tcp_sack.c revision 1.11
1 /* $NetBSD: tcp_sack.c,v 1.11 2005/03/18 21:25:09 kurahone 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 1.11 2005/03/18 21:25:09 kurahone 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_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 179 { 180 if (TCP_SACK_ENABLED(tp)) { 181 tp->rcv_dsack_block.left = seq; 182 tp->rcv_dsack_block.right = seq + len; 183 tp->rcv_sack_flags |= TCPSACK_HAVED; 184 } 185 } 186 187 void 188 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 189 { 190 struct sackblk 191 t_sack_block[(MAX_TCPOPTLEN - 2) / (sizeof(u_int32_t) * 2)]; 192 struct sackblk *sack = NULL; 193 struct sackhole *cur = NULL; 194 struct sackhole *tmp = NULL; 195 u_int32_t *lp = (u_int32_t *) (cp + 2); 196 int i, j, num_sack_blks; 197 tcp_seq left, right, acked; 198 199 /* 200 * If we aren't processing SACK responses, this is not an ACK 201 * or the peer sends us a sack option with invalid length, don't 202 * update the scoreboard. 203 */ 204 if (!TCP_SACK_ENABLED(tp) || ((th->th_flags & TH_ACK) == 0) || 205 (optlen % 8 != 2 || optlen < 10)) { 206 return; 207 } 208 209 /* If the ACK is outside [snd_una, snd_max], ignore the SACK options. */ 210 if (SEQ_LT(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max)) 211 return; 212 213 /* 214 * Extract SACK blocks. 215 * 216 * Note that t_sack_block is sorted so that we only need to do 217 * one pass over the sequence number space. (SACK "fast-path") 218 */ 219 num_sack_blks = optlen / 8; 220 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 221 for (i = 0; i < num_sack_blks; i++, lp += 2) { 222 memcpy(&left, lp, sizeof(*lp)); 223 memcpy(&right, lp + 1, sizeof(*lp)); 224 left = ntohl(left); 225 right = ntohl(right); 226 227 if (SEQ_LEQ(right, acked) || SEQ_GEQ(left, tp->snd_max) || 228 SEQ_GEQ(left, right)) { 229 /* SACK entry that's old, or invalid. */ 230 i--; 231 num_sack_blks--; 232 continue; 233 } 234 235 /* Insertion sort. */ 236 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 237 j--) { 238 t_sack_block[j].left = t_sack_block[j - 1].left; 239 t_sack_block[j].right = t_sack_block[j - 1].right; 240 } 241 t_sack_block[j].left = left; 242 t_sack_block[j].right = right; 243 } 244 245 /* Update the scoreboard. */ 246 cur = TAILQ_FIRST(&tp->snd_holes); 247 for (i = 0; i < num_sack_blks; i++) { 248 sack = &t_sack_block[i]; 249 /* 250 * FACK TCP. Update snd_fack so we can enter Fast 251 * Recovery early. 252 */ 253 if (SEQ_GEQ(sack->right, tp->snd_fack)) 254 tp->snd_fack = sack->right; 255 256 if (TAILQ_EMPTY(&tp->snd_holes)) { 257 /* First hole. */ 258 cur = (struct sackhole *) 259 pool_get(&sackhole_pool, PR_NOWAIT); 260 if (cur == NULL) { 261 /* ENOBUFS, bail out*/ 262 return; 263 } 264 cur->start = th->th_ack; 265 cur->end = sack->left; 266 cur->rxmit = cur->start; 267 tp->rcv_lastsack = sack->right; 268 TAILQ_INSERT_HEAD(&tp->snd_holes, cur, sackhole_q); 269 continue; /* With next sack block */ 270 } 271 272 /* Go through the list of holes. */ 273 while (cur) { 274 if (SEQ_LEQ(sack->right, cur->start)) 275 /* SACKs data before the current hole */ 276 break; /* No use going through more holes */ 277 278 if (SEQ_GEQ(sack->left, cur->end)) { 279 /* SACKs data beyond the current hole */ 280 cur = TAILQ_NEXT(cur, sackhole_q); 281 continue; 282 } 283 284 if (SEQ_LEQ(sack->left, cur->start)) { 285 /* Data acks at least the beginning of hole */ 286 if (SEQ_GEQ(sack->right, cur->end)) { 287 /* Acks entire hole, so delete hole */ 288 tmp = cur; 289 cur = TAILQ_NEXT(cur, sackhole_q); 290 TAILQ_REMOVE(&tp->snd_holes, tmp, 291 sackhole_q); 292 pool_put(&sackhole_pool, tmp); 293 break; 294 } 295 296 /* Otherwise, move start of hole forward */ 297 cur->start = sack->right; 298 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 299 break; 300 } 301 302 if (SEQ_GEQ(sack->right, cur->end)) { 303 /* Move end of hole backward. */ 304 cur->end = sack->left; 305 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 306 cur = TAILQ_NEXT(cur, sackhole_q); 307 break; 308 } 309 310 if (SEQ_LT(cur->start, sack->left) && 311 SEQ_GT(cur->end, sack->right)) { 312 /* 313 * ACKs some data in middle of a hole; need to 314 * split current hole 315 */ 316 tmp = (struct sackhole *) 317 pool_get(&sackhole_pool, PR_NOWAIT); 318 if (tmp == NULL) { 319 /* ENOBUFS, bail out. */ 320 return; 321 } 322 tmp->start = sack->right; 323 tmp->end = cur->end; 324 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 325 cur->end = sack->left; 326 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 327 TAILQ_INSERT_AFTER(&tp->snd_holes, cur, tmp, 328 sackhole_q); 329 cur = tmp; 330 break; 331 } 332 } 333 334 /* At this point, we have reached the tail of the list. */ 335 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 336 /* 337 * Need to append new hole at end. 338 */ 339 tmp = (struct sackhole *) 340 pool_get(&sackhole_pool, PR_NOWAIT); 341 if (tmp == NULL) 342 continue; /* ENOBUFS */ 343 tmp->start = tp->rcv_lastsack; 344 tmp->end = sack->left; 345 tmp->rxmit = tmp->start; 346 TAILQ_INSERT_TAIL(&tp->snd_holes, tmp, sackhole_q); 347 cur = tmp; 348 } 349 if (SEQ_LT(tp->rcv_lastsack, sack->right)) { 350 tp->rcv_lastsack = sack->right; 351 } 352 } 353 } 354 355 void 356 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th) 357 { 358 /* Max because this could be an older ack that just arrived. */ 359 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 360 th->th_ack : tp->snd_una; 361 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 362 struct sackhole *tmp; 363 364 while (cur) { 365 if (SEQ_LEQ(cur->end, lastack)) { 366 tmp = cur; 367 cur = TAILQ_NEXT(cur, sackhole_q); 368 TAILQ_REMOVE(&tp->snd_holes, tmp, sackhole_q); 369 pool_put(&sackhole_pool, tmp); 370 } else if (SEQ_LT(cur->start, lastack)) { 371 cur->start = lastack; 372 if (SEQ_LT(cur->rxmit, cur->start)) 373 cur->rxmit = cur->start; 374 break; 375 } else 376 break; 377 378 } 379 } 380 381 void 382 tcp_free_sackholes(struct tcpcb *tp) 383 { 384 struct sackhole *sack; 385 386 /* Free up the SACK hole list. */ 387 while (!TAILQ_EMPTY(&tp->snd_holes)) { 388 sack = TAILQ_FIRST(&tp->snd_holes); 389 TAILQ_REMOVE(&tp->snd_holes, sack, sackhole_q); 390 pool_put(&sackhole_pool, sack); 391 } 392 } 393 394 /* 395 * Implements the SACK response to a new ack, checking for partial acks 396 * in fast recovery. 397 */ 398 void 399 tcp_sack_newack(struct tcpcb *tp, struct tcphdr *th) 400 { 401 if (tp->t_partialacks < 0) { 402 /* 403 * Not in fast recovery. Reset the duplicate ack 404 * counter. 405 */ 406 tp->t_dupacks = 0; 407 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 408 /* 409 * Partial ack handling within a sack recovery episode. 410 * Keeping this very simple for now. When a partial ack 411 * is received, force snd_cwnd to a value that will allow 412 * the sender to transmit no more than 2 segments. 413 * If necessary, a fancier scheme can be adopted at a 414 * later point, but for now, the goal is to prevent the 415 * sender from bursting a large amount of data in the midst 416 * of sack recovery. 417 */ 418 int num_segs = 1; 419 int sack_bytes_rxmt = 0; 420 421 tp->t_partialacks++; 422 TCP_TIMER_DISARM(tp, TCPT_REXMT); 423 tp->t_rtttime = 0; 424 425 /* 426 * send one or 2 segments based on how much new data was acked 427 */ 428 if (((th->th_ack - tp->snd_una) / tp->t_segsz) > 2) 429 num_segs = 2; 430 (void)tcp_sack_output(tp, &sack_bytes_rxmt); 431 tp->snd_cwnd = sack_bytes_rxmt + 432 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_segsz; 433 tp->t_flags |= TF_ACKNOW; 434 (void) tcp_output(tp); 435 } else { 436 /* 437 * Complete ack, inflate the congestion window to 438 * ssthresh and exit fast recovery. 439 * 440 * Window inflation should have left us with approx. 441 * snd_ssthresh outstanding data. But in case we 442 * would be inclined to send a burst, better to do 443 * it via the slow start mechanism. 444 */ 445 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 446 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 447 + tp->t_segsz; 448 else 449 tp->snd_cwnd = tp->snd_ssthresh; 450 tp->t_partialacks = -1; 451 tp->t_dupacks = 0; 452 if (SEQ_GT(th->th_ack, tp->snd_fack)) 453 tp->snd_fack = th->th_ack; 454 } 455 } 456 457 /* 458 * Returns pointer to a sackhole if there are any pending retransmissions; 459 * NULL otherwise. 460 */ 461 struct sackhole * 462 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 463 { 464 struct sackhole *cur = NULL; 465 466 if(!TCP_SACK_ENABLED(tp)) 467 return (NULL); 468 469 *sack_bytes_rexmt = 0; 470 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 471 if (SEQ_LT(cur->rxmit, cur->end)) { 472 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 473 /* old SACK hole */ 474 continue; 475 } 476 *sack_bytes_rexmt += (cur->rxmit - cur->start); 477 break; 478 } 479 *sack_bytes_rexmt += (cur->rxmit - cur->start); 480 } 481 482 return (cur); 483 } 484 485 /* 486 * After a timeout, the SACK list may be rebuilt. This SACK information 487 * should be used to avoid retransmitting SACKed data. This function 488 * traverses the SACK list to see if snd_nxt should be moved forward. 489 */ 490 void 491 tcp_sack_adjust(struct tcpcb *tp) 492 { 493 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 494 struct sackhole *n = NULL; 495 496 if (TAILQ_EMPTY(&tp->snd_holes)) 497 return; /* No holes */ 498 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 499 return; /* We're already beyond any SACKed blocks */ 500 501 /* 502 * Two cases for which we want to advance snd_nxt: 503 * i) snd_nxt lies between end of one hole and beginning of another 504 * ii) snd_nxt lies between end of last hole and rcv_lastsack 505 */ 506 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 507 if (SEQ_LT(tp->snd_nxt, cur->end)) 508 return; 509 if (SEQ_GEQ(tp->snd_nxt, n->start)) 510 cur = n; 511 else { 512 tp->snd_nxt = n->start; 513 return; 514 } 515 } 516 if (SEQ_LT(tp->snd_nxt, cur->end)) 517 return; 518 tp->snd_nxt = tp->rcv_lastsack; 519 520 return; 521 } 522 523 int 524 tcp_sack_numblks(const struct tcpcb *tp) 525 { 526 int numblks; 527 528 if (!TCP_SACK_ENABLED(tp)) { 529 return 0; 530 } 531 532 numblks = (((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) ? 1 : 0) + 533 tp->t_segqlen; 534 535 if (numblks == 0) { 536 return 0; 537 } 538 539 if (numblks > TCP_SACK_MAX) { 540 numblks = TCP_SACK_MAX; 541 } 542 543 return numblks; 544 } 545
Indexes created Mon Sep 22 05:09:51 GMT 2025