tcp_sack.c revision 190948
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 4. Neither the name of the University nor the names of its contributors 15 * may be used to endorse or promote products derived from this software 16 * without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95 31 */ 32 33/*- 34 * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995 35 * 36 * NRL grants permission for redistribution and use in source and binary 37 * forms, with or without modification, of the software and documentation 38 * created at NRL provided that the following conditions are met: 39 * 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. All advertising materials mentioning features or use of this software 46 * must display the following acknowledgements: 47 * This product includes software developed by the University of 48 * California, Berkeley and its contributors. 49 * This product includes software developed at the Information 50 * Technology Division, US Naval Research Laboratory. 51 * 4. Neither the name of the NRL nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 56 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 57 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 58 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 59 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 60 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 61 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 62 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 63 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 64 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 65 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 66 * 67 * The views and conclusions contained in the software and documentation 68 * are those of the authors and should not be interpreted as representing 69 * official policies, either expressed or implied, of the US Naval 70 * Research Laboratory (NRL). 71 */ 72 73#include <sys/cdefs.h> 74__FBSDID("$FreeBSD: head/sys/netinet/tcp_sack.c 190948 2009-04-11 22:07:19Z rwatson $"); 75 76#include "opt_inet.h" 77#include "opt_inet6.h" 78#include "opt_tcpdebug.h" 79 80#include <sys/param.h> 81#include <sys/systm.h> 82#include <sys/kernel.h> 83#include <sys/sysctl.h> 84#include <sys/malloc.h> 85#include <sys/mbuf.h> 86#include <sys/proc.h> /* for proc0 declaration */ 87#include <sys/protosw.h> 88#include <sys/socket.h> 89#include <sys/socketvar.h> 90#include <sys/syslog.h> 91#include <sys/systm.h> 92#include <sys/vimage.h> 93 94#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 95 96#include <vm/uma.h> 97 98#include <net/if.h> 99#include <net/route.h> 100 101#include <netinet/in.h> 102#include <netinet/in_systm.h> 103#include <netinet/ip.h> 104#include <netinet/in_var.h> 105#include <netinet/in_pcb.h> 106#include <netinet/ip_var.h> 107#include <netinet/ip6.h> 108#include <netinet/icmp6.h> 109#include <netinet6/nd6.h> 110#include <netinet6/ip6_var.h> 111#include <netinet6/in6_pcb.h> 112#include <netinet/tcp.h> 113#include <netinet/tcp_fsm.h> 114#include <netinet/tcp_seq.h> 115#include <netinet/tcp_timer.h> 116#include <netinet/tcp_var.h> 117#include <netinet6/tcp6_var.h> 118#include <netinet/tcpip.h> 119#ifdef TCPDEBUG 120#include <netinet/tcp_debug.h> 121#endif /* TCPDEBUG */ 122#include <netinet/vinet.h> 123 124#include <machine/in_cksum.h> 125 126#ifdef VIMAGE_GLOBALS 127extern struct uma_zone *sack_hole_zone; 128int tcp_do_sack; 129int tcp_sack_maxholes; 130int tcp_sack_globalmaxholes; 131int tcp_sack_globalholes; 132#endif 133 134SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 135SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, enable, 136 CTLFLAG_RW, tcp_do_sack, 0, "Enable/Disable TCP SACK support"); 137 138SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, maxholes, 139 CTLFLAG_RW, tcp_sack_maxholes, 0, 140 "Maximum number of TCP SACK holes allowed per connection"); 141 142SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, globalmaxholes, 143 CTLFLAG_RW, tcp_sack_globalmaxholes, 0, 144 "Global maximum number of TCP SACK holes"); 145 146SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, globalholes, 147 CTLFLAG_RD, tcp_sack_globalholes, 0, 148 "Global number of TCP SACK holes currently allocated"); 149 150/* 151 * This function is called upon receipt of new valid data (while not in 152 * header prediction mode), and it updates the ordered list of sacks. 153 */ 154void 155tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) 156{ 157 /* 158 * First reported block MUST be the most recent one. Subsequent 159 * blocks SHOULD be in the order in which they arrived at the 160 * receiver. These two conditions make the implementation fully 161 * compliant with RFC 2018. 162 */ 163 struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; 164 int num_head, num_saved, i; 165 166 INP_WLOCK_ASSERT(tp->t_inpcb); 167 168 /* Check arguments. */ 169 KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end")); 170 171 /* SACK block for the received segment. */ 172 head_blk.start = rcv_start; 173 head_blk.end = rcv_end; 174 175 /* 176 * Merge updated SACK blocks into head_blk, and save unchanged SACK 177 * blocks into saved_blks[]. num_saved will have the number of the 178 * saved SACK blocks. 179 */ 180 num_saved = 0; 181 for (i = 0; i < tp->rcv_numsacks; i++) { 182 tcp_seq start = tp->sackblks[i].start; 183 tcp_seq end = tp->sackblks[i].end; 184 if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { 185 /* 186 * Discard this SACK block. 187 */ 188 } else if (SEQ_LEQ(head_blk.start, end) && 189 SEQ_GEQ(head_blk.end, start)) { 190 /* 191 * Merge this SACK block into head_blk. This SACK 192 * block itself will be discarded. 193 */ 194 if (SEQ_GT(head_blk.start, start)) 195 head_blk.start = start; 196 if (SEQ_LT(head_blk.end, end)) 197 head_blk.end = end; 198 } else { 199 /* 200 * Save this SACK block. 201 */ 202 saved_blks[num_saved].start = start; 203 saved_blks[num_saved].end = end; 204 num_saved++; 205 } 206 } 207 208 /* 209 * Update SACK list in tp->sackblks[]. 210 */ 211 num_head = 0; 212 if (SEQ_GT(head_blk.start, tp->rcv_nxt)) { 213 /* 214 * The received data segment is an out-of-order segment. Put 215 * head_blk at the top of SACK list. 216 */ 217 tp->sackblks[0] = head_blk; 218 num_head = 1; 219 /* 220 * If the number of saved SACK blocks exceeds its limit, 221 * discard the last SACK block. 222 */ 223 if (num_saved >= MAX_SACK_BLKS) 224 num_saved--; 225 } 226 if (num_saved > 0) { 227 /* 228 * Copy the saved SACK blocks back. 229 */ 230 bcopy(saved_blks, &tp->sackblks[num_head], 231 sizeof(struct sackblk) * num_saved); 232 } 233 234 /* Save the number of SACK blocks. */ 235 tp->rcv_numsacks = num_head + num_saved; 236} 237 238/* 239 * Delete all receiver-side SACK information. 240 */ 241void 242tcp_clean_sackreport(struct tcpcb *tp) 243{ 244 int i; 245 246 INP_WLOCK_ASSERT(tp->t_inpcb); 247 tp->rcv_numsacks = 0; 248 for (i = 0; i < MAX_SACK_BLKS; i++) 249 tp->sackblks[i].start = tp->sackblks[i].end=0; 250} 251 252/* 253 * Allocate struct sackhole. 254 */ 255static struct sackhole * 256tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) 257{ 258 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 259 struct sackhole *hole; 260 261 if (tp->snd_numholes >= V_tcp_sack_maxholes || 262 V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) { 263 TCPSTAT_INC(tcps_sack_sboverflow); 264 return NULL; 265 } 266 267 hole = (struct sackhole *)uma_zalloc(V_sack_hole_zone, M_NOWAIT); 268 if (hole == NULL) 269 return NULL; 270 271 hole->start = start; 272 hole->end = end; 273 hole->rxmit = start; 274 275 tp->snd_numholes++; 276 V_tcp_sack_globalholes++; 277 278 return hole; 279} 280 281/* 282 * Free struct sackhole. 283 */ 284static void 285tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) 286{ 287 INIT_VNET_INET(tp->t_vnet); 288 289 uma_zfree(V_sack_hole_zone, hole); 290 291 tp->snd_numholes--; 292 V_tcp_sack_globalholes--; 293 294 KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0")); 295 KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0")); 296} 297 298/* 299 * Insert new SACK hole into scoreboard. 300 */ 301static struct sackhole * 302tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end, 303 struct sackhole *after) 304{ 305 struct sackhole *hole; 306 307 /* Allocate a new SACK hole. */ 308 hole = tcp_sackhole_alloc(tp, start, end); 309 if (hole == NULL) 310 return NULL; 311 312 /* Insert the new SACK hole into scoreboard. */ 313 if (after != NULL) 314 TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink); 315 else 316 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink); 317 318 /* Update SACK hint. */ 319 if (tp->sackhint.nexthole == NULL) 320 tp->sackhint.nexthole = hole; 321 322 return hole; 323} 324 325/* 326 * Remove SACK hole from scoreboard. 327 */ 328static void 329tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole) 330{ 331 332 /* Update SACK hint. */ 333 if (tp->sackhint.nexthole == hole) 334 tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink); 335 336 /* Remove this SACK hole. */ 337 TAILQ_REMOVE(&tp->snd_holes, hole, scblink); 338 339 /* Free this SACK hole. */ 340 tcp_sackhole_free(tp, hole); 341} 342 343/* 344 * Process cumulative ACK and the TCP SACK option to update the scoreboard. 345 * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of 346 * the sequence space). 347 */ 348void 349tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) 350{ 351 struct sackhole *cur, *temp; 352 struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; 353 int i, j, num_sack_blks; 354 355 INP_WLOCK_ASSERT(tp->t_inpcb); 356 357 num_sack_blks = 0; 358 /* 359 * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, 360 * treat [SND.UNA, SEG.ACK) as if it is a SACK block. 361 */ 362 if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { 363 sack_blocks[num_sack_blks].start = tp->snd_una; 364 sack_blocks[num_sack_blks++].end = th_ack; 365 } 366 /* 367 * Append received valid SACK blocks to sack_blocks[], but only if we 368 * received new blocks from the other side. 369 */ 370 if (to->to_flags & TOF_SACK) { 371 for (i = 0; i < to->to_nsacks; i++) { 372 bcopy((to->to_sacks + i * TCPOLEN_SACK), 373 &sack, sizeof(sack)); 374 sack.start = ntohl(sack.start); 375 sack.end = ntohl(sack.end); 376 if (SEQ_GT(sack.end, sack.start) && 377 SEQ_GT(sack.start, tp->snd_una) && 378 SEQ_GT(sack.start, th_ack) && 379 SEQ_LT(sack.start, tp->snd_max) && 380 SEQ_GT(sack.end, tp->snd_una) && 381 SEQ_LEQ(sack.end, tp->snd_max)) 382 sack_blocks[num_sack_blks++] = sack; 383 } 384 } 385 /* 386 * Return if SND.UNA is not advanced and no valid SACK block is 387 * received. 388 */ 389 if (num_sack_blks == 0) 390 return; 391 392 /* 393 * Sort the SACK blocks so we can update the scoreboard with just one 394 * pass. The overhead of sorting upto 4+1 elements is less than 395 * making upto 4+1 passes over the scoreboard. 396 */ 397 for (i = 0; i < num_sack_blks; i++) { 398 for (j = i + 1; j < num_sack_blks; j++) { 399 if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { 400 sack = sack_blocks[i]; 401 sack_blocks[i] = sack_blocks[j]; 402 sack_blocks[j] = sack; 403 } 404 } 405 } 406 if (TAILQ_EMPTY(&tp->snd_holes)) 407 /* 408 * Empty scoreboard. Need to initialize snd_fack (it may be 409 * uninitialized or have a bogus value). Scoreboard holes 410 * (from the sack blocks received) are created later below 411 * (in the logic that adds holes to the tail of the 412 * scoreboard). 413 */ 414 tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); 415 /* 416 * In the while-loop below, incoming SACK blocks (sack_blocks[]) and 417 * SACK holes (snd_holes) are traversed from their tails with just 418 * one pass in order to reduce the number of compares especially when 419 * the bandwidth-delay product is large. 420 * 421 * Note: Typically, in the first RTT of SACK recovery, the highest 422 * three or four SACK blocks with the same ack number are received. 423 * In the second RTT, if retransmitted data segments are not lost, 424 * the highest three or four SACK blocks with ack number advancing 425 * are received. 426 */ 427 sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ 428 if (SEQ_LT(tp->snd_fack, sblkp->start)) { 429 /* 430 * The highest SACK block is beyond fack. Append new SACK 431 * hole at the tail. If the second or later highest SACK 432 * blocks are also beyond the current fack, they will be 433 * inserted by way of hole splitting in the while-loop below. 434 */ 435 temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); 436 if (temp != NULL) { 437 tp->snd_fack = sblkp->end; 438 /* Go to the previous sack block. */ 439 sblkp--; 440 } else { 441 /* 442 * We failed to add a new hole based on the current 443 * sack block. Skip over all the sack blocks that 444 * fall completely to the right of snd_fack and 445 * proceed to trim the scoreboard based on the 446 * remaining sack blocks. This also trims the 447 * scoreboard for th_ack (which is sack_blocks[0]). 448 */ 449 while (sblkp >= sack_blocks && 450 SEQ_LT(tp->snd_fack, sblkp->start)) 451 sblkp--; 452 if (sblkp >= sack_blocks && 453 SEQ_LT(tp->snd_fack, sblkp->end)) 454 tp->snd_fack = sblkp->end; 455 } 456 } else if (SEQ_LT(tp->snd_fack, sblkp->end)) 457 /* fack is advanced. */ 458 tp->snd_fack = sblkp->end; 459 /* We must have at least one SACK hole in scoreboard. */ 460 KASSERT(!TAILQ_EMPTY(&tp->snd_holes), 461 ("SACK scoreboard must not be empty")); 462 cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */ 463 /* 464 * Since the incoming sack blocks are sorted, we can process them 465 * making one sweep of the scoreboard. 466 */ 467 while (sblkp >= sack_blocks && cur != NULL) { 468 if (SEQ_GEQ(sblkp->start, cur->end)) { 469 /* 470 * SACKs data beyond the current hole. Go to the 471 * previous sack block. 472 */ 473 sblkp--; 474 continue; 475 } 476 if (SEQ_LEQ(sblkp->end, cur->start)) { 477 /* 478 * SACKs data before the current hole. Go to the 479 * previous hole. 480 */ 481 cur = TAILQ_PREV(cur, sackhole_head, scblink); 482 continue; 483 } 484 tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start); 485 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 486 ("sackhint bytes rtx >= 0")); 487 if (SEQ_LEQ(sblkp->start, cur->start)) { 488 /* Data acks at least the beginning of hole. */ 489 if (SEQ_GEQ(sblkp->end, cur->end)) { 490 /* Acks entire hole, so delete hole. */ 491 temp = cur; 492 cur = TAILQ_PREV(cur, sackhole_head, scblink); 493 tcp_sackhole_remove(tp, temp); 494 /* 495 * The sack block may ack all or part of the 496 * next hole too, so continue onto the next 497 * hole. 498 */ 499 continue; 500 } else { 501 /* Move start of hole forward. */ 502 cur->start = sblkp->end; 503 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 504 } 505 } else { 506 /* Data acks at least the end of hole. */ 507 if (SEQ_GEQ(sblkp->end, cur->end)) { 508 /* Move end of hole backward. */ 509 cur->end = sblkp->start; 510 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 511 } else { 512 /* 513 * ACKs some data in middle of a hole; need 514 * to split current hole 515 */ 516 temp = tcp_sackhole_insert(tp, sblkp->end, 517 cur->end, cur); 518 if (temp != NULL) { 519 if (SEQ_GT(cur->rxmit, temp->rxmit)) { 520 temp->rxmit = cur->rxmit; 521 tp->sackhint.sack_bytes_rexmit 522 += (temp->rxmit 523 - temp->start); 524 } 525 cur->end = sblkp->start; 526 cur->rxmit = SEQ_MIN(cur->rxmit, 527 cur->end); 528 } 529 } 530 } 531 tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start); 532 /* 533 * Testing sblkp->start against cur->start tells us whether 534 * we're done with the sack block or the sack hole. 535 * Accordingly, we advance one or the other. 536 */ 537 if (SEQ_LEQ(sblkp->start, cur->start)) 538 cur = TAILQ_PREV(cur, sackhole_head, scblink); 539 else 540 sblkp--; 541 } 542} 543 544/* 545 * Free all SACK holes to clear the scoreboard. 546 */ 547void 548tcp_free_sackholes(struct tcpcb *tp) 549{ 550 struct sackhole *q; 551 552 INP_WLOCK_ASSERT(tp->t_inpcb); 553 while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) 554 tcp_sackhole_remove(tp, q); 555 tp->sackhint.sack_bytes_rexmit = 0; 556 557 KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); 558 KASSERT(tp->sackhint.nexthole == NULL, 559 ("tp->sackhint.nexthole == NULL")); 560} 561 562/* 563 * Partial ack handling within a sack recovery episode. Keeping this very 564 * simple for now. When a partial ack is received, force snd_cwnd to a value 565 * that will allow the sender to transmit no more than 2 segments. If 566 * necessary, a better scheme can be adopted at a later point, but for now, 567 * the goal is to prevent the sender from bursting a large amount of data in 568 * the midst of sack recovery. 569 */ 570void 571tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) 572{ 573 int num_segs = 1; 574 575 INP_WLOCK_ASSERT(tp->t_inpcb); 576 tcp_timer_activate(tp, TT_REXMT, 0); 577 tp->t_rtttime = 0; 578 /* Send one or 2 segments based on how much new data was acked. */ 579 if (((th->th_ack - tp->snd_una) / tp->t_maxseg) > 2) 580 num_segs = 2; 581 tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + 582 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_maxseg); 583 if (tp->snd_cwnd > tp->snd_ssthresh) 584 tp->snd_cwnd = tp->snd_ssthresh; 585 tp->t_flags |= TF_ACKNOW; 586 (void) tcp_output(tp); 587} 588 589#if 0 590/* 591 * Debug version of tcp_sack_output() that walks the scoreboard. Used for 592 * now to sanity check the hint. 593 */ 594static struct sackhole * 595tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) 596{ 597 struct sackhole *p; 598 599 INP_WLOCK_ASSERT(tp->t_inpcb); 600 *sack_bytes_rexmt = 0; 601 TAILQ_FOREACH(p, &tp->snd_holes, scblink) { 602 if (SEQ_LT(p->rxmit, p->end)) { 603 if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ 604 continue; 605 } 606 *sack_bytes_rexmt += (p->rxmit - p->start); 607 break; 608 } 609 *sack_bytes_rexmt += (p->rxmit - p->start); 610 } 611 return (p); 612} 613#endif 614 615/* 616 * Returns the next hole to retransmit and the number of retransmitted bytes 617 * from the scoreboard. We store both the next hole and the number of 618 * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK 619 * reception). This avoids scoreboard traversals completely. 620 * 621 * The loop here will traverse *at most* one link. Here's the argument. For 622 * the loop to traverse more than 1 link before finding the next hole to 623 * retransmit, we would need to have at least 1 node following the current 624 * hint with (rxmit == end). But, for all holes following the current hint, 625 * (start == rxmit), since we have not yet retransmitted from them. 626 * Therefore, in order to traverse more 1 link in the loop below, we need to 627 * have at least one node following the current hint with (start == rxmit == 628 * end). But that can't happen, (start == end) means that all the data in 629 * that hole has been sacked, in which case, the hole would have been removed 630 * from the scoreboard. 631 */ 632struct sackhole * 633tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 634{ 635 struct sackhole *hole = NULL; 636 637 INP_WLOCK_ASSERT(tp->t_inpcb); 638 *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; 639 hole = tp->sackhint.nexthole; 640 if (hole == NULL || SEQ_LT(hole->rxmit, hole->end)) 641 goto out; 642 while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) { 643 if (SEQ_LT(hole->rxmit, hole->end)) { 644 tp->sackhint.nexthole = hole; 645 break; 646 } 647 } 648out: 649 return (hole); 650} 651 652/* 653 * After a timeout, the SACK list may be rebuilt. This SACK information 654 * should be used to avoid retransmitting SACKed data. This function 655 * traverses the SACK list to see if snd_nxt should be moved forward. 656 */ 657void 658tcp_sack_adjust(struct tcpcb *tp) 659{ 660 struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); 661 662 INP_WLOCK_ASSERT(tp->t_inpcb); 663 if (cur == NULL) 664 return; /* No holes */ 665 if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) 666 return; /* We're already beyond any SACKed blocks */ 667 /*- 668 * Two cases for which we want to advance snd_nxt: 669 * i) snd_nxt lies between end of one hole and beginning of another 670 * ii) snd_nxt lies between end of last hole and snd_fack 671 */ 672 while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { 673 if (SEQ_LT(tp->snd_nxt, cur->end)) 674 return; 675 if (SEQ_GEQ(tp->snd_nxt, p->start)) 676 cur = p; 677 else { 678 tp->snd_nxt = p->start; 679 return; 680 } 681 } 682 if (SEQ_LT(tp->snd_nxt, cur->end)) 683 return; 684 tp->snd_nxt = tp->snd_fack; 685} 686