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_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, enable, 130 CTLFLAG_RW, 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_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, maxholes, 135 CTLFLAG_RW, tcp_sack_maxholes, 0, 136 "Maximum number of TCP SACK holes allowed per connection"); 137 138 static int tcp_sack_globalmaxholes = 65536; 139 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, globalmaxholes, 140 CTLFLAG_RW, tcp_sack_globalmaxholes, 0, 141 "Global maximum number of TCP SACK holes"); 142 143 static int tcp_sack_globalholes = 0; 144 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_sack, OID_AUTO, globalholes, 145 CTLFLAG_RD, 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 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 257 struct sackhole *hole; 258 259 if (tp->snd_numholes >= V_tcp_sack_maxholes || 260 V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) { 261 V_tcpstat.tcps_sack_sboverflow++; 262 return NULL; 263 } 264 265 hole = (struct sackhole *)uma_zalloc(sack_hole_zone, M_NOWAIT); 266 if (hole == NULL) 267 return NULL; 268 269 hole->start = start; 270 hole->end = end; 271 hole->rxmit = start; 272 273 tp->snd_numholes++; 274 V_tcp_sack_globalholes++; 275 276 return hole; 277 } 278 279 /* 280 * Free struct sackhole. 281 */ 282 static void 283 tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) 284 { 285 INIT_VNET_INET(tp->t_vnet); 286 287 uma_zfree(sack_hole_zone, hole); 288 289 tp->snd_numholes--; 290 V_tcp_sack_globalholes--; 291 292 KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0")); 293 KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0")); 294 } 295 296 /* 297 * Insert new SACK hole into scoreboard. 298 */ 299 static struct sackhole * 300 tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end, 301 struct sackhole *after) 302 { 303 struct sackhole *hole; 304 305 /* Allocate a new SACK hole. */ 306 hole = tcp_sackhole_alloc(tp, start, end); 307 if (hole == NULL) 308 return NULL; 309 310 /* Insert the new SACK hole into scoreboard. */ 311 if (after != NULL) 312 TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink); 313 else 314 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink); 315 316 /* Update SACK hint. */ 317 if (tp->sackhint.nexthole == NULL) 318 tp->sackhint.nexthole = hole; 319 320 return hole; 321 } 322 323 /* 324 * Remove SACK hole from scoreboard. 325 */ 326 static void 327 tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole) 328 { 329 330 /* Update SACK hint. */ 331 if (tp->sackhint.nexthole == hole) 332 tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink); 333 334 /* Remove this SACK hole. */ 335 TAILQ_REMOVE(&tp->snd_holes, hole, scblink); 336 337 /* Free this SACK hole. */ 338 tcp_sackhole_free(tp, hole); 339 } 340 341 /* 342 * Process cumulative ACK and the TCP SACK option to update the scoreboard. 343 * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of 344 * the sequence space). 345 */ 346 void 347 tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) 348 { 349 struct sackhole *cur, *temp; 350 struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; 351 int i, j, num_sack_blks; 352 353 INP_WLOCK_ASSERT(tp->t_inpcb); 354 355 num_sack_blks = 0; 356 /* 357 * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, 358 * treat [SND.UNA, SEG.ACK) as if it is a SACK block. 359 */ 360 if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { 361 sack_blocks[num_sack_blks].start = tp->snd_una; 362 sack_blocks[num_sack_blks++].end = th_ack; 363 } 364 /* 365 * Append received valid SACK blocks to sack_blocks[], but only if we 366 * received new blocks from the other side. 367 */ 368 if (to->to_flags & TOF_SACK) { 369 for (i = 0; i < to->to_nsacks; i++) { 370 bcopy((to->to_sacks + i * TCPOLEN_SACK), 371 &sack, sizeof(sack)); 372 sack.start = ntohl(sack.start); 373 sack.end = ntohl(sack.end); 374 if (SEQ_GT(sack.end, sack.start) && 375 SEQ_GT(sack.start, tp->snd_una) && 376 SEQ_GT(sack.start, th_ack) && 377 SEQ_LT(sack.start, tp->snd_max) && 378 SEQ_GT(sack.end, tp->snd_una) && 379 SEQ_LEQ(sack.end, tp->snd_max)) 380 sack_blocks[num_sack_blks++] = sack; 381 } 382 } 383 /* 384 * Return if SND.UNA is not advanced and no valid SACK block is 385 * received. 386 */ 387 if (num_sack_blks == 0) 388 return; 389 390 /* 391 * Sort the SACK blocks so we can update the scoreboard with just one 392 * pass. The overhead of sorting upto 4+1 elements is less than 393 * making upto 4+1 passes over the scoreboard. 394 */ 395 for (i = 0; i < num_sack_blks; i++) { 396 for (j = i + 1; j < num_sack_blks; j++) { 397 if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { 398 sack = sack_blocks[i]; 399 sack_blocks[i] = sack_blocks[j]; 400 sack_blocks[j] = sack; 401 } 402 } 403 } 404 if (TAILQ_EMPTY(&tp->snd_holes)) 405 /* 406 * Empty scoreboard. Need to initialize snd_fack (it may be 407 * uninitialized or have a bogus value). Scoreboard holes 408 * (from the sack blocks received) are created later below 409 * (in the logic that adds holes to the tail of the 410 * scoreboard). 411 */ 412 tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); 413 /* 414 * In the while-loop below, incoming SACK blocks (sack_blocks[]) and 415 * SACK holes (snd_holes) are traversed from their tails with just 416 * one pass in order to reduce the number of compares especially when 417 * the bandwidth-delay product is large. 418 * 419 * Note: Typically, in the first RTT of SACK recovery, the highest 420 * three or four SACK blocks with the same ack number are received. 421 * In the second RTT, if retransmitted data segments are not lost, 422 * the highest three or four SACK blocks with ack number advancing 423 * are received. 424 */ 425 sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ 426 if (SEQ_LT(tp->snd_fack, sblkp->start)) { 427 /* 428 * The highest SACK block is beyond fack. Append new SACK 429 * hole at the tail. If the second or later highest SACK 430 * blocks are also beyond the current fack, they will be 431 * inserted by way of hole splitting in the while-loop below. 432 */ 433 temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); 434 if (temp != NULL) { 435 tp->snd_fack = sblkp->end; 436 /* Go to the previous sack block. */ 437 sblkp--; 438 } else { 439 /* 440 * We failed to add a new hole based on the current 441 * sack block. Skip over all the sack blocks that 442 * fall completely to the right of snd_fack and 443 * proceed to trim the scoreboard based on the 444 * remaining sack blocks. This also trims the 445 * scoreboard for th_ack (which is sack_blocks[0]). 446 */ 447 while (sblkp >= sack_blocks && 448 SEQ_LT(tp->snd_fack, sblkp->start)) 449 sblkp--; 450 if (sblkp >= sack_blocks && 451 SEQ_LT(tp->snd_fack, sblkp->end)) 452 tp->snd_fack = sblkp->end; 453 } 454 } else if (SEQ_LT(tp->snd_fack, sblkp->end)) 455 /* fack is advanced. */ 456 tp->snd_fack = sblkp->end; 457 /* We must have at least one SACK hole in scoreboard. */ 458 KASSERT(!TAILQ_EMPTY(&tp->snd_holes), 459 ("SACK scoreboard must not be empty")); 460 cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */ 461 /* 462 * Since the incoming sack blocks are sorted, we can process them 463 * making one sweep of the scoreboard. 464 */ 465 while (sblkp >= sack_blocks && cur != NULL) { 466 if (SEQ_GEQ(sblkp->start, cur->end)) { 467 /* 468 * SACKs data beyond the current hole. Go to the 469 * previous sack block. 470 */ 471 sblkp--; 472 continue; 473 } 474 if (SEQ_LEQ(sblkp->end, cur->start)) { 475 /* 476 * SACKs data before the current hole. Go to the 477 * previous hole. 478 */ 479 cur = TAILQ_PREV(cur, sackhole_head, scblink); 480 continue; 481 } 482 tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start); 483 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 484 ("sackhint bytes rtx >= 0")); 485 if (SEQ_LEQ(sblkp->start, cur->start)) { 486 /* Data acks at least the beginning of hole. */ 487 if (SEQ_GEQ(sblkp->end, cur->end)) { 488 /* Acks entire hole, so delete hole. */ 489 temp = cur; 490 cur = TAILQ_PREV(cur, sackhole_head, scblink); 491 tcp_sackhole_remove(tp, temp); 492 /* 493 * The sack block may ack all or part of the 494 * next hole too, so continue onto the next 495 * hole. 496 */ 497 continue; 498 } else { 499 /* Move start of hole forward. */ 500 cur->start = sblkp->end; 501 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 502 } 503 } else { 504 /* Data acks at least the end of hole. */ 505 if (SEQ_GEQ(sblkp->end, cur->end)) { 506 /* Move end of hole backward. */ 507 cur->end = sblkp->start; 508 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 509 } else { 510 /* 511 * ACKs some data in middle of a hole; need 512 * to split current hole 513 */ 514 temp = tcp_sackhole_insert(tp, sblkp->end, 515 cur->end, cur); 516 if (temp != NULL) { 517 if (SEQ_GT(cur->rxmit, temp->rxmit)) { 518 temp->rxmit = cur->rxmit; 519 tp->sackhint.sack_bytes_rexmit 520 += (temp->rxmit 521 - temp->start); 522 } 523 cur->end = sblkp->start; 524 cur->rxmit = SEQ_MIN(cur->rxmit, 525 cur->end); 526 } 527 } 528 } 529 tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start); 530 /* 531 * Testing sblkp->start against cur->start tells us whether 532 * we're done with the sack block or the sack hole. 533 * Accordingly, we advance one or the other. 534 */ 535 if (SEQ_LEQ(sblkp->start, cur->start)) 536 cur = TAILQ_PREV(cur, sackhole_head, scblink); 537 else 538 sblkp--; 539 } 540 } 541 542 /* 543 * Free all SACK holes to clear the scoreboard. 544 */ 545 void 546 tcp_free_sackholes(struct tcpcb *tp) 547 { 548 struct sackhole *q; 549 550 INP_WLOCK_ASSERT(tp->t_inpcb); 551 while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) 552 tcp_sackhole_remove(tp, q); 553 tp->sackhint.sack_bytes_rexmit = 0; 554 555 KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); 556 KASSERT(tp->sackhint.nexthole == NULL, 557 ("tp->sackhint.nexthole == NULL")); 558 } 559 560 /* 561 * Partial ack handling within a sack recovery episode. Keeping this very 562 * simple for now. When a partial ack is received, force snd_cwnd to a value 563 * that will allow the sender to transmit no more than 2 segments. If 564 * necessary, a better scheme can be adopted at a later point, but for now, 565 * the goal is to prevent the sender from bursting a large amount of data in 566 * the midst of sack recovery. 567 */ 568 void 569 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) 570 { 571 int num_segs = 1; 572 573 INP_WLOCK_ASSERT(tp->t_inpcb); 574 tcp_timer_activate(tp, TT_REXMT, 0); 575 tp->t_rtttime = 0; 576 /* Send one or 2 segments based on how much new data was acked. */ 577 if (((th->th_ack - tp->snd_una) / tp->t_maxseg) > 2) 578 num_segs = 2; 579 tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + 580 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_maxseg); 581 if (tp->snd_cwnd > tp->snd_ssthresh) 582 tp->snd_cwnd = tp->snd_ssthresh; 583 tp->t_flags |= TF_ACKNOW; 584 (void) tcp_output(tp); 585 } 586 587 #if 0 588 /* 589 * Debug version of tcp_sack_output() that walks the scoreboard. Used for 590 * now to sanity check the hint. 591 */ 592 static struct sackhole * 593 tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) 594 { 595 struct sackhole *p; 596 597 INP_WLOCK_ASSERT(tp->t_inpcb); 598 *sack_bytes_rexmt = 0; 599 TAILQ_FOREACH(p, &tp->snd_holes, scblink) { 600 if (SEQ_LT(p->rxmit, p->end)) { 601 if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ 602 continue; 603 } 604 *sack_bytes_rexmt += (p->rxmit - p->start); 605 break; 606 } 607 *sack_bytes_rexmt += (p->rxmit - p->start); 608 } 609 return (p); 610 } 611 #endif 612 613 /* 614 * Returns the next hole to retransmit and the number of retransmitted bytes 615 * from the scoreboard. We store both the next hole and the number of 616 * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK 617 * reception). This avoids scoreboard traversals completely. 618 * 619 * The loop here will traverse *at most* one link. Here's the argument. For 620 * the loop to traverse more than 1 link before finding the next hole to 621 * retransmit, we would need to have at least 1 node following the current 622 * hint with (rxmit == end). But, for all holes following the current hint, 623 * (start == rxmit), since we have not yet retransmitted from them. 624 * Therefore, in order to traverse more 1 link in the loop below, we need to 625 * have at least one node following the current hint with (start == rxmit == 626 * end). But that can't happen, (start == end) means that all the data in 627 * that hole has been sacked, in which case, the hole would have been removed 628 * from the scoreboard. 629 */ 630 struct sackhole * 631 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 632 { 633 struct sackhole *hole = NULL; 634 635 INP_WLOCK_ASSERT(tp->t_inpcb); 636 *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; 637 hole = tp->sackhint.nexthole; 638 if (hole == NULL || SEQ_LT(hole->rxmit, hole->end)) 639 goto out; 640 while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) { 641 if (SEQ_LT(hole->rxmit, hole->end)) { 642 tp->sackhint.nexthole = hole; 643 break; 644 } 645 } 646 out: 647 return (hole); 648 } 649 650 /* 651 * After a timeout, the SACK list may be rebuilt. This SACK information 652 * should be used to avoid retransmitting SACKed data. This function 653 * traverses the SACK list to see if snd_nxt should be moved forward. 654 */ 655 void 656 tcp_sack_adjust(struct tcpcb *tp) 657 { 658 struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); 659 660 INP_WLOCK_ASSERT(tp->t_inpcb); 661 if (cur == NULL) 662 return; /* No holes */ 663 if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) 664 return; /* We're already beyond any SACKed blocks */ 665 /*- 666 * Two cases for which we want to advance snd_nxt: 667 * i) snd_nxt lies between end of one hole and beginning of another 668 * ii) snd_nxt lies between end of last hole and snd_fack 669 */ 670 while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { 671 if (SEQ_LT(tp->snd_nxt, cur->end)) 672 return; 673 if (SEQ_GEQ(tp->snd_nxt, p->start)) 674 cur = p; 675 else { 676 tp->snd_nxt = p->start; 677 return; 678 } 679 } 680 if (SEQ_LT(tp->snd_nxt, cur->end)) 681 return; 682 tp->snd_nxt = tp->snd_fack; 683 } 684