1 /* $NetBSD: tcp_sack.c,v 1.34 2018/03/29 07:46:43 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 1.34 2018/03/29 07:46:43 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/tcpip.h> 157 #include <netinet/tcp_debug.h> 158 159 /* SACK block pool. */ 160 static struct pool sackhole_pool; 161 162 void 163 tcp_sack_init(void) 164 { 165 166 pool_init(&sackhole_pool, sizeof(struct sackhole), 0, 0, 0, 167 "sackholepl", NULL, IPL_SOFTNET); 168 } 169 170 static struct sackhole * 171 sack_allochole(struct tcpcb *tp) 172 { 173 struct sackhole *hole; 174 175 if (tp->snd_numholes >= tcp_sack_tp_maxholes || 176 tcp_sack_globalholes >= tcp_sack_globalmaxholes) { 177 return NULL; 178 } 179 hole = pool_get(&sackhole_pool, PR_NOWAIT); 180 if (hole == NULL) { 181 return NULL; 182 } 183 tp->snd_numholes++; 184 tcp_sack_globalholes++; 185 186 return hole; 187 } 188 189 static struct sackhole * 190 sack_inserthole(struct tcpcb *tp, tcp_seq start, tcp_seq end, 191 struct sackhole *prev) 192 { 193 struct sackhole *hole; 194 195 hole = sack_allochole(tp); 196 if (hole == NULL) { 197 return NULL; 198 } 199 hole->start = hole->rxmit = start; 200 hole->end = end; 201 if (prev != NULL) { 202 TAILQ_INSERT_AFTER(&tp->snd_holes, prev, hole, sackhole_q); 203 } else { 204 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, sackhole_q); 205 } 206 return hole; 207 } 208 209 static struct sackhole * 210 sack_removehole(struct tcpcb *tp, struct sackhole *hole) 211 { 212 struct sackhole *next; 213 214 next = TAILQ_NEXT(hole, sackhole_q); 215 tp->snd_numholes--; 216 tcp_sack_globalholes--; 217 TAILQ_REMOVE(&tp->snd_holes, hole, sackhole_q); 218 pool_put(&sackhole_pool, hole); 219 220 return next; 221 } 222 223 /* 224 * tcp_new_dsack: record the reception of a duplicated segment. 225 */ 226 227 void 228 tcp_new_dsack(struct tcpcb *tp, tcp_seq seq, u_int32_t len) 229 { 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 /* 239 * tcp_sack_option: parse the given SACK option and update the scoreboard. 240 */ 241 242 void 243 tcp_sack_option(struct tcpcb *tp, const struct tcphdr *th, const u_char *cp, 244 int optlen) 245 { 246 struct sackblk 247 t_sack_block[(MAX_TCPOPTLEN - 2) / (sizeof(u_int32_t) * 2)]; 248 struct sackblk *sack = NULL; 249 struct sackhole *cur = NULL; 250 struct sackhole *tmp = NULL; 251 const char *lp = cp + 2; 252 int i, j, num_sack_blks; 253 tcp_seq left, right, acked; 254 255 /* 256 * If we aren't processing SACK responses, this is not an ACK 257 * or the peer sends us a sack option with invalid length, don't 258 * update the scoreboard. 259 */ 260 if (!TCP_SACK_ENABLED(tp) || ((th->th_flags & TH_ACK) == 0) || 261 (optlen % 8 != 2 || optlen < 10)) { 262 return; 263 } 264 265 /* 266 * If we don't want any SACK holes to be allocated, just return. 267 */ 268 if (tcp_sack_globalmaxholes == 0 || tcp_sack_tp_maxholes == 0) { 269 return; 270 } 271 272 /* If the ACK is outside [snd_una, snd_max], ignore the SACK options. */ 273 if (SEQ_LT(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max)) 274 return; 275 276 /* 277 * Extract SACK blocks. 278 * 279 * Note that t_sack_block is sorted so that we only need to do 280 * one pass over the sequence number space. (SACK "fast-path") 281 */ 282 num_sack_blks = optlen / 8; 283 acked = (SEQ_GT(th->th_ack, tp->snd_una)) ? th->th_ack : tp->snd_una; 284 for (i = 0; i < num_sack_blks; i++, lp += sizeof(uint32_t) * 2) { 285 memcpy(&left, lp, sizeof(uint32_t)); 286 memcpy(&right, lp + sizeof(uint32_t), sizeof(uint32_t)); 287 left = ntohl(left); 288 right = ntohl(right); 289 290 if (SEQ_LEQ(right, acked) || SEQ_GT(right, tp->snd_max) || 291 SEQ_GEQ(left, right)) { 292 /* SACK entry that's old, or invalid. */ 293 i--; 294 num_sack_blks--; 295 continue; 296 } 297 298 /* Insertion sort. */ 299 for (j = i; (j > 0) && SEQ_LT(left, t_sack_block[j - 1].left); 300 j--) { 301 t_sack_block[j].left = t_sack_block[j - 1].left; 302 t_sack_block[j].right = t_sack_block[j - 1].right; 303 } 304 t_sack_block[j].left = left; 305 t_sack_block[j].right = right; 306 } 307 308 /* Update the scoreboard. */ 309 cur = TAILQ_FIRST(&tp->snd_holes); 310 for (i = 0; i < num_sack_blks; i++) { 311 sack = &t_sack_block[i]; 312 /* 313 * FACK TCP. Update snd_fack so we can enter Fast 314 * Recovery early. 315 */ 316 if (SEQ_GEQ(sack->right, tp->snd_fack)) 317 tp->snd_fack = sack->right; 318 319 if (TAILQ_EMPTY(&tp->snd_holes)) { 320 /* First hole. */ 321 cur = sack_inserthole(tp, th->th_ack, sack->left, NULL); 322 if (cur == NULL) { 323 /* ENOBUFS, bail out*/ 324 return; 325 } 326 tp->rcv_lastsack = sack->right; 327 continue; /* With next sack block */ 328 } 329 330 /* Go through the list of holes. */ 331 while (cur) { 332 if (SEQ_LEQ(sack->right, cur->start)) 333 /* SACKs data before the current hole */ 334 break; /* No use going through more holes */ 335 336 if (SEQ_GEQ(sack->left, cur->end)) { 337 /* SACKs data beyond the current hole */ 338 cur = TAILQ_NEXT(cur, sackhole_q); 339 continue; 340 } 341 342 if (SEQ_LEQ(sack->left, cur->start)) { 343 /* Data acks at least the beginning of hole */ 344 if (SEQ_GEQ(sack->right, cur->end)) { 345 /* Acks entire hole, so delete hole */ 346 cur = sack_removehole(tp, cur); 347 break; 348 } 349 350 /* Otherwise, move start of hole forward */ 351 cur->start = sack->right; 352 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 353 break; 354 } 355 356 if (SEQ_GEQ(sack->right, cur->end)) { 357 /* Move end of hole backward. */ 358 cur->end = sack->left; 359 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 360 cur = TAILQ_NEXT(cur, sackhole_q); 361 break; 362 } 363 364 if (SEQ_LT(cur->start, sack->left) && 365 SEQ_GT(cur->end, sack->right)) { 366 /* 367 * ACKs some data in middle of a hole; need to 368 * split current hole 369 */ 370 tmp = sack_inserthole(tp, sack->right, cur->end, 371 cur); 372 if (tmp == NULL) { 373 return; 374 } 375 tmp->rxmit = SEQ_MAX(cur->rxmit, tmp->start); 376 cur->end = sack->left; 377 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 378 cur = tmp; 379 break; 380 } 381 } 382 383 /* At this point, we have reached the tail of the list. */ 384 if (SEQ_LT(tp->rcv_lastsack, sack->left)) { 385 /* 386 * Need to append new hole at end. 387 */ 388 cur = sack_inserthole(tp, tp->rcv_lastsack, sack->left, 389 NULL); 390 if (cur == NULL) { 391 return; 392 } 393 } 394 if (SEQ_LT(tp->rcv_lastsack, sack->right)) { 395 tp->rcv_lastsack = sack->right; 396 } 397 } 398 } 399 400 /* 401 * tcp_del_sackholes: remove holes covered by a cumulative ACK. 402 */ 403 404 void 405 tcp_del_sackholes(struct tcpcb *tp, const struct tcphdr *th) 406 { 407 /* Max because this could be an older ack that just arrived. */ 408 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 409 th->th_ack : tp->snd_una; 410 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 411 412 while (cur) { 413 if (SEQ_LEQ(cur->end, lastack)) { 414 cur = sack_removehole(tp, cur); 415 } else if (SEQ_LT(cur->start, lastack)) { 416 cur->start = lastack; 417 if (SEQ_LT(cur->rxmit, cur->start)) 418 cur->rxmit = cur->start; 419 break; 420 } else 421 break; 422 } 423 } 424 425 /* 426 * tcp_free_sackholes: clear the scoreboard. 427 */ 428 429 void 430 tcp_free_sackholes(struct tcpcb *tp) 431 { 432 struct sackhole *sack; 433 434 /* Free up the SACK hole list. */ 435 while ((sack = TAILQ_FIRST(&tp->snd_holes)) != NULL) { 436 sack_removehole(tp, sack); 437 } 438 KASSERT(tp->snd_numholes == 0); 439 } 440 441 /* 442 * Returns pointer to a sackhole if there are any pending retransmissions; 443 * NULL otherwise. 444 */ 445 struct sackhole * 446 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 447 { 448 struct sackhole *cur = NULL; 449 450 if (!TCP_SACK_ENABLED(tp)) 451 return (NULL); 452 453 *sack_bytes_rexmt = 0; 454 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 455 if (SEQ_LT(cur->rxmit, cur->end)) { 456 if (SEQ_LT(cur->rxmit, tp->snd_una)) { 457 /* old SACK hole */ 458 continue; 459 } 460 *sack_bytes_rexmt += (cur->rxmit - cur->start); 461 break; 462 } 463 *sack_bytes_rexmt += (cur->rxmit - cur->start); 464 } 465 466 return (cur); 467 } 468 469 /* 470 * After a timeout, the SACK list may be rebuilt. This SACK information 471 * should be used to avoid retransmitting SACKed data. This function 472 * traverses the SACK list to see if snd_nxt should be moved forward. 473 */ 474 void 475 tcp_sack_adjust(struct tcpcb *tp) 476 { 477 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 478 struct sackhole *n = NULL; 479 480 if (TAILQ_EMPTY(&tp->snd_holes)) 481 return; /* No holes */ 482 if (SEQ_GEQ(tp->snd_nxt, tp->rcv_lastsack)) 483 return; /* We're already beyond any SACKed blocks */ 484 485 /* 486 * Two cases for which we want to advance snd_nxt: 487 * i) snd_nxt lies between end of one hole and beginning of another 488 * ii) snd_nxt lies between end of last hole and rcv_lastsack 489 */ 490 while ((n = TAILQ_NEXT(cur, sackhole_q)) != NULL) { 491 if (SEQ_LT(tp->snd_nxt, cur->end)) 492 return; 493 if (SEQ_GEQ(tp->snd_nxt, n->start)) 494 cur = n; 495 else { 496 tp->snd_nxt = n->start; 497 return; 498 } 499 } 500 if (SEQ_LT(tp->snd_nxt, cur->end)) 501 return; 502 tp->snd_nxt = tp->rcv_lastsack; 503 504 return; 505 } 506 507 /* 508 * tcp_sack_numblks: return the number of SACK blocks to send. 509 */ 510 511 int 512 tcp_sack_numblks(const struct tcpcb *tp) 513 { 514 int numblks; 515 516 if (!TCP_SACK_ENABLED(tp)) { 517 return 0; 518 } 519 520 numblks = (((tp->rcv_sack_flags & TCPSACK_HAVED) != 0) ? 1 : 0) + 521 tp->t_segqlen; 522 523 if (numblks == 0) { 524 return 0; 525 } 526 527 if (numblks > TCP_SACK_MAX) { 528 numblks = TCP_SACK_MAX; 529 } 530 531 return numblks; 532 } 533 534 #if defined(DDB) 535 void sack_dump(const struct tcpcb *); 536 537 void 538 sack_dump(const struct tcpcb *tp) 539 { 540 const struct sackhole *cur; 541 542 printf("snd_una=%" PRIu32 ", snd_max=%" PRIu32 "\n", 543 tp->snd_una, tp->snd_max); 544 printf("rcv_lastsack=%" PRIu32 ", snd_fack=%" PRIu32 "\n", 545 tp->rcv_lastsack, tp->snd_fack); 546 printf("numholes=%d\n", tp->snd_numholes); 547 TAILQ_FOREACH(cur, &tp->snd_holes, sackhole_q) { 548 printf("\t%" PRIu32 "-%" PRIu32 ", rxmit=%" PRIu32 "\n", 549 cur->start, cur->end, cur->rxmit); 550 } 551 } 552 #endif /* defined(DDB) */ 553