1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * $DragonFly: src/sys/netinet/tcp_sack.c,v 1.8 2008/08/15 21:37:16 nth Exp $ 34 */ 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/kernel.h> 39 #include <sys/malloc.h> 40 #include <sys/queue.h> 41 #include <sys/thread.h> 42 #include <sys/types.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.h> 45 46 #include <net/if.h> 47 48 #include <netinet/in.h> 49 #include <netinet/in_systm.h> 50 #include <netinet/ip.h> 51 #include <netinet/in_var.h> 52 #include <netinet/in_pcb.h> 53 #include <netinet/ip_var.h> 54 #include <netinet/tcp.h> 55 #include <netinet/tcp_seq.h> 56 #include <netinet/tcp_var.h> 57 58 /* 59 * Implemented: 60 * 61 * RFC 2018 62 * RFC 2883 63 * RFC 3517 64 */ 65 66 struct sackblock { 67 tcp_seq sblk_start; 68 tcp_seq sblk_end; 69 TAILQ_ENTRY(sackblock) sblk_list; 70 }; 71 72 #define MAXSAVEDBLOCKS 8 /* per connection limit */ 73 74 static int insert_block(struct scoreboard *scb, 75 const struct raw_sackblock *raw_sb, boolean_t *update); 76 77 static MALLOC_DEFINE(M_SACKBLOCK, "sblk", "sackblock struct"); 78 79 /* 80 * Per-tcpcb initialization. 81 */ 82 void 83 tcp_sack_tcpcb_init(struct tcpcb *tp) 84 { 85 struct scoreboard *scb = &tp->scb; 86 87 scb->nblocks = 0; 88 TAILQ_INIT(&scb->sackblocks); 89 scb->lastfound = NULL; 90 } 91 92 /* 93 * Find the SACK block containing or immediately preceding "seq". 94 * The boolean result indicates whether the sequence is actually 95 * contained in the SACK block. 96 */ 97 static boolean_t 98 sack_block_lookup(struct scoreboard *scb, tcp_seq seq, struct sackblock **sb) 99 { 100 struct sackblock *hint = scb->lastfound; 101 struct sackblock *cur, *last, *prev; 102 103 if (TAILQ_EMPTY(&scb->sackblocks)) { 104 *sb = NULL; 105 return FALSE; 106 } 107 108 if (hint == NULL) { 109 /* No hint. Search from start to end. */ 110 cur = TAILQ_FIRST(&scb->sackblocks); 111 last = NULL; 112 prev = TAILQ_LAST(&scb->sackblocks, sackblock_list); 113 } else { 114 if (SEQ_GEQ(seq, hint->sblk_start)) { 115 /* Search from hint to end of list. */ 116 cur = hint; 117 last = NULL; 118 prev = TAILQ_LAST(&scb->sackblocks, sackblock_list); 119 } else { 120 /* Search from front of list to hint. */ 121 cur = TAILQ_FIRST(&scb->sackblocks); 122 last = hint; 123 prev = TAILQ_PREV(hint, sackblock_list, sblk_list); 124 } 125 } 126 127 do { 128 if (SEQ_GT(cur->sblk_end, seq)) { 129 if (SEQ_GEQ(seq, cur->sblk_start)) { 130 *sb = scb->lastfound = cur; 131 return TRUE; 132 } else { 133 *sb = scb->lastfound = 134 TAILQ_PREV(cur, sackblock_list, sblk_list); 135 return FALSE; 136 } 137 } 138 cur = TAILQ_NEXT(cur, sblk_list); 139 } while (cur != last); 140 141 *sb = scb->lastfound = prev; 142 return FALSE; 143 } 144 145 /* 146 * Allocate a SACK block. 147 */ 148 static __inline struct sackblock * 149 alloc_sackblock(struct scoreboard *scb, const struct raw_sackblock *raw_sb) 150 { 151 struct sackblock *sb; 152 153 if (scb->freecache != NULL) { 154 sb = scb->freecache; 155 scb->freecache = NULL; 156 tcpstat.tcps_sacksbfast++; 157 } else { 158 sb = kmalloc(sizeof(struct sackblock), M_SACKBLOCK, M_NOWAIT); 159 if (sb == NULL) { 160 tcpstat.tcps_sacksbfailed++; 161 return NULL; 162 } 163 } 164 sb->sblk_start = raw_sb->rblk_start; 165 sb->sblk_end = raw_sb->rblk_end; 166 return sb; 167 } 168 169 static __inline struct sackblock * 170 alloc_sackblock_limit(struct scoreboard *scb, 171 const struct raw_sackblock *raw_sb) 172 { 173 if (scb->nblocks == MAXSAVEDBLOCKS) { 174 /* 175 * Should try to kick out older blocks XXX JH 176 * May be able to coalesce with existing block. 177 * Or, go other way and free all blocks if we hit 178 * this limit. 179 */ 180 tcpstat.tcps_sacksboverflow++; 181 return NULL; 182 } 183 return alloc_sackblock(scb, raw_sb); 184 } 185 186 /* 187 * Free a SACK block. 188 */ 189 static __inline void 190 free_sackblock(struct scoreboard *scb, struct sackblock *s) 191 { 192 if (scb->freecache == NULL) { 193 /* YYY Maybe use the latest freed block? */ 194 scb->freecache = s; 195 return; 196 } 197 kfree(s, M_SACKBLOCK); 198 } 199 200 /* 201 * Free up SACK blocks for data that's been acked. 202 */ 203 static void 204 tcp_sack_ack_blocks(struct scoreboard *scb, tcp_seq th_ack) 205 { 206 struct sackblock *sb, *nb; 207 208 sb = TAILQ_FIRST(&scb->sackblocks); 209 while (sb && SEQ_LEQ(sb->sblk_end, th_ack)) { 210 nb = TAILQ_NEXT(sb, sblk_list); 211 if (scb->lastfound == sb) 212 scb->lastfound = NULL; 213 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list); 214 free_sackblock(scb, sb); 215 --scb->nblocks; 216 KASSERT(scb->nblocks >= 0, 217 ("SACK block count underflow: %d < 0", scb->nblocks)); 218 sb = nb; 219 } 220 if (sb && SEQ_GT(th_ack, sb->sblk_start)) 221 sb->sblk_start = th_ack; /* other side reneged? XXX */ 222 } 223 224 /* 225 * Delete and free SACK blocks saved in scoreboard. 226 */ 227 void 228 tcp_sack_cleanup(struct scoreboard *scb) 229 { 230 struct sackblock *sb, *nb; 231 232 TAILQ_FOREACH_MUTABLE(sb, &scb->sackblocks, sblk_list, nb) { 233 free_sackblock(scb, sb); 234 --scb->nblocks; 235 } 236 KASSERT(scb->nblocks == 0, 237 ("SACK block %d count not zero", scb->nblocks)); 238 TAILQ_INIT(&scb->sackblocks); 239 scb->lastfound = NULL; 240 } 241 242 /* 243 * Delete and free SACK blocks saved in scoreboard. 244 * Delete the one slot block cache. 245 */ 246 void 247 tcp_sack_destroy(struct scoreboard *scb) 248 { 249 tcp_sack_cleanup(scb); 250 if (scb->freecache != NULL) { 251 kfree(scb->freecache, M_SACKBLOCK); 252 scb->freecache = NULL; 253 } 254 } 255 256 /* 257 * Cleanup the reported SACK block information 258 */ 259 void 260 tcp_sack_report_cleanup(struct tcpcb *tp) 261 { 262 tp->sack_flags &= 263 ~(TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT); 264 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 265 } 266 267 /* 268 * Returns 0 if not D-SACK block, 269 * 1 if D-SACK, 270 * 2 if duplicate of out-of-order D-SACK block. 271 */ 272 int 273 tcp_sack_ndsack_blocks(struct raw_sackblock *blocks, const int numblocks, 274 tcp_seq snd_una) 275 { 276 if (numblocks == 0) 277 return 0; 278 279 if (SEQ_LT(blocks[0].rblk_start, snd_una)) 280 return 1; 281 282 /* block 0 inside block 1 */ 283 if (numblocks > 1 && 284 SEQ_GEQ(blocks[0].rblk_start, blocks[1].rblk_start) && 285 SEQ_LEQ(blocks[0].rblk_end, blocks[1].rblk_end)) 286 return 2; 287 288 return 0; 289 } 290 291 /* 292 * Update scoreboard on new incoming ACK. 293 */ 294 static void 295 tcp_sack_add_blocks(struct tcpcb *tp, struct tcpopt *to) 296 { 297 const int numblocks = to->to_nsackblocks; 298 struct raw_sackblock *blocks = to->to_sackblocks; 299 struct scoreboard *scb = &tp->scb; 300 int startblock, i; 301 302 if (tcp_sack_ndsack_blocks(blocks, numblocks, tp->snd_una) > 0) 303 startblock = 1; 304 else 305 startblock = 0; 306 307 to->to_flags |= TOF_SACK_REDUNDANT; 308 for (i = startblock; i < numblocks; i++) { 309 struct raw_sackblock *newsackblock = &blocks[i]; 310 boolean_t update; 311 int error; 312 313 /* Guard against ACK reordering */ 314 if (SEQ_LT(newsackblock->rblk_start, tp->snd_una)) 315 continue; 316 317 /* Don't accept bad SACK blocks */ 318 if (SEQ_GT(newsackblock->rblk_end, tp->snd_max)) { 319 tcpstat.tcps_rcvbadsackopt++; 320 break; /* skip all other blocks */ 321 } 322 tcpstat.tcps_sacksbupdate++; 323 324 error = insert_block(scb, newsackblock, &update); 325 if (update) 326 to->to_flags &= ~TOF_SACK_REDUNDANT; 327 if (error) 328 break; 329 } 330 } 331 332 void 333 tcp_sack_update_scoreboard(struct tcpcb *tp, struct tcpopt *to) 334 { 335 struct scoreboard *scb = &tp->scb; 336 int rexmt_high_update = 0; 337 338 tcp_sack_ack_blocks(scb, tp->snd_una); 339 tcp_sack_add_blocks(tp, to); 340 tcp_sack_update_lostseq(scb, tp->snd_una, tp->t_maxseg, 341 tp->t_rxtthresh); 342 if (SEQ_LT(tp->rexmt_high, tp->snd_una)) { 343 tp->rexmt_high = tp->snd_una; 344 rexmt_high_update = 1; 345 } 346 if (tp->sack_flags & TSACK_F_SACKRESCUED) { 347 if (SEQ_LT(tp->rexmt_rescue, tp->snd_una)) { 348 tp->sack_flags &= ~TSACK_F_SACKRESCUED; 349 } else if (tcp_aggressive_rescuesack && rexmt_high_update && 350 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) { 351 /* Drag RescueRxt along with HighRxt */ 352 tp->rexmt_rescue = tp->rexmt_high; 353 } 354 } 355 } 356 357 /* 358 * Insert SACK block into sender's scoreboard. 359 */ 360 static int 361 insert_block(struct scoreboard *scb, const struct raw_sackblock *raw_sb, 362 boolean_t *update) 363 { 364 struct sackblock *sb, *workingblock; 365 boolean_t overlap_front; 366 367 *update = TRUE; 368 if (TAILQ_EMPTY(&scb->sackblocks)) { 369 struct sackblock *newblock; 370 371 KASSERT(scb->nblocks == 0, ("emply scb w/ blocks")); 372 373 newblock = alloc_sackblock(scb, raw_sb); 374 if (newblock == NULL) 375 return ENOMEM; 376 TAILQ_INSERT_HEAD(&scb->sackblocks, newblock, sblk_list); 377 scb->nblocks = 1; 378 return 0; 379 } 380 381 KASSERT(scb->nblocks > 0, ("insert_block() called w/ no blocks")); 382 KASSERT(scb->nblocks <= MAXSAVEDBLOCKS, 383 ("too many SACK blocks %d", scb->nblocks)); 384 385 overlap_front = sack_block_lookup(scb, raw_sb->rblk_start, &sb); 386 387 if (sb == NULL) { 388 workingblock = alloc_sackblock_limit(scb, raw_sb); 389 if (workingblock == NULL) 390 return ENOMEM; 391 TAILQ_INSERT_HEAD(&scb->sackblocks, workingblock, sblk_list); 392 ++scb->nblocks; 393 } else { 394 if (overlap_front || sb->sblk_end == raw_sb->rblk_start) { 395 /* Extend old block */ 396 workingblock = sb; 397 if (SEQ_GT(raw_sb->rblk_end, sb->sblk_end)) 398 sb->sblk_end = raw_sb->rblk_end; 399 else 400 *update = FALSE; 401 tcpstat.tcps_sacksbreused++; 402 } else { 403 workingblock = alloc_sackblock_limit(scb, raw_sb); 404 if (workingblock == NULL) 405 return ENOMEM; 406 TAILQ_INSERT_AFTER(&scb->sackblocks, sb, workingblock, 407 sblk_list); 408 ++scb->nblocks; 409 } 410 } 411 412 /* Consolidate right-hand side. */ 413 sb = TAILQ_NEXT(workingblock, sblk_list); 414 while (sb != NULL && 415 SEQ_GEQ(workingblock->sblk_end, sb->sblk_end)) { 416 struct sackblock *nextblock; 417 418 nextblock = TAILQ_NEXT(sb, sblk_list); 419 if (scb->lastfound == sb) 420 scb->lastfound = NULL; 421 /* Remove completely overlapped block */ 422 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list); 423 free_sackblock(scb, sb); 424 --scb->nblocks; 425 KASSERT(scb->nblocks > 0, 426 ("removed overlapped block: %d blocks left", scb->nblocks)); 427 sb = nextblock; 428 } 429 if (sb != NULL && 430 SEQ_GEQ(workingblock->sblk_end, sb->sblk_start)) { 431 /* Extend new block to cover partially overlapped old block. */ 432 workingblock->sblk_end = sb->sblk_end; 433 if (scb->lastfound == sb) 434 scb->lastfound = NULL; 435 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list); 436 free_sackblock(scb, sb); 437 --scb->nblocks; 438 KASSERT(scb->nblocks > 0, 439 ("removed partial right: %d blocks left", scb->nblocks)); 440 } 441 return 0; 442 } 443 444 #ifdef DEBUG_SACK_BLOCKS 445 static void 446 tcp_sack_dump_blocks(struct scoreboard *scb) 447 { 448 struct sackblock *sb; 449 450 kprintf("%d blocks:", scb->nblocks); 451 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) 452 kprintf(" [%u, %u)", sb->sblk_start, sb->sblk_end); 453 kprintf("\n"); 454 } 455 #else 456 static __inline void 457 tcp_sack_dump_blocks(struct scoreboard *scb) 458 { 459 } 460 #endif 461 462 /* 463 * Optimization to quickly determine which packets are lost. 464 */ 465 void 466 tcp_sack_update_lostseq(struct scoreboard *scb, tcp_seq snd_una, u_int maxseg, 467 int rxtthresh) 468 { 469 struct sackblock *sb; 470 int nsackblocks = 0; 471 int bytes_sacked = 0; 472 int rxtthresh_bytes; 473 474 /* 475 * XXX 476 * The RFC3517bis recommends to reduce the byte threshold. 477 * However, it will cause extra spurious retransmit if 478 * segments are reordered. Before certain DupThresh adaptive 479 * algorithm is implemented, we don't reduce the byte 480 * threshold (tcp_rfc3517bis_rxt is off by default). 481 */ 482 if (tcp_do_rfc3517bis && tcp_rfc3517bis_rxt) 483 rxtthresh_bytes = (rxtthresh - 1) * maxseg; 484 else 485 rxtthresh_bytes = rxtthresh * maxseg; 486 487 sb = TAILQ_LAST(&scb->sackblocks, sackblock_list); 488 while (sb != NULL) { 489 ++nsackblocks; 490 bytes_sacked += sb->sblk_end - sb->sblk_start; 491 if (nsackblocks == rxtthresh || 492 bytes_sacked >= rxtthresh_bytes) { 493 scb->lostseq = sb->sblk_start; 494 return; 495 } 496 sb = TAILQ_PREV(sb, sackblock_list, sblk_list); 497 } 498 scb->lostseq = snd_una; 499 } 500 501 /* 502 * Return whether the given sequence number is considered lost. 503 */ 504 boolean_t 505 tcp_sack_islost(struct scoreboard *scb, tcp_seq seqnum) 506 { 507 return SEQ_LT(seqnum, scb->lostseq); 508 } 509 510 /* 511 * True if at least "amount" has been SACKed. Used by Early Retransmit. 512 */ 513 boolean_t 514 tcp_sack_has_sacked(struct scoreboard *scb, u_int amount) 515 { 516 struct sackblock *sb; 517 int bytes_sacked = 0; 518 519 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) { 520 bytes_sacked += sb->sblk_end - sb->sblk_start; 521 if (bytes_sacked >= amount) 522 return TRUE; 523 } 524 return FALSE; 525 } 526 527 /* 528 * Number of bytes SACKed below seq. 529 */ 530 int 531 tcp_sack_bytes_below(struct scoreboard *scb, tcp_seq seq) 532 { 533 struct sackblock *sb; 534 int bytes_sacked = 0; 535 536 sb = TAILQ_FIRST(&scb->sackblocks); 537 while (sb && SEQ_GT(seq, sb->sblk_start)) { 538 bytes_sacked += seq_min(seq, sb->sblk_end) - sb->sblk_start; 539 sb = TAILQ_NEXT(sb, sblk_list); 540 } 541 return bytes_sacked; 542 } 543 544 /* 545 * Return estimate of the number of bytes outstanding in the network. 546 */ 547 uint32_t 548 tcp_sack_compute_pipe(struct tcpcb *tp) 549 { 550 struct scoreboard *scb = &tp->scb; 551 struct sackblock *sb; 552 int nlost, nretransmitted; 553 tcp_seq end; 554 555 nlost = tp->snd_max - scb->lostseq; 556 nretransmitted = tp->rexmt_high - tp->snd_una; 557 558 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) { 559 if (SEQ_LT(sb->sblk_start, tp->rexmt_high)) { 560 end = seq_min(sb->sblk_end, tp->rexmt_high); 561 nretransmitted -= end - sb->sblk_start; 562 } 563 if (SEQ_GEQ(sb->sblk_start, scb->lostseq)) 564 nlost -= sb->sblk_end - sb->sblk_start; 565 } 566 567 return (nlost + nretransmitted); 568 } 569 570 /* 571 * Return the sequence number and length of the next segment to transmit 572 * when in Fast Recovery. 573 */ 574 boolean_t 575 tcp_sack_nextseg(struct tcpcb *tp, tcp_seq *nextrexmt, uint32_t *plen, 576 boolean_t *rescue) 577 { 578 struct scoreboard *scb = &tp->scb; 579 struct socket *so = tp->t_inpcb->inp_socket; 580 struct sackblock *sb; 581 const struct sackblock *lastblock = 582 TAILQ_LAST(&scb->sackblocks, sackblock_list); 583 tcp_seq torexmt; 584 long len, off; 585 586 /* skip SACKed data */ 587 tcp_sack_skip_sacked(scb, &tp->rexmt_high); 588 589 /* Look for lost data. */ 590 torexmt = tp->rexmt_high; 591 *rescue = FALSE; 592 if (lastblock != NULL) { 593 if (SEQ_LT(torexmt, lastblock->sblk_end) && 594 tcp_sack_islost(scb, torexmt)) { 595 sendunsacked: 596 *nextrexmt = torexmt; 597 /* If the left-hand edge has been SACKed, pull it in. */ 598 if (sack_block_lookup(scb, torexmt + tp->t_maxseg, &sb)) 599 *plen = sb->sblk_start - torexmt; 600 else 601 *plen = tp->t_maxseg; 602 return TRUE; 603 } 604 } 605 606 /* See if unsent data available within send window. */ 607 off = tp->snd_max - tp->snd_una; 608 len = (long) ulmin(so->so_snd.ssb_cc, tp->snd_wnd) - off; 609 if (len > 0) { 610 *nextrexmt = tp->snd_max; /* Send new data. */ 611 *plen = tp->t_maxseg; 612 return TRUE; 613 } 614 615 /* We're less certain this data has been lost. */ 616 if (lastblock != NULL && SEQ_LT(torexmt, lastblock->sblk_end)) 617 goto sendunsacked; 618 619 /* Rescue retransmission */ 620 if (tcp_do_rescuesack || tcp_do_rfc3517bis) { 621 tcpstat.tcps_sackrescue_try++; 622 if (tp->sack_flags & TSACK_F_SACKRESCUED) { 623 if (!tcp_aggressive_rescuesack) 624 return FALSE; 625 626 /* 627 * Aggressive variant of the rescue retransmission. 628 * 629 * The idea of the rescue retransmission is to sustain 630 * the ACK clock thus to avoid timeout retransmission. 631 * 632 * Under some situations, the conservative approach 633 * suggested in the draft 634 * http://tools.ietf.org/html/ 635 * draft-nishida-tcpm-rescue-retransmission-00 636 * could not sustain ACK clock, since it only allows 637 * one rescue retransmission before a cumulative ACK 638 * covers the segement transmitted by rescue 639 * retransmission. 640 * 641 * We try to locate the next unSACKed segment which 642 * follows the previously sent rescue segment. If 643 * there is no such segment, we loop back to the first 644 * unacknowledged segment. 645 */ 646 647 /* 648 * Skip SACKed data, but here we follow 649 * the last transmitted rescue segment. 650 */ 651 torexmt = tp->rexmt_rescue; 652 tcp_sack_skip_sacked(scb, &torexmt); 653 if (torexmt == tp->snd_max) { 654 /* Nothing left to retransmit; restart */ 655 torexmt = tp->snd_una; 656 } 657 } 658 *rescue = TRUE; 659 goto sendunsacked; 660 } else if (tcp_do_smartsack && lastblock == NULL) { 661 tcpstat.tcps_sackrescue_try++; 662 *rescue = TRUE; 663 goto sendunsacked; 664 } 665 666 return FALSE; 667 } 668 669 /* 670 * Return the next sequence number higher than "*prexmt" that has 671 * not been SACKed. 672 */ 673 void 674 tcp_sack_skip_sacked(struct scoreboard *scb, tcp_seq *prexmt) 675 { 676 struct sackblock *sb; 677 678 /* skip SACKed data */ 679 if (sack_block_lookup(scb, *prexmt, &sb)) 680 *prexmt = sb->sblk_end; 681 } 682 683 #ifdef later 684 void 685 tcp_sack_save_scoreboard(struct scoreboard *scb) 686 { 687 struct scoreboard *scb = &tp->scb; 688 689 scb->sackblocks_prev = scb->sackblocks; 690 TAILQ_INIT(&scb->sackblocks); 691 } 692 693 void 694 tcp_sack_revert_scoreboard(struct scoreboard *scb, tcp_seq snd_una, 695 u_int maxseg) 696 { 697 struct sackblock *sb; 698 699 scb->sackblocks = scb->sackblocks_prev; 700 scb->nblocks = 0; 701 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) 702 ++scb->nblocks; 703 tcp_sack_ack_blocks(scb, snd_una); 704 scb->lastfound = NULL; 705 } 706 #endif 707 708 #ifdef DEBUG_SACK_HISTORY 709 static void 710 tcp_sack_dump_history(char *msg, struct tcpcb *tp) 711 { 712 int i; 713 static int ndumped; 714 715 /* only need a couple of these to debug most problems */ 716 if (++ndumped > 900) 717 return; 718 719 kprintf("%s:\tnsackhistory %d: ", msg, tp->nsackhistory); 720 for (i = 0; i < tp->nsackhistory; ++i) 721 kprintf("[%u, %u) ", tp->sackhistory[i].rblk_start, 722 tp->sackhistory[i].rblk_end); 723 kprintf("\n"); 724 } 725 #else 726 static __inline void 727 tcp_sack_dump_history(char *msg, struct tcpcb *tp) 728 { 729 } 730 #endif 731 732 /* 733 * Remove old SACK blocks from the SACK history that have already been ACKed. 734 */ 735 static void 736 tcp_sack_ack_history(struct tcpcb *tp) 737 { 738 int i, nblocks, openslot; 739 740 tcp_sack_dump_history("before tcp_sack_ack_history", tp); 741 nblocks = tp->nsackhistory; 742 for (i = openslot = 0; i < nblocks; ++i) { 743 if (SEQ_LEQ(tp->sackhistory[i].rblk_end, tp->rcv_nxt)) { 744 --tp->nsackhistory; 745 continue; 746 } 747 if (SEQ_LT(tp->sackhistory[i].rblk_start, tp->rcv_nxt)) 748 tp->sackhistory[i].rblk_start = tp->rcv_nxt; 749 if (i == openslot) 750 ++openslot; 751 else 752 tp->sackhistory[openslot++] = tp->sackhistory[i]; 753 } 754 tcp_sack_dump_history("after tcp_sack_ack_history", tp); 755 KASSERT(openslot == tp->nsackhistory, 756 ("tcp_sack_ack_history miscounted: %d != %d", 757 openslot, tp->nsackhistory)); 758 } 759 760 /* 761 * Add or merge newblock into reported history. 762 * Also remove or update SACK blocks that will be acked. 763 */ 764 static void 765 tcp_sack_update_reported_history(struct tcpcb *tp, tcp_seq start, tcp_seq end) 766 { 767 struct raw_sackblock copy[MAX_SACK_REPORT_BLOCKS]; 768 int i, cindex; 769 770 tcp_sack_dump_history("before tcp_sack_update_reported_history", tp); 771 /* 772 * Six cases: 773 * 0) no overlap 774 * 1) newblock == oldblock 775 * 2) oldblock contains newblock 776 * 3) newblock contains oldblock 777 * 4) tail of oldblock overlaps or abuts start of newblock 778 * 5) tail of newblock overlaps or abuts head of oldblock 779 */ 780 for (i = cindex = 0; i < tp->nsackhistory; ++i) { 781 struct raw_sackblock *oldblock = &tp->sackhistory[i]; 782 tcp_seq old_start = oldblock->rblk_start; 783 tcp_seq old_end = oldblock->rblk_end; 784 785 if (SEQ_LT(end, old_start) || SEQ_GT(start, old_end)) { 786 /* Case 0: no overlap. Copy old block. */ 787 copy[cindex++] = *oldblock; 788 continue; 789 } 790 791 if (SEQ_GEQ(start, old_start) && SEQ_LEQ(end, old_end)) { 792 /* Cases 1 & 2. Move block to front of history. */ 793 int j; 794 795 start = old_start; 796 end = old_end; 797 /* no need to check rest of blocks */ 798 for (j = i + 1; j < tp->nsackhistory; ++j) 799 copy[cindex++] = tp->sackhistory[j]; 800 break; 801 } 802 803 if (SEQ_GEQ(old_end, start) && SEQ_LT(old_start, start)) { 804 /* Case 4: extend start of new block. */ 805 start = old_start; 806 } else if (SEQ_GEQ(end, old_start) && SEQ_GT(old_end, end)) { 807 /* Case 5: extend end of new block */ 808 end = old_end; 809 } else { 810 /* Case 3. Delete old block by not copying it. */ 811 KASSERT(SEQ_LEQ(start, old_start) && 812 SEQ_GEQ(end, old_end), 813 ("bad logic: old [%u, %u), new [%u, %u)", 814 old_start, old_end, start, end)); 815 } 816 } 817 818 /* insert new block */ 819 tp->sackhistory[0].rblk_start = start; 820 tp->sackhistory[0].rblk_end = end; 821 cindex = min(cindex, MAX_SACK_REPORT_BLOCKS - 1); 822 for (i = 0; i < cindex; ++i) 823 tp->sackhistory[i + 1] = copy[i]; 824 tp->nsackhistory = cindex + 1; 825 tcp_sack_dump_history("after tcp_sack_update_reported_history", tp); 826 } 827 828 /* 829 * Fill in SACK report to return to data sender. 830 */ 831 void 832 tcp_sack_fill_report(struct tcpcb *tp, u_char *opt, u_int *plen) 833 { 834 u_int optlen = *plen; 835 uint32_t *lp = (uint32_t *)(opt + optlen); 836 uint32_t *olp; 837 tcp_seq hstart = tp->rcv_nxt, hend; 838 int nblocks; 839 840 KASSERT(TCP_MAXOLEN - optlen >= 841 TCPOLEN_SACK_ALIGNED + TCPOLEN_SACK_BLOCK, 842 ("no room for SACK header and one block: optlen %d", optlen)); 843 844 if (tp->sack_flags & TSACK_F_DUPSEG) 845 tcpstat.tcps_snddsackopt++; 846 else 847 tcpstat.tcps_sndsackopt++; 848 849 olp = lp++; 850 optlen += TCPOLEN_SACK_ALIGNED; 851 852 tcp_sack_ack_history(tp); 853 if (tp->reportblk.rblk_start != tp->reportblk.rblk_end) { 854 *lp++ = htonl(tp->reportblk.rblk_start); 855 *lp++ = htonl(tp->reportblk.rblk_end); 856 optlen += TCPOLEN_SACK_BLOCK; 857 hstart = tp->reportblk.rblk_start; 858 hend = tp->reportblk.rblk_end; 859 if (tp->sack_flags & TSACK_F_ENCLOSESEG) { 860 KASSERT(TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK, 861 ("no room for enclosing SACK block: oplen %d", 862 optlen)); 863 *lp++ = htonl(tp->encloseblk.rblk_start); 864 *lp++ = htonl(tp->encloseblk.rblk_end); 865 optlen += TCPOLEN_SACK_BLOCK; 866 hstart = tp->encloseblk.rblk_start; 867 hend = tp->encloseblk.rblk_end; 868 } 869 if (SEQ_GT(hstart, tp->rcv_nxt)) 870 tcp_sack_update_reported_history(tp, hstart, hend); 871 } 872 if (tcp_do_smartsack && (tp->sack_flags & TSACK_F_SACKLEFT)) { 873 /* Fill in from left! Walk re-assembly queue. */ 874 struct tseg_qent *q; 875 876 q = TAILQ_FIRST(&tp->t_segq); 877 while (q != NULL && 878 TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) { 879 *lp++ = htonl(q->tqe_th->th_seq); 880 *lp++ = htonl(TCP_SACK_BLKEND( 881 q->tqe_th->th_seq + q->tqe_len, 882 q->tqe_th->th_flags)); 883 optlen += TCPOLEN_SACK_BLOCK; 884 q = TAILQ_NEXT(q, tqe_q); 885 } 886 } else { 887 int n = 0; 888 889 /* Fill in SACK blocks from right side. */ 890 while (n < tp->nsackhistory && 891 TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) { 892 if (tp->sackhistory[n].rblk_start != hstart) { 893 *lp++ = htonl(tp->sackhistory[n].rblk_start); 894 *lp++ = htonl(tp->sackhistory[n].rblk_end); 895 optlen += TCPOLEN_SACK_BLOCK; 896 } 897 ++n; 898 } 899 } 900 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 901 tp->sack_flags &= 902 ~(TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT); 903 nblocks = (lp - olp - 1) / 2; 904 *olp = htonl(TCPOPT_SACK_ALIGNED | 905 (TCPOLEN_SACK + nblocks * TCPOLEN_SACK_BLOCK)); 906 *plen = optlen; 907 } 908