sys/netinet/tcp_input.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                                 /*
 841                                  * Parse the TCP options here because
 842                                  * syncookies need access to the reflected
 843                                  * timestamp.
 844                                  */
 845                                 tcp_dooptions(&to, optp, optlen, 0);
 846                                 if (!syncache_expand(&inc, &to, th, &so, m)) {
 847                                         /*
 848                                          * No syncache entry, or ACK was not
 849                                          * for our SYN/ACK.  Send a RST.
 850                                          */
 851                                         tcpstat.tcps_badsyn++;
 852                                         rstreason = BANDLIM_RST_OPENPORT;
 853                                         goto dropwithreset;
 854                                 }
 855                                 if (so == NULL) {
 856                                         /*
 857                                          * Could not complete 3-way handshake,
 858                                          * connection is being closed down, and
 859                                          * syncache has free'd mbuf.
 860                                          */
 861                                         INP_UNLOCK(inp);
 862                                         INP_INFO_WUNLOCK(&tcbinfo);
 863                                         return;
 864                                 }
 865                                 /*
 866                                  * Socket is created in state SYN_RECEIVED.
 867                                  * Continue processing segment.
 868                                  */
 869                                 INP_UNLOCK(inp);
 870                                 inp = sotoinpcb(so);
 871                                 INP_LOCK(inp);
 872                                 tp = intotcpcb(inp);
 873                                 /*
 874                                  * This is what would have happened in
 875                                  * tcp_output() when the SYN,ACK was sent.
 876                                  */
 877                                 tp->snd_up = tp->snd_una;
 878                                 tp->snd_max = tp->snd_nxt = tp->iss + 1;
 879                                 tp->last_ack_sent = tp->rcv_nxt;
 880                                 goto after_listen;
 881                         }
 882                         if (thflags & TH_RST) {
 883                                 syncache_chkrst(&inc, th);
 884                                 goto drop;
 885                         }
 886                         if (thflags & TH_ACK) {
 887                                 syncache_badack(&inc);
 888                                 tcpstat.tcps_badsyn++;
 889                                 rstreason = BANDLIM_RST_OPENPORT;
 890                                 goto dropwithreset;
 891                         }
 892                         goto drop;
 893                 }
 894
 895                 /*
 896                  * Segment's flags are (SYN) or (SYN|FIN).
 897                  */
 898 #ifdef INET6
 899                 /*
 900                  * If deprecated address is forbidden,
 901                  * we do not accept SYN to deprecated interface
 902                  * address to prevent any new inbound connection from
 903                  * getting established.
 904                  * When we do not accept SYN, we send a TCP RST,
 905                  * with deprecated source address (instead of dropping
 906                  * it).  We compromise it as it is much better for peer
 907                  * to send a RST, and RST will be the final packet
 908                  * for the exchange.
 909                  *
 910                  * If we do not forbid deprecated addresses, we accept
 911                  * the SYN packet.  RFC2462 does not suggest dropping
 912                  * SYN in this case.
 913                  * If we decipher RFC2462 5.5.4, it says like this:
 914                  * 1. use of deprecated addr with existing
 915                  *    communication is okay - "SHOULD continue to be
 916                  *    used"
 917                  * 2. use of it with new communication:
 918                  *   (2a) "SHOULD NOT be used if alternate address
 919                  *        with sufficient scope is available"
 920                  *   (2b) nothing mentioned otherwise.
 921                  * Here we fall into (2b) case as we have no choice in
 922                  * our source address selection - we must obey the peer.
 923                  *
 924                  * The wording in RFC2462 is confusing, and there are
 925                  * multiple description text for deprecated address
 926                  * handling - worse, they are not exactly the same.
 927                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
 928                  */
 929                 if (isipv6 && !ip6_use_deprecated) {
 930                         struct in6_ifaddr *ia6;
 931
 932                         if ((ia6 = ip6_getdstifaddr(m)) &&
 933                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
 934                                 INP_UNLOCK(inp);
 935                                 tp = NULL;
 936                                 rstreason = BANDLIM_RST_OPENPORT;
 937                                 goto dropwithreset;
 938                         }
 939                 }
 940 #endif
 941                 /*
 942                  * If it is from this socket, drop it, it must be forged.
 943                  * Don't bother responding if the destination was a broadcast.
 944                  */
 945                 if (th->th_dport == th->th_sport) {
 946                         if (isipv6) {
 947                                 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
 948                                                        &ip6->ip6_src))
 949                                         goto drop;
 950                         } else {
 951                                 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
 952                                         goto drop;
 953                         }
 954                 }
 955                 /*
 956                  * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
 957                  *
 958                  * Note that it is quite possible to receive unicast
 959                  * link-layer packets with a broadcast IP address. Use
 960                  * in_broadcast() to find them.
 961                  */
 962                 if (m->m_flags & (M_BCAST|M_MCAST))
 963                         goto drop;
 964                 if (isipv6) {
 965                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
 966                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
 967                                 goto drop;
 968                 } else {
 969                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 970                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 971                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
 972                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
 973                                 goto drop;
 974                 }
 975                 /*
 976                  * SYN appears to be valid; create compressed TCP state
 977                  * for syncache, or perform t/tcp connection.
 978                  */
 979                 if (so->so_qlen <= so->so_qlimit) {
 980 #ifdef TCPDEBUG
 981                         if (so->so_options & SO_DEBUG)
 982                                 tcp_trace(TA_INPUT, ostate, tp,
 983                                     (void *)tcp_saveipgen, &tcp_savetcp, 0);
 984 #endif
 985                         tcp_dooptions(&to, optp, optlen, TO_SYN);
 986                         if (!syncache_add(&inc, &to, th, inp, &so, m))
 987                                 goto drop;      /* XXX: does not happen */
 988                         if (so == NULL) {
 989                                 /*
 990                                  * Entry added to syncache, mbuf used to
 991                                  * send SYN,ACK packet.  Everything unlocked
 992                                  * already.
 993                                  */
 994                                 return;
 995                         }
 996                         panic("T/TCP not supported at the moment");
 997 #if 0 /* T/TCP */
 998                         /*
 999                          * Segment passed TAO tests.
1000                          * XXX: Can't happen at the moment.
1001                          */
1002                         INP_UNLOCK(inp);
1003                         inp = sotoinpcb(so);
1004                         INP_LOCK(inp);
1005                         tp = intotcpcb(inp);
1006                         tp->t_starttime = ticks;
1007                         tp->t_state = TCPS_ESTABLISHED;
1008
1009                         /*
1010                          * T/TCP logic:
1011                          * If there is a FIN or if there is data, then
1012                          * delay SYN,ACK(SYN) in the hope of piggy-backing
1013                          * it on a response segment.  Otherwise must send
1014                          * ACK now in case the other side is slow starting.
1015                          */
1016                         if (thflags & TH_FIN || tlen != 0)
1017                                 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1018                         else
1019                                 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1020                         tiwin = th->th_win << tp->snd_scale;
1021                         tcpstat.tcps_connects++;
1022                         soisconnected(so);
1023                         goto trimthenstep6;
1024 #endif  /* T/TCP */
1025                 }
1026                 goto drop;
1027         }
1028 after_listen:
1029         KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1030         INP_LOCK_ASSERT(inp);
1031
1032         /* Syncache takes care of sockets in the listen state. */
1033         KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1034
1035         /*
1036          * This is the second part of the MSS DoS prevention code (after
1037          * minmss on the sending side) and it deals with too many too small
1038          * tcp packets in a too short timeframe (1 second).
1039          *
1040          * For every full second we count the number of received packets
1041          * and bytes. If we get a lot of packets per second for this connection
1042          * (tcp_minmssoverload) we take a closer look at it and compute the
1043          * average packet size for the past second. If that is less than
1044          * tcp_minmss we get too many packets with very small payload which
1045          * is not good and burdens our system (and every packet generates
1046          * a wakeup to the process connected to our socket). We can reasonable
1047          * expect this to be small packet DoS attack to exhaust our CPU
1048          * cycles.
1049          *
1050          * Care has to be taken for the minimum packet overload value. This
1051          * value defines the minimum number of packets per second before we
1052          * start to worry. This must not be too low to avoid killing for
1053          * example interactive connections with many small packets like
1054          * telnet or SSH.
1055          *
1056          * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1057          * this check.
1058          *
1059          * Account for packet if payload packet, skip over ACK, etc.
1060          */
1061         if (tcp_minmss && tcp_minmssoverload &&
1062             tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1063                 if ((unsigned int)(tp->rcv_second - ticks) < hz) {
1064                         tp->rcv_pps++;
1065                         tp->rcv_byps += tlen + off;
1066                         if (tp->rcv_pps > tcp_minmssoverload) {
1067                                 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1068                                         printf("too many small tcp packets from "
1069                                                "%s:%u, av. %lubyte/packet, "
1070                                                "dropping connection\n",
1071 #ifdef INET6
1072                                                 isipv6 ?
1073                                                 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1074 #endif
1075                                                 inet_ntoa(inp->inp_inc.inc_faddr),
1076                                                 inp->inp_inc.inc_fport,
1077                                                 tp->rcv_byps / tp->rcv_pps);
1078                                         KASSERT(headlocked, ("tcp_input: "
1079                                             "after_listen: tcp_drop: head "
1080                                             "not locked"));
1081                                         tp = tcp_drop(tp, ECONNRESET);
1082                                         tcpstat.tcps_minmssdrops++;
1083                                         goto drop;
1084                                 }
1085                         }
1086                 } else {
1087                         tp->rcv_second = ticks + hz;
1088                         tp->rcv_pps = 1;
1089                         tp->rcv_byps = tlen + off;
1090                 }
1091         }
1092
1093         /*
1094          * Segment received on connection.
1095          * Reset idle time and keep-alive timer.
1096          */
1097         tp->t_rcvtime = ticks;
1098         if (TCPS_HAVEESTABLISHED(tp->t_state))
1099                 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1100
1101         /*
1102          * Unscale the window into a 32-bit value.
1103          * This value is bogus for the TCPS_SYN_SENT state
1104          * and is overwritten later.
1105          */
1106         tiwin = th->th_win << tp->snd_scale;
1107
1108         /*
1109          * Parse options on any incoming segment.
1110          */
1111         tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0);
1112
1113         /*
1114          * If echoed timestamp is later than the current time,
1115          * fall back to non RFC1323 RTT calculation.  Normalize
1116          * timestamp if syncookies were used when this connection
1117          * was established.
1118          */
1119         if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1120                 to.to_tsecr =- tp->ts_offset;
1121                 if (TSTMP_GT(to.to_tsecr, ticks))
1122                         to.to_tsecr = 0;
1123         }
1124
1125         /*
1126          * Process options only when we get SYN/ACK back. The SYN case
1127          * for incoming connections is handled in tcp_syncache.
1128          * XXX this is traditional behavior, may need to be cleaned up.
1129          */
1130         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1131                 if ((to.to_flags & TOF_SCALE) &&
1132                     (tp->t_flags & TF_REQ_SCALE)) {
1133                         tp->t_flags |= TF_RCVD_SCALE;
1134                         tp->snd_scale = to.to_requested_s_scale;
1135                         tp->snd_wnd = th->th_win << tp->snd_scale;
1136                         tiwin = tp->snd_wnd;
1137                 }
1138                 if (to.to_flags & TOF_TS) {
1139                         tp->t_flags |= TF_RCVD_TSTMP;
1140                         tp->ts_recent = to.to_tsval;
1141                         tp->ts_recent_age = ticks;
1142                 }
1143                 if (to.to_flags & TOF_MSS)
1144                         tcp_mss(tp, to.to_mss);
1145                 if (tp->sack_enable) {
1146                         if (!(to.to_flags & TOF_SACK))
1147                                 tp->sack_enable = 0;
1148                         else
1149                                 tp->t_flags |= TF_SACK_PERMIT;
1150                 }
1151
1152         }
1153
1154         /*
1155          * Header prediction: check for the two common cases
1156          * of a uni-directional data xfer.  If the packet has
1157          * no control flags, is in-sequence, the window didn't
1158          * change and we're not retransmitting, it's a
1159          * candidate.  If the length is zero and the ack moved
1160          * forward, we're the sender side of the xfer.  Just
1161          * free the data acked & wake any higher level process
1162          * that was blocked waiting for space.  If the length
1163          * is non-zero and the ack didn't move, we're the
1164          * receiver side.  If we're getting packets in-order
1165          * (the reassembly queue is empty), add the data to
1166          * the socket buffer and note that we need a delayed ack.
1167          * Make sure that the hidden state-flags are also off.
1168          * Since we check for TCPS_ESTABLISHED above, it can only
1169          * be TH_NEEDSYN.
1170          */
1171         if (tp->t_state == TCPS_ESTABLISHED &&
1172             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1173             ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1174             ((to.to_flags & TOF_TS) == 0 ||
1175              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1176              th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1177              tp->snd_nxt == tp->snd_max) {
1178
1179                 /*
1180                  * If last ACK falls within this segment's sequence numbers,
1181                  * record the timestamp.
1182                  * NOTE that the test is modified according to the latest
1183                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1184                  */
1185                 if ((to.to_flags & TOF_TS) != 0 &&
1186                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1187                         tp->ts_recent_age = ticks;
1188                         tp->ts_recent = to.to_tsval;
1189                 }
1190
1191                 if (tlen == 0) {
1192                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1193                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1194                             tp->snd_cwnd >= tp->snd_wnd &&
1195                             ((!tcp_do_newreno && !tp->sack_enable &&
1196                               tp->t_dupacks < tcprexmtthresh) ||
1197                              ((tcp_do_newreno || tp->sack_enable) &&
1198                               !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1199                               TAILQ_EMPTY(&tp->snd_holes)))) {
1200                                 KASSERT(headlocked, ("headlocked"));
1201                                 INP_INFO_WUNLOCK(&tcbinfo);
1202                                 headlocked = 0;
1203                                 /*
1204                                  * this is a pure ack for outstanding data.
1205                                  */
1206                                 ++tcpstat.tcps_predack;
1207                                 /*
1208                                  * "bad retransmit" recovery
1209                                  */
1210                                 if (tp->t_rxtshift == 1 &&
1211                                     ticks < tp->t_badrxtwin) {
1212                                         ++tcpstat.tcps_sndrexmitbad;
1213                                         tp->snd_cwnd = tp->snd_cwnd_prev;
1214                                         tp->snd_ssthresh =
1215                                             tp->snd_ssthresh_prev;
1216                                         tp->snd_recover = tp->snd_recover_prev;
1217                                         if (tp->t_flags & TF_WASFRECOVERY)
1218                                             ENTER_FASTRECOVERY(tp);
1219                                         tp->snd_nxt = tp->snd_max;
1220                                         tp->t_badrxtwin = 0;
1221                                 }
1222
1223                                 /*
1224                                  * Recalculate the transmit timer / rtt.
1225                                  *
1226                                  * Some boxes send broken timestamp replies
1227                                  * during the SYN+ACK phase, ignore
1228                                  * timestamps of 0 or we could calculate a
1229                                  * huge RTT and blow up the retransmit timer.
1230                                  */
1231                                 if ((to.to_flags & TOF_TS) != 0 &&
1232                                     to.to_tsecr) {
1233                                         if (!tp->t_rttlow ||
1234                                             tp->t_rttlow > ticks - to.to_tsecr)
1235                                                 tp->t_rttlow = ticks - to.to_tsecr;
1236                                         tcp_xmit_timer(tp,
1237                                             ticks - to.to_tsecr + 1);
1238                                 } else if (tp->t_rtttime &&
1239                                             SEQ_GT(th->th_ack, tp->t_rtseq)) {
1240                                         if (!tp->t_rttlow ||
1241                                             tp->t_rttlow > ticks - tp->t_rtttime)
1242                                                 tp->t_rttlow = ticks - tp->t_rtttime;
1243                                         tcp_xmit_timer(tp,
1244                                                         ticks - tp->t_rtttime);
1245                                 }
1246                                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1247                                 acked = th->th_ack - tp->snd_una;
1248                                 tcpstat.tcps_rcvackpack++;
1249                                 tcpstat.tcps_rcvackbyte += acked;
1250                                 sbdrop(&so->so_snd, acked);
1251                                 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1252                                     SEQ_LEQ(th->th_ack, tp->snd_recover))
1253                                         tp->snd_recover = th->th_ack - 1;
1254                                 tp->snd_una = th->th_ack;
1255                                 /*
1256                                  * pull snd_wl2 up to prevent seq wrap relative
1257                                  * to th_ack.
1258                                  */
1259                                 tp->snd_wl2 = th->th_ack;
1260                                 tp->t_dupacks = 0;
1261                                 m_freem(m);
1262                                 ND6_HINT(tp); /* some progress has been done */
1263
1264                                 /*
1265                                  * If all outstanding data are acked, stop
1266                                  * retransmit timer, otherwise restart timer
1267                                  * using current (possibly backed-off) value.
1268                                  * If process is waiting for space,
1269                                  * wakeup/selwakeup/signal.  If data
1270                                  * are ready to send, let tcp_output
1271                                  * decide between more output or persist.
1272
1273 #ifdef TCPDEBUG
1274                                 if (so->so_options & SO_DEBUG)
1275                                         tcp_trace(TA_INPUT, ostate, tp,
1276                                             (void *)tcp_saveipgen,
1277                                             &tcp_savetcp, 0);
1278 #endif
1279                                  */
1280                                 if (tp->snd_una == tp->snd_max)
1281                                         callout_stop(tp->tt_rexmt);
1282                                 else if (!callout_active(tp->tt_persist))
1283                                         callout_reset(tp->tt_rexmt,
1284                                                       tp->t_rxtcur,
1285                                                       tcp_timer_rexmt, tp);
1286
1287                                 sowwakeup(so);
1288                                 if (so->so_snd.sb_cc)
1289                                         (void) tcp_output(tp);
1290                                 goto check_delack;
1291                         }
1292                 } else if (th->th_ack == tp->snd_una &&
1293                     LIST_EMPTY(&tp->t_segq) &&
1294                     tlen <= sbspace(&so->so_rcv)) {
1295                         KASSERT(headlocked, ("headlocked"));
1296                         INP_INFO_WUNLOCK(&tcbinfo);
1297                         headlocked = 0;
1298                         /*
1299                          * this is a pure, in-sequence data packet
1300                          * with nothing on the reassembly queue and
1301                          * we have enough buffer space to take it.
1302                          */
1303                         /* Clean receiver SACK report if present */
1304                         if (tp->sack_enable && tp->rcv_numsacks)
1305                                 tcp_clean_sackreport(tp);
1306                         ++tcpstat.tcps_preddat;
1307                         tp->rcv_nxt += tlen;
1308                         /*
1309                          * Pull snd_wl1 up to prevent seq wrap relative to
1310                          * th_seq.
1311                          */
1312                         tp->snd_wl1 = th->th_seq;
1313                         /*
1314                          * Pull rcv_up up to prevent seq wrap relative to
1315                          * rcv_nxt.
1316                          */
1317                         tp->rcv_up = tp->rcv_nxt;
1318                         tcpstat.tcps_rcvpack++;
1319                         tcpstat.tcps_rcvbyte += tlen;
1320                         ND6_HINT(tp);   /* some progress has been done */
1321                         /*
1322 #ifdef TCPDEBUG
1323                         if (so->so_options & SO_DEBUG)
1324                                 tcp_trace(TA_INPUT, ostate, tp,
1325                                     (void *)tcp_saveipgen, &tcp_savetcp, 0);
1326 #endif
1327                          * Add data to socket buffer.
1328                          */
1329                         SOCKBUF_LOCK(&so->so_rcv);
1330                         if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1331                                 m_freem(m);
1332                         } else {
1333                                 m_adj(m, drop_hdrlen);  /* delayed header drop */
1334                                 sbappendstream_locked(&so->so_rcv, m);
1335                         }
1336                         sorwakeup_locked(so);
1337                         if (DELAY_ACK(tp)) {
1338                                 tp->t_flags |= TF_DELACK;
1339                         } else {
1340                                 tp->t_flags |= TF_ACKNOW;
1341                                 tcp_output(tp);
1342                         }
1343                         goto check_delack;
1344                 }
1345         }
1346
1347         /*
1348          * Calculate amount of space in receive window,
1349          * and then do TCP input processing.
1350          * Receive window is amount of space in rcv queue,
1351          * but not less than advertised window.
1352          */
1353         { int win;
1354
1355         win = sbspace(&so->so_rcv);
1356         if (win < 0)
1357                 win = 0;
1358         tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1359         }
1360
1361         switch (tp->t_state) {
1362
1363         /*
1364          * If the state is SYN_RECEIVED:
1365          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1366          */
1367         case TCPS_SYN_RECEIVED:
1368                 if ((thflags & TH_ACK) &&
1369                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1370                      SEQ_GT(th->th_ack, tp->snd_max))) {
1371                                 rstreason = BANDLIM_RST_OPENPORT;
1372                                 goto dropwithreset;
1373                 }
1374                 break;
1375
1376         /*
1377          * If the state is SYN_SENT:
1378          *      if seg contains an ACK, but not for our SYN, drop the input.
1379          *      if seg contains a RST, then drop the connection.
1380          *      if seg does not contain SYN, then drop it.
1381          * Otherwise this is an acceptable SYN segment
1382          *      initialize tp->rcv_nxt and tp->irs
1383          *      if seg contains ack then advance tp->snd_una
1384          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1385          *      arrange for segment to be acked (eventually)
1386          *      continue processing rest of data/controls, beginning with URG
1387          */
1388         case TCPS_SYN_SENT:
1389                 if ((thflags & TH_ACK) &&
1390                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1391                      SEQ_GT(th->th_ack, tp->snd_max))) {
1392                         rstreason = BANDLIM_UNLIMITED;
1393                         goto dropwithreset;
1394                 }
1395                 if (thflags & TH_RST) {
1396                         if (thflags & TH_ACK) {
1397                                 KASSERT(headlocked, ("tcp_input: after_listen"
1398                                     ": tcp_drop.2: head not locked"));
1399                                 tp = tcp_drop(tp, ECONNREFUSED);
1400                         }
1401                         goto drop;
1402                 }
1403                 if ((thflags & TH_SYN) == 0)
1404                         goto drop;
1405
1406                 /* Initial send window, already scaled. */
1407                 tp->snd_wnd = th->th_win;
1408
1409                 tp->irs = th->th_seq;
1410                 tcp_rcvseqinit(tp);
1411                 if (thflags & TH_ACK) {
1412                         tcpstat.tcps_connects++;
1413                         soisconnected(so);
1414 #ifdef MAC
1415                         SOCK_LOCK(so);
1416                         mac_set_socket_peer_from_mbuf(m, so);
1417                         SOCK_UNLOCK(so);
1418 #endif
1419                         /* Do window scaling on this connection? */
1420                         if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1421                                 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1422                                 tp->rcv_scale = tp->request_r_scale;
1423                         }
1424                         tp->rcv_adv += tp->rcv_wnd;
1425                         tp->snd_una++;          /* SYN is acked */
1426                         /*
1427                          * If there's data, delay ACK; if there's also a FIN
1428                          * ACKNOW will be turned on later.
1429                          */
1430                         if (DELAY_ACK(tp) && tlen != 0)
1431                                 callout_reset(tp->tt_delack, tcp_delacktime,
1432                                     tcp_timer_delack, tp);
1433                         else
1434                                 tp->t_flags |= TF_ACKNOW;
1435                         /*
1436                          * Received <SYN,ACK> in SYN_SENT[*] state.
1437                          * Transitions:
1438                          *      SYN_SENT  --> ESTABLISHED
1439                          *      SYN_SENT* --> FIN_WAIT_1
1440                          */
1441                         tp->t_starttime = ticks;
1442                         if (tp->t_flags & TF_NEEDFIN) {
1443                                 tp->t_state = TCPS_FIN_WAIT_1;
1444                                 tp->t_flags &= ~TF_NEEDFIN;
1445                                 thflags &= ~TH_SYN;
1446                         } else {
1447                                 tp->t_state = TCPS_ESTABLISHED;
1448                                 callout_reset(tp->tt_keep, tcp_keepidle,
1449                                               tcp_timer_keep, tp);
1450                         }
1451                 } else {
1452                         /*
1453                          * Received initial SYN in SYN-SENT[*] state =>
1454                          * simultaneous open.  If segment contains CC option
1455                          * and there is a cached CC, apply TAO test.
1456                          * If it succeeds, connection is * half-synchronized.
1457                          * Otherwise, do 3-way handshake:
1458                          *        SYN-SENT -> SYN-RECEIVED
1459                          *        SYN-SENT* -> SYN-RECEIVED*
1460                          * If there was no CC option, clear cached CC value.
1461                          */
1462                         tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1463                         callout_stop(tp->tt_rexmt);
1464                         tp->t_state = TCPS_SYN_RECEIVED;
1465                 }
1466
1467 #if 0 /* T/TCP */
1468 trimthenstep6:
1469 #endif
1470                 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1471                     "locked"));
1472                 INP_LOCK_ASSERT(inp);
1473
1474                 /*
1475                  * Advance th->th_seq to correspond to first data byte.
1476                  * If data, trim to stay within window,
1477                  * dropping FIN if necessary.
1478                  */
1479                 th->th_seq++;
1480                 if (tlen > tp->rcv_wnd) {
1481                         todrop = tlen - tp->rcv_wnd;
1482                         m_adj(m, -todrop);
1483                         tlen = tp->rcv_wnd;
1484                         thflags &= ~TH_FIN;
1485                         tcpstat.tcps_rcvpackafterwin++;
1486                         tcpstat.tcps_rcvbyteafterwin += todrop;
1487                 }
1488                 tp->snd_wl1 = th->th_seq - 1;
1489                 tp->rcv_up = th->th_seq;
1490                 /*
1491                  * Client side of transaction: already sent SYN and data.
1492                  * If the remote host used T/TCP to validate the SYN,
1493                  * our data will be ACK'd; if so, enter normal data segment
1494                  * processing in the middle of step 5, ack processing.
1495                  * Otherwise, goto step 6.
1496                  */
1497                 if (thflags & TH_ACK)
1498                         goto process_ACK;
1499
1500                 goto step6;
1501
1502         /*
1503          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1504          *      do normal processing.
1505          *
1506          * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
1507          */
1508         case TCPS_LAST_ACK:
1509         case TCPS_CLOSING:
1510         case TCPS_TIME_WAIT:
1511                 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1512                 break;  /* continue normal processing */
1513         }
1514
1515         /*
1516          * States other than LISTEN or SYN_SENT.
1517          * First check the RST flag and sequence number since reset segments
1518          * are exempt from the timestamp and connection count tests.  This
1519          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1520          * below which allowed reset segments in half the sequence space
1521          * to fall though and be processed (which gives forged reset
1522          * segments with a random sequence number a 50 percent chance of
1523          * killing a connection).
1524          * Then check timestamp, if present.
1525          * Then check the connection count, if present.
1526          * Then check that at least some bytes of segment are within
1527          * receive window.  If segment begins before rcv_nxt,
1528          * drop leading data (and SYN); if nothing left, just ack.
1529          *
1530          *
1531          * If the RST bit is set, check the sequence number to see
1532          * if this is a valid reset segment.
1533          * RFC 793 page 37:
1534          *   In all states except SYN-SENT, all reset (RST) segments
1535          *   are validated by checking their SEQ-fields.  A reset is
1536          *   valid if its sequence number is in the window.
1537          * Note: this does not take into account delayed ACKs, so
1538          *   we should test against last_ack_sent instead of rcv_nxt.
1539          *   The sequence number in the reset segment is normally an
1540          *   echo of our outgoing acknowlegement numbers, but some hosts
1541          *   send a reset with the sequence number at the rightmost edge
1542          *   of our receive window, and we have to handle this case.
1543          * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1544          *   that brute force RST attacks are possible.  To combat this,
1545          *   we use a much stricter check while in the ESTABLISHED state,
1546          *   only accepting RSTs where the sequence number is equal to
1547          *   last_ack_sent.  In all other states (the states in which a
1548          *   RST is more likely), the more permissive check is used.
1549          * If we have multiple segments in flight, the intial reset
1550          * segment sequence numbers will be to the left of last_ack_sent,
1551          * but they will eventually catch up.
1552          * In any case, it never made sense to trim reset segments to
1553          * fit the receive window since RFC 1122 says:
1554          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1555          *
1556          *    A TCP SHOULD allow a received RST segment to include data.
1557          *
1558          *    DISCUSSION
1559          *         It has been suggested that a RST segment could contain
1560          *         ASCII text that encoded and explained the cause of the
1561          *         RST.  No standard has yet been established for such
1562          *         data.
1563          *
1564          * If the reset segment passes the sequence number test examine
1565          * the state:
1566          *    SYN_RECEIVED STATE:
1567          *      If passive open, return to LISTEN state.
1568          *      If active open, inform user that connection was refused.
1569          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1570          *      Inform user that connection was reset, and close tcb.
1571          *    CLOSING, LAST_ACK STATES:
1572          *      Close the tcb.
1573          *    TIME_WAIT STATE:
1574          *      Drop the segment - see Stevens, vol. 2, p. 964 and
1575          *      RFC 1337.
1576          */
1577         if (thflags & TH_RST) {
1578                 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1579                     SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1580                     (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1581                         switch (tp->t_state) {
1582
1583                         case TCPS_SYN_RECEIVED:
1584                                 so->so_error = ECONNREFUSED;
1585                                 goto close;
1586
1587                         case TCPS_ESTABLISHED:
1588                                 if (tp->last_ack_sent != th->th_seq &&
1589                                     tcp_insecure_rst == 0) {
1590                                         tcpstat.tcps_badrst++;
1591                                         goto drop;
1592                                 }
1593                         case TCPS_FIN_WAIT_1:
1594                         case TCPS_FIN_WAIT_2:
1595                         case TCPS_CLOSE_WAIT:
1596                                 so->so_error = ECONNRESET;
1597                         close:
1598                                 tp->t_state = TCPS_CLOSED;
1599                                 tcpstat.tcps_drops++;
1600                                 KASSERT(headlocked, ("tcp_input: "
1601                                     "trimthenstep6: tcp_close: head not "
1602                                     "locked"));
1603                                 tp = tcp_close(tp);
1604                                 break;
1605
1606                         case TCPS_CLOSING:
1607                         case TCPS_LAST_ACK:
1608                                 KASSERT(headlocked, ("trimthenstep6: "
1609                                     "tcp_close.2: head not locked"));
1610                                 tp = tcp_close(tp);
1611                                 break;
1612
1613                         case TCPS_TIME_WAIT:
1614                                 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1615                                     ("timewait"));
1616                                 break;
1617                         }
1618                 }
1619                 goto drop;
1620         }
1621
1622         /*
1623          * RFC 1323 PAWS: If we have a timestamp reply on this segment
1624          * and it's less than ts_recent, drop it.
1625          */
1626         if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1627             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1628
1629                 /* Check to see if ts_recent is over 24 days old.  */
1630                 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1631                         /*
1632                          * Invalidate ts_recent.  If this segment updates
1633                          * ts_recent, the age will be reset later and ts_recent
1634                          * will get a valid value.  If it does not, setting
1635                          * ts_recent to zero will at least satisfy the
1636                          * requirement that zero be placed in the timestamp
1637                          * echo reply when ts_recent isn't valid.  The
1638                          * age isn't reset until we get a valid ts_recent
1639                          * because we don't want out-of-order segments to be
1640                          * dropped when ts_recent is old.
1641                          */
1642                         tp->ts_recent = 0;
1643                 } else {
1644                         tcpstat.tcps_rcvduppack++;
1645                         tcpstat.tcps_rcvdupbyte += tlen;
1646                         tcpstat.tcps_pawsdrop++;
1647                         if (tlen)
1648                                 goto dropafterack;
1649                         goto drop;
1650                 }
1651         }
1652
1653         /*
1654          * In the SYN-RECEIVED state, validate that the packet belongs to
1655          * this connection before trimming the data to fit the receive
1656          * window.  Check the sequence number versus IRS since we know
1657          * the sequence numbers haven't wrapped.  This is a partial fix
1658          * for the "LAND" DoS attack.
1659          */
1660         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1661                 rstreason = BANDLIM_RST_OPENPORT;
1662                 goto dropwithreset;
1663         }
1664
1665         todrop = tp->rcv_nxt - th->th_seq;
1666         if (todrop > 0) {
1667                 if (thflags & TH_SYN) {
1668                         thflags &= ~TH_SYN;
1669                         th->th_seq++;
1670                         if (th->th_urp > 1)
1671                                 th->th_urp--;
1672                         else
1673                                 thflags &= ~TH_URG;
1674                         todrop--;
1675                 }
1676                 /*
1677                  * Following if statement from Stevens, vol. 2, p. 960.
1678                  */
1679                 if (todrop > tlen
1680                     || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1681                         /*
1682                          * Any valid FIN must be to the left of the window.
1683                          * At this point the FIN must be a duplicate or out
1684                          * of sequence; drop it.
1685                          */
1686                         thflags &= ~TH_FIN;
1687
1688                         /*
1689                          * Send an ACK to resynchronize and drop any data.
1690                          * But keep on processing for RST or ACK.
1691                          */
1692                         tp->t_flags |= TF_ACKNOW;
1693                         todrop = tlen;
1694                         tcpstat.tcps_rcvduppack++;
1695                         tcpstat.tcps_rcvdupbyte += todrop;
1696                 } else {
1697                         tcpstat.tcps_rcvpartduppack++;
1698                         tcpstat.tcps_rcvpartdupbyte += todrop;
1699                 }
1700                 drop_hdrlen += todrop;  /* drop from the top afterwards */
1701                 th->th_seq += todrop;
1702                 tlen -= todrop;
1703                 if (th->th_urp > todrop)
1704                         th->th_urp -= todrop;
1705                 else {
1706                         thflags &= ~TH_URG;
1707                         th->th_urp = 0;
1708                 }
1709         }
1710
1711         /*
1712          * If new data are received on a connection after the
1713          * user processes are gone, then RST the other end.
1714          */
1715         if ((so->so_state & SS_NOFDREF) &&
1716             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1717                 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1718                     "locked"));
1719                 tp = tcp_close(tp);
1720                 tcpstat.tcps_rcvafterclose++;
1721                 rstreason = BANDLIM_UNLIMITED;
1722                 goto dropwithreset;
1723         }
1724
1725         /*
1726          * If segment ends after window, drop trailing data
1727          * (and PUSH and FIN); if nothing left, just ACK.
1728          */
1729         todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1730         if (todrop > 0) {
1731                 tcpstat.tcps_rcvpackafterwin++;
1732                 if (todrop >= tlen) {
1733                         tcpstat.tcps_rcvbyteafterwin += tlen;
1734                         /*
1735                          * If a new connection request is received
1736                          * while in TIME_WAIT, drop the old connection
1737                          * and start over if the sequence numbers
1738                          * are above the previous ones.
1739                          */
1740                         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1741                         if (thflags & TH_SYN &&
1742                             tp->t_state == TCPS_TIME_WAIT &&
1743                             SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1744                                 KASSERT(headlocked, ("trimthenstep6: "
1745                                     "tcp_close.4: head not locked"));
1746                                 tp = tcp_close(tp);
1747                                 goto findpcb;
1748                         }
1749                         /*
1750                          * If window is closed can only take segments at
1751                          * window edge, and have to drop data and PUSH from
1752                          * incoming segments.  Continue processing, but
1753                          * remember to ack.  Otherwise, drop segment
1754                          * and ack.
1755                          */
1756                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1757                                 tp->t_flags |= TF_ACKNOW;
1758                                 tcpstat.tcps_rcvwinprobe++;
1759                         } else
1760                                 goto dropafterack;
1761                 } else
1762                         tcpstat.tcps_rcvbyteafterwin += todrop;
1763                 m_adj(m, -todrop);
1764                 tlen -= todrop;
1765                 thflags &= ~(TH_PUSH|TH_FIN);
1766         }
1767
1768         /*
1769          * If last ACK falls within this segment's sequence numbers,
1770          * record its timestamp.
1771          * NOTE:
1772          * 1) That the test incorporates suggestions from the latest
1773          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1774          * 2) That updating only on newer timestamps interferes with
1775          *    our earlier PAWS tests, so this check should be solely
1776          *    predicated on the sequence space of this segment.
1777          * 3) That we modify the segment boundary check to be
1778          *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
1779          *    instead of RFC1323's
1780          *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
1781          *    This modified check allows us to overcome RFC1323's
1782          *    limitations as described in Stevens TCP/IP Illustrated
1783          *    Vol. 2 p.869. In such cases, we can still calculate the
1784          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1785          */
1786         if ((to.to_flags & TOF_TS) != 0 &&
1787             SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1788             SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1789                 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1790                 tp->ts_recent_age = ticks;
1791                 tp->ts_recent = to.to_tsval;
1792         }
1793
1794         /*
1795          * If a SYN is in the window, then this is an
1796          * error and we send an RST and drop the connection.
1797          */
1798         if (thflags & TH_SYN) {
1799                 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1800                     "head not locked"));
1801                 tp = tcp_drop(tp, ECONNRESET);
1802                 rstreason = BANDLIM_UNLIMITED;
1803                 goto drop;
1804         }
1805
1806         /*
1807          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1808          * flag is on (half-synchronized state), then queue data for
1809          * later processing; else drop segment and return.
1810          */
1811         if ((thflags & TH_ACK) == 0) {
1812                 if (tp->t_state == TCPS_SYN_RECEIVED ||
1813                     (tp->t_flags & TF_NEEDSYN))
1814                         goto step6;
1815                 else if (tp->t_flags & TF_ACKNOW)
1816                         goto dropafterack;
1817                 else
1818                         goto drop;
1819         }
1820
1821         /*
1822          * Ack processing.
1823          */
1824         switch (tp->t_state) {
1825
1826         /*
1827          * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1828          * ESTABLISHED state and continue processing.
1829          * The ACK was checked above.
1830          */
1831         case TCPS_SYN_RECEIVED:
1832
1833                 tcpstat.tcps_connects++;
1834                 soisconnected(so);
1835                 /* Do window scaling? */
1836                 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1837                         (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1838                         tp->rcv_scale = tp->request_r_scale;
1839                         tp->snd_wnd = tiwin;
1840                 }
1841                 /*
1842                  * Make transitions:
1843                  *      SYN-RECEIVED  -> ESTABLISHED
1844                  *      SYN-RECEIVED* -> FIN-WAIT-1
1845                  */
1846                 tp->t_starttime = ticks;
1847                 if (tp->t_flags & TF_NEEDFIN) {
1848                         tp->t_state = TCPS_FIN_WAIT_1;
1849                         tp->t_flags &= ~TF_NEEDFIN;
1850                 } else {
1851                         tp->t_state = TCPS_ESTABLISHED;
1852                         callout_reset(tp->tt_keep, tcp_keepidle,
1853                                       tcp_timer_keep, tp);
1854                 }
1855                 /*
1856                  * If segment contains data or ACK, will call tcp_reass()
1857                  * later; if not, do so now to pass queued data to user.
1858                  */
1859                 if (tlen == 0 && (thflags & TH_FIN) == 0)
1860                         (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1861                             (struct mbuf *)0);
1862                 tp->snd_wl1 = th->th_seq - 1;
1863                 /* FALLTHROUGH */
1864
1865         /*
1866          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1867          * ACKs.  If the ack is in the range
1868          *      tp->snd_una < th->th_ack <= tp->snd_max
1869          * then advance tp->snd_una to th->th_ack and drop
1870          * data from the retransmission queue.  If this ACK reflects
1871          * more up to date window information we update our window information.
1872          */
1873         case TCPS_ESTABLISHED:
1874         case TCPS_FIN_WAIT_1:
1875         case TCPS_FIN_WAIT_2:
1876         case TCPS_CLOSE_WAIT:
1877         case TCPS_CLOSING:
1878         case TCPS_LAST_ACK:
1879         case TCPS_TIME_WAIT:
1880                 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1881                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1882                         tcpstat.tcps_rcvacktoomuch++;
1883                         goto dropafterack;
1884                 }
1885                 if (tp->sack_enable &&
1886                     (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1887                         tcp_sack_doack(tp, &to, th->th_ack);
1888                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1889                         if (tlen == 0 && tiwin == tp->snd_wnd) {
1890                                 tcpstat.tcps_rcvdupack++;
1891                                 /*
1892                                  * If we have outstanding data (other than
1893                                  * a window probe), this is a completely
1894                                  * duplicate ack (ie, window info didn't
1895                                  * change), the ack is the biggest we've
1896                                  * seen and we've seen exactly our rexmt
1897                                  * threshhold of them, assume a packet
1898                                  * has been dropped and retransmit it.
1899                                  * Kludge snd_nxt & the congestion
1900                                  * window so we send only this one
1901                                  * packet.
1902                                  *
1903                                  * We know we're losing at the current
1904                                  * window size so do congestion avoidance
1905                                  * (set ssthresh to half the current window
1906                                  * and pull our congestion window back to
1907                                  * the new ssthresh).
1908                                  *
1909                                  * Dup acks mean that packets have left the
1910                                  * network (they're now cached at the receiver)
1911                                  * so bump cwnd by the amount in the receiver
1912                                  * to keep a constant cwnd packets in the
1913                                  * network.
1914                                  */
1915                                 if (!callout_active(tp->tt_rexmt) ||
1916                                     th->th_ack != tp->snd_una)
1917                                         tp->t_dupacks = 0;
1918                                 else if (++tp->t_dupacks > tcprexmtthresh ||
1919                                          ((tcp_do_newreno || tp->sack_enable) &&
1920                                           IN_FASTRECOVERY(tp))) {
1921                                         if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1922                                                 int awnd;
1923
1924                                                 /*
1925                                                  * Compute the amount of data in flight first.
1926                                                  * We can inject new data into the pipe iff
1927                                                  * we have less than 1/2 the original window's
1928                                                  * worth of data in flight.
1929                                                  */
1930                                                 awnd = (tp->snd_nxt - tp->snd_fack) +
1931                                                         tp->sackhint.sack_bytes_rexmit;
1932                                                 if (awnd < tp->snd_ssthresh) {
1933                                                         tp->snd_cwnd += tp->t_maxseg;
1934                                                         if (tp->snd_cwnd > tp->snd_ssthresh)
1935                                                                 tp->snd_cwnd = tp->snd_ssthresh;
1936                                                 }
1937                                         } else
1938                                                 tp->snd_cwnd += tp->t_maxseg;
1939                                         (void) tcp_output(tp);
1940                                         goto drop;
1941                                 } else if (tp->t_dupacks == tcprexmtthresh) {
1942                                         tcp_seq onxt = tp->snd_nxt;
1943                                         u_int win;
1944
1945                                         /*
1946                                          * If we're doing sack, check to
1947                                          * see if we're already in sack
1948                                          * recovery. If we're not doing sack,
1949                                          * check to see if we're in newreno
1950                                          * recovery.
1951                                          */
1952                                         if (tp->sack_enable) {
1953                                                 if (IN_FASTRECOVERY(tp)) {
1954                                                         tp->t_dupacks = 0;
1955                                                         break;
1956                                                 }
1957                                         } else if (tcp_do_newreno) {
1958                                                 if (SEQ_LEQ(th->th_ack,
1959                                                     tp->snd_recover)) {
1960                                                         tp->t_dupacks = 0;
1961                                                         break;
1962                                                 }
1963                                         }
1964                                         win = min(tp->snd_wnd, tp->snd_cwnd) /
1965                                             2 / tp->t_maxseg;
1966                                         if (win < 2)
1967                                                 win = 2;
1968                                         tp->snd_ssthresh = win * tp->t_maxseg;
1969                                         ENTER_FASTRECOVERY(tp);
1970                                         tp->snd_recover = tp->snd_max;
1971                                         callout_stop(tp->tt_rexmt);
1972                                         tp->t_rtttime = 0;
1973                                         if (tp->sack_enable) {
1974                                                 tcpstat.tcps_sack_recovery_episode++;
1975                                                 tp->sack_newdata = tp->snd_nxt;
1976                                                 tp->snd_cwnd = tp->t_maxseg;
1977                                                 (void) tcp_output(tp);
1978                                                 goto drop;
1979                                         }
1980                                         tp->snd_nxt = th->th_ack;
1981                                         tp->snd_cwnd = tp->t_maxseg;
1982                                         (void) tcp_output(tp);
1983                                         KASSERT(tp->snd_limited <= 2,
1984                                             ("tp->snd_limited too big"));
1985                                         tp->snd_cwnd = tp->snd_ssthresh +
1986                                              tp->t_maxseg *
1987                                              (tp->t_dupacks - tp->snd_limited);
1988                                         if (SEQ_GT(onxt, tp->snd_nxt))
1989                                                 tp->snd_nxt = onxt;
1990                                         goto drop;
1991                                 } else if (tcp_do_rfc3042) {
1992                                         u_long oldcwnd = tp->snd_cwnd;
1993                                         tcp_seq oldsndmax = tp->snd_max;
1994                                         u_int sent;
1995
1996                                         KASSERT(tp->t_dupacks == 1 ||
1997                                             tp->t_dupacks == 2,
1998                                             ("dupacks not 1 or 2"));
1999                                         if (tp->t_dupacks == 1)
2000                                                 tp->snd_limited = 0;
2001                                         tp->snd_cwnd =
2002                                             (tp->snd_nxt - tp->snd_una) +
2003                                             (tp->t_dupacks - tp->snd_limited) *
2004                                             tp->t_maxseg;
2005                                         (void) tcp_output(tp);
2006                                         sent = tp->snd_max - oldsndmax;
2007                                         if (sent > tp->t_maxseg) {
2008                                                 KASSERT((tp->t_dupacks == 2 &&
2009                                                     tp->snd_limited == 0) ||
2010                                                    (sent == tp->t_maxseg + 1 &&
2011                                                     tp->t_flags & TF_SENTFIN),
2012                                                     ("sent too much"));
2013                                                 tp->snd_limited = 2;
2014                                         } else if (sent > 0)
2015                                                 ++tp->snd_limited;
2016                                         tp->snd_cwnd = oldcwnd;
2017                                         goto drop;
2018                                 }
2019                         } else
2020                                 tp->t_dupacks = 0;
2021                         break;
2022                 }
2023
2024                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2025
2026                 /*
2027                  * If the congestion window was inflated to account
2028                  * for the other side's cached packets, retract it.
2029                  */
2030                 if (tcp_do_newreno || tp->sack_enable) {
2031                         if (IN_FASTRECOVERY(tp)) {
2032                                 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2033                                         if (tp->sack_enable)
2034                                                 tcp_sack_partialack(tp, th);
2035                                         else
2036                                                 tcp_newreno_partial_ack(tp, th);
2037                                 } else {
2038                                         /*
2039                                          * Out of fast recovery.
2040                                          * Window inflation should have left us
2041                                          * with approximately snd_ssthresh
2042                                          * outstanding data.
2043                                          * But in case we would be inclined to
2044                                          * send a burst, better to do it via
2045                                          * the slow start mechanism.
2046                                          */
2047                                         if (SEQ_GT(th->th_ack +
2048                                                         tp->snd_ssthresh,
2049                                                    tp->snd_max))
2050                                                 tp->snd_cwnd = tp->snd_max -
2051                                                                 th->th_ack +
2052                                                                 tp->t_maxseg;
2053                                         else
2054                                                 tp->snd_cwnd = tp->snd_ssthresh;
2055                                 }
2056                         }
2057                 } else {
2058                         if (tp->t_dupacks >= tcprexmtthresh &&
2059                             tp->snd_cwnd > tp->snd_ssthresh)
2060                                 tp->snd_cwnd = tp->snd_ssthresh;
2061                 }
2062                 tp->t_dupacks = 0;
2063                 /*
2064                  * If we reach this point, ACK is not a duplicate,
2065                  *     i.e., it ACKs something we sent.
2066                  */
2067                 if (tp->t_flags & TF_NEEDSYN) {
2068                         /*
2069                          * T/TCP: Connection was half-synchronized, and our
2070                          * SYN has been ACK'd (so connection is now fully
2071                          * synchronized).  Go to non-starred state,
2072                          * increment snd_una for ACK of SYN, and check if
2073                          * we can do window scaling.
2074                          */
2075                         tp->t_flags &= ~TF_NEEDSYN;
2076                         tp->snd_una++;
2077                         /* Do window scaling? */
2078                         if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2079                                 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2080                                 tp->rcv_scale = tp->request_r_scale;
2081                                 /* Send window already scaled. */
2082                         }
2083                 }
2084
2085 process_ACK:
2086                 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2087                     "locked"));
2088                 INP_LOCK_ASSERT(inp);
2089
2090                 acked = th->th_ack - tp->snd_una;
2091                 tcpstat.tcps_rcvackpack++;
2092                 tcpstat.tcps_rcvackbyte += acked;
2093
2094                 /*
2095                  * If we just performed our first retransmit, and the ACK
2096                  * arrives within our recovery window, then it was a mistake
2097                  * to do the retransmit in the first place.  Recover our
2098                  * original cwnd and ssthresh, and proceed to transmit where
2099                  * we left off.
2100                  */
2101                 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2102                         ++tcpstat.tcps_sndrexmitbad;
2103                         tp->snd_cwnd = tp->snd_cwnd_prev;
2104                         tp->snd_ssthresh = tp->snd_ssthresh_prev;
2105                         tp->snd_recover = tp->snd_recover_prev;
2106                         if (tp->t_flags & TF_WASFRECOVERY)
2107                                 ENTER_FASTRECOVERY(tp);
2108                         tp->snd_nxt = tp->snd_max;
2109                         tp->t_badrxtwin = 0;    /* XXX probably not required */
2110                 }
2111
2112                 /*
2113                  * If we have a timestamp reply, update smoothed
2114                  * round trip time.  If no timestamp is present but
2115                  * transmit timer is running and timed sequence
2116                  * number was acked, update smoothed round trip time.
2117                  * Since we now have an rtt measurement, cancel the
2118                  * timer backoff (cf., Phil Karn's retransmit alg.).
2119                  * Recompute the initial retransmit timer.
2120                  *
2121                  * Some boxes send broken timestamp replies
2122                  * during the SYN+ACK phase, ignore
2123                  * timestamps of 0 or we could calculate a
2124                  * huge RTT and blow up the retransmit timer.
2125                  */
2126                 if ((to.to_flags & TOF_TS) != 0 &&
2127                     to.to_tsecr) {
2128                         if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2129                                 tp->t_rttlow = ticks - to.to_tsecr;
2130                         tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2131                 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2132                         if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2133                                 tp->t_rttlow = ticks - tp->t_rtttime;
2134                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2135                 }
2136                 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2137
2138                 /*
2139                  * If all outstanding data is acked, stop retransmit
2140                  * timer and remember to restart (more output or persist).
2141                  * If there is more data to be acked, restart retransmit
2142                  * timer, using current (possibly backed-off) value.
2143                  */
2144                 if (th->th_ack == tp->snd_max) {
2145                         callout_stop(tp->tt_rexmt);
2146                         needoutput = 1;
2147                 } else if (!callout_active(tp->tt_persist))
2148                         callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2149                                       tcp_timer_rexmt, tp);
2150
2151                 /*
2152                  * If no data (only SYN) was ACK'd,
2153                  *    skip rest of ACK processing.
2154                  */
2155                 if (acked == 0)
2156                         goto step6;
2157
2158                 /*
2159                  * When new data is acked, open the congestion window.
2160                  * If the window gives us less than ssthresh packets
2161                  * in flight, open exponentially (maxseg per packet).
2162                  * Otherwise open linearly: maxseg per window
2163                  * (maxseg^2 / cwnd per packet).
2164                  */
2165                 if ((!tcp_do_newreno && !tp->sack_enable) ||
2166                     !IN_FASTRECOVERY(tp)) {
2167                         register u_int cw = tp->snd_cwnd;
2168                         register u_int incr = tp->t_maxseg;
2169                         if (cw > tp->snd_ssthresh)
2170                                 incr = incr * incr / cw;
2171                         tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2172                 }
2173                 SOCKBUF_LOCK(&so->so_snd);
2174                 if (acked > so->so_snd.sb_cc) {
2175                         tp->snd_wnd -= so->so_snd.sb_cc;
2176                         sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2177                         ourfinisacked = 1;
2178                 } else {
2179                         sbdrop_locked(&so->so_snd, acked);
2180                         tp->snd_wnd -= acked;
2181                         ourfinisacked = 0;
2182                 }
2183                 sowwakeup_locked(so);
2184                 /* detect una wraparound */
2185                 if ((tcp_do_newreno || tp->sack_enable) &&
2186                     !IN_FASTRECOVERY(tp) &&
2187                     SEQ_GT(tp->snd_una, tp->snd_recover) &&
2188                     SEQ_LEQ(th->th_ack, tp->snd_recover))
2189                         tp->snd_recover = th->th_ack - 1;
2190                 if ((tcp_do_newreno || tp->sack_enable) &&
2191                     IN_FASTRECOVERY(tp) &&
2192                     SEQ_GEQ(th->th_ack, tp->snd_recover))
2193                         EXIT_FASTRECOVERY(tp);
2194                 tp->snd_una = th->th_ack;
2195                 if (tp->sack_enable) {
2196                         if (SEQ_GT(tp->snd_una, tp->snd_recover))
2197                                 tp->snd_recover = tp->snd_una;
2198                 }
2199                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2200                         tp->snd_nxt = tp->snd_una;
2201
2202                 switch (tp->t_state) {
2203
2204                 /*
2205                  * In FIN_WAIT_1 STATE in addition to the processing
2206                  * for the ESTABLISHED state if our FIN is now acknowledged
2207                  * then enter FIN_WAIT_2.
2208                  */
2209                 case TCPS_FIN_WAIT_1:
2210                         if (ourfinisacked) {
2211                                 /*
2212                                  * If we can't receive any more
2213                                  * data, then closing user can proceed.
2214                                  * Starting the timer is contrary to the
2215                                  * specification, but if we don't get a FIN
2216                                  * we'll hang forever.
2217                                  */
2218                 /* XXXjl
2219                  * we should release the tp also, and use a
2220                  * compressed state.
2221                  */
2222                                 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2223                                         soisdisconnected(so);
2224                                         callout_reset(tp->tt_2msl, tcp_maxidle,
2225                                                       tcp_timer_2msl, tp);
2226                                 }
2227                                 tp->t_state = TCPS_FIN_WAIT_2;
2228                         }
2229                         break;
2230
2231                 /*
2232                  * In CLOSING STATE in addition to the processing for
2233                  * the ESTABLISHED state if the ACK acknowledges our FIN
2234                  * then enter the TIME-WAIT state, otherwise ignore
2235                  * the segment.
2236                  */
2237                 case TCPS_CLOSING:
2238                         if (ourfinisacked) {
2239                                 KASSERT(headlocked, ("tcp_input: process_ACK: "
2240                                     "head not locked"));
2241                                 tcp_twstart(tp);
2242                                 INP_INFO_WUNLOCK(&tcbinfo);
2243                                 m_freem(m);
2244                                 return;
2245                         }
2246                         break;
2247
2248                 /*
2249                  * In LAST_ACK, we may still be waiting for data to drain
2250                  * and/or to be acked, as well as for the ack of our FIN.
2251                  * If our FIN is now acknowledged, delete the TCB,
2252                  * enter the closed state and return.
2253                  */
2254                 case TCPS_LAST_ACK:
2255                         if (ourfinisacked) {
2256                                 KASSERT(headlocked, ("tcp_input: process_ACK:"
2257                                     " tcp_close: head not locked"));
2258                                 tp = tcp_close(tp);
2259                                 goto drop;
2260                         }
2261                         break;
2262
2263                 /*
2264                  * In TIME_WAIT state the only thing that should arrive
2265                  * is a retransmission of the remote FIN.  Acknowledge
2266                  * it and restart the finack timer.
2267                  */
2268                 case TCPS_TIME_WAIT:
2269                         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2270                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2271                                       tcp_timer_2msl, tp);
2272                         goto dropafterack;
2273                 }
2274         }
2275
2276 step6:
2277         KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2278         INP_LOCK_ASSERT(inp);
2279
2280         /*
2281          * Update window information.
2282          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2283          */
2284         if ((thflags & TH_ACK) &&
2285             (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2286             (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2287              (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2288                 /* keep track of pure window updates */
2289                 if (tlen == 0 &&
2290                     tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2291                         tcpstat.tcps_rcvwinupd++;
2292                 tp->snd_wnd = tiwin;
2293                 tp->snd_wl1 = th->th_seq;
2294                 tp->snd_wl2 = th->th_ack;
2295                 if (tp->snd_wnd > tp->max_sndwnd)
2296                         tp->max_sndwnd = tp->snd_wnd;
2297                 needoutput = 1;
2298         }
2299
2300         /*
2301          * Process segments with URG.
2302          */
2303         if ((thflags & TH_URG) && th->th_urp &&
2304             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2305                 /*
2306                  * This is a kludge, but if we receive and accept
2307                  * random urgent pointers, we'll crash in
2308                  * soreceive.  It's hard to imagine someone
2309                  * actually wanting to send this much urgent data.
2310                  */
2311                 SOCKBUF_LOCK(&so->so_rcv);
2312                 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2313                         th->th_urp = 0;                 /* XXX */
2314                         thflags &= ~TH_URG;             /* XXX */
2315                         SOCKBUF_UNLOCK(&so->so_rcv);    /* XXX */
2316                         goto dodata;                    /* XXX */
2317                 }
2318                 /*
2319                  * If this segment advances the known urgent pointer,
2320                  * then mark the data stream.  This should not happen
2321                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2322                  * a FIN has been received from the remote side.
2323                  * In these states we ignore the URG.
2324                  *
2325                  * According to RFC961 (Assigned Protocols),
2326                  * the urgent pointer points to the last octet
2327                  * of urgent data.  We continue, however,
2328                  * to consider it to indicate the first octet
2329                  * of data past the urgent section as the original
2330                  * spec states (in one of two places).
2331                  */
2332                 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2333                         tp->rcv_up = th->th_seq + th->th_urp;
2334                         so->so_oobmark = so->so_rcv.sb_cc +
2335                             (tp->rcv_up - tp->rcv_nxt) - 1;
2336                         if (so->so_oobmark == 0)
2337                                 so->so_rcv.sb_state |= SBS_RCVATMARK;
2338                         sohasoutofband(so);
2339                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2340                 }
2341                 SOCKBUF_UNLOCK(&so->so_rcv);
2342                 /*
2343                  * Remove out of band data so doesn't get presented to user.
2344                  * This can happen independent of advancing the URG pointer,
2345                  * but if two URG's are pending at once, some out-of-band
2346                  * data may creep in... ick.
2347                  */
2348                 if (th->th_urp <= (u_long)tlen &&
2349                     !(so->so_options & SO_OOBINLINE)) {
2350                         /* hdr drop is delayed */
2351                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2352                 }
2353         } else {
2354                 /*
2355                  * If no out of band data is expected,
2356                  * pull receive urgent pointer along
2357                  * with the receive window.
2358                  */
2359                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2360                         tp->rcv_up = tp->rcv_nxt;
2361         }
2362 dodata:                                                 /* XXX */
2363         KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2364         INP_LOCK_ASSERT(inp);
2365
2366         /*
2367          * Process the segment text, merging it into the TCP sequencing queue,
2368          * and arranging for acknowledgment of receipt if necessary.
2369          * This process logically involves adjusting tp->rcv_wnd as data
2370          * is presented to the user (this happens in tcp_usrreq.c,
2371          * case PRU_RCVD).  If a FIN has already been received on this
2372          * connection then we just ignore the text.
2373          */
2374         if ((tlen || (thflags & TH_FIN)) &&
2375             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2376                 tcp_seq save_start = th->th_seq;
2377                 tcp_seq save_end = th->th_seq + tlen;
2378                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2379                 /*
2380                  * Insert segment which includes th into TCP reassembly queue
2381                  * with control block tp.  Set thflags to whether reassembly now
2382                  * includes a segment with FIN.  This handles the common case
2383                  * inline (segment is the next to be received on an established
2384                  * connection, and the queue is empty), avoiding linkage into
2385                  * and removal from the queue and repetition of various
2386                  * conversions.
2387                  * Set DELACK for segments received in order, but ack
2388                  * immediately when segments are out of order (so
2389                  * fast retransmit can work).
2390                  */
2391                 if (th->th_seq == tp->rcv_nxt &&
2392                     LIST_EMPTY(&tp->t_segq) &&
2393                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2394                         if (DELAY_ACK(tp))
2395                                 tp->t_flags |= TF_DELACK;
2396                         else
2397                                 tp->t_flags |= TF_ACKNOW;
2398                         tp->rcv_nxt += tlen;
2399                         thflags = th->th_flags & TH_FIN;
2400                         tcpstat.tcps_rcvpack++;
2401                         tcpstat.tcps_rcvbyte += tlen;
2402                         ND6_HINT(tp);
2403                         SOCKBUF_LOCK(&so->so_rcv);
2404                         if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2405                                 m_freem(m);
2406                         else
2407                                 sbappendstream_locked(&so->so_rcv, m);
2408                         sorwakeup_locked(so);
2409                 } else {
2410                         thflags = tcp_reass(tp, th, &tlen, m);
2411                         tp->t_flags |= TF_ACKNOW;
2412                 }
2413                 if (tlen > 0 && tp->sack_enable)
2414                         tcp_update_sack_list(tp, save_start, save_end);
2415                 /*
2416                  * Note the amount of data that peer has sent into
2417                  * our window, in order to estimate the sender's
2418                  * buffer size.
2419                  */
2420                 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2421         } else {
2422                 m_freem(m);
2423                 thflags &= ~TH_FIN;
2424         }
2425
2426         /*
2427          * If FIN is received ACK the FIN and let the user know
2428          * that the connection is closing.
2429          */
2430         if (thflags & TH_FIN) {
2431                 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2432                         socantrcvmore(so);
2433                         /*
2434                          * If connection is half-synchronized
2435                          * (ie NEEDSYN flag on) then delay ACK,
2436                          * so it may be piggybacked when SYN is sent.
2437                          * Otherwise, since we received a FIN then no
2438                          * more input can be expected, send ACK now.
2439                          */
2440                         if (tp->t_flags & TF_NEEDSYN)
2441                                 tp->t_flags |= TF_DELACK;
2442                         else
2443                                 tp->t_flags |= TF_ACKNOW;
2444                         tp->rcv_nxt++;
2445                 }
2446                 switch (tp->t_state) {
2447
2448                 /*
2449                  * In SYN_RECEIVED and ESTABLISHED STATES
2450                  * enter the CLOSE_WAIT state.
2451                  */
2452                 case TCPS_SYN_RECEIVED:
2453                         tp->t_starttime = ticks;
2454                         /*FALLTHROUGH*/
2455                 case TCPS_ESTABLISHED:
2456                         tp->t_state = TCPS_CLOSE_WAIT;
2457                         break;
2458
2459                 /*
2460                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2461                  * enter the CLOSING state.
2462                  */
2463                 case TCPS_FIN_WAIT_1:
2464                         tp->t_state = TCPS_CLOSING;
2465                         break;
2466
2467                 /*
2468                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
2469                  * starting the time-wait timer, turning off the other
2470                  * standard timers.
2471                  */
2472                 case TCPS_FIN_WAIT_2:
2473                         KASSERT(headlocked == 1, ("tcp_input: dodata: "
2474                             "TCP_FIN_WAIT_2: head not locked"));
2475                         tcp_twstart(tp);
2476                         INP_INFO_WUNLOCK(&tcbinfo);
2477                         return;
2478
2479                 /*
2480                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
2481                  */
2482                 case TCPS_TIME_WAIT:
2483                         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2484                         callout_reset(tp->tt_2msl, 2 * tcp_msl,
2485                                       tcp_timer_2msl, tp);
2486                         break;
2487                 }
2488         }
2489         INP_INFO_WUNLOCK(&tcbinfo);
2490         headlocked = 0;
2491 #ifdef TCPDEBUG
2492         if (so->so_options & SO_DEBUG)
2493                 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2494                           &tcp_savetcp, 0);
2495 #endif
2496
2497         /*
2498          * Return any desired output.
2499          */
2500         if (needoutput || (tp->t_flags & TF_ACKNOW))
2501                 (void) tcp_output(tp);
2502
2503 check_delack:
2504         KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2505         INP_LOCK_ASSERT(inp);
2506         if (tp->t_flags & TF_DELACK) {
2507                 tp->t_flags &= ~TF_DELACK;
2508                 callout_reset(tp->tt_delack, tcp_delacktime,
2509                     tcp_timer_delack, tp);
2510         }
2511         INP_UNLOCK(inp);
2512         return;
2513
2514 dropafterack:
2515         KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2516         /*
2517          * Generate an ACK dropping incoming segment if it occupies
2518          * sequence space, where the ACK reflects our state.
2519          *
2520          * We can now skip the test for the RST flag since all
2521          * paths to this code happen after packets containing
2522          * RST have been dropped.
2523          *
2524          * In the SYN-RECEIVED state, don't send an ACK unless the
2525          * segment we received passes the SYN-RECEIVED ACK test.
2526          * If it fails send a RST.  This breaks the loop in the
2527          * "LAND" DoS attack, and also prevents an ACK storm
2528          * between two listening ports that have been sent forged
2529          * SYN segments, each with the source address of the other.
2530          */
2531         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2532             (SEQ_GT(tp->snd_una, th->th_ack) ||
2533              SEQ_GT(th->th_ack, tp->snd_max)) ) {
2534                 rstreason = BANDLIM_RST_OPENPORT;
2535                 goto dropwithreset;
2536         }
2537 #ifdef TCPDEBUG
2538         if (so->so_options & SO_DEBUG)
2539                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2540                           &tcp_savetcp, 0);
2541 #endif
2542         KASSERT(headlocked, ("headlocked should be 1"));
2543         INP_INFO_WUNLOCK(&tcbinfo);
2544         tp->t_flags |= TF_ACKNOW;
2545         (void) tcp_output(tp);
2546         INP_UNLOCK(inp);
2547         m_freem(m);
2548         return;
2549
2550 dropwithreset:
2551         KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2552         /*
2553          * Generate a RST, dropping incoming segment.
2554          * Make ACK acceptable to originator of segment.
2555          * Don't bother to respond if destination was broadcast/multicast.
2556          */
2557         if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2558                 goto drop;
2559         if (isipv6) {
2560                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2561                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2562                         goto drop;
2563         } else {
2564                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2565                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2566                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2567                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2568                         goto drop;
2569         }
2570         /* IPv6 anycast check is done at tcp6_input() */
2571
2572         /*
2573          * Perform bandwidth limiting.
2574          */
2575         if (badport_bandlim(rstreason) < 0)
2576                 goto drop;
2577
2578 #ifdef TCPDEBUG
2579         if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2580                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2581                           &tcp_savetcp, 0);
2582 #endif
2583
2584         if (thflags & TH_ACK)
2585                 /* mtod() below is safe as long as hdr dropping is delayed */
2586                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2587                             TH_RST);
2588         else {
2589                 if (thflags & TH_SYN)
2590                         tlen++;
2591                 /* mtod() below is safe as long as hdr dropping is delayed */
2592                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2593                             (tcp_seq)0, TH_RST|TH_ACK);
2594         }
2595
2596         if (tp != NULL)
2597                 INP_UNLOCK(inp);
2598         if (headlocked)
2599                 INP_INFO_WUNLOCK(&tcbinfo);
2600         return;
2601
2602 drop:
2603         /*
2604          * Drop space held by incoming segment and return.
2605          */
2606 #ifdef TCPDEBUG
2607         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2608                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2609                           &tcp_savetcp, 0);
2610 #endif
2611         if (tp != NULL)
2612                 INP_UNLOCK(inp);
2613         if (headlocked)
2614                 INP_INFO_WUNLOCK(&tcbinfo);
2615         m_freem(m);
2616         return;
2617 }
2618
2619 /*
2620  * Parse TCP options and place in tcpopt.
2621  */
2622 static void
2623 tcp_dooptions(to, cp, cnt, flags)
2624         struct tcpopt *to;
2625         u_char *cp;
2626         int cnt;
2627         int flags;
2628 {
2629         int opt, optlen;
2630
2631         to->to_flags = 0;
2632         for (; cnt > 0; cnt -= optlen, cp += optlen) {
2633                 opt = cp[0];
2634                 if (opt == TCPOPT_EOL)
2635                         break;
2636                 if (opt == TCPOPT_NOP)
2637                         optlen = 1;
2638                 else {
2639                         if (cnt < 2)
2640                                 break;
2641                         optlen = cp[1];
2642                         if (optlen < 2 || optlen > cnt)
2643                                 break;
2644                 }
2645                 switch (opt) {
2646                 case TCPOPT_MAXSEG:
2647                         if (optlen != TCPOLEN_MAXSEG)
2648                                 continue;
2649                         if (!(flags & TO_SYN))
2650                                 continue;
2651                         to->to_flags |= TOF_MSS;
2652                         bcopy((char *)cp + 2,
2653                             (char *)&to->to_mss, sizeof(to->to_mss));
2654                         to->to_mss = ntohs(to->to_mss);
2655                         break;
2656                 case TCPOPT_WINDOW:
2657                         if (optlen != TCPOLEN_WINDOW)
2658                                 continue;
2659                         if (!(flags & TO_SYN))
2660                                 continue;
2661                         to->to_flags |= TOF_SCALE;
2662                         to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2663                         break;
2664                 case TCPOPT_TIMESTAMP:
2665                         if (optlen != TCPOLEN_TIMESTAMP)
2666                                 continue;
2667                         to->to_flags |= TOF_TS;
2668                         bcopy((char *)cp + 2,
2669                             (char *)&to->to_tsval, sizeof(to->to_tsval));
2670                         to->to_tsval = ntohl(to->to_tsval);
2671                         bcopy((char *)cp + 6,
2672                             (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2673                         to->to_tsecr = ntohl(to->to_tsecr);
2674                         break;
2675 #ifdef TCP_SIGNATURE
2676                 /*
2677                  * XXX In order to reply to a host which has set the
2678                  * TCP_SIGNATURE option in its initial SYN, we have to
2679                  * record the fact that the option was observed here
2680                  * for the syncache code to perform the correct response.
2681                  */
2682                 case TCPOPT_SIGNATURE:
2683                         if (optlen != TCPOLEN_SIGNATURE)
2684                                 continue;
2685                         to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2686                         break;
2687 #endif
2688                 case TCPOPT_SACK_PERMITTED:
2689                         if (optlen != TCPOLEN_SACK_PERMITTED)
2690                                 continue;
2691                         if (!(flags & TO_SYN))
2692                                 continue;
2693                         if (!tcp_do_sack)
2694                                 continue;
2695                         to->to_flags |= TOF_SACK;
2696                         break;
2697                 case TCPOPT_SACK:
2698                         if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2699                                 continue;
2700                         to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2701                         to->to_sacks = cp + 2;
2702                         tcpstat.tcps_sack_rcv_blocks++;
2703                         break;
2704                 default:
2705                         continue;
2706                 }
2707         }
2708 }
2709
2710 /*
2711  * Pull out of band byte out of a segment so
2712  * it doesn't appear in the user's data queue.
2713  * It is still reflected in the segment length for
2714  * sequencing purposes.
2715  */
2716 static void
2717 tcp_pulloutofband(so, th, m, off)
2718         struct socket *so;
2719         struct tcphdr *th;
2720         register struct mbuf *m;
2721         int off;                /* delayed to be droped hdrlen */
2722 {
2723         int cnt = off + th->th_urp - 1;
2724
2725         while (cnt >= 0) {
2726                 if (m->m_len > cnt) {
2727                         char *cp = mtod(m, caddr_t) + cnt;
2728                         struct tcpcb *tp = sototcpcb(so);
2729
2730                         tp->t_iobc = *cp;
2731                         tp->t_oobflags |= TCPOOB_HAVEDATA;
2732                         bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2733                         m->m_len--;
2734                         if (m->m_flags & M_PKTHDR)
2735                                 m->m_pkthdr.len--;
2736                         return;
2737                 }
2738                 cnt -= m->m_len;
2739                 m = m->m_next;
2740                 if (m == 0)
2741                         break;
2742         }
2743         panic("tcp_pulloutofband");
2744 }
2745
2746 /*
2747  * Collect new round-trip time estimate
2748  * and update averages and current timeout.
2749  */
2750 static void
2751 tcp_xmit_timer(tp, rtt)
2752         register struct tcpcb *tp;
2753         int rtt;
2754 {
2755         register int delta;
2756
2757         INP_LOCK_ASSERT(tp->t_inpcb);
2758
2759         tcpstat.tcps_rttupdated++;
2760         tp->t_rttupdated++;
2761         if (tp->t_srtt != 0) {
2762                 /*
2763                  * srtt is stored as fixed point with 5 bits after the
2764                  * binary point (i.e., scaled by 8).  The following magic
2765                  * is equivalent to the smoothing algorithm in rfc793 with
2766                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2767                  * point).  Adjust rtt to origin 0.
2768                  */
2769                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2770                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2771
2772                 if ((tp->t_srtt += delta) <= 0)
2773                         tp->t_srtt = 1;
2774
2775                 /*
2776                  * We accumulate a smoothed rtt variance (actually, a
2777                  * smoothed mean difference), then set the retransmit
2778                  * timer to smoothed rtt + 4 times the smoothed variance.
2779                  * rttvar is stored as fixed point with 4 bits after the
2780                  * binary point (scaled by 16).  The following is
2781                  * equivalent to rfc793 smoothing with an alpha of .75
2782                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2783                  * rfc793's wired-in beta.
2784                  */
2785                 if (delta < 0)
2786                         delta = -delta;
2787                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2788                 if ((tp->t_rttvar += delta) <= 0)
2789                         tp->t_rttvar = 1;
2790                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2791                     tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2792         } else {
2793                 /*
2794                  * No rtt measurement yet - use the unsmoothed rtt.
2795                  * Set the variance to half the rtt (so our first
2796                  * retransmit happens at 3*rtt).
2797                  */
2798                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2799                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2800                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2801         }
2802         tp->t_rtttime = 0;
2803         tp->t_rxtshift = 0;
2804
2805         /*
2806          * the retransmit should happen at rtt + 4 * rttvar.
2807          * Because of the way we do the smoothing, srtt and rttvar
2808          * will each average +1/2 tick of bias.  When we compute
2809          * the retransmit timer, we want 1/2 tick of rounding and
2810          * 1 extra tick because of +-1/2 tick uncertainty in the
2811          * firing of the timer.  The bias will give us exactly the
2812          * 1.5 tick we need.  But, because the bias is
2813          * statistical, we have to test that we don't drop below
2814          * the minimum feasible timer (which is 2 ticks).
2815          */
2816         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2817                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2818
2819         /*
2820          * We received an ack for a packet that wasn't retransmitted;
2821          * it is probably safe to discard any error indications we've
2822          * received recently.  This isn't quite right, but close enough
2823          * for now (a route might have failed after we sent a segment,
2824          * and the return path might not be symmetrical).
2825          */
2826         tp->t_softerror = 0;
2827 }
2828
2829 /*
2830  * Determine a reasonable value for maxseg size.
2831  * If the route is known, check route for mtu.
2832  * If none, use an mss that can be handled on the outgoing
2833  * interface without forcing IP to fragment; if bigger than
2834  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2835  * to utilize large mbufs.  If no route is found, route has no mtu,
2836  * or the destination isn't local, use a default, hopefully conservative
2837  * size (usually 512 or the default IP max size, but no more than the mtu
2838  * of the interface), as we can't discover anything about intervening
2839  * gateways or networks.  We also initialize the congestion/slow start
2840  * window to be a single segment if the destination isn't local.
2841  * While looking at the routing entry, we also initialize other path-dependent
2842  * parameters from pre-set or cached values in the routing entry.
2843  *
2844  * Also take into account the space needed for options that we
2845  * send regularly.  Make maxseg shorter by that amount to assure
2846  * that we can send maxseg amount of data even when the options
2847  * are present.  Store the upper limit of the length of options plus
2848  * data in maxopd.
2849  *
2850  *
2851  * In case of T/TCP, we call this routine during implicit connection
2852  * setup as well (offer = -1), to initialize maxseg from the cached
2853  * MSS of our peer.
2854  *
2855  * NOTE that this routine is only called when we process an incoming
2856  * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2857  */
2858 void
2859 tcp_mss(tp, offer)
2860         struct tcpcb *tp;
2861         int offer;
2862 {
2863         int rtt, mss;
2864         u_long bufsize;
2865         u_long maxmtu;
2866         struct inpcb *inp = tp->t_inpcb;
2867         struct socket *so;
2868         struct hc_metrics_lite metrics;
2869         int origoffer = offer;
2870         int mtuflags = 0;
2871 #ifdef INET6
2872         int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2873         size_t min_protoh = isipv6 ?
2874                             sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2875                             sizeof (struct tcpiphdr);
2876 #else
2877         const size_t min_protoh = sizeof(struct tcpiphdr);
2878 #endif
2879
2880         /* initialize */
2881 #ifdef INET6
2882         if (isipv6) {
2883                 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags);
2884                 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2885         } else
2886 #endif
2887         {
2888                 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags);
2889                 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2890         }
2891         so = inp->inp_socket;
2892
2893         /*
2894          * no route to sender, stay with default mss and return
2895          */
2896         if (maxmtu == 0)
2897                 return;
2898
2899         /* what have we got? */
2900         switch (offer) {
2901                 case 0:
2902                         /*
2903                          * Offer == 0 means that there was no MSS on the SYN
2904                          * segment, in this case we use tcp_mssdflt.
2905                          */
2906                         offer =
2907 #ifdef INET6
2908                                 isipv6 ? tcp_v6mssdflt :
2909 #endif
2910                                 tcp_mssdflt;
2911                         break;
2912
2913                 case -1:
2914                         /*
2915                          * Offer == -1 means that we didn't receive SYN yet.
2916                          */
2917                         /* FALLTHROUGH */
2918
2919                 default:
2920                         /*
2921                          * Prevent DoS attack with too small MSS. Round up
2922                          * to at least minmss.
2923                          */
2924                         offer = max(offer, tcp_minmss);
2925                         /*
2926                          * Sanity check: make sure that maxopd will be large
2927                          * enough to allow some data on segments even if the
2928                          * all the option space is used (40bytes).  Otherwise
2929                          * funny things may happen in tcp_output.
2930                          */
2931                         offer = max(offer, 64);
2932         }
2933
2934         /*
2935          * rmx information is now retrieved from tcp_hostcache
2936          */
2937         tcp_hc_get(&inp->inp_inc, &metrics);
2938
2939         /*
2940          * if there's a discovered mtu int tcp hostcache, use it
2941          * else, use the link mtu.
2942          */
2943         if (metrics.rmx_mtu)
2944                 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2945         else {
2946 #ifdef INET6
2947                 if (isipv6) {
2948                         mss = maxmtu - min_protoh;
2949                         if (!path_mtu_discovery &&
2950                             !in6_localaddr(&inp->in6p_faddr))
2951                                 mss = min(mss, tcp_v6mssdflt);
2952                 } else
2953 #endif
2954                 {
2955                         mss = maxmtu - min_protoh;
2956                         if (!path_mtu_discovery &&
2957                             !in_localaddr(inp->inp_faddr))
2958                                 mss = min(mss, tcp_mssdflt);
2959                 }
2960         }
2961         mss = min(mss, offer);
2962
2963         /*
2964          * maxopd stores the maximum length of data AND options
2965          * in a segment; maxseg is the amount of data in a normal
2966          * segment.  We need to store this value (maxopd) apart
2967          * from maxseg, because now every segment carries options
2968          * and thus we normally have somewhat less data in segments.
2969          */
2970         tp->t_maxopd = mss;
2971
2972         /*
2973          * origoffer==-1 indicates, that no segments were received yet.
2974          * In this case we just guess.
2975          */
2976         if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2977             (origoffer == -1 ||
2978              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2979                 mss -= TCPOLEN_TSTAMP_APPA;
2980         tp->t_maxseg = mss;
2981
2982 #if     (MCLBYTES & (MCLBYTES - 1)) == 0
2983                 if (mss > MCLBYTES)
2984                         mss &= ~(MCLBYTES-1);
2985 #else
2986                 if (mss > MCLBYTES)
2987                         mss = mss / MCLBYTES * MCLBYTES;
2988 #endif
2989         tp->t_maxseg = mss;
2990
2991         /*
2992          * If there's a pipesize, change the socket buffer to that size,
2993          * don't change if sb_hiwat is different than default (then it
2994          * has been changed on purpose with setsockopt).
2995          * Make the socket buffers an integral number of mss units;
2996          * if the mss is larger than the socket buffer, decrease the mss.
2997          */
2998         SOCKBUF_LOCK(&so->so_snd);
2999         if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3000                 bufsize = metrics.rmx_sendpipe;
3001         else
3002                 bufsize = so->so_snd.sb_hiwat;
3003         if (bufsize < mss)
3004                 mss = bufsize;
3005         else {
3006                 bufsize = roundup(bufsize, mss);
3007                 if (bufsize > sb_max)
3008                         bufsize = sb_max;
3009                 if (bufsize > so->so_snd.sb_hiwat)
3010                         (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3011         }
3012         SOCKBUF_UNLOCK(&so->so_snd);
3013         tp->t_maxseg = mss;
3014
3015         SOCKBUF_LOCK(&so->so_rcv);
3016         if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3017                 bufsize = metrics.rmx_recvpipe;
3018         else
3019                 bufsize = so->so_rcv.sb_hiwat;
3020         if (bufsize > mss) {
3021                 bufsize = roundup(bufsize, mss);
3022                 if (bufsize > sb_max)
3023                         bufsize = sb_max;
3024                 if (bufsize > so->so_rcv.sb_hiwat)
3025                         (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3026         }
3027         SOCKBUF_UNLOCK(&so->so_rcv);
3028         /*
3029          * While we're here, check the others too
3030          */
3031         if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3032                 tp->t_srtt = rtt;
3033                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3034                 tcpstat.tcps_usedrtt++;
3035                 if (metrics.rmx_rttvar) {
3036                         tp->t_rttvar = metrics.rmx_rttvar;
3037                         tcpstat.tcps_usedrttvar++;
3038                 } else {
3039                         /* default variation is +- 1 rtt */
3040                         tp->t_rttvar =
3041                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3042                 }
3043                 TCPT_RANGESET(tp->t_rxtcur,
3044                               ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3045                               tp->t_rttmin, TCPTV_REXMTMAX);
3046         }
3047         if (metrics.rmx_ssthresh) {
3048                 /*
3049                  * There's some sort of gateway or interface
3050                  * buffer limit on the path.  Use this to set
3051                  * the slow start threshhold, but set the
3052                  * threshold to no less than 2*mss.
3053                  */
3054                 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3055                 tcpstat.tcps_usedssthresh++;
3056         }
3057         if (metrics.rmx_bandwidth)
3058                 tp->snd_bandwidth = metrics.rmx_bandwidth;
3059
3060         /*
3061          * Set the slow-start flight size depending on whether this
3062          * is a local network or not.
3063          *
3064          * Extend this so we cache the cwnd too and retrieve it here.
3065          * Make cwnd even bigger than RFC3390 suggests but only if we
3066          * have previous experience with the remote host. Be careful
3067          * not make cwnd bigger than remote receive window or our own
3068          * send socket buffer. Maybe put some additional upper bound
3069          * on the retrieved cwnd. Should do incremental updates to
3070          * hostcache when cwnd collapses so next connection doesn't
3071          * overloads the path again.
3072          *
3073          * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3074          * We currently check only in syncache_socket for that.
3075          */
3076 #define TCP_METRICS_CWND
3077 #ifdef TCP_METRICS_CWND
3078         if (metrics.rmx_cwnd)
3079                 tp->snd_cwnd = max(mss,
3080                                 min(metrics.rmx_cwnd / 2,
3081                                  min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3082         else
3083 #endif
3084         if (tcp_do_rfc3390)
3085                 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3086 #ifdef INET6
3087         else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3088                  (!isipv6 && in_localaddr(inp->inp_faddr)))
3089 #else
3090         else if (in_localaddr(inp->inp_faddr))
3091 #endif
3092                 tp->snd_cwnd = mss * ss_fltsz_local;
3093         else
3094                 tp->snd_cwnd = mss * ss_fltsz;
3095
3096         /* Check the interface for TSO capabilities. */
3097         if (mtuflags & CSUM_TSO)
3098                 tp->t_flags |= TF_TSO;
3099 }
3100
3101 /*
3102  * Determine the MSS option to send on an outgoing SYN.
3103  */
3104 int
3105 tcp_mssopt(inc)
3106         struct in_conninfo *inc;
3107 {
3108         int mss = 0;
3109         u_long maxmtu = 0;
3110         u_long thcmtu = 0;
3111         size_t min_protoh;
3112 #ifdef INET6
3113         int isipv6 = inc->inc_isipv6 ? 1 : 0;
3114 #endif
3115
3116         KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3117
3118 #ifdef INET6
3119         if (isipv6) {
3120                 mss = tcp_v6mssdflt;
3121                 maxmtu = tcp_maxmtu6(inc, NULL);
3122                 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3123                 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3124         } else
3125 #endif
3126         {
3127                 mss = tcp_mssdflt;
3128                 maxmtu = tcp_maxmtu(inc, NULL);
3129                 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3130                 min_protoh = sizeof(struct tcpiphdr);
3131         }
3132         if (maxmtu && thcmtu)
3133                 mss = min(maxmtu, thcmtu) - min_protoh;
3134         else if (maxmtu || thcmtu)
3135                 mss = max(maxmtu, thcmtu) - min_protoh;
3136
3137         return (mss);
3138 }
3139
3140
3141 /*
3142  * On a partial ack arrives, force the retransmission of the
3143  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3144  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3145  * be started again.
3146  */
3147 static void
3148 tcp_newreno_partial_ack(tp, th)
3149         struct tcpcb *tp;
3150         struct tcphdr *th;
3151 {
3152         tcp_seq onxt = tp->snd_nxt;
3153         u_long  ocwnd = tp->snd_cwnd;
3154
3155         callout_stop(tp->tt_rexmt);
3156         tp->t_rtttime = 0;
3157         tp->snd_nxt = th->th_ack;
3158         /*
3159          * Set snd_cwnd to one segment beyond acknowledged offset.
3160          * (tp->snd_una has not yet been updated when this function is called.)
3161          */
3162         tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3163         tp->t_flags |= TF_ACKNOW;
3164         (void) tcp_output(tp);
3165         tp->snd_cwnd = ocwnd;
3166         if (SEQ_GT(onxt, tp->snd_nxt))
3167                 tp->snd_nxt = onxt;
3168         /*
3169          * Partial window deflation.  Relies on fact that tp->snd_una
3170          * not updated yet.
3171          */
3172         if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3173                 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3174         else
3175                 tp->snd_cwnd = 0;
3176         tp->snd_cwnd += tp->t_maxseg;
3177 }
3178
3179 /*
3180  * Returns 1 if the TIME_WAIT state was killed and we should start over,
3181  * looking for a pcb in the listen state.  Returns 0 otherwise.
3182  */
3183 static int
3184 tcp_timewait(inp, to, th, m, tlen)
3185         struct inpcb *inp;
3186         struct tcpopt *to;
3187         struct tcphdr *th;
3188         struct mbuf *m;
3189         int tlen;
3190 {
3191         struct tcptw *tw;
3192         int thflags;
3193         tcp_seq seq;
3194 #ifdef INET6
3195         int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3196 #else
3197         const int isipv6 = 0;
3198 #endif
3199
3200         /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */
3201         INP_INFO_WLOCK_ASSERT(&tcbinfo);
3202         INP_LOCK_ASSERT(inp);
3203
3204         /*
3205          * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
3206          * still present.  This is undesirable, but temporarily necessary
3207          * until we work out how to handle inpcb's who's timewait state has
3208          * been removed.
3209          */
3210         tw = intotw(inp);
3211         if (tw == NULL)
3212                 goto drop;
3213
3214         thflags = th->th_flags;
3215
3216         /*
3217          * NOTE: for FIN_WAIT_2 (to be added later),
3218          * must validate sequence number before accepting RST
3219          */
3220
3221         /*
3222          * If the segment contains RST:
3223          *      Drop the segment - see Stevens, vol. 2, p. 964 and
3224          *      RFC 1337.
3225          */
3226         if (thflags & TH_RST)
3227                 goto drop;
3228
3229 #if 0
3230 /* PAWS not needed at the moment */
3231         /*
3232          * RFC 1323 PAWS: If we have a timestamp reply on this segment
3233          * and it's less than ts_recent, drop it.
3234          */
3235         if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3236             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3237                 if ((thflags & TH_ACK) == 0)
3238                         goto drop;
3239                 goto ack;
3240         }
3241         /*
3242          * ts_recent is never updated because we never accept new segments.
3243          */
3244 #endif
3245
3246         /*
3247          * If a new connection request is received
3248          * while in TIME_WAIT, drop the old connection
3249          * and start over if the sequence numbers
3250          * are above the previous ones.
3251          */
3252         if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3253                 tcp_twclose(tw, 0);
3254                 return (1);
3255         }
3256
3257         /*
3258          * Drop the the segment if it does not contain an ACK.
3259          */
3260         if ((thflags & TH_ACK) == 0)
3261                 goto drop;
3262
3263         /*
3264          * Reset the 2MSL timer if this is a duplicate FIN.
3265          */
3266         if (thflags & TH_FIN) {
3267                 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3268                 if (seq + 1 == tw->rcv_nxt)
3269                         tcp_timer_2msl_reset(tw, 1);
3270         }
3271
3272         /*
3273          * Acknowledge the segment if it has data or is not a duplicate ACK.
3274          */
3275         if (thflags != TH_ACK || tlen != 0 ||
3276             th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3277                 tcp_twrespond(tw, TH_ACK);
3278         goto drop;
3279
3280         /*
3281          * Generate a RST, dropping incoming segment.
3282          * Make ACK acceptable to originator of segment.
3283          * Don't bother to respond if destination was broadcast/multicast.
3284          */
3285         if (m->m_flags & (M_BCAST|M_MCAST))
3286                 goto drop;
3287         if (isipv6) {
3288                 struct ip6_hdr *ip6;
3289
3290                 /* IPv6 anycast check is done at tcp6_input() */
3291                 ip6 = mtod(m, struct ip6_hdr *);
3292                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3293                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3294                         goto drop;
3295         } else {
3296                 struct ip *ip;
3297
3298                 ip = mtod(m, struct ip *);
3299                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3300                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3301                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3302                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3303                         goto drop;
3304         }
3305         if (thflags & TH_ACK) {
3306                 tcp_respond(NULL,
3307                     mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3308         } else {
3309                 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3310                 tcp_respond(NULL,
3311                     mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3312         }
3313         INP_UNLOCK(inp);
3314         return (0);
3315
3316 drop:
3317         INP_UNLOCK(inp);
3318         m_freem(m);
3319         return (0);
3320 }