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 * RFC 6675 65 */ 66 67 struct sackblock { 68 tcp_seq sblk_start; 69 tcp_seq sblk_end; 70 TAILQ_ENTRY(sackblock) sblk_list; 71 }; 72 73 #define MAXSAVEDBLOCKS 8 /* per connection limit */ 74 75 static int insert_block(struct scoreboard *scb, 76 const struct raw_sackblock *raw_sb, boolean_t *update); 77 78 static MALLOC_DEFINE(M_SACKBLOCK, "sblk", "sackblock struct"); 79 80 /* 81 * Per-tcpcb initialization. 82 */ 83 void 84 tcp_sack_tcpcb_init(struct tcpcb *tp) 85 { 86 struct scoreboard *scb = &tp->scb; 87 88 scb->nblocks = 0; 89 TAILQ_INIT(&scb->sackblocks); 90 scb->lastfound = NULL; 91 } 92 93 /* 94 * Find the SACK block containing or immediately preceding "seq". 95 * The boolean result indicates whether the sequence is actually 96 * contained in the SACK block. 97 */ 98 static boolean_t 99 sack_block_lookup(struct scoreboard *scb, tcp_seq seq, struct sackblock **sb) 100 { 101 struct sackblock *hint = scb->lastfound; 102 struct sackblock *cur, *last, *prev; 103 104 if (TAILQ_EMPTY(&scb->sackblocks)) { 105 *sb = NULL; 106 return FALSE; 107 } 108 109 if (hint == NULL) { 110 /* No hint. Search from start to end. */ 111 cur = TAILQ_FIRST(&scb->sackblocks); 112 last = NULL; 113 prev = TAILQ_LAST(&scb->sackblocks, sackblock_list); 114 } else { 115 if (SEQ_GEQ(seq, hint->sblk_start)) { 116 /* Search from hint to end of list. */ 117 cur = hint; 118 last = NULL; 119 prev = TAILQ_LAST(&scb->sackblocks, sackblock_list); 120 } else { 121 /* Search from front of list to hint. */ 122 cur = TAILQ_FIRST(&scb->sackblocks); 123 last = hint; 124 prev = TAILQ_PREV(hint, sackblock_list, sblk_list); 125 } 126 } 127 128 do { 129 if (SEQ_GT(cur->sblk_end, seq)) { 130 if (SEQ_GEQ(seq, cur->sblk_start)) { 131 *sb = scb->lastfound = cur; 132 return TRUE; 133 } else { 134 *sb = scb->lastfound = 135 TAILQ_PREV(cur, sackblock_list, sblk_list); 136 return FALSE; 137 } 138 } 139 cur = TAILQ_NEXT(cur, sblk_list); 140 } while (cur != last); 141 142 *sb = scb->lastfound = prev; 143 return FALSE; 144 } 145 146 /* 147 * Allocate a SACK block. 148 */ 149 static __inline struct sackblock * 150 alloc_sackblock(struct scoreboard *scb, const struct raw_sackblock *raw_sb) 151 { 152 struct sackblock *sb; 153 154 if (scb->freecache != NULL) { 155 sb = scb->freecache; 156 scb->freecache = NULL; 157 tcpstat.tcps_sacksbfast++; 158 } else { 159 sb = kmalloc(sizeof(struct sackblock), M_SACKBLOCK, M_NOWAIT); 160 if (sb == NULL) { 161 tcpstat.tcps_sacksbfailed++; 162 return NULL; 163 } 164 } 165 sb->sblk_start = raw_sb->rblk_start; 166 sb->sblk_end = raw_sb->rblk_end; 167 return sb; 168 } 169 170 static __inline struct sackblock * 171 alloc_sackblock_limit(struct scoreboard *scb, 172 const struct raw_sackblock *raw_sb) 173 { 174 if (scb->nblocks == MAXSAVEDBLOCKS) { 175 /* 176 * Should try to kick out older blocks XXX JH 177 * May be able to coalesce with existing block. 178 * Or, go other way and free all blocks if we hit 179 * this limit. 180 */ 181 tcpstat.tcps_sacksboverflow++; 182 return NULL; 183 } 184 return alloc_sackblock(scb, raw_sb); 185 } 186 187 /* 188 * Free a SACK block. 189 */ 190 static __inline void 191 free_sackblock(struct scoreboard *scb, struct sackblock *s) 192 { 193 if (scb->freecache == NULL) { 194 /* YYY Maybe use the latest freed block? */ 195 scb->freecache = s; 196 return; 197 } 198 kfree(s, M_SACKBLOCK); 199 } 200 201 /* 202 * Free up SACK blocks for data that's been acked. 203 */ 204 static void 205 tcp_sack_ack_blocks(struct tcpcb *tp, tcp_seq th_ack) 206 { 207 struct scoreboard *scb = &tp->scb; 208 struct sackblock *sb, *nb; 209 210 sb = TAILQ_FIRST(&scb->sackblocks); 211 while (sb && SEQ_LEQ(sb->sblk_end, th_ack)) { 212 nb = TAILQ_NEXT(sb, sblk_list); 213 if (scb->lastfound == sb) 214 scb->lastfound = NULL; 215 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list); 216 free_sackblock(scb, sb); 217 --scb->nblocks; 218 KASSERT(scb->nblocks >= 0, 219 ("SACK block count underflow: %d < 0", scb->nblocks)); 220 sb = nb; 221 } 222 if (sb && SEQ_GEQ(th_ack, sb->sblk_start)) { 223 /* Other side reneged? XXX */ 224 tcpstat.tcps_sackrenege++; 225 tcp_sack_discard(tp); 226 } 227 } 228 229 /* 230 * Delete and free SACK blocks saved in scoreboard. 231 */ 232 static void 233 tcp_sack_cleanup(struct scoreboard *scb) 234 { 235 struct sackblock *sb, *nb; 236 237 TAILQ_FOREACH_MUTABLE(sb, &scb->sackblocks, sblk_list, nb) { 238 free_sackblock(scb, sb); 239 --scb->nblocks; 240 } 241 KASSERT(scb->nblocks == 0, 242 ("SACK block %d count not zero", scb->nblocks)); 243 TAILQ_INIT(&scb->sackblocks); 244 scb->lastfound = NULL; 245 } 246 247 /* 248 * Discard SACK scoreboard, HighRxt, RescueRxt and LostSeq. 249 */ 250 void 251 tcp_sack_discard(struct tcpcb *tp) 252 { 253 tcp_sack_cleanup(&tp->scb); 254 tp->rexmt_high = tp->snd_una; 255 tp->sack_flags &= ~TSACK_F_SACKRESCUED; 256 tp->scb.lostseq = tp->snd_una; 257 } 258 259 /* 260 * Delete and free SACK blocks saved in scoreboard. 261 * Delete the one slot block cache. 262 */ 263 void 264 tcp_sack_destroy(struct scoreboard *scb) 265 { 266 tcp_sack_cleanup(scb); 267 if (scb->freecache != NULL) { 268 kfree(scb->freecache, M_SACKBLOCK); 269 scb->freecache = NULL; 270 } 271 } 272 273 /* 274 * Cleanup the reported SACK block information 275 */ 276 void 277 tcp_sack_report_cleanup(struct tcpcb *tp) 278 { 279 tp->sack_flags &= 280 ~(TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT); 281 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 282 } 283 284 /* 285 * Whether SACK report is needed or not 286 */ 287 boolean_t 288 tcp_sack_report_needed(const struct tcpcb *tp) 289 { 290 if ((tp->sack_flags & 291 (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT)) || 292 tp->reportblk.rblk_start != tp->reportblk.rblk_end) 293 return TRUE; 294 else 295 return FALSE; 296 } 297 298 /* 299 * Returns 0 if not D-SACK block, 300 * 1 if D-SACK, 301 * 2 if duplicate of out-of-order D-SACK block. 302 */ 303 int 304 tcp_sack_ndsack_blocks(const struct raw_sackblock *blocks, const int numblocks, 305 tcp_seq snd_una) 306 { 307 if (numblocks == 0) 308 return 0; 309 310 if (SEQ_LT(blocks[0].rblk_start, snd_una)) 311 return 1; 312 313 /* block 0 inside block 1 */ 314 if (numblocks > 1 && 315 SEQ_GEQ(blocks[0].rblk_start, blocks[1].rblk_start) && 316 SEQ_LEQ(blocks[0].rblk_end, blocks[1].rblk_end)) 317 return 2; 318 319 return 0; 320 } 321 322 /* 323 * Update scoreboard on new incoming ACK. 324 */ 325 static void 326 tcp_sack_add_blocks(struct tcpcb *tp, struct tcpopt *to) 327 { 328 const int numblocks = to->to_nsackblocks; 329 struct raw_sackblock *blocks = to->to_sackblocks; 330 struct scoreboard *scb = &tp->scb; 331 int startblock, i; 332 333 if (tcp_sack_ndsack_blocks(blocks, numblocks, tp->snd_una) > 0) 334 startblock = 1; 335 else 336 startblock = 0; 337 338 to->to_flags |= TOF_SACK_REDUNDANT; 339 for (i = startblock; i < numblocks; i++) { 340 struct raw_sackblock *newsackblock = &blocks[i]; 341 boolean_t update; 342 int error; 343 344 /* Guard against ACK reordering */ 345 if (SEQ_LEQ(newsackblock->rblk_start, tp->snd_una)) 346 continue; 347 348 /* Don't accept bad SACK blocks */ 349 if (SEQ_GT(newsackblock->rblk_end, tp->snd_max)) { 350 tcpstat.tcps_rcvbadsackopt++; 351 break; /* skip all other blocks */ 352 } 353 tcpstat.tcps_sacksbupdate++; 354 355 error = insert_block(scb, newsackblock, &update); 356 if (update) 357 to->to_flags &= ~TOF_SACK_REDUNDANT; 358 if (error) 359 break; 360 } 361 } 362 363 void 364 tcp_sack_update_scoreboard(struct tcpcb *tp, struct tcpopt *to) 365 { 366 struct scoreboard *scb = &tp->scb; 367 int rexmt_high_update = 0; 368 369 tcp_sack_ack_blocks(tp, tp->snd_una); 370 tcp_sack_add_blocks(tp, to); 371 tcp_sack_update_lostseq(scb, tp->snd_una, tp->t_maxseg, 372 tp->t_rxtthresh); 373 if (SEQ_LT(tp->rexmt_high, tp->snd_una)) { 374 tp->rexmt_high = tp->snd_una; 375 rexmt_high_update = 1; 376 } 377 if (tp->sack_flags & TSACK_F_SACKRESCUED) { 378 if (SEQ_LEQ(tp->rexmt_rescue, tp->snd_una)) { 379 tp->sack_flags &= ~TSACK_F_SACKRESCUED; 380 } else if (tcp_aggressive_rescuesack && rexmt_high_update && 381 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) { 382 /* Drag RescueRxt along with HighRxt */ 383 tp->rexmt_rescue = tp->rexmt_high; 384 } 385 } 386 } 387 388 /* 389 * Insert SACK block into sender's scoreboard. 390 */ 391 static int 392 insert_block(struct scoreboard *scb, const struct raw_sackblock *raw_sb, 393 boolean_t *update) 394 { 395 struct sackblock *sb, *workingblock; 396 boolean_t overlap_front; 397 398 *update = TRUE; 399 if (TAILQ_EMPTY(&scb->sackblocks)) { 400 struct sackblock *newblock; 401 402 KASSERT(scb->nblocks == 0, ("emply scb w/ blocks")); 403 404 newblock = alloc_sackblock(scb, raw_sb); 405 if (newblock == NULL) 406 return ENOMEM; 407 TAILQ_INSERT_HEAD(&scb->sackblocks, newblock, sblk_list); 408 scb->nblocks = 1; 409 return 0; 410 } 411 412 KASSERT(scb->nblocks > 0, ("insert_block() called w/ no blocks")); 413 KASSERT(scb->nblocks <= MAXSAVEDBLOCKS, 414 ("too many SACK blocks %d", scb->nblocks)); 415 416 overlap_front = sack_block_lookup(scb, raw_sb->rblk_start, &sb); 417 418 if (sb == NULL) { 419 workingblock = alloc_sackblock_limit(scb, raw_sb); 420 if (workingblock == NULL) 421 return ENOMEM; 422 TAILQ_INSERT_HEAD(&scb->sackblocks, workingblock, sblk_list); 423 ++scb->nblocks; 424 } else { 425 if (overlap_front || sb->sblk_end == raw_sb->rblk_start) { 426 tcpstat.tcps_sacksbreused++; 427 428 /* Extend old block */ 429 workingblock = sb; 430 if (SEQ_GT(raw_sb->rblk_end, sb->sblk_end)) { 431 sb->sblk_end = raw_sb->rblk_end; 432 } else { 433 /* Exact match, nothing to consolidate */ 434 *update = FALSE; 435 return 0; 436 } 437 } else { 438 workingblock = alloc_sackblock_limit(scb, raw_sb); 439 if (workingblock == NULL) 440 return ENOMEM; 441 TAILQ_INSERT_AFTER(&scb->sackblocks, sb, workingblock, 442 sblk_list); 443 ++scb->nblocks; 444 } 445 } 446 447 /* Consolidate right-hand side. */ 448 sb = TAILQ_NEXT(workingblock, sblk_list); 449 while (sb != NULL && 450 SEQ_GEQ(workingblock->sblk_end, sb->sblk_end)) { 451 struct sackblock *nextblock; 452 453 nextblock = TAILQ_NEXT(sb, sblk_list); 454 if (scb->lastfound == sb) 455 scb->lastfound = NULL; 456 /* Remove completely overlapped block */ 457 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list); 458 free_sackblock(scb, sb); 459 --scb->nblocks; 460 KASSERT(scb->nblocks > 0, 461 ("removed overlapped block: %d blocks left", scb->nblocks)); 462 sb = nextblock; 463 } 464 if (sb != NULL && 465 SEQ_GEQ(workingblock->sblk_end, sb->sblk_start)) { 466 /* Extend new block to cover partially overlapped old block. */ 467 workingblock->sblk_end = sb->sblk_end; 468 if (scb->lastfound == sb) 469 scb->lastfound = NULL; 470 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list); 471 free_sackblock(scb, sb); 472 --scb->nblocks; 473 KASSERT(scb->nblocks > 0, 474 ("removed partial right: %d blocks left", scb->nblocks)); 475 } 476 return 0; 477 } 478 479 #ifdef DEBUG_SACK_BLOCKS 480 static void 481 tcp_sack_dump_blocks(const struct scoreboard *scb) 482 { 483 const struct sackblock *sb; 484 485 kprintf("%d blocks:", scb->nblocks); 486 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) 487 kprintf(" [%u, %u)", sb->sblk_start, sb->sblk_end); 488 kprintf("\n"); 489 } 490 #else 491 static __inline void 492 tcp_sack_dump_blocks(const struct scoreboard *scb) 493 { 494 } 495 #endif 496 497 /* 498 * Optimization to quickly determine which packets are lost. 499 */ 500 void 501 tcp_sack_update_lostseq(struct scoreboard *scb, tcp_seq snd_una, u_int maxseg, 502 int rxtthresh) 503 { 504 struct sackblock *sb; 505 int nsackblocks = 0; 506 int bytes_sacked = 0; 507 int rxtthresh_bytes; 508 509 if (tcp_do_rfc6675) 510 rxtthresh_bytes = (rxtthresh - 1) * maxseg; 511 else 512 rxtthresh_bytes = rxtthresh * maxseg; 513 514 sb = TAILQ_LAST(&scb->sackblocks, sackblock_list); 515 while (sb != NULL) { 516 ++nsackblocks; 517 bytes_sacked += sb->sblk_end - sb->sblk_start; 518 if (nsackblocks == rxtthresh || 519 bytes_sacked >= rxtthresh_bytes) { 520 scb->lostseq = sb->sblk_start; 521 return; 522 } 523 sb = TAILQ_PREV(sb, sackblock_list, sblk_list); 524 } 525 scb->lostseq = snd_una; 526 } 527 528 /* 529 * Return whether the given sequence number is considered lost. 530 */ 531 boolean_t 532 tcp_sack_islost(const struct scoreboard *scb, tcp_seq seqnum) 533 { 534 return SEQ_LT(seqnum, scb->lostseq); 535 } 536 537 /* 538 * True if at least "amount" has been SACKed. Used by Early Retransmit. 539 */ 540 boolean_t 541 tcp_sack_has_sacked(const struct scoreboard *scb, u_int amount) 542 { 543 const struct sackblock *sb; 544 int bytes_sacked = 0; 545 546 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) { 547 bytes_sacked += sb->sblk_end - sb->sblk_start; 548 if (bytes_sacked >= amount) 549 return TRUE; 550 } 551 return FALSE; 552 } 553 554 /* 555 * Number of bytes SACKed below seq. 556 */ 557 int 558 tcp_sack_bytes_below(const struct scoreboard *scb, tcp_seq seq) 559 { 560 const struct sackblock *sb; 561 int bytes_sacked = 0; 562 563 sb = TAILQ_FIRST(&scb->sackblocks); 564 while (sb && SEQ_GT(seq, sb->sblk_start)) { 565 bytes_sacked += seq_min(seq, sb->sblk_end) - sb->sblk_start; 566 sb = TAILQ_NEXT(sb, sblk_list); 567 } 568 return bytes_sacked; 569 } 570 571 /* 572 * Return estimate of the number of bytes outstanding in the network. 573 */ 574 uint32_t 575 tcp_sack_compute_pipe(const struct tcpcb *tp) 576 { 577 const struct scoreboard *scb = &tp->scb; 578 const struct sackblock *sb; 579 int nlost, nretransmitted; 580 tcp_seq end; 581 582 nlost = tp->snd_max - scb->lostseq; 583 nretransmitted = tp->rexmt_high - tp->snd_una; 584 585 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) { 586 if (SEQ_LT(sb->sblk_start, tp->rexmt_high)) { 587 end = seq_min(sb->sblk_end, tp->rexmt_high); 588 nretransmitted -= end - sb->sblk_start; 589 } 590 if (SEQ_GEQ(sb->sblk_start, scb->lostseq)) 591 nlost -= sb->sblk_end - sb->sblk_start; 592 } 593 594 return (nlost + nretransmitted); 595 } 596 597 /* 598 * Return the sequence number and length of the next segment to transmit 599 * when in Fast Recovery. 600 */ 601 boolean_t 602 tcp_sack_nextseg(struct tcpcb *tp, tcp_seq *nextrexmt, uint32_t *plen, 603 boolean_t *rescue) 604 { 605 struct scoreboard *scb = &tp->scb; 606 struct socket *so = tp->t_inpcb->inp_socket; 607 struct sackblock *sb; 608 const struct sackblock *lastblock = 609 TAILQ_LAST(&scb->sackblocks, sackblock_list); 610 tcp_seq torexmt; 611 long len, off, sendwin; 612 613 /* skip SACKed data */ 614 tcp_sack_skip_sacked(scb, &tp->rexmt_high); 615 616 /* Look for lost data. */ 617 torexmt = tp->rexmt_high; 618 *rescue = FALSE; 619 if (lastblock != NULL) { 620 if (SEQ_LT(torexmt, lastblock->sblk_end) && 621 tcp_sack_islost(scb, torexmt)) { 622 sendunsacked: 623 *nextrexmt = torexmt; 624 /* If the left-hand edge has been SACKed, pull it in. */ 625 if (sack_block_lookup(scb, torexmt + tp->t_maxseg, &sb)) 626 *plen = sb->sblk_start - torexmt; 627 else 628 *plen = tp->t_maxseg; 629 return TRUE; 630 } 631 } 632 633 /* See if unsent data available within send window. */ 634 off = tp->snd_max - tp->snd_una; 635 sendwin = min(tp->snd_wnd, tp->snd_bwnd); 636 len = (long) ulmin(so->so_snd.ssb_cc, sendwin) - off; 637 if (len > 0) { 638 *nextrexmt = tp->snd_max; /* Send new data. */ 639 *plen = tp->t_maxseg; 640 return TRUE; 641 } 642 643 /* We're less certain this data has been lost. */ 644 if (lastblock != NULL && SEQ_LT(torexmt, lastblock->sblk_end)) 645 goto sendunsacked; 646 647 /* Rescue retransmission */ 648 if (tcp_do_rescuesack || tcp_do_rfc6675) { 649 tcpstat.tcps_sackrescue_try++; 650 if (tp->sack_flags & TSACK_F_SACKRESCUED) { 651 if (!tcp_aggressive_rescuesack) 652 return FALSE; 653 654 /* 655 * Aggressive variant of the rescue retransmission. 656 * 657 * The idea of the rescue retransmission is to sustain 658 * the ACK clock thus to avoid timeout retransmission. 659 * 660 * Under some situations, the conservative approach 661 * suggested in the draft 662 * http://tools.ietf.org/html/ 663 * draft-nishida-tcpm-rescue-retransmission-00 664 * could not sustain ACK clock, since it only allows 665 * one rescue retransmission before a cumulative ACK 666 * covers the segement transmitted by rescue 667 * retransmission. 668 * 669 * We try to locate the next unSACKed segment which 670 * follows the previously sent rescue segment. If 671 * there is no such segment, we loop back to the first 672 * unacknowledged segment. 673 */ 674 675 /* 676 * Skip SACKed data, but here we follow 677 * the last transmitted rescue segment. 678 */ 679 torexmt = tp->rexmt_rescue; 680 tcp_sack_skip_sacked(scb, &torexmt); 681 } 682 if (torexmt == tp->snd_max) { 683 /* Nothing left to retransmit; restart */ 684 torexmt = tp->snd_una; 685 } 686 *rescue = TRUE; 687 goto sendunsacked; 688 } else if (tcp_do_smartsack && lastblock == NULL) { 689 tcpstat.tcps_sackrescue_try++; 690 *rescue = TRUE; 691 goto sendunsacked; 692 } 693 694 return FALSE; 695 } 696 697 /* 698 * Return the next sequence number higher than "*prexmt" that has 699 * not been SACKed. 700 */ 701 void 702 tcp_sack_skip_sacked(struct scoreboard *scb, tcp_seq *prexmt) 703 { 704 struct sackblock *sb; 705 706 /* skip SACKed data */ 707 if (sack_block_lookup(scb, *prexmt, &sb)) 708 *prexmt = sb->sblk_end; 709 } 710 711 /* 712 * The length of the first amount of unSACKed data 713 */ 714 uint32_t 715 tcp_sack_first_unsacked_len(const struct tcpcb *tp) 716 { 717 const struct sackblock *sb; 718 719 sb = TAILQ_FIRST(&tp->scb.sackblocks); 720 if (sb == NULL) 721 return tp->t_maxseg; 722 723 KASSERT(SEQ_LT(tp->snd_una, sb->sblk_start), 724 ("invalid sb start %u, snd_una %u", 725 sb->sblk_start, tp->snd_una)); 726 return (sb->sblk_start - tp->snd_una); 727 } 728 729 #ifdef later 730 void 731 tcp_sack_save_scoreboard(struct scoreboard *scb) 732 { 733 struct scoreboard *scb = &tp->scb; 734 735 scb->sackblocks_prev = scb->sackblocks; 736 TAILQ_INIT(&scb->sackblocks); 737 } 738 739 void 740 tcp_sack_revert_scoreboard(struct scoreboard *scb, tcp_seq snd_una, 741 u_int maxseg) 742 { 743 struct sackblock *sb; 744 745 scb->sackblocks = scb->sackblocks_prev; 746 scb->nblocks = 0; 747 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) 748 ++scb->nblocks; 749 tcp_sack_ack_blocks(scb, snd_una); 750 scb->lastfound = NULL; 751 } 752 #endif 753 754 #ifdef DEBUG_SACK_HISTORY 755 static void 756 tcp_sack_dump_history(const char *msg, const struct tcpcb *tp) 757 { 758 int i; 759 static int ndumped; 760 761 /* only need a couple of these to debug most problems */ 762 if (++ndumped > 900) 763 return; 764 765 kprintf("%s:\tnsackhistory %d: ", msg, tp->nsackhistory); 766 for (i = 0; i < tp->nsackhistory; ++i) 767 kprintf("[%u, %u) ", tp->sackhistory[i].rblk_start, 768 tp->sackhistory[i].rblk_end); 769 kprintf("\n"); 770 } 771 #else 772 static __inline void 773 tcp_sack_dump_history(const char *msg, const struct tcpcb *tp) 774 { 775 } 776 #endif 777 778 /* 779 * Remove old SACK blocks from the SACK history that have already been ACKed. 780 */ 781 static void 782 tcp_sack_ack_history(struct tcpcb *tp) 783 { 784 int i, nblocks, openslot; 785 786 tcp_sack_dump_history("before tcp_sack_ack_history", tp); 787 nblocks = tp->nsackhistory; 788 for (i = openslot = 0; i < nblocks; ++i) { 789 if (SEQ_LEQ(tp->sackhistory[i].rblk_end, tp->rcv_nxt)) { 790 --tp->nsackhistory; 791 continue; 792 } 793 if (SEQ_LT(tp->sackhistory[i].rblk_start, tp->rcv_nxt)) 794 tp->sackhistory[i].rblk_start = tp->rcv_nxt; 795 if (i == openslot) 796 ++openslot; 797 else 798 tp->sackhistory[openslot++] = tp->sackhistory[i]; 799 } 800 tcp_sack_dump_history("after tcp_sack_ack_history", tp); 801 KASSERT(openslot == tp->nsackhistory, 802 ("tcp_sack_ack_history miscounted: %d != %d", 803 openslot, tp->nsackhistory)); 804 } 805 806 /* 807 * Add or merge newblock into reported history. 808 * Also remove or update SACK blocks that will be acked. 809 */ 810 static void 811 tcp_sack_update_reported_history(struct tcpcb *tp, tcp_seq start, tcp_seq end) 812 { 813 struct raw_sackblock copy[MAX_SACK_REPORT_BLOCKS]; 814 int i, cindex; 815 816 tcp_sack_dump_history("before tcp_sack_update_reported_history", tp); 817 /* 818 * Six cases: 819 * 0) no overlap 820 * 1) newblock == oldblock 821 * 2) oldblock contains newblock 822 * 3) newblock contains oldblock 823 * 4) tail of oldblock overlaps or abuts start of newblock 824 * 5) tail of newblock overlaps or abuts head of oldblock 825 */ 826 for (i = cindex = 0; i < tp->nsackhistory; ++i) { 827 struct raw_sackblock *oldblock = &tp->sackhistory[i]; 828 tcp_seq old_start = oldblock->rblk_start; 829 tcp_seq old_end = oldblock->rblk_end; 830 831 if (SEQ_LT(end, old_start) || SEQ_GT(start, old_end)) { 832 /* Case 0: no overlap. Copy old block. */ 833 copy[cindex++] = *oldblock; 834 continue; 835 } 836 837 if (SEQ_GEQ(start, old_start) && SEQ_LEQ(end, old_end)) { 838 /* Cases 1 & 2. Move block to front of history. */ 839 int j; 840 841 start = old_start; 842 end = old_end; 843 /* no need to check rest of blocks */ 844 for (j = i + 1; j < tp->nsackhistory; ++j) 845 copy[cindex++] = tp->sackhistory[j]; 846 break; 847 } 848 849 if (SEQ_GEQ(old_end, start) && SEQ_LT(old_start, start)) { 850 /* Case 4: extend start of new block. */ 851 start = old_start; 852 } else if (SEQ_GEQ(end, old_start) && SEQ_GT(old_end, end)) { 853 /* Case 5: extend end of new block */ 854 end = old_end; 855 } else { 856 /* Case 3. Delete old block by not copying it. */ 857 KASSERT(SEQ_LEQ(start, old_start) && 858 SEQ_GEQ(end, old_end), 859 ("bad logic: old [%u, %u), new [%u, %u)", 860 old_start, old_end, start, end)); 861 } 862 } 863 864 /* insert new block */ 865 tp->sackhistory[0].rblk_start = start; 866 tp->sackhistory[0].rblk_end = end; 867 cindex = min(cindex, MAX_SACK_REPORT_BLOCKS - 1); 868 for (i = 0; i < cindex; ++i) 869 tp->sackhistory[i + 1] = copy[i]; 870 tp->nsackhistory = cindex + 1; 871 tcp_sack_dump_history("after tcp_sack_update_reported_history", tp); 872 } 873 874 /* 875 * Fill in SACK report to return to data sender. 876 */ 877 void 878 tcp_sack_fill_report(struct tcpcb *tp, u_char *opt, u_int *plen) 879 { 880 u_int optlen = *plen; 881 uint32_t *lp = (uint32_t *)(opt + optlen); 882 uint32_t *olp; 883 tcp_seq hstart = tp->rcv_nxt, hend; 884 int nblocks; 885 886 KASSERT(TCP_MAXOLEN - optlen >= 887 TCPOLEN_SACK_ALIGNED + TCPOLEN_SACK_BLOCK, 888 ("no room for SACK header and one block: optlen %d", optlen)); 889 890 if (tp->sack_flags & TSACK_F_DUPSEG) 891 tcpstat.tcps_snddsackopt++; 892 else 893 tcpstat.tcps_sndsackopt++; 894 895 olp = lp++; 896 optlen += TCPOLEN_SACK_ALIGNED; 897 898 tcp_sack_ack_history(tp); 899 if (tp->reportblk.rblk_start != tp->reportblk.rblk_end) { 900 *lp++ = htonl(tp->reportblk.rblk_start); 901 *lp++ = htonl(tp->reportblk.rblk_end); 902 optlen += TCPOLEN_SACK_BLOCK; 903 hstart = tp->reportblk.rblk_start; 904 hend = tp->reportblk.rblk_end; 905 if (tp->sack_flags & TSACK_F_ENCLOSESEG) { 906 KASSERT(TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK, 907 ("no room for enclosing SACK block: oplen %d", 908 optlen)); 909 *lp++ = htonl(tp->encloseblk.rblk_start); 910 *lp++ = htonl(tp->encloseblk.rblk_end); 911 optlen += TCPOLEN_SACK_BLOCK; 912 hstart = tp->encloseblk.rblk_start; 913 hend = tp->encloseblk.rblk_end; 914 } 915 if (SEQ_GT(hstart, tp->rcv_nxt)) 916 tcp_sack_update_reported_history(tp, hstart, hend); 917 } 918 if (tcp_do_smartsack && (tp->sack_flags & TSACK_F_SACKLEFT)) { 919 /* Fill in from left! Walk re-assembly queue. */ 920 struct tseg_qent *q; 921 922 q = TAILQ_FIRST(&tp->t_segq); 923 while (q != NULL && 924 TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) { 925 *lp++ = htonl(q->tqe_th->th_seq); 926 *lp++ = htonl(TCP_SACK_BLKEND( 927 q->tqe_th->th_seq + q->tqe_len, 928 q->tqe_th->th_flags)); 929 optlen += TCPOLEN_SACK_BLOCK; 930 q = TAILQ_NEXT(q, tqe_q); 931 } 932 } else { 933 int n = 0; 934 935 /* Fill in SACK blocks from right side. */ 936 while (n < tp->nsackhistory && 937 TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) { 938 if (tp->sackhistory[n].rblk_start != hstart) { 939 *lp++ = htonl(tp->sackhistory[n].rblk_start); 940 *lp++ = htonl(tp->sackhistory[n].rblk_end); 941 optlen += TCPOLEN_SACK_BLOCK; 942 } 943 ++n; 944 } 945 } 946 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 947 tp->sack_flags &= 948 ~(TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT); 949 nblocks = (lp - olp - 1) / 2; 950 *olp = htonl(TCPOPT_SACK_ALIGNED | 951 (TCPOLEN_SACK + nblocks * TCPOLEN_SACK_BLOCK)); 952 *plen = optlen; 953 } 954