tcp_sack.c revision 1.3
1 /* $NetBSD: tcp_sack.c,v 1.3 2005/03/06 23:05:56 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 1.3 2005/03/06 23:05:56 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 #define SEQ_MIN(a, b) ((SEQ_LT(a, b)) ? (a) : (b)) 172 #define SEQ_MAX(a, b) ((SEQ_GT(a, b)) ? (a) : (b)) 173 174 /* SACK block pool. */ 175 POOL_INIT(sackhole_pool, sizeof(struct sackhole), 0, 0, 0, "sackholepl", NULL); 176 177 void 178 tcp_update_sack_list(struct tcpcb *tp) 179 { 180 int i = 0; 181 struct ipqent *tiqe = NULL; 182 183 if (!TCP_SACK_ENABLED(tp) || (tp->t_flags & TF_SIGNATURE)) { 184 /* Can't SACK this connection. */ 185 return; 186 } 187 188 /* 189 * If possible, tack on the D-SACK block. (RFC2883) 190 */ 191 if (tp->rcv_sack_flags & TCPSACK_HAVED) { 192 tp->rcv_sack_block[0].left = tp->rcv_dsack_block.left; 193 tp->rcv_sack_block[0].right = tp->rcv_dsack_block.right; 194 tp->rcv_sack_flags &= ~TCPSACK_HAVED; 195 i++; 196 } 197 198 /* 199 * Build up a list of holes in the TCP space. Note that 200 * the first SACK block is always the most recent segment 201 * received. 202 */ 203 TAILQ_FOREACH(tiqe, &tp->timeq, ipqe_timeq) { 204 tp->rcv_sack_block[i].left = tiqe->ipqe_seq; 205 tp->rcv_sack_block[i].right = tiqe->ipqe_seq + tiqe->ipqe_len; 206 i++; 207 if (i >= TCP_SACK_MAX) { 208 break; 209 } 210 } 211 212 /* If we can SACK, do so. */ 213 tp->rcv_sack_num = i; 214 } 215 216 void 217 tcp_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 218 { 219 if (TCP_SACK_ENABLED(tp)) { 220 tp->rcv_dsack_block.left = seq; 221 tp->rcv_dsack_block.right = seq + len; 222 tp->rcv_sack_flags |= TCPSACK_HAVED; 223 } 224 } 225 226 void 227 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 228 { 229 struct sackblk t_sack_block[TCP_SACK_MAX]; 230 struct sackblk *sack = NULL; 231 struct sackhole *cur = NULL; 232 struct sackhole *tmp = NULL; 233 u_int32_t *lp = (u_int32_t *) (cp + 2); 234 int i, j, num_sack_blks; 235 tcp_seq left, right, acked; 236 237 /* 238 * If we aren't processing SACK responses, or the peer 239 * sends us a sack option with invalid length, don't 240 * update the scoreboard. 241 */ 242 if (!TCP_SACK_ENABLED(tp) || 243 (optlen % 8 != 2 || optlen < 10)) { 244 return; 245 } 246 247 /* 248 * Extract SACK blocks. 249 * 250 * Note that t_sack_block is sorted so that we only need to do 251 * one pass over the sequence number space. (SACK "fast-path") 252 */ 253 num_sack_blks = optlen / 8; 254 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 255 for (i = 0; i < num_sack_blks; i++, lp += 2) { 256 memcpy(&left, lp, sizeof(*lp)); 257 memcpy(&right, lp + 1, sizeof(*lp)); 258 left = ntohl(left); 259 right = ntohl(right); 260 261 if ((SEQ_LEQ(right, acked)) || 262 SEQ_GEQ(left, tp->snd_max) || 263 SEQ_GEQ(left, right)) { 264 /* SACK entry that's old, or invalid. */ 265 i--; 266 num_sack_blks--; 267 continue; 268 } 269 270 /* Insertion sort. */ 271 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 272 j--) { 273 t_sack_block[j].left = t_sack_block[j - 1].left; 274 t_sack_block[j].right = t_sack_block[j - 1].right; 275 } 276 t_sack_block[j].left = left; 277 t_sack_block[j].right = right; 278 } 279 280 /* Update the scoreboard. */ 281 cur = TAILQ_FIRST(&tp->snd_holes); 282 for (i = 0; i < num_sack_blks; i++) { 283 sack = &t_sack_block[i]; 284 /* 285 * FACK TCP. Update snd_fack so we can enter Fast 286 * Recovery early. 287 */ 288 if (SEQ_GEQ(sack->right, tp->snd_fack)) 289 tp->snd_fack = sack->right; 290 291 if (TAILQ_EMPTY(&tp->snd_holes)) { 292 /* First hole. */ 293 cur = (struct sackhole *) 294 pool_get(&sackhole_pool, PR_NOWAIT); 295 if (cur == NULL) { 296 /* ENOBUFS, bail out*/ 297 return; 298 } 299 cur->start = th->th_ack; 300 cur->end = sack->left; 301 cur->rxmit = cur->start; 302 tp->rcv_lastsack = sack->right; 303 TAILQ_INSERT_HEAD(&tp->snd_holes, cur, sackhole_q); 304 continue; /* With next sack block */ 305 } 306 307 /* Go through the list of holes. */ 308 while (cur) { 309 if (SEQ_LEQ(sack->left, cur->start)) 310 /* SACKs data before the current hole */ 311 break; /* No use going through more holes */ 312 313 if (SEQ_GEQ(sack->left, cur->end)) { 314 /* SACKs data beyond the current hole */ 315 cur = TAILQ_NEXT(cur, sackhole_q); 316 continue; 317 } 318 319 if (SEQ_LEQ(sack->left, cur->start)) { 320 /* Data acks at least the beginning of hole */ 321 if (SEQ_GEQ(sack->right, cur->end)) { 322 /* Acks entire hole, so delete hole */ 323 tmp = cur; 324 cur = TAILQ_NEXT(cur, sackhole_q); 325 TAILQ_REMOVE(&tp->snd_holes, tmp, 326 sackhole_q); 327 pool_put(&sackhole_pool, tmp); 328 break; 329 } 330 331 /* Otherwise, move start of hole forward */ 332 cur->start = sack->right; 333 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 334 cur = TAILQ_NEXT(cur, sackhole_q); 335 break; 336 } 337 338 if (SEQ_GEQ(sack->right, cur->end)) { 339 /* Move end of hole backward. */ 340 cur->end = sack->left; 341 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 342 cur = TAILQ_NEXT(cur, sackhole_q); 343 break; 344 } 345 346 if (SEQ_LT(cur->start, sack->left) && 347 SEQ_GT(cur->end, sack->right)) { 348 /* 349 * ACKs some data in middle of a hole; need to 350 * split current hole 351 */ 352 tmp = (struct sackhole *) 353 pool_get(&sackhole_pool, PR_NOWAIT); 354 if (tmp == NULL) { 355 /* ENOBUFS, bail out. */ 356 return; 357 } 358 tmp->start = sack->right; 359 tmp->end = cur->end; 360 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 361 cur->end = sack->left; 362 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 363 TAILQ_INSERT_AFTER(&tp->snd_holes, cur, tmp, 364 sackhole_q); 365 cur = TAILQ_NEXT(tmp, sackhole_q); 366 break; 367 } 368 } 369 370 /* At this point, we have reached the tail of the list. */ 371 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 372 /* 373 * Need to append new hole at end. 374 */ 375 tmp = (struct sackhole *) 376 pool_get(&sackhole_pool, PR_NOWAIT); 377 if (tmp == NULL) 378 continue; /* ENOBUFS */ 379 tmp->start = tp->rcv_lastsack; 380 tmp->end = sack->left; 381 tmp->rxmit = tmp->start; 382 tp->rcv_lastsack = sack->right; 383 TAILQ_INSERT_TAIL(&tp->snd_holes, tmp, sackhole_q); 384 cur = tmp; 385 } 386 } 387 } 388 389 void 390 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th) 391 { 392 /* Max because this could be an older ack that just arrived. */ 393 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 394 th->th_ack : tp->snd_una; 395 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 396 struct sackhole *tmp; 397 398 while (cur) { 399 if (SEQ_LEQ(cur->end, lastack)) { 400 tmp = cur; 401 cur = TAILQ_NEXT(cur, sackhole_q); 402 TAILQ_REMOVE(&tp->snd_holes, tmp, sackhole_q); 403 pool_put(&sackhole_pool, tmp); 404 } else if (SEQ_LT(cur->start, lastack)) { 405 cur->start = lastack; 406 if (SEQ_LT(cur->rxmit, cur->start)) 407 cur->rxmit = cur->start; 408 break; 409 } else 410 break; 411 412 } 413 } 414 415 void 416 tcp_free_sackholes(struct tcpcb *tp) 417 { 418 struct sackhole *sack; 419 420 /* Free up the SACK hole list. */ 421 while (!TAILQ_EMPTY(&tp->snd_holes)) { 422 sack = TAILQ_FIRST(&tp->snd_holes); 423 TAILQ_REMOVE(&tp->snd_holes, sack, sackhole_q); 424 pool_put(&sackhole_pool, sack); 425 } 426 } 427 428 /* 429 * Implements the SACK response to a new ack, checking for partial acks 430 * in fast recovery. 431 */ 432 void 433 tcp_sack_newack(struct tcpcb *tp, struct tcphdr *th) 434 { 435 if (tp->t_partialacks < 0) { 436 /* 437 * Not in fast recovery. Reset the duplicate ack 438 * counter. 439 */ 440 tp->t_dupacks = 0; 441 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 442 /* 443 * Partial ack handling within a sack recovery episode. 444 * Keeping this very simple for now. When a partial ack 445 * is received, force snd_cwnd to a value that will allow 446 * the sender to transmit no more than 2 segments. 447 * If necessary, a fancier scheme can be adopted at a 448 * later point, but for now, the goal is to prevent the 449 * sender from bursting a large amount of data in the midst 450 * of sack recovery. 451 */ 452 int num_segs = 1; 453 int sack_bytes_rxmt = 0; 454 455 tp->t_partialacks++; 456 TCP_TIMER_DISARM(tp, TCPT_REXMT); 457 tp->t_rtttime = 0; 458 459 /* 460 * send one or 2 segments based on how much new data was acked 461 */ 462 if (((th->th_ack - tp->snd_una) / tp->t_segsz) > 2) 463 num_segs = 2; 464 (void)tcp_sack_output(tp, &sack_bytes_rxmt); 465 tp->snd_cwnd = sack_bytes_rxmt + 466 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_segsz; 467 tp->t_flags |= TF_ACKNOW; 468 (void) tcp_output(tp); 469 } else { 470 /* 471 * Complete ack, inflate the congestion window to 472 * ssthresh and exit fast recovery. 473 * 474 * Window inflation should have left us with approx. 475 * snd_ssthresh outstanding data. But in case we 476 * would be inclined to send a burst, better to do 477 * it via the slow start mechanism. 478 */ 479 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 480 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 481 + tp->t_segsz; 482 else 483 tp->snd_cwnd = tp->snd_ssthresh; 484 tp->t_partialacks = -1; 485 tp->t_dupacks = 0; 486 if (SEQ_GT(th->th_ack, tp->snd_fack)) 487 tp->snd_fack = th->th_ack; 488 } 489 } 490 491 /* 492 * Returns pointer to a sackhole if there are any pending retransmissions; 493 * NULL otherwise. 494 */ 495 struct sackhole * 496 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 497 { 498 struct sackhole *cur = NULL; 499 500 if(!TCP_SACK_ENABLED(tp)) 501 return (NULL); 502 503 *sack_bytes_rexmt = 0; 504 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 505 if (SEQ_LT(cur->rxmit, cur->end)) { 506 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 507 /* old SACK hole */ 508 continue; 509 } 510 *sack_bytes_rexmt += (cur->rxmit - cur->start); 511 break; 512 } 513 *sack_bytes_rexmt += (cur->rxmit - cur->start); 514 } 515 516 return (cur); 517 } 518 519 /* 520 * After a timeout, the SACK list may be rebuilt. This SACK information 521 * should be used to avoid retransmitting SACKed data. This function 522 * traverses the SACK list to see if snd_nxt should be moved forward. 523 */ 524 void 525 tcp_sack_adjust(struct tcpcb *tp) 526 { 527 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 528 struct sackhole *n = NULL; 529 530 if (TAILQ_EMPTY(&tp->snd_holes)) 531 return; /* No holes */ 532 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 533 return; /* We're already beyond any SACKed blocks */ 534 535 /* 536 * Two cases for which we want to advance snd_nxt: 537 * i) snd_nxt lies between end of one hole and beginning of another 538 * ii) snd_nxt lies between end of last hole and rcv_lastsack 539 */ 540 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 541 if (SEQ_LT(tp->snd_nxt, cur->end)) 542 return; 543 if (SEQ_GEQ(tp->snd_nxt, n->start)) 544 cur = n; 545 else { 546 tp->snd_nxt = n->start; 547 return; 548 } 549 } 550 if (SEQ_LT(tp->snd_nxt, cur->end)) 551 return; 552 tp->snd_nxt = tp->rcv_lastsack; 553 554 return; 555 } 556
Indexes created Mon Sep 22 13:09:51 GMT 2025