1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_compat.h" 38 #include "opt_tcpdebug.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/sysctl.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/protosw.h> 47 48 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 49 50 #include <net/route.h> 51 52 #include <netinet/in.h> 53 #include <netinet/in_systm.h> 54 #include <netinet/in_pcb.h> 55 #include <netinet/ip_var.h> 56 #include <netinet/tcp.h> 57 #include <netinet/tcp_fsm.h> 58 #include <netinet/tcp_seq.h> 59 #include <netinet/tcp_timer.h> 60 #include <netinet/tcp_var.h> 61 #include <netinet/tcpip.h> 62 #ifdef TCPDEBUG 63 #include <netinet/tcp_debug.h> 64 #endif 65 66 static int 67 sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS 68 { 69 int error, s, tt; 70 71 tt = *(int *)oidp->oid_arg1; 72 s = tt * 1000 / hz; 73 74 error = sysctl_handle_int(oidp, &s, 0, req); 75 if (error || !req->newptr) 76 return (error); 77 78 tt = s * hz / 1000; 79 if (tt < 1) 80 return (EINVAL); 81 82 *(int *)oidp->oid_arg1 = tt; 83 return (0); 84 } 85 86 int tcp_keepinit; 87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW, 88 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", ""); 89 90 int tcp_keepidle; 91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW, 92 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", ""); 93 94 int tcp_keepintvl; 95 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW, 96 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", ""); 97 98 int tcp_delacktime; 99 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, 100 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I", 101 "Time before a delayed ACK is sent"); 102 103 int tcp_msl; 104 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW, 105 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime"); 106 107 static int always_keepalive = 0; 108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW, 109 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections"); 110 111 static int tcp_keepcnt = TCPTV_KEEPCNT; 112 /* max idle probes */ 113 int tcp_maxpersistidle; 114 /* max idle time in persist */ 115 int tcp_maxidle; 116 117 /* 118 * Tcp protocol timeout routine called every 500 ms. 119 * Updates timestamps used for TCP 120 * causes finite state machine actions if timers expire. 121 */ 122 void 123 tcp_slowtimo() 124 { 125 int s; 126 127 s = splnet(); 128 129 tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 130 131 tcp_iss += TCP_ISSINCR/PR_SLOWHZ; /* increment iss */ 132 #ifdef TCP_COMPAT_42 133 if ((int)tcp_iss < 0) 134 tcp_iss = TCP_ISSINCR; /* XXX */ 135 #endif 136 splx(s); 137 } 138 139 /* 140 * Cancel all timers for TCP tp. 141 */ 142 void 143 tcp_canceltimers(tp) 144 struct tcpcb *tp; 145 { 146 callout_stop(tp->tt_2msl); 147 callout_stop(tp->tt_persist); 148 callout_stop(tp->tt_keep); 149 callout_stop(tp->tt_rexmt); 150 } 151 152 int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 153 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; 154 155 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ 156 157 /* 158 * TCP timer processing. 159 */ 160 void 161 tcp_timer_delack(xtp) 162 void *xtp; 163 { 164 struct tcpcb *tp = xtp; 165 int s; 166 167 s = splnet(); 168 if (callout_pending(tp->tt_delack)) { 169 splx(s); 170 return; 171 } 172 callout_deactivate(tp->tt_delack); 173 174 tp->t_flags |= TF_ACKNOW; 175 tcpstat.tcps_delack++; 176 (void) tcp_output(tp); 177 splx(s); 178 } 179 180 void 181 tcp_timer_2msl(xtp) 182 void *xtp; 183 { 184 struct tcpcb *tp = xtp; 185 int s; 186 #ifdef TCPDEBUG 187 int ostate; 188 189 ostate = tp->t_state; 190 #endif 191 s = splnet(); 192 if (callout_pending(tp->tt_2msl)) { 193 splx(s); 194 return; 195 } 196 callout_deactivate(tp->tt_2msl); 197 /* 198 * 2 MSL timeout in shutdown went off. If we're closed but 199 * still waiting for peer to close and connection has been idle 200 * too long, or if 2MSL time is up from TIME_WAIT, delete connection 201 * control block. Otherwise, check again in a bit. 202 */ 203 if (tp->t_state != TCPS_TIME_WAIT && 204 (ticks - tp->t_rcvtime) <= tcp_maxidle) 205 callout_reset(tp->tt_2msl, tcp_keepintvl, 206 tcp_timer_2msl, tp); 207 else 208 tp = tcp_close(tp); 209 210 #ifdef TCPDEBUG 211 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 212 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, 213 PRU_SLOWTIMO); 214 #endif 215 splx(s); 216 } 217 218 void 219 tcp_timer_keep(xtp) 220 void *xtp; 221 { 222 struct tcpcb *tp = xtp; 223 int s; 224 #ifdef TCPDEBUG 225 int ostate; 226 227 ostate = tp->t_state; 228 #endif 229 s = splnet(); 230 if (callout_pending(tp->tt_keep)) { 231 splx(s); 232 return; 233 } 234 callout_deactivate(tp->tt_keep); 235 /* 236 * Keep-alive timer went off; send something 237 * or drop connection if idle for too long. 238 */ 239 tcpstat.tcps_keeptimeo++; 240 if (tp->t_state < TCPS_ESTABLISHED) 241 goto dropit; 242 if ((always_keepalive || 243 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) && 244 tp->t_state <= TCPS_CLOSING) { 245 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle) 246 goto dropit; 247 /* 248 * Send a packet designed to force a response 249 * if the peer is up and reachable: 250 * either an ACK if the connection is still alive, 251 * or an RST if the peer has closed the connection 252 * due to timeout or reboot. 253 * Using sequence number tp->snd_una-1 254 * causes the transmitted zero-length segment 255 * to lie outside the receive window; 256 * by the protocol spec, this requires the 257 * correspondent TCP to respond. 258 */ 259 tcpstat.tcps_keepprobe++; 260 #ifdef TCP_COMPAT_42 261 /* 262 * The keepalive packet must have nonzero length 263 * to get a 4.2 host to respond. 264 */ 265 tcp_respond(tp, tp->t_template, (struct mbuf *)NULL, 266 tp->rcv_nxt - 1, tp->snd_una - 1, 0); 267 #else 268 tcp_respond(tp, tp->t_template, (struct mbuf *)NULL, 269 tp->rcv_nxt, tp->snd_una - 1, 0); 270 #endif 271 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp); 272 } else 273 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 274 275 #ifdef TCPDEBUG 276 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 277 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, 278 PRU_SLOWTIMO); 279 #endif 280 splx(s); 281 return; 282 283 dropit: 284 tcpstat.tcps_keepdrops++; 285 tp = tcp_drop(tp, ETIMEDOUT); 286 287 #ifdef TCPDEBUG 288 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 289 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, 290 PRU_SLOWTIMO); 291 #endif 292 splx(s); 293 } 294 295 void 296 tcp_timer_persist(xtp) 297 void *xtp; 298 { 299 struct tcpcb *tp = xtp; 300 int s; 301 #ifdef TCPDEBUG 302 int ostate; 303 304 ostate = tp->t_state; 305 #endif 306 s = splnet(); 307 if (callout_pending(tp->tt_persist)) { 308 splx(s); 309 return; 310 } 311 callout_deactivate(tp->tt_persist); 312 /* 313 * Persistance timer into zero window. 314 * Force a byte to be output, if possible. 315 */ 316 tcpstat.tcps_persisttimeo++; 317 /* 318 * Hack: if the peer is dead/unreachable, we do not 319 * time out if the window is closed. After a full 320 * backoff, drop the connection if the idle time 321 * (no responses to probes) reaches the maximum 322 * backoff that we would use if retransmitting. 323 */ 324 if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 325 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle || 326 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { 327 tcpstat.tcps_persistdrop++; 328 tp = tcp_drop(tp, ETIMEDOUT); 329 goto out; 330 } 331 tcp_setpersist(tp); 332 tp->t_force = 1; 333 (void) tcp_output(tp); 334 tp->t_force = 0; 335 336 out: 337 #ifdef TCPDEBUG 338 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 339 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, 340 PRU_SLOWTIMO); 341 #endif 342 splx(s); 343 } 344 345 void 346 tcp_timer_rexmt(xtp) 347 void *xtp; 348 { 349 struct tcpcb *tp = xtp; 350 int s; 351 int rexmt; 352 #ifdef TCPDEBUG 353 int ostate; 354 355 ostate = tp->t_state; 356 #endif 357 s = splnet(); 358 if (callout_pending(tp->tt_rexmt)) { 359 splx(s); 360 return; 361 } 362 callout_deactivate(tp->tt_rexmt); 363 /* 364 * Retransmission timer went off. Message has not 365 * been acked within retransmit interval. Back off 366 * to a longer retransmit interval and retransmit one segment. 367 */ 368 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 369 tp->t_rxtshift = TCP_MAXRXTSHIFT; 370 tcpstat.tcps_timeoutdrop++; 371 tp = tcp_drop(tp, tp->t_softerror ? 372 tp->t_softerror : ETIMEDOUT); 373 goto out; 374 } 375 if (tp->t_rxtshift == 1) { 376 /* 377 * first retransmit; record ssthresh and cwnd so they can 378 * be recovered if this turns out to be a "bad" retransmit. 379 * A retransmit is considered "bad" if an ACK for this 380 * segment is received within RTT/2 interval; the assumption 381 * here is that the ACK was already in flight. See 382 * "On Estimating End-to-End Network Path Properties" by 383 * Allman and Paxson for more details. 384 */ 385 tp->snd_cwnd_prev = tp->snd_cwnd; 386 tp->snd_ssthresh_prev = tp->snd_ssthresh; 387 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); 388 } 389 tcpstat.tcps_rexmttimeo++; 390 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 391 TCPT_RANGESET(tp->t_rxtcur, rexmt, 392 tp->t_rttmin, TCPTV_REXMTMAX); 393 /* 394 * If losing, let the lower level know and try for 395 * a better route. Also, if we backed off this far, 396 * our srtt estimate is probably bogus. Clobber it 397 * so we'll take the next rtt measurement as our srtt; 398 * move the current srtt into rttvar to keep the current 399 * retransmit times until then. 400 */ 401 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 402 in_losing(tp->t_inpcb); 403 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 404 tp->t_srtt = 0; 405 } 406 tp->snd_nxt = tp->snd_una; 407 /* 408 * Force a segment to be sent. 409 */ 410 tp->t_flags |= TF_ACKNOW; 411 /* 412 * If timing a segment in this window, stop the timer. 413 */ 414 tp->t_rtttime = 0; 415 /* 416 * Close the congestion window down to one segment 417 * (we'll open it by one segment for each ack we get). 418 * Since we probably have a window's worth of unacked 419 * data accumulated, this "slow start" keeps us from 420 * dumping all that data as back-to-back packets (which 421 * might overwhelm an intermediate gateway). 422 * 423 * There are two phases to the opening: Initially we 424 * open by one mss on each ack. This makes the window 425 * size increase exponentially with time. If the 426 * window is larger than the path can handle, this 427 * exponential growth results in dropped packet(s) 428 * almost immediately. To get more time between 429 * drops but still "push" the network to take advantage 430 * of improving conditions, we switch from exponential 431 * to linear window opening at some threshhold size. 432 * For a threshhold, we use half the current window 433 * size, truncated to a multiple of the mss. 434 * 435 * (the minimum cwnd that will give us exponential 436 * growth is 2 mss. We don't allow the threshhold 437 * to go below this.) 438 */ 439 { 440 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 441 if (win < 2) 442 win = 2; 443 tp->snd_cwnd = tp->t_maxseg; 444 tp->snd_ssthresh = win * tp->t_maxseg; 445 tp->t_dupacks = 0; 446 } 447 (void) tcp_output(tp); 448 449 out: 450 #ifdef TCPDEBUG 451 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 452 tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, 453 PRU_SLOWTIMO); 454 #endif 455 splx(s); 456 } 457