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("$FreeBSD$"); 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 */ 122 123 #include <machine/in_cksum.h> 124 125 extern struct uma_zone *sack_hole_zone; 126 127 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 128 int tcp_do_sack = 1; 129 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW, 130 &tcp_do_sack, 0, "Enable/Disable TCP SACK support"); 131 TUNABLE_INT("net.inet.tcp.sack.enable", &tcp_do_sack); 132 133 static int tcp_sack_maxholes = 128; 134 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_RW, 135 &tcp_sack_maxholes, 0, 136 "Maximum number of TCP SACK holes allowed per connection"); 137 138 static int tcp_sack_globalmaxholes = 65536; 139 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_RW, 140 &tcp_sack_globalmaxholes, 0, 141 "Global maximum number of TCP SACK holes"); 142 143 static int tcp_sack_globalholes = 0; 144 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_RD, 145 &tcp_sack_globalholes, 0, 146 "Global number of TCP SACK holes currently allocated"); 147 148 /* 149 * This function is called upon receipt of new valid data (while not in 150 * header prediction mode), and it updates the ordered list of sacks. 151 */ 152 void 153 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) 154 { 155 /* 156 * First reported block MUST be the most recent one. Subsequent 157 * blocks SHOULD be in the order in which they arrived at the 158 * receiver. These two conditions make the implementation fully 159 * compliant with RFC 2018. 160 */ 161 struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; 162 int num_head, num_saved, i; 163 164 INP_WLOCK_ASSERT(tp->t_inpcb); 165 166 /* Check arguments. */ 167 KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end")); 168 169 /* SACK block for the received segment. */ 170 head_blk.start = rcv_start; 171 head_blk.end = rcv_end; 172 173 /* 174 * Merge updated SACK blocks into head_blk, and save unchanged SACK 175 * blocks into saved_blks[]. num_saved will have the number of the 176 * saved SACK blocks. 177 */ 178 num_saved = 0; 179 for (i = 0; i < tp->rcv_numsacks; i++) { 180 tcp_seq start = tp->sackblks[i].start; 181 tcp_seq end = tp->sackblks[i].end; 182 if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { 183 /* 184 * Discard this SACK block. 185 */ 186 } else if (SEQ_LEQ(head_blk.start, end) && 187 SEQ_GEQ(head_blk.end, start)) { 188 /* 189 * Merge this SACK block into head_blk. This SACK 190 * block itself will be discarded. 191 */ 192 if (SEQ_GT(head_blk.start, start)) 193 head_blk.start = start; 194 if (SEQ_LT(head_blk.end, end)) 195 head_blk.end = end; 196 } else { 197 /* 198 * Save this SACK block. 199 */ 200 saved_blks[num_saved].start = start; 201 saved_blks[num_saved].end = end; 202 num_saved++; 203 } 204 } 205 206 /* 207 * Update SACK list in tp->sackblks[]. 208 */ 209 num_head = 0; 210 if (SEQ_GT(head_blk.start, tp->rcv_nxt)) { 211 /* 212 * The received data segment is an out-of-order segment. Put 213 * head_blk at the top of SACK list. 214 */ 215 tp->sackblks[0] = head_blk; 216 num_head = 1; 217 /* 218 * If the number of saved SACK blocks exceeds its limit, 219 * discard the last SACK block. 220 */ 221 if (num_saved >= MAX_SACK_BLKS) 222 num_saved--; 223 } 224 if (num_saved > 0) { 225 /* 226 * Copy the saved SACK blocks back. 227 */ 228 bcopy(saved_blks, &tp->sackblks[num_head], 229 sizeof(struct sackblk) * num_saved); 230 } 231 232 /* Save the number of SACK blocks. */ 233 tp->rcv_numsacks = num_head + num_saved; 234 } 235 236 /* 237 * Delete all receiver-side SACK information. 238 */ 239 void 240 tcp_clean_sackreport(struct tcpcb *tp) 241 { 242 int i; 243 244 INP_WLOCK_ASSERT(tp->t_inpcb); 245 tp->rcv_numsacks = 0; 246 for (i = 0; i < MAX_SACK_BLKS; i++) 247 tp->sackblks[i].start = tp->sackblks[i].end=0; 248 } 249 250 /* 251 * Allocate struct sackhole. 252 */ 253 static struct sackhole * 254 tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) 255 { 256 struct sackhole *hole; 257 258 if (tp->snd_numholes >= V_tcp_sack_maxholes || 259 V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) { 260 V_tcpstat.tcps_sack_sboverflow++; 261 return NULL; 262 } 263 264 hole = (struct sackhole *)uma_zalloc(sack_hole_zone, M_NOWAIT); 265 if (hole == NULL) 266 return NULL; 267 268 hole->start = start; 269 hole->end = end; 270 hole->rxmit = start; 271 272 tp->snd_numholes++; 273 V_tcp_sack_globalholes++; 274 275 return hole; 276 } 277 278 /* 279 * Free struct sackhole. 280 */ 281 static void 282 tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) 283 { 284 285 uma_zfree(sack_hole_zone, hole); 286 287 tp->snd_numholes--; 288 V_tcp_sack_globalholes--; 289 290 KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0")); 291 KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0")); 292 } 293 294 /* 295 * Insert new SACK hole into scoreboard. 296 */ 297 static struct sackhole * 298 tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end, 299 struct sackhole *after) 300 { 301 struct sackhole *hole; 302 303 /* Allocate a new SACK hole. */ 304 hole = tcp_sackhole_alloc(tp, start, end); 305 if (hole == NULL) 306 return NULL; 307 308 /* Insert the new SACK hole into scoreboard. */ 309 if (after != NULL) 310 TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink); 311 else 312 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink); 313 314 /* Update SACK hint. */ 315 if (tp->sackhint.nexthole == NULL) 316 tp->sackhint.nexthole = hole; 317 318 return hole; 319 } 320 321 /* 322 * Remove SACK hole from scoreboard. 323 */ 324 static void 325 tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole) 326 { 327 328 /* Update SACK hint. */ 329 if (tp->sackhint.nexthole == hole) 330 tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink); 331 332 /* Remove this SACK hole. */ 333 TAILQ_REMOVE(&tp->snd_holes, hole, scblink); 334 335 /* Free this SACK hole. */ 336 tcp_sackhole_free(tp, hole); 337 } 338 339 /* 340 * Process cumulative ACK and the TCP SACK option to update the scoreboard. 341 * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of 342 * the sequence space). 343 */ 344 void 345 tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) 346 { 347 struct sackhole *cur, *temp; 348 struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; 349 int i, j, num_sack_blks; 350 351 INP_WLOCK_ASSERT(tp->t_inpcb); 352 353 num_sack_blks = 0; 354 /* 355 * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, 356 * treat [SND.UNA, SEG.ACK) as if it is a SACK block. 357 */ 358 if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { 359 sack_blocks[num_sack_blks].start = tp->snd_una; 360 sack_blocks[num_sack_blks++].end = th_ack; 361 } 362 /* 363 * Append received valid SACK blocks to sack_blocks[], but only if we 364 * received new blocks from the other side. 365 */ 366 if (to->to_flags & TOF_SACK) { 367 for (i = 0; i < to->to_nsacks; i++) { 368 bcopy((to->to_sacks + i * TCPOLEN_SACK), 369 &sack, sizeof(sack)); 370 sack.start = ntohl(sack.start); 371 sack.end = ntohl(sack.end); 372 if (SEQ_GT(sack.end, sack.start) && 373 SEQ_GT(sack.start, tp->snd_una) && 374 SEQ_GT(sack.start, th_ack) && 375 SEQ_LT(sack.start, tp->snd_max) && 376 SEQ_GT(sack.end, tp->snd_una) && 377 SEQ_LEQ(sack.end, tp->snd_max)) 378 sack_blocks[num_sack_blks++] = sack; 379 } 380 } 381 /* 382 * Return if SND.UNA is not advanced and no valid SACK block is 383 * received. 384 */ 385 if (num_sack_blks == 0) 386 return; 387 388 /* 389 * Sort the SACK blocks so we can update the scoreboard with just one 390 * pass. The overhead of sorting upto 4+1 elements is less than 391 * making upto 4+1 passes over the scoreboard. 392 */ 393 for (i = 0; i < num_sack_blks; i++) { 394 for (j = i + 1; j < num_sack_blks; j++) { 395 if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { 396 sack = sack_blocks[i]; 397 sack_blocks[i] = sack_blocks[j]; 398 sack_blocks[j] = sack; 399 } 400 } 401 } 402 if (TAILQ_EMPTY(&tp->snd_holes)) 403 /* 404 * Empty scoreboard. Need to initialize snd_fack (it may be 405 * uninitialized or have a bogus value). Scoreboard holes 406 * (from the sack blocks received) are created later below 407 * (in the logic that adds holes to the tail of the 408 * scoreboard). 409 */ 410 tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); 411 /* 412 * In the while-loop below, incoming SACK blocks (sack_blocks[]) and 413 * SACK holes (snd_holes) are traversed from their tails with just 414 * one pass in order to reduce the number of compares especially when 415 * the bandwidth-delay product is large. 416 * 417 * Note: Typically, in the first RTT of SACK recovery, the highest 418 * three or four SACK blocks with the same ack number are received. 419 * In the second RTT, if retransmitted data segments are not lost, 420 * the highest three or four SACK blocks with ack number advancing 421 * are received. 422 */ 423 sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ 424 if (SEQ_LT(tp->snd_fack, sblkp->start)) { 425 /* 426 * The highest SACK block is beyond fack. Append new SACK 427 * hole at the tail. If the second or later highest SACK 428 * blocks are also beyond the current fack, they will be 429 * inserted by way of hole splitting in the while-loop below. 430 */ 431 temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); 432 if (temp != NULL) { 433 tp->snd_fack = sblkp->end; 434 /* Go to the previous sack block. */ 435 sblkp--; 436 } else { 437 /* 438 * We failed to add a new hole based on the current 439 * sack block. Skip over all the sack blocks that 440 * fall completely to the right of snd_fack and 441 * proceed to trim the scoreboard based on the 442 * remaining sack blocks. This also trims the 443 * scoreboard for th_ack (which is sack_blocks[0]). 444 */ 445 while (sblkp >= sack_blocks && 446 SEQ_LT(tp->snd_fack, sblkp->start)) 447 sblkp--; 448 if (sblkp >= sack_blocks && 449 SEQ_LT(tp->snd_fack, sblkp->end)) 450 tp->snd_fack = sblkp->end; 451 } 452 } else if (SEQ_LT(tp->snd_fack, sblkp->end)) 453 /* fack is advanced. */ 454 tp->snd_fack = sblkp->end; 455 /* We must have at least one SACK hole in scoreboard. */ 456 KASSERT(!TAILQ_EMPTY(&tp->snd_holes), 457 ("SACK scoreboard must not be empty")); 458 cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */ 459 /* 460 * Since the incoming sack blocks are sorted, we can process them 461 * making one sweep of the scoreboard. 462 */ 463 while (sblkp >= sack_blocks && cur != NULL) { 464 if (SEQ_GEQ(sblkp->start, cur->end)) { 465 /* 466 * SACKs data beyond the current hole. Go to the 467 * previous sack block. 468 */ 469 sblkp--; 470 continue; 471 } 472 if (SEQ_LEQ(sblkp->end, cur->start)) { 473 /* 474 * SACKs data before the current hole. Go to the 475 * previous hole. 476 */ 477 cur = TAILQ_PREV(cur, sackhole_head, scblink); 478 continue; 479 } 480 tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start); 481 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 482 ("sackhint bytes rtx >= 0")); 483 if (SEQ_LEQ(sblkp->start, cur->start)) { 484 /* Data acks at least the beginning of hole. */ 485 if (SEQ_GEQ(sblkp->end, cur->end)) { 486 /* Acks entire hole, so delete hole. */ 487 temp = cur; 488 cur = TAILQ_PREV(cur, sackhole_head, scblink); 489 tcp_sackhole_remove(tp, temp); 490 /* 491 * The sack block may ack all or part of the 492 * next hole too, so continue onto the next 493 * hole. 494 */ 495 continue; 496 } else { 497 /* Move start of hole forward. */ 498 cur->start = sblkp->end; 499 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 500 } 501 } else { 502 /* Data acks at least the end of hole. */ 503 if (SEQ_GEQ(sblkp->end, cur->end)) { 504 /* Move end of hole backward. */ 505 cur->end = sblkp->start; 506 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 507 } else { 508 /* 509 * ACKs some data in middle of a hole; need 510 * to split current hole 511 */ 512 temp = tcp_sackhole_insert(tp, sblkp->end, 513 cur->end, cur); 514 if (temp != NULL) { 515 if (SEQ_GT(cur->rxmit, temp->rxmit)) { 516 temp->rxmit = cur->rxmit; 517 tp->sackhint.sack_bytes_rexmit 518 += (temp->rxmit 519 - temp->start); 520 } 521 cur->end = sblkp->start; 522 cur->rxmit = SEQ_MIN(cur->rxmit, 523 cur->end); 524 } 525 } 526 } 527 tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start); 528 /* 529 * Testing sblkp->start against cur->start tells us whether 530 * we're done with the sack block or the sack hole. 531 * Accordingly, we advance one or the other. 532 */ 533 if (SEQ_LEQ(sblkp->start, cur->start)) 534 cur = TAILQ_PREV(cur, sackhole_head, scblink); 535 else 536 sblkp--; 537 } 538 } 539 540 /* 541 * Free all SACK holes to clear the scoreboard. 542 */ 543 void 544 tcp_free_sackholes(struct tcpcb *tp) 545 { 546 struct sackhole *q; 547 548 INP_WLOCK_ASSERT(tp->t_inpcb); 549 while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) 550 tcp_sackhole_remove(tp, q); 551 tp->sackhint.sack_bytes_rexmit = 0; 552 553 KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); 554 KASSERT(tp->sackhint.nexthole == NULL, 555 ("tp->sackhint.nexthole == NULL")); 556 } 557 558 /* 559 * Partial ack handling within a sack recovery episode. Keeping this very 560 * simple for now. When a partial ack is received, force snd_cwnd to a value 561 * that will allow the sender to transmit no more than 2 segments. If 562 * necessary, a better scheme can be adopted at a later point, but for now, 563 * the goal is to prevent the sender from bursting a large amount of data in 564 * the midst of sack recovery. 565 */ 566 void 567 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) 568 { 569 int num_segs = 1; 570 571 INP_WLOCK_ASSERT(tp->t_inpcb); 572 tcp_timer_activate(tp, TT_REXMT, 0); 573 tp->t_rtttime = 0; 574 /* Send one or 2 segments based on how much new data was acked. */ 575 if (((th->th_ack - tp->snd_una) / tp->t_maxseg) > 2) 576 num_segs = 2; 577 tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + 578 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_maxseg); 579 if (tp->snd_cwnd > tp->snd_ssthresh) 580 tp->snd_cwnd = tp->snd_ssthresh; 581 tp->t_flags |= TF_ACKNOW; 582 (void) tcp_output(tp); 583 } 584 585 #if 0 586 /* 587 * Debug version of tcp_sack_output() that walks the scoreboard. Used for 588 * now to sanity check the hint. 589 */ 590 static struct sackhole * 591 tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) 592 { 593 struct sackhole *p; 594 595 INP_WLOCK_ASSERT(tp->t_inpcb); 596 *sack_bytes_rexmt = 0; 597 TAILQ_FOREACH(p, &tp->snd_holes, scblink) { 598 if (SEQ_LT(p->rxmit, p->end)) { 599 if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ 600 continue; 601 } 602 *sack_bytes_rexmt += (p->rxmit - p->start); 603 break; 604 } 605 *sack_bytes_rexmt += (p->rxmit - p->start); 606 } 607 return (p); 608 } 609 #endif 610 611 /* 612 * Returns the next hole to retransmit and the number of retransmitted bytes 613 * from the scoreboard. We store both the next hole and the number of 614 * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK 615 * reception). This avoids scoreboard traversals completely. 616 * 617 * The loop here will traverse *at most* one link. Here's the argument. For 618 * the loop to traverse more than 1 link before finding the next hole to 619 * retransmit, we would need to have at least 1 node following the current 620 * hint with (rxmit == end). But, for all holes following the current hint, 621 * (start == rxmit), since we have not yet retransmitted from them. 622 * Therefore, in order to traverse more 1 link in the loop below, we need to 623 * have at least one node following the current hint with (start == rxmit == 624 * end). But that can't happen, (start == end) means that all the data in 625 * that hole has been sacked, in which case, the hole would have been removed 626 * from the scoreboard. 627 */ 628 struct sackhole * 629 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 630 { 631 struct sackhole *hole = NULL; 632 633 INP_WLOCK_ASSERT(tp->t_inpcb); 634 *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; 635 hole = tp->sackhint.nexthole; 636 if (hole == NULL || SEQ_LT(hole->rxmit, hole->end)) 637 goto out; 638 while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) { 639 if (SEQ_LT(hole->rxmit, hole->end)) { 640 tp->sackhint.nexthole = hole; 641 break; 642 } 643 } 644 out: 645 return (hole); 646 } 647 648 /* 649 * After a timeout, the SACK list may be rebuilt. This SACK information 650 * should be used to avoid retransmitting SACKed data. This function 651 * traverses the SACK list to see if snd_nxt should be moved forward. 652 */ 653 void 654 tcp_sack_adjust(struct tcpcb *tp) 655 { 656 struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); 657 658 INP_WLOCK_ASSERT(tp->t_inpcb); 659 if (cur == NULL) 660 return; /* No holes */ 661 if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) 662 return; /* We're already beyond any SACKed blocks */ 663 /*- 664 * Two cases for which we want to advance snd_nxt: 665 * i) snd_nxt lies between end of one hole and beginning of another 666 * ii) snd_nxt lies between end of last hole and snd_fack 667 */ 668 while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { 669 if (SEQ_LT(tp->snd_nxt, cur->end)) 670 return; 671 if (SEQ_GEQ(tp->snd_nxt, p->start)) 672 cur = p; 673 else { 674 tp->snd_nxt = p->start; 675 return; 676 } 677 } 678 if (SEQ_LT(tp->snd_nxt, cur->end)) 679 return; 680 tp->snd_nxt = tp->snd_fack; 681 return; 682 } 683