1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 /*- 34 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 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. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995 62 * 63 * NRL grants permission for redistribution and use in source and binary 64 * forms, with or without modification, of the software and documentation 65 * created at NRL provided that the following conditions are met: 66 * 67 * 1. Redistributions of source code must retain the above copyright 68 * notice, this list of conditions and the following disclaimer. 69 * 2. Redistributions in binary form must reproduce the above copyright 70 * notice, this list of conditions and the following disclaimer in the 71 * documentation and/or other materials provided with the distribution. 72 * 3. All advertising materials mentioning features or use of this software 73 * must display the following acknowledgements: 74 * This product includes software developed by the University of 75 * California, Berkeley and its contributors. 76 * This product includes software developed at the Information 77 * Technology Division, US Naval Research Laboratory. 78 * 4. Neither the name of the NRL nor the names of its contributors 79 * may be used to endorse or promote products derived from this software 80 * without specific prior written permission. 81 * 82 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 83 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 84 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 85 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 86 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 87 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 88 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 89 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 90 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 91 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 92 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 93 * 94 * The views and conclusions contained in the software and documentation 95 * are those of the authors and should not be interpreted as representing 96 * official policies, either expressed or implied, of the US Naval 97 * Research Laboratory (NRL). 98 */ 99 #include "opt_inet.h" 100 #include "opt_inet6.h" 101 #include "opt_ipsec.h" 102 #include "opt_tcpdebug.h" 103 #include "opt_tcp_input.h" 104 #include "opt_tcp_sack.h" 105 106 #include <sys/param.h> 107 #include <sys/systm.h> 108 #include <sys/kernel.h> 109 #include <sys/sysctl.h> 110 #include <sys/malloc.h> 111 #include <sys/mbuf.h> 112 #include <sys/proc.h> /* for proc0 declaration */ 113 #include <sys/protosw.h> 114 #include <sys/socket.h> 115 #include <sys/socketvar.h> 116 #include <sys/syslog.h> 117 #include <sys/systm.h> 118 119 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 120 121 #include <vm/uma.h> 122 123 #include <net/if.h> 124 #include <net/route.h> 125 126 #include <netinet/in.h> 127 #include <netinet/in_systm.h> 128 #include <netinet/ip.h> 129 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 130 #include <netinet/in_var.h> 131 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 132 #include <netinet/in_pcb.h> 133 #include <netinet/ip_var.h> 134 #include <netinet/ip6.h> 135 #include <netinet/icmp6.h> 136 #include <netinet6/nd6.h> 137 #include <netinet6/ip6_var.h> 138 #include <netinet6/in6_pcb.h> 139 #include <netinet/tcp.h> 140 #include <netinet/tcp_fsm.h> 141 #include <netinet/tcp_seq.h> 142 #include <netinet/tcp_timer.h> 143 #include <netinet/tcp_var.h> 144 #include <netinet6/tcp6_var.h> 145 #include <netinet/tcpip.h> 146 #ifdef TCPDEBUG 147 #include <netinet/tcp_debug.h> 148 #endif /* TCPDEBUG */ 149 150 #ifdef FAST_IPSEC 151 #include <netipsec/ipsec.h> 152 #include <netipsec/ipsec6.h> 153 #endif 154 155 #ifdef IPSEC 156 #include <netinet6/ipsec.h> 157 #include <netinet6/ipsec6.h> 158 #include <netkey/key.h> 159 #endif /*IPSEC*/ 160 #include <machine/in_cksum.h> 161 162 extern struct uma_zone *sack_hole_zone; 163 164 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 165 int tcp_do_sack = 1; 166 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW, 167 &tcp_do_sack, 0, "Enable/Disable TCP SACK support"); 168 TUNABLE_INT("net.inet.tcp.sack.enable", &tcp_do_sack); 169 170 static int tcp_sack_maxholes = 128; 171 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_RW, 172 &tcp_sack_maxholes, 0, 173 "Maximum number of TCP SACK holes allowed per connection"); 174 175 static int tcp_sack_globalmaxholes = 65536; 176 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_RW, 177 &tcp_sack_globalmaxholes, 0, 178 "Global maximum number of TCP SACK holes"); 179 180 static int tcp_sack_globalholes = 0; 181 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_RD, 182 &tcp_sack_globalholes, 0, 183 "Global number of TCP SACK holes currently allocated"); 184 185 /* 186 * This function is called upon receipt of new valid data (while not in header 187 * prediction mode), and it updates the ordered list of sacks. 188 */ 189 void 190 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) 191 { 192 /* 193 * First reported block MUST be the most recent one. Subsequent 194 * blocks SHOULD be in the order in which they arrived at the 195 * receiver. These two conditions make the implementation fully 196 * compliant with RFC 2018. 197 */ 198 struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; 199 int num_head, num_saved, i; 200 201 INP_LOCK_ASSERT(tp->t_inpcb); 202 203 /* Check arguments */ 204 KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end")); 205 206 /* SACK block for the received segment. */ 207 head_blk.start = rcv_start; 208 head_blk.end = rcv_end; 209 210 /* 211 * Merge updated SACK blocks into head_blk, and 212 * save unchanged SACK blocks into saved_blks[]. 213 * num_saved will have the number of the saved SACK blocks. 214 */ 215 num_saved = 0; 216 for (i = 0; i < tp->rcv_numsacks; i++) { 217 tcp_seq start = tp->sackblks[i].start; 218 tcp_seq end = tp->sackblks[i].end; 219 if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { 220 /* 221 * Discard this SACK block. 222 */ 223 } else if (SEQ_LEQ(head_blk.start, end) && 224 SEQ_GEQ(head_blk.end, start)) { 225 /* 226 * Merge this SACK block into head_blk. 227 * This SACK block itself will be discarded. 228 */ 229 if (SEQ_GT(head_blk.start, start)) 230 head_blk.start = start; 231 if (SEQ_LT(head_blk.end, end)) 232 head_blk.end = end; 233 } else { 234 /* 235 * Save this SACK block. 236 */ 237 saved_blks[num_saved].start = start; 238 saved_blks[num_saved].end = end; 239 num_saved++; 240 } 241 } 242 243 /* 244 * Update SACK list in tp->sackblks[]. 245 */ 246 num_head = 0; 247 if (SEQ_GT(head_blk.start, tp->rcv_nxt)) { 248 /* 249 * The received data segment is an out-of-order segment. 250 * Put head_blk at the top of SACK list. 251 */ 252 tp->sackblks[0] = head_blk; 253 num_head = 1; 254 /* 255 * If the number of saved SACK blocks exceeds its limit, 256 * discard the last SACK block. 257 */ 258 if (num_saved >= MAX_SACK_BLKS) 259 num_saved--; 260 } 261 if (num_saved > 0) { 262 /* 263 * Copy the saved SACK blocks back. 264 */ 265 bcopy(saved_blks, &tp->sackblks[num_head], 266 sizeof(struct sackblk) * num_saved); 267 } 268 269 /* Save the number of SACK blocks. */ 270 tp->rcv_numsacks = num_head + num_saved; 271 } 272 273 /* 274 * Delete all receiver-side SACK information. 275 */ 276 void 277 tcp_clean_sackreport(tp) 278 struct tcpcb *tp; 279 { 280 int i; 281 282 INP_LOCK_ASSERT(tp->t_inpcb); 283 tp->rcv_numsacks = 0; 284 for (i = 0; i < MAX_SACK_BLKS; i++) 285 tp->sackblks[i].start = tp->sackblks[i].end=0; 286 } 287 288 /* 289 * Allocate struct sackhole. 290 */ 291 static struct sackhole * 292 tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) 293 { 294 struct sackhole *hole; 295 296 if (tp->snd_numholes >= tcp_sack_maxholes || 297 tcp_sack_globalholes >= tcp_sack_globalmaxholes) { 298 tcpstat.tcps_sack_sboverflow++; 299 return NULL; 300 } 301 302 hole = (struct sackhole *)uma_zalloc(sack_hole_zone, M_NOWAIT); 303 if (hole == NULL) 304 return NULL; 305 306 hole->start = start; 307 hole->end = end; 308 hole->rxmit = start; 309 310 tp->snd_numholes++; 311 tcp_sack_globalholes++; 312 313 return hole; 314 } 315 316 /* 317 * Free struct sackhole. 318 */ 319 static void 320 tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) 321 { 322 uma_zfree(sack_hole_zone, hole); 323 324 tp->snd_numholes--; 325 tcp_sack_globalholes--; 326 327 KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0")); 328 KASSERT(tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0")); 329 } 330 331 /* 332 * Process the TCP SACK option. Returns 1 if tcp_dooptions() should continue, 333 * and 0 otherwise, if the option was fine. tp->snd_holes is an ordered list 334 * of holes (oldest to newest, in terms of the sequence space). 335 */ 336 int 337 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 338 { 339 int tmp_olen; 340 u_char *tmp_cp; 341 struct sackhole *cur, *temp; 342 struct sackblk sack, sack_blocks[TCP_MAX_SACK]; 343 int i, j, next_sack_blk, num_sack_blks; 344 345 INP_LOCK_ASSERT(tp->t_inpcb); 346 if (!tp->sack_enable) 347 return (1); 348 if ((th->th_flags & TH_ACK) == 0) 349 return (1); 350 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 351 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 352 return (1); 353 /* If ack is outside [snd_una, snd_max], ignore the SACK options */ 354 if (SEQ_LT(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max)) 355 return (1); 356 tmp_cp = cp + 2; 357 tmp_olen = optlen - 2; 358 tcpstat.tcps_sack_rcv_blocks++; 359 /* 360 * Sort the SACK blocks so we can update the scoreboard 361 * with just one pass. The overhead of sorting upto 4 elements 362 * is less than making 3 passes over the scoreboard. 363 */ 364 num_sack_blks = 0; 365 while (tmp_olen > 0) { 366 bcopy(tmp_cp, &sack, sizeof(sack)); 367 sack.start = ntohl(sack.start); 368 sack.end = ntohl(sack.end); 369 if (SEQ_GT(sack.end, sack.start) && 370 SEQ_GT(sack.start, tp->snd_una) && 371 SEQ_GT(sack.start, th->th_ack) && 372 SEQ_LEQ(sack.end, tp->snd_max)) 373 sack_blocks[num_sack_blks++] = sack; 374 tmp_olen -= TCPOLEN_SACK; 375 tmp_cp += TCPOLEN_SACK; 376 } 377 if (num_sack_blks == 0) 378 return 0; 379 /* Bubble sort */ 380 for (i = 0; i < num_sack_blks; i++) { 381 for (j = i + 1; j < num_sack_blks; j++) { 382 if (SEQ_GT(sack_blocks[i].start, 383 sack_blocks[j].start)){ 384 sack = sack_blocks[i]; 385 sack_blocks[i] = sack_blocks[j]; 386 sack_blocks[j] = sack; 387 } 388 } 389 } 390 if (TAILQ_EMPTY(&tp->snd_holes)) 391 /* 392 * Empty scoreboard. Need to initialize snd_fack (it may be 393 * uninitialized or have a bogus value). Scoreboard holes 394 * (from the sack blocks received) are created later below (in 395 * the logic that adds holes to the tail of the scoreboard). 396 */ 397 tp->snd_fack = tp->snd_una; 398 next_sack_blk = 0; 399 cur = TAILQ_FIRST(&tp->snd_holes); 400 /* 401 * Since the incoming sack blocks are sorted, we can process them 402 * making one sweep of the scoreboard. 403 */ 404 while ((next_sack_blk < num_sack_blks) && (cur != NULL)) { 405 sack = sack_blocks[next_sack_blk]; 406 if (SEQ_LT(tp->snd_fack, sack.start)) 407 /* 408 * The sack block acks data to the right of all the holes 409 * in the scoreboard. No need to iterate over the 410 * scoreboard anymore. 411 */ 412 break; 413 if (SEQ_LEQ(sack.end, cur->start)) { 414 /* 415 * SACKs data before the current hole. 416 * Ignore the sack block. Go to the next sack 417 * block. 418 */ 419 next_sack_blk++; 420 continue; 421 } 422 if (SEQ_GEQ(sack.start, cur->end)) { 423 /* 424 * SACKs data beyond the current hole. 425 * Go to the next hole. 426 */ 427 cur = TAILQ_NEXT(cur, scblink); 428 continue; 429 } 430 tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start); 431 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, 432 ("sackhint bytes rtx >= 0")); 433 if (SEQ_LEQ(sack.start, cur->start)) { 434 /* Data acks at least the beginning of hole */ 435 if (SEQ_GEQ(sack.end, cur->end)) { 436 /* Acks entire hole, so delete hole */ 437 if (tp->sackhint.nexthole == cur) 438 tp->sackhint.nexthole = 439 TAILQ_NEXT(cur, scblink); 440 temp = cur; 441 cur = TAILQ_NEXT(cur, scblink); 442 TAILQ_REMOVE(&tp->snd_holes, temp, scblink); 443 tcp_sackhole_free(tp, temp); 444 /* 445 * The sack block may ack all or part of the next 446 * hole too, so continue onto the next hole. 447 */ 448 continue; 449 } else { 450 /* Move start of hole forward */ 451 cur->start = sack.end; 452 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 453 } 454 } else { 455 if (SEQ_GEQ(sack.end, cur->end)) { 456 /* Move end of hole backward */ 457 cur->end = sack.start; 458 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 459 } else { 460 /* 461 * ACKs some data in middle of a hole; need to 462 * split current hole 463 */ 464 temp = tcp_sackhole_alloc(tp, sack.end, 465 cur->end); 466 if (temp != NULL) { 467 if (SEQ_GT(cur->rxmit, temp->rxmit)) 468 temp->rxmit = cur->rxmit; 469 TAILQ_INSERT_AFTER(&tp->snd_holes, 470 cur, temp, scblink); 471 cur->end = sack.start; 472 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 473 tp->sackhint.sack_bytes_rexmit += 474 (cur->rxmit - cur->start); 475 cur = temp; 476 } 477 } 478 } 479 tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start); 480 /* 481 * Testing sack.end against cur->end tells us whether we're done 482 * with the sack block or the sack hole. Accordingly, we advance 483 * one or the other. 484 */ 485 if (SEQ_GEQ(sack.end, cur->end)) 486 cur = TAILQ_NEXT(cur, scblink); 487 else 488 next_sack_blk++; 489 } 490 /* Iterated all the holes in the scoreboard. Add new holes. */ 491 for ( ; next_sack_blk < num_sack_blks ; next_sack_blk++) { 492 sack = sack_blocks[next_sack_blk]; 493 /* 494 * The two SEQ_LT() checks here that test rcv_laststart against 495 * sack.start and sack.end seem redundant, but they're necessary 496 * to deal with overlapping sack blocks. 497 */ 498 if (SEQ_LT(tp->snd_fack, sack.start)) { 499 /* Need to append new hole at end. */ 500 temp = tcp_sackhole_alloc(tp, tp->snd_fack, 501 sack.start); 502 if (temp == NULL) 503 continue; /* ENOBUFS */ 504 TAILQ_INSERT_TAIL(&tp->snd_holes, temp, scblink); 505 tp->snd_fack = sack.end; 506 if (tp->sackhint.nexthole == NULL) 507 tp->sackhint.nexthole = temp; 508 } 509 if (SEQ_LT(tp->snd_fack, sack.end)) 510 tp->snd_fack = sack.end; 511 } 512 return (0); 513 } 514 515 /* 516 * Delete stale (i.e, cumulatively ack'd) holes. Hole is deleted only if 517 * it is completely acked; otherwise, tcp_sack_option(), called from 518 * tcp_dooptions(), will fix up the hole. 519 */ 520 void 521 tcp_del_sackholes(tp, th) 522 struct tcpcb *tp; 523 struct tcphdr *th; 524 { 525 INP_LOCK_ASSERT(tp->t_inpcb); 526 if (tp->sack_enable && tp->t_state != TCPS_LISTEN) { 527 /* max because this could be an older ack just arrived */ 528 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 529 th->th_ack : tp->snd_una; 530 struct sackhole *cur = TAILQ_FIRST(&tp->snd_holes); 531 struct sackhole *prev; 532 while (cur) 533 if (SEQ_LEQ(cur->end, lastack)) { 534 prev = cur; 535 tp->sackhint.sack_bytes_rexmit -= 536 (cur->rxmit - cur->start); 537 if (tp->sackhint.nexthole == cur) 538 tp->sackhint.nexthole = 539 TAILQ_NEXT(cur, scblink); 540 cur = TAILQ_NEXT(cur, scblink); 541 TAILQ_REMOVE(&tp->snd_holes, prev, scblink); 542 tcp_sackhole_free(tp, prev); 543 } else if (SEQ_LT(cur->start, lastack)) { 544 if (SEQ_LT(cur->rxmit, lastack)) { 545 tp->sackhint.sack_bytes_rexmit -= 546 (cur->rxmit - cur->start); 547 cur->rxmit = lastack; 548 } else 549 tp->sackhint.sack_bytes_rexmit -= 550 (lastack - cur->start); 551 cur->start = lastack; 552 break; 553 } else 554 break; 555 } 556 } 557 558 void 559 tcp_free_sackholes(struct tcpcb *tp) 560 { 561 struct sackhole *q; 562 563 INP_LOCK_ASSERT(tp->t_inpcb); 564 while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) { 565 TAILQ_REMOVE(&tp->snd_holes, q, scblink); 566 tcp_sackhole_free(tp, q); 567 } 568 tp->sackhint.nexthole = NULL; 569 tp->sackhint.sack_bytes_rexmit = 0; 570 571 KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); 572 } 573 574 /* 575 * Partial ack handling within a sack recovery episode. 576 * Keeping this very simple for now. When a partial ack 577 * is received, force snd_cwnd to a value that will allow 578 * the sender to transmit no more than 2 segments. 579 * If necessary, a better scheme can be adopted at a 580 * later point, but for now, the goal is to prevent the 581 * sender from bursting a large amount of data in the midst 582 * of sack recovery. 583 */ 584 void 585 tcp_sack_partialack(tp, th) 586 struct tcpcb *tp; 587 struct tcphdr *th; 588 { 589 int num_segs = 1; 590 591 INP_LOCK_ASSERT(tp->t_inpcb); 592 callout_stop(tp->tt_rexmt); 593 tp->t_rtttime = 0; 594 /* send one or 2 segments based on how much new data was acked */ 595 if (((th->th_ack - tp->snd_una) / tp->t_maxseg) > 2) 596 num_segs = 2; 597 tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + 598 (tp->snd_nxt - tp->sack_newdata) + 599 num_segs * tp->t_maxseg); 600 if (tp->snd_cwnd > tp->snd_ssthresh) 601 tp->snd_cwnd = tp->snd_ssthresh; 602 tp->t_flags |= TF_ACKNOW; 603 (void) tcp_output(tp); 604 } 605 606 /* 607 * Debug version of tcp_sack_output() that walks the scoreboard. Used for 608 * now to sanity check the hint. 609 */ 610 static struct sackhole * 611 tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) 612 { 613 struct sackhole *p; 614 615 INP_LOCK_ASSERT(tp->t_inpcb); 616 *sack_bytes_rexmt = 0; 617 TAILQ_FOREACH(p, &tp->snd_holes, scblink) { 618 if (SEQ_LT(p->rxmit, p->end)) { 619 if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ 620 continue; 621 } 622 *sack_bytes_rexmt += (p->rxmit - p->start); 623 break; 624 } 625 *sack_bytes_rexmt += (p->rxmit - p->start); 626 } 627 return (p); 628 } 629 630 /* 631 * Returns the next hole to retransmit and the number of retransmitted bytes 632 * from the scoreboard. We store both the next hole and the number of 633 * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK 634 * reception). This avoids scoreboard traversals completely. 635 * 636 * The loop here will traverse *at most* one link. Here's the argument. 637 * For the loop to traverse more than 1 link before finding the next hole to 638 * retransmit, we would need to have at least 1 node following the current hint 639 * with (rxmit == end). But, for all holes following the current hint, 640 * (start == rxmit), since we have not yet retransmitted from them. Therefore, 641 * in order to traverse more 1 link in the loop below, we need to have at least 642 * one node following the current hint with (start == rxmit == end). 643 * But that can't happen, (start == end) means that all the data in that hole 644 * has been sacked, in which case, the hole would have been removed from the 645 * scoreboard. 646 */ 647 struct sackhole * 648 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) 649 { 650 struct sackhole *hole = NULL, *dbg_hole = NULL; 651 int dbg_bytes_rexmt; 652 653 INP_LOCK_ASSERT(tp->t_inpcb); 654 dbg_hole = tcp_sack_output_debug(tp, &dbg_bytes_rexmt); 655 *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; 656 hole = tp->sackhint.nexthole; 657 if (hole == NULL || SEQ_LT(hole->rxmit, hole->end)) 658 goto out; 659 while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) { 660 if (SEQ_LT(hole->rxmit, hole->end)) { 661 tp->sackhint.nexthole = hole; 662 break; 663 } 664 } 665 out: 666 if (dbg_hole != hole) { 667 printf("%s: Computed sack hole not the same as cached value\n", __func__); 668 hole = dbg_hole; 669 } 670 if (*sack_bytes_rexmt != dbg_bytes_rexmt) { 671 printf("%s: Computed sack_bytes_retransmitted (%d) not" 672 "the same as cached value (%d)\n", 673 __func__, dbg_bytes_rexmt, *sack_bytes_rexmt); 674 *sack_bytes_rexmt = dbg_bytes_rexmt; 675 } 676 return (hole); 677 } 678 679 /* 680 * After a timeout, the SACK list may be rebuilt. This SACK information 681 * should be used to avoid retransmitting SACKed data. This function 682 * traverses the SACK list to see if snd_nxt should be moved forward. 683 */ 684 void 685 tcp_sack_adjust(struct tcpcb *tp) 686 { 687 struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); 688 689 INP_LOCK_ASSERT(tp->t_inpcb); 690 if (cur == NULL) 691 return; /* No holes */ 692 if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) 693 return; /* We're already beyond any SACKed blocks */ 694 /* 695 * Two cases for which we want to advance snd_nxt: 696 * i) snd_nxt lies between end of one hole and beginning of another 697 * ii) snd_nxt lies between end of last hole and snd_fack 698 */ 699 while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { 700 if (SEQ_LT(tp->snd_nxt, cur->end)) 701 return; 702 if (SEQ_GEQ(tp->snd_nxt, p->start)) 703 cur = p; 704 else { 705 tp->snd_nxt = p->start; 706 return; 707 } 708 } 709 if (SEQ_LT(tp->snd_nxt, cur->end)) 710 return; 711 tp->snd_nxt = tp->snd_fack; 712 return; 713 } 714