1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. 6 * All rights reserved. 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 University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95 33 */ 34 35 /*- 36 * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995 37 * 38 * NRL grants permission for redistribution and use in source and binary 39 * forms, with or without modification, of the software and documentation 40 * created at NRL provided that the following conditions are met: 41 * 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. All advertising materials mentioning features or use of this software 48 * must display the following acknowledgements: 49 * This product includes software developed by the University of 50 * California, Berkeley and its contributors. 51 * This product includes software developed at the Information 52 * Technology Division, US Naval Research Laboratory. 53 * 4. Neither the name of the NRL nor the names of its contributors 54 * may be used to endorse or promote products derived from this software 55 * without specific prior written permission. 56 * 57 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 58 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 59 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 60 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 61 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 62 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 63 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 64 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 65 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 66 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 67 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 68 * 69 * The views and conclusions contained in the software and documentation 70 * are those of the authors and should not be interpreted as representing 71 * official policies, either expressed or implied, of the US Naval 72 * Research Laboratory (NRL). 73 */ 74 75 #include <sys/cdefs.h> 76 __FBSDID("$FreeBSD$"); 77 78 #include "opt_inet.h" 79 #include "opt_inet6.h" 80 #include "opt_tcpdebug.h" 81 82 #include <sys/param.h> 83 #include <sys/systm.h> 84 #include <sys/kernel.h> 85 #include <sys/sysctl.h> 86 #include <sys/malloc.h> 87 #include <sys/mbuf.h> 88 #include <sys/proc.h> /* for proc0 declaration */ 89 #include <sys/protosw.h> 90 #include <sys/socket.h> 91 #include <sys/socketvar.h> 92 #include <sys/syslog.h> 93 #include <sys/systm.h> 94 95 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 96 97 #include <vm/uma.h> 98 99 #include <net/if.h> 100 #include <net/if_var.h> 101 #include <net/route.h> 102 #include <net/vnet.h> 103 104 #include <netinet/in.h> 105 #include <netinet/in_systm.h> 106 #include <netinet/ip.h> 107 #include <netinet/in_var.h> 108 #include <netinet/in_pcb.h> 109 #include <netinet/ip_var.h> 110 #include <netinet/ip6.h> 111 #include <netinet/icmp6.h> 112 #include <netinet6/nd6.h> 113 #include <netinet6/ip6_var.h> 114 #include <netinet6/in6_pcb.h> 115 #include <netinet/tcp.h> 116 #include <netinet/tcp_fsm.h> 117 #include <netinet/tcp_seq.h> 118 #include <netinet/tcp_timer.h> 119 #include <netinet/tcp_var.h> 120 #include <netinet6/tcp6_var.h> 121 #include <netinet/tcpip.h> 122 #ifdef TCPDEBUG 123 #include <netinet/tcp_debug.h> 124 #endif /* TCPDEBUG */ 125 126 #include <machine/in_cksum.h> 127 128 VNET_DECLARE(struct uma_zone *, sack_hole_zone); 129 #define V_sack_hole_zone VNET(sack_hole_zone) 130 131 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 132 VNET_DEFINE(int, tcp_do_sack) = 1; 133 #define V_tcp_do_sack VNET(tcp_do_sack) 134 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 135 &VNET_NAME(tcp_do_sack), 0, "Enable/Disable TCP SACK support"); 136 137 VNET_DEFINE(int, tcp_sack_maxholes) = 128; 138 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(tcp_sack_maxholes), 0, 140 "Maximum number of TCP SACK holes allowed per connection"); 141 142 VNET_DEFINE(int, tcp_sack_globalmaxholes) = 65536; 143 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(tcp_sack_globalmaxholes), 0, 145 "Global maximum number of TCP SACK holes"); 146 147 VNET_DEFINE(int, tcp_sack_globalholes) = 0; 148 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_VNET | CTLFLAG_RD, 149 &VNET_NAME(tcp_sack_globalholes), 0, 150 "Global number of TCP SACK holes currently allocated"); 151 152 /* 153 * This function is called upon receipt of new valid data (while not in 154 * header prediction mode), and it updates the ordered list of sacks. 155 */ 156 void 157 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) 158 { 159 /* 160 * First reported block MUST be the most recent one. Subsequent 161 * blocks SHOULD be in the order in which they arrived at the 162 * receiver. These two conditions make the implementation fully 163 * compliant with RFC 2018. 164 */ 165 struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; 166 int num_head, num_saved, i; 167 168 INP_WLOCK_ASSERT(tp->t_inpcb); 169 170 /* Check arguments. */ 171 KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end")); 172 173 /* SACK block for the received segment. */ 174 head_blk.start = rcv_start; 175 head_blk.end = rcv_end; 176 177 /* 178 * Merge updated SACK blocks into head_blk, and save unchanged SACK 179 * blocks into saved_blks[]. num_saved will have the number of the 180 * saved SACK blocks. 181 */ 182 num_saved = 0; 183 for (i = 0; i < tp->rcv_numsacks; i++) { 184 tcp_seq start = tp->sackblks[i].start; 185 tcp_seq end = tp->sackblks[i].end; 186 if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { 187 /* 188 * Discard this SACK block. 189 */ 190 } else if (SEQ_LEQ(head_blk.start, end) && 191 SEQ_GEQ(head_blk.end, start)) { 192 /* 193 * Merge this SACK block into head_blk. This SACK 194 * block itself will be discarded. 195 */ 196 if (SEQ_GT(head_blk.start, start)) 197 head_blk.start = start; 198 if (SEQ_LT(head_blk.end, end)) 199 head_blk.end = end; 200 } else { 201 /* 202 * Save this SACK block. 203 */ 204 saved_blks[num_saved].start = start; 205 saved_blks[num_saved].end = end; 206 num_saved++; 207 } 208 } 209 210 /* 211 * Update SACK list in tp->sackblks[]. 212 */ 213 num_head = 0; 214 if (SEQ_GT(head_blk.start, tp->rcv_nxt)) { 215 /* 216 * The received data segment is an out-of-order segment. Put 217 * head_blk at the top of SACK list. 218 */ 219 tp->sackblks[0] = head_blk; 220 num_head = 1; 221 /* 222 * If the number of saved SACK blocks exceeds its limit, 223 * discard the last SACK block. 224 */ 225 if (num_saved >= MAX_SACK_BLKS) 226 num_saved--; 227 } 228 if (num_saved > 0) { 229 /* 230 * Copy the saved SACK blocks back. 231 */ 232 bcopy(saved_blks, &tp->sackblks[num_head], 233 sizeof(struct sackblk) * num_saved); 234 } 235 236 /* Save the number of SACK blocks. */ 237 tp->rcv_numsacks = num_head + num_saved; 238 } 239 240 /* 241 * Delete all receiver-side SACK information. 242 */ 243 void 244 tcp_clean_sackreport(struct tcpcb *tp) 245 { 246 int i; 247 248 INP_WLOCK_ASSERT(tp->t_inpcb); 249 tp->rcv_numsacks = 0; 250 for (i = 0; i < MAX_SACK_BLKS; i++) 251 tp->sackblks[i].start = tp->sackblks[i].end=0; 252 } 253 254 /* 255 * Allocate struct sackhole. 256 */ 257 static struct sackhole * 258 tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) 259 { 260 struct sackhole *hole; 261 262 if (tp->snd_numholes >= V_tcp_sack_maxholes || 263 V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) { 264 TCPSTAT_INC(tcps_sack_sboverflow); 265 return NULL; 266 } 267 268 hole = (struct sackhole *)uma_zalloc(V_sack_hole_zone, M_NOWAIT); 269 if (hole == NULL) 270 return NULL; 271 272 hole->start = start; 273 hole->end = end; 274 hole->rxmit = start; 275 276 tp->snd_numholes++; 277 atomic_add_int(&V_tcp_sack_globalholes, 1); 278 279 return hole; 280 } 281 282 /* 283 * Free struct sackhole. 284 */ 285 static void 286 tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) 287 { 288 289 uma_zfree(V_sack_hole_zone, hole); 290 291 tp->snd_numholes--; 292 atomic_subtract_int(&V_tcp_sack_globalholes, 1); 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 */ 301 static struct sackhole * 302 tcp_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 */ 328 static void 329 tcp_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 * Returns 1 if incoming ACK has previously unknown SACK information, 348 * 0 otherwise. Note: We treat (snd_una, th_ack) as a sack block so any changes 349 * to that (i.e. left edge moving) would also be considered a change in SACK 350 * information which is slightly different than rfc6675. 351 */ 352 int 353 tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) 354 { 355 struct sackhole *cur, *temp; 356 struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; 357 int i, j, num_sack_blks, sack_changed; 358 359 INP_WLOCK_ASSERT(tp->t_inpcb); 360 361 num_sack_blks = 0; 362 sack_changed = 0; 363 /* 364 * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, 365 * treat [SND.UNA, SEG.ACK) as if it is a SACK block. 366 */ 367 if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { 368 sack_blocks[num_sack_blks].start = tp->snd_una; 369 sack_blocks[num_sack_blks++].end = th_ack; 370 } 371 /* 372 * Append received valid SACK blocks to sack_blocks[], but only if we 373 * received new blocks from the other side. 374 */ 375 if (to->to_flags & TOF_SACK) { 376 tp->sackhint.sacked_bytes = 0; /* reset */ 377 for (i = 0; i < to->to_nsacks; i++) { 378 bcopy((to->to_sacks + i * TCPOLEN_SACK), 379 &sack, sizeof(sack)); 380 sack.start = ntohl(sack.start); 381 sack.end = ntohl(sack.end); 382 if (SEQ_GT(sack.end, sack.start) && 383 SEQ_GT(sack.start, tp->snd_una) && 384 SEQ_GT(sack.start, th_ack) && 385 SEQ_LT(sack.start, tp->snd_max) && 386 SEQ_GT(sack.end, tp->snd_una) && 387 SEQ_LEQ(sack.end, tp->snd_max)) { 388 sack_blocks[num_sack_blks++] = sack; 389 tp->sackhint.sacked_bytes += 390 (sack.end-sack.start); 391 } 392 } 393 } 394 /* 395 * Return if SND.UNA is not advanced and no valid SACK block is 396 * received. 397 */ 398 if (num_sack_blks == 0) 399 return (sack_changed); 400 401 /* 402 * Sort the SACK blocks so we can update the scoreboard with just one 403 * pass. The overhead of sorting up to 4+1 elements is less than 404 * making up to 4+1 passes over the scoreboard. 405 */ 406 for (i = 0; i < num_sack_blks; i++) { 407 for (j = i + 1; j < num_sack_blks; j++) { 408 if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { 409 sack = sack_blocks[i]; 410 sack_blocks[i] = sack_blocks[j]; 411 sack_blocks[j] = sack; 412 } 413 } 414 } 415 if (TAILQ_EMPTY(&tp->snd_holes)) 416 /* 417 * Empty scoreboard. Need to initialize snd_fack (it may be 418 * uninitialized or have a bogus value). Scoreboard holes 419 * (from the sack blocks received) are created later below 420 * (in the logic that adds holes to the tail of the 421 * scoreboard). 422 */ 423 tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); 424 /* 425 * In the while-loop below, incoming SACK blocks (sack_blocks[]) and 426 * SACK holes (snd_holes) are traversed from their tails with just 427 * one pass in order to reduce the number of compares especially when 428 * the bandwidth-delay product is large. 429 * 430 * Note: Typically, in the first RTT of SACK recovery, the highest 431 * three or four SACK blocks with the same ack number are received. 432 * In the second RTT, if retransmitted data segments are not lost, 433 * the highest three or four SACK blocks with ack number advancing 434 * are received. 435 */ 436 sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ 437 tp->sackhint.last_sack_ack = sblkp->end; 438 if (SEQ_LT(tp->snd_fack, sblkp->start)) { 439 /* 440 * The highest SACK block is beyond fack. Append new SACK 441 * hole at the tail. If the second or later highest SACK 442 * blocks are also beyond the current fack, they will be 443 * inserted by way of hole splitting in the while-loop below. 444 */ 445 temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); 446 if (temp != NULL) { 447 tp->snd_fack = sblkp->end; 448 /* Go to the previous sack block. */ 449 sblkp--; 450 sack_changed = 1; 451 } else { 452 /* 453 * We failed to add a new hole based on the current 454 * sack block. Skip over all the sack blocks that 455 * fall completely to the right of snd_fack and 456 * proceed to trim the scoreboard based on the 457 * remaining sack blocks. This also trims the 458 * scoreboard for th_ack (which is sack_blocks[0]). 459 */ 460 while (sblkp >= sack_blocks && 461 SEQ_LT(tp->snd_fack, sblkp->start)) 462 sblkp--; 463 if (sblkp >= sack_blocks && 464 SEQ_LT(tp->snd_fack, sblkp->end)) 465 tp->snd_fack = sblkp->end; 466 } 467 } else if (SEQ_LT(tp->snd_fack, sblkp->end)) { 468 /* fack is advanced. */ 469 tp->snd_fack = sblkp->end; 470 sack_changed = 1; 471 } 472 cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */ 473 /* 474 * Since the incoming sack blocks are sorted, we can process them 475 * making one sweep of the scoreboard. 476 */ 477 while (sblkp >= sack_blocks && cur != NULL) { 478 if (SEQ_GEQ(sblkp->start, cur->end)) { 479 /* 480 * SACKs data beyond the current hole. Go to the 481 * previous sack block. 482 */ 483 sblkp--; 484 continue; 485 } 486 if (SEQ_LEQ(sblkp->end, cur->start)) { 487 /* 488 * SACKs data before the current hole. Go to the 489 * previous hole. 490 */ 491 cur = TAILQ_PREV(cur, sackhole_head, scblink); 492 continue; 493 } 494 tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start); 495 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 496 ("sackhint bytes rtx >= 0")); 497 sack_changed = 1; 498 if (SEQ_LEQ(sblkp->start, cur->start)) { 499 /* Data acks at least the beginning of hole. */ 500 if (SEQ_GEQ(sblkp->end, cur->end)) { 501 /* Acks entire hole, so delete hole. */ 502 temp = cur; 503 cur = TAILQ_PREV(cur, sackhole_head, scblink); 504 tcp_sackhole_remove(tp, temp); 505 /* 506 * The sack block may ack all or part of the 507 * next hole too, so continue onto the next 508 * hole. 509 */ 510 continue; 511 } else { 512 /* Move start of hole forward. */ 513 cur->start = sblkp->end; 514 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 515 } 516 } else { 517 /* Data acks at least the end of hole. */ 518 if (SEQ_GEQ(sblkp->end, cur->end)) { 519 /* Move end of hole backward. */ 520 cur->end = sblkp->start; 521 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 522 } else { 523 /* 524 * ACKs some data in middle of a hole; need 525 * to split current hole 526 */ 527 temp = tcp_sackhole_insert(tp, sblkp->end, 528 cur->end, cur); 529 if (temp != NULL) { 530 if (SEQ_GT(cur->rxmit, temp->rxmit)) { 531 temp->rxmit = cur->rxmit; 532 tp->sackhint.sack_bytes_rexmit 533 += (temp->rxmit 534 - temp->start); 535 } 536 cur->end = sblkp->start; 537 cur->rxmit = SEQ_MIN(cur->rxmit, 538 cur->end); 539 } 540 } 541 } 542 tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start); 543 /* 544 * Testing sblkp->start against cur->start tells us whether 545 * we're done with the sack block or the sack hole. 546 * Accordingly, we advance one or the other. 547 */ 548 if (SEQ_LEQ(sblkp->start, cur->start)) 549 cur = TAILQ_PREV(cur, sackhole_head, scblink); 550 else 551 sblkp--; 552 } 553 return (sack_changed); 554 } 555 556 /* 557 * Free all SACK holes to clear the scoreboard. 558 */ 559 void 560 tcp_free_sackholes(struct tcpcb *tp) 561 { 562 struct sackhole *q; 563 564 INP_WLOCK_ASSERT(tp->t_inpcb); 565 while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) 566 tcp_sackhole_remove(tp, q); 567 tp->sackhint.sack_bytes_rexmit = 0; 568 569 KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); 570 KASSERT(tp->sackhint.nexthole == NULL, 571 ("tp->sackhint.nexthole == NULL")); 572 } 573 574 /* 575 * Partial ack handling within a sack recovery episode. Keeping this very 576 * simple for now. When a partial ack is received, force snd_cwnd to a value 577 * that will allow the sender to transmit no more than 2 segments. If 578 * necessary, a better scheme can be adopted at a later point, but for now, 579 * the goal is to prevent the sender from bursting a large amount of data in 580 * the midst of sack recovery. 581 */ 582 void 583 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) 584 { 585 int num_segs = 1; 586 587 INP_WLOCK_ASSERT(tp->t_inpcb); 588 tcp_timer_activate(tp, TT_REXMT, 0); 589 tp->t_rtttime = 0; 590 /* Send one or 2 segments based on how much new data was acked. */ 591 if ((BYTES_THIS_ACK(tp, th) / tp->t_maxseg) >= 2) 592 num_segs = 2; 593 tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + 594 (tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_maxseg); 595 if (tp->snd_cwnd > tp->snd_ssthresh) 596 tp->snd_cwnd = tp->snd_ssthresh; 597 tp->t_flags |= TF_ACKNOW; 598 (void) tp->t_fb->tfb_tcp_output(tp); 599 } 600 601 #if 0 602 /* 603 * Debug version of tcp_sack_output() that walks the scoreboard. Used for 604 * now to sanity check the hint. 605 */ 606 static struct sackhole * 607 tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) 608 { 609 struct sackhole *p; 610 611 INP_WLOCK_ASSERT(tp->t_inpcb); 612 *sack_bytes_rexmt = 0; 613 TAILQ_FOREACH(p, &tp->snd_holes, scblink) { 614 if (SEQ_LT(p->rxmit, p->end)) { 615 if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ 616 continue; 617 } 618 *sack_bytes_rexmt += (p->rxmit - p->start); 619 break; 620 } 621 *sack_bytes_rexmt += (p->rxmit - p->start); 622 } 623 return (p); 624 } 625 #endif 626 627 /* 628 * Returns the next hole to retransmit and the number of retransmitted bytes 629 * from the scoreboard. We store both the next hole and the number of 630 * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK 631 * reception). This avoids scoreboard traversals completely. 632 * 633 * The loop here will traverse *at most* one link. Here's the argument. For 634 * the loop to traverse more than 1 link before finding the next hole to 635 * retransmit, we would need to have at least 1 node following the current 636 * hint with (rxmit == end). But, for all holes following the current hint, 637 * (start == rxmit), since we have not yet retransmitted from them. 638 * Therefore, in order to traverse more 1 link in the loop below, we need to 639 * have at least one node following the current hint with (start == rxmit == 640 * end). But that can't happen, (start == end) means that all the data in 641 * that hole has been sacked, in which case, the hole would have been removed 642 * from the scoreboard. 643 */ 644 struct sackhole * 645 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 646 { 647 struct sackhole *hole = NULL; 648 649 INP_WLOCK_ASSERT(tp->t_inpcb); 650 *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; 651 hole = tp->sackhint.nexthole; 652 if (hole == NULL || SEQ_LT(hole->rxmit, hole->end)) 653 goto out; 654 while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) { 655 if (SEQ_LT(hole->rxmit, hole->end)) { 656 tp->sackhint.nexthole = hole; 657 break; 658 } 659 } 660 out: 661 return (hole); 662 } 663 664 /* 665 * After a timeout, the SACK list may be rebuilt. This SACK information 666 * should be used to avoid retransmitting SACKed data. This function 667 * traverses the SACK list to see if snd_nxt should be moved forward. 668 */ 669 void 670 tcp_sack_adjust(struct tcpcb *tp) 671 { 672 struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); 673 674 INP_WLOCK_ASSERT(tp->t_inpcb); 675 if (cur == NULL) 676 return; /* No holes */ 677 if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) 678 return; /* We're already beyond any SACKed blocks */ 679 /*- 680 * Two cases for which we want to advance snd_nxt: 681 * i) snd_nxt lies between end of one hole and beginning of another 682 * ii) snd_nxt lies between end of last hole and snd_fack 683 */ 684 while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { 685 if (SEQ_LT(tp->snd_nxt, cur->end)) 686 return; 687 if (SEQ_GEQ(tp->snd_nxt, p->start)) 688 cur = p; 689 else { 690 tp->snd_nxt = p->start; 691 return; 692 } 693 } 694 if (SEQ_LT(tp->snd_nxt, cur->end)) 695 return; 696 tp->snd_nxt = tp->snd_fack; 697 } 698