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  * Copyright (c) 2007-2008,2010
   5  *      Swinburne University of Technology, Melbourne, Australia.
   6  * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
   7  * Copyright (c) 2010 The FreeBSD Foundation
   8  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed at the Centre for Advanced Internet
  12  * Architectures, Swinburne University of Technology, by Lawrence Stewart,
  13  * James Healy and David Hayes, made possible in part by a grant from the Cisco
  14  * University Research Program Fund at Community Foundation Silicon Valley.
  15  *
  16  * Portions of this software were developed at the Centre for Advanced
  17  * Internet Architectures, Swinburne University of Technology, Melbourne,
  18  * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
  19  *
  20  * Portions of this software were developed by Robert N. M. Watson under
  21  * contract to Juniper Networks, Inc.
  22  *
  23  * Redistribution and use in source and binary forms, with or without
  24  * modification, are permitted provided that the following conditions
  25  * are met:
  26  * 1. Redistributions of source code must retain the above copyright
  27  *    notice, this list of conditions and the following disclaimer.
  28  * 2. Redistributions in binary form must reproduce the above copyright
  29  *    notice, this list of conditions and the following disclaimer in the
  30  *    documentation and/or other materials provided with the distribution.
  31  * 4. Neither the name of the University nor the names of its contributors
  32  *    may be used to endorse or promote products derived from this software
  33  *    without specific prior written permission.
  34  *
  35  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  36  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  37  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  38  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  39  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  40  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  41  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  42  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  43  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  44  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  45  * SUCH DAMAGE.
  46  *
  47  *      @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
  48  */
  49
  50 #include <sys/cdefs.h>
  51 __FBSDID("$FreeBSD$");
  52
  53 #include "opt_ipfw.h"           /* for ipfw_fwd */
  54 #include "opt_inet.h"
  55 #include "opt_inet6.h"
  56 #include "opt_ipsec.h"
  57 #include "opt_tcpdebug.h"
  58
  59 #include <sys/param.h>
  60 #include <sys/kernel.h>
  61 #include <sys/hhook.h>
  62 #include <sys/malloc.h>
  63 #include <sys/mbuf.h>
  64 #include <sys/proc.h>           /* for proc0 declaration */
  65 #include <sys/protosw.h>
  66 #include <sys/sdt.h>
  67 #include <sys/signalvar.h>
  68 #include <sys/socket.h>
  69 #include <sys/socketvar.h>
  70 #include <sys/sysctl.h>
  71 #include <sys/syslog.h>
  72 #include <sys/systm.h>
  73
  74 #include <machine/cpu.h>        /* before tcp_seq.h, for tcp_random18() */
  75
  76 #include <vm/uma.h>
  77
  78 #include <net/if.h>
  79 #include <net/if_var.h>
  80 #include <net/route.h>
  81 #include <net/vnet.h>
  82
  83 #define TCPSTATES               /* for logging */
  84
  85 #include <netinet/cc.h>
  86 #include <netinet/in.h>
  87 #include <netinet/in_kdtrace.h>
  88 #include <netinet/in_pcb.h>
  89 #include <netinet/in_systm.h>
  90 #include <netinet/in_var.h>
  91 #include <netinet/ip.h>
  92 #include <netinet/ip_icmp.h>    /* required for icmp_var.h */
  93 #include <netinet/icmp_var.h>   /* for ICMP_BANDLIM */
  94 #include <netinet/ip_var.h>
  95 #include <netinet/ip_options.h>
  96 #include <netinet/ip6.h>
  97 #include <netinet/icmp6.h>
  98 #include <netinet6/in6_pcb.h>
  99 #include <netinet6/ip6_var.h>
 100 #include <netinet6/nd6.h>
 101 #include <netinet/tcp_fsm.h>
 102 #include <netinet/tcp_seq.h>
 103 #include <netinet/tcp_timer.h>
 104 #include <netinet/tcp_var.h>
 105 #include <netinet6/tcp6_var.h>
 106 #include <netinet/tcpip.h>
 107 #include <netinet/tcp_syncache.h>
 108 #ifdef TCPDEBUG
 109 #include <netinet/tcp_debug.h>
 110 #endif /* TCPDEBUG */
 111 #ifdef TCP_OFFLOAD
 112 #include <netinet/tcp_offload.h>
 113 #endif
 114
 115 #ifdef IPSEC
 116 #include <netipsec/ipsec.h>
 117 #include <netipsec/ipsec6.h>
 118 #endif /*IPSEC*/
 119
 120 #include <machine/in_cksum.h>
 121
 122 #include <security/mac/mac_framework.h>
 123
 124 const int tcprexmtthresh = 3;
 125
 126 int tcp_log_in_vain = 0;
 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
 128     &tcp_log_in_vain, 0,
 129     "Log all incoming TCP segments to closed ports");
 130
 131 VNET_DEFINE(int, blackhole) = 0;
 132 #define V_blackhole             VNET(blackhole)
 133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
 134     &VNET_NAME(blackhole), 0,
 135     "Do not send RST on segments to closed ports");
 136
 137 VNET_DEFINE(int, tcp_delack_enabled) = 1;
 138 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
 139     &VNET_NAME(tcp_delack_enabled), 0,
 140     "Delay ACK to try and piggyback it onto a data packet");
 141
 142 VNET_DEFINE(int, drop_synfin) = 0;
 143 #define V_drop_synfin           VNET(drop_synfin)
 144 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
 145     &VNET_NAME(drop_synfin), 0,
 146     "Drop TCP packets with SYN+FIN set");
 147
 148 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
 149 #define V_tcp_do_rfc3042        VNET(tcp_do_rfc3042)
 150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
 151     &VNET_NAME(tcp_do_rfc3042), 0,
 152     "Enable RFC 3042 (Limited Transmit)");
 153
 154 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
 155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
 156     &VNET_NAME(tcp_do_rfc3390), 0,
 157     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
 158
 159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
 160     "Experimental TCP extensions");
 161
 162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1;
 163 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
 164     &VNET_NAME(tcp_do_initcwnd10), 0,
 165     "Enable RFC 6928 (Increasing initial CWND to 10)");
 166
 167 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
 168 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
 169     &VNET_NAME(tcp_do_rfc3465), 0,
 170     "Enable RFC 3465 (Appropriate Byte Counting)");
 171
 172 VNET_DEFINE(int, tcp_abc_l_var) = 2;
 173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
 174     &VNET_NAME(tcp_abc_l_var), 2,
 175     "Cap the max cwnd increment during slow-start to this number of segments");
 176
 177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
 178
 179 VNET_DEFINE(int, tcp_do_ecn) = 0;
 180 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
 181     &VNET_NAME(tcp_do_ecn), 0,
 182     "TCP ECN support");
 183
 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
 185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
 186     &VNET_NAME(tcp_ecn_maxretries), 0,
 187     "Max retries before giving up on ECN");
 188
 189 VNET_DEFINE(int, tcp_insecure_rst) = 0;
 190 #define V_tcp_insecure_rst      VNET(tcp_insecure_rst)
 191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
 192     &VNET_NAME(tcp_insecure_rst), 0,
 193     "Follow the old (insecure) criteria for accepting RST packets");
 194
 195 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
 196 #define V_tcp_recvspace VNET(tcp_recvspace)
 197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
 198     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
 199
 200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
 201 #define V_tcp_do_autorcvbuf     VNET(tcp_do_autorcvbuf)
 202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
 203     &VNET_NAME(tcp_do_autorcvbuf), 0,
 204     "Enable automatic receive buffer sizing");
 205
 206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
 207 #define V_tcp_autorcvbuf_inc    VNET(tcp_autorcvbuf_inc)
 208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
 209     &VNET_NAME(tcp_autorcvbuf_inc), 0,
 210     "Incrementor step size of automatic receive buffer");
 211
 212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
 213 #define V_tcp_autorcvbuf_max    VNET(tcp_autorcvbuf_max)
 214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
 215     &VNET_NAME(tcp_autorcvbuf_max), 0,
 216     "Max size of automatic receive buffer");
 217
 218 VNET_DEFINE(struct inpcbhead, tcb);
 219 #define tcb6    tcb  /* for KAME src sync over BSD*'s */
 220 VNET_DEFINE(struct inpcbinfo, tcbinfo);
 221
 222 static void      tcp_dooptions(struct tcpopt *, u_char *, int, int);
 223 static void      tcp_do_segment(struct mbuf *, struct tcphdr *,
 224                      struct socket *, struct tcpcb *, int, int, uint8_t,
 225                      int);
 226 static void      tcp_dropwithreset(struct mbuf *, struct tcphdr *,
 227                      struct tcpcb *, int, int);
 228 static void      tcp_pulloutofband(struct socket *,
 229                      struct tcphdr *, struct mbuf *, int);
 230 static void      tcp_xmit_timer(struct tcpcb *, int);
 231 static void      tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
 232 static void inline      tcp_fields_to_host(struct tcphdr *);
 233 #ifdef TCP_SIGNATURE
 234 static void inline      tcp_fields_to_net(struct tcphdr *);
 235 static int inline       tcp_signature_verify_input(struct mbuf *, int, int,
 236                             int, struct tcpopt *, struct tcphdr *, u_int);
 237 #endif
 238 static void inline      cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
 239                             uint16_t type);
 240 static void inline      cc_conn_init(struct tcpcb *tp);
 241 static void inline      cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
 242 static void inline      hhook_run_tcp_est_in(struct tcpcb *tp,
 243                             struct tcphdr *th, struct tcpopt *to);
 244
 245 /*
 246  * TCP statistics are stored in an "array" of counter(9)s.
 247  */
 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
 249 VNET_PCPUSTAT_SYSINIT(tcpstat);
 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
 251     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
 252
 253 #ifdef VIMAGE
 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
 255 #endif /* VIMAGE */
 256 /*
 257  * Kernel module interface for updating tcpstat.  The argument is an index
 258  * into tcpstat treated as an array.
 259  */
 260 void
 261 kmod_tcpstat_inc(int statnum)
 262 {
 263
 264         counter_u64_add(VNET(tcpstat)[statnum], 1);
 265 }
 266
 267 /*
 268  * Wrapper for the TCP established input helper hook.
 269  */
 270 static void inline
 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
 272 {
 273         struct tcp_hhook_data hhook_data;
 274
 275         if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
 276                 hhook_data.tp = tp;
 277                 hhook_data.th = th;
 278                 hhook_data.to = to;
 279
 280                 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
 281                     tp->osd);
 282         }
 283 }
 284
 285 /*
 286  * CC wrapper hook functions
 287  */
 288 static void inline
 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
 290 {
 291         INP_WLOCK_ASSERT(tp->t_inpcb);
 292
 293         tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
 294         if (tp->snd_cwnd <= tp->snd_wnd)
 295                 tp->ccv->flags |= CCF_CWND_LIMITED;
 296         else
 297                 tp->ccv->flags &= ~CCF_CWND_LIMITED;
 298
 299         if (type == CC_ACK) {
 300                 if (tp->snd_cwnd > tp->snd_ssthresh) {
 301                         tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
 302                              V_tcp_abc_l_var * tp->t_maxseg);
 303                         if (tp->t_bytes_acked >= tp->snd_cwnd) {
 304                                 tp->t_bytes_acked -= tp->snd_cwnd;
 305                                 tp->ccv->flags |= CCF_ABC_SENTAWND;
 306                         }
 307                 } else {
 308                                 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
 309                                 tp->t_bytes_acked = 0;
 310                 }
 311         }
 312
 313         if (CC_ALGO(tp)->ack_received != NULL) {
 314                 /* XXXLAS: Find a way to live without this */
 315                 tp->ccv->curack = th->th_ack;
 316                 CC_ALGO(tp)->ack_received(tp->ccv, type);
 317         }
 318 }
 319
 320 static void inline
 321 cc_conn_init(struct tcpcb *tp)
 322 {
 323         struct hc_metrics_lite metrics;
 324         struct inpcb *inp = tp->t_inpcb;
 325         int rtt;
 326
 327         INP_WLOCK_ASSERT(tp->t_inpcb);
 328
 329         tcp_hc_get(&inp->inp_inc, &metrics);
 330
 331         if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
 332                 tp->t_srtt = rtt;
 333                 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
 334                 TCPSTAT_INC(tcps_usedrtt);
 335                 if (metrics.rmx_rttvar) {
 336                         tp->t_rttvar = metrics.rmx_rttvar;
 337                         TCPSTAT_INC(tcps_usedrttvar);
 338                 } else {
 339                         /* default variation is +- 1 rtt */
 340                         tp->t_rttvar =
 341                             tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
 342                 }
 343                 TCPT_RANGESET(tp->t_rxtcur,
 344                     ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
 345                     tp->t_rttmin, TCPTV_REXMTMAX);
 346         }
 347         if (metrics.rmx_ssthresh) {
 348                 /*
 349                  * There's some sort of gateway or interface
 350                  * buffer limit on the path.  Use this to set
 351                  * the slow start threshhold, but set the
 352                  * threshold to no less than 2*mss.
 353                  */
 354                 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
 355                 TCPSTAT_INC(tcps_usedssthresh);
 356         }
 357
 358         /*
 359          * Set the initial slow-start flight size.
 360          *
 361          * RFC5681 Section 3.1 specifies the default conservative values.
 362          * RFC3390 specifies slightly more aggressive values.
 363          * RFC6928 increases it to ten segments.
 364          *
 365          * If a SYN or SYN/ACK was lost and retransmitted, we have to
 366          * reduce the initial CWND to one segment as congestion is likely
 367          * requiring us to be cautious.
 368          */
 369         if (tp->snd_cwnd == 1)
 370                 tp->snd_cwnd = tp->t_maxseg;            /* SYN(-ACK) lost */
 371         else if (V_tcp_do_initcwnd10)
 372                 tp->snd_cwnd = min(10 * tp->t_maxseg,
 373                     max(2 * tp->t_maxseg, 14600));
 374         else if (V_tcp_do_rfc3390)
 375                 tp->snd_cwnd = min(4 * tp->t_maxseg,
 376                     max(2 * tp->t_maxseg, 4380));
 377         else {
 378                 /* Per RFC5681 Section 3.1 */
 379                 if (tp->t_maxseg > 2190)
 380                         tp->snd_cwnd = 2 * tp->t_maxseg;
 381                 else if (tp->t_maxseg > 1095)
 382                         tp->snd_cwnd = 3 * tp->t_maxseg;
 383                 else
 384                         tp->snd_cwnd = 4 * tp->t_maxseg;
 385         }
 386
 387         if (CC_ALGO(tp)->conn_init != NULL)
 388                 CC_ALGO(tp)->conn_init(tp->ccv);
 389 }
 390
 391 void inline
 392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
 393 {
 394         INP_WLOCK_ASSERT(tp->t_inpcb);
 395
 396         switch(type) {
 397         case CC_NDUPACK:
 398                 if (!IN_FASTRECOVERY(tp->t_flags)) {
 399                         tp->snd_recover = tp->snd_max;
 400                         if (tp->t_flags & TF_ECN_PERMIT)
 401                                 tp->t_flags |= TF_ECN_SND_CWR;
 402                 }
 403                 break;
 404         case CC_ECN:
 405                 if (!IN_CONGRECOVERY(tp->t_flags)) {
 406                         TCPSTAT_INC(tcps_ecn_rcwnd);
 407                         tp->snd_recover = tp->snd_max;
 408                         if (tp->t_flags & TF_ECN_PERMIT)
 409                                 tp->t_flags |= TF_ECN_SND_CWR;
 410                 }
 411                 break;
 412         case CC_RTO:
 413                 tp->t_dupacks = 0;
 414                 tp->t_bytes_acked = 0;
 415                 EXIT_RECOVERY(tp->t_flags);
 416                 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
 417                     tp->t_maxseg) * tp->t_maxseg;
 418                 tp->snd_cwnd = tp->t_maxseg;
 419                 break;
 420         case CC_RTO_ERR:
 421                 TCPSTAT_INC(tcps_sndrexmitbad);
 422                 /* RTO was unnecessary, so reset everything. */
 423                 tp->snd_cwnd = tp->snd_cwnd_prev;
 424                 tp->snd_ssthresh = tp->snd_ssthresh_prev;
 425                 tp->snd_recover = tp->snd_recover_prev;
 426                 if (tp->t_flags & TF_WASFRECOVERY)
 427                         ENTER_FASTRECOVERY(tp->t_flags);
 428                 if (tp->t_flags & TF_WASCRECOVERY)
 429                         ENTER_CONGRECOVERY(tp->t_flags);
 430                 tp->snd_nxt = tp->snd_max;
 431                 tp->t_flags &= ~TF_PREVVALID;
 432                 tp->t_badrxtwin = 0;
 433                 break;
 434         }
 435
 436         if (CC_ALGO(tp)->cong_signal != NULL) {
 437                 if (th != NULL)
 438                         tp->ccv->curack = th->th_ack;
 439                 CC_ALGO(tp)->cong_signal(tp->ccv, type);
 440         }
 441 }
 442
 443 static void inline
 444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
 445 {
 446         INP_WLOCK_ASSERT(tp->t_inpcb);
 447
 448         /* XXXLAS: KASSERT that we're in recovery? */
 449
 450         if (CC_ALGO(tp)->post_recovery != NULL) {
 451                 tp->ccv->curack = th->th_ack;
 452                 CC_ALGO(tp)->post_recovery(tp->ccv);
 453         }
 454         /* XXXLAS: EXIT_RECOVERY ? */
 455         tp->t_bytes_acked = 0;
 456 }
 457
 458 static inline void
 459 tcp_fields_to_host(struct tcphdr *th)
 460 {
 461
 462         th->th_seq = ntohl(th->th_seq);
 463         th->th_ack = ntohl(th->th_ack);
 464         th->th_win = ntohs(th->th_win);
 465         th->th_urp = ntohs(th->th_urp);
 466 }
 467
 468 #ifdef TCP_SIGNATURE
 469 static inline void
 470 tcp_fields_to_net(struct tcphdr *th)
 471 {
 472
 473         th->th_seq = htonl(th->th_seq);
 474         th->th_ack = htonl(th->th_ack);
 475         th->th_win = htons(th->th_win);
 476         th->th_urp = htons(th->th_urp);
 477 }
 478
 479 static inline int
 480 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
 481     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
 482 {
 483         int ret;
 484
 485         tcp_fields_to_net(th);
 486         ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
 487         tcp_fields_to_host(th);
 488         return (ret);
 489 }
 490 #endif
 491
 492 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
 493 #ifdef INET6
 494 #define ND6_HINT(tp) \
 495 do { \
 496         if ((tp) && (tp)->t_inpcb && \
 497             ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
 498                 nd6_nud_hint(NULL, NULL, 0); \
 499 } while (0)
 500 #else
 501 #define ND6_HINT(tp)
 502 #endif
 503
 504 /*
 505  * Indicate whether this ack should be delayed.  We can delay the ack if
 506  * following conditions are met:
 507  *      - There is no delayed ack timer in progress.
 508  *      - Our last ack wasn't a 0-sized window. We never want to delay
 509  *        the ack that opens up a 0-sized window.
 510  *      - LRO wasn't used for this segment. We make sure by checking that the
 511  *        segment size is not larger than the MSS.
 512  *      - Delayed acks are enabled or this is a half-synchronized T/TCP
 513  *        connection.
 514  */
 515 #define DELAY_ACK(tp, tlen)                                             \
 516         ((!tcp_timer_active(tp, TT_DELACK) &&                           \
 517             (tp->t_flags & TF_RXWIN0SENT) == 0) &&                      \
 518             (tlen <= tp->t_maxopd) &&                                   \
 519             (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
 520
 521 /*
 522  * TCP input handling is split into multiple parts:
 523  *   tcp6_input is a thin wrapper around tcp_input for the extended
 524  *      ip6_protox[] call format in ip6_input
 525  *   tcp_input handles primary segment validation, inpcb lookup and
 526  *      SYN processing on listen sockets
 527  *   tcp_do_segment processes the ACK and text of the segment for
 528  *      establishing, established and closing connections
 529  */
 530 #ifdef INET6
 531 int
 532 tcp6_input(struct mbuf **mp, int *offp, int proto)
 533 {
 534         struct mbuf *m = *mp;
 535         struct in6_ifaddr *ia6;
 536
 537         IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
 538
 539         /*
 540          * draft-itojun-ipv6-tcp-to-anycast
 541          * better place to put this in?
 542          */
 543         ia6 = ip6_getdstifaddr(m);
 544         if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
 545                 struct ip6_hdr *ip6;
 546
 547                 ifa_free(&ia6->ia_ifa);
 548                 ip6 = mtod(m, struct ip6_hdr *);
 549                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
 550                             (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
 551                 return IPPROTO_DONE;
 552         }
 553         if (ia6)
 554                 ifa_free(&ia6->ia_ifa);
 555
 556         tcp_input(m, *offp);
 557         return IPPROTO_DONE;
 558 }
 559 #endif /* INET6 */
 560
 561 void
 562 tcp_input(struct mbuf *m, int off0)
 563 {
 564         struct tcphdr *th = NULL;
 565         struct ip *ip = NULL;
 566         struct inpcb *inp = NULL;
 567         struct tcpcb *tp = NULL;
 568         struct socket *so = NULL;
 569         u_char *optp = NULL;
 570         int optlen = 0;
 571 #ifdef INET
 572         int len;
 573 #endif
 574         int tlen = 0, off;
 575         int drop_hdrlen;
 576         int thflags;
 577         int rstreason = 0;      /* For badport_bandlim accounting purposes */
 578 #ifdef TCP_SIGNATURE
 579         uint8_t sig_checked = 0;
 580 #endif
 581         uint8_t iptos = 0;
 582         struct m_tag *fwd_tag = NULL;
 583 #ifdef INET6
 584         struct ip6_hdr *ip6 = NULL;
 585         int isipv6;
 586 #else
 587         const void *ip6 = NULL;
 588 #endif /* INET6 */
 589         struct tcpopt to;               /* options in this segment */
 590         char *s = NULL;                 /* address and port logging */
 591         int ti_locked;
 592 #define TI_UNLOCKED     1
 593 #define TI_WLOCKED      2
 594
 595 #ifdef TCPDEBUG
 596         /*
 597          * The size of tcp_saveipgen must be the size of the max ip header,
 598          * now IPv6.
 599          */
 600         u_char tcp_saveipgen[IP6_HDR_LEN];
 601         struct tcphdr tcp_savetcp;
 602         short ostate = 0;
 603 #endif
 604
 605 #ifdef INET6
 606         isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
 607 #endif
 608
 609         to.to_flags = 0;
 610         TCPSTAT_INC(tcps_rcvtotal);
 611
 612 #ifdef INET6
 613         if (isipv6) {
 614                 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
 615
 616                 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
 617                         m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
 618                         if (m == NULL) {
 619                                 TCPSTAT_INC(tcps_rcvshort);
 620                                 return;
 621                         }
 622                 }
 623
 624                 ip6 = mtod(m, struct ip6_hdr *);
 625                 th = (struct tcphdr *)((caddr_t)ip6 + off0);
 626                 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
 627                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
 628                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 629                                 th->th_sum = m->m_pkthdr.csum_data;
 630                         else
 631                                 th->th_sum = in6_cksum_pseudo(ip6, tlen,
 632                                     IPPROTO_TCP, m->m_pkthdr.csum_data);
 633                         th->th_sum ^= 0xffff;
 634                 } else
 635                         th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
 636                 if (th->th_sum) {
 637                         TCPSTAT_INC(tcps_rcvbadsum);
 638                         goto drop;
 639                 }
 640
 641                 /*
 642                  * Be proactive about unspecified IPv6 address in source.
 643                  * As we use all-zero to indicate unbounded/unconnected pcb,
 644                  * unspecified IPv6 address can be used to confuse us.
 645                  *
 646                  * Note that packets with unspecified IPv6 destination is
 647                  * already dropped in ip6_input.
 648                  */
 649                 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
 650                         /* XXX stat */
 651                         goto drop;
 652                 }
 653         }
 654 #endif
 655 #if defined(INET) && defined(INET6)
 656         else
 657 #endif
 658 #ifdef INET
 659         {
 660                 /*
 661                  * Get IP and TCP header together in first mbuf.
 662                  * Note: IP leaves IP header in first mbuf.
 663                  */
 664                 if (off0 > sizeof (struct ip)) {
 665                         ip_stripoptions(m);
 666                         off0 = sizeof(struct ip);
 667                 }
 668                 if (m->m_len < sizeof (struct tcpiphdr)) {
 669                         if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
 670                             == NULL) {
 671                                 TCPSTAT_INC(tcps_rcvshort);
 672                                 return;
 673                         }
 674                 }
 675                 ip = mtod(m, struct ip *);
 676                 th = (struct tcphdr *)((caddr_t)ip + off0);
 677                 tlen = ntohs(ip->ip_len) - off0;
 678
 679                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
 680                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 681                                 th->th_sum = m->m_pkthdr.csum_data;
 682                         else
 683                                 th->th_sum = in_pseudo(ip->ip_src.s_addr,
 684                                     ip->ip_dst.s_addr,
 685                                     htonl(m->m_pkthdr.csum_data + tlen +
 686                                     IPPROTO_TCP));
 687                         th->th_sum ^= 0xffff;
 688                 } else {
 689                         struct ipovly *ipov = (struct ipovly *)ip;
 690
 691                         /*
 692                          * Checksum extended TCP header and data.
 693                          */
 694                         len = off0 + tlen;
 695                         bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
 696                         ipov->ih_len = htons(tlen);
 697                         th->th_sum = in_cksum(m, len);
 698                         /* Reset length for SDT probes. */
 699                         ip->ip_len = htons(tlen + off0);
 700                 }
 701
 702                 if (th->th_sum) {
 703                         TCPSTAT_INC(tcps_rcvbadsum);
 704                         goto drop;
 705                 }
 706                 /* Re-initialization for later version check */
 707                 ip->ip_v = IPVERSION;
 708         }
 709 #endif /* INET */
 710
 711 #ifdef INET6
 712         if (isipv6)
 713                 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
 714 #endif
 715 #if defined(INET) && defined(INET6)
 716         else
 717 #endif
 718 #ifdef INET
 719                 iptos = ip->ip_tos;
 720 #endif
 721
 722         /*
 723          * Check that TCP offset makes sense,
 724          * pull out TCP options and adjust length.              XXX
 725          */
 726         off = th->th_off << 2;
 727         if (off < sizeof (struct tcphdr) || off > tlen) {
 728                 TCPSTAT_INC(tcps_rcvbadoff);
 729                 goto drop;
 730         }
 731         tlen -= off;    /* tlen is used instead of ti->ti_len */
 732         if (off > sizeof (struct tcphdr)) {
 733 #ifdef INET6
 734                 if (isipv6) {
 735                         IP6_EXTHDR_CHECK(m, off0, off, );
 736                         ip6 = mtod(m, struct ip6_hdr *);
 737                         th = (struct tcphdr *)((caddr_t)ip6 + off0);
 738                 }
 739 #endif
 740 #if defined(INET) && defined(INET6)
 741                 else
 742 #endif
 743 #ifdef INET
 744                 {
 745                         if (m->m_len < sizeof(struct ip) + off) {
 746                                 if ((m = m_pullup(m, sizeof (struct ip) + off))
 747                                     == NULL) {
 748                                         TCPSTAT_INC(tcps_rcvshort);
 749                                         return;
 750                                 }
 751                                 ip = mtod(m, struct ip *);
 752                                 th = (struct tcphdr *)((caddr_t)ip + off0);
 753                         }
 754                 }
 755 #endif
 756                 optlen = off - sizeof (struct tcphdr);
 757                 optp = (u_char *)(th + 1);
 758         }
 759         thflags = th->th_flags;
 760
 761         /*
 762          * Convert TCP protocol specific fields to host format.
 763          */
 764         tcp_fields_to_host(th);
 765
 766         /*
 767          * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
 768          */
 769         drop_hdrlen = off0 + off;
 770
 771         /*
 772          * Locate pcb for segment; if we're likely to add or remove a
 773          * connection then first acquire pcbinfo lock.  There are two cases
 774          * where we might discover later we need a write lock despite the
 775          * flags: ACKs moving a connection out of the syncache, and ACKs for
 776          * a connection in TIMEWAIT.
 777          */
 778         if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
 779                 INP_INFO_WLOCK(&V_tcbinfo);
 780                 ti_locked = TI_WLOCKED;
 781         } else
 782                 ti_locked = TI_UNLOCKED;
 783
 784         /*
 785          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 786          */
 787         if (
 788 #ifdef INET6
 789             (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
 790 #ifdef INET
 791             || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
 792 #endif
 793 #endif
 794 #if defined(INET) && !defined(INET6)
 795             (m->m_flags & M_IP_NEXTHOP)
 796 #endif
 797             )
 798                 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 799
 800 findpcb:
 801 #ifdef INVARIANTS
 802         if (ti_locked == TI_WLOCKED) {
 803                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 804         } else {
 805                 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
 806         }
 807 #endif
 808 #ifdef INET6
 809         if (isipv6 && fwd_tag != NULL) {
 810                 struct sockaddr_in6 *next_hop6;
 811
 812                 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
 813                 /*
 814                  * Transparently forwarded. Pretend to be the destination.
 815                  * Already got one like this?
 816                  */
 817                 inp = in6_pcblookup_mbuf(&V_tcbinfo,
 818                     &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
 819                     INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
 820                 if (!inp) {
 821                         /*
 822                          * It's new.  Try to find the ambushing socket.
 823                          * Because we've rewritten the destination address,
 824                          * any hardware-generated hash is ignored.
 825                          */
 826                         inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
 827                             th->th_sport, &next_hop6->sin6_addr,
 828                             next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
 829                             th->th_dport, INPLOOKUP_WILDCARD |
 830                             INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
 831                 }
 832         } else if (isipv6) {
 833                 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
 834                     th->th_sport, &ip6->ip6_dst, th->th_dport,
 835                     INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
 836                     m->m_pkthdr.rcvif, m);
 837         }
 838 #endif /* INET6 */
 839 #if defined(INET6) && defined(INET)
 840         else
 841 #endif
 842 #ifdef INET
 843         if (fwd_tag != NULL) {
 844                 struct sockaddr_in *next_hop;
 845
 846                 next_hop = (struct sockaddr_in *)(fwd_tag+1);
 847                 /*
 848                  * Transparently forwarded. Pretend to be the destination.
 849                  * already got one like this?
 850                  */
 851                 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
 852                     ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
 853                     m->m_pkthdr.rcvif, m);
 854                 if (!inp) {
 855                         /*
 856                          * It's new.  Try to find the ambushing socket.
 857                          * Because we've rewritten the destination address,
 858                          * any hardware-generated hash is ignored.
 859                          */
 860                         inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
 861                             th->th_sport, next_hop->sin_addr,
 862                             next_hop->sin_port ? ntohs(next_hop->sin_port) :
 863                             th->th_dport, INPLOOKUP_WILDCARD |
 864                             INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
 865                 }
 866         } else
 867                 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
 868                     th->th_sport, ip->ip_dst, th->th_dport,
 869                     INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
 870                     m->m_pkthdr.rcvif, m);
 871 #endif /* INET */
 872
 873         /*
 874          * If the INPCB does not exist then all data in the incoming
 875          * segment is discarded and an appropriate RST is sent back.
 876          * XXX MRT Send RST using which routing table?
 877          */
 878         if (inp == NULL) {
 879                 /*
 880                  * Log communication attempts to ports that are not
 881                  * in use.
 882                  */
 883                 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
 884                     tcp_log_in_vain == 2) {
 885                         if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
 886                                 log(LOG_INFO, "%s; %s: Connection attempt "
 887                                     "to closed port\n", s, __func__);
 888                 }
 889                 /*
 890                  * When blackholing do not respond with a RST but
 891                  * completely ignore the segment and drop it.
 892                  */
 893                 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
 894                     V_blackhole == 2)
 895                         goto dropunlock;
 896
 897                 rstreason = BANDLIM_RST_CLOSEDPORT;
 898                 goto dropwithreset;
 899         }
 900         INP_WLOCK_ASSERT(inp);
 901         if (!(inp->inp_flags & INP_HW_FLOWID)
 902             && (m->m_flags & M_FLOWID)
 903             && ((inp->inp_socket == NULL)
 904                 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
 905                 inp->inp_flags |= INP_HW_FLOWID;
 906                 inp->inp_flags &= ~INP_SW_FLOWID;
 907                 inp->inp_flowid = m->m_pkthdr.flowid;
 908         }
 909 #ifdef IPSEC
 910 #ifdef INET6
 911         if (isipv6 && ipsec6_in_reject(m, inp)) {
 912                 IPSEC6STAT_INC(ips_in_polvio);
 913                 goto dropunlock;
 914         } else
 915 #endif /* INET6 */
 916         if (ipsec4_in_reject(m, inp) != 0) {
 917                 IPSECSTAT_INC(ips_in_polvio);
 918                 goto dropunlock;
 919         }
 920 #endif /* IPSEC */
 921
 922         /*
 923          * Check the minimum TTL for socket.
 924          */
 925         if (inp->inp_ip_minttl != 0) {
 926 #ifdef INET6
 927                 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
 928                         goto dropunlock;
 929                 else
 930 #endif
 931                 if (inp->inp_ip_minttl > ip->ip_ttl)
 932                         goto dropunlock;
 933         }
 934
 935         /*
 936          * A previous connection in TIMEWAIT state is supposed to catch stray
 937          * or duplicate segments arriving late.  If this segment was a
 938          * legitimate new connection attempt, the old INPCB gets removed and
 939          * we can try again to find a listening socket.
 940          *
 941          * At this point, due to earlier optimism, we may hold only an inpcb
 942          * lock, and not the inpcbinfo write lock.  If so, we need to try to
 943          * acquire it, or if that fails, acquire a reference on the inpcb,
 944          * drop all locks, acquire a global write lock, and then re-acquire
 945          * the inpcb lock.  We may at that point discover that another thread
 946          * has tried to free the inpcb, in which case we need to loop back
 947          * and try to find a new inpcb to deliver to.
 948          *
 949          * XXXRW: It may be time to rethink timewait locking.
 950          */
 951 relocked:
 952         if (inp->inp_flags & INP_TIMEWAIT) {
 953                 if (ti_locked == TI_UNLOCKED) {
 954                         if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
 955                                 in_pcbref(inp);
 956                                 INP_WUNLOCK(inp);
 957                                 INP_INFO_WLOCK(&V_tcbinfo);
 958                                 ti_locked = TI_WLOCKED;
 959                                 INP_WLOCK(inp);
 960                                 if (in_pcbrele_wlocked(inp)) {
 961                                         inp = NULL;
 962                                         goto findpcb;
 963                                 }
 964                         } else
 965                                 ti_locked = TI_WLOCKED;
 966                 }
 967                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 968
 969                 if (thflags & TH_SYN)
 970                         tcp_dooptions(&to, optp, optlen, TO_SYN);
 971                 /*
 972                  * NB: tcp_twcheck unlocks the INP and frees the mbuf.
 973                  */
 974                 if (tcp_twcheck(inp, &to, th, m, tlen))
 975                         goto findpcb;
 976                 INP_INFO_WUNLOCK(&V_tcbinfo);
 977                 return;
 978         }
 979         /*
 980          * The TCPCB may no longer exist if the connection is winding
 981          * down or it is in the CLOSED state.  Either way we drop the
 982          * segment and send an appropriate response.
 983          */
 984         tp = intotcpcb(inp);
 985         if (tp == NULL || tp->t_state == TCPS_CLOSED) {
 986                 rstreason = BANDLIM_RST_CLOSEDPORT;
 987                 goto dropwithreset;
 988         }
 989
 990 #ifdef TCP_OFFLOAD
 991         if (tp->t_flags & TF_TOE) {
 992                 tcp_offload_input(tp, m);
 993                 m = NULL;       /* consumed by the TOE driver */
 994                 goto dropunlock;
 995         }
 996 #endif
 997
 998         /*
 999          * We've identified a valid inpcb, but it could be that we need an
1000          * inpcbinfo write lock but don't hold it.  In this case, attempt to
1001          * acquire using the same strategy as the TIMEWAIT case above.  If we
1002          * relock, we have to jump back to 'relocked' as the connection might
1003          * now be in TIMEWAIT.
1004          */
1005 #ifdef INVARIANTS
1006         if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
1007                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1008 #endif
1009         if (tp->t_state != TCPS_ESTABLISHED) {
1010                 if (ti_locked == TI_UNLOCKED) {
1011                         if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
1012                                 in_pcbref(inp);
1013                                 INP_WUNLOCK(inp);
1014                                 INP_INFO_WLOCK(&V_tcbinfo);
1015                                 ti_locked = TI_WLOCKED;
1016                                 INP_WLOCK(inp);
1017                                 if (in_pcbrele_wlocked(inp)) {
1018                                         inp = NULL;
1019                                         goto findpcb;
1020                                 }
1021                                 goto relocked;
1022                         } else
1023                                 ti_locked = TI_WLOCKED;
1024                 }
1025                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1026         }
1027
1028 #ifdef MAC
1029         INP_WLOCK_ASSERT(inp);
1030         if (mac_inpcb_check_deliver(inp, m))
1031                 goto dropunlock;
1032 #endif
1033         so = inp->inp_socket;
1034         KASSERT(so != NULL, ("%s: so == NULL", __func__));
1035 #ifdef TCPDEBUG
1036         if (so->so_options & SO_DEBUG) {
1037                 ostate = tp->t_state;
1038 #ifdef INET6
1039                 if (isipv6) {
1040                         bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1041                 } else
1042 #endif
1043                         bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1044                 tcp_savetcp = *th;
1045         }
1046 #endif /* TCPDEBUG */
1047         /*
1048          * When the socket is accepting connections (the INPCB is in LISTEN
1049          * state) we look into the SYN cache if this is a new connection
1050          * attempt or the completion of a previous one.  Because listen
1051          * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1052          * held in this case.
1053          */
1054         if (so->so_options & SO_ACCEPTCONN) {
1055                 struct in_conninfo inc;
1056
1057                 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1058                     "tp not listening", __func__));
1059                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1060
1061                 bzero(&inc, sizeof(inc));
1062 #ifdef INET6
1063                 if (isipv6) {
1064                         inc.inc_flags |= INC_ISIPV6;
1065                         inc.inc6_faddr = ip6->ip6_src;
1066                         inc.inc6_laddr = ip6->ip6_dst;
1067                 } else
1068 #endif
1069                 {
1070                         inc.inc_faddr = ip->ip_src;
1071                         inc.inc_laddr = ip->ip_dst;
1072                 }
1073                 inc.inc_fport = th->th_sport;
1074                 inc.inc_lport = th->th_dport;
1075                 inc.inc_fibnum = so->so_fibnum;
1076
1077                 /*
1078                  * Check for an existing connection attempt in syncache if
1079                  * the flag is only ACK.  A successful lookup creates a new
1080                  * socket appended to the listen queue in SYN_RECEIVED state.
1081                  */
1082                 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1083                         /*
1084                          * Parse the TCP options here because
1085                          * syncookies need access to the reflected
1086                          * timestamp.
1087                          */
1088                         tcp_dooptions(&to, optp, optlen, 0);
1089                         /*
1090                          * NB: syncache_expand() doesn't unlock
1091                          * inp and tcpinfo locks.
1092                          */
1093                         if (!syncache_expand(&inc, &to, th, &so, m)) {
1094                                 /*
1095                                  * No syncache entry or ACK was not
1096                                  * for our SYN/ACK.  Send a RST.
1097                                  * NB: syncache did its own logging
1098                                  * of the failure cause.
1099                                  */
1100                                 rstreason = BANDLIM_RST_OPENPORT;
1101                                 goto dropwithreset;
1102                         }
1103                         if (so == NULL) {
1104                                 /*
1105                                  * We completed the 3-way handshake
1106                                  * but could not allocate a socket
1107                                  * either due to memory shortage,
1108                                  * listen queue length limits or
1109                                  * global socket limits.  Send RST
1110                                  * or wait and have the remote end
1111                                  * retransmit the ACK for another
1112                                  * try.
1113                                  */
1114                                 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1115                                         log(LOG_DEBUG, "%s; %s: Listen socket: "
1116                                             "Socket allocation failed due to "
1117                                             "limits or memory shortage, %s\n",
1118                                             s, __func__,
1119                                             V_tcp_sc_rst_sock_fail ?
1120                                             "sending RST" : "try again");
1121                                 if (V_tcp_sc_rst_sock_fail) {
1122                                         rstreason = BANDLIM_UNLIMITED;
1123                                         goto dropwithreset;
1124                                 } else
1125                                         goto dropunlock;
1126                         }
1127                         /*
1128                          * Socket is created in state SYN_RECEIVED.
1129                          * Unlock the listen socket, lock the newly
1130                          * created socket and update the tp variable.
1131                          */
1132                         INP_WUNLOCK(inp);       /* listen socket */
1133                         inp = sotoinpcb(so);
1134                         INP_WLOCK(inp);         /* new connection */
1135                         tp = intotcpcb(inp);
1136                         KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1137                             ("%s: ", __func__));
1138 #ifdef TCP_SIGNATURE
1139                         if (sig_checked == 0)  {
1140                                 tcp_dooptions(&to, optp, optlen,
1141                                     (thflags & TH_SYN) ? TO_SYN : 0);
1142                                 if (!tcp_signature_verify_input(m, off0, tlen,
1143                                     optlen, &to, th, tp->t_flags)) {
1144
1145                                         /*
1146                                          * In SYN_SENT state if it receives an
1147                                          * RST, it is allowed for further
1148                                          * processing.
1149                                          */
1150                                         if ((thflags & TH_RST) == 0 ||
1151                                             (tp->t_state == TCPS_SYN_SENT) == 0)
1152                                                 goto dropunlock;
1153                                 }
1154                                 sig_checked = 1;
1155                         }
1156 #endif
1157
1158                         /*
1159                          * Process the segment and the data it
1160                          * contains.  tcp_do_segment() consumes
1161                          * the mbuf chain and unlocks the inpcb.
1162                          */
1163                         tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1164                             iptos, ti_locked);
1165                         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1166                         return;
1167                 }
1168                 /*
1169                  * Segment flag validation for new connection attempts:
1170                  *
1171                  * Our (SYN|ACK) response was rejected.
1172                  * Check with syncache and remove entry to prevent
1173                  * retransmits.
1174                  *
1175                  * NB: syncache_chkrst does its own logging of failure
1176                  * causes.
1177                  */
1178                 if (thflags & TH_RST) {
1179                         syncache_chkrst(&inc, th);
1180                         goto dropunlock;
1181                 }
1182                 /*
1183                  * We can't do anything without SYN.
1184                  */
1185                 if ((thflags & TH_SYN) == 0) {
1186                         if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1187                                 log(LOG_DEBUG, "%s; %s: Listen socket: "
1188                                     "SYN is missing, segment ignored\n",
1189                                     s, __func__);
1190                         TCPSTAT_INC(tcps_badsyn);
1191                         goto dropunlock;
1192                 }
1193                 /*
1194                  * (SYN|ACK) is bogus on a listen socket.
1195                  */
1196                 if (thflags & TH_ACK) {
1197                         if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1198                                 log(LOG_DEBUG, "%s; %s: Listen socket: "
1199                                     "SYN|ACK invalid, segment rejected\n",
1200                                     s, __func__);
1201                         syncache_badack(&inc);  /* XXX: Not needed! */
1202                         TCPSTAT_INC(tcps_badsyn);
1203                         rstreason = BANDLIM_RST_OPENPORT;
1204                         goto dropwithreset;
1205                 }
1206                 /*
1207                  * If the drop_synfin option is enabled, drop all
1208                  * segments with both the SYN and FIN bits set.
1209                  * This prevents e.g. nmap from identifying the
1210                  * TCP/IP stack.
1211                  * XXX: Poor reasoning.  nmap has other methods
1212                  * and is constantly refining its stack detection
1213                  * strategies.
1214                  * XXX: This is a violation of the TCP specification
1215                  * and was used by RFC1644.
1216                  */
1217                 if ((thflags & TH_FIN) && V_drop_synfin) {
1218                         if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1219                                 log(LOG_DEBUG, "%s; %s: Listen socket: "
1220                                     "SYN|FIN segment ignored (based on "
1221                                     "sysctl setting)\n", s, __func__);
1222                         TCPSTAT_INC(tcps_badsyn);
1223                         goto dropunlock;
1224                 }
1225                 /*
1226                  * Segment's flags are (SYN) or (SYN|FIN).
1227                  *
1228                  * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1229                  * as they do not affect the state of the TCP FSM.
1230                  * The data pointed to by TH_URG and th_urp is ignored.
1231                  */
1232                 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1233                     ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1234                 KASSERT(thflags & (TH_SYN),
1235                     ("%s: Listen socket: TH_SYN not set", __func__));
1236 #ifdef INET6
1237                 /*
1238                  * If deprecated address is forbidden,
1239                  * we do not accept SYN to deprecated interface
1240                  * address to prevent any new inbound connection from
1241                  * getting established.
1242                  * When we do not accept SYN, we send a TCP RST,
1243                  * with deprecated source address (instead of dropping
1244                  * it).  We compromise it as it is much better for peer
1245                  * to send a RST, and RST will be the final packet
1246                  * for the exchange.
1247                  *
1248                  * If we do not forbid deprecated addresses, we accept
1249                  * the SYN packet.  RFC2462 does not suggest dropping
1250                  * SYN in this case.
1251                  * If we decipher RFC2462 5.5.4, it says like this:
1252                  * 1. use of deprecated addr with existing
1253                  *    communication is okay - "SHOULD continue to be
1254                  *    used"
1255                  * 2. use of it with new communication:
1256                  *   (2a) "SHOULD NOT be used if alternate address
1257                  *        with sufficient scope is available"
1258                  *   (2b) nothing mentioned otherwise.
1259                  * Here we fall into (2b) case as we have no choice in
1260                  * our source address selection - we must obey the peer.
1261                  *
1262                  * The wording in RFC2462 is confusing, and there are
1263                  * multiple description text for deprecated address
1264                  * handling - worse, they are not exactly the same.
1265                  * I believe 5.5.4 is the best one, so we follow 5.5.4.
1266                  */
1267                 if (isipv6 && !V_ip6_use_deprecated) {
1268                         struct in6_ifaddr *ia6;
1269
1270                         ia6 = ip6_getdstifaddr(m);
1271                         if (ia6 != NULL &&
1272                             (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1273                                 ifa_free(&ia6->ia_ifa);
1274                                 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1275                                     log(LOG_DEBUG, "%s; %s: Listen socket: "
1276                                         "Connection attempt to deprecated "
1277                                         "IPv6 address rejected\n",
1278                                         s, __func__);
1279                                 rstreason = BANDLIM_RST_OPENPORT;
1280                                 goto dropwithreset;
1281                         }
1282                         if (ia6)
1283                                 ifa_free(&ia6->ia_ifa);
1284                 }
1285 #endif /* INET6 */
1286                 /*
1287                  * Basic sanity checks on incoming SYN requests:
1288                  *   Don't respond if the destination is a link layer
1289                  *      broadcast according to RFC1122 4.2.3.10, p. 104.
1290                  *   If it is from this socket it must be forged.
1291                  *   Don't respond if the source or destination is a
1292                  *      global or subnet broad- or multicast address.
1293                  *   Note that it is quite possible to receive unicast
1294                  *      link-layer packets with a broadcast IP address. Use
1295                  *      in_broadcast() to find them.
1296                  */
1297                 if (m->m_flags & (M_BCAST|M_MCAST)) {
1298                         if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299                             log(LOG_DEBUG, "%s; %s: Listen socket: "
1300                                 "Connection attempt from broad- or multicast "
1301                                 "link layer address ignored\n", s, __func__);
1302                         goto dropunlock;
1303                 }
1304 #ifdef INET6
1305                 if (isipv6) {
1306                         if (th->th_dport == th->th_sport &&
1307                             IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1308                                 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309                                     log(LOG_DEBUG, "%s; %s: Listen socket: "
1310                                         "Connection attempt to/from self "
1311                                         "ignored\n", s, __func__);
1312                                 goto dropunlock;
1313                         }
1314                         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1315                             IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1316                                 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1317                                     log(LOG_DEBUG, "%s; %s: Listen socket: "
1318                                         "Connection attempt from/to multicast "
1319                                         "address ignored\n", s, __func__);
1320                                 goto dropunlock;
1321                         }
1322                 }
1323 #endif
1324 #if defined(INET) && defined(INET6)
1325                 else
1326 #endif
1327 #ifdef INET
1328                 {
1329                         if (th->th_dport == th->th_sport &&
1330                             ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1331                                 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1332                                     log(LOG_DEBUG, "%s; %s: Listen socket: "
1333                                         "Connection attempt from/to self "
1334                                         "ignored\n", s, __func__);
1335                                 goto dropunlock;
1336                         }
1337                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1338                             IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1339                             ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1340                             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1341                                 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1342                                     log(LOG_DEBUG, "%s; %s: Listen socket: "
1343                                         "Connection attempt from/to broad- "
1344                                         "or multicast address ignored\n",
1345                                         s, __func__);
1346                                 goto dropunlock;
1347                         }
1348                 }
1349 #endif
1350                 /*
1351                  * SYN appears to be valid.  Create compressed TCP state
1352                  * for syncache.
1353                  */
1354 #ifdef TCPDEBUG
1355                 if (so->so_options & SO_DEBUG)
1356                         tcp_trace(TA_INPUT, ostate, tp,
1357                             (void *)tcp_saveipgen, &tcp_savetcp, 0);
1358 #endif
1359                 tcp_dooptions(&to, optp, optlen, TO_SYN);
1360                 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1361                 /*
1362                  * Entry added to syncache and mbuf consumed.
1363                  * Everything already unlocked by syncache_add().
1364                  */
1365                 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1366                 return;
1367         } else if (tp->t_state == TCPS_LISTEN) {
1368                 /*
1369                  * When a listen socket is torn down the SO_ACCEPTCONN
1370                  * flag is removed first while connections are drained
1371                  * from the accept queue in a unlock/lock cycle of the
1372                  * ACCEPT_LOCK, opening a race condition allowing a SYN
1373                  * attempt go through unhandled.
1374                  */
1375                 goto dropunlock;
1376         }
1377
1378 #ifdef TCP_SIGNATURE
1379         if (sig_checked == 0)  {
1380                 tcp_dooptions(&to, optp, optlen,
1381                     (thflags & TH_SYN) ? TO_SYN : 0);
1382                 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1383                     th, tp->t_flags)) {
1384
1385                         /*
1386                          * In SYN_SENT state if it receives an RST, it is
1387                          * allowed for further processing.
1388                          */
1389                         if ((thflags & TH_RST) == 0 ||
1390                             (tp->t_state == TCPS_SYN_SENT) == 0)
1391                                 goto dropunlock;
1392                 }
1393                 sig_checked = 1;
1394         }
1395 #endif
1396
1397         TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1398
1399         /*
1400          * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1401          * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1402          * the inpcb, and unlocks pcbinfo.
1403          */
1404         tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1405         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1406         return;
1407
1408 dropwithreset:
1409         TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1410
1411         if (ti_locked == TI_WLOCKED) {
1412                 INP_INFO_WUNLOCK(&V_tcbinfo);
1413                 ti_locked = TI_UNLOCKED;
1414         }
1415 #ifdef INVARIANTS
1416         else {
1417                 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1418                     "ti_locked: %d", __func__, ti_locked));
1419                 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1420         }
1421 #endif
1422
1423         if (inp != NULL) {
1424                 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1425                 INP_WUNLOCK(inp);
1426         } else
1427                 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1428         m = NULL;       /* mbuf chain got consumed. */
1429         goto drop;
1430
1431 dropunlock:
1432         if (m != NULL)
1433                 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1434
1435         if (ti_locked == TI_WLOCKED) {
1436                 INP_INFO_WUNLOCK(&V_tcbinfo);
1437                 ti_locked = TI_UNLOCKED;
1438         }
1439 #ifdef INVARIANTS
1440         else {
1441                 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1442                     "ti_locked: %d", __func__, ti_locked));
1443                 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1444         }
1445 #endif
1446
1447         if (inp != NULL)
1448                 INP_WUNLOCK(inp);
1449
1450 drop:
1451         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1452         if (s != NULL)
1453                 free(s, M_TCPLOG);
1454         if (m != NULL)
1455                 m_freem(m);
1456 }
1457
1458 static void
1459 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1460     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1461     int ti_locked)
1462 {
1463         int thflags, acked, ourfinisacked, needoutput = 0;
1464         int rstreason, todrop, win;
1465         u_long tiwin;
1466         char *s;
1467         struct in_conninfo *inc;
1468         struct mbuf *mfree;
1469         struct tcpopt to;
1470
1471 #ifdef TCPDEBUG
1472         /*
1473          * The size of tcp_saveipgen must be the size of the max ip header,
1474          * now IPv6.
1475          */
1476         u_char tcp_saveipgen[IP6_HDR_LEN];
1477         struct tcphdr tcp_savetcp;
1478         short ostate = 0;
1479 #endif
1480         thflags = th->th_flags;
1481         inc = &tp->t_inpcb->inp_inc;
1482         tp->sackhint.last_sack_ack = 0;
1483
1484         /*
1485          * If this is either a state-changing packet or current state isn't
1486          * established, we require a write lock on tcbinfo.  Otherwise, we
1487          * allow the tcbinfo to be in either alocked or unlocked, as the
1488          * caller may have unnecessarily acquired a write lock due to a race.
1489          */
1490         if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1491             tp->t_state != TCPS_ESTABLISHED) {
1492                 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1493                     "SYN/FIN/RST/!EST", __func__, ti_locked));
1494                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1495         } else {
1496 #ifdef INVARIANTS
1497                 if (ti_locked == TI_WLOCKED)
1498                         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1499                 else {
1500                         KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1501                             "ti_locked: %d", __func__, ti_locked));
1502                         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1503                 }
1504 #endif
1505         }
1506         INP_WLOCK_ASSERT(tp->t_inpcb);
1507         KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1508             __func__));
1509         KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1510             __func__));
1511
1512         /*
1513          * Segment received on connection.
1514          * Reset idle time and keep-alive timer.
1515          * XXX: This should be done after segment
1516          * validation to ignore broken/spoofed segs.
1517          */
1518         tp->t_rcvtime = ticks;
1519         if (TCPS_HAVEESTABLISHED(tp->t_state))
1520                 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1521
1522         /*
1523          * Unscale the window into a 32-bit value.
1524          * For the SYN_SENT state the scale is zero.
1525          */
1526         tiwin = th->th_win << tp->snd_scale;
1527
1528         /*
1529          * TCP ECN processing.
1530          */
1531         if (tp->t_flags & TF_ECN_PERMIT) {
1532                 if (thflags & TH_CWR)
1533                         tp->t_flags &= ~TF_ECN_SND_ECE;
1534                 switch (iptos & IPTOS_ECN_MASK) {
1535                 case IPTOS_ECN_CE:
1536                         tp->t_flags |= TF_ECN_SND_ECE;
1537                         TCPSTAT_INC(tcps_ecn_ce);
1538                         break;
1539                 case IPTOS_ECN_ECT0:
1540                         TCPSTAT_INC(tcps_ecn_ect0);
1541                         break;
1542                 case IPTOS_ECN_ECT1:
1543                         TCPSTAT_INC(tcps_ecn_ect1);
1544                         break;
1545                 }
1546                 /* Congestion experienced. */
1547                 if (thflags & TH_ECE) {
1548                         cc_cong_signal(tp, th, CC_ECN);
1549                 }
1550         }
1551
1552         /*
1553          * Parse options on any incoming segment.
1554          */
1555         tcp_dooptions(&to, (u_char *)(th + 1),
1556             (th->th_off << 2) - sizeof(struct tcphdr),
1557             (thflags & TH_SYN) ? TO_SYN : 0);
1558
1559         /*
1560          * If echoed timestamp is later than the current time,
1561          * fall back to non RFC1323 RTT calculation.  Normalize
1562          * timestamp if syncookies were used when this connection
1563          * was established.
1564          */
1565         if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1566                 to.to_tsecr -= tp->ts_offset;
1567                 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1568                         to.to_tsecr = 0;
1569         }
1570         /*
1571          * If timestamps were negotiated during SYN/ACK they should
1572          * appear on every segment during this session and vice versa.
1573          */
1574         if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1575                 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1576                         log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1577                             "no action\n", s, __func__);
1578                         free(s, M_TCPLOG);
1579                 }
1580         }
1581         if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1582                 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1583                         log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1584                             "no action\n", s, __func__);
1585                         free(s, M_TCPLOG);
1586                 }
1587         }
1588
1589         /*
1590          * Process options only when we get SYN/ACK back. The SYN case
1591          * for incoming connections is handled in tcp_syncache.
1592          * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1593          * or <SYN,ACK>) segment itself is never scaled.
1594          * XXX this is traditional behavior, may need to be cleaned up.
1595          */
1596         if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1597                 if ((to.to_flags & TOF_SCALE) &&
1598                     (tp->t_flags & TF_REQ_SCALE)) {
1599                         tp->t_flags |= TF_RCVD_SCALE;
1600                         tp->snd_scale = to.to_wscale;
1601                 }
1602                 /*
1603                  * Initial send window.  It will be updated with
1604                  * the next incoming segment to the scaled value.
1605                  */
1606                 tp->snd_wnd = th->th_win;
1607                 if (to.to_flags & TOF_TS) {
1608                         tp->t_flags |= TF_RCVD_TSTMP;
1609                         tp->ts_recent = to.to_tsval;
1610                         tp->ts_recent_age = tcp_ts_getticks();
1611                 }
1612                 if (to.to_flags & TOF_MSS)
1613                         tcp_mss(tp, to.to_mss);
1614                 if ((tp->t_flags & TF_SACK_PERMIT) &&
1615                     (to.to_flags & TOF_SACKPERM) == 0)
1616                         tp->t_flags &= ~TF_SACK_PERMIT;
1617         }
1618
1619         /*
1620          * Header prediction: check for the two common cases
1621          * of a uni-directional data xfer.  If the packet has
1622          * no control flags, is in-sequence, the window didn't
1623          * change and we're not retransmitting, it's a
1624          * candidate.  If the length is zero and the ack moved
1625          * forward, we're the sender side of the xfer.  Just
1626          * free the data acked & wake any higher level process
1627          * that was blocked waiting for space.  If the length
1628          * is non-zero and the ack didn't move, we're the
1629          * receiver side.  If we're getting packets in-order
1630          * (the reassembly queue is empty), add the data to
1631          * the socket buffer and note that we need a delayed ack.
1632          * Make sure that the hidden state-flags are also off.
1633          * Since we check for TCPS_ESTABLISHED first, it can only
1634          * be TH_NEEDSYN.
1635          */
1636         if (tp->t_state == TCPS_ESTABLISHED &&
1637             th->th_seq == tp->rcv_nxt &&
1638             (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1639             tp->snd_nxt == tp->snd_max &&
1640             tiwin && tiwin == tp->snd_wnd &&
1641             ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1642             tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 ||
1643              TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1644
1645                 /*
1646                  * If last ACK falls within this segment's sequence numbers,
1647                  * record the timestamp.
1648                  * NOTE that the test is modified according to the latest
1649                  * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1650                  */
1651                 if ((to.to_flags & TOF_TS) != 0 &&
1652                     SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1653                         tp->ts_recent_age = tcp_ts_getticks();
1654                         tp->ts_recent = to.to_tsval;
1655                 }
1656
1657                 if (tlen == 0) {
1658                         if (SEQ_GT(th->th_ack, tp->snd_una) &&
1659                             SEQ_LEQ(th->th_ack, tp->snd_max) &&
1660                             !IN_RECOVERY(tp->t_flags) &&
1661                             (to.to_flags & TOF_SACK) == 0 &&
1662                             TAILQ_EMPTY(&tp->snd_holes)) {
1663                                 /*
1664                                  * This is a pure ack for outstanding data.
1665                                  */
1666                                 if (ti_locked == TI_WLOCKED)
1667                                         INP_INFO_WUNLOCK(&V_tcbinfo);
1668                                 ti_locked = TI_UNLOCKED;
1669
1670                                 TCPSTAT_INC(tcps_predack);
1671
1672                                 /*
1673                                  * "bad retransmit" recovery.
1674                                  */
1675                                 if (tp->t_rxtshift == 1 &&
1676                                     tp->t_flags & TF_PREVVALID &&
1677                                     (int)(ticks - tp->t_badrxtwin) < 0) {
1678                                         cc_cong_signal(tp, th, CC_RTO_ERR);
1679                                 }
1680
1681                                 /*
1682                                  * Recalculate the transmit timer / rtt.
1683                                  *
1684                                  * Some boxes send broken timestamp replies
1685                                  * during the SYN+ACK phase, ignore
1686                                  * timestamps of 0 or we could calculate a
1687                                  * huge RTT and blow up the retransmit timer.
1688                                  */
1689                                 if ((to.to_flags & TOF_TS) != 0 &&
1690                                     to.to_tsecr) {
1691                                         u_int t;
1692
1693                                         t = tcp_ts_getticks() - to.to_tsecr;
1694                                         if (!tp->t_rttlow || tp->t_rttlow > t)
1695                                                 tp->t_rttlow = t;
1696                                         tcp_xmit_timer(tp,
1697                                             TCP_TS_TO_TICKS(t) + 1);
1698                                 } else if (tp->t_rtttime &&
1699                                     SEQ_GT(th->th_ack, tp->t_rtseq)) {
1700                                         if (!tp->t_rttlow ||
1701                                             tp->t_rttlow > ticks - tp->t_rtttime)
1702                                                 tp->t_rttlow = ticks - tp->t_rtttime;
1703                                         tcp_xmit_timer(tp,
1704                                                         ticks - tp->t_rtttime);
1705                                 }
1706                                 acked = BYTES_THIS_ACK(tp, th);
1707
1708                                 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1709                                 hhook_run_tcp_est_in(tp, th, &to);
1710
1711                                 TCPSTAT_INC(tcps_rcvackpack);
1712                                 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1713                                 sbdrop(&so->so_snd, acked);
1714                                 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1715                                     SEQ_LEQ(th->th_ack, tp->snd_recover))
1716                                         tp->snd_recover = th->th_ack - 1;
1717
1718                                 /*
1719                                  * Let the congestion control algorithm update
1720                                  * congestion control related information. This
1721                                  * typically means increasing the congestion
1722                                  * window.
1723                                  */
1724                                 cc_ack_received(tp, th, CC_ACK);
1725
1726                                 tp->snd_una = th->th_ack;
1727                                 /*
1728                                  * Pull snd_wl2 up to prevent seq wrap relative
1729                                  * to th_ack.
1730                                  */
1731                                 tp->snd_wl2 = th->th_ack;
1732                                 tp->t_dupacks = 0;
1733                                 m_freem(m);
1734                                 ND6_HINT(tp); /* Some progress has been made. */
1735
1736                                 /*
1737                                  * If all outstanding data are acked, stop
1738                                  * retransmit timer, otherwise restart timer
1739                                  * using current (possibly backed-off) value.
1740                                  * If process is waiting for space,
1741                                  * wakeup/selwakeup/signal.  If data
1742                                  * are ready to send, let tcp_output
1743                                  * decide between more output or persist.
1744                                  */
1745 #ifdef TCPDEBUG
1746                                 if (so->so_options & SO_DEBUG)
1747                                         tcp_trace(TA_INPUT, ostate, tp,
1748                                             (void *)tcp_saveipgen,
1749                                             &tcp_savetcp, 0);
1750 #endif
1751                                 if (tp->snd_una == tp->snd_max)
1752                                         tcp_timer_activate(tp, TT_REXMT, 0);
1753                                 else if (!tcp_timer_active(tp, TT_PERSIST))
1754                                         tcp_timer_activate(tp, TT_REXMT,
1755                                                       tp->t_rxtcur);
1756                                 sowwakeup(so);
1757                                 if (so->so_snd.sb_cc)
1758                                         (void) tcp_output(tp);
1759                                 goto check_delack;
1760                         }
1761                 } else if (th->th_ack == tp->snd_una &&
1762                     tlen <= sbspace(&so->so_rcv)) {
1763                         int newsize = 0;        /* automatic sockbuf scaling */
1764
1765                         /*
1766                          * This is a pure, in-sequence data packet with
1767                          * nothing on the reassembly queue and we have enough
1768                          * buffer space to take it.
1769                          */
1770                         if (ti_locked == TI_WLOCKED)
1771                                 INP_INFO_WUNLOCK(&V_tcbinfo);
1772                         ti_locked = TI_UNLOCKED;
1773
1774                         /* Clean receiver SACK report if present */
1775                         if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1776                                 tcp_clean_sackreport(tp);
1777                         TCPSTAT_INC(tcps_preddat);
1778                         tp->rcv_nxt += tlen;
1779                         /*
1780                          * Pull snd_wl1 up to prevent seq wrap relative to
1781                          * th_seq.
1782                          */
1783                         tp->snd_wl1 = th->th_seq;
1784                         /*
1785                          * Pull rcv_up up to prevent seq wrap relative to
1786                          * rcv_nxt.
1787                          */
1788                         tp->rcv_up = tp->rcv_nxt;
1789                         TCPSTAT_INC(tcps_rcvpack);
1790                         TCPSTAT_ADD(tcps_rcvbyte, tlen);
1791                         ND6_HINT(tp);   /* Some progress has been made */
1792 #ifdef TCPDEBUG
1793                         if (so->so_options & SO_DEBUG)
1794                                 tcp_trace(TA_INPUT, ostate, tp,
1795                                     (void *)tcp_saveipgen, &tcp_savetcp, 0);
1796 #endif
1797                 /*
1798                  * Automatic sizing of receive socket buffer.  Often the send
1799                  * buffer size is not optimally adjusted to the actual network
1800                  * conditions at hand (delay bandwidth product).  Setting the
1801                  * buffer size too small limits throughput on links with high
1802                  * bandwidth and high delay (eg. trans-continental/oceanic links).
1803                  *
1804                  * On the receive side the socket buffer memory is only rarely
1805                  * used to any significant extent.  This allows us to be much
1806                  * more aggressive in scaling the receive socket buffer.  For
1807                  * the case that the buffer space is actually used to a large
1808                  * extent and we run out of kernel memory we can simply drop
1809                  * the new segments; TCP on the sender will just retransmit it
1810                  * later.  Setting the buffer size too big may only consume too
1811                  * much kernel memory if the application doesn't read() from
1812                  * the socket or packet loss or reordering makes use of the
1813                  * reassembly queue.
1814                  *
1815                  * The criteria to step up the receive buffer one notch are:
1816                  *  1. the number of bytes received during the time it takes
1817                  *     one timestamp to be reflected back to us (the RTT);
1818                  *  2. received bytes per RTT is within seven eighth of the
1819                  *     current socket buffer size;
1820                  *  3. receive buffer size has not hit maximal automatic size;
1821                  *
1822                  * This algorithm does one step per RTT at most and only if
1823                  * we receive a bulk stream w/o packet losses or reorderings.
1824                  * Shrinking the buffer during idle times is not necessary as
1825                  * it doesn't consume any memory when idle.
1826                  *
1827                  * TODO: Only step up if the application is actually serving
1828                  * the buffer to better manage the socket buffer resources.
1829                  */
1830                         if (V_tcp_do_autorcvbuf &&
1831                             to.to_tsecr &&
1832                             (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1833                                 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1834                                     to.to_tsecr - tp->rfbuf_ts < hz) {
1835                                         if (tp->rfbuf_cnt >
1836                                             (so->so_rcv.sb_hiwat / 8 * 7) &&
1837                                             so->so_rcv.sb_hiwat <
1838                                             V_tcp_autorcvbuf_max) {
1839                                                 newsize =
1840                                                     min(so->so_rcv.sb_hiwat +
1841                                                     V_tcp_autorcvbuf_inc,
1842                                                     V_tcp_autorcvbuf_max);
1843                                         }
1844                                         /* Start over with next RTT. */
1845                                         tp->rfbuf_ts = 0;
1846                                         tp->rfbuf_cnt = 0;
1847                                 } else
1848                                         tp->rfbuf_cnt += tlen;  /* add up */
1849                         }
1850
1851                         /* Add data to socket buffer. */
1852                         SOCKBUF_LOCK(&so->so_rcv);
1853                         if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1854                                 m_freem(m);
1855                         } else {
1856                                 /*
1857                                  * Set new socket buffer size.
1858                                  * Give up when limit is reached.
1859                                  */
1860                                 if (newsize)
1861                                         if (!sbreserve_locked(&so->so_rcv,
1862                                             newsize, so, NULL))
1863                                                 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1864                                 m_adj(m, drop_hdrlen);  /* delayed header drop */
1865                                 sbappendstream_locked(&so->so_rcv, m);
1866                         }
1867                         /* NB: sorwakeup_locked() does an implicit unlock. */
1868                         sorwakeup_locked(so);
1869                         if (DELAY_ACK(tp, tlen)) {
1870                                 tp->t_flags |= TF_DELACK;
1871                         } else {
1872                                 tp->t_flags |= TF_ACKNOW;
1873                                 tcp_output(tp);
1874                         }
1875                         goto check_delack;
1876                 }
1877         }
1878
1879         /*
1880          * Calculate amount of space in receive window,
1881          * and then do TCP input processing.
1882          * Receive window is amount of space in rcv queue,
1883          * but not less than advertised window.
1884          */
1885         win = sbspace(&so->so_rcv);
1886         if (win < 0)
1887                 win = 0;
1888         tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1889
1890         /* Reset receive buffer auto scaling when not in bulk receive mode. */
1891         tp->rfbuf_ts = 0;
1892         tp->rfbuf_cnt = 0;
1893
1894         switch (tp->t_state) {
1895
1896         /*
1897          * If the state is SYN_RECEIVED:
1898          *      if seg contains an ACK, but not for our SYN/ACK, send a RST.
1899          */
1900         case TCPS_SYN_RECEIVED:
1901                 if ((thflags & TH_ACK) &&
1902                     (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1903                      SEQ_GT(th->th_ack, tp->snd_max))) {
1904                                 rstreason = BANDLIM_RST_OPENPORT;
1905                                 goto dropwithreset;
1906                 }
1907                 break;
1908
1909         /*
1910          * If the state is SYN_SENT:
1911          *      if seg contains an ACK, but not for our SYN, drop the input.
1912          *      if seg contains a RST, then drop the connection.
1913          *      if seg does not contain SYN, then drop it.
1914          * Otherwise this is an acceptable SYN segment
1915          *      initialize tp->rcv_nxt and tp->irs
1916          *      if seg contains ack then advance tp->snd_una
1917          *      if seg contains an ECE and ECN support is enabled, the stream
1918          *          is ECN capable.
1919          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1920          *      arrange for segment to be acked (eventually)
1921          *      continue processing rest of data/controls, beginning with URG
1922          */
1923         case TCPS_SYN_SENT:
1924                 if ((thflags & TH_ACK) &&
1925                     (SEQ_LEQ(th->th_ack, tp->iss) ||
1926                      SEQ_GT(th->th_ack, tp->snd_max))) {
1927                         rstreason = BANDLIM_UNLIMITED;
1928                         goto dropwithreset;
1929                 }
1930                 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1931                         TCP_PROBE5(connect__refused, NULL, tp,
1932                             mtod(m, const char *), tp, th);
1933                         tp = tcp_drop(tp, ECONNREFUSED);
1934                 }
1935                 if (thflags & TH_RST)
1936                         goto drop;
1937                 if (!(thflags & TH_SYN))
1938                         goto drop;
1939
1940                 tp->irs = th->th_seq;
1941                 tcp_rcvseqinit(tp);
1942                 if (thflags & TH_ACK) {
1943                         TCPSTAT_INC(tcps_connects);
1944                         soisconnected(so);
1945 #ifdef MAC
1946                         mac_socketpeer_set_from_mbuf(m, so);
1947 #endif
1948                         /* Do window scaling on this connection? */
1949                         if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1950                                 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1951                                 tp->rcv_scale = tp->request_r_scale;
1952                         }
1953                         tp->rcv_adv += imin(tp->rcv_wnd,
1954                             TCP_MAXWIN << tp->rcv_scale);
1955                         tp->snd_una++;          /* SYN is acked */
1956                         /*
1957                          * If there's data, delay ACK; if there's also a FIN
1958                          * ACKNOW will be turned on later.
1959                          */
1960                         if (DELAY_ACK(tp, tlen) && tlen != 0)
1961                                 tcp_timer_activate(tp, TT_DELACK,
1962                                     tcp_delacktime);
1963                         else
1964                                 tp->t_flags |= TF_ACKNOW;
1965
1966                         if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1967                                 tp->t_flags |= TF_ECN_PERMIT;
1968                                 TCPSTAT_INC(tcps_ecn_shs);
1969                         }
1970
1971                         /*
1972                          * Received <SYN,ACK> in SYN_SENT[*] state.
1973                          * Transitions:
1974                          *      SYN_SENT  --> ESTABLISHED
1975                          *      SYN_SENT* --> FIN_WAIT_1
1976                          */
1977                         tp->t_starttime = ticks;
1978                         if (tp->t_flags & TF_NEEDFIN) {
1979                                 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1980                                 tp->t_flags &= ~TF_NEEDFIN;
1981                                 thflags &= ~TH_SYN;
1982                         } else {
1983                                 tcp_state_change(tp, TCPS_ESTABLISHED);
1984                                 TCP_PROBE5(connect__established, NULL, tp,
1985                                     mtod(m, const char *), tp, th);
1986                                 cc_conn_init(tp);
1987                                 tcp_timer_activate(tp, TT_KEEP,
1988                                     TP_KEEPIDLE(tp));
1989                         }
1990                 } else {
1991                         /*
1992                          * Received initial SYN in SYN-SENT[*] state =>
1993                          * simultaneous open.
1994                          * If it succeeds, connection is * half-synchronized.
1995                          * Otherwise, do 3-way handshake:
1996                          *        SYN-SENT -> SYN-RECEIVED
1997                          *        SYN-SENT* -> SYN-RECEIVED*
1998                          */
1999                         tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2000                         tcp_timer_activate(tp, TT_REXMT, 0);
2001                         tcp_state_change(tp, TCPS_SYN_RECEIVED);
2002                 }
2003
2004                 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
2005                     "ti_locked %d", __func__, ti_locked));
2006                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2007                 INP_WLOCK_ASSERT(tp->t_inpcb);
2008
2009                 /*
2010                  * Advance th->th_seq to correspond to first data byte.
2011                  * If data, trim to stay within window,
2012                  * dropping FIN if necessary.
2013                  */
2014                 th->th_seq++;
2015                 if (tlen > tp->rcv_wnd) {
2016                         todrop = tlen - tp->rcv_wnd;
2017                         m_adj(m, -todrop);
2018                         tlen = tp->rcv_wnd;
2019                         thflags &= ~TH_FIN;
2020                         TCPSTAT_INC(tcps_rcvpackafterwin);
2021                         TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2022                 }
2023                 tp->snd_wl1 = th->th_seq - 1;
2024                 tp->rcv_up = th->th_seq;
2025                 /*
2026                  * Client side of transaction: already sent SYN and data.
2027                  * If the remote host used T/TCP to validate the SYN,
2028                  * our data will be ACK'd; if so, enter normal data segment
2029                  * processing in the middle of step 5, ack processing.
2030                  * Otherwise, goto step 6.
2031                  */
2032                 if (thflags & TH_ACK)
2033                         goto process_ACK;
2034
2035                 goto step6;
2036
2037         /*
2038          * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2039          *      do normal processing.
2040          *
2041          * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2042          */
2043         case TCPS_LAST_ACK:
2044         case TCPS_CLOSING:
2045                 break;  /* continue normal processing */
2046         }
2047
2048         /*
2049          * States other than LISTEN or SYN_SENT.
2050          * First check the RST flag and sequence number since reset segments
2051          * are exempt from the timestamp and connection count tests.  This
2052          * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2053          * below which allowed reset segments in half the sequence space
2054          * to fall though and be processed (which gives forged reset
2055          * segments with a random sequence number a 50 percent chance of
2056          * killing a connection).
2057          * Then check timestamp, if present.
2058          * Then check the connection count, if present.
2059          * Then check that at least some bytes of segment are within
2060          * receive window.  If segment begins before rcv_nxt,
2061          * drop leading data (and SYN); if nothing left, just ack.
2062          *
2063          *
2064          * If the RST bit is set, check the sequence number to see
2065          * if this is a valid reset segment.
2066          * RFC 793 page 37:
2067          *   In all states except SYN-SENT, all reset (RST) segments
2068          *   are validated by checking their SEQ-fields.  A reset is
2069          *   valid if its sequence number is in the window.
2070          * Note: this does not take into account delayed ACKs, so
2071          *   we should test against last_ack_sent instead of rcv_nxt.
2072          *   The sequence number in the reset segment is normally an
2073          *   echo of our outgoing acknowlegement numbers, but some hosts
2074          *   send a reset with the sequence number at the rightmost edge
2075          *   of our receive window, and we have to handle this case.
2076          * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2077          *   that brute force RST attacks are possible.  To combat this,
2078          *   we use a much stricter check while in the ESTABLISHED state,
2079          *   only accepting RSTs where the sequence number is equal to
2080          *   last_ack_sent.  In all other states (the states in which a
2081          *   RST is more likely), the more permissive check is used.
2082          * If we have multiple segments in flight, the initial reset
2083          * segment sequence numbers will be to the left of last_ack_sent,
2084          * but they will eventually catch up.
2085          * In any case, it never made sense to trim reset segments to
2086          * fit the receive window since RFC 1122 says:
2087          *   4.2.2.12  RST Segment: RFC-793 Section 3.4
2088          *
2089          *    A TCP SHOULD allow a received RST segment to include data.
2090          *
2091          *    DISCUSSION
2092          *         It has been suggested that a RST segment could contain
2093          *         ASCII text that encoded and explained the cause of the
2094          *         RST.  No standard has yet been established for such
2095          *         data.
2096          *
2097          * If the reset segment passes the sequence number test examine
2098          * the state:
2099          *    SYN_RECEIVED STATE:
2100          *      If passive open, return to LISTEN state.
2101          *      If active open, inform user that connection was refused.
2102          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2103          *      Inform user that connection was reset, and close tcb.
2104          *    CLOSING, LAST_ACK STATES:
2105          *      Close the tcb.
2106          *    TIME_WAIT STATE:
2107          *      Drop the segment - see Stevens, vol. 2, p. 964 and
2108          *      RFC 1337.
2109          */
2110         if (thflags & TH_RST) {
2111                 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2112                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2113                         switch (tp->t_state) {
2114
2115                         case TCPS_SYN_RECEIVED:
2116                                 so->so_error = ECONNREFUSED;
2117                                 goto close;
2118
2119                         case TCPS_ESTABLISHED:
2120                                 if (V_tcp_insecure_rst == 0 &&
2121                                     !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2122                                     SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2123                                     !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2124                                     SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2125                                         TCPSTAT_INC(tcps_badrst);
2126                                         goto drop;
2127                                 }
2128                                 /* FALLTHROUGH */
2129                         case TCPS_FIN_WAIT_1:
2130                         case TCPS_FIN_WAIT_2:
2131                         case TCPS_CLOSE_WAIT:
2132                                 so->so_error = ECONNRESET;
2133                         close:
2134                                 KASSERT(ti_locked == TI_WLOCKED,
2135                                     ("tcp_do_segment: TH_RST 1 ti_locked %d",
2136                                     ti_locked));
2137                                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2138
2139                                 tcp_state_change(tp, TCPS_CLOSED);
2140                                 TCPSTAT_INC(tcps_drops);
2141                                 tp = tcp_close(tp);
2142                                 break;
2143
2144                         case TCPS_CLOSING:
2145                         case TCPS_LAST_ACK:
2146                                 KASSERT(ti_locked == TI_WLOCKED,
2147                                     ("tcp_do_segment: TH_RST 2 ti_locked %d",
2148                                     ti_locked));
2149                                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2150
2151                                 tp = tcp_close(tp);
2152                                 break;
2153                         }
2154                 }
2155                 goto drop;
2156         }
2157
2158         /*
2159          * RFC 1323 PAWS: If we have a timestamp reply on this segment
2160          * and it's less than ts_recent, drop it.
2161          */
2162         if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2163             TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2164
2165                 /* Check to see if ts_recent is over 24 days old.  */
2166                 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2167                         /*
2168                          * Invalidate ts_recent.  If this segment updates
2169                          * ts_recent, the age will be reset later and ts_recent
2170                          * will get a valid value.  If it does not, setting
2171                          * ts_recent to zero will at least satisfy the
2172                          * requirement that zero be placed in the timestamp
2173                          * echo reply when ts_recent isn't valid.  The
2174                          * age isn't reset until we get a valid ts_recent
2175                          * because we don't want out-of-order segments to be
2176                          * dropped when ts_recent is old.
2177                          */
2178                         tp->ts_recent = 0;
2179                 } else {
2180                         TCPSTAT_INC(tcps_rcvduppack);
2181                         TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2182                         TCPSTAT_INC(tcps_pawsdrop);
2183                         if (tlen)
2184                                 goto dropafterack;
2185                         goto drop;
2186                 }
2187         }
2188
2189         /*
2190          * In the SYN-RECEIVED state, validate that the packet belongs to
2191          * this connection before trimming the data to fit the receive
2192          * window.  Check the sequence number versus IRS since we know
2193          * the sequence numbers haven't wrapped.  This is a partial fix
2194          * for the "LAND" DoS attack.
2195          */
2196         if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2197                 rstreason = BANDLIM_RST_OPENPORT;
2198                 goto dropwithreset;
2199         }
2200
2201         todrop = tp->rcv_nxt - th->th_seq;
2202         if (todrop > 0) {
2203                 /*
2204                  * If this is a duplicate SYN for our current connection,
2205                  * advance over it and pretend and it's not a SYN.
2206                  */
2207                 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2208                         thflags &= ~TH_SYN;
2209                         th->th_seq++;
2210                         if (th->th_urp > 1)
2211                                 th->th_urp--;
2212                         else
2213                                 thflags &= ~TH_URG;
2214                         todrop--;
2215                 }
2216                 /*
2217                  * Following if statement from Stevens, vol. 2, p. 960.
2218                  */
2219                 if (todrop > tlen
2220                     || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2221                         /*
2222                          * Any valid FIN must be to the left of the window.
2223                          * At this point the FIN must be a duplicate or out
2224                          * of sequence; drop it.
2225                          */
2226                         thflags &= ~TH_FIN;
2227
2228                         /*
2229                          * Send an ACK to resynchronize and drop any data.
2230                          * But keep on processing for RST or ACK.
2231                          */
2232                         tp->t_flags |= TF_ACKNOW;
2233                         todrop = tlen;
2234                         TCPSTAT_INC(tcps_rcvduppack);
2235                         TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2236                 } else {
2237                         TCPSTAT_INC(tcps_rcvpartduppack);
2238                         TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2239                 }
2240                 drop_hdrlen += todrop;  /* drop from the top afterwards */
2241                 th->th_seq += todrop;
2242                 tlen -= todrop;
2243                 if (th->th_urp > todrop)
2244                         th->th_urp -= todrop;
2245                 else {
2246                         thflags &= ~TH_URG;
2247                         th->th_urp = 0;
2248                 }
2249         }
2250
2251         /*
2252          * If new data are received on a connection after the
2253          * user processes are gone, then RST the other end.
2254          */
2255         if ((so->so_state & SS_NOFDREF) &&
2256             tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2257                 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2258                     "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2259                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2260
2261                 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2262                         log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2263                             "after socket was closed, "
2264                             "sending RST and removing tcpcb\n",
2265                             s, __func__, tcpstates[tp->t_state], tlen);
2266                         free(s, M_TCPLOG);
2267                 }
2268                 tp = tcp_close(tp);
2269                 TCPSTAT_INC(tcps_rcvafterclose);
2270                 rstreason = BANDLIM_UNLIMITED;
2271                 goto dropwithreset;
2272         }
2273
2274         /*
2275          * If segment ends after window, drop trailing data
2276          * (and PUSH and FIN); if nothing left, just ACK.
2277          */
2278         todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2279         if (todrop > 0) {
2280                 TCPSTAT_INC(tcps_rcvpackafterwin);
2281                 if (todrop >= tlen) {
2282                         TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2283                         /*
2284                          * If window is closed can only take segments at
2285                          * window edge, and have to drop data and PUSH from
2286                          * incoming segments.  Continue processing, but
2287                          * remember to ack.  Otherwise, drop segment
2288                          * and ack.
2289                          */
2290                         if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2291                                 tp->t_flags |= TF_ACKNOW;
2292                                 TCPSTAT_INC(tcps_rcvwinprobe);
2293                         } else
2294                                 goto dropafterack;
2295                 } else
2296                         TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2297                 m_adj(m, -todrop);
2298                 tlen -= todrop;
2299                 thflags &= ~(TH_PUSH|TH_FIN);
2300         }
2301
2302         /*
2303          * If last ACK falls within this segment's sequence numbers,
2304          * record its timestamp.
2305          * NOTE:
2306          * 1) That the test incorporates suggestions from the latest
2307          *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2308          * 2) That updating only on newer timestamps interferes with
2309          *    our earlier PAWS tests, so this check should be solely
2310          *    predicated on the sequence space of this segment.
2311          * 3) That we modify the segment boundary check to be
2312          *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2313          *    instead of RFC1323's
2314          *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2315          *    This modified check allows us to overcome RFC1323's
2316          *    limitations as described in Stevens TCP/IP Illustrated
2317          *    Vol. 2 p.869. In such cases, we can still calculate the
2318          *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2319          */
2320         if ((to.to_flags & TOF_TS) != 0 &&
2321             SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2322             SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2323                 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2324                 tp->ts_recent_age = tcp_ts_getticks();
2325                 tp->ts_recent = to.to_tsval;
2326         }
2327
2328         /*
2329          * If a SYN is in the window, then this is an
2330          * error and we send an RST and drop the connection.
2331          */
2332         if (thflags & TH_SYN) {
2333                 KASSERT(ti_locked == TI_WLOCKED,
2334                     ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2335                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2336
2337                 tp = tcp_drop(tp, ECONNRESET);
2338                 rstreason = BANDLIM_UNLIMITED;
2339                 goto drop;
2340         }
2341
2342         /*
2343          * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2344          * flag is on (half-synchronized state), then queue data for
2345          * later processing; else drop segment and return.
2346          */
2347         if ((thflags & TH_ACK) == 0) {
2348                 if (tp->t_state == TCPS_SYN_RECEIVED ||
2349                     (tp->t_flags & TF_NEEDSYN))
2350                         goto step6;
2351                 else if (tp->t_flags & TF_ACKNOW)
2352                         goto dropafterack;
2353                 else
2354                         goto drop;
2355         }
2356
2357         /*
2358          * Ack processing.
2359          */
2360         switch (tp->t_state) {
2361
2362         /*
2363          * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2364          * ESTABLISHED state and continue processing.
2365          * The ACK was checked above.
2366          */
2367         case TCPS_SYN_RECEIVED:
2368
2369                 TCPSTAT_INC(tcps_connects);
2370                 soisconnected(so);
2371                 /* Do window scaling? */
2372                 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2373                         (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2374                         tp->rcv_scale = tp->request_r_scale;
2375                         tp->snd_wnd = tiwin;
2376                 }
2377                 /*
2378                  * Make transitions:
2379                  *      SYN-RECEIVED  -> ESTABLISHED
2380                  *      SYN-RECEIVED* -> FIN-WAIT-1
2381                  */
2382                 tp->t_starttime = ticks;
2383                 if (tp->t_flags & TF_NEEDFIN) {
2384                         tcp_state_change(tp, TCPS_FIN_WAIT_1);
2385                         tp->t_flags &= ~TF_NEEDFIN;
2386                 } else {
2387                         tcp_state_change(tp, TCPS_ESTABLISHED);
2388                         TCP_PROBE5(accept__established, NULL, tp,
2389                             mtod(m, const char *), tp, th);
2390                         cc_conn_init(tp);
2391                         tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2392                 }
2393                 /*
2394                  * If segment contains data or ACK, will call tcp_reass()
2395                  * later; if not, do so now to pass queued data to user.
2396                  */
2397                 if (tlen == 0 && (thflags & TH_FIN) == 0)
2398                         (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2399                             (struct mbuf *)0);
2400                 tp->snd_wl1 = th->th_seq - 1;
2401                 /* FALLTHROUGH */
2402
2403         /*
2404          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2405          * ACKs.  If the ack is in the range
2406          *      tp->snd_una < th->th_ack <= tp->snd_max
2407          * then advance tp->snd_una to th->th_ack and drop
2408          * data from the retransmission queue.  If this ACK reflects
2409          * more up to date window information we update our window information.
2410          */
2411         case TCPS_ESTABLISHED:
2412         case TCPS_FIN_WAIT_1:
2413         case TCPS_FIN_WAIT_2:
2414         case TCPS_CLOSE_WAIT:
2415         case TCPS_CLOSING:
2416         case TCPS_LAST_ACK:
2417                 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2418                         TCPSTAT_INC(tcps_rcvacktoomuch);
2419                         goto dropafterack;
2420                 }
2421                 if ((tp->t_flags & TF_SACK_PERMIT) &&
2422                     ((to.to_flags & TOF_SACK) ||
2423                      !TAILQ_EMPTY(&tp->snd_holes)))
2424                         tcp_sack_doack(tp, &to, th->th_ack);
2425
2426                 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2427                 hhook_run_tcp_est_in(tp, th, &to);
2428
2429                 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2430                         if (tlen == 0 && tiwin == tp->snd_wnd) {
2431                                 /*
2432                                  * If this is the first time we've seen a
2433                                  * FIN from the remote, this is not a
2434                                  * duplicate and it needs to be processed
2435                                  * normally.  This happens during a
2436                                  * simultaneous close.
2437                                  */
2438                                 if ((thflags & TH_FIN) &&
2439                                     (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2440                                         tp->t_dupacks = 0;
2441                                         break;
2442                                 }
2443                                 TCPSTAT_INC(tcps_rcvdupack);
2444                                 /*
2445                                  * If we have outstanding data (other than
2446                                  * a window probe), this is a completely
2447                                  * duplicate ack (ie, window info didn't
2448                                  * change and FIN isn't set),
2449                                  * the ack is the biggest we've
2450                                  * seen and we've seen exactly our rexmt
2451                                  * threshhold of them, assume a packet
2452                                  * has been dropped and retransmit it.
2453                                  * Kludge snd_nxt & the congestion
2454                                  * window so we send only this one
2455                                  * packet.
2456                                  *
2457                                  * We know we're losing at the current
2458                                  * window size so do congestion avoidance
2459                                  * (set ssthresh to half the current window
2460                                  * and pull our congestion window back to
2461                                  * the new ssthresh).
2462                                  *
2463                                  * Dup acks mean that packets have left the
2464                                  * network (they're now cached at the receiver)
2465                                  * so bump cwnd by the amount in the receiver
2466                                  * to keep a constant cwnd packets in the
2467                                  * network.
2468                                  *
2469                                  * When using TCP ECN, notify the peer that
2470                                  * we reduced the cwnd.
2471                                  */
2472                                 if (!tcp_timer_active(tp, TT_REXMT) ||
2473                                     th->th_ack != tp->snd_una)
2474                                         tp->t_dupacks = 0;
2475                                 else if (++tp->t_dupacks > tcprexmtthresh ||
2476                                      IN_FASTRECOVERY(tp->t_flags)) {
2477                                         cc_ack_received(tp, th, CC_DUPACK);
2478                                         if ((tp->t_flags & TF_SACK_PERMIT) &&
2479                                             IN_FASTRECOVERY(tp->t_flags)) {
2480                                                 int awnd;
2481
2482                                                 /*
2483                                                  * Compute the amount of data in flight first.
2484                                                  * We can inject new data into the pipe iff
2485                                                  * we have less than 1/2 the original window's
2486                                                  * worth of data in flight.
2487                                                  */
2488                                                 awnd = (tp->snd_nxt - tp->snd_fack) +
2489                                                         tp->sackhint.sack_bytes_rexmit;
2490                                                 if (awnd < tp->snd_ssthresh) {
2491                                                         tp->snd_cwnd += tp->t_maxseg;
2492                                                         if (tp->snd_cwnd > tp->snd_ssthresh)
2493                                                                 tp->snd_cwnd = tp->snd_ssthresh;
2494                                                 }
2495                                         } else
2496                                                 tp->snd_cwnd += tp->t_maxseg;
2497                                         (void) tcp_output(tp);
2498                                         goto drop;
2499                                 } else if (tp->t_dupacks == tcprexmtthresh) {
2500                                         tcp_seq onxt = tp->snd_nxt;
2501
2502                                         /*
2503                                          * If we're doing sack, check to
2504                                          * see if we're already in sack
2505                                          * recovery. If we're not doing sack,
2506                                          * check to see if we're in newreno
2507                                          * recovery.
2508                                          */
2509                                         if (tp->t_flags & TF_SACK_PERMIT) {
2510                                                 if (IN_FASTRECOVERY(tp->t_flags)) {
2511                                                         tp->t_dupacks = 0;
2512                                                         break;
2513                                                 }
2514                                         } else {
2515                                                 if (SEQ_LEQ(th->th_ack,
2516                                                     tp->snd_recover)) {
2517                                                         tp->t_dupacks = 0;
2518                                                         break;
2519                                                 }
2520                                         }
2521                                         /* Congestion signal before ack. */
2522                                         cc_cong_signal(tp, th, CC_NDUPACK);
2523                                         cc_ack_received(tp, th, CC_DUPACK);
2524                                         tcp_timer_activate(tp, TT_REXMT, 0);
2525                                         tp->t_rtttime = 0;
2526                                         if (tp->t_flags & TF_SACK_PERMIT) {
2527                                                 TCPSTAT_INC(
2528                                                     tcps_sack_recovery_episode);
2529                                                 tp->sack_newdata = tp->snd_nxt;
2530                                                 tp->snd_cwnd = tp->t_maxseg;
2531                                                 (void) tcp_output(tp);
2532                                                 goto drop;
2533                                         }
2534                                         tp->snd_nxt = th->th_ack;
2535                                         tp->snd_cwnd = tp->t_maxseg;
2536                                         (void) tcp_output(tp);
2537                                         KASSERT(tp->snd_limited <= 2,
2538                                             ("%s: tp->snd_limited too big",
2539                                             __func__));
2540                                         tp->snd_cwnd = tp->snd_ssthresh +
2541                                              tp->t_maxseg *
2542                                              (tp->t_dupacks - tp->snd_limited);
2543                                         if (SEQ_GT(onxt, tp->snd_nxt))
2544                                                 tp->snd_nxt = onxt;
2545                                         goto drop;
2546                                 } else if (V_tcp_do_rfc3042) {
2547                                         cc_ack_received(tp, th, CC_DUPACK);
2548                                         u_long oldcwnd = tp->snd_cwnd;
2549                                         tcp_seq oldsndmax = tp->snd_max;
2550                                         u_int sent;
2551                                         int avail;
2552
2553                                         KASSERT(tp->t_dupacks == 1 ||
2554                                             tp->t_dupacks == 2,
2555                                             ("%s: dupacks not 1 or 2",
2556                                             __func__));
2557                                         if (tp->t_dupacks == 1)
2558                                                 tp->snd_limited = 0;
2559                                         tp->snd_cwnd =
2560                                             (tp->snd_nxt - tp->snd_una) +
2561                                             (tp->t_dupacks - tp->snd_limited) *
2562                                             tp->t_maxseg;
2563                                         /*
2564                                          * Only call tcp_output when there
2565                                          * is new data available to be sent.
2566                                          * Otherwise we would send pure ACKs.
2567                                          */
2568                                         SOCKBUF_LOCK(&so->so_snd);
2569                                         avail = so->so_snd.sb_cc -
2570                                             (tp->snd_nxt - tp->snd_una);
2571                                         SOCKBUF_UNLOCK(&so->so_snd);
2572                                         if (avail > 0)
2573                                                 (void) tcp_output(tp);
2574                                         sent = tp->snd_max - oldsndmax;
2575                                         if (sent > tp->t_maxseg) {
2576                                                 KASSERT((tp->t_dupacks == 2 &&
2577                                                     tp->snd_limited == 0) ||
2578                                                    (sent == tp->t_maxseg + 1 &&
2579                                                     tp->t_flags & TF_SENTFIN),
2580                                                     ("%s: sent too much",
2581                                                     __func__));
2582                                                 tp->snd_limited = 2;
2583                                         } else if (sent > 0)
2584                                                 ++tp->snd_limited;
2585                                         tp->snd_cwnd = oldcwnd;
2586                                         goto drop;
2587                                 }
2588                         } else
2589                                 tp->t_dupacks = 0;
2590                         break;
2591                 }
2592
2593                 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2594                     ("%s: th_ack <= snd_una", __func__));
2595
2596                 /*
2597                  * If the congestion window was inflated to account
2598                  * for the other side's cached packets, retract it.
2599                  */
2600                 if (IN_FASTRECOVERY(tp->t_flags)) {
2601                         if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2602                                 if (tp->t_flags & TF_SACK_PERMIT)
2603                                         tcp_sack_partialack(tp, th);
2604                                 else
2605                                         tcp_newreno_partial_ack(tp, th);
2606                         } else
2607                                 cc_post_recovery(tp, th);
2608                 }
2609                 tp->t_dupacks = 0;
2610                 /*
2611                  * If we reach this point, ACK is not a duplicate,
2612                  *     i.e., it ACKs something we sent.
2613                  */
2614                 if (tp->t_flags & TF_NEEDSYN) {
2615                         /*
2616                          * T/TCP: Connection was half-synchronized, and our
2617                          * SYN has been ACK'd (so connection is now fully
2618                          * synchronized).  Go to non-starred state,
2619                          * increment snd_una for ACK of SYN, and check if
2620                          * we can do window scaling.
2621                          */
2622                         tp->t_flags &= ~TF_NEEDSYN;
2623                         tp->snd_una++;
2624                         /* Do window scaling? */
2625                         if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2626                                 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2627                                 tp->rcv_scale = tp->request_r_scale;
2628                                 /* Send window already scaled. */
2629                         }
2630                 }
2631
2632 process_ACK:
2633                 INP_WLOCK_ASSERT(tp->t_inpcb);
2634
2635                 acked = BYTES_THIS_ACK(tp, th);
2636                 TCPSTAT_INC(tcps_rcvackpack);
2637                 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2638
2639                 /*
2640                  * If we just performed our first retransmit, and the ACK
2641                  * arrives within our recovery window, then it was a mistake
2642                  * to do the retransmit in the first place.  Recover our
2643                  * original cwnd and ssthresh, and proceed to transmit where
2644                  * we left off.
2645                  */
2646                 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2647                     (int)(ticks - tp->t_badrxtwin) < 0)
2648                         cc_cong_signal(tp, th, CC_RTO_ERR);
2649
2650                 /*
2651                  * If we have a timestamp reply, update smoothed
2652                  * round trip time.  If no timestamp is present but
2653                  * transmit timer is running and timed sequence
2654                  * number was acked, update smoothed round trip time.
2655                  * Since we now have an rtt measurement, cancel the
2656                  * timer backoff (cf., Phil Karn's retransmit alg.).
2657                  * Recompute the initial retransmit timer.
2658                  *
2659                  * Some boxes send broken timestamp replies
2660                  * during the SYN+ACK phase, ignore
2661                  * timestamps of 0 or we could calculate a
2662                  * huge RTT and blow up the retransmit timer.
2663                  */
2664                 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2665                         u_int t;
2666
2667                         t = tcp_ts_getticks() - to.to_tsecr;
2668                         if (!tp->t_rttlow || tp->t_rttlow > t)
2669                                 tp->t_rttlow = t;
2670                         tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2671                 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2672                         if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2673                                 tp->t_rttlow = ticks - tp->t_rtttime;
2674                         tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2675                 }
2676
2677                 /*
2678                  * If all outstanding data is acked, stop retransmit
2679                  * timer and remember to restart (more output or persist).
2680                  * If there is more data to be acked, restart retransmit
2681                  * timer, using current (possibly backed-off) value.
2682                  */
2683                 if (th->th_ack == tp->snd_max) {
2684                         tcp_timer_activate(tp, TT_REXMT, 0);
2685                         needoutput = 1;
2686                 } else if (!tcp_timer_active(tp, TT_PERSIST))
2687                         tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2688
2689                 /*
2690                  * If no data (only SYN) was ACK'd,
2691                  *    skip rest of ACK processing.
2692                  */
2693                 if (acked == 0)
2694                         goto step6;
2695
2696                 /*
2697                  * Let the congestion control algorithm update congestion
2698                  * control related information. This typically means increasing
2699                  * the congestion window.
2700                  */
2701                 cc_ack_received(tp, th, CC_ACK);
2702
2703                 SOCKBUF_LOCK(&so->so_snd);
2704                 if (acked > so->so_snd.sb_cc) {
2705                         tp->snd_wnd -= so->so_snd.sb_cc;
2706                         mfree = sbcut_locked(&so->so_snd,
2707                             (int)so->so_snd.sb_cc);
2708                         ourfinisacked = 1;
2709                 } else {
2710                         mfree = sbcut_locked(&so->so_snd, acked);
2711                         tp->snd_wnd -= acked;
2712                         ourfinisacked = 0;
2713                 }
2714                 /* NB: sowwakeup_locked() does an implicit unlock. */
2715                 sowwakeup_locked(so);
2716                 m_freem(mfree);
2717                 /* Detect una wraparound. */
2718                 if (!IN_RECOVERY(tp->t_flags) &&
2719                     SEQ_GT(tp->snd_una, tp->snd_recover) &&
2720                     SEQ_LEQ(th->th_ack, tp->snd_recover))
2721                         tp->snd_recover = th->th_ack - 1;
2722                 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2723                 if (IN_RECOVERY(tp->t_flags) &&
2724                     SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2725                         EXIT_RECOVERY(tp->t_flags);
2726                 }
2727                 tp->snd_una = th->th_ack;
2728                 if (tp->t_flags & TF_SACK_PERMIT) {
2729                         if (SEQ_GT(tp->snd_una, tp->snd_recover))
2730                                 tp->snd_recover = tp->snd_una;
2731                 }
2732                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2733                         tp->snd_nxt = tp->snd_una;
2734
2735                 switch (tp->t_state) {
2736
2737                 /*
2738                  * In FIN_WAIT_1 STATE in addition to the processing
2739                  * for the ESTABLISHED state if our FIN is now acknowledged
2740                  * then enter FIN_WAIT_2.
2741                  */
2742                 case TCPS_FIN_WAIT_1:
2743                         if (ourfinisacked) {
2744                                 /*
2745                                  * If we can't receive any more
2746                                  * data, then closing user can proceed.
2747                                  * Starting the timer is contrary to the
2748                                  * specification, but if we don't get a FIN
2749                                  * we'll hang forever.
2750                                  *
2751                                  * XXXjl:
2752                                  * we should release the tp also, and use a
2753                                  * compressed state.
2754                                  */
2755                                 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2756                                         soisdisconnected(so);
2757                                         tcp_timer_activate(tp, TT_2MSL,
2758                                             (tcp_fast_finwait2_recycle ?
2759                                             tcp_finwait2_timeout :
2760                                             TP_MAXIDLE(tp)));
2761                                 }
2762                                 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2763                         }
2764                         break;
2765
2766                 /*
2767                  * In CLOSING STATE in addition to the processing for
2768                  * the ESTABLISHED state if the ACK acknowledges our FIN
2769                  * then enter the TIME-WAIT state, otherwise ignore
2770                  * the segment.
2771                  */
2772                 case TCPS_CLOSING:
2773                         if (ourfinisacked) {
2774                                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2775                                 tcp_twstart(tp);
2776                                 INP_INFO_WUNLOCK(&V_tcbinfo);
2777                                 m_freem(m);
2778                                 return;
2779                         }
2780                         break;
2781
2782                 /*
2783                  * In LAST_ACK, we may still be waiting for data to drain
2784                  * and/or to be acked, as well as for the ack of our FIN.
2785                  * If our FIN is now acknowledged, delete the TCB,
2786                  * enter the closed state and return.
2787                  */
2788                 case TCPS_LAST_ACK:
2789                         if (ourfinisacked) {
2790                                 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2791                                 tp = tcp_close(tp);
2792                                 goto drop;
2793                         }
2794                         break;
2795                 }
2796         }
2797
2798 step6:
2799         INP_WLOCK_ASSERT(tp->t_inpcb);
2800
2801         /*
2802          * Update window information.
2803          * Don't look at window if no ACK: TAC's send garbage on first SYN.
2804          */
2805         if ((thflags & TH_ACK) &&
2806             (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2807             (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2808              (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2809                 /* keep track of pure window updates */
2810                 if (tlen == 0 &&
2811                     tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2812                         TCPSTAT_INC(tcps_rcvwinupd);
2813                 tp->snd_wnd = tiwin;
2814                 tp->snd_wl1 = th->th_seq;
2815                 tp->snd_wl2 = th->th_ack;
2816                 if (tp->snd_wnd > tp->max_sndwnd)
2817                         tp->max_sndwnd = tp->snd_wnd;
2818                 needoutput = 1;
2819         }
2820
2821         /*
2822          * Process segments with URG.
2823          */
2824         if ((thflags & TH_URG) && th->th_urp &&
2825             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2826                 /*
2827                  * This is a kludge, but if we receive and accept
2828                  * random urgent pointers, we'll crash in
2829                  * soreceive.  It's hard to imagine someone
2830                  * actually wanting to send this much urgent data.
2831                  */
2832                 SOCKBUF_LOCK(&so->so_rcv);
2833                 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2834                         th->th_urp = 0;                 /* XXX */
2835                         thflags &= ~TH_URG;             /* XXX */
2836                         SOCKBUF_UNLOCK(&so->so_rcv);    /* XXX */
2837                         goto dodata;                    /* XXX */
2838                 }
2839                 /*
2840                  * If this segment advances the known urgent pointer,
2841                  * then mark the data stream.  This should not happen
2842                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2843                  * a FIN has been received from the remote side.
2844                  * In these states we ignore the URG.
2845                  *
2846                  * According to RFC961 (Assigned Protocols),
2847                  * the urgent pointer points to the last octet
2848                  * of urgent data.  We continue, however,
2849                  * to consider it to indicate the first octet
2850                  * of data past the urgent section as the original
2851                  * spec states (in one of two places).
2852                  */
2853                 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2854                         tp->rcv_up = th->th_seq + th->th_urp;
2855                         so->so_oobmark = so->so_rcv.sb_cc +
2856                             (tp->rcv_up - tp->rcv_nxt) - 1;
2857                         if (so->so_oobmark == 0)
2858                                 so->so_rcv.sb_state |= SBS_RCVATMARK;
2859                         sohasoutofband(so);
2860                         tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2861                 }
2862                 SOCKBUF_UNLOCK(&so->so_rcv);
2863                 /*
2864                  * Remove out of band data so doesn't get presented to user.
2865                  * This can happen independent of advancing the URG pointer,
2866                  * but if two URG's are pending at once, some out-of-band
2867                  * data may creep in... ick.
2868                  */
2869                 if (th->th_urp <= (u_long)tlen &&
2870                     !(so->so_options & SO_OOBINLINE)) {
2871                         /* hdr drop is delayed */
2872                         tcp_pulloutofband(so, th, m, drop_hdrlen);
2873                 }
2874         } else {
2875                 /*
2876                  * If no out of band data is expected,
2877                  * pull receive urgent pointer along
2878                  * with the receive window.
2879                  */
2880                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2881                         tp->rcv_up = tp->rcv_nxt;
2882         }
2883 dodata:                                                 /* XXX */
2884         INP_WLOCK_ASSERT(tp->t_inpcb);
2885
2886         /*
2887          * Process the segment text, merging it into the TCP sequencing queue,
2888          * and arranging for acknowledgment of receipt if necessary.
2889          * This process logically involves adjusting tp->rcv_wnd as data
2890          * is presented to the user (this happens in tcp_usrreq.c,
2891          * case PRU_RCVD).  If a FIN has already been received on this
2892          * connection then we just ignore the text.
2893          */
2894         if ((tlen || (thflags & TH_FIN)) &&
2895             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2896                 tcp_seq save_start = th->th_seq;
2897                 m_adj(m, drop_hdrlen);  /* delayed header drop */
2898                 /*
2899                  * Insert segment which includes th into TCP reassembly queue
2900                  * with control block tp.  Set thflags to whether reassembly now
2901                  * includes a segment with FIN.  This handles the common case
2902                  * inline (segment is the next to be received on an established
2903                  * connection, and the queue is empty), avoiding linkage into
2904                  * and removal from the queue and repetition of various
2905                  * conversions.
2906                  * Set DELACK for segments received in order, but ack
2907                  * immediately when segments are out of order (so
2908                  * fast retransmit can work).
2909                  */
2910                 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL &&
2911                     TCPS_HAVEESTABLISHED(tp->t_state)) {
2912                         if (DELAY_ACK(tp, tlen))
2913                                 tp->t_flags |= TF_DELACK;
2914                         else
2915                                 tp->t_flags |= TF_ACKNOW;
2916                         tp->rcv_nxt += tlen;
2917                         thflags = th->th_flags & TH_FIN;
2918                         TCPSTAT_INC(tcps_rcvpack);
2919                         TCPSTAT_ADD(tcps_rcvbyte, tlen);
2920                         ND6_HINT(tp);
2921                         SOCKBUF_LOCK(&so->so_rcv);
2922                         if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2923                                 m_freem(m);
2924                         else
2925                                 sbappendstream_locked(&so->so_rcv, m);
2926                         /* NB: sorwakeup_locked() does an implicit unlock. */
2927                         sorwakeup_locked(so);
2928                 } else {
2929                         /*
2930                          * XXX: Due to the header drop above "th" is
2931                          * theoretically invalid by now.  Fortunately
2932                          * m_adj() doesn't actually frees any mbufs
2933                          * when trimming from the head.
2934                          */
2935                         thflags = tcp_reass(tp, th, &tlen, m);
2936                         tp->t_flags |= TF_ACKNOW;
2937                 }
2938                 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2939                         tcp_update_sack_list(tp, save_start, save_start + tlen);
2940 #if 0
2941                 /*
2942                  * Note the amount of data that peer has sent into
2943                  * our window, in order to estimate the sender's
2944                  * buffer size.
2945                  * XXX: Unused.
2946                  */
2947                 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2948                         len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2949                 else
2950                         len = so->so_rcv.sb_hiwat;
2951 #endif
2952         } else {
2953                 m_freem(m);
2954                 thflags &= ~TH_FIN;
2955         }
2956
2957         /*
2958          * If FIN is received ACK the FIN and let the user know
2959          * that the connection is closing.
2960          */
2961         if (thflags & TH_FIN) {
2962                 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2963                         socantrcvmore(so);
2964                         /*
2965                          * If connection is half-synchronized
2966                          * (ie NEEDSYN flag on) then delay ACK,
2967                          * so it may be piggybacked when SYN is sent.
2968                          * Otherwise, since we received a FIN then no
2969                          * more input can be expected, send ACK now.
2970                          */
2971                         if (tp->t_flags & TF_NEEDSYN)
2972                                 tp->t_flags |= TF_DELACK;
2973                         else
2974                                 tp->t_flags |= TF_ACKNOW;
2975                         tp->rcv_nxt++;
2976                 }
2977                 switch (tp->t_state) {
2978
2979                 /*
2980                  * In SYN_RECEIVED and ESTABLISHED STATES
2981                  * enter the CLOSE_WAIT state.
2982                  */
2983                 case TCPS_SYN_RECEIVED:
2984                         tp->t_starttime = ticks;
2985                         /* FALLTHROUGH */
2986                 case TCPS_ESTABLISHED:
2987                         tcp_state_change(tp, TCPS_CLOSE_WAIT);
2988                         break;
2989
2990                 /*
2991                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
2992                  * enter the CLOSING state.
2993                  */
2994                 case TCPS_FIN_WAIT_1:
2995                         tcp_state_change(tp, TCPS_CLOSING);
2996                         break;
2997
2998                 /*
2999                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
3000                  * starting the time-wait timer, turning off the other
3001                  * standard timers.
3002                  */
3003                 case TCPS_FIN_WAIT_2:
3004                         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
3005                         KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
3006                             "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3007                             ti_locked));
3008
3009                         tcp_twstart(tp);
3010                         INP_INFO_WUNLOCK(&V_tcbinfo);
3011                         return;
3012                 }
3013         }
3014         if (ti_locked == TI_WLOCKED)
3015                 INP_INFO_WUNLOCK(&V_tcbinfo);
3016         ti_locked = TI_UNLOCKED;
3017
3018 #ifdef TCPDEBUG
3019         if (so->so_options & SO_DEBUG)
3020                 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3021                           &tcp_savetcp, 0);
3022 #endif
3023
3024         /*
3025          * Return any desired output.
3026          */
3027         if (needoutput || (tp->t_flags & TF_ACKNOW))
3028                 (void) tcp_output(tp);
3029
3030 check_delack:
3031         KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3032             __func__, ti_locked));
3033         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3034         INP_WLOCK_ASSERT(tp->t_inpcb);
3035
3036         if (tp->t_flags & TF_DELACK) {
3037                 tp->t_flags &= ~TF_DELACK;
3038                 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3039         }
3040         INP_WUNLOCK(tp->t_inpcb);
3041         return;
3042
3043 dropafterack:
3044         /*
3045          * Generate an ACK dropping incoming segment if it occupies
3046          * sequence space, where the ACK reflects our state.
3047          *
3048          * We can now skip the test for the RST flag since all
3049          * paths to this code happen after packets containing
3050          * RST have been dropped.
3051          *
3052          * In the SYN-RECEIVED state, don't send an ACK unless the
3053          * segment we received passes the SYN-RECEIVED ACK test.
3054          * If it fails send a RST.  This breaks the loop in the
3055          * "LAND" DoS attack, and also prevents an ACK storm
3056          * between two listening ports that have been sent forged
3057          * SYN segments, each with the source address of the other.
3058          */
3059         if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3060             (SEQ_GT(tp->snd_una, th->th_ack) ||
3061              SEQ_GT(th->th_ack, tp->snd_max)) ) {
3062                 rstreason = BANDLIM_RST_OPENPORT;
3063                 goto dropwithreset;
3064         }
3065 #ifdef TCPDEBUG
3066         if (so->so_options & SO_DEBUG)
3067                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3068                           &tcp_savetcp, 0);
3069 #endif
3070         if (ti_locked == TI_WLOCKED)
3071                 INP_INFO_WUNLOCK(&V_tcbinfo);
3072         ti_locked = TI_UNLOCKED;
3073
3074         tp->t_flags |= TF_ACKNOW;
3075         (void) tcp_output(tp);
3076         INP_WUNLOCK(tp->t_inpcb);
3077         m_freem(m);
3078         return;
3079
3080 dropwithreset:
3081         if (ti_locked == TI_WLOCKED)
3082                 INP_INFO_WUNLOCK(&V_tcbinfo);
3083         ti_locked = TI_UNLOCKED;
3084
3085         if (tp != NULL) {
3086                 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3087                 INP_WUNLOCK(tp->t_inpcb);
3088         } else
3089                 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3090         return;
3091
3092 drop:
3093         if (ti_locked == TI_WLOCKED) {
3094                 INP_INFO_WUNLOCK(&V_tcbinfo);
3095                 ti_locked = TI_UNLOCKED;
3096         }
3097 #ifdef INVARIANTS
3098         else
3099                 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3100 #endif
3101
3102         /*
3103          * Drop space held by incoming segment and return.
3104          */
3105 #ifdef TCPDEBUG
3106         if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3107                 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3108                           &tcp_savetcp, 0);
3109 #endif
3110         if (tp != NULL)
3111                 INP_WUNLOCK(tp->t_inpcb);
3112         m_freem(m);
3113 }
3114
3115 /*
3116  * Issue RST and make ACK acceptable to originator of segment.
3117  * The mbuf must still include the original packet header.
3118  * tp may be NULL.
3119  */
3120 static void
3121 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3122     int tlen, int rstreason)
3123 {
3124 #ifdef INET
3125         struct ip *ip;
3126 #endif
3127 #ifdef INET6
3128         struct ip6_hdr *ip6;
3129 #endif
3130
3131         if (tp != NULL) {
3132                 INP_WLOCK_ASSERT(tp->t_inpcb);
3133         }
3134
3135         /* Don't bother if destination was broadcast/multicast. */
3136         if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3137                 goto drop;
3138 #ifdef INET6
3139         if (mtod(m, struct ip *)->ip_v == 6) {
3140                 ip6 = mtod(m, struct ip6_hdr *);
3141                 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3142                     IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3143                         goto drop;
3144                 /* IPv6 anycast check is done at tcp6_input() */
3145         }
3146 #endif
3147 #if defined(INET) && defined(INET6)
3148         else
3149 #endif
3150 #ifdef INET
3151         {
3152                 ip = mtod(m, struct ip *);
3153                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3154                     IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3155                     ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3156                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3157                         goto drop;
3158         }
3159 #endif
3160
3161         /* Perform bandwidth limiting. */
3162         if (badport_bandlim(rstreason) < 0)
3163                 goto drop;
3164
3165         /* tcp_respond consumes the mbuf chain. */
3166         if (th->th_flags & TH_ACK) {
3167                 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3168                     th->th_ack, TH_RST);
3169         } else {
3170                 if (th->th_flags & TH_SYN)
3171                         tlen++;
3172                 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3173                     (tcp_seq)0, TH_RST|TH_ACK);
3174         }
3175         return;
3176 drop:
3177         m_freem(m);
3178 }
3179
3180 /*
3181  * Parse TCP options and place in tcpopt.
3182  */
3183 static void
3184 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3185 {
3186         int opt, optlen;
3187
3188         to->to_flags = 0;
3189         for (; cnt > 0; cnt -= optlen, cp += optlen) {
3190                 opt = cp[0];
3191                 if (opt == TCPOPT_EOL)
3192                         break;
3193                 if (opt == TCPOPT_NOP)
3194                         optlen = 1;
3195                 else {
3196                         if (cnt < 2)
3197                                 break;
3198                         optlen = cp[1];
3199                         if (optlen < 2 || optlen > cnt)
3200                                 break;
3201                 }
3202                 switch (opt) {
3203                 case TCPOPT_MAXSEG:
3204                         if (optlen != TCPOLEN_MAXSEG)
3205                                 continue;
3206                         if (!(flags & TO_SYN))
3207                                 continue;
3208                         to->to_flags |= TOF_MSS;
3209                         bcopy((char *)cp + 2,
3210                             (char *)&to->to_mss, sizeof(to->to_mss));
3211                         to->to_mss = ntohs(to->to_mss);
3212                         break;
3213                 case TCPOPT_WINDOW:
3214                         if (optlen != TCPOLEN_WINDOW)
3215                                 continue;
3216                         if (!(flags & TO_SYN))
3217                                 continue;
3218                         to->to_flags |= TOF_SCALE;
3219                         to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3220                         break;
3221                 case TCPOPT_TIMESTAMP:
3222                         if (optlen != TCPOLEN_TIMESTAMP)
3223                                 continue;
3224                         to->to_flags |= TOF_TS;
3225                         bcopy((char *)cp + 2,
3226                             (char *)&to->to_tsval, sizeof(to->to_tsval));
3227                         to->to_tsval = ntohl(to->to_tsval);
3228                         bcopy((char *)cp + 6,
3229                             (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3230                         to->to_tsecr = ntohl(to->to_tsecr);
3231                         break;
3232 #ifdef TCP_SIGNATURE
3233                 /*
3234                  * XXX In order to reply to a host which has set the
3235                  * TCP_SIGNATURE option in its initial SYN, we have to
3236                  * record the fact that the option was observed here
3237                  * for the syncache code to perform the correct response.
3238                  */
3239                 case TCPOPT_SIGNATURE:
3240                         if (optlen != TCPOLEN_SIGNATURE)
3241                                 continue;
3242                         to->to_flags |= TOF_SIGNATURE;
3243                         to->to_signature = cp + 2;
3244                         break;
3245 #endif
3246                 case TCPOPT_SACK_PERMITTED:
3247                         if (optlen != TCPOLEN_SACK_PERMITTED)
3248                                 continue;
3249                         if (!(flags & TO_SYN))
3250                                 continue;
3251                         if (!V_tcp_do_sack)
3252                                 continue;
3253                         to->to_flags |= TOF_SACKPERM;
3254                         break;
3255                 case TCPOPT_SACK:
3256                         if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3257                                 continue;
3258                         if (flags & TO_SYN)
3259                                 continue;
3260                         to->to_flags |= TOF_SACK;
3261                         to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3262                         to->to_sacks = cp + 2;
3263                         TCPSTAT_INC(tcps_sack_rcv_blocks);
3264                         break;
3265                 default:
3266                         continue;
3267                 }
3268         }
3269 }
3270
3271 /*
3272  * Pull out of band byte out of a segment so
3273  * it doesn't appear in the user's data queue.
3274  * It is still reflected in the segment length for
3275  * sequencing purposes.
3276  */
3277 static void
3278 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3279     int off)
3280 {
3281         int cnt = off + th->th_urp - 1;
3282
3283         while (cnt >= 0) {
3284                 if (m->m_len > cnt) {
3285                         char *cp = mtod(m, caddr_t) + cnt;
3286                         struct tcpcb *tp = sototcpcb(so);
3287
3288                         INP_WLOCK_ASSERT(tp->t_inpcb);
3289
3290                         tp->t_iobc = *cp;
3291                         tp->t_oobflags |= TCPOOB_HAVEDATA;
3292                         bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3293                         m->m_len--;
3294                         if (m->m_flags & M_PKTHDR)
3295                                 m->m_pkthdr.len--;
3296                         return;
3297                 }
3298                 cnt -= m->m_len;
3299                 m = m->m_next;
3300                 if (m == NULL)
3301                         break;
3302         }
3303         panic("tcp_pulloutofband");
3304 }
3305
3306 /*
3307  * Collect new round-trip time estimate
3308  * and update averages and current timeout.
3309  */
3310 static void
3311 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3312 {
3313         int delta;
3314
3315         INP_WLOCK_ASSERT(tp->t_inpcb);
3316
3317         TCPSTAT_INC(tcps_rttupdated);
3318         tp->t_rttupdated++;
3319         if (tp->t_srtt != 0) {
3320                 /*
3321                  * srtt is stored as fixed point with 5 bits after the
3322                  * binary point (i.e., scaled by 8).  The following magic
3323                  * is equivalent to the smoothing algorithm in rfc793 with
3324                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3325                  * point).  Adjust rtt to origin 0.
3326                  */
3327                 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3328                         - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3329
3330                 if ((tp->t_srtt += delta) <= 0)
3331                         tp->t_srtt = 1;
3332
3333                 /*
3334                  * We accumulate a smoothed rtt variance (actually, a
3335                  * smoothed mean difference), then set the retransmit
3336                  * timer to smoothed rtt + 4 times the smoothed variance.
3337                  * rttvar is stored as fixed point with 4 bits after the
3338                  * binary point (scaled by 16).  The following is
3339                  * equivalent to rfc793 smoothing with an alpha of .75
3340                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3341                  * rfc793's wired-in beta.
3342                  */
3343                 if (delta < 0)
3344                         delta = -delta;
3345                 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3346                 if ((tp->t_rttvar += delta) <= 0)
3347                         tp->t_rttvar = 1;
3348                 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3349                     tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3350         } else {
3351                 /*
3352                  * No rtt measurement yet - use the unsmoothed rtt.
3353                  * Set the variance to half the rtt (so our first
3354                  * retransmit happens at 3*rtt).
3355                  */
3356                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3357                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3358                 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3359         }
3360         tp->t_rtttime = 0;
3361         tp->t_rxtshift = 0;
3362
3363         /*
3364          * the retransmit should happen at rtt + 4 * rttvar.
3365          * Because of the way we do the smoothing, srtt and rttvar
3366          * will each average +1/2 tick of bias.  When we compute
3367          * the retransmit timer, we want 1/2 tick of rounding and
3368          * 1 extra tick because of +-1/2 tick uncertainty in the
3369          * firing of the timer.  The bias will give us exactly the
3370          * 1.5 tick we need.  But, because the bias is
3371          * statistical, we have to test that we don't drop below
3372          * the minimum feasible timer (which is 2 ticks).
3373          */
3374         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3375                       max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3376
3377         /*
3378          * We received an ack for a packet that wasn't retransmitted;
3379          * it is probably safe to discard any error indications we've
3380          * received recently.  This isn't quite right, but close enough
3381          * for now (a route might have failed after we sent a segment,
3382          * and the return path might not be symmetrical).
3383          */
3384         tp->t_softerror = 0;
3385 }
3386
3387 /*
3388  * Determine a reasonable value for maxseg size.
3389  * If the route is known, check route for mtu.
3390  * If none, use an mss that can be handled on the outgoing interface
3391  * without forcing IP to fragment.  If no route is found, route has no mtu,
3392  * or the destination isn't local, use a default, hopefully conservative
3393  * size (usually 512 or the default IP max size, but no more than the mtu
3394  * of the interface), as we can't discover anything about intervening
3395  * gateways or networks.  We also initialize the congestion/slow start
3396  * window to be a single segment if the destination isn't local.
3397  * While looking at the routing entry, we also initialize other path-dependent
3398  * parameters from pre-set or cached values in the routing entry.
3399  *
3400  * Also take into account the space needed for options that we
3401  * send regularly.  Make maxseg shorter by that amount to assure
3402  * that we can send maxseg amount of data even when the options
3403  * are present.  Store the upper limit of the length of options plus
3404  * data in maxopd.
3405  *
3406  * NOTE that this routine is only called when we process an incoming
3407  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3408  * settings are handled in tcp_mssopt().
3409  */
3410 void
3411 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3412     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3413 {
3414         int mss = 0;
3415         u_long maxmtu = 0;
3416         struct inpcb *inp = tp->t_inpcb;
3417         struct hc_metrics_lite metrics;
3418         int origoffer;
3419 #ifdef INET6
3420         int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3421         size_t min_protoh = isipv6 ?
3422                             sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3423                             sizeof (struct tcpiphdr);
3424 #else
3425         const size_t min_protoh = sizeof(struct tcpiphdr);
3426 #endif
3427
3428         INP_WLOCK_ASSERT(tp->t_inpcb);
3429
3430         if (mtuoffer != -1) {
3431                 KASSERT(offer == -1, ("%s: conflict", __func__));
3432                 offer = mtuoffer - min_protoh;
3433         }
3434         origoffer = offer;
3435
3436         /* Initialize. */
3437 #ifdef INET6
3438         if (isipv6) {
3439                 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3440                 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3441         }
3442 #endif
3443 #if defined(INET) && defined(INET6)
3444         else
3445 #endif
3446 #ifdef INET
3447         {
3448                 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3449                 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3450         }
3451 #endif
3452
3453         /*
3454          * No route to sender, stay with default mss and return.
3455          */
3456         if (maxmtu == 0) {
3457                 /*
3458                  * In case we return early we need to initialize metrics
3459                  * to a defined state as tcp_hc_get() would do for us
3460                  * if there was no cache hit.
3461                  */
3462                 if (metricptr != NULL)
3463                         bzero(metricptr, sizeof(struct hc_metrics_lite));
3464                 return;
3465         }
3466
3467         /* What have we got? */
3468         switch (offer) {
3469                 case 0:
3470                         /*
3471                          * Offer == 0 means that there was no MSS on the SYN
3472                          * segment, in this case we use tcp_mssdflt as
3473                          * already assigned to t_maxopd above.
3474                          */
3475                         offer = tp->t_maxopd;
3476                         break;
3477
3478                 case -1:
3479                         /*
3480                          * Offer == -1 means that we didn't receive SYN yet.
3481                          */
3482                         /* FALLTHROUGH */
3483
3484                 default:
3485                         /*
3486                          * Prevent DoS attack with too small MSS. Round up
3487                          * to at least minmss.
3488                          */
3489                         offer = max(offer, V_tcp_minmss);
3490         }
3491
3492         /*
3493          * rmx information is now retrieved from tcp_hostcache.
3494          */
3495         tcp_hc_get(&inp->inp_inc, &metrics);
3496         if (metricptr != NULL)
3497                 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3498
3499         /*
3500          * If there's a discovered mtu int tcp hostcache, use it
3501          * else, use the link mtu.
3502          */
3503         if (metrics.rmx_mtu)
3504                 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3505         else {
3506 #ifdef INET6
3507                 if (isipv6) {
3508                         mss = maxmtu - min_protoh;
3509                         if (!V_path_mtu_discovery &&
3510                             !in6_localaddr(&inp->in6p_faddr))
3511                                 mss = min(mss, V_tcp_v6mssdflt);
3512                 }
3513 #endif
3514 #if defined(INET) && defined(INET6)
3515                 else
3516 #endif
3517 #ifdef INET
3518                 {
3519                         mss = maxmtu - min_protoh;
3520                         if (!V_path_mtu_discovery &&
3521                             !in_localaddr(inp->inp_faddr))
3522                                 mss = min(mss, V_tcp_mssdflt);
3523                 }
3524 #endif
3525                 /*
3526                  * XXX - The above conditional (mss = maxmtu - min_protoh)
3527                  * probably violates the TCP spec.
3528                  * The problem is that, since we don't know the
3529                  * other end's MSS, we are supposed to use a conservative
3530                  * default.  But, if we do that, then MTU discovery will
3531                  * never actually take place, because the conservative
3532                  * default is much less than the MTUs typically seen
3533                  * on the Internet today.  For the moment, we'll sweep
3534                  * this under the carpet.
3535                  *
3536                  * The conservative default might not actually be a problem
3537                  * if the only case this occurs is when sending an initial
3538                  * SYN with options and data to a host we've never talked
3539                  * to before.  Then, they will reply with an MSS value which
3540                  * will get recorded and the new parameters should get
3541                  * recomputed.  For Further Study.
3542                  */
3543         }
3544         mss = min(mss, offer);
3545
3546         /*
3547          * Sanity check: make sure that maxopd will be large
3548          * enough to allow some data on segments even if the
3549          * all the option space is used (40bytes).  Otherwise
3550          * funny things may happen in tcp_output.
3551          */
3552         mss = max(mss, 64);
3553
3554         /*
3555          * maxopd stores the maximum length of data AND options
3556          * in a segment; maxseg is the amount of data in a normal
3557          * segment.  We need to store this value (maxopd) apart
3558          * from maxseg, because now every segment carries options
3559          * and thus we normally have somewhat less data in segments.
3560          */
3561         tp->t_maxopd = mss;
3562
3563         /*
3564          * origoffer==-1 indicates that no segments were received yet.
3565          * In this case we just guess.
3566          */
3567         if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3568             (origoffer == -1 ||
3569              (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3570                 mss -= TCPOLEN_TSTAMP_APPA;
3571
3572         tp->t_maxseg = mss;
3573 }
3574
3575 void
3576 tcp_mss(struct tcpcb *tp, int offer)
3577 {
3578         int mss;
3579         u_long bufsize;
3580         struct inpcb *inp;
3581         struct socket *so;
3582         struct hc_metrics_lite metrics;
3583         struct tcp_ifcap cap;
3584
3585         KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3586
3587         bzero(&cap, sizeof(cap));
3588         tcp_mss_update(tp, offer, -1, &metrics, &cap);
3589
3590         mss = tp->t_maxseg;
3591         inp = tp->t_inpcb;
3592
3593         /*
3594          * If there's a pipesize, change the socket buffer to that size,
3595          * don't change if sb_hiwat is different than default (then it
3596          * has been changed on purpose with setsockopt).
3597          * Make the socket buffers an integral number of mss units;
3598          * if the mss is larger than the socket buffer, decrease the mss.
3599          */
3600         so = inp->inp_socket;
3601         SOCKBUF_LOCK(&so->so_snd);
3602         if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3603                 bufsize = metrics.rmx_sendpipe;
3604         else
3605                 bufsize = so->so_snd.sb_hiwat;
3606         if (bufsize < mss)
3607                 mss = bufsize;
3608         else {
3609                 bufsize = roundup(bufsize, mss);
3610                 if (bufsize > sb_max)
3611                         bufsize = sb_max;
3612                 if (bufsize > so->so_snd.sb_hiwat)
3613                         (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3614         }
3615         SOCKBUF_UNLOCK(&so->so_snd);
3616         tp->t_maxseg = mss;
3617
3618         SOCKBUF_LOCK(&so->so_rcv);
3619         if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3620                 bufsize = metrics.rmx_recvpipe;
3621         else
3622                 bufsize = so->so_rcv.sb_hiwat;
3623         if (bufsize > mss) {
3624                 bufsize = roundup(bufsize, mss);
3625                 if (bufsize > sb_max)
3626                         bufsize = sb_max;
3627                 if (bufsize > so->so_rcv.sb_hiwat)
3628                         (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3629         }
3630         SOCKBUF_UNLOCK(&so->so_rcv);
3631
3632         /* Check the interface for TSO capabilities. */
3633         if (cap.ifcap & CSUM_TSO) {
3634                 tp->t_flags |= TF_TSO;
3635                 tp->t_tsomax = cap.tsomax;
3636         }
3637 }
3638
3639 /*
3640  * Determine the MSS option to send on an outgoing SYN.
3641  */
3642 int
3643 tcp_mssopt(struct in_conninfo *inc)
3644 {
3645         int mss = 0;
3646         u_long maxmtu = 0;
3647         u_long thcmtu = 0;
3648         size_t min_protoh;
3649
3650         KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3651
3652 #ifdef INET6
3653         if (inc->inc_flags & INC_ISIPV6) {
3654                 mss = V_tcp_v6mssdflt;
3655                 maxmtu = tcp_maxmtu6(inc, NULL);
3656                 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3657         }
3658 #endif
3659 #if defined(INET) && defined(INET6)
3660         else
3661 #endif
3662 #ifdef INET
3663         {
3664                 mss = V_tcp_mssdflt;
3665                 maxmtu = tcp_maxmtu(inc, NULL);
3666                 min_protoh = sizeof(struct tcpiphdr);
3667         }
3668 #endif
3669 #if defined(INET6) || defined(INET)
3670         thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3671 #endif
3672
3673         if (maxmtu && thcmtu)
3674                 mss = min(maxmtu, thcmtu) - min_protoh;
3675         else if (maxmtu || thcmtu)
3676                 mss = max(maxmtu, thcmtu) - min_protoh;
3677
3678         return (mss);
3679 }
3680
3681
3682 /*
3683  * On a partial ack arrives, force the retransmission of the
3684  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3685  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3686  * be started again.
3687  */
3688 static void
3689 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3690 {
3691         tcp_seq onxt = tp->snd_nxt;
3692         u_long  ocwnd = tp->snd_cwnd;
3693
3694         INP_WLOCK_ASSERT(tp->t_inpcb);
3695
3696         tcp_timer_activate(tp, TT_REXMT, 0);
3697         tp->t_rtttime = 0;
3698         tp->snd_nxt = th->th_ack;
3699         /*
3700          * Set snd_cwnd to one segment beyond acknowledged offset.
3701          * (tp->snd_una has not yet been updated when this function is called.)
3702          */
3703         tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3704         tp->t_flags |= TF_ACKNOW;
3705         (void) tcp_output(tp);
3706         tp->snd_cwnd = ocwnd;
3707         if (SEQ_GT(onxt, tp->snd_nxt))
3708                 tp->snd_nxt = onxt;
3709         /*
3710          * Partial window deflation.  Relies on fact that tp->snd_una
3711          * not updated yet.
3712          */
3713         if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3714                 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3715         else
3716                 tp->snd_cwnd = 0;
3717         tp->snd_cwnd += tp->t_maxseg;
3718 }