tcp_sack.c revision 1.21
1 /* $NetBSD: tcp_sack.c,v 1.21 2006/10/21 10:08:54 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 1.21 2006/10/21 10:08:54 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. */ 172 static POOL_INIT(sackhole_pool, sizeof(struct sackhole), 0, 0, 0, "sackholepl", 173 NULL); 174 175 static struct sackhole * 176 sack_allochole(struct tcpcb *tp) 177 { 178 struct sackhole *hole; 179 180 if (tp->snd_numholes >= tcp_sack_tp_maxholes || 181 tcp_sack_globalholes >= tcp_sack_globalmaxholes) { 182 return NULL; 183 } 184 hole = pool_get(&sackhole_pool, PR_NOWAIT); 185 if (hole == NULL) { 186 return NULL; 187 } 188 tp->snd_numholes++; 189 tcp_sack_globalholes++; 190 191 return hole; 192 } 193 194 static struct sackhole * 195 sack_inserthole(struct tcpcb *tp, tcp_seq start, tcp_seq end, 196 struct sackhole *prev) 197 { 198 struct sackhole *hole; 199 200 hole = sack_allochole(tp); 201 if (hole == NULL) { 202 return NULL; 203 } 204 hole->start = hole->rxmit = start; 205 hole->end = end; 206 if (prev != NULL) { 207 TAILQ_INSERT_AFTER(&tp->snd_holes, prev, hole, sackhole_q); 208 } else { 209 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, sackhole_q); 210 } 211 return hole; 212 } 213 214 static struct sackhole * 215 sack_removehole(struct tcpcb *tp, struct sackhole *hole) 216 { 217 struct sackhole *next; 218 219 next = TAILQ_NEXT(hole, sackhole_q); 220 tp->snd_numholes--; 221 tcp_sack_globalholes--; 222 TAILQ_REMOVE(&tp->snd_holes, hole, sackhole_q); 223 pool_put(&sackhole_pool, hole); 224 225 return next; 226 } 227 228 void 229 tcp_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 230 { 231 if (TCP_SACK_ENABLED(tp)) { 232 tp->rcv_dsack_block.left = seq; 233 tp->rcv_dsack_block.right = seq + len; 234 tp->rcv_sack_flags |= TCPSACK_HAVED; 235 } 236 } 237 238 void 239 tcp_sack_option(struct tcpcb *tp, const struct tcphdr *th, const u_char *cp, 240 int optlen) 241 { 242 struct sackblk 243 t_sack_block[(MAX_TCPOPTLEN - 2) / (sizeof(u_int32_t) * 2)]; 244 struct sackblk *sack = NULL; 245 struct sackhole *cur = NULL; 246 struct sackhole *tmp = NULL; 247 const char *lp = cp + 2; 248 int i, j, num_sack_blks; 249 tcp_seq left, right, acked; 250 251 /* 252 * If we aren't processing SACK responses, this is not an ACK 253 * or the peer sends us a sack option with invalid length, don't 254 * update the scoreboard. 255 */ 256 if (!TCP_SACK_ENABLED(tp) || ((th->th_flags & TH_ACK) == 0) || 257 (optlen % 8 != 2 || optlen < 10)) { 258 return; 259 } 260 261 /* 262 * If we don't want any SACK holes to be allocated, just return. 263 */ 264 if (tcp_sack_globalmaxholes == 0 || tcp_sack_tp_maxholes == 0) { 265 return; 266 } 267 268 /* If the ACK is outside [snd_una, snd_max], ignore the SACK options. */ 269 if (SEQ_LT(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max)) 270 return; 271 272 /* 273 * Extract SACK blocks. 274 * 275 * Note that t_sack_block is sorted so that we only need to do 276 * one pass over the sequence number space. (SACK "fast-path") 277 */ 278 num_sack_blks = optlen / 8; 279 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 280 for (i = 0; i < num_sack_blks; i++, lp += sizeof(uint32_t) * 2) { 281 memcpy(&left, lp, sizeof(uint32_t)); 282 memcpy(&right, lp + sizeof(uint32_t), sizeof(uint32_t)); 283 left = ntohl(left); 284 right = ntohl(right); 285 286 if (SEQ_LEQ(right, acked) || SEQ_GT(right, tp->snd_max) || 287 SEQ_GEQ(left, right)) { 288 /* SACK entry that's old, or invalid. */ 289 i--; 290 num_sack_blks--; 291 continue; 292 } 293 294 /* Insertion sort. */ 295 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 296 j--) { 297 t_sack_block[j].left = t_sack_block[j - 1].left; 298 t_sack_block[j].right = t_sack_block[j - 1].right; 299 } 300 t_sack_block[j].left = left; 301 t_sack_block[j].right = right; 302 } 303 304 /* Update the scoreboard. */ 305 cur = TAILQ_FIRST(&tp->snd_holes); 306 for (i = 0; i < num_sack_blks; i++) { 307 sack = &t_sack_block[i]; 308 /* 309 * FACK TCP. Update snd_fack so we can enter Fast 310 * Recovery early. 311 */ 312 if (SEQ_GEQ(sack->right, tp->snd_fack)) 313 tp->snd_fack = sack->right; 314 315 if (TAILQ_EMPTY(&tp->snd_holes)) { 316 /* First hole. */ 317 cur = sack_inserthole(tp, th->th_ack, sack->left, NULL); 318 if (cur == NULL) { 319 /* ENOBUFS, bail out*/ 320 return; 321 } 322 tp->rcv_lastsack = sack->right; 323 continue; /* With next sack block */ 324 } 325 326 /* Go through the list of holes. */ 327 while (cur) { 328 if (SEQ_LEQ(sack->right, cur->start)) 329 /* SACKs data before the current hole */ 330 break; /* No use going through more holes */ 331 332 if (SEQ_GEQ(sack->left, cur->end)) { 333 /* SACKs data beyond the current hole */ 334 cur = TAILQ_NEXT(cur, sackhole_q); 335 continue; 336 } 337 338 if (SEQ_LEQ(sack->left, cur->start)) { 339 /* Data acks at least the beginning of hole */ 340 if (SEQ_GEQ(sack->right, cur->end)) { 341 /* Acks entire hole, so delete hole */ 342 cur = sack_removehole(tp, cur); 343 break; 344 } 345 346 /* Otherwise, move start of hole forward */ 347 cur->start = sack->right; 348 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 349 break; 350 } 351 352 if (SEQ_GEQ(sack->right, cur->end)) { 353 /* Move end of hole backward. */ 354 cur->end = sack->left; 355 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 356 cur = TAILQ_NEXT(cur, sackhole_q); 357 break; 358 } 359 360 if (SEQ_LT(cur->start, sack->left) && 361 SEQ_GT(cur->end, sack->right)) { 362 /* 363 * ACKs some data in middle of a hole; need to 364 * split current hole 365 */ 366 tmp = sack_inserthole(tp, sack->right, cur->end, 367 cur); 368 if (tmp == NULL) { 369 return; 370 } 371 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 372 cur->end = sack->left; 373 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 374 cur = tmp; 375 break; 376 } 377 } 378 379 /* At this point, we have reached the tail of the list. */ 380 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 381 /* 382 * Need to append new hole at end. 383 */ 384 cur = sack_inserthole(tp, tp->rcv_lastsack, sack->left, 385 NULL); 386 if (cur == NULL) { 387 return; 388 } 389 } 390 if (SEQ_LT(tp->rcv_lastsack, sack->right)) { 391 tp->rcv_lastsack = sack->right; 392 } 393 } 394 } 395 396 void 397 tcp_del_sackholes(struct tcpcb *tp, const struct tcphdr *th) 398 { 399 /* Max because this could be an older ack that just arrived. */ 400 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 401 th->th_ack : tp->snd_una; 402 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 403 404 while (cur) { 405 if (SEQ_LEQ(cur->end, lastack)) { 406 cur = sack_removehole(tp, cur); 407 } else if (SEQ_LT(cur->start, lastack)) { 408 cur->start = lastack; 409 if (SEQ_LT(cur->rxmit, cur->start)) 410 cur->rxmit = cur->start; 411 break; 412 } else 413 break; 414 } 415 } 416 417 void 418 tcp_free_sackholes(struct tcpcb *tp) 419 { 420 struct sackhole *sack; 421 422 /* Free up the SACK hole list. */ 423 while ((sack = TAILQ_FIRST(&tp->snd_holes)) != NULL) { 424 sack_removehole(tp, sack); 425 } 426 KASSERT(tp->snd_numholes == 0); 427 } 428 429 /* 430 * Implements the SACK response to a new ack, checking for partial acks 431 * in fast recovery. 432 */ 433 void 434 tcp_sack_newack(struct tcpcb *tp, const struct tcphdr *th) 435 { 436 if (tp->t_partialacks < 0) { 437 /* 438 * Not in fast recovery. Reset the duplicate ack 439 * counter. 440 */ 441 tp->t_dupacks = 0; 442 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 443 /* 444 * Partial ack handling within a sack recovery episode. 445 * Keeping this very simple for now. When a partial ack 446 * is received, force snd_cwnd to a value that will allow 447 * the sender to transmit no more than 2 segments. 448 * If necessary, a fancier scheme can be adopted at a 449 * later point, but for now, the goal is to prevent the 450 * sender from bursting a large amount of data in the midst 451 * of sack recovery. 452 */ 453 int num_segs = 1; 454 int sack_bytes_rxmt = 0; 455 456 tp->t_partialacks++; 457 TCP_TIMER_DISARM(tp, TCPT_REXMT); 458 tp->t_rtttime = 0; 459 460 /* 461 * send one or 2 segments based on how much new data was acked 462 */ 463 if (((th->th_ack - tp->snd_una) / tp->t_segsz) > 2) 464 num_segs = 2; 465 (void)tcp_sack_output(tp, &sack_bytes_rxmt); 466 tp->snd_cwnd = sack_bytes_rxmt + 467 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_segsz; 468 tp->t_flags |= TF_ACKNOW; 469 (void) tcp_output(tp); 470 } else { 471 /* 472 * Complete ack, inflate the congestion window to 473 * ssthresh and exit fast recovery. 474 * 475 * Window inflation should have left us with approx. 476 * snd_ssthresh outstanding data. But in case we 477 * would be inclined to send a burst, better to do 478 * it via the slow start mechanism. 479 */ 480 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 481 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 482 + tp->t_segsz; 483 else 484 tp->snd_cwnd = tp->snd_ssthresh; 485 tp->t_partialacks = -1; 486 tp->t_dupacks = 0; 487 if (SEQ_GT(th->th_ack, tp->snd_fack)) 488 tp->snd_fack = th->th_ack; 489 } 490 } 491 492 /* 493 * Returns pointer to a sackhole if there are any pending retransmissions; 494 * NULL otherwise. 495 */ 496 struct sackhole * 497 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 498 { 499 struct sackhole *cur = NULL; 500 501 if (!TCP_SACK_ENABLED(tp)) 502 return (NULL); 503 504 *sack_bytes_rexmt = 0; 505 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 506 if (SEQ_LT(cur->rxmit, cur->end)) { 507 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 508 /* old SACK hole */ 509 continue; 510 } 511 *sack_bytes_rexmt += (cur->rxmit - cur->start); 512 break; 513 } 514 *sack_bytes_rexmt += (cur->rxmit - cur->start); 515 } 516 517 return (cur); 518 } 519 520 /* 521 * After a timeout, the SACK list may be rebuilt. This SACK information 522 * should be used to avoid retransmitting SACKed data. This function 523 * traverses the SACK list to see if snd_nxt should be moved forward. 524 */ 525 void 526 tcp_sack_adjust(struct tcpcb *tp) 527 { 528 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 529 struct sackhole *n = NULL; 530 531 if (TAILQ_EMPTY(&tp->snd_holes)) 532 return; /* No holes */ 533 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 534 return; /* We're already beyond any SACKed blocks */ 535 536 /* 537 * Two cases for which we want to advance snd_nxt: 538 * i) snd_nxt lies between end of one hole and beginning of another 539 * ii) snd_nxt lies between end of last hole and rcv_lastsack 540 */ 541 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 542 if (SEQ_LT(tp->snd_nxt, cur->end)) 543 return; 544 if (SEQ_GEQ(tp->snd_nxt, n->start)) 545 cur = n; 546 else { 547 tp->snd_nxt = n->start; 548 return; 549 } 550 } 551 if (SEQ_LT(tp->snd_nxt, cur->end)) 552 return; 553 tp->snd_nxt = tp->rcv_lastsack; 554 555 return; 556 } 557 558 int 559 tcp_sack_numblks(const struct tcpcb *tp) 560 { 561 int numblks; 562 563 if (!TCP_SACK_ENABLED(tp)) { 564 return 0; 565 } 566 567 numblks = (((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) ? 1 : 0) + 568 tp->t_segqlen; 569 570 if (numblks == 0) { 571 return 0; 572 } 573 574 if (numblks > TCP_SACK_MAX) { 575 numblks = TCP_SACK_MAX; 576 } 577 578 return numblks; 579 } 580
Indexes created Tue Sep 23 07:09:52 GMT 2025