tcp_sack.c revision 1.18
1/* $NetBSD: tcp_sack.c,v 1.18 2006/10/07 19:56:14 yamt Exp $ */ 2 3/* 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Kentaro A. Kurahone. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39/* 40 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 41 * The Regents of the University of California. All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 4. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95 68 * $FreeBSD: src/sys/netinet/tcp_sack.c,v 1.3.2.2 2004/12/25 23:02:57 rwatson Exp $ 69 */ 70 71/* 72 * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995 73 * 74 * NRL grants permission for redistribution and use in source and binary 75 * forms, with or without modification, of the software and documentation 76 * created at NRL provided that the following conditions are met: 77 * 78 * 1. Redistributions of source code must retain the above copyright 79 * notice, this list of conditions and the following disclaimer. 80 * 2. Redistributions in binary form must reproduce the above copyright 81 * notice, this list of conditions and the following disclaimer in the 82 * documentation and/or other materials provided with the distribution. 83 * 3. All advertising materials mentioning features or use of this software 84 * must display the following acknowledgements: 85 * This product includes software developed by the University of 86 * California, Berkeley and its contributors. 87 * This product includes software developed at the Information 88 * Technology Division, US Naval Research Laboratory. 89 * 4. Neither the name of the NRL nor the names of its contributors 90 * may be used to endorse or promote products derived from this software 91 * without specific prior written permission. 92 * 93 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 94 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 95 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 96 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 97 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 98 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 99 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 100 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 101 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 102 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 103 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 104 * 105 * The views and conclusions contained in the software and documentation 106 * are those of the authors and should not be interpreted as representing 107 * official policies, either expressed or implied, of the US Naval 108 * Research Laboratory (NRL). 109 */ 110 111#include <sys/cdefs.h> 112__KERNEL_RCSID(0, "$NetBSD: tcp_sack.c,v 1.18 2006/10/07 19:56:14 yamt Exp $"); 113 114#include "opt_inet.h" 115#include "opt_ipsec.h" 116#include "opt_inet_csum.h" 117#include "opt_tcp_debug.h" 118 119#include <sys/param.h> 120#include <sys/systm.h> 121#include <sys/malloc.h> 122#include <sys/mbuf.h> 123#include <sys/protosw.h> 124#include <sys/socket.h> 125#include <sys/socketvar.h> 126#include <sys/errno.h> 127#include <sys/syslog.h> 128#include <sys/pool.h> 129#include <sys/domain.h> 130#include <sys/kernel.h> 131 132#include <net/if.h> 133#include <net/route.h> 134#include <net/if_types.h> 135 136#include <netinet/in.h> 137#include <netinet/in_systm.h> 138#include <netinet/ip.h> 139#include <netinet/in_pcb.h> 140#include <netinet/in_var.h> 141#include <netinet/ip_var.h> 142 143#ifdef INET6 144#ifndef INET 145#include <netinet/in.h> 146#endif 147#include <netinet/ip6.h> 148#include <netinet6/ip6_var.h> 149#include <netinet6/in6_pcb.h> 150#include <netinet6/ip6_var.h> 151#include <netinet6/in6_var.h> 152#include <netinet/icmp6.h> 153#include <netinet6/nd6.h> 154#endif 155 156#ifndef INET6 157/* always need ip6.h for IP6_EXTHDR_GET */ 158#include <netinet/ip6.h> 159#endif 160 161#include <netinet/tcp.h> 162#include <netinet/tcp_fsm.h> 163#include <netinet/tcp_seq.h> 164#include <netinet/tcp_timer.h> 165#include <netinet/tcp_var.h> 166#include <netinet/tcpip.h> 167#include <netinet/tcp_debug.h> 168 169#include <machine/stdarg.h> 170 171/* SACK block pool. */ 172POOL_INIT(sackhole_pool, sizeof(struct sackhole), 0, 0, 0, "sackholepl", NULL); 173 174void 175tcp_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 176{ 177 if (TCP_SACK_ENABLED(tp)) { 178 tp->rcv_dsack_block.left = seq; 179 tp->rcv_dsack_block.right = seq + len; 180 tp->rcv_sack_flags |= TCPSACK_HAVED; 181 } 182} 183 184void 185tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 186{ 187 struct sackblk 188 t_sack_block[(MAX_TCPOPTLEN - 2) / (sizeof(u_int32_t) * 2)]; 189 struct sackblk *sack = NULL; 190 struct sackhole *cur = NULL; 191 struct sackhole *tmp = NULL; 192 u_int32_t *lp = (u_int32_t *) (cp + 2); 193 int i, j, num_sack_blks; 194 tcp_seq left, right, acked; 195 196 /* 197 * If we aren't processing SACK responses, this is not an ACK 198 * or the peer sends us a sack option with invalid length, don't 199 * update the scoreboard. 200 */ 201 if (!TCP_SACK_ENABLED(tp) || ((th->th_flags & TH_ACK) == 0) || 202 (optlen % 8 != 2 || optlen < 10)) { 203 return; 204 } 205 206 /* 207 * If we don't want any SACK holes to be allocated, just return. 208 */ 209 if (tcp_sack_globalmaxholes == 0 || tcp_sack_tp_maxholes == 0) { 210 return; 211 } 212 213 /* If the ACK is outside [snd_una, snd_max], ignore the SACK options. */ 214 if (SEQ_LT(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max)) 215 return; 216 217 /* 218 * Extract SACK blocks. 219 * 220 * Note that t_sack_block is sorted so that we only need to do 221 * one pass over the sequence number space. (SACK "fast-path") 222 */ 223 num_sack_blks = optlen / 8; 224 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 225 for (i = 0; i < num_sack_blks; i++, lp += 2) { 226 memcpy(&left, lp, sizeof(*lp)); 227 memcpy(&right, lp + 1, sizeof(*lp)); 228 left = ntohl(left); 229 right = ntohl(right); 230 231 if (SEQ_LEQ(right, acked) || SEQ_GT(right, tp->snd_max) || 232 SEQ_GEQ(left, right)) { 233 /* SACK entry that's old, or invalid. */ 234 i--; 235 num_sack_blks--; 236 continue; 237 } 238 239 /* Insertion sort. */ 240 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 241 j--) { 242 t_sack_block[j].left = t_sack_block[j - 1].left; 243 t_sack_block[j].right = t_sack_block[j - 1].right; 244 } 245 t_sack_block[j].left = left; 246 t_sack_block[j].right = right; 247 } 248 249 /* Update the scoreboard. */ 250 cur = TAILQ_FIRST(&tp->snd_holes); 251 for (i = 0; i < num_sack_blks; i++) { 252 sack = &t_sack_block[i]; 253 /* 254 * FACK TCP. Update snd_fack so we can enter Fast 255 * Recovery early. 256 */ 257 if (SEQ_GEQ(sack->right, tp->snd_fack)) 258 tp->snd_fack = sack->right; 259 260 if (TAILQ_EMPTY(&tp->snd_holes)) { 261 /* First hole. */ 262 if (tcp_sack_globalholes >= tcp_sack_globalmaxholes) { 263 return; 264 } 265 cur = (struct sackhole *) 266 pool_get(&sackhole_pool, PR_NOWAIT); 267 if (cur == NULL) { 268 /* ENOBUFS, bail out*/ 269 return; 270 } 271 cur->start = th->th_ack; 272 cur->end = sack->left; 273 cur->rxmit = cur->start; 274 tp->rcv_lastsack = sack->right; 275 tp->snd_numholes++; 276 tcp_sack_globalholes++; 277 TAILQ_INSERT_HEAD(&tp->snd_holes, cur, sackhole_q); 278 continue; /* With next sack block */ 279 } 280 281 /* Go through the list of holes. */ 282 while (cur) { 283 if (SEQ_LEQ(sack->right, cur->start)) 284 /* SACKs data before the current hole */ 285 break; /* No use going through more holes */ 286 287 if (SEQ_GEQ(sack->left, cur->end)) { 288 /* SACKs data beyond the current hole */ 289 cur = TAILQ_NEXT(cur, sackhole_q); 290 continue; 291 } 292 293 if (SEQ_LEQ(sack->left, cur->start)) { 294 /* Data acks at least the beginning of hole */ 295 if (SEQ_GEQ(sack->right, cur->end)) { 296 /* Acks entire hole, so delete hole */ 297 tmp = cur; 298 cur = TAILQ_NEXT(cur, sackhole_q); 299 tp->snd_numholes--; 300 tcp_sack_globalholes--; 301 TAILQ_REMOVE(&tp->snd_holes, tmp, 302 sackhole_q); 303 pool_put(&sackhole_pool, tmp); 304 break; 305 } 306 307 /* Otherwise, move start of hole forward */ 308 cur->start = sack->right; 309 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 310 break; 311 } 312 313 if (SEQ_GEQ(sack->right, cur->end)) { 314 /* Move end of hole backward. */ 315 cur->end = sack->left; 316 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 317 cur = TAILQ_NEXT(cur, sackhole_q); 318 break; 319 } 320 321 if (SEQ_LT(cur->start, sack->left) && 322 SEQ_GT(cur->end, sack->right)) { 323 /* 324 * ACKs some data in middle of a hole; need to 325 * split current hole 326 */ 327 if (tcp_sack_globalholes >= 328 tcp_sack_globalmaxholes || 329 tp->snd_numholes >= 330 tcp_sack_tp_maxholes) { 331 return; 332 } 333 tmp = (struct sackhole *) 334 pool_get(&sackhole_pool, PR_NOWAIT); 335 if (tmp == NULL) { 336 /* ENOBUFS, bail out. */ 337 return; 338 } 339 tmp->start = sack->right; 340 tmp->end = cur->end; 341 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 342 cur->end = sack->left; 343 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 344 tp->snd_numholes++; 345 tcp_sack_globalholes++; 346 TAILQ_INSERT_AFTER(&tp->snd_holes, cur, tmp, 347 sackhole_q); 348 cur = tmp; 349 break; 350 } 351 } 352 353 /* At this point, we have reached the tail of the list. */ 354 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 355 /* 356 * Need to append new hole at end. 357 */ 358 if (tcp_sack_globalholes >= 359 tcp_sack_globalmaxholes || 360 tp->snd_numholes >= 361 tcp_sack_tp_maxholes) { 362 return; 363 } 364 tmp = (struct sackhole *) 365 pool_get(&sackhole_pool, PR_NOWAIT); 366 if (tmp == NULL) 367 continue; /* ENOBUFS */ 368 tmp->start = tp->rcv_lastsack; 369 tmp->end = sack->left; 370 tmp->rxmit = tmp->start; 371 tp->snd_numholes++; 372 tcp_sack_globalholes++; 373 TAILQ_INSERT_TAIL(&tp->snd_holes, tmp, sackhole_q); 374 cur = tmp; 375 } 376 if (SEQ_LT(tp->rcv_lastsack, sack->right)) { 377 tp->rcv_lastsack = sack->right; 378 } 379 } 380} 381 382void 383tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th) 384{ 385 /* Max because this could be an older ack that just arrived. */ 386 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 387 th->th_ack : tp->snd_una; 388 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 389 struct sackhole *tmp; 390 391 while (cur) { 392 if (SEQ_LEQ(cur->end, lastack)) { 393 tmp = cur; 394 cur = TAILQ_NEXT(cur, sackhole_q); 395 tp->snd_numholes--; 396 tcp_sack_globalholes--; 397 TAILQ_REMOVE(&tp->snd_holes, tmp, sackhole_q); 398 pool_put(&sackhole_pool, tmp); 399 } else if (SEQ_LT(cur->start, lastack)) { 400 cur->start = lastack; 401 if (SEQ_LT(cur->rxmit, cur->start)) 402 cur->rxmit = cur->start; 403 break; 404 } else 405 break; 406 } 407} 408 409void 410tcp_free_sackholes(struct tcpcb *tp) 411{ 412 struct sackhole *sack; 413 414 /* Free up the SACK hole list. */ 415 while (!TAILQ_EMPTY(&tp->snd_holes)) { 416 sack = TAILQ_FIRST(&tp->snd_holes); 417 tcp_sack_globalholes--; 418 TAILQ_REMOVE(&tp->snd_holes, sack, sackhole_q); 419 pool_put(&sackhole_pool, sack); 420 } 421 422 tp->snd_numholes = 0; 423} 424 425/* 426 * Implements the SACK response to a new ack, checking for partial acks 427 * in fast recovery. 428 */ 429void 430tcp_sack_newack(struct tcpcb *tp, struct tcphdr *th) 431{ 432 if (tp->t_partialacks < 0) { 433 /* 434 * Not in fast recovery. Reset the duplicate ack 435 * counter. 436 */ 437 tp->t_dupacks = 0; 438 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 439 /* 440 * Partial ack handling within a sack recovery episode. 441 * Keeping this very simple for now. When a partial ack 442 * is received, force snd_cwnd to a value that will allow 443 * the sender to transmit no more than 2 segments. 444 * If necessary, a fancier scheme can be adopted at a 445 * later point, but for now, the goal is to prevent the 446 * sender from bursting a large amount of data in the midst 447 * of sack recovery. 448 */ 449 int num_segs = 1; 450 int sack_bytes_rxmt = 0; 451 452 tp->t_partialacks++; 453 TCP_TIMER_DISARM(tp, TCPT_REXMT); 454 tp->t_rtttime = 0; 455 456 /* 457 * send one or 2 segments based on how much new data was acked 458 */ 459 if (((th->th_ack - tp->snd_una) / tp->t_segsz) > 2) 460 num_segs = 2; 461 (void)tcp_sack_output(tp, &sack_bytes_rxmt); 462 tp->snd_cwnd = sack_bytes_rxmt + 463 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_segsz; 464 tp->t_flags |= TF_ACKNOW; 465 (void) tcp_output(tp); 466 } else { 467 /* 468 * Complete ack, inflate the congestion window to 469 * ssthresh and exit fast recovery. 470 * 471 * Window inflation should have left us with approx. 472 * snd_ssthresh outstanding data. But in case we 473 * would be inclined to send a burst, better to do 474 * it via the slow start mechanism. 475 */ 476 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 477 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 478 + tp->t_segsz; 479 else 480 tp->snd_cwnd = tp->snd_ssthresh; 481 tp->t_partialacks = -1; 482 tp->t_dupacks = 0; 483 if (SEQ_GT(th->th_ack, tp->snd_fack)) 484 tp->snd_fack = th->th_ack; 485 } 486} 487 488/* 489 * Returns pointer to a sackhole if there are any pending retransmissions; 490 * NULL otherwise. 491 */ 492struct sackhole * 493tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 494{ 495 struct sackhole *cur = NULL; 496 497 if (!TCP_SACK_ENABLED(tp)) 498 return (NULL); 499 500 *sack_bytes_rexmt = 0; 501 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 502 if (SEQ_LT(cur->rxmit, cur->end)) { 503 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 504 /* old SACK hole */ 505 continue; 506 } 507 *sack_bytes_rexmt += (cur->rxmit - cur->start); 508 break; 509 } 510 *sack_bytes_rexmt += (cur->rxmit - cur->start); 511 } 512 513 return (cur); 514} 515 516/* 517 * After a timeout, the SACK list may be rebuilt. This SACK information 518 * should be used to avoid retransmitting SACKed data. This function 519 * traverses the SACK list to see if snd_nxt should be moved forward. 520 */ 521void 522tcp_sack_adjust(struct tcpcb *tp) 523{ 524 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 525 struct sackhole *n = NULL; 526 527 if (TAILQ_EMPTY(&tp->snd_holes)) 528 return; /* No holes */ 529 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 530 return; /* We're already beyond any SACKed blocks */ 531 532 /* 533 * Two cases for which we want to advance snd_nxt: 534 * i) snd_nxt lies between end of one hole and beginning of another 535 * ii) snd_nxt lies between end of last hole and rcv_lastsack 536 */ 537 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 538 if (SEQ_LT(tp->snd_nxt, cur->end)) 539 return; 540 if (SEQ_GEQ(tp->snd_nxt, n->start)) 541 cur = n; 542 else { 543 tp->snd_nxt = n->start; 544 return; 545 } 546 } 547 if (SEQ_LT(tp->snd_nxt, cur->end)) 548 return; 549 tp->snd_nxt = tp->rcv_lastsack; 550 551 return; 552} 553 554int 555tcp_sack_numblks(const struct tcpcb *tp) 556{ 557 int numblks; 558 559 if (!TCP_SACK_ENABLED(tp)) { 560 return 0; 561 } 562 563 numblks = (((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) ? 1 : 0) + 564 tp->t_segqlen; 565 566 if (numblks == 0) { 567 return 0; 568 } 569 570 if (numblks > TCP_SACK_MAX) { 571 numblks = TCP_SACK_MAX; 572 } 573 574 return numblks; 575} 576