| 1 | /* |
| 2 | * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. |
| 3 | * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. |
| 4 | * |
| 5 | * This code is derived from software contributed to The DragonFly Project |
| 6 | * by Jeffrey M. Hsu. |
| 7 | * |
| 8 | * Redistribution and use in source and binary forms, with or without |
| 9 | * modification, are permitted provided that the following conditions |
| 10 | * are met: |
| 11 | * 1. Redistributions of source code must retain the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer. |
| 13 | * 2. Redistributions in binary form must reproduce the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer in the |
| 15 | * documentation and/or other materials provided with the distribution. |
| 16 | * 3. Neither the name of The DragonFly Project nor the names of its |
| 17 | * contributors may be used to endorse or promote products derived |
| 18 | * from this software without specific, prior written permission. |
| 19 | * |
| 20 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 21 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 22 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 23 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 24 | * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 25 | * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 26 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 27 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 28 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 29 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 30 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 31 | * SUCH DAMAGE. |
| 32 | */ |
| 33 | |
| 34 | /* |
| 35 | * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 |
| 36 | * The Regents of the University of California. All rights reserved. |
| 37 | * |
| 38 | * Redistribution and use in source and binary forms, with or without |
| 39 | * modification, are permitted provided that the following conditions |
| 40 | * are met: |
| 41 | * 1. Redistributions of source code must retain the above copyright |
| 42 | * notice, this list of conditions and the following disclaimer. |
| 43 | * 2. Redistributions in binary form must reproduce the above copyright |
| 44 | * notice, this list of conditions and the following disclaimer in the |
| 45 | * documentation and/or other materials provided with the distribution. |
| 46 | * 3. All advertising materials mentioning features or use of this software |
| 47 | * must display the following acknowledgement: |
| 48 | * This product includes software developed by the University of |
| 49 | * California, Berkeley and its contributors. |
| 50 | * 4. Neither the name of the University nor the names of its contributors |
| 51 | * may be used to endorse or promote products derived from this software |
| 52 | * without specific prior written permission. |
| 53 | * |
| 54 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 55 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 56 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 57 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 58 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 59 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 60 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 61 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 62 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 63 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 64 | * SUCH DAMAGE. |
| 65 | * |
| 66 | * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95 |
| 67 | * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $ |
| 68 | * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.17 2008/03/30 20:39:01 dillon Exp $ |
| 69 | */ |
| 70 | |
| 71 | #include "opt_compat.h" |
| 72 | #include "opt_inet6.h" |
| 73 | #include "opt_tcpdebug.h" |
| 74 | |
| 75 | #include <sys/param.h> |
| 76 | #include <sys/systm.h> |
| 77 | #include <sys/kernel.h> |
| 78 | #include <sys/mbuf.h> |
| 79 | #include <sys/sysctl.h> |
| 80 | #include <sys/socket.h> |
| 81 | #include <sys/socketvar.h> |
| 82 | #include <sys/protosw.h> |
| 83 | #include <sys/thread.h> |
| 84 | #include <sys/globaldata.h> |
| 85 | #include <sys/thread2.h> |
| 86 | #include <sys/msgport2.h> |
| 87 | |
| 88 | #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ |
| 89 | |
| 90 | #include <net/route.h> |
| 91 | #include <net/netmsg2.h> |
| 92 | |
| 93 | #include <netinet/in.h> |
| 94 | #include <netinet/in_systm.h> |
| 95 | #include <netinet/in_pcb.h> |
| 96 | #ifdef INET6 |
| 97 | #include <netinet6/in6_pcb.h> |
| 98 | #endif |
| 99 | #include <netinet/ip_var.h> |
| 100 | #include <netinet/tcp.h> |
| 101 | #include <netinet/tcp_fsm.h> |
| 102 | #include <netinet/tcp_seq.h> |
| 103 | #include <netinet/tcp_timer.h> |
| 104 | #include <netinet/tcp_timer2.h> |
| 105 | #include <netinet/tcp_var.h> |
| 106 | #include <netinet/tcpip.h> |
| 107 | #ifdef TCPDEBUG |
| 108 | #include <netinet/tcp_debug.h> |
| 109 | #endif |
| 110 | |
| 111 | #define TCP_TIMER_REXMT 0x01 |
| 112 | #define TCP_TIMER_PERSIST 0x02 |
| 113 | #define TCP_TIMER_KEEP 0x04 |
| 114 | #define TCP_TIMER_2MSL 0x08 |
| 115 | #define TCP_TIMER_DELACK 0x10 |
| 116 | |
| 117 | static struct tcpcb *tcp_timer_rexmt_handler(struct tcpcb *); |
| 118 | static struct tcpcb *tcp_timer_persist_handler(struct tcpcb *); |
| 119 | static struct tcpcb *tcp_timer_keep_handler(struct tcpcb *); |
| 120 | static struct tcpcb *tcp_timer_2msl_handler(struct tcpcb *); |
| 121 | static struct tcpcb *tcp_timer_delack_handler(struct tcpcb *); |
| 122 | |
| 123 | static const struct tcp_timer { |
| 124 | uint32_t tt_task; |
| 125 | struct tcpcb *(*tt_handler)(struct tcpcb *); |
| 126 | } tcp_timer_handlers[] = { |
| 127 | { TCP_TIMER_DELACK, tcp_timer_delack_handler }, |
| 128 | { TCP_TIMER_REXMT, tcp_timer_rexmt_handler }, |
| 129 | { TCP_TIMER_PERSIST, tcp_timer_persist_handler }, |
| 130 | { TCP_TIMER_KEEP, tcp_timer_keep_handler }, |
| 131 | { TCP_TIMER_2MSL, tcp_timer_2msl_handler }, |
| 132 | { 0, NULL } |
| 133 | }; |
| 134 | |
| 135 | static int |
| 136 | sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS) |
| 137 | { |
| 138 | int error, s, tt; |
| 139 | |
| 140 | tt = *(int *)oidp->oid_arg1; |
| 141 | s = (int)((int64_t)tt * 1000 / hz); |
| 142 | |
| 143 | error = sysctl_handle_int(oidp, &s, 0, req); |
| 144 | if (error || !req->newptr) |
| 145 | return (error); |
| 146 | |
| 147 | tt = (int)((int64_t)s * hz / 1000); |
| 148 | if (tt < 1) |
| 149 | return (EINVAL); |
| 150 | |
| 151 | *(int *)oidp->oid_arg1 = tt; |
| 152 | return (0); |
| 153 | } |
| 154 | |
| 155 | int tcp_keepinit; |
| 156 | SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW, |
| 157 | &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "Time to establish TCP connection"); |
| 158 | |
| 159 | int tcp_keepidle; |
| 160 | SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW, |
| 161 | &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "Time before TCP keepalive probes begin"); |
| 162 | |
| 163 | int tcp_keepintvl; |
| 164 | SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW, |
| 165 | &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "Time between TCP keepalive probes"); |
| 166 | |
| 167 | int tcp_delacktime; |
| 168 | SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, |
| 169 | CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I", |
| 170 | "Time before a delayed ACK is sent"); |
| 171 | |
| 172 | int tcp_msl; |
| 173 | SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW, |
| 174 | &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime"); |
| 175 | |
| 176 | int tcp_rexmit_min; |
| 177 | SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW, |
| 178 | &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout"); |
| 179 | |
| 180 | int tcp_rexmit_slop; |
| 181 | SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW, |
| 182 | &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", |
| 183 | "Retransmission Timer Slop"); |
| 184 | |
| 185 | static int always_keepalive = 1; |
| 186 | SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW, |
| 187 | &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections"); |
| 188 | |
| 189 | /* max idle probes */ |
| 190 | int tcp_keepcnt = TCPTV_KEEPCNT; |
| 191 | SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, |
| 192 | &tcp_keepcnt, 0, "Maximum number of keepalive probes to be sent"); |
| 193 | |
| 194 | /* max idle time in persist */ |
| 195 | int tcp_maxpersistidle; |
| 196 | |
| 197 | /* |
| 198 | * Cancel all timers for TCP tp. |
| 199 | */ |
| 200 | void |
| 201 | tcp_canceltimers(struct tcpcb *tp) |
| 202 | { |
| 203 | tcp_callout_stop(tp, tp->tt_2msl); |
| 204 | tcp_callout_stop(tp, tp->tt_persist); |
| 205 | tcp_callout_stop(tp, tp->tt_keep); |
| 206 | tcp_callout_stop(tp, tp->tt_rexmt); |
| 207 | } |
| 208 | |
| 209 | /* |
| 210 | * Caller should be in critical section |
| 211 | */ |
| 212 | static void |
| 213 | tcp_send_timermsg(struct tcpcb *tp, uint32_t task) |
| 214 | { |
| 215 | struct netmsg_tcp_timer *tmsg = tp->tt_msg; |
| 216 | |
| 217 | KKASSERT(tmsg != NULL && tmsg->tt_cpuid == mycpuid && |
| 218 | tmsg->tt_tcb != NULL); |
| 219 | |
| 220 | tmsg->tt_tasks |= task; |
| 221 | if (tmsg->tt_msg.lmsg.ms_flags & MSGF_DONE) |
| 222 | lwkt_sendmsg(tmsg->tt_msgport, &tmsg->tt_msg.lmsg); |
| 223 | } |
| 224 | |
| 225 | int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] = |
| 226 | { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 }; |
| 227 | |
| 228 | int tcp_syn_backoff_low[TCP_MAXRXTSHIFT + 1] = |
| 229 | { 1, 1, 2, 4, 8, 8, 16, 16, 32, 64, 64, 64, 64 }; |
| 230 | |
| 231 | int tcp_backoff[TCP_MAXRXTSHIFT + 1] = |
| 232 | { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 }; |
| 233 | |
| 234 | static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ |
| 235 | |
| 236 | /* Caller should be in critical section */ |
| 237 | static struct tcpcb * |
| 238 | tcp_timer_delack_handler(struct tcpcb *tp) |
| 239 | { |
| 240 | tp->t_flags |= TF_ACKNOW; |
| 241 | tcpstat.tcps_delack++; |
| 242 | tcp_output(tp); |
| 243 | return tp; |
| 244 | } |
| 245 | |
| 246 | /* |
| 247 | * TCP timer processing. |
| 248 | */ |
| 249 | void |
| 250 | tcp_timer_delack(void *xtp) |
| 251 | { |
| 252 | struct tcpcb *tp = xtp; |
| 253 | struct callout *co = &tp->tt_delack->tc_callout; |
| 254 | |
| 255 | crit_enter(); |
| 256 | if (callout_pending(co) || !callout_active(co)) { |
| 257 | crit_exit(); |
| 258 | return; |
| 259 | } |
| 260 | callout_deactivate(co); |
| 261 | tcp_send_timermsg(tp, TCP_TIMER_DELACK); |
| 262 | crit_exit(); |
| 263 | } |
| 264 | |
| 265 | /* Caller should be in critical section */ |
| 266 | static struct tcpcb * |
| 267 | tcp_timer_2msl_handler(struct tcpcb *tp) |
| 268 | { |
| 269 | #ifdef TCPDEBUG |
| 270 | int ostate; |
| 271 | #endif |
| 272 | |
| 273 | #ifdef TCPDEBUG |
| 274 | ostate = tp->t_state; |
| 275 | #endif |
| 276 | /* |
| 277 | * 2 MSL timeout in shutdown went off. If we're closed but |
| 278 | * still waiting for peer to close and connection has been idle |
| 279 | * too long, or if 2MSL time is up from TIME_WAIT, delete connection |
| 280 | * control block. Otherwise, check again in a bit. |
| 281 | */ |
| 282 | if (tp->t_state != TCPS_TIME_WAIT && |
| 283 | (ticks - tp->t_rcvtime) <= tp->t_maxidle) { |
| 284 | tcp_callout_reset(tp, tp->tt_2msl, tp->t_keepintvl, |
| 285 | tcp_timer_2msl); |
| 286 | } else { |
| 287 | tp = tcp_close(tp); |
| 288 | } |
| 289 | |
| 290 | #ifdef TCPDEBUG |
| 291 | if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) |
| 292 | tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); |
| 293 | #endif |
| 294 | return tp; |
| 295 | } |
| 296 | |
| 297 | void |
| 298 | tcp_timer_2msl(void *xtp) |
| 299 | { |
| 300 | struct tcpcb *tp = xtp; |
| 301 | struct callout *co = &tp->tt_2msl->tc_callout; |
| 302 | |
| 303 | crit_enter(); |
| 304 | if (callout_pending(co) || !callout_active(co)) { |
| 305 | crit_exit(); |
| 306 | return; |
| 307 | } |
| 308 | callout_deactivate(co); |
| 309 | tcp_send_timermsg(tp, TCP_TIMER_2MSL); |
| 310 | crit_exit(); |
| 311 | } |
| 312 | |
| 313 | /* Caller should be in critical section */ |
| 314 | static struct tcpcb * |
| 315 | tcp_timer_keep_handler(struct tcpcb *tp) |
| 316 | { |
| 317 | struct tcptemp *t_template; |
| 318 | #ifdef TCPDEBUG |
| 319 | int ostate = tp->t_state; |
| 320 | #endif |
| 321 | |
| 322 | /* |
| 323 | * Keep-alive timer went off; send something |
| 324 | * or drop connection if idle for too long. |
| 325 | */ |
| 326 | tcpstat.tcps_keeptimeo++; |
| 327 | if (tp->t_state < TCPS_ESTABLISHED) |
| 328 | goto dropit; |
| 329 | if ((always_keepalive || (tp->t_flags & TF_KEEPALIVE) || |
| 330 | (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE)) && |
| 331 | tp->t_state <= TCPS_CLOSING) { |
| 332 | if ((ticks - tp->t_rcvtime) >= tp->t_keepidle + tp->t_maxidle) |
| 333 | goto dropit; |
| 334 | /* |
| 335 | * Send a packet designed to force a response |
| 336 | * if the peer is up and reachable: |
| 337 | * either an ACK if the connection is still alive, |
| 338 | * or an RST if the peer has closed the connection |
| 339 | * due to timeout or reboot. |
| 340 | * Using sequence number tp->snd_una-1 |
| 341 | * causes the transmitted zero-length segment |
| 342 | * to lie outside the receive window; |
| 343 | * by the protocol spec, this requires the |
| 344 | * correspondent TCP to respond. |
| 345 | */ |
| 346 | tcpstat.tcps_keepprobe++; |
| 347 | t_template = tcp_maketemplate(tp); |
| 348 | if (t_template) { |
| 349 | tcp_respond(tp, t_template->tt_ipgen, |
| 350 | &t_template->tt_t, NULL, |
| 351 | tp->rcv_nxt, tp->snd_una - 1, 0); |
| 352 | tcp_freetemplate(t_template); |
| 353 | } |
| 354 | tcp_callout_reset(tp, tp->tt_keep, tp->t_keepintvl, |
| 355 | tcp_timer_keep); |
| 356 | } else { |
| 357 | tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, |
| 358 | tcp_timer_keep); |
| 359 | } |
| 360 | |
| 361 | #ifdef TCPDEBUG |
| 362 | if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) |
| 363 | tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); |
| 364 | #endif |
| 365 | return tp; |
| 366 | |
| 367 | dropit: |
| 368 | tcpstat.tcps_keepdrops++; |
| 369 | tp = tcp_drop(tp, ETIMEDOUT); |
| 370 | |
| 371 | #ifdef TCPDEBUG |
| 372 | if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) |
| 373 | tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); |
| 374 | #endif |
| 375 | return tp; |
| 376 | } |
| 377 | |
| 378 | void |
| 379 | tcp_timer_keep(void *xtp) |
| 380 | { |
| 381 | struct tcpcb *tp = xtp; |
| 382 | struct callout *co = &tp->tt_keep->tc_callout; |
| 383 | |
| 384 | crit_enter(); |
| 385 | if (callout_pending(co) || !callout_active(co)) { |
| 386 | crit_exit(); |
| 387 | return; |
| 388 | } |
| 389 | callout_deactivate(co); |
| 390 | tcp_send_timermsg(tp, TCP_TIMER_KEEP); |
| 391 | crit_exit(); |
| 392 | } |
| 393 | |
| 394 | /* Caller should be in critical section */ |
| 395 | static struct tcpcb * |
| 396 | tcp_timer_persist_handler(struct tcpcb *tp) |
| 397 | { |
| 398 | #ifdef TCPDEBUG |
| 399 | int ostate; |
| 400 | #endif |
| 401 | |
| 402 | #ifdef TCPDEBUG |
| 403 | ostate = tp->t_state; |
| 404 | #endif |
| 405 | /* |
| 406 | * Persistance timer into zero window. |
| 407 | * Force a byte to be output, if possible. |
| 408 | */ |
| 409 | tcpstat.tcps_persisttimeo++; |
| 410 | /* |
| 411 | * Hack: if the peer is dead/unreachable, we do not |
| 412 | * time out if the window is closed. After a full |
| 413 | * backoff, drop the connection if the idle time |
| 414 | * (no responses to probes) reaches the maximum |
| 415 | * backoff that we would use if retransmitting. |
| 416 | */ |
| 417 | if (tp->t_rxtshift == TCP_MAXRXTSHIFT && |
| 418 | ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle || |
| 419 | (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { |
| 420 | tcpstat.tcps_persistdrop++; |
| 421 | tp = tcp_drop(tp, ETIMEDOUT); |
| 422 | goto out; |
| 423 | } |
| 424 | tcp_setpersist(tp); |
| 425 | tp->t_flags |= TF_FORCE; |
| 426 | tcp_output(tp); |
| 427 | tp->t_flags &= ~TF_FORCE; |
| 428 | |
| 429 | out: |
| 430 | #ifdef TCPDEBUG |
| 431 | if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG) |
| 432 | tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); |
| 433 | #endif |
| 434 | return tp; |
| 435 | } |
| 436 | |
| 437 | void |
| 438 | tcp_timer_persist(void *xtp) |
| 439 | { |
| 440 | struct tcpcb *tp = xtp; |
| 441 | struct callout *co = &tp->tt_persist->tc_callout; |
| 442 | |
| 443 | crit_enter(); |
| 444 | if (callout_pending(co) || !callout_active(co)){ |
| 445 | crit_exit(); |
| 446 | return; |
| 447 | } |
| 448 | callout_deactivate(co); |
| 449 | tcp_send_timermsg(tp, TCP_TIMER_PERSIST); |
| 450 | crit_exit(); |
| 451 | } |
| 452 | |
| 453 | void |
| 454 | tcp_save_congestion_state(struct tcpcb *tp) |
| 455 | { |
| 456 | tp->snd_cwnd_prev = tp->snd_cwnd; |
| 457 | tp->snd_wacked_prev = tp->snd_wacked; |
| 458 | tp->snd_ssthresh_prev = tp->snd_ssthresh; |
| 459 | tp->snd_recover_prev = tp->snd_recover; |
| 460 | if (IN_FASTRECOVERY(tp)) |
| 461 | tp->t_flags |= TF_WASFRECOVERY; |
| 462 | else |
| 463 | tp->t_flags &= ~TF_WASFRECOVERY; |
| 464 | if (tp->t_flags & TF_RCVD_TSTMP) { |
| 465 | tp->t_rexmtTS = ticks; |
| 466 | tp->t_flags |= TF_FIRSTACCACK; |
| 467 | } |
| 468 | #ifdef later |
| 469 | tcp_sack_save_scoreboard(&tp->scb); |
| 470 | #endif |
| 471 | } |
| 472 | |
| 473 | void |
| 474 | tcp_revert_congestion_state(struct tcpcb *tp) |
| 475 | { |
| 476 | tp->snd_cwnd = tp->snd_cwnd_prev; |
| 477 | tp->snd_wacked = tp->snd_wacked_prev; |
| 478 | tp->snd_ssthresh = tp->snd_ssthresh_prev; |
| 479 | tp->snd_recover = tp->snd_recover_prev; |
| 480 | if (tp->t_flags & TF_WASFRECOVERY) |
| 481 | ENTER_FASTRECOVERY(tp); |
| 482 | if (tp->t_flags & TF_FASTREXMT) { |
| 483 | ++tcpstat.tcps_sndfastrexmitbad; |
| 484 | if (tp->t_flags & TF_EARLYREXMT) |
| 485 | ++tcpstat.tcps_sndearlyrexmitbad; |
| 486 | } else |
| 487 | ++tcpstat.tcps_sndrtobad; |
| 488 | tp->t_badrxtwin = 0; |
| 489 | tp->t_rxtshift = 0; |
| 490 | tp->snd_nxt = tp->snd_max; |
| 491 | #ifdef later |
| 492 | tcp_sack_revert_scoreboard(&tp->scb, tp->snd_una); |
| 493 | #endif |
| 494 | } |
| 495 | |
| 496 | /* Caller should be in critical section */ |
| 497 | static struct tcpcb * |
| 498 | tcp_timer_rexmt_handler(struct tcpcb *tp) |
| 499 | { |
| 500 | int rexmt; |
| 501 | #ifdef TCPDEBUG |
| 502 | int ostate; |
| 503 | #endif |
| 504 | |
| 505 | #ifdef TCPDEBUG |
| 506 | ostate = tp->t_state; |
| 507 | #endif |
| 508 | /* |
| 509 | * Retransmission timer went off. Message has not |
| 510 | * been acked within retransmit interval. Back off |
| 511 | * to a longer retransmit interval and retransmit one segment. |
| 512 | */ |
| 513 | if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { |
| 514 | tp->t_rxtshift = TCP_MAXRXTSHIFT; |
| 515 | tcpstat.tcps_timeoutdrop++; |
| 516 | tp = tcp_drop(tp, tp->t_softerror ? |
| 517 | tp->t_softerror : ETIMEDOUT); |
| 518 | goto out; |
| 519 | } |
| 520 | if (tp->t_rxtshift == 1) { |
| 521 | /* |
| 522 | * first retransmit; record ssthresh and cwnd so they can |
| 523 | * be recovered if this turns out to be a "bad" retransmit. |
| 524 | * A retransmit is considered "bad" if an ACK for this |
| 525 | * segment is received within RTT/2 interval; the assumption |
| 526 | * here is that the ACK was already in flight. See |
| 527 | * "On Estimating End-to-End Network Path Properties" by |
| 528 | * Allman and Paxson for more details. |
| 529 | */ |
| 530 | tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); |
| 531 | tcp_save_congestion_state(tp); |
| 532 | tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT); |
| 533 | } |
| 534 | if (tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED) { |
| 535 | /* |
| 536 | * Record the time that we spent in SYN or SYN|ACK |
| 537 | * retransmition. |
| 538 | * |
| 539 | * Needed by RFC3390 and RFC6298. |
| 540 | */ |
| 541 | tp->t_rxtsyn += tp->t_rxtcur; |
| 542 | } |
| 543 | /* Throw away SACK blocks on a RTO, as specified by RFC2018. */ |
| 544 | tcp_sack_cleanup(&tp->scb); |
| 545 | tcpstat.tcps_rexmttimeo++; |
| 546 | if (tp->t_state == TCPS_SYN_SENT) { |
| 547 | if (tcp_low_rtobase) { |
| 548 | rexmt = TCP_REXMTVAL(tp) * |
| 549 | tcp_syn_backoff_low[tp->t_rxtshift]; |
| 550 | } else { |
| 551 | rexmt = TCP_REXMTVAL(tp) * |
| 552 | tcp_syn_backoff[tp->t_rxtshift]; |
| 553 | } |
| 554 | } else { |
| 555 | rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; |
| 556 | } |
| 557 | TCPT_RANGESET(tp->t_rxtcur, rexmt, |
| 558 | tp->t_rttmin, TCPTV_REXMTMAX); |
| 559 | /* |
| 560 | * If losing, let the lower level know and try for |
| 561 | * a better route. Also, if we backed off this far, |
| 562 | * our srtt estimate is probably bogus. Clobber it |
| 563 | * so we'll take the next rtt measurement as our srtt; |
| 564 | * move the current srtt into rttvar to keep the current |
| 565 | * retransmit times until then. |
| 566 | */ |
| 567 | if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { |
| 568 | #ifdef INET6 |
| 569 | if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) |
| 570 | in6_losing(tp->t_inpcb); |
| 571 | else |
| 572 | #endif |
| 573 | in_losing(tp->t_inpcb); |
| 574 | tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); |
| 575 | tp->t_srtt = 0; |
| 576 | } |
| 577 | tp->snd_nxt = tp->snd_una; |
| 578 | tp->rexmt_high = tp->snd_una; |
| 579 | tp->snd_recover = tp->snd_max; |
| 580 | /* |
| 581 | * Force a segment to be sent. |
| 582 | */ |
| 583 | tp->t_flags |= TF_ACKNOW; |
| 584 | /* |
| 585 | * If timing a segment in this window, stop the timer. |
| 586 | */ |
| 587 | tp->t_rtttime = 0; |
| 588 | /* |
| 589 | * Close the congestion window down to one segment |
| 590 | * (we'll open it by one segment for each ack we get). |
| 591 | * Since we probably have a window's worth of unacked |
| 592 | * data accumulated, this "slow start" keeps us from |
| 593 | * dumping all that data as back-to-back packets (which |
| 594 | * might overwhelm an intermediate gateway). |
| 595 | * |
| 596 | * There are two phases to the opening: Initially we |
| 597 | * open by one mss on each ack. This makes the window |
| 598 | * size increase exponentially with time. If the |
| 599 | * window is larger than the path can handle, this |
| 600 | * exponential growth results in dropped packet(s) |
| 601 | * almost immediately. To get more time between |
| 602 | * drops but still "push" the network to take advantage |
| 603 | * of improving conditions, we switch from exponential |
| 604 | * to linear window opening at some threshhold size. |
| 605 | * For a threshhold, we use half the current window |
| 606 | * size, truncated to a multiple of the mss. |
| 607 | * |
| 608 | * (the minimum cwnd that will give us exponential |
| 609 | * growth is 2 mss. We don't allow the threshhold |
| 610 | * to go below this.) |
| 611 | */ |
| 612 | { |
| 613 | u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; |
| 614 | |
| 615 | if (win < 2) |
| 616 | win = 2; |
| 617 | tp->snd_cwnd = tp->t_maxseg; |
| 618 | tp->snd_wacked = 0; |
| 619 | tp->snd_ssthresh = win * tp->t_maxseg; |
| 620 | tp->t_dupacks = 0; |
| 621 | } |
| 622 | EXIT_FASTRECOVERY(tp); |
| 623 | tcp_output(tp); |
| 624 | |
| 625 | out: |
| 626 | #ifdef TCPDEBUG |
| 627 | if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) |
| 628 | tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); |
| 629 | #endif |
| 630 | return tp; |
| 631 | } |
| 632 | |
| 633 | void |
| 634 | tcp_timer_rexmt(void *xtp) |
| 635 | { |
| 636 | struct tcpcb *tp = xtp; |
| 637 | struct callout *co = &tp->tt_rexmt->tc_callout; |
| 638 | |
| 639 | crit_enter(); |
| 640 | if (callout_pending(co) || !callout_active(co)) { |
| 641 | crit_exit(); |
| 642 | return; |
| 643 | } |
| 644 | callout_deactivate(co); |
| 645 | tcp_send_timermsg(tp, TCP_TIMER_REXMT); |
| 646 | crit_exit(); |
| 647 | } |
| 648 | |
| 649 | static void |
| 650 | tcp_timer_handler(netmsg_t msg) |
| 651 | { |
| 652 | struct netmsg_tcp_timer *tmsg = (struct netmsg_tcp_timer *)msg; |
| 653 | const struct tcp_timer *tt; |
| 654 | struct tcpcb *tp; |
| 655 | |
| 656 | crit_enter(); |
| 657 | |
| 658 | KKASSERT(tmsg->tt_cpuid == mycpuid && tmsg->tt_tcb != NULL); |
| 659 | tp = tmsg->tt_tcb; |
| 660 | |
| 661 | /* Save pending tasks and reset the tasks in message */ |
| 662 | tmsg->tt_running_tasks = tmsg->tt_tasks; |
| 663 | tmsg->tt_prev_tasks = tmsg->tt_tasks; |
| 664 | tmsg->tt_tasks = 0; |
| 665 | |
| 666 | /* Reply ASAP */ |
| 667 | lwkt_replymsg(&tmsg->tt_msg.lmsg, 0); |
| 668 | |
| 669 | if (tmsg->tt_running_tasks == 0) { |
| 670 | /* |
| 671 | * All of the timers are cancelled when the message |
| 672 | * is pending; bail out. |
| 673 | */ |
| 674 | crit_exit(); |
| 675 | return; |
| 676 | } |
| 677 | |
| 678 | for (tt = tcp_timer_handlers; tt->tt_handler != NULL; ++tt) { |
| 679 | if ((tmsg->tt_running_tasks & tt->tt_task) == 0) |
| 680 | continue; |
| 681 | |
| 682 | tmsg->tt_running_tasks &= ~tt->tt_task; |
| 683 | tp = tt->tt_handler(tp); |
| 684 | if (tp == NULL) |
| 685 | break; |
| 686 | |
| 687 | if (tmsg->tt_running_tasks == 0) /* nothing left to do */ |
| 688 | break; |
| 689 | } |
| 690 | |
| 691 | crit_exit(); |
| 692 | } |
| 693 | |
| 694 | void |
| 695 | tcp_create_timermsg(struct tcpcb *tp, struct lwkt_port *msgport) |
| 696 | { |
| 697 | struct netmsg_tcp_timer *tmsg = tp->tt_msg; |
| 698 | |
| 699 | netmsg_init(&tmsg->tt_msg, NULL, &netisr_adone_rport, |
| 700 | MSGF_DROPABLE | MSGF_PRIORITY, tcp_timer_handler); |
| 701 | tmsg->tt_cpuid = mycpuid; |
| 702 | tmsg->tt_msgport = msgport; |
| 703 | tmsg->tt_tcb = tp; |
| 704 | tmsg->tt_tasks = 0; |
| 705 | } |
| 706 | |
| 707 | void |
| 708 | tcp_destroy_timermsg(struct tcpcb *tp) |
| 709 | { |
| 710 | struct netmsg_tcp_timer *tmsg = tp->tt_msg; |
| 711 | |
| 712 | if (tmsg == NULL || /* listen socket */ |
| 713 | tmsg->tt_tcb == NULL) /* only tcp_attach() is called */ |
| 714 | return; |
| 715 | |
| 716 | KKASSERT(tmsg->tt_cpuid == mycpuid); |
| 717 | crit_enter(); |
| 718 | if ((tmsg->tt_msg.lmsg.ms_flags & MSGF_DONE) == 0) { |
| 719 | /* |
| 720 | * This message is still pending to be processed; |
| 721 | * drop it. |
| 722 | */ |
| 723 | lwkt_dropmsg(&tmsg->tt_msg.lmsg); |
| 724 | } |
| 725 | crit_exit(); |
| 726 | } |
| 727 | |
| 728 | static __inline void |
| 729 | tcp_callout_init(struct tcp_callout *tc, uint32_t task) |
| 730 | { |
| 731 | callout_init_mp(&tc->tc_callout); |
| 732 | tc->tc_task = task; |
| 733 | } |
| 734 | |
| 735 | void |
| 736 | tcp_inittimers(struct tcpcb *tp) |
| 737 | { |
| 738 | tcp_callout_init(tp->tt_rexmt, TCP_TIMER_REXMT); |
| 739 | tcp_callout_init(tp->tt_persist, TCP_TIMER_PERSIST); |
| 740 | tcp_callout_init(tp->tt_keep, TCP_TIMER_KEEP); |
| 741 | tcp_callout_init(tp->tt_2msl, TCP_TIMER_2MSL); |
| 742 | tcp_callout_init(tp->tt_delack, TCP_TIMER_DELACK); |
| 743 | } |