tcp_sack.c revision 1.9
1 /* $NetBSD: tcp_sack.c,v 1.9 2005/03/16 00:38:27 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 1.9 2005/03/16 00:38:27 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 i++; 195 } 196 197 /* 198 * Build up a list of holes in the TCP space. Note that 199 * the first SACK block is always the most recent segment 200 * received. 201 */ 202 TAILQ_FOREACH(tiqe, &tp->timeq, ipqe_timeq) { 203 tp->rcv_sack_block[i].left = tiqe->ipqe_seq; 204 tp->rcv_sack_block[i].right = tiqe->ipqe_seq + tiqe->ipqe_len; 205 i++; 206 if (i >= TCP_SACK_MAX) { 207 break; 208 } 209 } 210 211 /* If we can SACK, do so. */ 212 tp->rcv_sack_num = i; 213 } 214 215 void 216 tcp_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 217 { 218 if (TCP_SACK_ENABLED(tp)) { 219 tp->rcv_dsack_block.left = seq; 220 tp->rcv_dsack_block.right = seq + len; 221 tp->rcv_sack_flags |= TCPSACK_HAVED; 222 } 223 } 224 225 void 226 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 227 { 228 struct sackblk 229 t_sack_block[(MAX_TCPOPTLEN - 2) / (sizeof(u_int32_t) * 2)]; 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) || (optlen % 8 != 2 || optlen < 10)) { 243 return; 244 } 245 246 /* 247 * Extract SACK blocks. 248 * 249 * Note that t_sack_block is sorted so that we only need to do 250 * one pass over the sequence number space. (SACK "fast-path") 251 */ 252 num_sack_blks = optlen / 8; 253 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 254 for (i = 0; i < num_sack_blks; i++, lp += 2) { 255 memcpy(&left, lp, sizeof(*lp)); 256 memcpy(&right, lp + 1, sizeof(*lp)); 257 left = ntohl(left); 258 right = ntohl(right); 259 260 if (SEQ_LEQ(right, acked) || SEQ_GEQ(left, tp->snd_max) || 261 SEQ_GEQ(left, right)) { 262 /* SACK entry that's old, or invalid. */ 263 i--; 264 num_sack_blks--; 265 continue; 266 } 267 268 /* Insertion sort. */ 269 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 270 j--) { 271 t_sack_block[j].left = t_sack_block[j - 1].left; 272 t_sack_block[j].right = t_sack_block[j - 1].right; 273 } 274 t_sack_block[j].left = left; 275 t_sack_block[j].right = right; 276 } 277 278 /* Update the scoreboard. */ 279 cur = TAILQ_FIRST(&tp->snd_holes); 280 for (i = 0; i < num_sack_blks; i++) { 281 sack = &t_sack_block[i]; 282 /* 283 * FACK TCP. Update snd_fack so we can enter Fast 284 * Recovery early. 285 */ 286 if (SEQ_GEQ(sack->right, tp->snd_fack)) 287 tp->snd_fack = sack->right; 288 289 if (TAILQ_EMPTY(&tp->snd_holes)) { 290 /* First hole. */ 291 cur = (struct sackhole *) 292 pool_get(&sackhole_pool, PR_NOWAIT); 293 if (cur == NULL) { 294 /* ENOBUFS, bail out*/ 295 return; 296 } 297 cur->start = th->th_ack; 298 cur->end = sack->left; 299 cur->rxmit = cur->start; 300 tp->rcv_lastsack = sack->right; 301 TAILQ_INSERT_HEAD(&tp->snd_holes, cur, sackhole_q); 302 continue; /* With next sack block */ 303 } 304 305 /* Go through the list of holes. */ 306 while (cur) { 307 if (SEQ_LEQ(sack->right, cur->start)) 308 /* SACKs data before the current hole */ 309 break; /* No use going through more holes */ 310 311 if (SEQ_GEQ(sack->left, cur->end)) { 312 /* SACKs data beyond the current hole */ 313 cur = TAILQ_NEXT(cur, sackhole_q); 314 continue; 315 } 316 317 if (SEQ_LEQ(sack->left, cur->start)) { 318 /* Data acks at least the beginning of hole */ 319 if (SEQ_GEQ(sack->right, cur->end)) { 320 /* Acks entire hole, so delete hole */ 321 tmp = cur; 322 cur = TAILQ_NEXT(cur, sackhole_q); 323 TAILQ_REMOVE(&tp->snd_holes, tmp, 324 sackhole_q); 325 pool_put(&sackhole_pool, tmp); 326 break; 327 } 328 329 /* Otherwise, move start of hole forward */ 330 cur->start = sack->right; 331 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 332 break; 333 } 334 335 if (SEQ_GEQ(sack->right, cur->end)) { 336 /* Move end of hole backward. */ 337 cur->end = sack->left; 338 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 339 cur = TAILQ_NEXT(cur, sackhole_q); 340 break; 341 } 342 343 if (SEQ_LT(cur->start, sack->left) && 344 SEQ_GT(cur->end, sack->right)) { 345 /* 346 * ACKs some data in middle of a hole; need to 347 * split current hole 348 */ 349 tmp = (struct sackhole *) 350 pool_get(&sackhole_pool, PR_NOWAIT); 351 if (tmp == NULL) { 352 /* ENOBUFS, bail out. */ 353 return; 354 } 355 tmp->start = sack->right; 356 tmp->end = cur->end; 357 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 358 cur->end = sack->left; 359 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 360 TAILQ_INSERT_AFTER(&tp->snd_holes, cur, tmp, 361 sackhole_q); 362 cur = tmp; 363 break; 364 } 365 } 366 367 /* At this point, we have reached the tail of the list. */ 368 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 369 /* 370 * Need to append new hole at end. 371 */ 372 tmp = (struct sackhole *) 373 pool_get(&sackhole_pool, PR_NOWAIT); 374 if (tmp == NULL) 375 continue; /* ENOBUFS */ 376 tmp->start = tp->rcv_lastsack; 377 tmp->end = sack->left; 378 tmp->rxmit = tmp->start; 379 TAILQ_INSERT_TAIL(&tp->snd_holes, tmp, sackhole_q); 380 cur = tmp; 381 } 382 if (SEQ_LT(tp->rcv_lastsack, sack->right)) { 383 tp->rcv_lastsack = sack->right; 384 } 385 } 386 } 387 388 void 389 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th) 390 { 391 /* Max because this could be an older ack that just arrived. */ 392 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 393 th->th_ack : tp->snd_una; 394 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 395 struct sackhole *tmp; 396 397 while (cur) { 398 if (SEQ_LEQ(cur->end, lastack)) { 399 tmp = cur; 400 cur = TAILQ_NEXT(cur, sackhole_q); 401 TAILQ_REMOVE(&tp->snd_holes, tmp, sackhole_q); 402 pool_put(&sackhole_pool, tmp); 403 } else if (SEQ_LT(cur->start, lastack)) { 404 cur->start = lastack; 405 if (SEQ_LT(cur->rxmit, cur->start)) 406 cur->rxmit = cur->start; 407 break; 408 } else 409 break; 410 411 } 412 } 413 414 void 415 tcp_free_sackholes(struct tcpcb *tp) 416 { 417 struct sackhole *sack; 418 419 /* Free up the SACK hole list. */ 420 while (!TAILQ_EMPTY(&tp->snd_holes)) { 421 sack = TAILQ_FIRST(&tp->snd_holes); 422 TAILQ_REMOVE(&tp->snd_holes, sack, sackhole_q); 423 pool_put(&sackhole_pool, sack); 424 } 425 } 426 427 /* 428 * Implements the SACK response to a new ack, checking for partial acks 429 * in fast recovery. 430 */ 431 void 432 tcp_sack_newack(struct tcpcb *tp, struct tcphdr *th) 433 { 434 if (tp->t_partialacks < 0) { 435 /* 436 * Not in fast recovery. Reset the duplicate ack 437 * counter. 438 */ 439 tp->t_dupacks = 0; 440 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 441 /* 442 * Partial ack handling within a sack recovery episode. 443 * Keeping this very simple for now. When a partial ack 444 * is received, force snd_cwnd to a value that will allow 445 * the sender to transmit no more than 2 segments. 446 * If necessary, a fancier scheme can be adopted at a 447 * later point, but for now, the goal is to prevent the 448 * sender from bursting a large amount of data in the midst 449 * of sack recovery. 450 */ 451 int num_segs = 1; 452 int sack_bytes_rxmt = 0; 453 454 tp->t_partialacks++; 455 TCP_TIMER_DISARM(tp, TCPT_REXMT); 456 tp->t_rtttime = 0; 457 458 /* 459 * send one or 2 segments based on how much new data was acked 460 */ 461 if (((th->th_ack - tp->snd_una) / tp->t_segsz) > 2) 462 num_segs = 2; 463 (void)tcp_sack_output(tp, &sack_bytes_rxmt); 464 tp->snd_cwnd = sack_bytes_rxmt + 465 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_segsz; 466 tp->t_flags |= TF_ACKNOW; 467 (void) tcp_output(tp); 468 } else { 469 /* 470 * Complete ack, inflate the congestion window to 471 * ssthresh and exit fast recovery. 472 * 473 * Window inflation should have left us with approx. 474 * snd_ssthresh outstanding data. But in case we 475 * would be inclined to send a burst, better to do 476 * it via the slow start mechanism. 477 */ 478 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 479 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 480 + tp->t_segsz; 481 else 482 tp->snd_cwnd = tp->snd_ssthresh; 483 tp->t_partialacks = -1; 484 tp->t_dupacks = 0; 485 if (SEQ_GT(th->th_ack, tp->snd_fack)) 486 tp->snd_fack = th->th_ack; 487 } 488 } 489 490 /* 491 * Returns pointer to a sackhole if there are any pending retransmissions; 492 * NULL otherwise. 493 */ 494 struct sackhole * 495 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 496 { 497 struct sackhole *cur = NULL; 498 499 if(!TCP_SACK_ENABLED(tp)) 500 return (NULL); 501 502 *sack_bytes_rexmt = 0; 503 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 504 if (SEQ_LT(cur->rxmit, cur->end)) { 505 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 506 /* old SACK hole */ 507 continue; 508 } 509 *sack_bytes_rexmt += (cur->rxmit - cur->start); 510 break; 511 } 512 *sack_bytes_rexmt += (cur->rxmit - cur->start); 513 } 514 515 return (cur); 516 } 517 518 /* 519 * After a timeout, the SACK list may be rebuilt. This SACK information 520 * should be used to avoid retransmitting SACKed data. This function 521 * traverses the SACK list to see if snd_nxt should be moved forward. 522 */ 523 void 524 tcp_sack_adjust(struct tcpcb *tp) 525 { 526 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 527 struct sackhole *n = NULL; 528 529 if (TAILQ_EMPTY(&tp->snd_holes)) 530 return; /* No holes */ 531 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 532 return; /* We're already beyond any SACKed blocks */ 533 534 /* 535 * Two cases for which we want to advance snd_nxt: 536 * i) snd_nxt lies between end of one hole and beginning of another 537 * ii) snd_nxt lies between end of last hole and rcv_lastsack 538 */ 539 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 540 if (SEQ_LT(tp->snd_nxt, cur->end)) 541 return; 542 if (SEQ_GEQ(tp->snd_nxt, n->start)) 543 cur = n; 544 else { 545 tp->snd_nxt = n->start; 546 return; 547 } 548 } 549 if (SEQ_LT(tp->snd_nxt, cur->end)) 550 return; 551 tp->snd_nxt = tp->rcv_lastsack; 552 553 return; 554 } 555 556 int 557 tcp_sack_optlen(struct tcpcb *tp) 558 { 559 560 if (!TCP_SACK_ENABLED(tp) || tp->rcv_sack_num == 0) { 561 return 0; 562 } 563 564 return tp->rcv_sack_num * 8 + 2 + 2; 565 } 566
Indexes created Mon Sep 22 21:09:42 GMT 2025