sys/netinet/tcp_sack.c

   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 }