tcp_congctl.c revision 1.9
1 /* $NetBSD: tcp_congctl.c,v 1.9 2006/10/19 11:42:02 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 9 * Facility, NASA Ames Research Center. 10 * This code is derived from software contributed to The NetBSD Foundation 11 * by Charles M. Hannum. 12 * This code is derived from software contributed to The NetBSD Foundation 13 * by Rui Paulo. 14 * 15 * Redistribution and use in source and binary forms, with or without 16 * modification, are permitted provided that the following conditions 17 * are met: 18 * 1. Redistributions of source code must retain the above copyright 19 * notice, this list of conditions and the following disclaimer. 20 * 2. Redistributions in binary form must reproduce the above copyright 21 * notice, this list of conditions and the following disclaimer in the 22 * documentation and/or other materials provided with the distribution. 23 * 3. All advertising materials mentioning features or use of this software 24 * must display the following acknowledgement: 25 * This product includes software developed by the NetBSD 26 * Foundation, Inc. and its contributors. 27 * 4. Neither the name of The NetBSD Foundation nor the names of its 28 * contributors may be used to endorse or promote products derived 29 * from this software without specific prior written permission. 30 * 31 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 32 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 33 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 34 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 35 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 36 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 37 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 38 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 39 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 40 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 41 * POSSIBILITY OF SUCH DAMAGE. 42 */ 43 44 /* 45 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 46 * All rights reserved. 47 * 48 * Redistribution and use in source and binary forms, with or without 49 * modification, are permitted provided that the following conditions 50 * are met: 51 * 1. Redistributions of source code must retain the above copyright 52 * notice, this list of conditions and the following disclaimer. 53 * 2. Redistributions in binary form must reproduce the above copyright 54 * notice, this list of conditions and the following disclaimer in the 55 * documentation and/or other materials provided with the distribution. 56 * 3. Neither the name of the project nor the names of its contributors 57 * may be used to endorse or promote products derived from this software 58 * without specific prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 61 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 62 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 63 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 64 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 65 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 66 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 70 * SUCH DAMAGE. 71 */ 72 73 /* 74 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 75 * 76 * NRL grants permission for redistribution and use in source and binary 77 * forms, with or without modification, of the software and documentation 78 * created at NRL provided that the following conditions are met: 79 * 80 * 1. Redistributions of source code must retain the above copyright 81 * notice, this list of conditions and the following disclaimer. 82 * 2. Redistributions in binary form must reproduce the above copyright 83 * notice, this list of conditions and the following disclaimer in the 84 * documentation and/or other materials provided with the distribution. 85 * 3. All advertising materials mentioning features or use of this software 86 * must display the following acknowledgements: 87 * This product includes software developed by the University of 88 * California, Berkeley and its contributors. 89 * This product includes software developed at the Information 90 * Technology Division, US Naval Research Laboratory. 91 * 4. Neither the name of the NRL nor the names of its contributors 92 * may be used to endorse or promote products derived from this software 93 * without specific prior written permission. 94 * 95 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 96 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 97 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 98 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 99 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 100 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 101 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 102 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 103 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 104 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 105 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 106 * 107 * The views and conclusions contained in the software and documentation 108 * are those of the authors and should not be interpreted as representing 109 * official policies, either expressed or implied, of the US Naval 110 * Research Laboratory (NRL). 111 */ 112 113 /* 114 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 115 * The Regents of the University of California. All rights reserved. 116 * 117 * Redistribution and use in source and binary forms, with or without 118 * modification, are permitted provided that the following conditions 119 * are met: 120 * 1. Redistributions of source code must retain the above copyright 121 * notice, this list of conditions and the following disclaimer. 122 * 2. Redistributions in binary form must reproduce the above copyright 123 * notice, this list of conditions and the following disclaimer in the 124 * documentation and/or other materials provided with the distribution. 125 * 3. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 #include <sys/cdefs.h> 145 __KERNEL_RCSID(0, "$NetBSD: tcp_congctl.c,v 1.9 2006/10/19 11:42:02 yamt Exp $"); 146 147 #include "opt_inet.h" 148 #include "opt_tcp_debug.h" 149 #include "opt_tcp_congctl.h" 150 151 #include <sys/param.h> 152 #include <sys/systm.h> 153 #include <sys/malloc.h> 154 #include <sys/mbuf.h> 155 #include <sys/protosw.h> 156 #include <sys/socket.h> 157 #include <sys/socketvar.h> 158 #include <sys/errno.h> 159 #include <sys/syslog.h> 160 #include <sys/pool.h> 161 #include <sys/domain.h> 162 #include <sys/kernel.h> 163 #include <sys/lock.h> 164 165 #include <net/if.h> 166 #include <net/route.h> 167 168 #include <netinet/in.h> 169 #include <netinet/in_systm.h> 170 #include <netinet/ip.h> 171 #include <netinet/in_pcb.h> 172 #include <netinet/in_var.h> 173 #include <netinet/ip_var.h> 174 175 #ifdef INET6 176 #ifndef INET 177 #include <netinet/in.h> 178 #endif 179 #include <netinet/ip6.h> 180 #include <netinet6/ip6_var.h> 181 #include <netinet6/in6_pcb.h> 182 #include <netinet6/ip6_var.h> 183 #include <netinet6/in6_var.h> 184 #include <netinet/icmp6.h> 185 #include <netinet6/nd6.h> 186 #endif 187 188 #include <netinet/tcp.h> 189 #include <netinet/tcp_fsm.h> 190 #include <netinet/tcp_seq.h> 191 #include <netinet/tcp_timer.h> 192 #include <netinet/tcp_var.h> 193 #include <netinet/tcpip.h> 194 #include <netinet/tcp_congctl.h> 195 #ifdef TCP_DEBUG 196 #include <netinet/tcp_debug.h> 197 #endif 198 199 /* 200 * TODO: 201 * consider separating the actual implementations in another file. 202 */ 203 204 static int tcp_reno_fast_retransmit(struct tcpcb *, struct tcphdr *); 205 static void tcp_reno_slow_retransmit(struct tcpcb *); 206 static void tcp_reno_fast_retransmit_newack(struct tcpcb *, struct tcphdr *); 207 static void tcp_reno_newack(struct tcpcb *, struct tcphdr *); 208 static void tcp_reno_congestion_exp(struct tcpcb *tp); 209 210 static int tcp_newreno_fast_retransmit(struct tcpcb *, struct tcphdr *); 211 static void tcp_newreno_fast_retransmit_newack(struct tcpcb *, 212 struct tcphdr *); 213 static void tcp_newreno_newack(struct tcpcb *, struct tcphdr *); 214 215 216 static void tcp_congctl_fillnames(void); 217 218 extern int tcprexmtthresh; 219 220 MALLOC_DEFINE(M_TCPCONGCTL, "tcpcongctl", "TCP congestion control structures"); 221 222 /* 223 * Used to list the available congestion control algorithms. 224 */ 225 struct tcp_congctlent { 226 TAILQ_ENTRY(tcp_congctlent) congctl_ent; 227 char congctl_name[TCPCC_MAXLEN]; 228 struct tcp_congctl *congctl_ctl; 229 }; 230 TAILQ_HEAD(, tcp_congctlent) tcp_congctlhd; 231 232 struct simplelock tcp_congctl_slock; 233 234 void 235 tcp_congctl_init(void) 236 { 237 int r; 238 239 TAILQ_INIT(&tcp_congctlhd); 240 simple_lock_init(&tcp_congctl_slock); 241 242 /* Base algorithms. */ 243 r = tcp_congctl_register("reno", &tcp_reno_ctl); 244 KASSERT(r == 0); 245 r = tcp_congctl_register("newreno", &tcp_newreno_ctl); 246 KASSERT(r == 0); 247 248 /* NewReno is the default. */ 249 #ifndef TCP_CONGCTL_DEFAULT 250 #define TCP_CONGCTL_DEFAULT "newreno" 251 #endif 252 253 r = tcp_congctl_select(NULL, TCP_CONGCTL_DEFAULT); 254 KASSERT(r == 0); 255 } 256 257 /* 258 * Register a congestion algorithm and select it if we have none. 259 */ 260 int 261 tcp_congctl_register(const char *name, struct tcp_congctl *tcc) 262 { 263 struct tcp_congctlent *ntcc, *tccp; 264 265 TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) 266 if (!strcmp(name, tccp->congctl_name)) { 267 /* name already registered */ 268 return EEXIST; 269 } 270 271 ntcc = malloc(sizeof(*ntcc), M_TCPCONGCTL, M_WAITOK); 272 273 strlcpy(ntcc->congctl_name, name, sizeof(ntcc->congctl_name) - 1); 274 ntcc->congctl_ctl = tcc; 275 276 TAILQ_INSERT_TAIL(&tcp_congctlhd, ntcc, congctl_ent); 277 tcp_congctl_fillnames(); 278 279 if (TAILQ_FIRST(&tcp_congctlhd) == ntcc) 280 tcp_congctl_select(NULL, name); 281 282 return 0; 283 } 284 285 int 286 tcp_congctl_unregister(const char *name) 287 { 288 struct tcp_congctlent *tccp, *rtccp; 289 unsigned int size; 290 291 rtccp = NULL; 292 size = 0; 293 TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) { 294 if (!strcmp(name, tccp->congctl_name)) 295 rtccp = tccp; 296 size++; 297 } 298 299 if (!rtccp) 300 return ENOENT; 301 302 if (size <= 1 || tcp_congctl_global == rtccp->congctl_ctl || 303 rtccp->congctl_ctl->refcnt) 304 return EBUSY; 305 306 TAILQ_REMOVE(&tcp_congctlhd, rtccp, congctl_ent); 307 free(rtccp, M_TCPCONGCTL); 308 tcp_congctl_fillnames(); 309 310 return 0; 311 } 312 313 /* 314 * Select a congestion algorithm by name. 315 */ 316 int 317 tcp_congctl_select(struct tcpcb *tp, const char *name) 318 { 319 struct tcp_congctlent *tccp; 320 321 KASSERT(name); 322 323 TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) 324 if (!strcmp(name, tccp->congctl_name)) { 325 if (tp) { 326 simple_lock(&tcp_congctl_slock); 327 tp->t_congctl->refcnt--; 328 tp->t_congctl = tccp->congctl_ctl; 329 tp->t_congctl->refcnt++; 330 simple_unlock(&tcp_congctl_slock); 331 } else { 332 tcp_congctl_global = tccp->congctl_ctl; 333 strlcpy(tcp_congctl_global_name, 334 tccp->congctl_name, 335 sizeof(tcp_congctl_global_name) - 1); 336 } 337 return 0; 338 } 339 340 return EINVAL; 341 } 342 343 /* 344 * Returns the name of a congestion algorithm. 345 */ 346 const char * 347 tcp_congctl_bystruct(const struct tcp_congctl *tcc) 348 { 349 struct tcp_congctlent *tccp; 350 351 KASSERT(tcc); 352 353 TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) 354 if (tccp->congctl_ctl == tcc) 355 return tccp->congctl_name; 356 357 return NULL; 358 } 359 360 static void 361 tcp_congctl_fillnames(void) 362 { 363 struct tcp_congctlent *tccp; 364 const char *delim = " "; 365 366 tcp_congctl_avail[0] = '\0'; 367 TAILQ_FOREACH(tccp, &tcp_congctlhd, congctl_ent) { 368 strlcat(tcp_congctl_avail, tccp->congctl_name, 369 sizeof(tcp_congctl_avail) - 1); 370 if (TAILQ_NEXT(tccp, congctl_ent)) 371 strlcat(tcp_congctl_avail, delim, 372 sizeof(tcp_congctl_avail) - 1); 373 } 374 375 } 376 377 /* ------------------------------------------------------------------------ */ 378 379 /* 380 * TCP/Reno congestion control. 381 */ 382 static void 383 tcp_reno_congestion_exp(struct tcpcb *tp) 384 { 385 u_int win; 386 387 /* 388 * Halve the congestion window and reduce the 389 * slow start threshold. 390 */ 391 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz; 392 if (win < 2) 393 win = 2; 394 395 tp->snd_ssthresh = win * tp->t_segsz; 396 tp->snd_recover = tp->snd_max; 397 tp->snd_cwnd = tp->snd_ssthresh; 398 399 /* 400 * When using TCP ECN, notify the peer that 401 * we reduced the cwnd. 402 */ 403 if (TCP_ECN_ALLOWED(tp)) 404 tp->t_flags |= TF_ECN_SND_CWR; 405 } 406 407 408 409 static int 410 tcp_reno_fast_retransmit(struct tcpcb *tp, struct tcphdr *th) 411 { 412 /* 413 * We know we're losing at the current 414 * window size so do congestion avoidance 415 * (set ssthresh to half the current window 416 * and pull our congestion window back to 417 * the new ssthresh). 418 * 419 * Dup acks mean that packets have left the 420 * network (they're now cached at the receiver) 421 * so bump cwnd by the amount in the receiver 422 * to keep a constant cwnd packets in the 423 * network. 424 * 425 * If we are using TCP/SACK, then enter 426 * Fast Recovery if the receiver SACKs 427 * data that is tcprexmtthresh * MSS 428 * bytes past the last ACKed segment, 429 * irrespective of the number of DupAcks. 430 */ 431 432 tcp_seq onxt; 433 434 onxt = tp->snd_nxt; 435 tcp_reno_congestion_exp(tp); 436 tp->t_partialacks = 0; 437 TCP_TIMER_DISARM(tp, TCPT_REXMT); 438 tp->t_rtttime = 0; 439 if (TCP_SACK_ENABLED(tp)) { 440 tp->t_dupacks = tcprexmtthresh; 441 tp->sack_newdata = tp->snd_nxt; 442 tp->snd_cwnd = tp->t_segsz; 443 (void) tcp_output(tp); 444 return 0; 445 } 446 tp->snd_nxt = th->th_ack; 447 tp->snd_cwnd = tp->t_segsz; 448 (void) tcp_output(tp); 449 tp->snd_cwnd = tp->snd_ssthresh + tp->t_segsz * tp->t_dupacks; 450 if (SEQ_GT(onxt, tp->snd_nxt)) 451 tp->snd_nxt = onxt; 452 453 return 0; 454 } 455 456 static void 457 tcp_reno_slow_retransmit(struct tcpcb *tp) 458 { 459 u_int win; 460 461 /* 462 * Close the congestion window down to one segment 463 * (we'll open it by one segment for each ack we get). 464 * Since we probably have a window's worth of unacked 465 * data accumulated, this "slow start" keeps us from 466 * dumping all that data as back-to-back packets (which 467 * might overwhelm an intermediate gateway). 468 * 469 * There are two phases to the opening: Initially we 470 * open by one mss on each ack. This makes the window 471 * size increase exponentially with time. If the 472 * window is larger than the path can handle, this 473 * exponential growth results in dropped packet(s) 474 * almost immediately. To get more time between 475 * drops but still "push" the network to take advantage 476 * of improving conditions, we switch from exponential 477 * to linear window opening at some threshhold size. 478 * For a threshhold, we use half the current window 479 * size, truncated to a multiple of the mss. 480 * 481 * (the minimum cwnd that will give us exponential 482 * growth is 2 mss. We don't allow the threshhold 483 * to go below this.) 484 */ 485 486 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz; 487 if (win < 2) 488 win = 2; 489 /* Loss Window MUST be one segment. */ 490 tp->snd_cwnd = tp->t_segsz; 491 tp->snd_ssthresh = win * tp->t_segsz; 492 tp->t_partialacks = -1; 493 tp->t_dupacks = 0; 494 tp->t_bytes_acked = 0; 495 } 496 497 static void 498 tcp_reno_fast_retransmit_newack(struct tcpcb *tp, struct tcphdr *th __unused) 499 { 500 if (tp->t_partialacks < 0) { 501 /* 502 * We were not in fast recovery. Reset the duplicate ack 503 * counter. 504 */ 505 tp->t_dupacks = 0; 506 } else { 507 /* 508 * Clamp the congestion window to the crossover point and 509 * exit fast recovery. 510 */ 511 if (tp->snd_cwnd > tp->snd_ssthresh) 512 tp->snd_cwnd = tp->snd_ssthresh; 513 tp->t_partialacks = -1; 514 tp->t_dupacks = 0; 515 tp->t_bytes_acked = 0; 516 } 517 } 518 519 static void 520 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th __unused) 521 { 522 /* 523 * When new data is acked, open the congestion window. 524 */ 525 526 u_int cw = tp->snd_cwnd; 527 u_int incr = tp->t_segsz; 528 529 if (tcp_do_abc) { 530 531 /* 532 * RFC 3465 Appropriate Byte Counting (ABC) 533 */ 534 535 int acked = th->th_ack - tp->snd_una; 536 537 if (cw >= tp->snd_ssthresh) { 538 tp->t_bytes_acked += acked; 539 if (tp->t_bytes_acked >= cw) { 540 /* Time to increase the window. */ 541 tp->t_bytes_acked -= cw; 542 } else { 543 /* No need to increase yet. */ 544 incr = 0; 545 } 546 } else { 547 /* 548 * use 2*SMSS or 1*SMSS for the "L" param, 549 * depending on sysctl setting. 550 * 551 * (See RFC 3465 2.3 Choosing the Limit) 552 */ 553 u_int abc_lim; 554 555 abc_lim = (tcp_abc_aggressive == 0 || 556 tp->snd_nxt != tp->snd_max) ? incr : incr * 2; 557 incr = min(acked, abc_lim); 558 } 559 } else { 560 561 /* 562 * If the window gives us less than ssthresh packets 563 * in flight, open exponentially (segsz per packet). 564 * Otherwise open linearly: segsz per window 565 * (segsz^2 / cwnd per packet). 566 */ 567 568 if (cw >= tp->snd_ssthresh) { 569 incr = incr * incr / cw; 570 } 571 } 572 573 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale); 574 } 575 576 struct tcp_congctl tcp_reno_ctl = { 577 .fast_retransmit = tcp_reno_fast_retransmit, 578 .slow_retransmit = tcp_reno_slow_retransmit, 579 .fast_retransmit_newack = tcp_reno_fast_retransmit_newack, 580 .newack = tcp_reno_newack, 581 .cong_exp = tcp_reno_congestion_exp, 582 }; 583 584 /* 585 * TCP/NewReno Congestion control. 586 */ 587 static int 588 tcp_newreno_fast_retransmit(struct tcpcb *tp, struct tcphdr *th) 589 { 590 if (SEQ_LT(th->th_ack, tp->snd_high)) { 591 /* 592 * False fast retransmit after timeout. 593 * Do not enter fast recovery 594 */ 595 tp->t_dupacks = 0; 596 return 1; 597 } else { 598 /* 599 * Fast retransmit is same as reno. 600 */ 601 return tcp_reno_fast_retransmit(tp, th); 602 } 603 604 return 0; 605 } 606 607 /* 608 * Implement the NewReno response to a new ack, checking for partial acks in 609 * fast recovery. 610 */ 611 static void 612 tcp_newreno_fast_retransmit_newack(struct tcpcb *tp, struct tcphdr *th) 613 { 614 if (tp->t_partialacks < 0) { 615 /* 616 * We were not in fast recovery. Reset the duplicate ack 617 * counter. 618 */ 619 tp->t_dupacks = 0; 620 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 621 /* 622 * This is a partial ack. Retransmit the first unacknowledged 623 * segment and deflate the congestion window by the amount of 624 * acknowledged data. Do not exit fast recovery. 625 */ 626 tcp_seq onxt = tp->snd_nxt; 627 u_long ocwnd = tp->snd_cwnd; 628 629 /* 630 * snd_una has not yet been updated and the socket's send 631 * buffer has not yet drained off the ACK'd data, so we 632 * have to leave snd_una as it was to get the correct data 633 * offset in tcp_output(). 634 */ 635 if (++tp->t_partialacks == 1) 636 TCP_TIMER_DISARM(tp, TCPT_REXMT); 637 tp->t_rtttime = 0; 638 tp->snd_nxt = th->th_ack; 639 /* 640 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 641 * is not yet updated when we're called. 642 */ 643 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 644 (void) tcp_output(tp); 645 tp->snd_cwnd = ocwnd; 646 if (SEQ_GT(onxt, tp->snd_nxt)) 647 tp->snd_nxt = onxt; 648 /* 649 * Partial window deflation. Relies on fact that tp->snd_una 650 * not updated yet. 651 */ 652 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 653 } else { 654 /* 655 * Complete ack. Inflate the congestion window to ssthresh 656 * and exit fast recovery. 657 * 658 * Window inflation should have left us with approx. 659 * snd_ssthresh outstanding data. But in case we 660 * would be inclined to send a burst, better to do 661 * it via the slow start mechanism. 662 */ 663 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 664 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 665 + tp->t_segsz; 666 else 667 tp->snd_cwnd = tp->snd_ssthresh; 668 tp->t_partialacks = -1; 669 tp->t_dupacks = 0; 670 tp->t_bytes_acked = 0; 671 } 672 } 673 674 static void 675 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 676 { 677 /* 678 * If we are still in fast recovery (meaning we are using 679 * NewReno and we have only received partial acks), do not 680 * inflate the window yet. 681 */ 682 if (tp->t_partialacks < 0) 683 tcp_reno_newack(tp, th); 684 } 685 686 687 struct tcp_congctl tcp_newreno_ctl = { 688 .fast_retransmit = tcp_newreno_fast_retransmit, 689 .slow_retransmit = tcp_reno_slow_retransmit, 690 .fast_retransmit_newack = tcp_newreno_fast_retransmit_newack, 691 .newack = tcp_newreno_newack, 692 .cong_exp = tcp_reno_congestion_exp, 693 }; 694 695 696
Indexes created Thu Sep 25 08:09:54 GMT 2025