bsd/netinet/tcp_var.h

   1 /*
   2  * Copyright (c) 2000-2011 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * Copyright (c) 1982, 1986, 1993, 1994, 1995
  30  *      The Regents of the University of California.  All rights reserved.
  31  *
  32  * Redistribution and use in source and binary forms, with or without
  33  * modification, are permitted provided that the following conditions
  34  * are met:
  35  * 1. Redistributions of source code must retain the above copyright
  36  *    notice, this list of conditions and the following disclaimer.
  37  * 2. Redistributions in binary form must reproduce the above copyright
  38  *    notice, this list of conditions and the following disclaimer in the
  39  *    documentation and/or other materials provided with the distribution.
  40  * 3. All advertising materials mentioning features or use of this software
  41  *    must display the following acknowledgement:
  42  *      This product includes software developed by the University of
  43  *      California, Berkeley and its contributors.
  44  * 4. Neither the name of the University nor the names of its contributors
  45  *    may be used to endorse or promote products derived from this software
  46  *    without specific prior written permission.
  47  *
  48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  58  * SUCH DAMAGE.
  59  *
  60  *      @(#)tcp_var.h   8.4 (Berkeley) 5/24/95
  61  * $FreeBSD: src/sys/netinet/tcp_var.h,v 1.56.2.8 2001/08/22 00:59:13 silby Exp $
  62  */
  63
  64 #ifndef _NETINET_TCP_VAR_H_
  65 #define _NETINET_TCP_VAR_H_
  66 #include <sys/appleapiopts.h>
  67 #include <sys/queue.h>
  68 #include <netinet/in_pcb.h>
  69 #include <netinet/tcp_timer.h>
  70
  71 #if defined(__LP64__)
  72 #define _TCPCB_PTR(x)                   u_int32_t
  73 #define _TCPCB_LIST_HEAD(name, type)    \
  74 struct name {                           \
  75         u_int32_t       lh_first;       \
  76 };
  77 #else
  78 #define _TCPCB_PTR(x)                   x
  79 #define _TCPCB_LIST_HEAD(name, type)    LIST_HEAD(name, type)
  80 #endif
  81
  82 #define TCP_RETRANSHZ   1000    /* granularity of TCP timestamps, 1ms */
  83 #define TCP_TIMERHZ     100             /* frequency of TCP fast timer, 100 ms */
  84
  85 /* Minimum time quantum within which the timers are coalesced */
  86 #define TCP_FASTTIMER_QUANTUM   TCP_TIMERHZ     /* fast mode, once every 100ms */
  87 #define TCP_SLOWTIMER_QUANTUM   TCP_RETRANSHZ / PR_SLOWHZ       /* slow mode, once every 500ms */
  88
  89 #define TCP_RETRANSHZ_TO_USEC 1000
  90
  91 #ifdef KERNEL_PRIVATE
  92 #define N_TIME_WAIT_SLOTS   128         /* must be power of 2 */
  93
  94 /* Base RTT is stored for N_MIN_RTT_HISTORY slots. This is used to
  95  * estimate expected minimum RTT for delay based congestion control
  96  * algorithms.
  97  */
  98 #define N_RTT_BASE      5
  99
 100 /* Always allow at least 4 packets worth of recv window when adjusting
 101  * recv window using inter-packet arrival jitter.
 102  */
 103 #define MIN_IAJ_WIN 4
 104
 105 /* A variation in delay of this many milliseconds is tolerable. This limit has to
 106  * be low but greater than zero. We also use standard deviation on jitter to adjust
 107  * this limit for different link and connection types.
 108  */
 109 #define ALLOWED_IAJ 5
 110
 111 /* Ignore the first few packets on a connection until the ACK clock gets going
 112  */
 113 #define IAJ_IGNORE_PKTCNT 40
 114
 115 /* Let the accumulated IAJ value increase by this threshold at most. This limit
 116  * will control how many ALLOWED_IAJ measurements a receiver will have to see
 117  * before opening the receive window
 118  */
 119 #define ACC_IAJ_HIGH_THRESH 100
 120
 121 /* When accumulated IAJ reaches this value, the receiver starts to react by
 122  * closing the window
 123  */
 124 #define ACC_IAJ_REACT_LIMIT 200
 125
 126 /* If the number of small packets (smaller than IAJ packet size) seen on a
 127  * connection is more than this threshold, reset the size and learn it again.
 128  * This is needed because the sender might send smaller segments after PMTU
 129  * discovery and the receiver has to learn the new size.
 130  */
 131 #define RESET_IAJ_SIZE_THRESH 20
 132
 133 /*
 134  * Kernel variables for tcp.
 135  */
 136
 137 /* TCP segment queue entry */
 138 struct tseg_qent {
 139         LIST_ENTRY(tseg_qent) tqe_q;
 140         int     tqe_len;                /* TCP segment data length */
 141         struct  tcphdr *tqe_th;         /* a pointer to tcp header */
 142         struct  mbuf    *tqe_m;         /* mbuf contains packet */
 143 };
 144 LIST_HEAD(tsegqe_head, tseg_qent);
 145 extern int      tcp_reass_maxseg;
 146 extern int      tcp_reass_qsize;
 147 #ifdef MALLOC_DECLARE
 148 MALLOC_DECLARE(M_TSEGQ);
 149 #endif
 150
 151 struct sackblk {
 152         tcp_seq start;          /* start seq no. of sack block */
 153         tcp_seq end;            /* end seq no. */
 154 };
 155
 156 struct sackhole {
 157         tcp_seq start;          /* start seq no. of hole */
 158         tcp_seq end;            /* end seq no. */
 159         tcp_seq rxmit;          /* next seq. no in hole to be retransmitted */
 160         TAILQ_ENTRY(sackhole) scblink;  /* scoreboard linkage */
 161 };
 162
 163 struct sackhint {
 164         struct sackhole *nexthole;
 165         int     sack_bytes_rexmit;
 166 };
 167
 168 struct tcptemp {
 169         u_char  tt_ipgen[40]; /* the size must be of max ip header, now IPv6 */
 170         struct  tcphdr tt_t;
 171 };
 172
 173 #define tcp6cb          tcpcb  /* for KAME src sync over BSD*'s */
 174
 175 /*
 176  * Tcp control block, one per tcp; fields:
 177  * Organized for 16 byte cacheline efficiency.
 178  */
 179 struct tcpcb {
 180         struct  tsegqe_head t_segq;
 181         int     t_dupacks;              /* consecutive dup acks recd */
 182         uint32_t t_timer[TCPT_NTIMERS]; /* tcp timers */
 183         struct tcptimerentry tentry;    /* entry in timer list */
 184
 185         struct  inpcb *t_inpcb;         /* back pointer to internet pcb */
 186         int     t_state;                /* state of this connection */
 187         u_int   t_flags;
 188 #define TF_ACKNOW       0x00001         /* ack peer immediately */
 189 #define TF_DELACK       0x00002         /* ack, but try to delay it */
 190 #define TF_NODELAY      0x00004         /* don't delay packets to coalesce */
 191 #define TF_NOOPT        0x00008         /* don't use tcp options */
 192 #define TF_SENTFIN      0x00010         /* have sent FIN */
 193 #define TF_REQ_SCALE    0x00020         /* have/will request window scaling */
 194 #define TF_RCVD_SCALE   0x00040         /* other side has requested scaling */
 195 #define TF_REQ_TSTMP    0x00080         /* have/will request timestamps */
 196 #define TF_RCVD_TSTMP   0x00100         /* a timestamp was received in SYN */
 197 #define TF_SACK_PERMIT  0x00200         /* other side said I could SACK */
 198 #define TF_NEEDSYN      0x00400         /* send SYN (implicit state) */
 199 #define TF_NEEDFIN      0x00800         /* send FIN (implicit state) */
 200 #define TF_NOPUSH       0x01000         /* don't push */
 201 #define TF_REQ_CC       0x02000         /* have/will request CC */
 202 #define TF_RCVD_CC      0x04000         /* a CC was received in SYN */
 203 #define TF_SENDCCNEW    0x08000         /* send CCnew instead of CC in SYN */
 204 #define TF_MORETOCOME   0x10000         /* More data to be appended to sock */
 205 #define TF_LOCAL        0x20000         /* connection to a host on local link */
 206 #define TF_RXWIN0SENT   0x40000         /* sent a receiver win 0 in response */
 207 #define TF_SLOWLINK     0x80000         /* route is a on a modem speed link */
 208 #define TF_LASTIDLE     0x100000        /* connection was previously idle */
 209 #define TF_FASTRECOVERY 0x200000        /* in NewReno Fast Recovery */
 210 #define TF_WASFRECOVERY 0x400000        /* was in NewReno Fast Recovery */
 211 #define TF_SIGNATURE    0x800000        /* require MD5 digests (RFC2385) */
 212 #define TF_MAXSEGSNT    0x1000000       /* last segment sent was a full segment */
 213 #define TF_SENDINPROG   0x2000000       /* send is in progress */
 214 #define TF_PMTUD        0x4000000       /* Perform Path MTU Discovery for this connection */
 215 #define TF_CLOSING      0x8000000       /* pending tcp close */
 216 #define TF_TSO          0x10000000      /* TCP Segment Offloading is enable on this connection */
 217 #define TF_BLACKHOLE    0x20000000      /* Path MTU Discovery Black Hole detection */
 218 #define TF_TIMER_ONLIST 0x40000000      /* pcb is on tcp_timer_list */
 219 #define TF_STRETCHACK   0x80000000      /* receiver is going to delay acks */
 220
 221         int     t_force;                /* 1 if forcing out a byte */
 222
 223         tcp_seq snd_una;                /* send unacknowledged */
 224         tcp_seq snd_max;                /* highest sequence number sent;
 225                                          * used to recognize retransmits
 226                                          */
 227         tcp_seq snd_nxt;                /* send next */
 228         tcp_seq snd_up;                 /* send urgent pointer */
 229
 230         tcp_seq snd_wl1;                /* window update seg seq number */
 231         tcp_seq snd_wl2;                /* window update seg ack number */
 232         tcp_seq iss;                    /* initial send sequence number */
 233         tcp_seq irs;                    /* initial receive sequence number */
 234
 235         tcp_seq rcv_nxt;                /* receive next */
 236         tcp_seq rcv_adv;                /* advertised window */
 237         u_int32_t       rcv_wnd;                /* receive window */
 238         tcp_seq rcv_up;                 /* receive urgent pointer */
 239
 240         u_int32_t       snd_wnd;                /* send window */
 241         u_int32_t       snd_cwnd;               /* congestion-controlled window */
 242         u_int32_t       snd_bwnd;               /* bandwidth-controlled window */
 243         u_int32_t       snd_ssthresh;           /* snd_cwnd size threshold for
 244                                          * for slow start exponential to
 245                                          * linear switch
 246                                          */
 247         u_int32_t       snd_bandwidth;  /* calculated bandwidth or 0 */
 248         tcp_seq snd_recover;            /* for use in NewReno Fast Recovery */
 249
 250         u_int   t_maxopd;               /* mss plus options */
 251
 252         u_int32_t       t_rcvtime;      /* time at which a packet was received */
 253         u_int32_t       t_starttime;    /* time connection was established */
 254         int     t_rtttime;              /* tcp clock when rtt calculation was started */
 255         tcp_seq t_rtseq;                /* sequence number being timed */
 256
 257         int     t_bw_rtttime;           /* used for bandwidth calculation */
 258         tcp_seq t_bw_rtseq;             /* used for bandwidth calculation */
 259
 260         int     t_rxtcur;               /* current retransmit value (ticks) */
 261         u_int   t_maxseg;               /* maximum segment size */
 262         int     t_srtt;                 /* smoothed round-trip time */
 263         int     t_rttvar;               /* variance in round-trip time */
 264
 265         int     t_rxtshift;             /* log(2) of rexmt exp. backoff */
 266         u_int   t_rttmin;               /* minimum rtt allowed */
 267         u_int   t_rttbest;              /* best rtt we've seen */
 268         u_int   t_rttcur;               /* most recent value of rtt */
 269         u_int32_t       t_rttupdated;           /* number of times rtt sampled */
 270         u_int32_t       rxt_conndroptime;       /* retxmt conn gets dropped after this time, when set */
 271         u_int32_t       rxt_start;              /* time at a connection starts retransmitting */
 272         u_int32_t       max_sndwnd;             /* largest window peer has offered */
 273
 274         int     t_softerror;            /* possible error not yet reported */
 275 /* out-of-band data */
 276         char    t_oobflags;             /* have some */
 277         char    t_iobc;                 /* input character */
 278 #define TCPOOB_HAVEDATA 0x01
 279 #define TCPOOB_HADDATA  0x02
 280 /* RFC 1323 variables */
 281         u_char  snd_scale;              /* window scaling for send window */
 282         u_char  rcv_scale;              /* window scaling for recv window */
 283         u_char  request_r_scale;        /* pending window scaling */
 284         u_char  requested_s_scale;
 285         u_int16_t       tcp_cc_index;   /* index of congestion control algorithm */
 286         u_int32_t       ts_recent;              /* timestamp echo data */
 287
 288         u_int32_t       ts_recent_age;          /* when last updated */
 289         tcp_seq last_ack_sent;
 290 /* RFC 1644 variables */
 291         tcp_cc  cc_send;                /* send connection count */
 292         tcp_cc  cc_recv;                /* receive connection count */
 293 /* RFC 3465 variables */
 294         u_int32_t       t_bytes_acked;          /* ABC "bytes_acked" parameter */
 295 /* experimental */
 296         u_int32_t       snd_cwnd_prev;          /* cwnd prior to retransmit */
 297         u_int32_t       snd_ssthresh_prev;      /* ssthresh prior to retransmit */
 298         u_int32_t       t_badrxtwin;            /* window for retransmit recovery */
 299
 300         int     t_keepidle;             /* keepalive idle timer (override global if > 0) */
 301         int     t_lastchain;            /* amount of packets chained last time around */
 302         int     t_unacksegs;            /* received but unacked segments: used for delaying acks */
 303         u_int32_t       t_persist_timeout;      /* ZWP persistence limit as set by PERSIST_TIMEOUT */
 304         u_int32_t       t_persist_stop;         /* persistence limit deadline if triggered by ZWP */
 305
 306
 307 /* 3529618 MSS overload prevention */
 308         u_int32_t       rcv_reset;
 309         u_int32_t       rcv_pps;
 310         u_int32_t       rcv_byps;
 311         u_int32_t       rcv_maxbyps;
 312
 313 /* Receiver state for stretch-ack algorithm */
 314         u_int32_t       rcv_unackwin;   /* to measure win for stretching acks */
 315         u_int32_t       rcv_by_unackwin; /* bytes seen during the last ack-stretching win */
 316         u_int16_t       rcv_waitforss;  /* wait for packets during slow-start */
 317         u_int16_t               ecn_flags;
 318 #define TE_SETUPSENT            0x01    /* Indicate we have sent ECN-SETUP SYN or SYN-ACK */
 319 #define TE_SETUPRECEIVED        0x02    /* Indicate we have received ECN-SETUP SYN or SYN-ACK */
 320 #define TE_SENDIPECT            0x04    /* Indicate we haven't sent or received non-ECN-setup SYN or SYN-ACK */
 321 #define TE_SENDCWR              0x08    /* Indicate that the next non-retransmit should have the TCP CWR flag set */
 322 #define TE_SENDECE              0x10    /* Indicate that the next packet should have the TCP ECE flag set */
 323         tcp_seq snd_high;               /* for use in NewReno Fast Recovery */
 324         tcp_seq snd_high_prev;  /* snd_high prior to retransmit */
 325         tcp_seq snd_recover_prev;       /* snd_recover prior to retransmit */
 326         u_char  snd_limited;            /* segments limited transmitted */
 327 /* anti DoS counters */
 328         u_int32_t       rcv_second;             /* start of interval second */
 329
 330 /* SACK related state */
 331         int     sack_enable;            /* enable SACK for this connection */
 332         int     snd_numholes;           /* number of holes seen by sender */
 333         TAILQ_HEAD(sackhole_head, sackhole) snd_holes;
 334                                                 /* SACK scoreboard (sorted) */
 335         tcp_seq snd_fack;               /* last seq number(+1) sack'd by rcv'r*/
 336         int     rcv_numsacks;           /* # distinct sack blks present */
 337         struct sackblk sackblks[MAX_SACK_BLKS]; /* seq nos. of sack blocks */
 338         tcp_seq sack_newdata;           /* New data xmitted in this recovery
 339                                            episode starts at this seq number */
 340         struct sackhint sackhint;       /* SACK scoreboard hint */
 341
 342         u_int32_t       t_pktlist_sentlen; /* total bytes in transmit chain */
 343         struct mbuf     *t_pktlist_head; /* First packet in transmit chain */
 344         struct mbuf     *t_pktlist_tail; /* Last packet in transmit chain */
 345
 346         int             t_keepinit; /* connection timeout, i.e. idle time in SYN_SENT or SYN_RECV state */
 347         u_int32_t       tso_max_segment_size;   /* TCP Segment Offloading maximum segment unit for NIC */
 348         u_int           t_pmtud_saved_maxopd;   /* MSS saved before performing PMTU-D BlackHole detection */
 349
 350         struct
 351         {
 352                 u_int32_t       rxduplicatebytes;
 353                 u_int32_t       rxoutoforderbytes;
 354                 u_int32_t       txretransmitbytes;
 355                 u_int32_t       unused_pad_to_8;
 356         } t_stat;
 357
 358         /* Background congestion related state */
 359         uint32_t        rtt_hist[N_RTT_BASE];   /* history of minimum RTT */
 360         uint32_t        rtt_count;              /* Number of RTT samples in recent base history */
 361         uint32_t        bg_ssthresh;            /* Slow start threshold until delay increases */
 362         uint32_t        t_flagsext;             /* Another field to accommodate more flags */
 363 #define TF_RXTFINDROP   0x1                     /* Drop conn after retransmitting FIN 3 times */
 364 #define TF_RCVUNACK_WAITSS      0x2             /* set when the receiver should not stretch acks */
 365
 366 #if TRAFFIC_MGT
 367         /* Inter-arrival jitter related state */
 368         uint32_t        iaj_rcv_ts;             /* tcp clock when the first packet was received */
 369         uint16_t        iaj_size;               /* Size of packet for iaj measurement */
 370         uint16_t        iaj_small_pkt;          /* Count of packets smaller than iaj_size */
 371         uint16_t        iaj_pktcnt;             /* packet count, to avoid throttling initially */
 372         uint16_t        acc_iaj;                /* Accumulated iaj */
 373         tcp_seq         iaj_rwintop;            /* recent max advertised window */
 374         uint32_t        avg_iaj;                /* Mean */
 375         uint32_t        std_dev_iaj;            /* Standard deviation */
 376 #endif /* TRAFFIC_MGT */
 377 };
 378
 379 #define IN_FASTRECOVERY(tp)     (tp->t_flags & TF_FASTRECOVERY)
 380 #define ENTER_FASTRECOVERY(tp)  tp->t_flags |= TF_FASTRECOVERY
 381 #define EXIT_FASTRECOVERY(tp)   tp->t_flags &= ~TF_FASTRECOVERY
 382
 383 #if CONFIG_DTRACE
 384 enum tcp_cc_event {
 385         TCP_CC_CWND_INIT,
 386         TCP_CC_INSEQ_ACK_RCVD,
 387         TCP_CC_ACK_RCVD,
 388         TCP_CC_ENTER_FASTRECOVERY,
 389         TCP_CC_IN_FASTRECOVERY,
 390         TCP_CC_EXIT_FASTRECOVERY,
 391         TCP_CC_PARTIAL_ACK,
 392         TCP_CC_IDLE_TIMEOUT,
 393         TCP_CC_REXMT_TIMEOUT,
 394         TCP_CC_ECN_RCVD,
 395         TCP_CC_BAD_REXMT_RECOVERY,
 396         TCP_CC_OUTPUT_ERROR,
 397         TCP_CC_CHANGE_ALGO
 398 };
 399 #endif /* CONFIG_DTRACE */
 400
 401 /*
 402  * Structure to hold TCP options that are only used during segment
 403  * processing (in tcp_input), but not held in the tcpcb.
 404  * It's basically used to reduce the number of parameters
 405  * to tcp_dooptions.
 406  */
 407 struct tcpopt {
 408         u_int32_t       to_flags;               /* which options are present */
 409 #define TOF_TS          0x0001          /* timestamp */
 410 #define TOF_MSS         0x0010
 411 #define TOF_SCALE       0x0020
 412 #define TOF_SIGNATURE   0x0040  /* signature option present */
 413 #define TOF_SIGLEN      0x0080  /* signature length valid (RFC2385) */
 414 #define TOF_SACK        0x0100          /* Peer sent SACK option */
 415         u_int32_t               to_tsval;
 416         u_int32_t               to_tsecr;
 417         u_int16_t       to_mss;
 418         u_int8_t        to_requested_s_scale;
 419         u_int8_t        to_nsacks;      /* number of SACK blocks */
 420         u_char          *to_sacks;      /* pointer to the first SACK blocks */
 421 };
 422
 423 /*
 424  * The TAO cache entry which is stored in the protocol family specific
 425  * portion of the route metrics.
 426  */
 427 struct rmxp_tao {
 428         tcp_cc  tao_cc;                 /* latest CC in valid SYN */
 429         tcp_cc  tao_ccsent;             /* latest CC sent to peer */
 430         u_short tao_mssopt;             /* peer's cached MSS */
 431 #ifdef notyet
 432         u_short tao_flags;              /* cache status flags */
 433 #define TAOF_DONT       0x0001          /* peer doesn't understand rfc1644 */
 434 #define TAOF_OK         0x0002          /* peer does understand rfc1644 */
 435 #define TAOF_UNDEF      0               /* we don't know yet */
 436 #endif /* notyet */
 437 };
 438 #define rmx_taop(r)     ((struct rmxp_tao *)(r).rmx_filler)
 439
 440 #define intotcpcb(ip)   ((struct tcpcb *)(ip)->inp_ppcb)
 441 #define sototcpcb(so)   (intotcpcb(sotoinpcb(so)))
 442
 443 /*
 444  * The rtt measured is in milliseconds as the timestamp granularity is
 445  * a millisecond. The smoothed round-trip time and estimated variance
 446  * are stored as fixed point numbers scaled by the values below.
 447  * For convenience, these scales are also used in smoothing the average
 448  * (smoothed = (1/scale)sample + ((scale-1)/scale)smoothed).
 449  * With these scales, srtt has 5 bits to the right of the binary point,
 450  * and thus an "ALPHA" of 0.875.  rttvar has 4 bits to the right of the
 451  * binary point, and is smoothed with an ALPHA of 0.75.
 452  */
 453 #define TCP_RTT_SCALE           32      /* multiplier for srtt; 3 bits frac. */
 454 #define TCP_RTT_SHIFT           5       /* shift for srtt; 5 bits frac. */
 455 #define TCP_RTTVAR_SCALE        16      /* multiplier for rttvar; 4 bits */
 456 #define TCP_RTTVAR_SHIFT        4       /* shift for rttvar; 4 bits */
 457 #define TCP_DELTA_SHIFT         2       /* see tcp_input.c */
 458
 459 /*
 460  * The initial retransmission should happen at rtt + 4 * rttvar.
 461  * Because of the way we do the smoothing, srtt and rttvar
 462  * will each average +1/2 tick of bias.  When we compute
 463  * the retransmit timer, we want 1/2 tick of rounding and
 464  * 1 extra tick because of +-1/2 tick uncertainty in the
 465  * firing of the timer.  The bias will give us exactly the
 466  * 1.5 tick we need.  But, because the bias is
 467  * statistical, we have to test that we don't drop below
 468  * the minimum feasible timer (which is 2 ticks).
 469  * This version of the macro adapted from a paper by Lawrence
 470  * Brakmo and Larry Peterson which outlines a problem caused
 471  * by insufficient precision in the original implementation,
 472  * which results in inappropriately large RTO values for very
 473  * fast networks.
 474  */
 475 #define TCP_REXMTVAL(tp) \
 476         max((tp)->t_rttmin, (((tp)->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT))  \
 477           + (tp)->t_rttvar) >> TCP_DELTA_SHIFT)
 478
 479 /*
 480  * Jaguar compatible TCP control block, for xtcpcb
 481  * Does not have the old fields
 482  */
 483 struct otcpcb {
 484 #else
 485 struct tseg_qent;
 486 _TCPCB_LIST_HEAD(tsegqe_head, tseg_qent);
 487
 488 struct tcpcb {
 489 #endif /* KERNEL_PRIVATE */
 490 #if defined(KERNEL_PRIVATE)
 491         u_int32_t t_segq;
 492 #else
 493         struct  tsegqe_head t_segq;
 494 #endif /* KERNEL_PRIVATE */
 495         int     t_dupacks;              /* consecutive dup acks recd */
 496         u_int32_t unused;               /* unused now: was t_template */
 497
 498         int     t_timer[TCPT_NTIMERS_EXT];      /* tcp timers */
 499
 500         _TCPCB_PTR(struct inpcb *) t_inpcb;     /* back pointer to internet pcb */
 501         int     t_state;                /* state of this connection */
 502         u_int   t_flags;
 503 #define TF_ACKNOW       0x00001         /* ack peer immediately */
 504 #define TF_DELACK       0x00002         /* ack, but try to delay it */
 505 #define TF_NODELAY      0x00004         /* don't delay packets to coalesce */
 506 #define TF_NOOPT        0x00008         /* don't use tcp options */
 507 #define TF_SENTFIN      0x00010         /* have sent FIN */
 508 #define TF_REQ_SCALE    0x00020         /* have/will request window scaling */
 509 #define TF_RCVD_SCALE   0x00040         /* other side has requested scaling */
 510 #define TF_REQ_TSTMP    0x00080         /* have/will request timestamps */
 511 #define TF_RCVD_TSTMP   0x00100         /* a timestamp was received in SYN */
 512 #define TF_SACK_PERMIT  0x00200         /* other side said I could SACK */
 513 #define TF_NEEDSYN      0x00400         /* send SYN (implicit state) */
 514 #define TF_NEEDFIN      0x00800         /* send FIN (implicit state) */
 515 #define TF_NOPUSH       0x01000         /* don't push */
 516 #define TF_REQ_CC       0x02000         /* have/will request CC */
 517 #define TF_RCVD_CC      0x04000         /* a CC was received in SYN */
 518 #define TF_SENDCCNEW    0x08000         /* send CCnew instead of CC in SYN */
 519 #define TF_MORETOCOME   0x10000         /* More data to be appended to sock */
 520 #define TF_LQ_OVERFLOW  0x20000         /* listen queue overflow */
 521 #define TF_RXWIN0SENT   0x40000         /* sent a receiver win 0 in response */
 522 #define TF_SLOWLINK     0x80000         /* route is a on a modem speed link */
 523
 524         int     t_force;                /* 1 if forcing out a byte */
 525
 526         tcp_seq snd_una;                /* send unacknowledged */
 527         tcp_seq snd_max;                /* highest sequence number sent;
 528                                          * used to recognize retransmits
 529                                          */
 530         tcp_seq snd_nxt;                /* send next */
 531         tcp_seq snd_up;                 /* send urgent pointer */
 532
 533         tcp_seq snd_wl1;                /* window update seg seq number */
 534         tcp_seq snd_wl2;                /* window update seg ack number */
 535         tcp_seq iss;                    /* initial send sequence number */
 536         tcp_seq irs;                    /* initial receive sequence number */
 537
 538         tcp_seq rcv_nxt;                /* receive next */
 539         tcp_seq rcv_adv;                /* advertised window */
 540         u_int32_t rcv_wnd;              /* receive window */
 541         tcp_seq rcv_up;                 /* receive urgent pointer */
 542
 543         u_int32_t snd_wnd;              /* send window */
 544         u_int32_t snd_cwnd;             /* congestion-controlled window */
 545         u_int32_t snd_ssthresh;         /* snd_cwnd size threshold for
 546                                          * for slow start exponential to
 547                                          * linear switch
 548                                          */
 549         u_int   t_maxopd;               /* mss plus options */
 550
 551         u_int32_t t_rcvtime;            /* time at which a packet was received */
 552         u_int32_t t_starttime;          /* time connection was established */
 553         int     t_rtttime;              /* round trip time */
 554         tcp_seq t_rtseq;                /* sequence number being timed */
 555
 556         int     t_rxtcur;               /* current retransmit value (ticks) */
 557         u_int   t_maxseg;               /* maximum segment size */
 558         int     t_srtt;                 /* smoothed round-trip time */
 559         int     t_rttvar;               /* variance in round-trip time */
 560
 561         int     t_rxtshift;             /* log(2) of rexmt exp. backoff */
 562         u_int   t_rttmin;               /* minimum rtt allowed */
 563         u_int32_t t_rttupdated;         /* number of times rtt sampled */
 564         u_int32_t max_sndwnd;           /* largest window peer has offered */
 565
 566         int     t_softerror;            /* possible error not yet reported */
 567 /* out-of-band data */
 568         char    t_oobflags;             /* have some */
 569         char    t_iobc;                 /* input character */
 570 #define TCPOOB_HAVEDATA 0x01
 571 #define TCPOOB_HADDATA  0x02
 572 /* RFC 1323 variables */
 573         u_char  snd_scale;              /* window scaling for send window */
 574         u_char  rcv_scale;              /* window scaling for recv window */
 575         u_char  request_r_scale;        /* pending window scaling */
 576         u_char  requested_s_scale;
 577         u_int32_t ts_recent;            /* timestamp echo data */
 578
 579         u_int32_t ts_recent_age;        /* when last updated */
 580         tcp_seq last_ack_sent;
 581 /* RFC 1644 variables */
 582         tcp_cc  cc_send;                /* send connection count */
 583         tcp_cc  cc_recv;                /* receive connection count */
 584         tcp_seq snd_recover;            /* for use in fast recovery */
 585 /* experimental */
 586         u_int32_t snd_cwnd_prev;        /* cwnd prior to retransmit */
 587         u_int32_t snd_ssthresh_prev;    /* ssthresh prior to retransmit */
 588         u_int32_t t_badrxtwin;          /* window for retransmit recovery */
 589 };
 590
 591
 592 /*
 593  * TCP statistics.
 594  * Many of these should be kept per connection,
 595  * but that's inconvenient at the moment.
 596  */
 597 struct  tcpstat {
 598         u_int32_t       tcps_connattempt;       /* connections initiated */
 599         u_int32_t       tcps_accepts;           /* connections accepted */
 600         u_int32_t       tcps_connects;          /* connections established */
 601         u_int32_t       tcps_drops;             /* connections dropped */
 602         u_int32_t       tcps_conndrops;         /* embryonic connections dropped */
 603         u_int32_t       tcps_closed;            /* conn. closed (includes drops) */
 604         u_int32_t       tcps_segstimed;         /* segs where we tried to get rtt */
 605         u_int32_t       tcps_rttupdated;        /* times we succeeded */
 606         u_int32_t       tcps_delack;            /* delayed acks sent */
 607         u_int32_t       tcps_timeoutdrop;       /* conn. dropped in rxmt timeout */
 608         u_int32_t       tcps_rexmttimeo;        /* retransmit timeouts */
 609         u_int32_t       tcps_persisttimeo;      /* persist timeouts */
 610         u_int32_t       tcps_keeptimeo;         /* keepalive timeouts */
 611         u_int32_t       tcps_keepprobe;         /* keepalive probes sent */
 612         u_int32_t       tcps_keepdrops;         /* connections dropped in keepalive */
 613
 614         u_int32_t       tcps_sndtotal;          /* total packets sent */
 615         u_int32_t       tcps_sndpack;           /* data packets sent */
 616         u_int32_t       tcps_sndbyte;           /* data bytes sent */
 617         u_int32_t       tcps_sndrexmitpack;     /* data packets retransmitted */
 618         u_int32_t       tcps_sndrexmitbyte;     /* data bytes retransmitted */
 619         u_int32_t       tcps_sndacks;           /* ack-only packets sent */
 620         u_int32_t       tcps_sndprobe;          /* window probes sent */
 621         u_int32_t       tcps_sndurg;            /* packets sent with URG only */
 622         u_int32_t       tcps_sndwinup;          /* window update-only packets sent */
 623         u_int32_t       tcps_sndctrl;           /* control (SYN|FIN|RST) packets sent */
 624
 625         u_int32_t       tcps_rcvtotal;          /* total packets received */
 626         u_int32_t       tcps_rcvpack;           /* packets received in sequence */
 627         u_int32_t       tcps_rcvbyte;           /* bytes received in sequence */
 628         u_int32_t       tcps_rcvbadsum;         /* packets received with ccksum errs */
 629         u_int32_t       tcps_rcvbadoff;         /* packets received with bad offset */
 630         u_int32_t       tcps_rcvmemdrop;        /* packets dropped for lack of memory */
 631         u_int32_t       tcps_rcvshort;          /* packets received too short */
 632         u_int32_t       tcps_rcvduppack;        /* duplicate-only packets received */
 633         u_int32_t       tcps_rcvdupbyte;        /* duplicate-only bytes received */
 634         u_int32_t       tcps_rcvpartduppack;    /* packets with some duplicate data */
 635         u_int32_t       tcps_rcvpartdupbyte;    /* dup. bytes in part-dup. packets */
 636         u_int32_t       tcps_rcvoopack;         /* out-of-order packets received */
 637         u_int32_t       tcps_rcvoobyte;         /* out-of-order bytes received */
 638         u_int32_t       tcps_rcvpackafterwin;   /* packets with data after window */
 639         u_int32_t       tcps_rcvbyteafterwin;   /* bytes rcvd after window */
 640         u_int32_t       tcps_rcvafterclose;     /* packets rcvd after "close" */
 641         u_int32_t       tcps_rcvwinprobe;       /* rcvd window probe packets */
 642         u_int32_t       tcps_rcvdupack;         /* rcvd duplicate acks */
 643         u_int32_t       tcps_rcvacktoomuch;     /* rcvd acks for unsent data */
 644         u_int32_t       tcps_rcvackpack;        /* rcvd ack packets */
 645         u_int32_t       tcps_rcvackbyte;        /* bytes acked by rcvd acks */
 646         u_int32_t       tcps_rcvwinupd;         /* rcvd window update packets */
 647         u_int32_t       tcps_pawsdrop;          /* segments dropped due to PAWS */
 648         u_int32_t       tcps_predack;           /* times hdr predict ok for acks */
 649         u_int32_t       tcps_preddat;           /* times hdr predict ok for data pkts */
 650         u_int32_t       tcps_pcbcachemiss;
 651         u_int32_t       tcps_cachedrtt;         /* times cached RTT in route updated */
 652         u_int32_t       tcps_cachedrttvar;      /* times cached rttvar updated */
 653         u_int32_t       tcps_cachedssthresh;    /* times cached ssthresh updated */
 654         u_int32_t       tcps_usedrtt;           /* times RTT initialized from route */
 655         u_int32_t       tcps_usedrttvar;        /* times RTTVAR initialized from rt */
 656         u_int32_t       tcps_usedssthresh;      /* times ssthresh initialized from rt*/
 657         u_int32_t       tcps_persistdrop;       /* timeout in persist state */
 658         u_int32_t       tcps_badsyn;            /* bogus SYN, e.g. premature ACK */
 659         u_int32_t       tcps_mturesent;         /* resends due to MTU discovery */
 660         u_int32_t       tcps_listendrop;        /* listen queue overflows */
 661
 662         /* new stats from FreeBSD 5.4 sync up */
 663         u_int32_t       tcps_minmssdrops;       /* average minmss too low drops */
 664         u_int32_t       tcps_sndrexmitbad;      /* unnecessary packet retransmissions */
 665         u_int32_t       tcps_badrst;            /* ignored RSTs in the window */
 666
 667         u_int32_t       tcps_sc_added;          /* entry added to syncache */
 668         u_int32_t       tcps_sc_retransmitted;  /* syncache entry was retransmitted */
 669         u_int32_t       tcps_sc_dupsyn;         /* duplicate SYN packet */
 670         u_int32_t       tcps_sc_dropped;        /* could not reply to packet */
 671         u_int32_t       tcps_sc_completed;      /* successful extraction of entry */
 672         u_int32_t       tcps_sc_bucketoverflow; /* syncache per-bucket limit hit */
 673         u_int32_t       tcps_sc_cacheoverflow;  /* syncache cache limit hit */
 674         u_int32_t       tcps_sc_reset;          /* RST removed entry from syncache */
 675         u_int32_t       tcps_sc_stale;          /* timed out or listen socket gone */
 676         u_int32_t       tcps_sc_aborted;        /* syncache entry aborted */
 677         u_int32_t       tcps_sc_badack;         /* removed due to bad ACK */
 678         u_int32_t       tcps_sc_unreach;        /* ICMP unreachable received */
 679         u_int32_t       tcps_sc_zonefail;       /* zalloc() failed */
 680         u_int32_t       tcps_sc_sendcookie;     /* SYN cookie sent */
 681         u_int32_t       tcps_sc_recvcookie;     /* SYN cookie received */
 682
 683         u_int32_t       tcps_hc_added;          /* entry added to hostcache */
 684         u_int32_t       tcps_hc_bucketoverflow; /* hostcache per bucket limit hit */
 685
 686         /* SACK related stats */
 687         u_int32_t       tcps_sack_recovery_episode; /* SACK recovery episodes */
 688         u_int32_t  tcps_sack_rexmits;       /* SACK rexmit segments   */
 689         u_int32_t  tcps_sack_rexmit_bytes;          /* SACK rexmit bytes      */
 690         u_int32_t  tcps_sack_rcv_blocks;            /* SACK blocks (options) received */
 691         u_int32_t  tcps_sack_send_blocks;           /* SACK blocks (options) sent     */
 692         u_int32_t  tcps_sack_sboverflow;            /* SACK sendblock overflow   */
 693
 694         u_int32_t       tcps_bg_rcvtotal;       /* total background packets received */
 695         u_int32_t       tcps_rxtfindrop;        /* drop conn after retransmitting FIN */
 696 };
 697
 698 #pragma pack(4)
 699
 700 /*
 701  * TCB structure exported to user-land via sysctl(3).
 702  * Evil hack: declare only if in_pcb.h and sys/socketvar.h have been
 703  * included.  Not all of our clients do.
 704  */
 705
 706 struct  xtcpcb {
 707         u_int32_t       xt_len;
 708 #ifdef KERNEL_PRIVATE
 709         struct  inpcb_compat    xt_inp;
 710 #else
 711         struct  inpcb   xt_inp;
 712 #endif
 713 #ifdef KERNEL_PRIVATE
 714         struct  otcpcb  xt_tp;
 715 #else
 716         struct  tcpcb   xt_tp;
 717 #endif
 718         struct  xsocket xt_socket;
 719         u_quad_t        xt_alignment_hack;
 720 };
 721
 722 #if !CONFIG_EMBEDDED
 723
 724 struct  xtcpcb64 {
 725         u_int32_t               xt_len;
 726         struct xinpcb64         xt_inpcb;
 727
 728         u_int64_t t_segq;
 729         int     t_dupacks;              /* consecutive dup acks recd */
 730
 731         int t_timer[TCPT_NTIMERS_EXT];  /* tcp timers */
 732
 733         int     t_state;                /* state of this connection */
 734         u_int   t_flags;
 735
 736         int     t_force;                /* 1 if forcing out a byte */
 737
 738         tcp_seq snd_una;                /* send unacknowledged */
 739         tcp_seq snd_max;                /* highest sequence number sent;
 740                                          * used to recognize retransmits
 741                                          */
 742         tcp_seq snd_nxt;                /* send next */
 743         tcp_seq snd_up;                 /* send urgent pointer */
 744
 745         tcp_seq snd_wl1;                /* window update seg seq number */
 746         tcp_seq snd_wl2;                /* window update seg ack number */
 747         tcp_seq iss;                    /* initial send sequence number */
 748         tcp_seq irs;                    /* initial receive sequence number */
 749
 750         tcp_seq rcv_nxt;                /* receive next */
 751         tcp_seq rcv_adv;                /* advertised window */
 752         u_int32_t rcv_wnd;              /* receive window */
 753         tcp_seq rcv_up;                 /* receive urgent pointer */
 754
 755         u_int32_t snd_wnd;              /* send window */
 756         u_int32_t snd_cwnd;             /* congestion-controlled window */
 757         u_int32_t snd_ssthresh;         /* snd_cwnd size threshold for
 758                                          * for slow start exponential to
 759                                          * linear switch
 760                                          */
 761         u_int   t_maxopd;               /* mss plus options */
 762
 763         u_int32_t t_rcvtime;            /* time at which a packet was received */
 764         u_int32_t t_starttime;          /* time connection was established */
 765         int     t_rtttime;              /* round trip time */
 766         tcp_seq t_rtseq;                /* sequence number being timed */
 767
 768         int     t_rxtcur;               /* current retransmit value (ticks) */
 769         u_int   t_maxseg;               /* maximum segment size */
 770         int     t_srtt;                 /* smoothed round-trip time */
 771         int     t_rttvar;               /* variance in round-trip time */
 772
 773         int     t_rxtshift;             /* log(2) of rexmt exp. backoff */
 774         u_int   t_rttmin;               /* minimum rtt allowed */
 775         u_int32_t t_rttupdated;         /* number of times rtt sampled */
 776         u_int32_t max_sndwnd;           /* largest window peer has offered */
 777
 778         int     t_softerror;            /* possible error not yet reported */
 779 /* out-of-band data */
 780         char    t_oobflags;             /* have some */
 781         char    t_iobc;                 /* input character */
 782 /* RFC 1323 variables */
 783         u_char  snd_scale;              /* window scaling for send window */
 784         u_char  rcv_scale;              /* window scaling for recv window */
 785         u_char  request_r_scale;        /* pending window scaling */
 786         u_char  requested_s_scale;
 787         u_int32_t ts_recent;            /* timestamp echo data */
 788
 789         u_int32_t ts_recent_age;        /* when last updated */
 790         tcp_seq last_ack_sent;
 791 /* RFC 1644 variables */
 792         tcp_cc  cc_send;                /* send connection count */
 793         tcp_cc  cc_recv;                /* receive connection count */
 794         tcp_seq snd_recover;            /* for use in fast recovery */
 795 /* experimental */
 796         u_int32_t snd_cwnd_prev;        /* cwnd prior to retransmit */
 797         u_int32_t snd_ssthresh_prev;    /* ssthresh prior to retransmit */
 798         u_int32_t t_badrxtwin;          /* window for retransmit recovery */
 799
 800         u_quad_t                xt_alignment_hack;
 801 };
 802
 803 #endif /* !CONFIG_EMBEDDED */
 804
 805 #ifdef PRIVATE
 806
 807 struct  xtcpcb_n {
 808         u_int32_t               xt_len;
 809         u_int32_t                       xt_kind;                /* XSO_TCPCB */
 810
 811         u_int64_t t_segq;
 812         int     t_dupacks;              /* consecutive dup acks recd */
 813
 814         int t_timer[TCPT_NTIMERS_EXT];  /* tcp timers */
 815
 816         int     t_state;                /* state of this connection */
 817         u_int   t_flags;
 818
 819         int     t_force;                /* 1 if forcing out a byte */
 820
 821         tcp_seq snd_una;                /* send unacknowledged */
 822         tcp_seq snd_max;                /* highest sequence number sent;
 823                                                                          * used to recognize retransmits
 824                                                                          */
 825         tcp_seq snd_nxt;                /* send next */
 826         tcp_seq snd_up;                 /* send urgent pointer */
 827
 828         tcp_seq snd_wl1;                /* window update seg seq number */
 829         tcp_seq snd_wl2;                /* window update seg ack number */
 830         tcp_seq iss;                    /* initial send sequence number */
 831         tcp_seq irs;                    /* initial receive sequence number */
 832
 833         tcp_seq rcv_nxt;                /* receive next */
 834         tcp_seq rcv_adv;                /* advertised window */
 835         u_int32_t rcv_wnd;              /* receive window */
 836         tcp_seq rcv_up;                 /* receive urgent pointer */
 837
 838         u_int32_t snd_wnd;              /* send window */
 839         u_int32_t snd_cwnd;             /* congestion-controlled window */
 840         u_int32_t snd_ssthresh;         /* snd_cwnd size threshold for
 841                                                                          * for slow start exponential to
 842                                                                          * linear switch
 843                                                                          */
 844         u_int   t_maxopd;               /* mss plus options */
 845
 846         u_int32_t t_rcvtime;            /* time at which a packet was received */
 847         u_int32_t t_starttime;          /* time connection was established */
 848         int     t_rtttime;              /* round trip time */
 849         tcp_seq t_rtseq;                /* sequence number being timed */
 850
 851         int     t_rxtcur;               /* current retransmit value (ticks) */
 852         u_int   t_maxseg;               /* maximum segment size */
 853         int     t_srtt;                 /* smoothed round-trip time */
 854         int     t_rttvar;               /* variance in round-trip time */
 855
 856         int     t_rxtshift;             /* log(2) of rexmt exp. backoff */
 857         u_int   t_rttmin;               /* minimum rtt allowed */
 858         u_int32_t t_rttupdated;         /* number of times rtt sampled */
 859         u_int32_t max_sndwnd;           /* largest window peer has offered */
 860
 861         int     t_softerror;            /* possible error not yet reported */
 862         /* out-of-band data */
 863         char    t_oobflags;             /* have some */
 864         char    t_iobc;                 /* input character */
 865         /* RFC 1323 variables */
 866         u_char  snd_scale;              /* window scaling for send window */
 867         u_char  rcv_scale;              /* window scaling for recv window */
 868         u_char  request_r_scale;        /* pending window scaling */
 869         u_char  requested_s_scale;
 870         u_int32_t ts_recent;            /* timestamp echo data */
 871
 872         u_int32_t ts_recent_age;        /* when last updated */
 873         tcp_seq last_ack_sent;
 874         /* RFC 1644 variables */
 875         tcp_cc  cc_send;                /* send connection count */
 876         tcp_cc  cc_recv;                /* receive connection count */
 877         tcp_seq snd_recover;            /* for use in fast recovery */
 878         /* experimental */
 879         u_int32_t snd_cwnd_prev;        /* cwnd prior to retransmit */
 880         u_int32_t snd_ssthresh_prev;    /* ssthresh prior to retransmit */
 881         u_int32_t t_badrxtwin;          /* window for retransmit recovery */
 882 };
 883
 884 #endif /* PRIVATE */
 885
 886 #pragma pack()
 887
 888 /*
 889  * Names for TCP sysctl objects
 890  */
 891 #define TCPCTL_DO_RFC1323       1       /* use RFC-1323 extensions */
 892 #define TCPCTL_DO_RFC1644       2       /* use RFC-1644 extensions */
 893 #define TCPCTL_MSSDFLT          3       /* MSS default */
 894 #define TCPCTL_STATS            4       /* statistics (read-only) */
 895 #define TCPCTL_RTTDFLT          5       /* default RTT estimate */
 896 #define TCPCTL_KEEPIDLE         6       /* keepalive idle timer */
 897 #define TCPCTL_KEEPINTVL        7       /* interval to send keepalives */
 898 #define TCPCTL_SENDSPACE        8       /* send buffer space */
 899 #define TCPCTL_RECVSPACE        9       /* receive buffer space */
 900 #define TCPCTL_KEEPINIT         10      /* timeout for establishing syn */
 901 #define TCPCTL_PCBLIST          11      /* list of all outstanding PCBs */
 902 #define TCPCTL_DELACKTIME       12      /* time before sending delayed ACK */
 903 #define TCPCTL_V6MSSDFLT        13      /* MSS default for IPv6 */
 904 #define TCPCTL_MAXID            14
 905
 906 #ifdef KERNEL_PRIVATE
 907 #define TCP_PKTLIST_CLEAR(tp) {                                         \
 908         (tp)->t_pktlist_head = (tp)->t_pktlist_tail = NULL;             \
 909         (tp)->t_lastchain = (tp)->t_pktlist_sentlen = 0;                \
 910 }
 911
 912 #define TCPCTL_NAMES { \
 913         { 0, 0 }, \
 914         { "rfc1323", CTLTYPE_INT }, \
 915         { "rfc1644", CTLTYPE_INT }, \
 916         { "mssdflt", CTLTYPE_INT }, \
 917         { "stats", CTLTYPE_STRUCT }, \
 918         { "rttdflt", CTLTYPE_INT }, \
 919         { "keepidle", CTLTYPE_INT }, \
 920         { "keepintvl", CTLTYPE_INT }, \
 921         { "sendspace", CTLTYPE_INT }, \
 922         { "recvspace", CTLTYPE_INT }, \
 923         { "keepinit", CTLTYPE_INT }, \
 924         { "pcblist", CTLTYPE_STRUCT }, \
 925         { "delacktime", CTLTYPE_INT }, \
 926         { "v6mssdflt", CTLTYPE_INT }, \
 927 }
 928
 929 #ifdef SYSCTL_DECL
 930 SYSCTL_DECL(_net_inet_tcp);
 931 #endif /* SYSCTL_DECL */
 932
 933 extern  struct inpcbhead tcb;           /* head of queue of active tcpcb's */
 934 extern  struct inpcbinfo tcbinfo;
 935 extern  struct tcpstat tcpstat; /* tcp statistics */
 936 extern  int tcp_mssdflt;        /* XXX */
 937 extern  int tcp_minmss;
 938 extern  int tcp_minmssoverload;
 939 extern  int ss_fltsz;
 940 extern  int ss_fltsz_local;
 941 extern  int tcp_do_rfc3390;             /* Calculate ss_fltsz according to RFC 3390 */
 942 #ifdef __APPLE__
 943 extern  u_int32_t tcp_now;              /* for RFC 1323 timestamps */
 944 extern struct timeval tcp_uptime;
 945 extern lck_spin_t *tcp_uptime_lock;
 946
 947 extern  int tcp_delack_enabled;
 948 #endif /* __APPLE__ */
 949
 950 extern  int tcp_do_sack;        /* SACK enabled/disabled */
 951
 952 #if CONFIG_IFEF_NOWINDOWSCALE
 953 extern int tcp_obey_ifef_nowindowscale;
 954 #endif
 955
 956 void     tcp_canceltimers(struct tcpcb *);
 957 struct tcpcb *
 958          tcp_close(struct tcpcb *);
 959 void     tcp_ctlinput(int, struct sockaddr *, void *);
 960 int      tcp_ctloutput(struct socket *, struct sockopt *);
 961 struct tcpcb *
 962          tcp_drop(struct tcpcb *, int);
 963 void     tcp_drain(void);
 964 struct rmxp_tao *
 965          tcp_gettaocache(struct inpcb *);
 966 void     tcp_init(void) __attribute__((section("__TEXT, initcode")));
 967 void     tcp_input(struct mbuf *, int);
 968 void     tcp_mss(struct tcpcb *, int, unsigned int);
 969 int      tcp_mssopt(struct tcpcb *);
 970 void     tcp_drop_syn_sent(struct inpcb *, int);
 971 void     tcp_mtudisc(struct inpcb *, int);
 972 struct tcpcb *
 973          tcp_newtcpcb(struct inpcb *);
 974 int      tcp_output(struct tcpcb *);
 975 void     tcp_respond(struct tcpcb *, void *,
 976             struct tcphdr *, struct mbuf *, tcp_seq, tcp_seq, int,
 977             unsigned int, unsigned int);
 978 struct rtentry *tcp_rtlookup(struct inpcb *, unsigned int);
 979 void     tcp_setpersist(struct tcpcb *);
 980 void     tcp_slowtimo(void);
 981 void     tcp_check_timer_state(struct tcpcb *tp);
 982 void     tcp_run_timerlist(void *arg1, void *arg2);
 983
 984 struct tcptemp *
 985          tcp_maketemplate(struct tcpcb *);
 986 void     tcp_fillheaders(struct tcpcb *, void *, void *);
 987 struct tcpcb *
 988          tcp_timers(struct tcpcb *, int);
 989 void     tcp_trace(int, int, struct tcpcb *, void *, struct tcphdr *, int);
 990
 991 void tcp_sack_doack(struct tcpcb *, struct tcpopt *, tcp_seq);
 992 void     tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart, tcp_seq rcv_lastend);
 993 void     tcp_clean_sackreport(struct tcpcb *tp);
 994 void     tcp_sack_adjust(struct tcpcb *tp);
 995 struct sackhole *tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt);
 996 void     tcp_sack_partialack(struct tcpcb *, struct tcphdr *);
 997 void     tcp_free_sackholes(struct tcpcb *tp);
 998 int32_t  tcp_sbspace(struct tcpcb *tp);
 999 void     tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp);
1000 void     tcp_reset_stretch_ack(struct tcpcb *tp);
1001
1002 #if TRAFFIC_MGT
1003 void     reset_acc_iaj(struct tcpcb *tp);
1004 #endif /* TRAFFIC_MGT */
1005
1006 int      tcp_lock (struct socket *, int, void *);
1007 int      tcp_unlock (struct socket *, int, void *);
1008 void     calculate_tcp_clock(void);
1009
1010 #ifdef _KERN_LOCKS_H_
1011 lck_mtx_t *      tcp_getlock (struct socket *, int);
1012 #else
1013 void *   tcp_getlock (struct socket *, int);
1014 #endif
1015
1016
1017 extern  struct pr_usrreqs tcp_usrreqs;
1018 extern  u_int32_t tcp_sendspace;
1019 extern  u_int32_t tcp_recvspace;
1020 tcp_seq tcp_new_isn(struct tcpcb *);
1021
1022 #endif /* KERNEL_RPIVATE */
1023
1024 #endif /* _NETINET_TCP_VAR_H_ */