1 /* $NetBSD: tcp_sack.c,v 1.35 2018/05/03 07:13:48 maxv 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 34 * The Regents of the University of California. All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 4. Neither the name of the University nor the names of its contributors 45 * may be used to endorse or promote products derived from this software 46 * without specific prior written permission. 47 * 48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 * 60 * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95 61 * $FreeBSD: src/sys/netinet/tcp_sack.c,v 1.3.2.2 2004/12/25 23:02:57 rwatson Exp $ 62 */ 63 64 /* 65 * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995 66 * 67 * NRL grants permission for redistribution and use in source and binary 68 * forms, with or without modification, of the software and documentation 69 * created at NRL provided that the following conditions are met: 70 * 71 * 1. Redistributions of source code must retain the above copyright 72 * notice, this list of conditions and the following disclaimer. 73 * 2. Redistributions in binary form must reproduce the above copyright 74 * notice, this list of conditions and the following disclaimer in the 75 * documentation and/or other materials provided with the distribution. 76 * 3. All advertising materials mentioning features or use of this software 77 * must display the following acknowledgements: 78 * This product includes software developed by the University of 79 * California, Berkeley and its contributors. 80 * This product includes software developed at the Information 81 * Technology Division, US Naval Research Laboratory. 82 * 4. Neither the name of the NRL nor the names of its contributors 83 * may be used to endorse or promote products derived from this software 84 * without specific prior written permission. 85 * 86 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 87 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 88 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 89 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 90 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 91 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 92 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 93 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 94 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 95 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 96 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 97 * 98 * The views and conclusions contained in the software and documentation 99 * are those of the authors and should not be interpreted as representing 100 * official policies, either expressed or implied, of the US Naval 101 * Research Laboratory (NRL). 102 */ 103 104 #include <sys/cdefs.h> 105 __KERNEL_RCSID(0, "$NetBSD: tcp_sack.c,v 1.35 2018/05/03 07:13:48 maxv Exp $"); 106 107 #ifdef _KERNEL_OPT 108 #include "opt_inet.h" 109 #include "opt_inet_csum.h" 110 #include "opt_tcp_debug.h" 111 #include "opt_ddb.h" 112 #endif 113 114 #include <sys/param.h> 115 #include <sys/systm.h> 116 #include <sys/mbuf.h> 117 #include <sys/protosw.h> 118 #include <sys/socket.h> 119 #include <sys/socketvar.h> 120 #include <sys/errno.h> 121 #include <sys/syslog.h> 122 #include <sys/pool.h> 123 #include <sys/domain.h> 124 #include <sys/kernel.h> 125 126 #include <net/if.h> 127 #include <net/route.h> 128 #include <net/if_types.h> 129 130 #include <netinet/in.h> 131 #include <netinet/in_systm.h> 132 #include <netinet/ip.h> 133 #include <netinet/in_pcb.h> 134 #include <netinet/in_var.h> 135 #include <netinet/ip_var.h> 136 137 #ifdef INET6 138 #include <netinet/ip6.h> 139 #include <netinet6/ip6_var.h> 140 #include <netinet6/in6_pcb.h> 141 #include <netinet6/ip6_var.h> 142 #include <netinet6/in6_var.h> 143 #include <netinet/icmp6.h> 144 #endif 145 146 #ifndef INET6 147 /* always need ip6.h for IP6_EXTHDR_GET */ 148 #include <netinet/ip6.h> 149 #endif 150 151 #include <netinet/tcp.h> 152 #include <netinet/tcp_fsm.h> 153 #include <netinet/tcp_seq.h> 154 #include <netinet/tcp_timer.h> 155 #include <netinet/tcp_var.h> 156 #include <netinet/tcp_debug.h> 157 158 /* SACK block pool. */ 159 static struct pool sackhole_pool; 160 161 void 162 tcp_sack_init(void) 163 { 164 165 pool_init(&sackhole_pool, sizeof(struct sackhole), 0, 0, 0, 166 "sackholepl", NULL, IPL_SOFTNET); 167 } 168 169 static struct sackhole * 170 sack_allochole(struct tcpcb *tp) 171 { 172 struct sackhole *hole; 173 174 if (tp->snd_numholes >= tcp_sack_tp_maxholes || 175 tcp_sack_globalholes >= tcp_sack_globalmaxholes) { 176 return NULL; 177 } 178 hole = pool_get(&sackhole_pool, PR_NOWAIT); 179 if (hole == NULL) { 180 return NULL; 181 } 182 tp->snd_numholes++; 183 tcp_sack_globalholes++; 184 185 return hole; 186 } 187 188 static struct sackhole * 189 sack_inserthole(struct tcpcb *tp, tcp_seq start, tcp_seq end, 190 struct sackhole *prev) 191 { 192 struct sackhole *hole; 193 194 hole = sack_allochole(tp); 195 if (hole == NULL) { 196 return NULL; 197 } 198 hole->start = hole->rxmit = start; 199 hole->end = end; 200 if (prev != NULL) { 201 TAILQ_INSERT_AFTER(&tp->snd_holes, prev, hole, sackhole_q); 202 } else { 203 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, sackhole_q); 204 } 205 return hole; 206 } 207 208 static struct sackhole * 209 sack_removehole(struct tcpcb *tp, struct sackhole *hole) 210 { 211 struct sackhole *next; 212 213 next = TAILQ_NEXT(hole, sackhole_q); 214 tp->snd_numholes--; 215 tcp_sack_globalholes--; 216 TAILQ_REMOVE(&tp->snd_holes, hole, sackhole_q); 217 pool_put(&sackhole_pool, hole); 218 219 return next; 220 } 221 222 /* 223 * tcp_new_dsack: record the reception of a duplicated segment. 224 */ 225 226 void 227 tcp_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 228 { 229 230 if (TCP_SACK_ENABLED(tp)) { 231 tp->rcv_dsack_block.left = seq; 232 tp->rcv_dsack_block.right = seq + len; 233 tp->rcv_sack_flags |= TCPSACK_HAVED; 234 } 235 } 236 237 /* 238 * tcp_sack_option: parse the given SACK option and update the scoreboard. 239 */ 240 241 void 242 tcp_sack_option(struct tcpcb *tp, const struct tcphdr *th, const u_char *cp, 243 int optlen) 244 { 245 struct sackblk 246 t_sack_block[(MAX_TCPOPTLEN - 2) / (sizeof(u_int32_t) * 2)]; 247 struct sackblk *sack = NULL; 248 struct sackhole *cur = NULL; 249 struct sackhole *tmp = NULL; 250 const char *lp = cp + 2; 251 int i, j, num_sack_blks; 252 tcp_seq left, right, acked; 253 254 /* 255 * If we aren't processing SACK responses, this is not an ACK 256 * or the peer sends us a sack option with invalid length, don't 257 * update the scoreboard. 258 */ 259 if (!TCP_SACK_ENABLED(tp) || ((th->th_flags & TH_ACK) == 0) || 260 (optlen % 8 != 2 || optlen < 10)) { 261 return; 262 } 263 264 /* 265 * If we don't want any SACK holes to be allocated, just return. 266 */ 267 if (tcp_sack_globalmaxholes == 0 || tcp_sack_tp_maxholes == 0) { 268 return; 269 } 270 271 /* If the ACK is outside [snd_una, snd_max], ignore the SACK options. */ 272 if (SEQ_LT(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max)) 273 return; 274 275 /* 276 * Extract SACK blocks. 277 * 278 * Note that t_sack_block is sorted so that we only need to do 279 * one pass over the sequence number space. (SACK "fast-path") 280 */ 281 num_sack_blks = optlen / 8; 282 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 283 for (i = 0; i < num_sack_blks; i++, lp += sizeof(uint32_t) * 2) { 284 memcpy(&left, lp, sizeof(uint32_t)); 285 memcpy(&right, lp + sizeof(uint32_t), sizeof(uint32_t)); 286 left = ntohl(left); 287 right = ntohl(right); 288 289 if (SEQ_LEQ(right, acked) || SEQ_GT(right, tp->snd_max) || 290 SEQ_GEQ(left, right)) { 291 /* SACK entry that's old, or invalid. */ 292 i--; 293 num_sack_blks--; 294 continue; 295 } 296 297 /* Insertion sort. */ 298 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 299 j--) { 300 t_sack_block[j].left = t_sack_block[j - 1].left; 301 t_sack_block[j].right = t_sack_block[j - 1].right; 302 } 303 t_sack_block[j].left = left; 304 t_sack_block[j].right = right; 305 } 306 307 /* Update the scoreboard. */ 308 cur = TAILQ_FIRST(&tp->snd_holes); 309 for (i = 0; i < num_sack_blks; i++) { 310 sack = &t_sack_block[i]; 311 /* 312 * FACK TCP. Update snd_fack so we can enter Fast 313 * Recovery early. 314 */ 315 if (SEQ_GEQ(sack->right, tp->snd_fack)) 316 tp->snd_fack = sack->right; 317 318 if (TAILQ_EMPTY(&tp->snd_holes)) { 319 /* First hole. */ 320 cur = sack_inserthole(tp, th->th_ack, sack->left, NULL); 321 if (cur == NULL) { 322 /* ENOBUFS, bail out*/ 323 return; 324 } 325 tp->rcv_lastsack = sack->right; 326 continue; /* With next sack block */ 327 } 328 329 /* Go through the list of holes. */ 330 while (cur) { 331 if (SEQ_LEQ(sack->right, cur->start)) 332 /* SACKs data before the current hole */ 333 break; /* No use going through more holes */ 334 335 if (SEQ_GEQ(sack->left, cur->end)) { 336 /* SACKs data beyond the current hole */ 337 cur = TAILQ_NEXT(cur, sackhole_q); 338 continue; 339 } 340 341 if (SEQ_LEQ(sack->left, cur->start)) { 342 /* Data acks at least the beginning of hole */ 343 if (SEQ_GEQ(sack->right, cur->end)) { 344 /* Acks entire hole, so delete hole */ 345 cur = sack_removehole(tp, cur); 346 break; 347 } 348 349 /* Otherwise, move start of hole forward */ 350 cur->start = sack->right; 351 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 352 break; 353 } 354 355 if (SEQ_GEQ(sack->right, cur->end)) { 356 /* Move end of hole backward. */ 357 cur->end = sack->left; 358 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 359 cur = TAILQ_NEXT(cur, sackhole_q); 360 break; 361 } 362 363 if (SEQ_LT(cur->start, sack->left) && 364 SEQ_GT(cur->end, sack->right)) { 365 /* 366 * ACKs some data in middle of a hole; need to 367 * split current hole 368 */ 369 tmp = sack_inserthole(tp, sack->right, cur->end, 370 cur); 371 if (tmp == NULL) { 372 return; 373 } 374 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 375 cur->end = sack->left; 376 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 377 cur = tmp; 378 break; 379 } 380 } 381 382 /* At this point, we have reached the tail of the list. */ 383 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 384 /* 385 * Need to append new hole at end. 386 */ 387 cur = sack_inserthole(tp, tp->rcv_lastsack, sack->left, 388 NULL); 389 if (cur == NULL) { 390 return; 391 } 392 } 393 if (SEQ_LT(tp->rcv_lastsack, sack->right)) { 394 tp->rcv_lastsack = sack->right; 395 } 396 } 397 } 398 399 /* 400 * tcp_del_sackholes: remove holes covered by a cumulative ACK. 401 */ 402 403 void 404 tcp_del_sackholes(struct tcpcb *tp, const struct tcphdr *th) 405 { 406 /* Max because this could be an older ack that just arrived. */ 407 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 408 th->th_ack : tp->snd_una; 409 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 410 411 while (cur) { 412 if (SEQ_LEQ(cur->end, lastack)) { 413 cur = sack_removehole(tp, cur); 414 } else if (SEQ_LT(cur->start, lastack)) { 415 cur->start = lastack; 416 if (SEQ_LT(cur->rxmit, cur->start)) 417 cur->rxmit = cur->start; 418 break; 419 } else 420 break; 421 } 422 } 423 424 /* 425 * tcp_free_sackholes: clear the scoreboard. 426 */ 427 428 void 429 tcp_free_sackholes(struct tcpcb *tp) 430 { 431 struct sackhole *sack; 432 433 /* Free up the SACK hole list. */ 434 while ((sack = TAILQ_FIRST(&tp->snd_holes)) != NULL) { 435 sack_removehole(tp, sack); 436 } 437 KASSERT(tp->snd_numholes == 0); 438 } 439 440 /* 441 * Returns pointer to a sackhole if there are any pending retransmissions; 442 * NULL otherwise. 443 */ 444 struct sackhole * 445 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 446 { 447 struct sackhole *cur = NULL; 448 449 if (!TCP_SACK_ENABLED(tp)) 450 return (NULL); 451 452 *sack_bytes_rexmt = 0; 453 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 454 if (SEQ_LT(cur->rxmit, cur->end)) { 455 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 456 /* old SACK hole */ 457 continue; 458 } 459 *sack_bytes_rexmt += (cur->rxmit - cur->start); 460 break; 461 } 462 *sack_bytes_rexmt += (cur->rxmit - cur->start); 463 } 464 465 return (cur); 466 } 467 468 /* 469 * After a timeout, the SACK list may be rebuilt. This SACK information 470 * should be used to avoid retransmitting SACKed data. This function 471 * traverses the SACK list to see if snd_nxt should be moved forward. 472 */ 473 void 474 tcp_sack_adjust(struct tcpcb *tp) 475 { 476 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 477 struct sackhole *n = NULL; 478 479 if (TAILQ_EMPTY(&tp->snd_holes)) 480 return; /* No holes */ 481 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 482 return; /* We're already beyond any SACKed blocks */ 483 484 /* 485 * Two cases for which we want to advance snd_nxt: 486 * i) snd_nxt lies between end of one hole and beginning of another 487 * ii) snd_nxt lies between end of last hole and rcv_lastsack 488 */ 489 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 490 if (SEQ_LT(tp->snd_nxt, cur->end)) 491 return; 492 if (SEQ_GEQ(tp->snd_nxt, n->start)) 493 cur = n; 494 else { 495 tp->snd_nxt = n->start; 496 return; 497 } 498 } 499 if (SEQ_LT(tp->snd_nxt, cur->end)) 500 return; 501 tp->snd_nxt = tp->rcv_lastsack; 502 503 return; 504 } 505 506 /* 507 * tcp_sack_numblks: return the number of SACK blocks to send. 508 */ 509 510 int 511 tcp_sack_numblks(const struct tcpcb *tp) 512 { 513 int numblks; 514 515 if (!TCP_SACK_ENABLED(tp)) { 516 return 0; 517 } 518 519 numblks = (((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) ? 1 : 0) + 520 tp->t_segqlen; 521 522 if (numblks == 0) { 523 return 0; 524 } 525 526 if (numblks > TCP_SACK_MAX) { 527 numblks = TCP_SACK_MAX; 528 } 529 530 return numblks; 531 } 532 533 #if defined(DDB) 534 void sack_dump(const struct tcpcb *); 535 536 void 537 sack_dump(const struct tcpcb *tp) 538 { 539 const struct sackhole *cur; 540 541 printf("snd_una=%" PRIu32 ", snd_max=%" PRIu32 "\n", 542 tp->snd_una, tp->snd_max); 543 printf("rcv_lastsack=%" PRIu32 ", snd_fack=%" PRIu32 "\n", 544 tp->rcv_lastsack, tp->snd_fack); 545 printf("numholes=%d\n", tp->snd_numholes); 546 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 547 printf("\t%" PRIu32 "-%" PRIu32 ", rxmit=%" PRIu32 "\n", 548 cur->start, cur->end, cur->rxmit); 549 } 550 } 551 #endif /* defined(DDB) */ 552