bsd/netinet/tcp_timer.h

   1 /*
   2  * Copyright (c) 2000-2010 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 KERNEL
  69 #include <kern/thread_call.h>
  70 #endif /* KERNEL */
  71
  72 /*
  73  * Definitions of the TCP timers.
  74  */
  75 #define TCPT_NTIMERS    5
  76
  77 /* Keep the external definition the same for binary compatibility */
  78 #define TCPT_NTIMERS_EXT        4
  79
  80 #define TCPT_REXMT      0               /* retransmit */
  81 #define TCPT_PERSIST    1               /* retransmit persistence */
  82 #define TCPT_KEEP       2               /* keep alive */
  83 #define TCPT_2MSL       3               /* 2*msl quiet time timer */
  84 #define TCPT_DELACK     4               /* delayed ack timer */
  85 #define TCPT_MAX        4
  86 #define TCPT_NONE       (TCPT_MAX + 1)
  87
  88 /*
  89  * The TCPT_REXMT timer is used to force retransmissions.
  90  * The TCP has the TCPT_REXMT timer set whenever segments
  91  * have been sent for which ACKs are expected but not yet
  92  * received.  If an ACK is received which advances tp->snd_una,
  93  * then the retransmit timer is cleared (if there are no more
  94  * outstanding segments) or reset to the base value (if there
  95  * are more ACKs expected).  Whenever the retransmit timer goes off,
  96  * we retransmit one unacknowledged segment, and do a backoff
  97  * on the retransmit timer.
  98  *
  99  * The TCPT_PERSIST timer is used to keep window size information
 100  * flowing even if the window goes shut.  If all previous transmissions
 101  * have been acknowledged (so that there are no retransmissions in progress),
 102  * and the window is too small to bother sending anything, then we start
 103  * the TCPT_PERSIST timer.  When it expires, if the window is nonzero,
 104  * we go to transmit state.  Otherwise, at intervals send a single byte
 105  * into the peer's window to force him to update our window information.
 106  * We do this at most as often as TCPT_PERSMIN time intervals,
 107  * but no more frequently than the current estimate of round-trip
 108  * packet time.  The TCPT_PERSIST timer is cleared whenever we receive
 109  * a window update from the peer.
 110  *
 111  * The TCPT_KEEP timer is used to keep connections alive.  If an
 112  * connection is idle (no segments received) for TCPTV_KEEP_INIT amount of time,
 113  * but not yet established, then we drop the connection.  Once the connection
 114  * is established, if the connection is idle for TCPTV_KEEP_IDLE time
 115  * (and keepalives have been enabled on the socket), we begin to probe
 116  * the connection.  We force the peer to send us a 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 TCPT_MAXIDLE
 121  * amount of time probing, then we drop the connection.
 122  */
 123
 124 #ifdef PRIVATE
 125
 126 /*
 127  * Time constants.
 128  */
 129 #define TCPTV_MSL       ( 15*TCP_RETRANSHZ)             /* max seg lifetime (hah!) */
 130 #define TCPTV_SRTTBASE  0                       /* base roundtrip time;
 131                                                    if  0, no idea yet */
 132 #define TCPTV_RTOBASE   (  1*TCP_RETRANSHZ)             /* assumed RTO if no info */
 133 #define TCPTV_SRTTDFLT  (  1*TCP_RETRANSHZ)             /* assumed RTT if no info */
 134
 135 #define TCPTV_PERSMIN   (  5*TCP_RETRANSHZ)             /* retransmit persistence */
 136 #define TCPTV_PERSMAX   ( 60*TCP_RETRANSHZ)             /* maximum persist interval */
 137
 138 #define TCPTV_KEEP_INIT ( 75*TCP_RETRANSHZ)             /* initial connect keep alive */
 139 #define TCPTV_KEEP_IDLE (120*60*TCP_RETRANSHZ)  /* dflt time before probing */
 140 #define TCPTV_KEEPINTVL ( 75*TCP_RETRANSHZ)             /* default probe interval */
 141 #define TCPTV_KEEPCNT   8                       /* max probes before drop */
 142
 143 #define TCPTV_REXMTMAX  ( 64*TCP_RETRANSHZ )    /* max allowable REXMT value */
 144 #define TCPTV_REXMTMIN  ( TCP_RETRANSHZ/33 )    /* min REXMT for non-local connections */
 145 #define TCPTV_UNACKWIN  ( TCP_RETRANSHZ/10 )    /* Window for counting rcv bytes to see if
 146                                                    ack-stretching can start (default 100 ms) */
 147 #define TCPTV_MAXRCVIDLE (TCP_RETRANSHZ/5 )     /* Receiver idle time, avoid ack-stretching after that*/
 148 #define TCPTV_RCVBUFIDLE (TCP_RETRANSHZ/2)      /* Receiver idle time, for rcv socket buffer resizing */
 149
 150 /* No ack stretching during slow-start, until we see some packets.
 151  * By the time the receiver gets 512 packets, the senders cwnd
 152  * should open by a few hundred packets consdering the
 153  * slow-start progression.
 154  */
 155 #define TCP_RCV_SS_PKTCOUNT     512
 156
 157 #define TCPTV_TWTRUNC   8                       /* RTO factor to truncate TW */
 158
 159 #define TCP_LINGERTIME  120                     /* linger at most 2 minutes */
 160
 161 #define TCP_MAXRXTSHIFT 12                      /* maximum retransmits */
 162
 163 #ifdef  TCPTIMERS
 164 static char *tcptimers[] =
 165     { "REXMT", "PERSIST", "KEEP", "2MSL" , "DELACK"};
 166 #endif
 167
 168 #ifdef KERNEL
 169
 170 /* We consider persist, keep and 2msl as slow timers which can be coalesced
 171  * at a higher granularity (500 ms). Rexmt and delayed ack are considered fast
 172  * timers which fire in the order of 100ms.
 173  *
 174  * The following conditional is to check if a timer is one of the slow timers. This
 175  * is fast and works well for now. If we add more slow timers for any reason,
 176  * we may need to change this.
 177  */
 178 #define IS_TIMER_SLOW(ind) ((ind & 0x3) != 0)
 179
 180 struct tcptimerlist;
 181
 182 struct tcptimerentry {
 183         LIST_ENTRY(tcptimerentry) le;   /* links for timer list */
 184         uint32_t timer_start;           /* tcp clock when the timer was started */
 185         uint16_t index;                 /* index of lowest timer that needs to run first */
 186         uint32_t runtime;               /* deadline at which the first timer has to fire */
 187 };
 188
 189 LIST_HEAD(timerlisthead, tcptimerentry);
 190
 191 struct tcptimerlist {
 192         struct timerlisthead lhead;     /* head of the list of timer entries */
 193         lck_mtx_t *mtx;                 /* lock to protect the list */
 194         lck_attr_t *mtx_attr;           /* mutex attributes */
 195         lck_grp_t *mtx_grp;             /* mutex group definition */
 196         lck_grp_attr_t *mtx_grp_attr;   /* mutex group attributes */
 197         uint32_t fast_quantum;          /* minimum time quantum to coalesce fast timers */
 198         uint32_t slow_quantum;          /* minimum time quantum to coalesce slow timers */
 199         thread_call_t call;             /* call entry */
 200         uint32_t runtime;               /* time at which this list is going to run */
 201         uint32_t entries;               /* Number of entries on the list */
 202         uint32_t maxentries;            /* Max number of entries at any time */
 203
 204         /* Set desired mode when timer list running */
 205         boolean_t running;              /* Set when timer list is being processed */
 206 #define TCP_TIMERLIST_FASTMODE 0x1
 207 #define TCP_TIMERLIST_SLOWMODE 0x2
 208         uint32_t mode;                  /* Current mode, fast or slow */
 209         uint32_t pref_mode;             /* Preferred mode set by a connection, fast or slow */
 210         uint32_t pref_offset;           /* Preferred offset set by a connection */
 211         uint32_t idlegen;               /* Number of times the list has been idle in fast mode */
 212         struct tcptimerentry *next_te;  /* Store the next timer entry pointer to process */
 213
 214 };
 215
 216 #define TCP_FASTMODE_IDLEGEN_MAX 20     /* Approximately 2 seconds */
 217
 218 /*
 219  * Minimum retransmit timeout is set to 30ms. We add a slop of
 220  * 200 ms to the retransmit value to account for processing
 221  * variance and delayed ack. This extra 200ms will help to avoid
 222  * spurious retransmits by taking into consideration the receivers
 223  * that wait for delayed ack timer instead of generating an ack
 224  * for every two packets.
 225  *
 226  * On a local link, the minimum retransmit timeout is 100ms and
 227  * variance is set to 0. This will make the sender a little bit more
 228  * aggressive on local link. When the connection is not established yet,
 229  * there is no need to add an extra 200ms to retransmit timeout because
 230  * the initial value is high (1s) and delayed ack is not a problem in
 231  * that case.
 232  */
 233 #define TCPTV_REXMTSLOP ( TCP_RETRANSHZ/5 )     /* rexmt slop allowed (200 ms) */
 234
 235 /* macro to decide when retransmit slop (described above) should be added */
 236 #define TCP_ADD_REXMTSLOP(tp) (tp->t_state >= TCPS_ESTABLISHED)
 237
 238 #define TCPT_RANGESET(tv, value, tvmin, tvmax, addslop) do { \
 239         (tv) = ((addslop) ? tcp_rexmt_slop : 0) + (value); \
 240         if ((uint32_t)(tv) < (uint32_t)(tvmin)) \
 241                 (tv) = (tvmin); \
 242         else if ((uint32_t)(tv) > (uint32_t)(tvmax)) \
 243                 (tv) = (tvmax); \
 244 } while(0)
 245
 246 #define TCP_KEEPIDLE(tp) \
 247         (tp->t_keepidle && (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) ? \
 248                 tp->t_keepidle : tcp_keepidle)
 249
 250 /* Since we did not add rexmt slop for local connections, we should add
 251  * it to idle timeout. Otherwise local connections will reach idle state
 252  * quickly
 253  */
 254 #define TCP_IDLETIMEOUT(tp) \
 255         (((TCP_ADD_REXMTSLOP(tp)) ? 0 : tcp_rexmt_slop) + tp->t_rxtcur)
 256
 257 extern int tcp_keepinit;                /* time to establish connection */
 258 extern int tcp_keepidle;                /* time before keepalive probes begin */
 259 extern int tcp_keepintvl;               /* time between keepalive probes */
 260 extern int tcp_maxidle;                 /* time to drop after starting probes */
 261 extern int tcp_delack;                  /* delayed ack timer */
 262 extern int tcp_maxpersistidle;
 263 extern int tcp_msl;
 264 extern int tcp_ttl;                     /* time to live for TCP segs */
 265 extern int tcp_backoff[];
 266 extern int tcp_rexmt_slop;
 267 extern u_int32_t tcp_max_persist_timeout;       /* Maximum persistence for Zero Window Probes */
 268
 269 #define OFFSET_FROM_START(tp, off) ((tcp_now + (off)) - (tp)->tentry.timer_start)
 270
 271 #endif /* KERNEL */
 272 #endif /* PRIVATE */
 273 #endif /* !_NETINET_TCP_TIMER_H_ */
 274