                  * but if two URG's are pending at once, some out-of-band
2336                  * data may creep in... ick.
2337                  */
2338                 if (th->th_urp <= (u_long)tlen &&
2339                     !(so->so_options & SO_OOBINLINE)) {
2340                         /* hdr drop is delayed */
2341                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2342                 }
2343         } else {
2344                 /*
2345                  * If no out of band data is expected,
2346                  * pull receive urgent pointer along
2347                  * with the receive window.
2348                  */
2349                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2350                         tp->rcv_up = tp->rcv_nxt;
2351         }
2352 dodata:                                                 /* XXX */
2353         KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2354         INP_LOCK_ASSERT(inp);
2355
2356         /*
2357          * Process the segment text, merging it into the TCP sequencing queue,
2358          * and arranging for acknowledgment of receipt if necessary.
2359          * This process logically involves adjusting tp->rcv_wnd as data
2360          * is presented to the user (this happens in tcp_usrreq.c,
2361          * case PRU_RCVD).  If a FIN has already been received on this
2362          * connection then we just ignore the text.
2363          */
2364         if ((tlen || (thflags & TH_FIN)) &&
2365             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2366                 tcp_seq save_start = th->th_seq;
2367                 tcp_seq save_end = th->th_seq + tlen;
2368                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2369                 /*
2370                  * Insert segment which includes th into TCP reassembly queue
2371                  * with control block tp.  Set thflags to whether reassembly now
2372                  * includes a segment with FIN.  This handles the common case
2373                  * inline (segment is the next to be received on an established
2374                  * connection, and the queue is empty), avoiding linkage into
2375                  * and removal from the queue and repetition of various
2376                  * conversions.
2377                  * Set DELACK for segments received in order, but ack
2378                  * immediately when segments are out of order (so
2379                  * fast retransmit can work).
2380                  */
2381                 if (th->th_seq == tp->rcv_nxt &&
2382                     LIST_EMPTY(&tp->t_segq) &&
2383                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2384                         if (DELAY_ACK(tp))
2385                                 tp->t_flags |= TF_DELACK;
2386                         else
2387                                 tp->t_flags |= TF_ACKNOW;
2388                         tp->rcv_nxt += tlen;
2389                         thflags = th->th_flags & TH_FIN;
2390                         tcpstat.tcps_rcvpack++;
2391                         tcpstat.tcps_rcvbyte += tlen;
2392                         ND6_HINT(tp);
2393                         SOCKBUF_LOCK(&so->so_rcv);
2394                         if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2395                                 m_freem(m);
2396                         else
2397                                 sbappendstream_locked(&so->so_rcv, m);
2398                         sorwakeup_locked(so);
2399                 } else {
2400                         thflags = tcp_reass(tp, th, &tlen, m);
2401                         tp->t_flags |= TF_ACKNOW;
2402                 }
2403                 if (tlen > 0 && tp->sack_enable)
2404                         tcp_update_sack_list(tp, save_start, save_end);
2405                 /*
2406                  * Note the amount of data that peer has sent into
2407                  * our window, in order to estimate the sender's
2408                  * buffer size.
2409                  */
2410                 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2411         } else {
2412                 m_freem(m);
2413                 thflags &= ~TH_FIN;
2414         }
2415
2416         /*
2417          * If FIN is received ACK the FIN and let the user know
2418          * that the connection is closing.
2419          */
2420         if (thflags & TH_FIN) {
2421                 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2422                         socantrcvmore(so);
2423                         /*
2424                          * If connection is half-synchronized
2425                          * (ie NEEDSYN flag on) then delay ACK,
2426                          * so it may be piggybacked when SYN is sent.
2427                          * Otherwise, since we received a FIN then no
2428                          * more input can be expected, send ACK now.
2429                          */
2430                         if (tp->t_flags & TF_NEEDSYN)
2431                                 tp->t_flags |= TF_DELACK;
2432                         else
2433                                 tp->t_flags |= TF_ACKNOW;
2434                         tp->rcv_nxt++;
2435                 }
2436                 switch (tp->t_state) {
2437
2438                 /*
2439                  * In SYN_RECEIVED and ESTABLISHED STATES
2440                  * enter the CLOSE_WAIT state.
2441                  */
2442                 case TCPS_SYN_RECEIVED:
2443                         tp->t_starttime = ticks;
2444                         /*FALLTHROUGH*/
2445                 case TCPS_ESTABLISHED:
2446                         tp->t_state = TCPS_CLOSE_WAIT;
2447                         break;
2448
2449                 /*
2450                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2451                  * enter the CLOSING state.
2452                  */
2453                 case TCPS_FIN_WAIT_1:
2454                         tp->t_state = TCPS_CLOSING;
2455                         break;
2456
2457                 /*
2458                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2459                  * starting the time-wait timer, turning off the other
2460                  * standard timers.
2461                  */
2462                 case TCPS_FIN_WAIT_2:
2463                         KASSERT(headlocked == 1, ("tcp_input: dodata: "
2464                             "TCP_FIN_WAIT_2: head not locked"));
2465                         tcp_twstart(tp);
2466                         INP_INFO_WUNLOCK(&tcbinfo);
2467                         return;
2468
2469                 /*
2470                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2471                  */
2472                 case TCPS_TIME_WAIT:
2473                         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2474                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2475                                       tcp_timer_2msl, tp);
2476                         break;
2477                 }
2478         }
2479         INP_INFO_WUNLOCK(&tcbinfo);
2480         headlocked = 0;
2481 #ifdef TCPDEBUG
2482         if (so->so_options & SO_DEBUG)
2483                 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2484                           &tcp_savetcp, 0);
2485 #endif
2486
2487         /*
2488          * Return any desired output.
2489          */
2490         if (needoutput || (tp->t_flags & TF_ACKNOW))
2491                 (void) tcp_output(tp);
2492
2493 check_delack:
2494         KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2495         INP_LOCK_ASSERT(inp);
2496         if (tp->t_flags & TF_DELACK) {
2497                 tp->t_flags &= ~TF_DELACK;
2498                 callout_reset(tp->tt_delack, tcp_delacktime,
2499                     tcp_timer_delack, tp);
2500         }
2501         INP_UNLOCK(inp);
2502         return;
2503
2504 dropafterack:
2505         KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2506         /*
2507          * Generate an ACK dropping incoming segment if it occupies
2508          * sequence space, where the ACK reflects our state.
2509          *
2510          * We can now skip the test for the RST flag since all
2511          * paths to this code happen after packets containing
2512          * RST have been dropped.
2513          *
2514          * In the SYN-RECEIVED state, don't send an ACK unless the
2515          * segment we received passes the SYN-RECEIVED ACK test.
2516          * If it fails send a RST.  This breaks the loop in the
2517          * "LAND" DoS attack, and also prevents an ACK storm
2518          * between two listening ports that have been sent forged
2519          * SYN segments, each with the source address of the other.
2520          */
2521         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2522             (SEQ_GT(tp->snd_una, th->th_ack) ||
2523              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2524                 rstreason = BANDLIM_RST_OPENPORT;
2525                 goto dropwithreset;
2526         }
2527 #ifdef TCPDEBUG
2528         if (so->so_options & SO_DEBUG)
2529                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2530                           &tcp_savetcp, 0);
2531 #endif
2532         KASSERT(headlocked, ("headlocked should be 1"));
2533         INP_INFO_WUNLOCK(&tcbinfo);
2534         tp->t_flags |= TF_ACKNOW;
2535         (void) tcp_output(tp);
2536         INP_UNLOCK(inp);
2537         m_freem(m);
2538         return;
2539
2540 dropwithreset:
2541         KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2542         /*
2543          * Generate a RST, dropping incoming segment.
2544          * Make ACK acceptable to originator of segment.
2545          * Don't bother to respond if destination was broadcast/multicast.
2546          */
2547         if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2548                 goto drop;
2549         if (isipv6) {
2550                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2551                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2552                         goto drop;
2553         } else {
2554                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2555                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2556                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2557                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2558                         goto drop;
2559         }
2560         /* IPv6 anycast check is done at tcp6_input() */
2561
2562         /*
2563          * Perform bandwidth limiting.
2564          */
2565         if (badport_bandlim(rstreason) < 0)
2566                 goto drop;
2567
2568 #ifdef TCPDEBUG
2569         if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2570                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2571                           &tcp_savetcp, 0);
2572 #endif
2573
2574         if (thflags & TH_ACK)
2575                 /* mtod() below is safe as long as hdr dropping is delayed */
2576                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2577                             TH_RST);
2578         else {
2579                 if (thflags & TH_SYN)
2580                         tlen++;
2581                 /* mtod() below is safe as long as hdr dropping is delayed */
2582                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2583                             (tcp_seq)0, TH_RST|TH_ACK);
2584         }
2585
2586         if (tp != NULL)
2587                 INP_UNLOCK(inp);
2588         if (headlocked)
2589                 INP_INFO_WUNLOCK(&tcbinfo);
2590         return;
2591
2592 drop:
2593         /*
2594          * Drop space held by incoming segment and return.
2595          */
2596 #ifdef TCPDEBUG
2597         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2598                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2599                           &tcp_savetcp, 0);
2600 #endif
2601         if (tp != NULL)
2602                 INP_UNLOCK(inp);
2603         if (headlocked)
2604                 INP_INFO_WUNLOCK(&tcbinfo);
2605         m_freem(m);
2606         return;
2607 }
2608
2609 /*
2610  * Parse TCP options and place in tcpopt.
2611  */
2612 static void
2613 tcp_dooptions(to, cp, cnt, flags)
2614         struct tcpopt *to;
2615         u_char *cp;
2616         int cnt;
2617         int flags;
2618 {
2619         int opt, optlen;
2620
2621         to->to_flags = 0;
2622         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2623                 opt = cp[0];
2624                 if (opt == TCPOPT_EOL)
2625                         break;
2626                 if (opt == TCPOPT_NOP)
2627                         optlen = 1;
2628                 else {
2629                         if (cnt < 2)
2630                                 break;
2631                         optlen = cp[1];
2632                         if (optlen < 2 || optlen > cnt)
2633                                 break;
2634                 }
2635                 switch (opt) {
2636                 case TCPOPT_MAXSEG:
2637                         if (optlen != TCPOLEN_MAXSEG)
2638                                 continue;
2639                         if (!(flags & TO_SYN))
2640                                 continue;
2641                         to->to_flags |= TOF_MSS;
2642                         bcopy((char *)cp + 2,
2643                             (char *)&to->to_mss, sizeof(to->to_mss));
2644                         to->to_mss = ntohs(to->to_mss);
2645                         break;
2646                 case TCPOPT_WINDOW:
2647                         if (optlen != TCPOLEN_WINDOW)
2648                                 continue;
2649                         if (!(flags & TO_SYN))
2650                                 continue;
2651                         to->to_flags |= TOF_SCALE;
2652                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2653                         break;
2654                 case TCPOPT_TIMESTAMP:
2655                         if (optlen != TCPOLEN_TIMESTAMP)
2656                                 continue;
2657                         to->to_flags |= TOF_TS;
2658                         bcopy((char *)cp + 2,
2659                             (char *)&to->to_tsval, sizeof(to->to_tsval));
2660                         to->to_tsval = ntohl(to->to_tsval);
2661                         bcopy((char *)cp + 6,
2662                             (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2663                         to->to_tsecr = ntohl(to->to_tsecr);
2664                         break;
2665 #ifdef TCP_SIGNATURE
2666                 /*
2667                  * XXX In order to reply to a host which has set the
2668                  * TCP_SIGNATURE option in its initial SYN, we have to
2669                  * record the fact that the option was observed here
2670                  * for the syncache code to perform the correct response.
2671                  */
2672                 case TCPOPT_SIGNATURE:
2673                         if (optlen != TCPOLEN_SIGNATURE)
2674                                 continue;
2675                         to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2676                         break;
2677 #endif
2678                 case TCPOPT_SACK_PERMITTED:
2679                         if (optlen != TCPOLEN_SACK_PERMITTED)
2680                                 continue;
2681                         if (!(flags & TO_SYN))
2682                                 continue;
2683                         if (!tcp_do_sack)
2684                                 continue;
2685                         to->to_flags |= TOF_SACK;
2686                         break;
2687                 case TCPOPT_SACK:
2688                         if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2689                                 continue;
2690                         to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2691                         to->to_sacks = cp + 2;
2692                         tcpstat.tcps_sack_rcv_blocks++;
2693                         break;
2694                 default:
2695                         continue;
2696                 }
2697         }
2698 }
2699
2700 /*
2701  * Pull out of band byte out of a segment so
2702  * it doesn't appear in the user's data queue.
2703  * It is still reflected in the segment length for
2704  * sequencing purposes.
2705  */
2706 static void
2707 tcp_pulloutofband(so, th, m, off)
2708         struct socket *so;
2709         struct tcphdr *th;
2710         register struct mbuf *m;
2711         int off;                /* delayed to be droped hdrlen */
2712 {
2713         int cnt = off + th->th_urp - 1;
2714
2715         while (cnt >= 0) {
2716                 if (m->m_len > cnt) {
2717                         char *cp = mtod(m, caddr_t) + cnt;
2718                         struct tcpcb *tp = sototcpcb(so);
2719
2720                         tp->t_iobc = *cp;
2721                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2722                         bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2723                         m->m_len--;
2724                         if (m->m_flags & M_PKTHDR)
2725                                 m->m_pkthdr.len--;
2726                         return;
2727                 }
2728                 cnt -= m->m_len;
2729                 m = m->m_next;
2730                 if (m == 0)
2731                         break;
2732         }
2733         panic("tcp_pulloutofband");
2734 }
2735
2736 /*
2737  * Collect new round-trip time estimate
2738  * and update averages and current timeout.
2739  */
2740 static void
2741 tcp_xmit_timer(tp, rtt)
2742         register struct tcpcb *tp;
2743         int rtt;
2744 {
2745         register int delta;
2746
2747         INP_LOCK_ASSERT(tp->t_inpcb);
2748
2749         tcpstat.tcps_rttupdated++;
2750         tp->t_rttupdated++;
2751         if (tp->t_srtt != 0) {
2752                 /*
2753                  * srtt is stored as fixed point with 5 bits after the
2754                  * binary point (i.e., scaled by 8).  The following magic
2755                  * is equivalent to the smoothing algorithm in rfc793 with
2756                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2757                  * point).  Adjust rtt to origin 0.
2758                  */
2759                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2760                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2761
2762                 if ((tp->t_srtt += delta) <= 0)
2763                         tp->t_srtt = 1;
2764
2765                 /*
2766                  * We accumulate a smoothed rtt variance (actually, a
2767                  * smoothed mean difference), then set the retransmit
2768                  * timer to smoothed rtt + 4 times the smoothed variance.
2769                  * rttvar is stored as fixed point with 4 bits after the
2770                  * binary point (scaled by 16).  The following is
2771                  * equivalent to rfc793 smoothing with an alpha of .75
2772                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2773                  * rfc793's wired-in beta.
2774                  */
2775                 if (delta < 0)
2776                         delta = -delta;
2777                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2778                 if ((tp->t_rttvar += delta) <= 0)
2779                         tp->t_rttvar = 1;
2780                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2781                     tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2782         } else {
2783                 /*
2784                  * No rtt measurement yet - use the unsmoothed rtt.
2785                  * Set the variance to half the rtt (so our first
2786                  * retransmit happens at 3*rtt).
2787                  */
2788                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2789                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2790                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2791         }
2792         tp->t_rtttime = 0;
2793         tp->t_rxtshift = 0;
2794
2795         /*
2796          * the retransmit should happen at rtt + 4 * rttvar.
2797          * Because of the way we do the smoothing, srtt and rttvar
2798          * will each average +1/2 tick of bias.  When we compute
2799          * the retransmit timer, we want 1/2 tick of rounding and
2800          * 1 extra tick because of +-1/2 tick uncertainty in the
2801          * firing of the timer.  The bias will give us exactly the
2802          * 1.5 tick we need.  But, because the bias is
2803          * statistical, we have to test that we don't drop below
2804          * the minimum feasible timer (which is 2 ticks).
2805          */
2806         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2807                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2808
2809         /*
2810          * We received an ack for a packet that wasn't retransmitted;
2811          * it is probably safe to discard any error indications we've
2812          * received recently.  This isn't quite right, but close enough
2813          * for now (a route might have failed after we sent a segment,
2814          * and the return path might not be symmetrical).
2815          */
2816         tp->t_softerror = 0;
2817 }
2818
2819 /*
2820  * Determine a reasonable value for maxseg size.
2821  * If the route is known, check route for mtu.
2822  * If none, use an mss that can be handled on the outgoing
2823  * interface without forcing IP to fragment; if bigger than
2824  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2825  * to utilize large mbufs.  If no route is found, route has no mtu,
2826  * or the destination isn't local, use a default, hopefully conservative
2827  * size (usually 512 or the default IP max size, but no more than the mtu
2828  * of the interface), as we can't discover anything about intervening
2829  * gateways or networks.  We also initialize the congestion/slow start
2830  * window to be a single segment if the destination isn't local.
2831  * While looking at the routing entry, we also initialize other path-dependent
2832  * parameters from pre-set or cached values in the routing entry.
2833  *
2834  * Also take into account the space needed for options that we
2835  * send regularly.  Make maxseg shorter by that amount to assure
2836  * that we can send maxseg amount of data even when the options
2837  * are present.  Store the upper limit of the length of options plus
2838  * data in maxopd.
2839  *
2840  *
2841  * In case of T/TCP, we call this routine during implicit connection
2842  * setup as well (offer = -1), to initialize maxseg from the cached
2843  * MSS of our peer.
2844  *
2845  * NOTE that this routine is only called when we process an incoming
2846  * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2847  */
2848 void
2849 tcp_mss(tp, offer)
2850         struct tcpcb *tp;
2851         int offer;
2852 {
2853         int rtt, mss;
2854         u_long bufsize;
2855         u_long maxmtu;
2856         struct inpcb *inp = tp->t_inpcb;
2857         struct socket *so;
2858         struct hc_metrics_lite metrics;
2859         int origoffer = offer;
2860 #ifdef INET6
2861         int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2862         size_t min_protoh = isipv6 ?
2863                             sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2864                             sizeof (struct tcpiphdr);
2865 #else
2866         const size_t min_protoh = sizeof(struct tcpiphdr);
2867 #endif
2868
2869         /* initialize */
2870 #ifdef INET6
2871         if (isipv6) {
2872                 maxmtu = tcp_maxmtu6(&inp->inp_inc);
2873                 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2874         } else
2875 #endif
2876         {
2877                 maxmtu = tcp_maxmtu(&inp->inp_inc);
2878                 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2879         }
2880         so = inp->inp_socket;
2881
2882         /*
2883          * no route to sender, stay with default mss and return
2884          */
2885         if (maxmtu == 0)
2886                 return;
2887
2888         /* what have we got? */
2889         switch (offer) {
2890                 case 0:
2891                         /*
2892                          * Offer == 0 means that there was no MSS on the SYN
2893                          * segment, in this case we use tcp_mssdflt.
2894                          */
2895                         offer =
2896 #ifdef INET6
2897                                 isipv6 ? tcp_v6mssdflt :
2898 #endif
2899                                 tcp_mssdflt;
2900                         break;
2901
2902                 case -1:
2903                         /*
2904                          * Offer == -1 means that we didn't receive SYN yet.
2905                          */
2906                         /* FALLTHROUGH */
2907
2908                 default:
2909                         /*
2910                          * Prevent DoS attack with too small MSS. Round up
2911                          * to at least minmss.
2912                          */
2913                         offer = max(offer, tcp_minmss);
2914                         /*
2915                          * Sanity check: make sure that maxopd will be large
2916                          * enough to allow some data on segments even if the
2917                          * all the option space is used (40bytes).  Otherwise
2918                          * funny things may happen in tcp_output.
2919                          */
2920                         offer = max(offer, 64);
2921         }
2922
2923         /*
2924          * rmx information is now retrieved from tcp_hostcache
2925          */
2926         tcp_hc_get(&inp->inp_inc, &metrics);
2927
2928         /*
2929          * if there's a discovered mtu int tcp hostcache, use it
2930          * else, use the link mtu.
2931          */
2932         if (metrics.rmx_mtu)
2933                 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2934         else {
2935 #ifdef INET6
2936                 if (isipv6) {
2937                         mss = maxmtu - min_protoh;
2938                         if (!path_mtu_discovery &&
2939                             !in6_localaddr(&inp->in6p_faddr))
2940                                 mss = min(mss, tcp_v6mssdflt);
2941                 } else
2942 #endif
2943                 {
2944                         mss = maxmtu - min_protoh;
2945                         if (!path_mtu_discovery &&
2946                             !in_localaddr(inp->inp_faddr))
2947                                 mss = min(mss, tcp_mssdflt);
2948                 }
2949         }
2950         mss = min(mss, offer);
2951
2952         /*
2953          * maxopd stores the maximum length of data AND options
2954          * in a segment; maxseg is the amount of data in a normal
2955          * segment.  We need to store this value (maxopd) apart
2956          * from maxseg, because now every segment carries options
2957          * and thus we normally have somewhat less data in segments.
2958          */
2959         tp->t_maxopd = mss;
2960
2961         /*
2962          * origoffer==-1 indicates, that no segments were received yet.
2963          * In this case we just guess.
2964          */
2965         if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2966             (origoffer == -1 ||
2967              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2968                 mss -= TCPOLEN_TSTAMP_APPA;
2969         tp->t_maxseg = mss;
2970
2971 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
2972                 if (mss > MCLBYTES)
2973                         mss &= ~(MCLBYTES-1);
2974 #else
2975                 if (mss > MCLBYTES)
2976                         mss = mss / MCLBYTES * MCLBYTES;
2977 #endif
2978         tp->t_maxseg = mss;
2979
2980         /*
2981          * If there's a pipesize, change the socket buffer to that size,
2982          * don't change if sb_hiwat is different than default (then it
2983          * has been changed on purpose with setsockopt).
2984          * Make the socket buffers an integral number of mss units;
2985          * if the mss is larger than the socket buffer, decrease the mss.
2986          */
2987         SOCKBUF_LOCK(&so->so_snd);
2988         if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2989                 bufsize = metrics.rmx_sendpipe;
2990         else
2991                 bufsize = so->so_snd.sb_hiwat;
2992         if (bufsize < mss)
2993                 mss = bufsize;
2994         else {
2995                 bufsize = roundup(bufsize, mss);
2996                 if (bufsize > sb_max)
2997                         bufsize = sb_max;
2998                 if (bufsize > so->so_snd.sb_hiwat)
2999                         (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3000         }
3001         SOCKBUF_UNLOCK(&so->so_snd);
3002         tp->t_maxseg = mss;
3003
3004         SOCKBUF_LOCK(&so->so_rcv);
3005         if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3006                 bufsize = metrics.rmx_recvpipe;
3007         else
3008                 bufsize = so->so_rcv.sb_hiwat;
3009         if (bufsize > mss) {
3010                 bufsize = roundup(bufsize, mss);
3011                 if (bufsize > sb_max)
3012                         bufsize = sb_max;
3013                 if (bufsize > so->so_rcv.sb_hiwat)
3014                         (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3015         }
3016         SOCKBUF_UNLOCK(&so->so_rcv);
3017         /*
3018          * While we're here, check the others too
3019          */
3020         if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3021                 tp->t_srtt = rtt;
3022                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3023                 tcpstat.tcps_usedrtt++;
3024                 if (metrics.rmx_rttvar) {
3025                         tp->t_rttvar = metrics.rmx_rttvar;
3026                         tcpstat.tcps_usedrttvar++;
3027                 } else {
3028                         /* default variation is +- 1 rtt */
3029                         tp->t_rttvar =
3030                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3031                 }
3032                 TCPT_RANGESET(tp->t_rxtcur,
3033                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3034                               tp->t_rttmin, TCPTV_REXMTMAX);
3035         }
3036         if (metrics.rmx_ssthresh) {
3037                 /*
3038                  * There's some sort of gateway or interface
3039                  * buffer limit on the path.  Use this to set
3040                  * the slow start threshhold, but set the
3041                  * threshold to no less than 2*mss.
3042                  */
3043                 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3044                 tcpstat.tcps_usedssthresh++;
3045         }
3046         if (metrics.rmx_bandwidth)
3047                 tp->snd_bandwidth = metrics.rmx_bandwidth;
3048
3049         /*
3050          * Set the slow-start flight size depending on whether this
3051          * is a local network or not.
3052          *
3053          * Extend this so we cache the cwnd too and retrieve it here.
3054          * Make cwnd even bigger than RFC3390 suggests but only if we
3055          * have previous experience with the remote host. Be careful
3056          * not make cwnd bigger than remote receive window or our own
3057          * send socket buffer. Maybe put some additional upper bound
3058          * on the retrieved cwnd. Should do incremental updates to
3059          * hostcache when cwnd collapses so next connection doesn't
3060          * overloads the path again.
3061          *
3062          * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3063          * We currently check only in syncache_socket for that.
3064          */
3065 #define TCP_METRICS_CWND
3066 #ifdef TCP_METRICS_CWND
3067         if (metrics.rmx_cwnd)
3068                 tp->snd_cwnd = max(mss,
3069                                 min(metrics.rmx_cwnd / 2,
3070                                  min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3071         else
3072 #endif
3073         if (tcp_do_rfc3390)
3074                 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3075 #ifdef INET6
3076         else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3077                  (!isipv6 && in_localaddr(inp->inp_faddr)))
3078 #else
3079         else if (in_localaddr(inp->inp_faddr))
3080 #endif
3081                 tp->snd_cwnd = mss * ss_fltsz_local;
3082         else
3083                 tp->snd_cwnd = mss * ss_fltsz;
3084 }
3085
3086 /*
3087  * Determine the MSS option to send on an outgoing SYN.
3088  */
3089 int
3090 tcp_mssopt(inc)
3091         struct in_conninfo *inc;
3092 {
3093         int mss = 0;
3094         u_long maxmtu = 0;
3095         u_long thcmtu = 0;
3096         size_t min_protoh;
3097 #ifdef INET6
3098         int isipv6 = inc->inc_isipv6 ? 1 : 0;
3099 #endif
3100
3101         KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3102
3103 #ifdef INET6
3104         if (isipv6) {
3105                 mss = tcp_v6mssdflt;
3106                 maxmtu = tcp_maxmtu6(inc);
3107                 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3108                 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3109         } else
3110 #endif
3111         {
3112                 mss = tcp_mssdflt;
3113                 maxmtu = tcp_maxmtu(inc);
3114                 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3115                 min_protoh = sizeof(struct tcpiphdr);
3116         }
3117         if (maxmtu && thcmtu)
3118                 mss = min(maxmtu, thcmtu) - min_protoh;
3119         else if (maxmtu || thcmtu)
3120                 mss = max(maxmtu, thcmtu) - min_protoh;
3121
3122         return (mss);
3123 }
3124
3125
3126 /*
3127  * On a partial ack arrives, force the retransmission of the
3128  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3129  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3130  * be started again.
3131  */
3132 static void
3133 tcp_newreno_partial_ack(tp, th)
3134         struct tcpcb *tp;
3135         struct tcphdr *th;
3136 {
3137         tcp_seq onxt = tp->snd_nxt;
3138         u_long  ocwnd = tp->snd_cwnd;
3139
3140         callout_stop(tp->tt_rexmt);
3141         tp->t_rtttime = 0;
3142         tp->snd_nxt = th->th_ack;
3143         /*
3144          * Set snd_cwnd to one segment beyond acknowledged offset.
3145          * (tp->snd_una has not yet been updated when this function is called.)
3146          */
3147         tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3148         tp->t_flags |= TF_ACKNOW;
3149         (void) tcp_output(tp);
3150         tp->snd_cwnd = ocwnd;
3151         if (SEQ_GT(onxt, tp->snd_nxt))
3152                 tp->snd_nxt = onxt;
3153         /*
3154          * Partial window deflation.  Relies on fact that tp->snd_una
3155          * not updated yet.
3156          */
3157         if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3158                 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3159         else
3160                 tp->snd_cwnd = 0;
3161         tp->snd_cwnd += tp->t_maxseg;
3162 }
3163
3164 /*
3165  * Returns 1 if the TIME_WAIT state was killed and we should start over,
3166  * looking for a pcb in the listen state.  Returns 0 otherwise.
3167  */
3168 static int
3169 tcp_timewait(inp, to, th, m, tlen)
3170         struct inpcb *inp;
3171         struct tcpopt *to;
3172         struct tcphdr *th;
3173         struct mbuf *m;
3174         int tlen;
3175 {
3176         struct tcptw *tw;
3177         int thflags;
3178         tcp_seq seq;
3179 #ifdef INET6
3180         int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3181 #else
3182         const int isipv6 = 0;
3183 #endif
3184
3185         /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */
3186         INP_INFO_WLOCK_ASSERT(&tcbinfo);
3187         INP_LOCK_ASSERT(inp);
3188
3189         /*
3190          * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
3191          * still present.  This is undesirable, but temporarily necessary
3192          * until we work out how to handle inpcb's who's timewait state has
3193          * been removed.
3194          */
3195         tw = intotw(inp);
3196         if (tw == NULL)
3197                 goto drop;
3198
3199         thflags = th->th_flags;
3200
3201         /*
3202          * NOTE: for FIN_WAIT_2 (to be added later),
3203          * must validate sequence number before accepting RST
3204          */
3205
3206         /*
3207          * If the segment contains RST:
3208          *      Drop the segment - see Stevens, vol. 2, p. 964 and
3209          *      RFC 1337.
3210          */
3211         if (thflags & TH_RST)
3212                 goto drop;
3213
3214 #if 0
3215 /* PAWS not needed at the moment */
3216         /*
3217          * RFC 1323 PAWS: If we have a timestamp reply on this segment
3218          * and it's less than ts_recent, drop it.
3219          */
3220         if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3221             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3222                 if ((thflags & TH_ACK) == 0)
3223                         goto drop;
3224                 goto ack;
3225         }
3226         /*
3227          * ts_recent is never updated because we never accept new segments.
3228          */
3229 #endif
3230
3231         /*
3232          * If a new connection request is received
3233          * while in TIME_WAIT, drop the old connection
3234          * and start over if the sequence numbers
3235          * are above the previous ones.
3236          */
3237         if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3238                 tcp_twclose(tw, 0);
3239                 return (1);
3240         }
3241
3242         /*
3243          * Drop the the segment if it does not contain an ACK.
3244          */
3245         if ((thflags & TH_ACK) == 0)
3246                 goto drop;
3247
3248         /*
3249          * Reset the 2MSL timer if this is a duplicate FIN.
3250          */
3251         if (thflags & TH_FIN) {
3252                 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3253                 if (seq + 1 == tw->rcv_nxt)
3254                         tcp_timer_2msl_reset(tw, 2 * tcp_msl, 1);
3255         }
3256
3257         /*
3258          * Acknowledge the segment if it has data or is not a duplicate ACK.
3259          */
3260         if (thflags != TH_ACK || tlen != 0 ||
3261             th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3262                 tcp_twrespond(tw, TH_ACK);
3263         goto drop;
3264
3265         /*
3266          * Generate a RST, dropping incoming segment.
3267          * Make ACK acceptable to originator of segment.
3268          * Don't bother to respond if destination was broadcast/multicast.
3269          */
3270         if (m->m_flags & (M_BCAST|M_MCAST))
3271                 goto drop;
3272         if (isipv6) {
3273                 struct ip6_hdr *ip6;
3274
3275                 /* IPv6 anycast check is done at tcp6_input() */
3276                 ip6 = mtod(m, struct ip6_hdr *);
3277                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3278                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3279                         goto drop;
3280         } else {
3281                 struct ip *ip;
3282
3283                 ip = mtod(m, struct ip *);
3284                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3285                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3286                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3287                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3288                         goto drop;
3289         }
3290         if (thflags & TH_ACK) {
3291                 tcp_respond(NULL,
3292                     mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3293         } else {
3294                 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3295                 tcp_respond(NULL,
3296                     mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3297         }
3298         INP_UNLOCK(inp);
3299         return (0);
3300
3301 drop:
3302         INP_UNLOCK(inp);
3303         m_freem(m);
3304         return (0);
3305 }