bsd/netinet/tcp_timer.h

   1 /*
   2  * Copyright (c) 2000-2014 Apple Computer, 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
  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_timer.h 8.1 (Berkeley) 6/10/93
  61  * $FreeBSD: src/sys/netinet/tcp_timer.h,v 1.18 1999/12/29 04:41:03 peter Exp $
  62  */
  63
  64 #ifndef _NETINET_TCP_TIMER_H_
  65 #define _NETINET_TCP_TIMER_H_
  66 #include <sys/appleapiopts.h>
  67
  68 #ifdef BSD_KERNEL_PRIVATE
  69 #include <kern/thread_call.h>
  70 #endif /* BSD_KERNEL_PRIVATE */
  71
  72 /* Keep the external definition the same for binary compatibility */
  73 #define TCPT_NTIMERS_EXT        4
  74
  75 /*
  76  * Definitions of the TCP timers.
  77  *
  78  * The TCPT_PTO timer is used for probing for a tail loss in a send window.
  79  * If this probe gets acknowledged using SACK, it will allow the connection
  80  * to enter fast-recovery instead of hitting a retransmit timeout. A probe
  81  * timeout will send the last unacknowledged segment to generate more acks
  82  * with SACK information which can be used for fast-retransmiting the lost
  83  * packets. This will fire in the order of 10ms.
  84  *
  85  * The TCPT_REXMT timer is used to force retransmissions.
  86  * The TCP has the TCPT_REXMT timer set whenever segments
  87  * have been sent for which ACKs are expected but not yet
  88  * received.  If an ACK is received which advances tp->snd_una,
  89  * then the retransmit timer is cleared (if there are no more
  90  * outstanding segments) or reset to the base value (if there
  91  * are more ACKs expected).  Whenever the retransmit timer goes off,
  92  * we retransmit one unacknowledged segment, and do a backoff
  93  * on the retransmit timer.
  94  *
  95  * The TCPT_DELACK timer is used for transmitting delayed acknowledgements
  96  * if an acknowledgement was delayed in anticipation of a new segment.
  97  *
  98  * The TCPT_PERSIST timer is used to keep window size information
  99  * flowing even if the window goes shut.  If all previous transmissions
 100  * have been acknowledged(so that there are no retransmissions in progress),
 101  * and the window is too small to bother sending anything, then we start
 102  * the TCPT_PERSIST timer.  When it expires, if the window is nonzero,
 103  * we go to transmit state.  Otherwise, at intervals send a single byte
 104  * into the peer's window to force him to update our window information.
 105  * We do this at most as often as TCPT_PERSMIN time intervals,
 106  * but no more frequently than the current estimate of round-trip
 107  * packet time.  The TCPT_PERSIST timer is cleared whenever we receive
 108  * a window update from the peer.
 109  *
 110  * The TCPT_KEEP timer is used to keep connections alive.  If an
 111  * connection is idle (no segments received) for TCPTV_KEEP_INIT amount
 112  * of time, but not yet established, then we drop the connection.
 113  * Once the connection is established, if the connection is idle for
 114  * TCPTV_KEEP_IDLE time (and keepalives have been enabled on the socket),
 115  * we begin to probe the connection.  We force the peer to send us a
 116  * segment by sending:
 117  *      <SEQ=SND.UNA-1><ACK=RCV.NXT><CTL=ACK>
 118  * This segment is (deliberately) outside the window, and should elicit
 119  * an ack segment in response from the peer.  If, despite the TCPT_KEEP
 120  * initiated segments we cannot elicit a response from a peer in
 121  * TCPT_MAXIDLE amount of time probing, then we drop the connection.
 122  *
 123  * The TCPT_2MSL timer is used for keeping the conenction in Time-wait state
 124  * before fully closing it so that the connection 4-tuple can be reused.
 125  */
 126 #ifdef BSD_KERNEL_PRIVATE
 127
 128 #define TCPT_PTO        0       /* Probe timeout */
 129 #define TCPT_DELAYFR    1       /* Delay recovery if there is reordering */
 130 #define TCPT_REXMT      2       /* retransmit */
 131 #define TCPT_DELACK     3       /* delayed ack */
 132 #define TCPT_PERSIST    4       /* retransmit persistence */
 133 #define TCPT_KEEP       5       /* keep alive */
 134 #define TCPT_2MSL       6       /* 2*msl quiet time timer */
 135 #if MPTCP
 136 #define TCPT_JACK_RXMT  7       /* retransmit timer for join ack */
 137 #define TCPT_MAX        7
 138 #else /* MPTCP */
 139 #define TCPT_MAX        6
 140 #endif /* !MPTCP */
 141
 142 #define TCPT_NONE       (TCPT_MAX + 1)
 143 #define TCPT_NTIMERS    (TCPT_MAX + 1)
 144
 145 /* External definitions */
 146 #define TCPT_REXMT_EXT          0
 147 #define TCPT_PERSIST_EXT        1
 148 #define TCPT_KEEP_EXT           2
 149 #define TCPT_2MSL_EXT           3
 150 #define TCPT_DELACK_EXT         4
 151
 152 #else /* !BSD_KERNEL_PRIVATE */
 153 #define TCPT_REXMT      0               /* retransmit */
 154 #define TCPT_PERSIST    1               /* retransmit persistence */
 155 #define TCPT_KEEP       2               /* keep alive */
 156 #define TCPT_2MSL       3               /* 2*msl quiet time timer */
 157 #define TCPT_DELACK     4               /* delayed ack timer */
 158 #if MPTCP
 159 #define TCPT_JACK_RXMT  5       /* retransmit timer for join ack */
 160 #define TCPT_MAX        5
 161 #else /* MPTCP */
 162 #define TCPT_MAX        4
 163 #endif /* !MPTCP */
 164 #define TCPT_NONE       (TCPT_MAX + 1)
 165 #define TCPT_NTIMERS    (TCPT_MAX + 1)
 166
 167 #endif /* BSD_KERNEL_PRIVATE */
 168
 169 #ifdef BSD_KERNEL_PRIVATE
 170 /*
 171  * Time constants.
 172  */
 173 #define TCPTV_MSL       ( 15*TCP_RETRANSHZ)     /* max seg lifetime */
 174 #define TCPTV_SRTTBASE  0       /* base roundtrip time; if 0, no idea yet */
 175 #define TCPTV_RTOBASE   (  1*TCP_RETRANSHZ)     /* assumed RTO if no info */
 176 #define TCPTV_SRTTDFLT  (  1*TCP_RETRANSHZ)     /* assumed RTT if no info */
 177 #define TCPTV_PERSMIN   (  5*TCP_RETRANSHZ)     /* retransmit persistence */
 178 #define TCPTV_PERSMAX   ( 60*TCP_RETRANSHZ)     /* maximum persist interval */
 179
 180 #define TCPTV_KEEP_INIT ( 75*TCP_RETRANSHZ)     /* connect keep alive */
 181 #define TCPTV_KEEP_IDLE (120*60*TCP_RETRANSHZ)  /* time before probing */
 182 #define TCPTV_KEEPINTVL ( 75*TCP_RETRANSHZ)     /* default probe interval */
 183 #define TCPTV_KEEPCNT   8                       /* max probes before drop */
 184
 185 #define TCPTV_REXMTMAX  ( 64*TCP_RETRANSHZ )    /* max REXMT value */
 186 #define TCPTV_REXMTMIN  ( TCP_RETRANSHZ/33 )    /* min REXMT for non-local connections */
 187
 188 /*
 189  * Window for counting received bytes to see if ack-stretching
 190  * can start (default 100 ms)
 191  */
 192 #define TCPTV_UNACKWIN  ( TCP_RETRANSHZ/10 )
 193
 194 /* Receiver idle time, avoid ack-stretching after this idle time */
 195 #define TCPTV_MAXRCVIDLE (TCP_RETRANSHZ/5 )
 196
 197 /*
 198  * No ack stretching during slow-start, until we see some packets.
 199  * By the time the receiver gets 512 packets, the senders cwnd
 200  * should open by a few hundred packets consdering the
 201  * slow-start progression.
 202  */
 203 #define TCP_RCV_SS_PKTCOUNT     512
 204
 205 /* Receiver idle time, for rcv socket buffer resizing */
 206 #define TCPTV_RCVBUFIDLE (TCP_RETRANSHZ/2)
 207 #define TCPTV_TWTRUNC   8               /* RTO factor to truncate TW */
 208
 209 #define TCP_LINGERTIME  120             /* linger at most 2 minutes */
 210
 211 #define TCP_MAXRXTSHIFT 12              /* maximum retransmits */
 212
 213 #ifdef  TCPTIMERS
 214 static char *tcptimers[] =
 215     { "REXMT", "PERSIST", "KEEP", "2MSL" , "DELACK"};
 216 #endif /* TCPTIMERS */
 217
 218 /*
 219  * Persist, keep, 2msl and MPTCP's join-ack timer as slow timers which can
 220  * be coalesced at a higher granularity (500 ms).
 221  *
 222  * Rexmt and delayed ack timers are considered as fast timers which run
 223  * in the order of 100ms.
 224  *
 225  * Probe timeout is a quick timer which will run in the order of 10ms.
 226  */
 227 #define IS_TIMER_HZ_500MS(i)    ((i) >= TCPT_PERSIST)
 228 #define IS_TIMER_HZ_100MS(i)    ((i) >= TCPT_REXMT && (i) < TCPT_PERSIST)
 229 #define IS_TIMER_HZ_10MS(i)     ((i) < TCPT_REXMT)
 230
 231 struct tcptimerlist;
 232
 233 struct tcptimerentry {
 234         LIST_ENTRY(tcptimerentry) le;   /* links for timer list */
 235         uint32_t timer_start;   /* tcp clock when the timer was started */
 236         uint16_t index;         /* index of lowest timer that needs to run first */
 237         uint16_t mode;          /* Bit-wise OR of timers that are active */
 238         uint32_t runtime;       /* deadline at which the first timer has to fire */
 239 };
 240
 241 LIST_HEAD(timerlisthead, tcptimerentry);
 242
 243 struct tcptimerlist {
 244         struct timerlisthead lhead;     /* head of the list */
 245         lck_mtx_t *mtx;         /* lock to protect the list */
 246         lck_attr_t *mtx_attr;   /* mutex attributes */
 247         lck_grp_t *mtx_grp;     /* mutex group definition */
 248         lck_grp_attr_t *mtx_grp_attr;   /* mutex group attributes */
 249         thread_call_t call;     /* call entry */
 250         uint32_t runtime;       /* time at which this list is going to run */
 251         uint32_t entries;       /* Number of entries on the list */
 252         uint32_t maxentries;    /* Max number of entries at any time */
 253
 254         /* Set desired mode when timer list running */
 255         boolean_t running;      /* Set when timer list is being processed */
 256         boolean_t scheduled;    /* set when the timer is scheduled */
 257 #define TCP_TIMERLIST_10MS_MODE 0x1
 258 #define TCP_TIMERLIST_100MS_MODE 0x2
 259 #define TCP_TIMERLIST_500MS_MODE 0x4
 260         uint32_t mode;          /* Current mode of the timer */
 261         uint32_t pref_mode;     /* Preferred mode set by a connection */
 262         uint32_t pref_offset;   /* Preferred offset set by a connection */
 263         uint32_t idleruns;      /* Number of times the list has been idle in fast mode */
 264         struct tcptimerentry *next_te;  /* next timer entry pointer to process */
 265         u_int16_t probe_if_index; /* Interface index that needs to send probes */
 266
 267 };
 268
 269 /* number of idle runs allowed for TCP timer list in fast or quick modes */
 270 #define TCP_FASTMODE_IDLERUN_MAX 10
 271
 272 /*
 273  * Minimum retransmit timeout is set to 30ms. We add a slop of
 274  * 200 ms to the retransmit value to account for processing
 275  * variance and delayed ack. This extra 200ms will help to avoid
 276  * spurious retransmits by taking into consideration the receivers
 277  * that wait for delayed ack timer instead of generating an ack
 278  * for every two packets.
 279  *
 280  * On a local link, the minimum retransmit timeout is 100ms and
 281  * variance is set to 0. This will make the sender a little bit more
 282  * aggressive on local link. When the connection is not established yet,
 283  * there is no need to add an extra 200ms to retransmit timeout because
 284  * the initial value is high (1s) and delayed ack is not a problem in
 285  * that case.
 286  */
 287 #define TCPTV_REXMTSLOP ( TCP_RETRANSHZ/5 )     /* extra 200 ms slop */
 288
 289 /* macro to decide when retransmit slop (described above) should be added */
 290 #define TCP_ADD_REXMTSLOP(tp) (tp->t_state >= TCPS_ESTABLISHED)
 291
 292 #define TCPT_RANGESET(tv, value, tvmin, tvmax, addslop) do { \
 293         (tv) = ((addslop) ? tcp_rexmt_slop : 0) + (value); \
 294         if ((uint32_t)(tv) < (uint32_t)(tvmin)) \
 295                 (tv) = (tvmin); \
 296         else if ((uint32_t)(tv) > (uint32_t)(tvmax)) \
 297                 (tv) = (tvmax); \
 298 } while(0)
 299
 300 #define TCP_CONN_KEEPIDLE(tp) \
 301         ((tp)->t_keepidle && \
 302         ((tp)->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) ? \
 303                 (tp)->t_keepidle : tcp_keepidle)
 304 #define TCP_CONN_KEEPINIT(tp) \
 305         (((tp)->t_keepinit > 0) ? (tp)->t_keepinit : tcp_keepinit)
 306 #define TCP_CONN_KEEPCNT(tp) \
 307         (((tp)->t_keepcnt > 0) ? (tp)->t_keepcnt : tcp_keepcnt)
 308 #define TCP_CONN_KEEPINTVL(tp) \
 309         (((tp)->t_keepintvl > 0) ? (tp)->t_keepintvl : tcp_keepintvl)
 310 #define TCP_CONN_MAXIDLE(tp) \
 311         (TCP_CONN_KEEPCNT(tp) * TCP_CONN_KEEPINTVL(tp))
 312
 313 #define TCP_IDLETIMEOUT(tp) \
 314         (((TCP_ADD_REXMTSLOP(tp)) ? 0 : tcp_rexmt_slop) + tp->t_rxtcur)
 315
 316 TAILQ_HEAD(tcptailq, tcpcb);
 317
 318 extern int tcp_keepinit;        /* time to establish connection */
 319 extern int tcp_keepidle;        /* time before keepalive probes begin */
 320 extern int tcp_keepintvl;       /* time between keepalive probes */
 321 extern int tcp_keepcnt;         /* number of keepalives */
 322 extern int tcp_delack;          /* delayed ack timer */
 323 extern int tcp_maxpersistidle;
 324 extern int tcp_msl;
 325 extern int tcp_ttl;             /* time to live for TCP segs */
 326 extern int tcp_backoff[];
 327 extern int tcp_rexmt_slop;
 328 extern u_int32_t tcp_max_persist_timeout;       /* Maximum persistence for Zero Window Probes */
 329
 330 #define OFFSET_FROM_START(tp, off) ((tcp_now + (off)) - (tp)->tentry.timer_start)
 331
 332 #endif /* BSD_KERNEL_PRIVATE */
 333 #endif /* !_NETINET_TCP_TIMER_H_ */
 334