1 /*- 2 * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions are met: 6 * 7 * a) Redistributions of source code must retain the above copyright notice, 8 * this list of conditions and the following disclaimer. 9 * 10 * b) Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in 12 * the documentation and/or other materials provided with the distribution. 13 * 14 * c) Neither the name of Cisco Systems, Inc. nor the names of its 15 * contributors may be used to endorse or promote products derived 16 * from this software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 20 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 28 * THE POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31 #include <netinet/sctp_os.h> 32 #include <netinet/sctp_var.h> 33 #include <netinet/sctp_sysctl.h> 34 #include <netinet/sctp_pcb.h> 35 #include <netinet/sctp_header.h> 36 #include <netinet/sctputil.h> 37 #include <netinet/sctp_output.h> 38 #include <netinet/sctp_input.h> 39 #include <netinet/sctp_indata.h> 40 #include <netinet/sctp_uio.h> 41 #include <netinet/sctp_timer.h> 42 #include <netinet/sctp_auth.h> 43 #include <netinet/sctp_asconf.h> 44 #include <netinet/sctp_cc_functions.h> 45 #include <netinet/sctp_dtrace_declare.h> 46 #include <sys/cdefs.h> 47 __FBSDID("$FreeBSD$"); 48 49 void 50 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 51 { 52 struct sctp_association *assoc; 53 uint32_t cwnd_in_mtu; 54 55 assoc = &stcb->asoc; 56 /* 57 * We take the minimum of the burst limit and the initial congestion 58 * window. The initial congestion window is at least two times the 59 * MTU. 60 */ 61 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 62 if ((assoc->max_burst > 0) && (cwnd_in_mtu > assoc->max_burst)) 63 cwnd_in_mtu = assoc->max_burst; 64 net->cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 65 net->ssthresh = assoc->peers_rwnd; 66 67 SDT_PROBE(sctp, cwnd, net, init, 68 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 69 0, net->cwnd); 70 if (SCTP_BASE_SYSCTL(sctp_logging_level) & 71 (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 72 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 73 } 74 } 75 76 void 77 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb, 78 struct sctp_association *asoc) 79 { 80 struct sctp_nets *net; 81 82 /*- 83 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 84 * (net->fast_retran_loss_recovery == 0))) 85 */ 86 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 87 if ((asoc->fast_retran_loss_recovery == 0) || 88 (asoc->sctp_cmt_on_off == 1)) { 89 /* out of a RFC2582 Fast recovery window? */ 90 if (net->net_ack > 0) { 91 /* 92 * per section 7.2.3, are there any 93 * destinations that had a fast retransmit 94 * to them. If so what we need to do is 95 * adjust ssthresh and cwnd. 96 */ 97 struct sctp_tmit_chunk *lchk; 98 int old_cwnd = net->cwnd; 99 100 net->ssthresh = net->cwnd / 2; 101 if (net->ssthresh < (net->mtu * 2)) { 102 net->ssthresh = 2 * net->mtu; 103 } 104 net->cwnd = net->ssthresh; 105 SDT_PROBE(sctp, cwnd, net, fr, 106 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 107 old_cwnd, net->cwnd); 108 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 109 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 110 SCTP_CWND_LOG_FROM_FR); 111 } 112 lchk = TAILQ_FIRST(&asoc->send_queue); 113 114 net->partial_bytes_acked = 0; 115 /* Turn on fast recovery window */ 116 asoc->fast_retran_loss_recovery = 1; 117 if (lchk == NULL) { 118 /* Mark end of the window */ 119 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 120 } else { 121 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 122 } 123 124 /* 125 * CMT fast recovery -- per destination 126 * recovery variable. 127 */ 128 net->fast_retran_loss_recovery = 1; 129 130 if (lchk == NULL) { 131 /* Mark end of the window */ 132 net->fast_recovery_tsn = asoc->sending_seq - 1; 133 } else { 134 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 135 } 136 137 /* 138 * Disable Nonce Sum Checking and store the 139 * resync tsn 140 */ 141 asoc->nonce_sum_check = 0; 142 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 143 144 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 145 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 146 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 147 stcb->sctp_ep, stcb, net); 148 } 149 } else if (net->net_ack > 0) { 150 /* 151 * Mark a peg that we WOULD have done a cwnd 152 * reduction but RFC2582 prevented this action. 153 */ 154 SCTP_STAT_INCR(sctps_fastretransinrtt); 155 } 156 } 157 } 158 159 void 160 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 161 struct sctp_association *asoc, 162 int accum_moved, int reneged_all, int will_exit) 163 { 164 struct sctp_nets *net; 165 int old_cwnd; 166 167 /******************************/ 168 /* update cwnd and Early FR */ 169 /******************************/ 170 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 171 172 #ifdef JANA_CMT_FAST_RECOVERY 173 /* 174 * CMT fast recovery code. Need to debug. 175 */ 176 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 177 if (compare_with_wrap(asoc->last_acked_seq, 178 net->fast_recovery_tsn, MAX_TSN) || 179 (asoc->last_acked_seq == net->fast_recovery_tsn) || 180 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 181 (net->pseudo_cumack == net->fast_recovery_tsn)) { 182 net->will_exit_fast_recovery = 1; 183 } 184 } 185 #endif 186 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 187 /* 188 * So, first of all do we need to have a Early FR 189 * timer running? 190 */ 191 if ((!TAILQ_EMPTY(&asoc->sent_queue) && 192 (net->ref_count > 1) && 193 (net->flight_size < net->cwnd)) || 194 (reneged_all)) { 195 /* 196 * yes, so in this case stop it if its 197 * running, and then restart it. Reneging 198 * all is a special case where we want to 199 * run the Early FR timer and then force the 200 * last few unacked to be sent, causing us 201 * to illicit a sack with gaps to force out 202 * the others. 203 */ 204 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 205 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 206 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 207 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 208 } 209 SCTP_STAT_INCR(sctps_earlyfrstrid); 210 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 211 } else { 212 /* No, stop it if its running */ 213 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 214 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 215 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 216 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 217 } 218 } 219 } 220 /* if nothing was acked on this destination skip it */ 221 if (net->net_ack == 0) { 222 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 223 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 224 } 225 continue; 226 } 227 if (net->net_ack2 > 0) { 228 /* 229 * Karn's rule applies to clearing error count, this 230 * is optional. 231 */ 232 net->error_count = 0; 233 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 234 SCTP_ADDR_NOT_REACHABLE) { 235 /* addr came good */ 236 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 237 net->dest_state |= SCTP_ADDR_REACHABLE; 238 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 239 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 240 /* now was it the primary? if so restore */ 241 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 242 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 243 } 244 } 245 /* 246 * JRS 5/14/07 - If CMT PF is on and the destination 247 * is in PF state, set the destination to active 248 * state and set the cwnd to one or two MTU's based 249 * on whether PF1 or PF2 is being used. 250 * 251 * Should we stop any running T3 timer here? 252 */ 253 if ((asoc->sctp_cmt_on_off == 1) && 254 (asoc->sctp_cmt_pf > 0) && 255 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 256 net->dest_state &= ~SCTP_ADDR_PF; 257 old_cwnd = net->cwnd; 258 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 259 SDT_PROBE(sctp, cwnd, net, ack, 260 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 261 old_cwnd, net->cwnd); 262 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 263 net, net->cwnd); 264 /* 265 * Since the cwnd value is explicitly set, 266 * skip the code that updates the cwnd 267 * value. 268 */ 269 goto skip_cwnd_update; 270 } 271 } 272 #ifdef JANA_CMT_FAST_RECOVERY 273 /* 274 * CMT fast recovery code 275 */ 276 /* 277 * if (sctp_cmt_on_off == 1 && 278 * net->fast_retran_loss_recovery && 279 * net->will_exit_fast_recovery == 0) { @@@ Do something } 280 * else if (sctp_cmt_on_off == 0 && 281 * asoc->fast_retran_loss_recovery && will_exit == 0) { 282 */ 283 #endif 284 285 if (asoc->fast_retran_loss_recovery && 286 (will_exit == 0) && 287 (asoc->sctp_cmt_on_off == 0)) { 288 /* 289 * If we are in loss recovery we skip any cwnd 290 * update 291 */ 292 goto skip_cwnd_update; 293 } 294 /* 295 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 296 * moved. 297 */ 298 if (accum_moved || 299 ((asoc->sctp_cmt_on_off == 1) && net->new_pseudo_cumack)) { 300 /* If the cumulative ack moved we can proceed */ 301 if (net->cwnd <= net->ssthresh) { 302 /* We are in slow start */ 303 if (net->flight_size + net->net_ack >= net->cwnd) { 304 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 305 old_cwnd = net->cwnd; 306 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 307 SDT_PROBE(sctp, cwnd, net, ack, 308 stcb->asoc.my_vtag, 309 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 310 net, 311 old_cwnd, net->cwnd); 312 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 313 sctp_log_cwnd(stcb, net, net->mtu, 314 SCTP_CWND_LOG_FROM_SS); 315 } 316 } else { 317 old_cwnd = net->cwnd; 318 net->cwnd += net->net_ack; 319 SDT_PROBE(sctp, cwnd, net, ack, 320 stcb->asoc.my_vtag, 321 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 322 net, 323 old_cwnd, net->cwnd); 324 325 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 326 sctp_log_cwnd(stcb, net, net->net_ack, 327 SCTP_CWND_LOG_FROM_SS); 328 } 329 } 330 } else { 331 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 332 sctp_log_cwnd(stcb, net, net->net_ack, 333 SCTP_CWND_LOG_NOADV_SS); 334 } 335 } 336 } else { 337 /* We are in congestion avoidance */ 338 /* 339 * Add to pba 340 */ 341 net->partial_bytes_acked += net->net_ack; 342 343 if ((net->flight_size + net->net_ack >= net->cwnd) && 344 (net->partial_bytes_acked >= net->cwnd)) { 345 net->partial_bytes_acked -= net->cwnd; 346 old_cwnd = net->cwnd; 347 net->cwnd += net->mtu; 348 SDT_PROBE(sctp, cwnd, net, ack, 349 stcb->asoc.my_vtag, 350 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 351 net, 352 old_cwnd, net->cwnd); 353 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 354 sctp_log_cwnd(stcb, net, net->mtu, 355 SCTP_CWND_LOG_FROM_CA); 356 } 357 } else { 358 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 359 sctp_log_cwnd(stcb, net, net->net_ack, 360 SCTP_CWND_LOG_NOADV_CA); 361 } 362 } 363 } 364 } else { 365 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 366 sctp_log_cwnd(stcb, net, net->mtu, 367 SCTP_CWND_LOG_NO_CUMACK); 368 } 369 } 370 skip_cwnd_update: 371 /* 372 * NOW, according to Karn's rule do we need to restore the 373 * RTO timer back? Check our net_ack2. If not set then we 374 * have a ambiguity.. i.e. all data ack'd was sent to more 375 * than one place. 376 */ 377 if (net->net_ack2) { 378 /* restore any doubled timers */ 379 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 380 if (net->RTO < stcb->asoc.minrto) { 381 net->RTO = stcb->asoc.minrto; 382 } 383 if (net->RTO > stcb->asoc.maxrto) { 384 net->RTO = stcb->asoc.maxrto; 385 } 386 } 387 } 388 } 389 390 void 391 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 392 { 393 int old_cwnd = net->cwnd; 394 395 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 396 net->cwnd = net->mtu; 397 net->partial_bytes_acked = 0; 398 SDT_PROBE(sctp, cwnd, net, to, 399 stcb->asoc.my_vtag, 400 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 401 net, 402 old_cwnd, net->cwnd); 403 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 404 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 405 } 406 } 407 408 void 409 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net) 410 { 411 int old_cwnd = net->cwnd; 412 413 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 414 net->ssthresh = net->cwnd / 2; 415 if (net->ssthresh < net->mtu) { 416 net->ssthresh = net->mtu; 417 /* here back off the timer as well, to slow us down */ 418 net->RTO <<= 1; 419 } 420 net->cwnd = net->ssthresh; 421 SDT_PROBE(sctp, cwnd, net, ecn, 422 stcb->asoc.my_vtag, 423 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 424 net, 425 old_cwnd, net->cwnd); 426 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 427 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 428 } 429 } 430 431 void 432 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 433 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 434 uint32_t * bottle_bw, uint32_t * on_queue) 435 { 436 uint32_t bw_avail; 437 int rtt, incr; 438 int old_cwnd = net->cwnd; 439 440 /* need real RTT for this calc */ 441 rtt = ((net->lastsa >> 2) + net->lastsv) >> 1; 442 /* get bottle neck bw */ 443 *bottle_bw = ntohl(cp->bottle_bw); 444 /* and whats on queue */ 445 *on_queue = ntohl(cp->current_onq); 446 /* 447 * adjust the on-queue if our flight is more it could be that the 448 * router has not yet gotten data "in-flight" to it 449 */ 450 if (*on_queue < net->flight_size) 451 *on_queue = net->flight_size; 452 /* calculate the available space */ 453 bw_avail = (*bottle_bw * rtt) / 1000; 454 if (bw_avail > *bottle_bw) { 455 /* 456 * Cap the growth to no more than the bottle neck. This can 457 * happen as RTT slides up due to queues. It also means if 458 * you have more than a 1 second RTT with a empty queue you 459 * will be limited to the bottle_bw per second no matter if 460 * other points have 1/2 the RTT and you could get more 461 * out... 462 */ 463 bw_avail = *bottle_bw; 464 } 465 if (*on_queue > bw_avail) { 466 /* 467 * No room for anything else don't allow anything else to be 468 * "added to the fire". 469 */ 470 int seg_inflight, seg_onqueue, my_portion; 471 472 net->partial_bytes_acked = 0; 473 474 /* how much are we over queue size? */ 475 incr = *on_queue - bw_avail; 476 if (stcb->asoc.seen_a_sack_this_pkt) { 477 /* 478 * undo any cwnd adjustment that the sack might have 479 * made 480 */ 481 net->cwnd = net->prev_cwnd; 482 } 483 /* Now how much of that is mine? */ 484 seg_inflight = net->flight_size / net->mtu; 485 seg_onqueue = *on_queue / net->mtu; 486 my_portion = (incr * seg_inflight) / seg_onqueue; 487 488 /* Have I made an adjustment already */ 489 if (net->cwnd > net->flight_size) { 490 /* 491 * for this flight I made an adjustment we need to 492 * decrease the portion by a share our previous 493 * adjustment. 494 */ 495 int diff_adj; 496 497 diff_adj = net->cwnd - net->flight_size; 498 if (diff_adj > my_portion) 499 my_portion = 0; 500 else 501 my_portion -= diff_adj; 502 } 503 /* 504 * back down to the previous cwnd (assume we have had a sack 505 * before this packet). minus what ever portion of the 506 * overage is my fault. 507 */ 508 net->cwnd -= my_portion; 509 510 /* we will NOT back down more than 1 MTU */ 511 if (net->cwnd <= net->mtu) { 512 net->cwnd = net->mtu; 513 } 514 /* force into CA */ 515 net->ssthresh = net->cwnd - 1; 516 } else { 517 /* 518 * Take 1/4 of the space left or max burst up .. whichever 519 * is less. 520 */ 521 incr = min((bw_avail - *on_queue) >> 2, 522 stcb->asoc.max_burst * net->mtu); 523 net->cwnd += incr; 524 } 525 if (net->cwnd > bw_avail) { 526 /* We can't exceed the pipe size */ 527 net->cwnd = bw_avail; 528 } 529 if (net->cwnd < net->mtu) { 530 /* We always have 1 MTU */ 531 net->cwnd = net->mtu; 532 } 533 if (net->cwnd - old_cwnd != 0) { 534 /* log only changes */ 535 SDT_PROBE(sctp, cwnd, net, pd, 536 stcb->asoc.my_vtag, 537 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 538 net, 539 old_cwnd, net->cwnd); 540 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 541 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 542 SCTP_CWND_LOG_FROM_SAT); 543 } 544 } 545 } 546 547 void 548 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 549 struct sctp_nets *net, int burst_limit) 550 { 551 int old_cwnd = net->cwnd; 552 553 if (net->ssthresh < net->cwnd) 554 net->ssthresh = net->cwnd; 555 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 556 SDT_PROBE(sctp, cwnd, net, bl, 557 stcb->asoc.my_vtag, 558 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 559 net, 560 old_cwnd, net->cwnd); 561 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 562 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 563 } 564 } 565 566 void 567 sctp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 568 struct sctp_tcb *stcb, struct sctp_nets *net) 569 { 570 int old_cwnd = net->cwnd; 571 572 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 573 /* 574 * make a small adjustment to cwnd and force to CA. 575 */ 576 if (net->cwnd > net->mtu) 577 /* drop down one MTU after sending */ 578 net->cwnd -= net->mtu; 579 if (net->cwnd < net->ssthresh) 580 /* still in SS move to CA */ 581 net->ssthresh = net->cwnd - 1; 582 SDT_PROBE(sctp, cwnd, net, fr, 583 stcb->asoc.my_vtag, 584 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 585 net, 586 old_cwnd, net->cwnd); 587 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 588 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 589 } 590 } 591 592 struct sctp_hs_raise_drop { 593 int32_t cwnd; 594 int32_t increase; 595 int32_t drop_percent; 596 }; 597 598 #define SCTP_HS_TABLE_SIZE 73 599 600 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 601 {38, 1, 50}, /* 0 */ 602 {118, 2, 44}, /* 1 */ 603 {221, 3, 41}, /* 2 */ 604 {347, 4, 38}, /* 3 */ 605 {495, 5, 37}, /* 4 */ 606 {663, 6, 35}, /* 5 */ 607 {851, 7, 34}, /* 6 */ 608 {1058, 8, 33}, /* 7 */ 609 {1284, 9, 32}, /* 8 */ 610 {1529, 10, 31}, /* 9 */ 611 {1793, 11, 30}, /* 10 */ 612 {2076, 12, 29}, /* 11 */ 613 {2378, 13, 28}, /* 12 */ 614 {2699, 14, 28}, /* 13 */ 615 {3039, 15, 27}, /* 14 */ 616 {3399, 16, 27}, /* 15 */ 617 {3778, 17, 26}, /* 16 */ 618 {4177, 18, 26}, /* 17 */ 619 {4596, 19, 25}, /* 18 */ 620 {5036, 20, 25}, /* 19 */ 621 {5497, 21, 24}, /* 20 */ 622 {5979, 22, 24}, /* 21 */ 623 {6483, 23, 23}, /* 22 */ 624 {7009, 24, 23}, /* 23 */ 625 {7558, 25, 22}, /* 24 */ 626 {8130, 26, 22}, /* 25 */ 627 {8726, 27, 22}, /* 26 */ 628 {9346, 28, 21}, /* 27 */ 629 {9991, 29, 21}, /* 28 */ 630 {10661, 30, 21}, /* 29 */ 631 {11358, 31, 20}, /* 30 */ 632 {12082, 32, 20}, /* 31 */ 633 {12834, 33, 20}, /* 32 */ 634 {13614, 34, 19}, /* 33 */ 635 {14424, 35, 19}, /* 34 */ 636 {15265, 36, 19}, /* 35 */ 637 {16137, 37, 19}, /* 36 */ 638 {17042, 38, 18}, /* 37 */ 639 {17981, 39, 18}, /* 38 */ 640 {18955, 40, 18}, /* 39 */ 641 {19965, 41, 17}, /* 40 */ 642 {21013, 42, 17}, /* 41 */ 643 {22101, 43, 17}, /* 42 */ 644 {23230, 44, 17}, /* 43 */ 645 {24402, 45, 16}, /* 44 */ 646 {25618, 46, 16}, /* 45 */ 647 {26881, 47, 16}, /* 46 */ 648 {28193, 48, 16}, /* 47 */ 649 {29557, 49, 15}, /* 48 */ 650 {30975, 50, 15}, /* 49 */ 651 {32450, 51, 15}, /* 50 */ 652 {33986, 52, 15}, /* 51 */ 653 {35586, 53, 14}, /* 52 */ 654 {37253, 54, 14}, /* 53 */ 655 {38992, 55, 14}, /* 54 */ 656 {40808, 56, 14}, /* 55 */ 657 {42707, 57, 13}, /* 56 */ 658 {44694, 58, 13}, /* 57 */ 659 {46776, 59, 13}, /* 58 */ 660 {48961, 60, 13}, /* 59 */ 661 {51258, 61, 13}, /* 60 */ 662 {53677, 62, 12}, /* 61 */ 663 {56230, 63, 12}, /* 62 */ 664 {58932, 64, 12}, /* 63 */ 665 {61799, 65, 12}, /* 64 */ 666 {64851, 66, 11}, /* 65 */ 667 {68113, 67, 11}, /* 66 */ 668 {71617, 68, 11}, /* 67 */ 669 {75401, 69, 10}, /* 68 */ 670 {79517, 70, 10}, /* 69 */ 671 {84035, 71, 10}, /* 70 */ 672 {89053, 72, 10}, /* 71 */ 673 {94717, 73, 9} /* 72 */ 674 }; 675 676 static void 677 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 678 { 679 int cur_val, i, indx, incr; 680 681 cur_val = net->cwnd >> 10; 682 indx = SCTP_HS_TABLE_SIZE - 1; 683 #ifdef SCTP_DEBUG 684 printf("HS CC CAlled.\n"); 685 #endif 686 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 687 /* normal mode */ 688 if (net->net_ack > net->mtu) { 689 net->cwnd += net->mtu; 690 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 691 sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS); 692 } 693 } else { 694 net->cwnd += net->net_ack; 695 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 696 sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS); 697 } 698 } 699 } else { 700 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 701 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 702 indx = i; 703 break; 704 } 705 } 706 net->last_hs_used = indx; 707 incr = ((sctp_cwnd_adjust[indx].increase) << 10); 708 net->cwnd += incr; 709 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 710 sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS); 711 } 712 } 713 } 714 715 static void 716 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 717 { 718 int cur_val, i, indx; 719 int old_cwnd = net->cwnd; 720 721 cur_val = net->cwnd >> 10; 722 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 723 /* normal mode */ 724 net->ssthresh = net->cwnd / 2; 725 if (net->ssthresh < (net->mtu * 2)) { 726 net->ssthresh = 2 * net->mtu; 727 } 728 net->cwnd = net->ssthresh; 729 } else { 730 /* drop by the proper amount */ 731 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 732 sctp_cwnd_adjust[net->last_hs_used].drop_percent); 733 net->cwnd = net->ssthresh; 734 /* now where are we */ 735 indx = net->last_hs_used; 736 cur_val = net->cwnd >> 10; 737 /* reset where we are in the table */ 738 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 739 /* feel out of hs */ 740 net->last_hs_used = 0; 741 } else { 742 for (i = indx; i >= 1; i--) { 743 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 744 break; 745 } 746 } 747 net->last_hs_used = indx; 748 } 749 } 750 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 751 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 752 } 753 } 754 755 void 756 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 757 struct sctp_association *asoc) 758 { 759 struct sctp_nets *net; 760 761 /* 762 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 763 * (net->fast_retran_loss_recovery == 0))) 764 */ 765 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 766 if ((asoc->fast_retran_loss_recovery == 0) || 767 (asoc->sctp_cmt_on_off == 1)) { 768 /* out of a RFC2582 Fast recovery window? */ 769 if (net->net_ack > 0) { 770 /* 771 * per section 7.2.3, are there any 772 * destinations that had a fast retransmit 773 * to them. If so what we need to do is 774 * adjust ssthresh and cwnd. 775 */ 776 struct sctp_tmit_chunk *lchk; 777 778 sctp_hs_cwnd_decrease(stcb, net); 779 780 lchk = TAILQ_FIRST(&asoc->send_queue); 781 782 net->partial_bytes_acked = 0; 783 /* Turn on fast recovery window */ 784 asoc->fast_retran_loss_recovery = 1; 785 if (lchk == NULL) { 786 /* Mark end of the window */ 787 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 788 } else { 789 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 790 } 791 792 /* 793 * CMT fast recovery -- per destination 794 * recovery variable. 795 */ 796 net->fast_retran_loss_recovery = 1; 797 798 if (lchk == NULL) { 799 /* Mark end of the window */ 800 net->fast_recovery_tsn = asoc->sending_seq - 1; 801 } else { 802 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 803 } 804 805 /* 806 * Disable Nonce Sum Checking and store the 807 * resync tsn 808 */ 809 asoc->nonce_sum_check = 0; 810 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 811 812 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 813 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 814 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 815 stcb->sctp_ep, stcb, net); 816 } 817 } else if (net->net_ack > 0) { 818 /* 819 * Mark a peg that we WOULD have done a cwnd 820 * reduction but RFC2582 prevented this action. 821 */ 822 SCTP_STAT_INCR(sctps_fastretransinrtt); 823 } 824 } 825 } 826 827 void 828 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 829 struct sctp_association *asoc, 830 int accum_moved, int reneged_all, int will_exit) 831 { 832 struct sctp_nets *net; 833 834 /******************************/ 835 /* update cwnd and Early FR */ 836 /******************************/ 837 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 838 839 #ifdef JANA_CMT_FAST_RECOVERY 840 /* 841 * CMT fast recovery code. Need to debug. 842 */ 843 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 844 if (compare_with_wrap(asoc->last_acked_seq, 845 net->fast_recovery_tsn, MAX_TSN) || 846 (asoc->last_acked_seq == net->fast_recovery_tsn) || 847 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 848 (net->pseudo_cumack == net->fast_recovery_tsn)) { 849 net->will_exit_fast_recovery = 1; 850 } 851 } 852 #endif 853 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 854 /* 855 * So, first of all do we need to have a Early FR 856 * timer running? 857 */ 858 if ((!TAILQ_EMPTY(&asoc->sent_queue) && 859 (net->ref_count > 1) && 860 (net->flight_size < net->cwnd)) || 861 (reneged_all)) { 862 /* 863 * yes, so in this case stop it if its 864 * running, and then restart it. Reneging 865 * all is a special case where we want to 866 * run the Early FR timer and then force the 867 * last few unacked to be sent, causing us 868 * to illicit a sack with gaps to force out 869 * the others. 870 */ 871 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 872 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 873 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 874 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 875 } 876 SCTP_STAT_INCR(sctps_earlyfrstrid); 877 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 878 } else { 879 /* No, stop it if its running */ 880 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 881 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 882 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 883 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 884 } 885 } 886 } 887 /* if nothing was acked on this destination skip it */ 888 if (net->net_ack == 0) { 889 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 890 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 891 } 892 continue; 893 } 894 if (net->net_ack2 > 0) { 895 /* 896 * Karn's rule applies to clearing error count, this 897 * is optional. 898 */ 899 net->error_count = 0; 900 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 901 SCTP_ADDR_NOT_REACHABLE) { 902 /* addr came good */ 903 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 904 net->dest_state |= SCTP_ADDR_REACHABLE; 905 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 906 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 907 /* now was it the primary? if so restore */ 908 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 909 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 910 } 911 } 912 /* 913 * JRS 5/14/07 - If CMT PF is on and the destination 914 * is in PF state, set the destination to active 915 * state and set the cwnd to one or two MTU's based 916 * on whether PF1 or PF2 is being used. 917 * 918 * Should we stop any running T3 timer here? 919 */ 920 if ((asoc->sctp_cmt_on_off == 1) && 921 (asoc->sctp_cmt_pf > 0) && 922 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 923 net->dest_state &= ~SCTP_ADDR_PF; 924 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 925 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 926 net, net->cwnd); 927 /* 928 * Since the cwnd value is explicitly set, 929 * skip the code that updates the cwnd 930 * value. 931 */ 932 goto skip_cwnd_update; 933 } 934 } 935 #ifdef JANA_CMT_FAST_RECOVERY 936 /* 937 * CMT fast recovery code 938 */ 939 /* 940 * if (sctp_cmt_on_off == 1 && 941 * net->fast_retran_loss_recovery && 942 * net->will_exit_fast_recovery == 0) { @@@ Do something } 943 * else if (sctp_cmt_on_off == 0 && 944 * asoc->fast_retran_loss_recovery && will_exit == 0) { 945 */ 946 #endif 947 948 if (asoc->fast_retran_loss_recovery && 949 (will_exit == 0) && 950 (asoc->sctp_cmt_on_off == 0)) { 951 /* 952 * If we are in loss recovery we skip any cwnd 953 * update 954 */ 955 goto skip_cwnd_update; 956 } 957 /* 958 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 959 * moved. 960 */ 961 if (accum_moved || 962 ((asoc->sctp_cmt_on_off == 1) && net->new_pseudo_cumack)) { 963 /* If the cumulative ack moved we can proceed */ 964 if (net->cwnd <= net->ssthresh) { 965 /* We are in slow start */ 966 if (net->flight_size + net->net_ack >= net->cwnd) { 967 968 sctp_hs_cwnd_increase(stcb, net); 969 970 } else { 971 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 972 sctp_log_cwnd(stcb, net, net->net_ack, 973 SCTP_CWND_LOG_NOADV_SS); 974 } 975 } 976 } else { 977 /* We are in congestion avoidance */ 978 net->partial_bytes_acked += net->net_ack; 979 if ((net->flight_size + net->net_ack >= net->cwnd) && 980 (net->partial_bytes_acked >= net->cwnd)) { 981 net->partial_bytes_acked -= net->cwnd; 982 net->cwnd += net->mtu; 983 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 984 sctp_log_cwnd(stcb, net, net->mtu, 985 SCTP_CWND_LOG_FROM_CA); 986 } 987 } else { 988 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 989 sctp_log_cwnd(stcb, net, net->net_ack, 990 SCTP_CWND_LOG_NOADV_CA); 991 } 992 } 993 } 994 } else { 995 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 996 sctp_log_cwnd(stcb, net, net->mtu, 997 SCTP_CWND_LOG_NO_CUMACK); 998 } 999 } 1000 skip_cwnd_update: 1001 /* 1002 * NOW, according to Karn's rule do we need to restore the 1003 * RTO timer back? Check our net_ack2. If not set then we 1004 * have a ambiguity.. i.e. all data ack'd was sent to more 1005 * than one place. 1006 */ 1007 if (net->net_ack2) { 1008 /* restore any doubled timers */ 1009 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 1010 if (net->RTO < stcb->asoc.minrto) { 1011 net->RTO = stcb->asoc.minrto; 1012 } 1013 if (net->RTO > stcb->asoc.maxrto) { 1014 net->RTO = stcb->asoc.maxrto; 1015 } 1016 } 1017 } 1018 } 1019 1020 1021 /* 1022 * H-TCP congestion control. The algorithm is detailed in: 1023 * R.N.Shorten, D.J.Leith: 1024 * "H-TCP: TCP for high-speed and long-distance networks" 1025 * Proc. PFLDnet, Argonne, 2004. 1026 * http://www.hamilton.ie/net/htcp3.pdf 1027 */ 1028 1029 1030 static int use_rtt_scaling = 1; 1031 static int use_bandwidth_switch = 1; 1032 1033 static inline int 1034 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1035 { 1036 return seq3 - seq2 >= seq1 - seq2; 1037 } 1038 1039 static inline uint32_t 1040 htcp_cong_time(struct htcp *ca) 1041 { 1042 return sctp_get_tick_count() - ca->last_cong; 1043 } 1044 1045 static inline uint32_t 1046 htcp_ccount(struct htcp *ca) 1047 { 1048 return htcp_cong_time(ca) / ca->minRTT; 1049 } 1050 1051 static inline void 1052 htcp_reset(struct htcp *ca) 1053 { 1054 ca->undo_last_cong = ca->last_cong; 1055 ca->undo_maxRTT = ca->maxRTT; 1056 ca->undo_old_maxB = ca->old_maxB; 1057 ca->last_cong = sctp_get_tick_count(); 1058 } 1059 1060 #ifdef SCTP_NOT_USED 1061 1062 static uint32_t 1063 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1064 { 1065 net->htcp_ca.last_cong = net->htcp_ca.undo_last_cong; 1066 net->htcp_ca.maxRTT = net->htcp_ca.undo_maxRTT; 1067 net->htcp_ca.old_maxB = net->htcp_ca.undo_old_maxB; 1068 return max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->htcp_ca.beta) * net->mtu); 1069 } 1070 1071 #endif 1072 1073 static inline void 1074 measure_rtt(struct sctp_tcb *stcb, struct sctp_nets *net) 1075 { 1076 uint32_t srtt = net->lastsa >> 3; 1077 1078 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1079 if (net->htcp_ca.minRTT > srtt || !net->htcp_ca.minRTT) 1080 net->htcp_ca.minRTT = srtt; 1081 1082 /* max RTT */ 1083 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->htcp_ca) > 3) { 1084 if (net->htcp_ca.maxRTT < net->htcp_ca.minRTT) 1085 net->htcp_ca.maxRTT = net->htcp_ca.minRTT; 1086 if (net->htcp_ca.maxRTT < srtt && srtt <= net->htcp_ca.maxRTT + MSEC_TO_TICKS(20)) 1087 net->htcp_ca.maxRTT = srtt; 1088 } 1089 } 1090 1091 static void 1092 measure_achieved_throughput(struct sctp_tcb *stcb, struct sctp_nets *net) 1093 { 1094 uint32_t now = sctp_get_tick_count(); 1095 1096 if (net->fast_retran_ip == 0) 1097 net->htcp_ca.bytes_acked = net->net_ack; 1098 1099 if (!use_bandwidth_switch) 1100 return; 1101 1102 /* achieved throughput calculations */ 1103 /* JRS - not 100% sure of this statement */ 1104 if (net->fast_retran_ip == 1) { 1105 net->htcp_ca.bytecount = 0; 1106 net->htcp_ca.lasttime = now; 1107 return; 1108 } 1109 net->htcp_ca.bytecount += net->net_ack; 1110 1111 if (net->htcp_ca.bytecount >= net->cwnd - ((net->htcp_ca.alpha >> 7 ? : 1) * net->mtu) 1112 && now - net->htcp_ca.lasttime >= net->htcp_ca.minRTT 1113 && net->htcp_ca.minRTT > 0) { 1114 uint32_t cur_Bi = net->htcp_ca.bytecount / net->mtu * hz / (now - net->htcp_ca.lasttime); 1115 1116 if (htcp_ccount(&net->htcp_ca) <= 3) { 1117 /* just after backoff */ 1118 net->htcp_ca.minB = net->htcp_ca.maxB = net->htcp_ca.Bi = cur_Bi; 1119 } else { 1120 net->htcp_ca.Bi = (3 * net->htcp_ca.Bi + cur_Bi) / 4; 1121 if (net->htcp_ca.Bi > net->htcp_ca.maxB) 1122 net->htcp_ca.maxB = net->htcp_ca.Bi; 1123 if (net->htcp_ca.minB > net->htcp_ca.maxB) 1124 net->htcp_ca.minB = net->htcp_ca.maxB; 1125 } 1126 net->htcp_ca.bytecount = 0; 1127 net->htcp_ca.lasttime = now; 1128 } 1129 } 1130 1131 static inline void 1132 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1133 { 1134 if (use_bandwidth_switch) { 1135 uint32_t maxB = ca->maxB; 1136 uint32_t old_maxB = ca->old_maxB; 1137 1138 ca->old_maxB = ca->maxB; 1139 1140 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1141 ca->beta = BETA_MIN; 1142 ca->modeswitch = 0; 1143 return; 1144 } 1145 } 1146 if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) { 1147 ca->beta = (minRTT << 7) / maxRTT; 1148 if (ca->beta < BETA_MIN) 1149 ca->beta = BETA_MIN; 1150 else if (ca->beta > BETA_MAX) 1151 ca->beta = BETA_MAX; 1152 } else { 1153 ca->beta = BETA_MIN; 1154 ca->modeswitch = 1; 1155 } 1156 } 1157 1158 static inline void 1159 htcp_alpha_update(struct htcp *ca) 1160 { 1161 uint32_t minRTT = ca->minRTT; 1162 uint32_t factor = 1; 1163 uint32_t diff = htcp_cong_time(ca); 1164 1165 if (diff > (uint32_t) hz) { 1166 diff -= hz; 1167 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1168 } 1169 if (use_rtt_scaling && minRTT) { 1170 uint32_t scale = (hz << 3) / (10 * minRTT); 1171 1172 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 1173 * interval [0.5,10]<<3 */ 1174 factor = (factor << 3) / scale; 1175 if (!factor) 1176 factor = 1; 1177 } 1178 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 1179 if (!ca->alpha) 1180 ca->alpha = ALPHA_BASE; 1181 } 1182 1183 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 1184 * rtt before we adjust our beta to ensure we are working from a consistent 1185 * data. 1186 * 1187 * This function should be called when we hit a congestion event since only at 1188 * that point do we really have a real sense of maxRTT (the queues en route 1189 * were getting just too full now). 1190 */ 1191 static void 1192 htcp_param_update(struct sctp_tcb *stcb, struct sctp_nets *net) 1193 { 1194 uint32_t minRTT = net->htcp_ca.minRTT; 1195 uint32_t maxRTT = net->htcp_ca.maxRTT; 1196 1197 htcp_beta_update(&net->htcp_ca, minRTT, maxRTT); 1198 htcp_alpha_update(&net->htcp_ca); 1199 1200 /* 1201 * add slowly fading memory for maxRTT to accommodate routing 1202 * changes etc 1203 */ 1204 if (minRTT > 0 && maxRTT > minRTT) 1205 net->htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 1206 } 1207 1208 static uint32_t 1209 htcp_recalc_ssthresh(struct sctp_tcb *stcb, struct sctp_nets *net) 1210 { 1211 htcp_param_update(stcb, net); 1212 return max(((net->cwnd / net->mtu * net->htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu); 1213 } 1214 1215 static void 1216 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 1217 { 1218 /*- 1219 * How to handle these functions? 1220 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 1221 * return; 1222 */ 1223 if (net->cwnd <= net->ssthresh) { 1224 /* We are in slow start */ 1225 if (net->flight_size + net->net_ack >= net->cwnd) { 1226 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 1227 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 1228 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1229 sctp_log_cwnd(stcb, net, net->mtu, 1230 SCTP_CWND_LOG_FROM_SS); 1231 } 1232 } else { 1233 net->cwnd += net->net_ack; 1234 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1235 sctp_log_cwnd(stcb, net, net->net_ack, 1236 SCTP_CWND_LOG_FROM_SS); 1237 } 1238 } 1239 } else { 1240 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1241 sctp_log_cwnd(stcb, net, net->net_ack, 1242 SCTP_CWND_LOG_NOADV_SS); 1243 } 1244 } 1245 } else { 1246 measure_rtt(stcb, net); 1247 1248 /* 1249 * In dangerous area, increase slowly. In theory this is 1250 * net->cwnd += alpha / net->cwnd 1251 */ 1252 /* What is snd_cwnd_cnt?? */ 1253 if (((net->partial_bytes_acked / net->mtu * net->htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 1254 /*- 1255 * Does SCTP have a cwnd clamp? 1256 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 1257 */ 1258 net->cwnd += net->mtu; 1259 net->partial_bytes_acked = 0; 1260 htcp_alpha_update(&net->htcp_ca); 1261 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1262 sctp_log_cwnd(stcb, net, net->mtu, 1263 SCTP_CWND_LOG_FROM_CA); 1264 } 1265 } else { 1266 net->partial_bytes_acked += net->net_ack; 1267 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1268 sctp_log_cwnd(stcb, net, net->net_ack, 1269 SCTP_CWND_LOG_NOADV_CA); 1270 } 1271 } 1272 1273 net->htcp_ca.bytes_acked = net->mtu; 1274 } 1275 } 1276 1277 #ifdef SCTP_NOT_USED 1278 /* Lower bound on congestion window. */ 1279 static uint32_t 1280 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 1281 { 1282 return net->ssthresh; 1283 } 1284 1285 #endif 1286 1287 static void 1288 htcp_init(struct sctp_tcb *stcb, struct sctp_nets *net) 1289 { 1290 memset(&net->htcp_ca, 0, sizeof(struct htcp)); 1291 net->htcp_ca.alpha = ALPHA_BASE; 1292 net->htcp_ca.beta = BETA_MIN; 1293 net->htcp_ca.bytes_acked = net->mtu; 1294 net->htcp_ca.last_cong = sctp_get_tick_count(); 1295 } 1296 1297 void 1298 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 1299 { 1300 /* 1301 * We take the max of the burst limit times a MTU or the 1302 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 1303 */ 1304 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1305 net->ssthresh = stcb->asoc.peers_rwnd; 1306 htcp_init(stcb, net); 1307 1308 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 1309 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 1310 } 1311 } 1312 1313 void 1314 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1315 struct sctp_association *asoc, 1316 int accum_moved, int reneged_all, int will_exit) 1317 { 1318 struct sctp_nets *net; 1319 1320 /******************************/ 1321 /* update cwnd and Early FR */ 1322 /******************************/ 1323 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1324 1325 #ifdef JANA_CMT_FAST_RECOVERY 1326 /* 1327 * CMT fast recovery code. Need to debug. 1328 */ 1329 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1330 if (compare_with_wrap(asoc->last_acked_seq, 1331 net->fast_recovery_tsn, MAX_TSN) || 1332 (asoc->last_acked_seq == net->fast_recovery_tsn) || 1333 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 1334 (net->pseudo_cumack == net->fast_recovery_tsn)) { 1335 net->will_exit_fast_recovery = 1; 1336 } 1337 } 1338 #endif 1339 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 1340 /* 1341 * So, first of all do we need to have a Early FR 1342 * timer running? 1343 */ 1344 if ((!TAILQ_EMPTY(&asoc->sent_queue) && 1345 (net->ref_count > 1) && 1346 (net->flight_size < net->cwnd)) || 1347 (reneged_all)) { 1348 /* 1349 * yes, so in this case stop it if its 1350 * running, and then restart it. Reneging 1351 * all is a special case where we want to 1352 * run the Early FR timer and then force the 1353 * last few unacked to be sent, causing us 1354 * to illicit a sack with gaps to force out 1355 * the others. 1356 */ 1357 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1358 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 1359 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1360 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 1361 } 1362 SCTP_STAT_INCR(sctps_earlyfrstrid); 1363 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 1364 } else { 1365 /* No, stop it if its running */ 1366 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1367 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 1368 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1369 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 1370 } 1371 } 1372 } 1373 /* if nothing was acked on this destination skip it */ 1374 if (net->net_ack == 0) { 1375 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1376 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1377 } 1378 continue; 1379 } 1380 if (net->net_ack2 > 0) { 1381 /* 1382 * Karn's rule applies to clearing error count, this 1383 * is optional. 1384 */ 1385 net->error_count = 0; 1386 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 1387 SCTP_ADDR_NOT_REACHABLE) { 1388 /* addr came good */ 1389 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 1390 net->dest_state |= SCTP_ADDR_REACHABLE; 1391 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 1392 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 1393 /* now was it the primary? if so restore */ 1394 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 1395 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 1396 } 1397 } 1398 /* 1399 * JRS 5/14/07 - If CMT PF is on and the destination 1400 * is in PF state, set the destination to active 1401 * state and set the cwnd to one or two MTU's based 1402 * on whether PF1 or PF2 is being used. 1403 * 1404 * Should we stop any running T3 timer here? 1405 */ 1406 if ((asoc->sctp_cmt_on_off == 1) && 1407 (asoc->sctp_cmt_pf > 0) && 1408 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 1409 net->dest_state &= ~SCTP_ADDR_PF; 1410 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 1411 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1412 net, net->cwnd); 1413 /* 1414 * Since the cwnd value is explicitly set, 1415 * skip the code that updates the cwnd 1416 * value. 1417 */ 1418 goto skip_cwnd_update; 1419 } 1420 } 1421 #ifdef JANA_CMT_FAST_RECOVERY 1422 /* 1423 * CMT fast recovery code 1424 */ 1425 /* 1426 * if (sctp_cmt_on_off == 1 && 1427 * net->fast_retran_loss_recovery && 1428 * net->will_exit_fast_recovery == 0) { @@@ Do something } 1429 * else if (sctp_cmt_on_off == 0 && 1430 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1431 */ 1432 #endif 1433 1434 if (asoc->fast_retran_loss_recovery && 1435 will_exit == 0 && 1436 (asoc->sctp_cmt_on_off == 0)) { 1437 /* 1438 * If we are in loss recovery we skip any cwnd 1439 * update 1440 */ 1441 goto skip_cwnd_update; 1442 } 1443 /* 1444 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1445 * moved. 1446 */ 1447 if (accum_moved || 1448 ((asoc->sctp_cmt_on_off == 1) && net->new_pseudo_cumack)) { 1449 htcp_cong_avoid(stcb, net); 1450 measure_achieved_throughput(stcb, net); 1451 } else { 1452 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1453 sctp_log_cwnd(stcb, net, net->mtu, 1454 SCTP_CWND_LOG_NO_CUMACK); 1455 } 1456 } 1457 skip_cwnd_update: 1458 /* 1459 * NOW, according to Karn's rule do we need to restore the 1460 * RTO timer back? Check our net_ack2. If not set then we 1461 * have a ambiguity.. i.e. all data ack'd was sent to more 1462 * than one place. 1463 */ 1464 if (net->net_ack2) { 1465 /* restore any doubled timers */ 1466 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 1467 if (net->RTO < stcb->asoc.minrto) { 1468 net->RTO = stcb->asoc.minrto; 1469 } 1470 if (net->RTO > stcb->asoc.maxrto) { 1471 net->RTO = stcb->asoc.maxrto; 1472 } 1473 } 1474 } 1475 } 1476 1477 void 1478 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 1479 struct sctp_association *asoc) 1480 { 1481 struct sctp_nets *net; 1482 1483 /* 1484 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 1485 * (net->fast_retran_loss_recovery == 0))) 1486 */ 1487 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1488 if ((asoc->fast_retran_loss_recovery == 0) || 1489 (asoc->sctp_cmt_on_off == 1)) { 1490 /* out of a RFC2582 Fast recovery window? */ 1491 if (net->net_ack > 0) { 1492 /* 1493 * per section 7.2.3, are there any 1494 * destinations that had a fast retransmit 1495 * to them. If so what we need to do is 1496 * adjust ssthresh and cwnd. 1497 */ 1498 struct sctp_tmit_chunk *lchk; 1499 int old_cwnd = net->cwnd; 1500 1501 /* JRS - reset as if state were changed */ 1502 htcp_reset(&net->htcp_ca); 1503 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1504 net->cwnd = net->ssthresh; 1505 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1506 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1507 SCTP_CWND_LOG_FROM_FR); 1508 } 1509 lchk = TAILQ_FIRST(&asoc->send_queue); 1510 1511 net->partial_bytes_acked = 0; 1512 /* Turn on fast recovery window */ 1513 asoc->fast_retran_loss_recovery = 1; 1514 if (lchk == NULL) { 1515 /* Mark end of the window */ 1516 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1517 } else { 1518 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1519 } 1520 1521 /* 1522 * CMT fast recovery -- per destination 1523 * recovery variable. 1524 */ 1525 net->fast_retran_loss_recovery = 1; 1526 1527 if (lchk == NULL) { 1528 /* Mark end of the window */ 1529 net->fast_recovery_tsn = asoc->sending_seq - 1; 1530 } else { 1531 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1532 } 1533 1534 /* 1535 * Disable Nonce Sum Checking and store the 1536 * resync tsn 1537 */ 1538 asoc->nonce_sum_check = 0; 1539 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 1540 1541 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1542 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 1543 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1544 stcb->sctp_ep, stcb, net); 1545 } 1546 } else if (net->net_ack > 0) { 1547 /* 1548 * Mark a peg that we WOULD have done a cwnd 1549 * reduction but RFC2582 prevented this action. 1550 */ 1551 SCTP_STAT_INCR(sctps_fastretransinrtt); 1552 } 1553 } 1554 } 1555 1556 void 1557 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 1558 struct sctp_nets *net) 1559 { 1560 int old_cwnd = net->cwnd; 1561 1562 /* JRS - reset as if the state were being changed to timeout */ 1563 htcp_reset(&net->htcp_ca); 1564 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1565 net->cwnd = net->mtu; 1566 net->partial_bytes_acked = 0; 1567 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1568 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1569 } 1570 } 1571 1572 void 1573 sctp_htcp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 1574 struct sctp_tcb *stcb, struct sctp_nets *net) 1575 { 1576 int old_cwnd; 1577 1578 old_cwnd = net->cwnd; 1579 1580 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 1581 net->htcp_ca.last_cong = sctp_get_tick_count(); 1582 /* 1583 * make a small adjustment to cwnd and force to CA. 1584 */ 1585 if (net->cwnd > net->mtu) 1586 /* drop down one MTU after sending */ 1587 net->cwnd -= net->mtu; 1588 if (net->cwnd < net->ssthresh) 1589 /* still in SS move to CA */ 1590 net->ssthresh = net->cwnd - 1; 1591 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1592 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 1593 } 1594 } 1595 1596 void 1597 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 1598 struct sctp_nets *net) 1599 { 1600 int old_cwnd; 1601 1602 old_cwnd = net->cwnd; 1603 1604 /* JRS - reset hctp as if state changed */ 1605 htcp_reset(&net->htcp_ca); 1606 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1607 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1608 if (net->ssthresh < net->mtu) { 1609 net->ssthresh = net->mtu; 1610 /* here back off the timer as well, to slow us down */ 1611 net->RTO <<= 1; 1612 } 1613 net->cwnd = net->ssthresh; 1614 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1615 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1616 } 1617 } 1618