1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved. 5 * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. 6 * Copyright (c) 2008-2012, by Michael Tuexen. 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 are met: 10 * 11 * a) Redistributions of source code must retain the above copyright notice, 12 * this list of conditions and the following disclaimer. 13 * 14 * b) Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the distribution. 17 * 18 * c) Neither the name of Cisco Systems, Inc. nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 24 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 26 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 32 * THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 #include <netinet/sctp_os.h> 39 #include <netinet/sctp_var.h> 40 #include <netinet/sctp_sysctl.h> 41 #include <netinet/sctp_pcb.h> 42 #include <netinet/sctp_header.h> 43 #include <netinet/sctputil.h> 44 #include <netinet/sctp_output.h> 45 #include <netinet/sctp_input.h> 46 #include <netinet/sctp_indata.h> 47 #include <netinet/sctp_uio.h> 48 #include <netinet/sctp_timer.h> 49 #include <netinet/sctp_auth.h> 50 #include <netinet/sctp_asconf.h> 51 #include <netinet/sctp_kdtrace.h> 52 53 #define SHIFT_MPTCP_MULTI_N 40 54 #define SHIFT_MPTCP_MULTI_Z 16 55 #define SHIFT_MPTCP_MULTI 8 56 57 static void 58 sctp_enforce_cwnd_limit(struct sctp_association *assoc, struct sctp_nets *net) 59 { 60 if ((assoc->max_cwnd > 0) && 61 (net->cwnd > assoc->max_cwnd) && 62 (net->cwnd > (net->mtu - sizeof(struct sctphdr)))) { 63 net->cwnd = assoc->max_cwnd; 64 if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) { 65 net->cwnd = net->mtu - sizeof(struct sctphdr); 66 } 67 } 68 } 69 70 static void 71 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 72 { 73 struct sctp_association *assoc; 74 uint32_t cwnd_in_mtu; 75 76 assoc = &stcb->asoc; 77 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 78 if (cwnd_in_mtu == 0) { 79 /* Using 0 means that the value of RFC 4960 is used. */ 80 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 81 } else { 82 /* 83 * We take the minimum of the burst limit and the initial 84 * congestion window. 85 */ 86 if ((assoc->max_burst > 0) && (cwnd_in_mtu > assoc->max_burst)) 87 cwnd_in_mtu = assoc->max_burst; 88 net->cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 89 } 90 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 91 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) { 92 /* In case of resource pooling initialize appropriately */ 93 net->cwnd /= assoc->numnets; 94 if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) { 95 net->cwnd = net->mtu - sizeof(struct sctphdr); 96 } 97 } 98 sctp_enforce_cwnd_limit(assoc, net); 99 net->ssthresh = assoc->peers_rwnd; 100 SDT_PROBE5(sctp, cwnd, net, init, 101 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 102 0, net->cwnd); 103 if (SCTP_BASE_SYSCTL(sctp_logging_level) & 104 (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 105 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 106 } 107 } 108 109 static void 110 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb, 111 struct sctp_association *asoc) 112 { 113 struct sctp_nets *net; 114 uint32_t t_ssthresh, t_cwnd; 115 uint64_t t_ucwnd_sbw; 116 117 /* MT FIXME: Don't compute this over and over again */ 118 t_ssthresh = 0; 119 t_cwnd = 0; 120 t_ucwnd_sbw = 0; 121 if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) || 122 (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) { 123 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 124 t_ssthresh += net->ssthresh; 125 t_cwnd += net->cwnd; 126 if (net->lastsa > 0) { 127 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)net->lastsa; 128 } 129 } 130 if (t_ucwnd_sbw == 0) { 131 t_ucwnd_sbw = 1; 132 } 133 } 134 135 /*- 136 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 137 * (net->fast_retran_loss_recovery == 0))) 138 */ 139 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 140 if ((asoc->fast_retran_loss_recovery == 0) || 141 (asoc->sctp_cmt_on_off > 0)) { 142 /* out of a RFC2582 Fast recovery window? */ 143 if (net->net_ack > 0) { 144 /* 145 * per section 7.2.3, are there any 146 * destinations that had a fast retransmit 147 * to them. If so what we need to do is 148 * adjust ssthresh and cwnd. 149 */ 150 struct sctp_tmit_chunk *lchk; 151 int old_cwnd = net->cwnd; 152 153 if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) || 154 (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) { 155 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) { 156 net->ssthresh = (uint32_t)(((uint64_t)4 * 157 (uint64_t)net->mtu * 158 (uint64_t)net->ssthresh) / 159 (uint64_t)t_ssthresh); 160 } 161 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2) { 162 uint32_t srtt; 163 164 srtt = net->lastsa; 165 /* 166 * lastsa>>3; we don't need 167 * to devide ... 168 */ 169 if (srtt == 0) { 170 srtt = 1; 171 } 172 /* 173 * Short Version => Equal to 174 * Contel Version MBe 175 */ 176 net->ssthresh = (uint32_t)(((uint64_t)4 * 177 (uint64_t)net->mtu * 178 (uint64_t)net->cwnd) / 179 ((uint64_t)srtt * 180 t_ucwnd_sbw)); 181 /* INCREASE FACTOR */ ; 182 } 183 if ((net->cwnd > t_cwnd / 2) && 184 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 185 net->ssthresh = net->cwnd - t_cwnd / 2; 186 } 187 if (net->ssthresh < net->mtu) { 188 net->ssthresh = net->mtu; 189 } 190 } else { 191 net->ssthresh = net->cwnd / 2; 192 if (net->ssthresh < (net->mtu * 2)) { 193 net->ssthresh = 2 * net->mtu; 194 } 195 } 196 net->cwnd = net->ssthresh; 197 sctp_enforce_cwnd_limit(asoc, net); 198 SDT_PROBE5(sctp, cwnd, net, fr, 199 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 200 old_cwnd, net->cwnd); 201 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 202 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 203 SCTP_CWND_LOG_FROM_FR); 204 } 205 lchk = TAILQ_FIRST(&asoc->send_queue); 206 207 net->partial_bytes_acked = 0; 208 /* Turn on fast recovery window */ 209 asoc->fast_retran_loss_recovery = 1; 210 if (lchk == NULL) { 211 /* Mark end of the window */ 212 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 213 } else { 214 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 215 } 216 217 /* 218 * CMT fast recovery -- per destination 219 * recovery variable. 220 */ 221 net->fast_retran_loss_recovery = 1; 222 223 if (lchk == NULL) { 224 /* Mark end of the window */ 225 net->fast_recovery_tsn = asoc->sending_seq - 1; 226 } else { 227 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 228 } 229 230 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 231 stcb->sctp_ep, stcb, net, 232 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_1); 233 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 234 stcb->sctp_ep, stcb, net); 235 } 236 } else if (net->net_ack > 0) { 237 /* 238 * Mark a peg that we WOULD have done a cwnd 239 * reduction but RFC2582 prevented this action. 240 */ 241 SCTP_STAT_INCR(sctps_fastretransinrtt); 242 } 243 } 244 } 245 246 /* Defines for instantaneous bw decisions */ 247 #define SCTP_INST_LOOSING 1 /* Losing to other flows */ 248 #define SCTP_INST_NEUTRAL 2 /* Neutral, no indication */ 249 #define SCTP_INST_GAINING 3 /* Gaining, step down possible */ 250 251 static int 252 cc_bw_same(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, 253 uint64_t rtt_offset, uint64_t vtag, uint8_t inst_ind) 254 { 255 uint64_t oth, probepoint; 256 257 probepoint = (((uint64_t)net->cwnd) << 32); 258 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 259 /* 260 * rtt increased we don't update bw.. so we don't update the 261 * rtt either. 262 */ 263 /* Probe point 5 */ 264 probepoint |= ((5 << 16) | 1); 265 SDT_PROBE5(sctp, cwnd, net, rttvar, 266 vtag, 267 ((net->cc_mod.rtcc.lbw << 32) | nbw), 268 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 269 net->flight_size, 270 probepoint); 271 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 272 if (net->cc_mod.rtcc.last_step_state == 5) 273 net->cc_mod.rtcc.step_cnt++; 274 else 275 net->cc_mod.rtcc.step_cnt = 1; 276 net->cc_mod.rtcc.last_step_state = 5; 277 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 278 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 279 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 280 /* Try a step down */ 281 oth = net->cc_mod.rtcc.vol_reduce; 282 oth <<= 16; 283 oth |= net->cc_mod.rtcc.step_cnt; 284 oth <<= 16; 285 oth |= net->cc_mod.rtcc.last_step_state; 286 SDT_PROBE5(sctp, cwnd, net, rttstep, 287 vtag, 288 ((net->cc_mod.rtcc.lbw << 32) | nbw), 289 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 290 oth, 291 probepoint); 292 if (net->cwnd > (4 * net->mtu)) { 293 net->cwnd -= net->mtu; 294 net->cc_mod.rtcc.vol_reduce++; 295 } else { 296 net->cc_mod.rtcc.step_cnt = 0; 297 } 298 } 299 } 300 return (1); 301 } 302 if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 303 /* 304 * rtt decreased, there could be more room. we update both 305 * the bw and the rtt here to lock this in as a good step 306 * down. 307 */ 308 /* Probe point 6 */ 309 probepoint |= ((6 << 16) | 0); 310 SDT_PROBE5(sctp, cwnd, net, rttvar, 311 vtag, 312 ((net->cc_mod.rtcc.lbw << 32) | nbw), 313 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 314 net->flight_size, 315 probepoint); 316 if (net->cc_mod.rtcc.steady_step) { 317 oth = net->cc_mod.rtcc.vol_reduce; 318 oth <<= 16; 319 oth |= net->cc_mod.rtcc.step_cnt; 320 oth <<= 16; 321 oth |= net->cc_mod.rtcc.last_step_state; 322 SDT_PROBE5(sctp, cwnd, net, rttstep, 323 vtag, 324 ((net->cc_mod.rtcc.lbw << 32) | nbw), 325 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 326 oth, 327 probepoint); 328 if ((net->cc_mod.rtcc.last_step_state == 5) && 329 (net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step)) { 330 /* Step down worked */ 331 net->cc_mod.rtcc.step_cnt = 0; 332 return (1); 333 } else { 334 net->cc_mod.rtcc.last_step_state = 6; 335 net->cc_mod.rtcc.step_cnt = 0; 336 } 337 } 338 net->cc_mod.rtcc.lbw = nbw; 339 net->cc_mod.rtcc.lbw_rtt = net->rtt; 340 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 341 if (inst_ind == SCTP_INST_GAINING) 342 return (1); 343 else if (inst_ind == SCTP_INST_NEUTRAL) 344 return (1); 345 else 346 return (0); 347 } 348 /* 349 * Ok bw and rtt remained the same .. no update to any 350 */ 351 /* Probe point 7 */ 352 probepoint |= ((7 << 16) | net->cc_mod.rtcc.ret_from_eq); 353 SDT_PROBE5(sctp, cwnd, net, rttvar, 354 vtag, 355 ((net->cc_mod.rtcc.lbw << 32) | nbw), 356 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 357 net->flight_size, 358 probepoint); 359 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 360 if (net->cc_mod.rtcc.last_step_state == 5) 361 net->cc_mod.rtcc.step_cnt++; 362 else 363 net->cc_mod.rtcc.step_cnt = 1; 364 net->cc_mod.rtcc.last_step_state = 5; 365 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 366 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 367 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 368 /* Try a step down */ 369 if (net->cwnd > (4 * net->mtu)) { 370 net->cwnd -= net->mtu; 371 net->cc_mod.rtcc.vol_reduce++; 372 return (1); 373 } else { 374 net->cc_mod.rtcc.step_cnt = 0; 375 } 376 } 377 } 378 if (inst_ind == SCTP_INST_GAINING) 379 return (1); 380 else if (inst_ind == SCTP_INST_NEUTRAL) 381 return (1); 382 else 383 return ((int)net->cc_mod.rtcc.ret_from_eq); 384 } 385 386 static int 387 cc_bw_decrease(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset, 388 uint64_t vtag, uint8_t inst_ind) 389 { 390 uint64_t oth, probepoint; 391 392 /* Bandwidth decreased. */ 393 probepoint = (((uint64_t)net->cwnd) << 32); 394 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 395 /* rtt increased */ 396 /* Did we add more */ 397 if ((net->cwnd > net->cc_mod.rtcc.cwnd_at_bw_set) && 398 (inst_ind != SCTP_INST_LOOSING)) { 399 /* We caused it maybe.. back off? */ 400 /* PROBE POINT 1 */ 401 probepoint |= ((1 << 16) | 1); 402 SDT_PROBE5(sctp, cwnd, net, rttvar, 403 vtag, 404 ((net->cc_mod.rtcc.lbw << 32) | nbw), 405 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 406 net->flight_size, 407 probepoint); 408 if (net->cc_mod.rtcc.ret_from_eq) { 409 /* 410 * Switch over to CA if we are less 411 * aggressive 412 */ 413 net->ssthresh = net->cwnd - 1; 414 net->partial_bytes_acked = 0; 415 } 416 return (1); 417 } 418 /* Probe point 2 */ 419 probepoint |= ((2 << 16) | 0); 420 SDT_PROBE5(sctp, cwnd, net, rttvar, 421 vtag, 422 ((net->cc_mod.rtcc.lbw << 32) | nbw), 423 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 424 net->flight_size, 425 probepoint); 426 /* Someone else - fight for more? */ 427 if (net->cc_mod.rtcc.steady_step) { 428 oth = net->cc_mod.rtcc.vol_reduce; 429 oth <<= 16; 430 oth |= net->cc_mod.rtcc.step_cnt; 431 oth <<= 16; 432 oth |= net->cc_mod.rtcc.last_step_state; 433 SDT_PROBE5(sctp, cwnd, net, rttstep, 434 vtag, 435 ((net->cc_mod.rtcc.lbw << 32) | nbw), 436 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 437 oth, 438 probepoint); 439 /* 440 * Did we voluntarily give up some? if so take one 441 * back please 442 */ 443 if ((net->cc_mod.rtcc.vol_reduce) && 444 (inst_ind != SCTP_INST_GAINING)) { 445 net->cwnd += net->mtu; 446 sctp_enforce_cwnd_limit(&stcb->asoc, net); 447 net->cc_mod.rtcc.vol_reduce--; 448 } 449 net->cc_mod.rtcc.last_step_state = 2; 450 net->cc_mod.rtcc.step_cnt = 0; 451 } 452 goto out_decision; 453 } else if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 454 /* bw & rtt decreased */ 455 /* Probe point 3 */ 456 probepoint |= ((3 << 16) | 0); 457 SDT_PROBE5(sctp, cwnd, net, rttvar, 458 vtag, 459 ((net->cc_mod.rtcc.lbw << 32) | nbw), 460 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 461 net->flight_size, 462 probepoint); 463 if (net->cc_mod.rtcc.steady_step) { 464 oth = net->cc_mod.rtcc.vol_reduce; 465 oth <<= 16; 466 oth |= net->cc_mod.rtcc.step_cnt; 467 oth <<= 16; 468 oth |= net->cc_mod.rtcc.last_step_state; 469 SDT_PROBE5(sctp, cwnd, net, rttstep, 470 vtag, 471 ((net->cc_mod.rtcc.lbw << 32) | nbw), 472 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 473 oth, 474 probepoint); 475 if ((net->cc_mod.rtcc.vol_reduce) && 476 (inst_ind != SCTP_INST_GAINING)) { 477 net->cwnd += net->mtu; 478 sctp_enforce_cwnd_limit(&stcb->asoc, net); 479 net->cc_mod.rtcc.vol_reduce--; 480 } 481 net->cc_mod.rtcc.last_step_state = 3; 482 net->cc_mod.rtcc.step_cnt = 0; 483 } 484 goto out_decision; 485 } 486 /* The bw decreased but rtt stayed the same */ 487 /* Probe point 4 */ 488 probepoint |= ((4 << 16) | 0); 489 SDT_PROBE5(sctp, cwnd, net, rttvar, 490 vtag, 491 ((net->cc_mod.rtcc.lbw << 32) | nbw), 492 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 493 net->flight_size, 494 probepoint); 495 if (net->cc_mod.rtcc.steady_step) { 496 oth = net->cc_mod.rtcc.vol_reduce; 497 oth <<= 16; 498 oth |= net->cc_mod.rtcc.step_cnt; 499 oth <<= 16; 500 oth |= net->cc_mod.rtcc.last_step_state; 501 SDT_PROBE5(sctp, cwnd, net, rttstep, 502 vtag, 503 ((net->cc_mod.rtcc.lbw << 32) | nbw), 504 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 505 oth, 506 probepoint); 507 if ((net->cc_mod.rtcc.vol_reduce) && 508 (inst_ind != SCTP_INST_GAINING)) { 509 net->cwnd += net->mtu; 510 sctp_enforce_cwnd_limit(&stcb->asoc, net); 511 net->cc_mod.rtcc.vol_reduce--; 512 } 513 net->cc_mod.rtcc.last_step_state = 4; 514 net->cc_mod.rtcc.step_cnt = 0; 515 } 516 out_decision: 517 net->cc_mod.rtcc.lbw = nbw; 518 net->cc_mod.rtcc.lbw_rtt = net->rtt; 519 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 520 if (inst_ind == SCTP_INST_GAINING) { 521 return (1); 522 } else { 523 return (0); 524 } 525 } 526 527 static int 528 cc_bw_increase(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t vtag) 529 { 530 uint64_t oth, probepoint; 531 532 /* 533 * BW increased, so update and return 0, since all actions in our 534 * table say to do the normal CC update. Note that we pay no 535 * attention to the inst_ind since our overall sum is increasing. 536 */ 537 /* PROBE POINT 0 */ 538 probepoint = (((uint64_t)net->cwnd) << 32); 539 SDT_PROBE5(sctp, cwnd, net, rttvar, 540 vtag, 541 ((net->cc_mod.rtcc.lbw << 32) | nbw), 542 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 543 net->flight_size, 544 probepoint); 545 if (net->cc_mod.rtcc.steady_step) { 546 oth = net->cc_mod.rtcc.vol_reduce; 547 oth <<= 16; 548 oth |= net->cc_mod.rtcc.step_cnt; 549 oth <<= 16; 550 oth |= net->cc_mod.rtcc.last_step_state; 551 SDT_PROBE5(sctp, cwnd, net, rttstep, 552 vtag, 553 ((net->cc_mod.rtcc.lbw << 32) | nbw), 554 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 555 oth, 556 probepoint); 557 net->cc_mod.rtcc.last_step_state = 0; 558 net->cc_mod.rtcc.step_cnt = 0; 559 net->cc_mod.rtcc.vol_reduce = 0; 560 } 561 net->cc_mod.rtcc.lbw = nbw; 562 net->cc_mod.rtcc.lbw_rtt = net->rtt; 563 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 564 return (0); 565 } 566 567 /* RTCC Algorithm to limit growth of cwnd, return 568 * true if you want to NOT allow cwnd growth 569 */ 570 static int 571 cc_bw_limit(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw) 572 { 573 uint64_t bw_offset, rtt_offset; 574 uint64_t probepoint, rtt, vtag; 575 uint64_t bytes_for_this_rtt, inst_bw; 576 uint64_t div, inst_off; 577 int bw_shift; 578 uint8_t inst_ind; 579 int ret; 580 581 /*- 582 * Here we need to see if we want 583 * to limit cwnd growth due to increase 584 * in overall rtt but no increase in bw. 585 * We use the following table to figure 586 * out what we should do. When we return 587 * 0, cc update goes on as planned. If we 588 * return 1, then no cc update happens and cwnd 589 * stays where it is at. 590 * ---------------------------------- 591 * BW | RTT | Action 592 * ********************************* 593 * INC | INC | return 0 594 * ---------------------------------- 595 * INC | SAME | return 0 596 * ---------------------------------- 597 * INC | DECR | return 0 598 * ---------------------------------- 599 * SAME | INC | return 1 600 * ---------------------------------- 601 * SAME | SAME | return 1 602 * ---------------------------------- 603 * SAME | DECR | return 0 604 * ---------------------------------- 605 * DECR | INC | return 0 or 1 based on if we caused. 606 * ---------------------------------- 607 * DECR | SAME | return 0 608 * ---------------------------------- 609 * DECR | DECR | return 0 610 * ---------------------------------- 611 * 612 * We are a bit fuzz on what an increase or 613 * decrease is. For BW it is the same if 614 * it did not change within 1/64th. For 615 * RTT it stayed the same if it did not 616 * change within 1/32nd 617 */ 618 bw_shift = SCTP_BASE_SYSCTL(sctp_rttvar_bw); 619 rtt = stcb->asoc.my_vtag; 620 vtag = (rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | (stcb->rport); 621 probepoint = (((uint64_t)net->cwnd) << 32); 622 rtt = net->rtt; 623 if (net->cc_mod.rtcc.rtt_set_this_sack) { 624 net->cc_mod.rtcc.rtt_set_this_sack = 0; 625 bytes_for_this_rtt = net->cc_mod.rtcc.bw_bytes - net->cc_mod.rtcc.bw_bytes_at_last_rttc; 626 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 627 if (net->rtt) { 628 div = net->rtt / 1000; 629 if (div) { 630 inst_bw = bytes_for_this_rtt / div; 631 inst_off = inst_bw >> bw_shift; 632 if (inst_bw > nbw) 633 inst_ind = SCTP_INST_GAINING; 634 else if ((inst_bw + inst_off) < nbw) 635 inst_ind = SCTP_INST_LOOSING; 636 else 637 inst_ind = SCTP_INST_NEUTRAL; 638 probepoint |= ((0xb << 16) | inst_ind); 639 } else { 640 inst_ind = net->cc_mod.rtcc.last_inst_ind; 641 inst_bw = bytes_for_this_rtt / (uint64_t)(net->rtt); 642 /* Can't determine do not change */ 643 probepoint |= ((0xc << 16) | inst_ind); 644 } 645 } else { 646 inst_ind = net->cc_mod.rtcc.last_inst_ind; 647 inst_bw = bytes_for_this_rtt; 648 /* Can't determine do not change */ 649 probepoint |= ((0xd << 16) | inst_ind); 650 } 651 SDT_PROBE5(sctp, cwnd, net, rttvar, 652 vtag, 653 ((nbw << 32) | inst_bw), 654 ((net->cc_mod.rtcc.lbw_rtt << 32) | rtt), 655 net->flight_size, 656 probepoint); 657 } else { 658 /* No rtt measurement, use last one */ 659 inst_ind = net->cc_mod.rtcc.last_inst_ind; 660 } 661 bw_offset = net->cc_mod.rtcc.lbw >> bw_shift; 662 if (nbw > net->cc_mod.rtcc.lbw + bw_offset) { 663 ret = cc_bw_increase(stcb, net, nbw, vtag); 664 goto out; 665 } 666 rtt_offset = net->cc_mod.rtcc.lbw_rtt >> SCTP_BASE_SYSCTL(sctp_rttvar_rtt); 667 if (nbw < net->cc_mod.rtcc.lbw - bw_offset) { 668 ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, vtag, inst_ind); 669 goto out; 670 } 671 /* 672 * If we reach here then we are in a situation where the bw stayed 673 * the same. 674 */ 675 ret = cc_bw_same(stcb, net, nbw, rtt_offset, vtag, inst_ind); 676 out: 677 net->cc_mod.rtcc.last_inst_ind = inst_ind; 678 return (ret); 679 } 680 681 static void 682 sctp_cwnd_update_after_sack_common(struct sctp_tcb *stcb, 683 struct sctp_association *asoc, 684 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit, int use_rtcc) 685 { 686 struct sctp_nets *net; 687 int old_cwnd; 688 uint32_t t_ssthresh, incr; 689 uint64_t t_ucwnd_sbw; 690 uint64_t t_path_mptcp; 691 uint64_t mptcp_like_alpha; 692 uint32_t srtt; 693 uint64_t max_path; 694 695 /* MT FIXME: Don't compute this over and over again */ 696 t_ssthresh = 0; 697 t_ucwnd_sbw = 0; 698 t_path_mptcp = 0; 699 mptcp_like_alpha = 1; 700 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 701 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2) || 702 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_MPTCP)) { 703 max_path = 0; 704 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 705 t_ssthresh += net->ssthresh; 706 /* lastsa>>3; we don't need to devide ... */ 707 srtt = net->lastsa; 708 if (srtt > 0) { 709 uint64_t tmp; 710 711 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)srtt; 712 t_path_mptcp += (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_Z) / 713 (((uint64_t)net->mtu) * (uint64_t)srtt); 714 tmp = (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_N) / 715 ((uint64_t)net->mtu * (uint64_t)(srtt * srtt)); 716 if (tmp > max_path) { 717 max_path = tmp; 718 } 719 } 720 } 721 if (t_path_mptcp > 0) { 722 mptcp_like_alpha = max_path / (t_path_mptcp * t_path_mptcp); 723 } else { 724 mptcp_like_alpha = 1; 725 } 726 } 727 if (t_ssthresh == 0) { 728 t_ssthresh = 1; 729 } 730 if (t_ucwnd_sbw == 0) { 731 t_ucwnd_sbw = 1; 732 } 733 /******************************/ 734 /* update cwnd and Early FR */ 735 /******************************/ 736 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 737 #ifdef JANA_CMT_FAST_RECOVERY 738 /* 739 * CMT fast recovery code. Need to debug. 740 */ 741 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 742 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 743 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 744 net->will_exit_fast_recovery = 1; 745 } 746 } 747 #endif 748 /* if nothing was acked on this destination skip it */ 749 if (net->net_ack == 0) { 750 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 751 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 752 } 753 continue; 754 } 755 #ifdef JANA_CMT_FAST_RECOVERY 756 /* 757 * CMT fast recovery code 758 */ 759 /* 760 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 761 * && net->will_exit_fast_recovery == 0) { @@@ Do something 762 * } else if (sctp_cmt_on_off == 0 && 763 * asoc->fast_retran_loss_recovery && will_exit == 0) { 764 */ 765 #endif 766 767 if (asoc->fast_retran_loss_recovery && 768 (will_exit == 0) && 769 (asoc->sctp_cmt_on_off == 0)) { 770 /* 771 * If we are in loss recovery we skip any cwnd 772 * update 773 */ 774 return; 775 } 776 /* 777 * Did any measurements go on for this network? 778 */ 779 if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) { 780 uint64_t nbw; 781 782 /* 783 * At this point our bw_bytes has been updated by 784 * incoming sack information. 785 * 786 * But our bw may not yet be set. 787 * 788 */ 789 if ((net->cc_mod.rtcc.new_tot_time / 1000) > 0) { 790 nbw = net->cc_mod.rtcc.bw_bytes / (net->cc_mod.rtcc.new_tot_time / 1000); 791 } else { 792 nbw = net->cc_mod.rtcc.bw_bytes; 793 } 794 if (net->cc_mod.rtcc.lbw) { 795 if (cc_bw_limit(stcb, net, nbw)) { 796 /* Hold here, no update */ 797 continue; 798 } 799 } else { 800 uint64_t vtag, probepoint; 801 802 probepoint = (((uint64_t)net->cwnd) << 32); 803 probepoint |= ((0xa << 16) | 0); 804 vtag = (net->rtt << 32) | 805 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 806 (stcb->rport); 807 808 SDT_PROBE5(sctp, cwnd, net, rttvar, 809 vtag, 810 nbw, 811 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 812 net->flight_size, 813 probepoint); 814 net->cc_mod.rtcc.lbw = nbw; 815 net->cc_mod.rtcc.lbw_rtt = net->rtt; 816 if (net->cc_mod.rtcc.rtt_set_this_sack) { 817 net->cc_mod.rtcc.rtt_set_this_sack = 0; 818 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 819 } 820 } 821 } 822 /* 823 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 824 * moved. 825 */ 826 if (accum_moved || 827 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 828 /* If the cumulative ack moved we can proceed */ 829 if (net->cwnd <= net->ssthresh) { 830 /* We are in slow start */ 831 if (net->flight_size + net->net_ack >= net->cwnd) { 832 uint32_t limit; 833 834 old_cwnd = net->cwnd; 835 switch (asoc->sctp_cmt_on_off) { 836 case SCTP_CMT_RPV1: 837 limit = (uint32_t)(((uint64_t)net->mtu * 838 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 839 (uint64_t)net->ssthresh) / 840 (uint64_t)t_ssthresh); 841 incr = (uint32_t)(((uint64_t)net->net_ack * 842 (uint64_t)net->ssthresh) / 843 (uint64_t)t_ssthresh); 844 if (incr > limit) { 845 incr = limit; 846 } 847 if (incr == 0) { 848 incr = 1; 849 } 850 break; 851 case SCTP_CMT_RPV2: 852 /* 853 * lastsa>>3; we don't need 854 * to divide ... 855 */ 856 srtt = net->lastsa; 857 if (srtt == 0) { 858 srtt = 1; 859 } 860 limit = (uint32_t)(((uint64_t)net->mtu * 861 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 862 (uint64_t)net->cwnd) / 863 ((uint64_t)srtt * t_ucwnd_sbw)); 864 /* INCREASE FACTOR */ 865 incr = (uint32_t)(((uint64_t)net->net_ack * 866 (uint64_t)net->cwnd) / 867 ((uint64_t)srtt * t_ucwnd_sbw)); 868 /* INCREASE FACTOR */ 869 if (incr > limit) { 870 incr = limit; 871 } 872 if (incr == 0) { 873 incr = 1; 874 } 875 break; 876 case SCTP_CMT_MPTCP: 877 limit = (uint32_t)(((uint64_t)net->mtu * 878 mptcp_like_alpha * 879 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) >> 880 SHIFT_MPTCP_MULTI); 881 incr = (uint32_t)(((uint64_t)net->net_ack * 882 mptcp_like_alpha) >> 883 SHIFT_MPTCP_MULTI); 884 if (incr > limit) { 885 incr = limit; 886 } 887 if (incr > net->net_ack) { 888 incr = net->net_ack; 889 } 890 if (incr > net->mtu) { 891 incr = net->mtu; 892 } 893 break; 894 default: 895 incr = net->net_ack; 896 if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) { 897 incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable); 898 } 899 break; 900 } 901 net->cwnd += incr; 902 sctp_enforce_cwnd_limit(asoc, net); 903 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 904 sctp_log_cwnd(stcb, net, incr, 905 SCTP_CWND_LOG_FROM_SS); 906 } 907 SDT_PROBE5(sctp, cwnd, net, ack, 908 stcb->asoc.my_vtag, 909 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 910 net, 911 old_cwnd, net->cwnd); 912 } else { 913 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 914 sctp_log_cwnd(stcb, net, net->net_ack, 915 SCTP_CWND_LOG_NOADV_SS); 916 } 917 } 918 } else { 919 /* We are in congestion avoidance */ 920 /* 921 * Add to pba 922 */ 923 net->partial_bytes_acked += net->net_ack; 924 925 if ((net->flight_size + net->net_ack >= net->cwnd) && 926 (net->partial_bytes_acked >= net->cwnd)) { 927 net->partial_bytes_acked -= net->cwnd; 928 old_cwnd = net->cwnd; 929 switch (asoc->sctp_cmt_on_off) { 930 case SCTP_CMT_RPV1: 931 incr = (uint32_t)(((uint64_t)net->mtu * 932 (uint64_t)net->ssthresh) / 933 (uint64_t)t_ssthresh); 934 if (incr == 0) { 935 incr = 1; 936 } 937 break; 938 case SCTP_CMT_RPV2: 939 /* 940 * lastsa>>3; we don't need 941 * to divide ... 942 */ 943 srtt = net->lastsa; 944 if (srtt == 0) { 945 srtt = 1; 946 } 947 incr = (uint32_t)((uint64_t)net->mtu * 948 (uint64_t)net->cwnd / 949 ((uint64_t)srtt * 950 t_ucwnd_sbw)); 951 /* INCREASE FACTOR */ 952 if (incr == 0) { 953 incr = 1; 954 } 955 break; 956 case SCTP_CMT_MPTCP: 957 incr = (uint32_t)((mptcp_like_alpha * 958 (uint64_t)net->cwnd) >> 959 SHIFT_MPTCP_MULTI); 960 if (incr > net->mtu) { 961 incr = net->mtu; 962 } 963 break; 964 default: 965 incr = net->mtu; 966 break; 967 } 968 net->cwnd += incr; 969 sctp_enforce_cwnd_limit(asoc, net); 970 SDT_PROBE5(sctp, cwnd, net, ack, 971 stcb->asoc.my_vtag, 972 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 973 net, 974 old_cwnd, net->cwnd); 975 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 976 sctp_log_cwnd(stcb, net, net->mtu, 977 SCTP_CWND_LOG_FROM_CA); 978 } 979 } else { 980 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 981 sctp_log_cwnd(stcb, net, net->net_ack, 982 SCTP_CWND_LOG_NOADV_CA); 983 } 984 } 985 } 986 } else { 987 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 988 sctp_log_cwnd(stcb, net, net->mtu, 989 SCTP_CWND_LOG_NO_CUMACK); 990 } 991 } 992 } 993 } 994 995 static void 996 sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb, struct sctp_nets *net) 997 { 998 int old_cwnd; 999 1000 old_cwnd = net->cwnd; 1001 net->cwnd = net->mtu; 1002 SDT_PROBE5(sctp, cwnd, net, ack, 1003 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 1004 old_cwnd, net->cwnd); 1005 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1006 (void *)net, net->cwnd); 1007 } 1008 1009 static void 1010 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 1011 { 1012 int old_cwnd = net->cwnd; 1013 uint32_t t_ssthresh, t_cwnd; 1014 uint64_t t_ucwnd_sbw; 1015 1016 /* MT FIXME: Don't compute this over and over again */ 1017 t_ssthresh = 0; 1018 t_cwnd = 0; 1019 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 1020 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) { 1021 struct sctp_nets *lnet; 1022 uint32_t srtt; 1023 1024 t_ucwnd_sbw = 0; 1025 TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) { 1026 t_ssthresh += lnet->ssthresh; 1027 t_cwnd += lnet->cwnd; 1028 srtt = lnet->lastsa; 1029 /* lastsa>>3; we don't need to divide ... */ 1030 if (srtt > 0) { 1031 t_ucwnd_sbw += (uint64_t)lnet->cwnd / (uint64_t)srtt; 1032 } 1033 } 1034 if (t_ssthresh < 1) { 1035 t_ssthresh = 1; 1036 } 1037 if (t_ucwnd_sbw < 1) { 1038 t_ucwnd_sbw = 1; 1039 } 1040 if (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) { 1041 net->ssthresh = (uint32_t)(((uint64_t)4 * 1042 (uint64_t)net->mtu * 1043 (uint64_t)net->ssthresh) / 1044 (uint64_t)t_ssthresh); 1045 } else { 1046 uint64_t cc_delta; 1047 1048 srtt = net->lastsa; 1049 /* lastsa>>3; we don't need to divide ... */ 1050 if (srtt == 0) { 1051 srtt = 1; 1052 } 1053 cc_delta = t_ucwnd_sbw * (uint64_t)srtt / 2; 1054 if (cc_delta < t_cwnd) { 1055 net->ssthresh = (uint32_t)((uint64_t)t_cwnd - cc_delta); 1056 } else { 1057 net->ssthresh = net->mtu; 1058 } 1059 } 1060 if ((net->cwnd > t_cwnd / 2) && 1061 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 1062 net->ssthresh = net->cwnd - t_cwnd / 2; 1063 } 1064 if (net->ssthresh < net->mtu) { 1065 net->ssthresh = net->mtu; 1066 } 1067 } else { 1068 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 1069 } 1070 net->cwnd = net->mtu; 1071 net->partial_bytes_acked = 0; 1072 SDT_PROBE5(sctp, cwnd, net, to, 1073 stcb->asoc.my_vtag, 1074 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1075 net, 1076 old_cwnd, net->cwnd); 1077 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1078 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1079 } 1080 } 1081 1082 static void 1083 sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net, 1084 int in_window, int num_pkt_lost, int use_rtcc) 1085 { 1086 int old_cwnd = net->cwnd; 1087 1088 if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) { 1089 /* Data center Congestion Control */ 1090 if (in_window == 0) { 1091 /* 1092 * Go to CA with the cwnd at the point we sent the 1093 * TSN that was marked with a CE. 1094 */ 1095 if (net->ecn_prev_cwnd < net->cwnd) { 1096 /* Restore to prev cwnd */ 1097 net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost); 1098 } else { 1099 /* Just cut in 1/2 */ 1100 net->cwnd /= 2; 1101 } 1102 /* Drop to CA */ 1103 net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu); 1104 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1105 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1106 } 1107 } else { 1108 /* 1109 * Further tuning down required over the drastic 1110 * original cut 1111 */ 1112 net->ssthresh -= (net->mtu * num_pkt_lost); 1113 net->cwnd -= (net->mtu * num_pkt_lost); 1114 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1115 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1116 } 1117 } 1118 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1119 } else { 1120 if (in_window == 0) { 1121 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1122 net->ssthresh = net->cwnd / 2; 1123 if (net->ssthresh < net->mtu) { 1124 net->ssthresh = net->mtu; 1125 /* 1126 * here back off the timer as well, to slow 1127 * us down 1128 */ 1129 net->RTO <<= 1; 1130 } 1131 net->cwnd = net->ssthresh; 1132 SDT_PROBE5(sctp, cwnd, net, ecn, 1133 stcb->asoc.my_vtag, 1134 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1135 net, 1136 old_cwnd, net->cwnd); 1137 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1138 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1139 } 1140 } 1141 } 1142 1143 } 1144 1145 static void 1146 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 1147 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 1148 uint32_t *bottle_bw, uint32_t *on_queue) 1149 { 1150 uint32_t bw_avail; 1151 unsigned int incr; 1152 int old_cwnd = net->cwnd; 1153 1154 /* get bottle neck bw */ 1155 *bottle_bw = ntohl(cp->bottle_bw); 1156 /* and whats on queue */ 1157 *on_queue = ntohl(cp->current_onq); 1158 /* 1159 * adjust the on-queue if our flight is more it could be that the 1160 * router has not yet gotten data "in-flight" to it 1161 */ 1162 if (*on_queue < net->flight_size) { 1163 *on_queue = net->flight_size; 1164 } 1165 /* rtt is measured in micro seconds, bottle_bw in bytes per second */ 1166 bw_avail = (uint32_t)(((uint64_t)(*bottle_bw) * net->rtt) / (uint64_t)1000000); 1167 if (bw_avail > *bottle_bw) { 1168 /* 1169 * Cap the growth to no more than the bottle neck. This can 1170 * happen as RTT slides up due to queues. It also means if 1171 * you have more than a 1 second RTT with a empty queue you 1172 * will be limited to the bottle_bw per second no matter if 1173 * other points have 1/2 the RTT and you could get more 1174 * out... 1175 */ 1176 bw_avail = *bottle_bw; 1177 } 1178 if (*on_queue > bw_avail) { 1179 /* 1180 * No room for anything else don't allow anything else to be 1181 * "added to the fire". 1182 */ 1183 int seg_inflight, seg_onqueue, my_portion; 1184 1185 net->partial_bytes_acked = 0; 1186 /* how much are we over queue size? */ 1187 incr = *on_queue - bw_avail; 1188 if (stcb->asoc.seen_a_sack_this_pkt) { 1189 /* 1190 * undo any cwnd adjustment that the sack might have 1191 * made 1192 */ 1193 net->cwnd = net->prev_cwnd; 1194 } 1195 /* Now how much of that is mine? */ 1196 seg_inflight = net->flight_size / net->mtu; 1197 seg_onqueue = *on_queue / net->mtu; 1198 my_portion = (incr * seg_inflight) / seg_onqueue; 1199 1200 /* Have I made an adjustment already */ 1201 if (net->cwnd > net->flight_size) { 1202 /* 1203 * for this flight I made an adjustment we need to 1204 * decrease the portion by a share our previous 1205 * adjustment. 1206 */ 1207 int diff_adj; 1208 1209 diff_adj = net->cwnd - net->flight_size; 1210 if (diff_adj > my_portion) 1211 my_portion = 0; 1212 else 1213 my_portion -= diff_adj; 1214 } 1215 /* 1216 * back down to the previous cwnd (assume we have had a sack 1217 * before this packet). minus what ever portion of the 1218 * overage is my fault. 1219 */ 1220 net->cwnd -= my_portion; 1221 1222 /* we will NOT back down more than 1 MTU */ 1223 if (net->cwnd <= net->mtu) { 1224 net->cwnd = net->mtu; 1225 } 1226 /* force into CA */ 1227 net->ssthresh = net->cwnd - 1; 1228 } else { 1229 /* 1230 * Take 1/4 of the space left or max burst up .. whichever 1231 * is less. 1232 */ 1233 incr = (bw_avail - *on_queue) >> 2; 1234 if ((stcb->asoc.max_burst > 0) && 1235 (stcb->asoc.max_burst * net->mtu < incr)) { 1236 incr = stcb->asoc.max_burst * net->mtu; 1237 } 1238 net->cwnd += incr; 1239 } 1240 if (net->cwnd > bw_avail) { 1241 /* We can't exceed the pipe size */ 1242 net->cwnd = bw_avail; 1243 } 1244 if (net->cwnd < net->mtu) { 1245 /* We always have 1 MTU */ 1246 net->cwnd = net->mtu; 1247 } 1248 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1249 if (net->cwnd - old_cwnd != 0) { 1250 /* log only changes */ 1251 SDT_PROBE5(sctp, cwnd, net, pd, 1252 stcb->asoc.my_vtag, 1253 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1254 net, 1255 old_cwnd, net->cwnd); 1256 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1257 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1258 SCTP_CWND_LOG_FROM_SAT); 1259 } 1260 } 1261 } 1262 1263 static void 1264 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 1265 struct sctp_nets *net, int burst_limit) 1266 { 1267 int old_cwnd = net->cwnd; 1268 1269 if (net->ssthresh < net->cwnd) 1270 net->ssthresh = net->cwnd; 1271 if (burst_limit) { 1272 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 1273 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1274 SDT_PROBE5(sctp, cwnd, net, bl, 1275 stcb->asoc.my_vtag, 1276 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1277 net, 1278 old_cwnd, net->cwnd); 1279 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1280 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 1281 } 1282 } 1283 } 1284 1285 static void 1286 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1287 struct sctp_association *asoc, 1288 int accum_moved, int reneged_all, int will_exit) 1289 { 1290 /* Passing a zero argument in last disables the rtcc algorithm */ 1291 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0); 1292 } 1293 1294 static void 1295 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1296 int in_window, int num_pkt_lost) 1297 { 1298 /* Passing a zero argument in last disables the rtcc algorithm */ 1299 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0); 1300 } 1301 1302 /* Here starts the RTCCVAR type CC invented by RRS which 1303 * is a slight mod to RFC2581. We reuse a common routine or 1304 * two since these algorithms are so close and need to 1305 * remain the same. 1306 */ 1307 static void 1308 sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1309 int in_window, int num_pkt_lost) 1310 { 1311 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1); 1312 } 1313 1314 static void 1315 sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net, 1316 struct sctp_tmit_chunk *tp1) 1317 { 1318 net->cc_mod.rtcc.bw_bytes += tp1->send_size; 1319 } 1320 1321 static void 1322 sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb SCTP_UNUSED, 1323 struct sctp_nets *net) 1324 { 1325 if (net->cc_mod.rtcc.tls_needs_set > 0) { 1326 /* We had a bw measurment going on */ 1327 struct timeval ltls; 1328 1329 SCTP_GETPTIME_TIMEVAL(<ls); 1330 timevalsub(<ls, &net->cc_mod.rtcc.tls); 1331 net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec; 1332 } 1333 } 1334 1335 static void 1336 sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb, 1337 struct sctp_nets *net) 1338 { 1339 uint64_t vtag, probepoint; 1340 1341 if (net->cc_mod.rtcc.lbw) { 1342 /* Clear the old bw.. we went to 0 in-flight */ 1343 vtag = (net->rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1344 (stcb->rport); 1345 probepoint = (((uint64_t)net->cwnd) << 32); 1346 /* Probe point 8 */ 1347 probepoint |= ((8 << 16) | 0); 1348 SDT_PROBE5(sctp, cwnd, net, rttvar, 1349 vtag, 1350 ((net->cc_mod.rtcc.lbw << 32) | 0), 1351 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 1352 net->flight_size, 1353 probepoint); 1354 net->cc_mod.rtcc.lbw_rtt = 0; 1355 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1356 net->cc_mod.rtcc.lbw = 0; 1357 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1358 net->cc_mod.rtcc.vol_reduce = 0; 1359 net->cc_mod.rtcc.bw_tot_time = 0; 1360 net->cc_mod.rtcc.bw_bytes = 0; 1361 net->cc_mod.rtcc.tls_needs_set = 0; 1362 if (net->cc_mod.rtcc.steady_step) { 1363 net->cc_mod.rtcc.vol_reduce = 0; 1364 net->cc_mod.rtcc.step_cnt = 0; 1365 net->cc_mod.rtcc.last_step_state = 0; 1366 } 1367 if (net->cc_mod.rtcc.ret_from_eq) { 1368 /* less aggressive one - reset cwnd too */ 1369 uint32_t cwnd_in_mtu, cwnd; 1370 1371 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 1372 if (cwnd_in_mtu == 0) { 1373 /* 1374 * Using 0 means that the value of RFC 4960 1375 * is used. 1376 */ 1377 cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1378 } else { 1379 /* 1380 * We take the minimum of the burst limit 1381 * and the initial congestion window. 1382 */ 1383 if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst)) 1384 cwnd_in_mtu = stcb->asoc.max_burst; 1385 cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 1386 } 1387 if (net->cwnd > cwnd) { 1388 /* 1389 * Only set if we are not a timeout (i.e. 1390 * down to 1 mtu) 1391 */ 1392 net->cwnd = cwnd; 1393 } 1394 } 1395 } 1396 } 1397 1398 static void 1399 sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb, 1400 struct sctp_nets *net) 1401 { 1402 uint64_t vtag, probepoint; 1403 1404 sctp_set_initial_cc_param(stcb, net); 1405 stcb->asoc.use_precise_time = 1; 1406 probepoint = (((uint64_t)net->cwnd) << 32); 1407 probepoint |= ((9 << 16) | 0); 1408 vtag = (net->rtt << 32) | 1409 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1410 (stcb->rport); 1411 SDT_PROBE5(sctp, cwnd, net, rttvar, 1412 vtag, 1413 0, 1414 0, 1415 0, 1416 probepoint); 1417 net->cc_mod.rtcc.lbw_rtt = 0; 1418 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1419 net->cc_mod.rtcc.vol_reduce = 0; 1420 net->cc_mod.rtcc.lbw = 0; 1421 net->cc_mod.rtcc.vol_reduce = 0; 1422 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1423 net->cc_mod.rtcc.bw_tot_time = 0; 1424 net->cc_mod.rtcc.bw_bytes = 0; 1425 net->cc_mod.rtcc.tls_needs_set = 0; 1426 net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret); 1427 net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step); 1428 net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn); 1429 net->cc_mod.rtcc.step_cnt = 0; 1430 net->cc_mod.rtcc.last_step_state = 0; 1431 } 1432 1433 static int 1434 sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget, 1435 struct sctp_cc_option *cc_opt) 1436 { 1437 struct sctp_nets *net; 1438 1439 if (setorget == 1) { 1440 /* a set */ 1441 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1442 if ((cc_opt->aid_value.assoc_value != 0) && 1443 (cc_opt->aid_value.assoc_value != 1)) { 1444 return (EINVAL); 1445 } 1446 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1447 net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value; 1448 } 1449 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1450 if ((cc_opt->aid_value.assoc_value != 0) && 1451 (cc_opt->aid_value.assoc_value != 1)) { 1452 return (EINVAL); 1453 } 1454 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1455 net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value; 1456 } 1457 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1458 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1459 net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value; 1460 } 1461 } else { 1462 return (EINVAL); 1463 } 1464 } else { 1465 /* a get */ 1466 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1467 net = TAILQ_FIRST(&stcb->asoc.nets); 1468 if (net == NULL) { 1469 return (EFAULT); 1470 } 1471 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.ret_from_eq; 1472 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1473 net = TAILQ_FIRST(&stcb->asoc.nets); 1474 if (net == NULL) { 1475 return (EFAULT); 1476 } 1477 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.use_dccc_ecn; 1478 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1479 net = TAILQ_FIRST(&stcb->asoc.nets); 1480 if (net == NULL) { 1481 return (EFAULT); 1482 } 1483 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step; 1484 } else { 1485 return (EINVAL); 1486 } 1487 } 1488 return (0); 1489 } 1490 1491 static void 1492 sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb SCTP_UNUSED, 1493 struct sctp_nets *net) 1494 { 1495 if (net->cc_mod.rtcc.tls_needs_set == 0) { 1496 SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls); 1497 net->cc_mod.rtcc.tls_needs_set = 2; 1498 } 1499 } 1500 1501 static void 1502 sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb, 1503 struct sctp_association *asoc, 1504 int accum_moved, int reneged_all, int will_exit) 1505 { 1506 /* Passing a one argument at the last enables the rtcc algorithm */ 1507 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1); 1508 } 1509 1510 static void 1511 sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb SCTP_UNUSED, 1512 struct sctp_nets *net, 1513 struct timeval *now SCTP_UNUSED) 1514 { 1515 net->cc_mod.rtcc.rtt_set_this_sack = 1; 1516 } 1517 1518 /* Here starts Sally Floyds HS-TCP */ 1519 1520 struct sctp_hs_raise_drop { 1521 int32_t cwnd; 1522 int8_t increase; 1523 int8_t drop_percent; 1524 }; 1525 1526 #define SCTP_HS_TABLE_SIZE 73 1527 1528 static const struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 1529 {38, 1, 50}, /* 0 */ 1530 {118, 2, 44}, /* 1 */ 1531 {221, 3, 41}, /* 2 */ 1532 {347, 4, 38}, /* 3 */ 1533 {495, 5, 37}, /* 4 */ 1534 {663, 6, 35}, /* 5 */ 1535 {851, 7, 34}, /* 6 */ 1536 {1058, 8, 33}, /* 7 */ 1537 {1284, 9, 32}, /* 8 */ 1538 {1529, 10, 31}, /* 9 */ 1539 {1793, 11, 30}, /* 10 */ 1540 {2076, 12, 29}, /* 11 */ 1541 {2378, 13, 28}, /* 12 */ 1542 {2699, 14, 28}, /* 13 */ 1543 {3039, 15, 27}, /* 14 */ 1544 {3399, 16, 27}, /* 15 */ 1545 {3778, 17, 26}, /* 16 */ 1546 {4177, 18, 26}, /* 17 */ 1547 {4596, 19, 25}, /* 18 */ 1548 {5036, 20, 25}, /* 19 */ 1549 {5497, 21, 24}, /* 20 */ 1550 {5979, 22, 24}, /* 21 */ 1551 {6483, 23, 23}, /* 22 */ 1552 {7009, 24, 23}, /* 23 */ 1553 {7558, 25, 22}, /* 24 */ 1554 {8130, 26, 22}, /* 25 */ 1555 {8726, 27, 22}, /* 26 */ 1556 {9346, 28, 21}, /* 27 */ 1557 {9991, 29, 21}, /* 28 */ 1558 {10661, 30, 21}, /* 29 */ 1559 {11358, 31, 20}, /* 30 */ 1560 {12082, 32, 20}, /* 31 */ 1561 {12834, 33, 20}, /* 32 */ 1562 {13614, 34, 19}, /* 33 */ 1563 {14424, 35, 19}, /* 34 */ 1564 {15265, 36, 19}, /* 35 */ 1565 {16137, 37, 19}, /* 36 */ 1566 {17042, 38, 18}, /* 37 */ 1567 {17981, 39, 18}, /* 38 */ 1568 {18955, 40, 18}, /* 39 */ 1569 {19965, 41, 17}, /* 40 */ 1570 {21013, 42, 17}, /* 41 */ 1571 {22101, 43, 17}, /* 42 */ 1572 {23230, 44, 17}, /* 43 */ 1573 {24402, 45, 16}, /* 44 */ 1574 {25618, 46, 16}, /* 45 */ 1575 {26881, 47, 16}, /* 46 */ 1576 {28193, 48, 16}, /* 47 */ 1577 {29557, 49, 15}, /* 48 */ 1578 {30975, 50, 15}, /* 49 */ 1579 {32450, 51, 15}, /* 50 */ 1580 {33986, 52, 15}, /* 51 */ 1581 {35586, 53, 14}, /* 52 */ 1582 {37253, 54, 14}, /* 53 */ 1583 {38992, 55, 14}, /* 54 */ 1584 {40808, 56, 14}, /* 55 */ 1585 {42707, 57, 13}, /* 56 */ 1586 {44694, 58, 13}, /* 57 */ 1587 {46776, 59, 13}, /* 58 */ 1588 {48961, 60, 13}, /* 59 */ 1589 {51258, 61, 13}, /* 60 */ 1590 {53677, 62, 12}, /* 61 */ 1591 {56230, 63, 12}, /* 62 */ 1592 {58932, 64, 12}, /* 63 */ 1593 {61799, 65, 12}, /* 64 */ 1594 {64851, 66, 11}, /* 65 */ 1595 {68113, 67, 11}, /* 66 */ 1596 {71617, 68, 11}, /* 67 */ 1597 {75401, 69, 10}, /* 68 */ 1598 {79517, 70, 10}, /* 69 */ 1599 {84035, 71, 10}, /* 70 */ 1600 {89053, 72, 10}, /* 71 */ 1601 {94717, 73, 9} /* 72 */ 1602 }; 1603 1604 static void 1605 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 1606 { 1607 int cur_val, i, indx, incr; 1608 int old_cwnd = net->cwnd; 1609 1610 cur_val = net->cwnd >> 10; 1611 indx = SCTP_HS_TABLE_SIZE - 1; 1612 1613 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1614 /* normal mode */ 1615 if (net->net_ack > net->mtu) { 1616 net->cwnd += net->mtu; 1617 } else { 1618 net->cwnd += net->net_ack; 1619 } 1620 } else { 1621 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 1622 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 1623 indx = i; 1624 break; 1625 } 1626 } 1627 net->last_hs_used = indx; 1628 incr = (((int32_t)sctp_cwnd_adjust[indx].increase) << 10); 1629 net->cwnd += incr; 1630 } 1631 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1632 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1633 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SS); 1634 } 1635 } 1636 1637 static void 1638 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 1639 { 1640 int cur_val, i, indx; 1641 int old_cwnd = net->cwnd; 1642 1643 cur_val = net->cwnd >> 10; 1644 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1645 /* normal mode */ 1646 net->ssthresh = net->cwnd / 2; 1647 if (net->ssthresh < (net->mtu * 2)) { 1648 net->ssthresh = 2 * net->mtu; 1649 } 1650 net->cwnd = net->ssthresh; 1651 } else { 1652 /* drop by the proper amount */ 1653 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 1654 (int32_t)sctp_cwnd_adjust[net->last_hs_used].drop_percent); 1655 net->cwnd = net->ssthresh; 1656 /* now where are we */ 1657 indx = net->last_hs_used; 1658 cur_val = net->cwnd >> 10; 1659 /* reset where we are in the table */ 1660 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1661 /* feel out of hs */ 1662 net->last_hs_used = 0; 1663 } else { 1664 for (i = indx; i >= 1; i--) { 1665 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 1666 break; 1667 } 1668 } 1669 net->last_hs_used = indx; 1670 } 1671 } 1672 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1673 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1674 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 1675 } 1676 } 1677 1678 static void 1679 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 1680 struct sctp_association *asoc) 1681 { 1682 struct sctp_nets *net; 1683 1684 /* 1685 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 1686 * (net->fast_retran_loss_recovery == 0))) 1687 */ 1688 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1689 if ((asoc->fast_retran_loss_recovery == 0) || 1690 (asoc->sctp_cmt_on_off > 0)) { 1691 /* out of a RFC2582 Fast recovery window? */ 1692 if (net->net_ack > 0) { 1693 /* 1694 * per section 7.2.3, are there any 1695 * destinations that had a fast retransmit 1696 * to them. If so what we need to do is 1697 * adjust ssthresh and cwnd. 1698 */ 1699 struct sctp_tmit_chunk *lchk; 1700 1701 sctp_hs_cwnd_decrease(stcb, net); 1702 1703 lchk = TAILQ_FIRST(&asoc->send_queue); 1704 1705 net->partial_bytes_acked = 0; 1706 /* Turn on fast recovery window */ 1707 asoc->fast_retran_loss_recovery = 1; 1708 if (lchk == NULL) { 1709 /* Mark end of the window */ 1710 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1711 } else { 1712 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1713 } 1714 1715 /* 1716 * CMT fast recovery -- per destination 1717 * recovery variable. 1718 */ 1719 net->fast_retran_loss_recovery = 1; 1720 1721 if (lchk == NULL) { 1722 /* Mark end of the window */ 1723 net->fast_recovery_tsn = asoc->sending_seq - 1; 1724 } else { 1725 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1726 } 1727 1728 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1729 stcb->sctp_ep, stcb, net, 1730 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_2); 1731 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1732 stcb->sctp_ep, stcb, net); 1733 } 1734 } else if (net->net_ack > 0) { 1735 /* 1736 * Mark a peg that we WOULD have done a cwnd 1737 * reduction but RFC2582 prevented this action. 1738 */ 1739 SCTP_STAT_INCR(sctps_fastretransinrtt); 1740 } 1741 } 1742 } 1743 1744 static void 1745 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 1746 struct sctp_association *asoc, 1747 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 1748 { 1749 struct sctp_nets *net; 1750 1751 /******************************/ 1752 /* update cwnd and Early FR */ 1753 /******************************/ 1754 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1755 #ifdef JANA_CMT_FAST_RECOVERY 1756 /* 1757 * CMT fast recovery code. Need to debug. 1758 */ 1759 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1760 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 1761 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 1762 net->will_exit_fast_recovery = 1; 1763 } 1764 } 1765 #endif 1766 /* if nothing was acked on this destination skip it */ 1767 if (net->net_ack == 0) { 1768 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1769 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1770 } 1771 continue; 1772 } 1773 #ifdef JANA_CMT_FAST_RECOVERY 1774 /* 1775 * CMT fast recovery code 1776 */ 1777 /* 1778 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 1779 * && net->will_exit_fast_recovery == 0) { @@@ Do something 1780 * } else if (sctp_cmt_on_off == 0 && 1781 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1782 */ 1783 #endif 1784 1785 if (asoc->fast_retran_loss_recovery && 1786 (will_exit == 0) && 1787 (asoc->sctp_cmt_on_off == 0)) { 1788 /* 1789 * If we are in loss recovery we skip any cwnd 1790 * update 1791 */ 1792 return; 1793 } 1794 /* 1795 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1796 * moved. 1797 */ 1798 if (accum_moved || 1799 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 1800 /* If the cumulative ack moved we can proceed */ 1801 if (net->cwnd <= net->ssthresh) { 1802 /* We are in slow start */ 1803 if (net->flight_size + net->net_ack >= net->cwnd) { 1804 sctp_hs_cwnd_increase(stcb, net); 1805 } else { 1806 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1807 sctp_log_cwnd(stcb, net, net->net_ack, 1808 SCTP_CWND_LOG_NOADV_SS); 1809 } 1810 } 1811 } else { 1812 /* We are in congestion avoidance */ 1813 net->partial_bytes_acked += net->net_ack; 1814 if ((net->flight_size + net->net_ack >= net->cwnd) && 1815 (net->partial_bytes_acked >= net->cwnd)) { 1816 net->partial_bytes_acked -= net->cwnd; 1817 net->cwnd += net->mtu; 1818 sctp_enforce_cwnd_limit(asoc, net); 1819 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1820 sctp_log_cwnd(stcb, net, net->mtu, 1821 SCTP_CWND_LOG_FROM_CA); 1822 } 1823 } else { 1824 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1825 sctp_log_cwnd(stcb, net, net->net_ack, 1826 SCTP_CWND_LOG_NOADV_CA); 1827 } 1828 } 1829 } 1830 } else { 1831 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1832 sctp_log_cwnd(stcb, net, net->mtu, 1833 SCTP_CWND_LOG_NO_CUMACK); 1834 } 1835 } 1836 } 1837 } 1838 1839 /* 1840 * H-TCP congestion control. The algorithm is detailed in: 1841 * R.N.Shorten, D.J.Leith: 1842 * "H-TCP: TCP for high-speed and long-distance networks" 1843 * Proc. PFLDnet, Argonne, 2004. 1844 * http://www.hamilton.ie/net/htcp3.pdf 1845 */ 1846 1847 static int use_rtt_scaling = 1; 1848 static int use_bandwidth_switch = 1; 1849 1850 static inline int 1851 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1852 { 1853 return (seq3 - seq2 >= seq1 - seq2); 1854 } 1855 1856 static inline uint32_t 1857 htcp_cong_time(struct htcp *ca) 1858 { 1859 return (sctp_get_tick_count() - ca->last_cong); 1860 } 1861 1862 static inline uint32_t 1863 htcp_ccount(struct htcp *ca) 1864 { 1865 return (ca->minRTT == 0 ? htcp_cong_time(ca) : htcp_cong_time(ca) / ca->minRTT); 1866 } 1867 1868 static inline void 1869 htcp_reset(struct htcp *ca) 1870 { 1871 ca->undo_last_cong = ca->last_cong; 1872 ca->undo_maxRTT = ca->maxRTT; 1873 ca->undo_old_maxB = ca->old_maxB; 1874 ca->last_cong = sctp_get_tick_count(); 1875 } 1876 1877 #ifdef SCTP_NOT_USED 1878 1879 static uint32_t 1880 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1881 { 1882 net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong; 1883 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT; 1884 net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB; 1885 return (max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->cc_mod.htcp_ca.beta) * net->mtu)); 1886 } 1887 1888 #endif 1889 1890 static inline void 1891 measure_rtt(struct sctp_nets *net) 1892 { 1893 uint32_t srtt = net->lastsa >> SCTP_RTT_SHIFT; 1894 1895 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1896 if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT) 1897 net->cc_mod.htcp_ca.minRTT = srtt; 1898 1899 /* max RTT */ 1900 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) { 1901 if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT) 1902 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT; 1903 if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT + sctp_msecs_to_ticks(20)) 1904 net->cc_mod.htcp_ca.maxRTT = srtt; 1905 } 1906 } 1907 1908 static void 1909 measure_achieved_throughput(struct sctp_nets *net) 1910 { 1911 uint32_t now = sctp_get_tick_count(); 1912 1913 if (net->fast_retran_ip == 0) 1914 net->cc_mod.htcp_ca.bytes_acked = net->net_ack; 1915 1916 if (!use_bandwidth_switch) 1917 return; 1918 1919 /* achieved throughput calculations */ 1920 /* JRS - not 100% sure of this statement */ 1921 if (net->fast_retran_ip == 1) { 1922 net->cc_mod.htcp_ca.bytecount = 0; 1923 net->cc_mod.htcp_ca.lasttime = now; 1924 return; 1925 } 1926 1927 net->cc_mod.htcp_ca.bytecount += net->net_ack; 1928 if ((net->cc_mod.htcp_ca.bytecount >= net->cwnd - (((net->cc_mod.htcp_ca.alpha >> 7) ? (net->cc_mod.htcp_ca.alpha >> 7) : 1) * net->mtu)) && 1929 (now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT) && 1930 (net->cc_mod.htcp_ca.minRTT > 0)) { 1931 uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount / net->mtu * hz / (now - net->cc_mod.htcp_ca.lasttime); 1932 1933 if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) { 1934 /* just after backoff */ 1935 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi; 1936 } else { 1937 net->cc_mod.htcp_ca.Bi = (3 * net->cc_mod.htcp_ca.Bi + cur_Bi) / 4; 1938 if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB) 1939 net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi; 1940 if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB) 1941 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB; 1942 } 1943 net->cc_mod.htcp_ca.bytecount = 0; 1944 net->cc_mod.htcp_ca.lasttime = now; 1945 } 1946 } 1947 1948 static inline void 1949 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1950 { 1951 if (use_bandwidth_switch) { 1952 uint32_t maxB = ca->maxB; 1953 uint32_t old_maxB = ca->old_maxB; 1954 1955 ca->old_maxB = ca->maxB; 1956 1957 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1958 ca->beta = BETA_MIN; 1959 ca->modeswitch = 0; 1960 return; 1961 } 1962 } 1963 1964 if (ca->modeswitch && minRTT > sctp_msecs_to_ticks(10) && maxRTT) { 1965 ca->beta = (minRTT << 7) / maxRTT; 1966 if (ca->beta < BETA_MIN) 1967 ca->beta = BETA_MIN; 1968 else if (ca->beta > BETA_MAX) 1969 ca->beta = BETA_MAX; 1970 } else { 1971 ca->beta = BETA_MIN; 1972 ca->modeswitch = 1; 1973 } 1974 } 1975 1976 static inline void 1977 htcp_alpha_update(struct htcp *ca) 1978 { 1979 uint32_t minRTT = ca->minRTT; 1980 uint32_t factor = 1; 1981 uint32_t diff = htcp_cong_time(ca); 1982 1983 if (diff > (uint32_t)hz) { 1984 diff -= hz; 1985 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1986 } 1987 1988 if (use_rtt_scaling && minRTT) { 1989 uint32_t scale = (hz << 3) / (10 * minRTT); 1990 1991 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 1992 * interval [0.5,10]<<3 */ 1993 factor = (factor << 3) / scale; 1994 if (factor != 0) 1995 factor = 1; 1996 } 1997 1998 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 1999 if (ca->alpha != 0) 2000 ca->alpha = ALPHA_BASE; 2001 } 2002 2003 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 2004 * rtt before we adjust our beta to ensure we are working from a consistent 2005 * data. 2006 * 2007 * This function should be called when we hit a congestion event since only at 2008 * that point do we really have a real sense of maxRTT (the queues en route 2009 * were getting just too full now). 2010 */ 2011 static void 2012 htcp_param_update(struct sctp_nets *net) 2013 { 2014 uint32_t minRTT = net->cc_mod.htcp_ca.minRTT; 2015 uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT; 2016 2017 htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT); 2018 htcp_alpha_update(&net->cc_mod.htcp_ca); 2019 2020 /* 2021 * add slowly fading memory for maxRTT to accommodate routing 2022 * changes etc 2023 */ 2024 if (minRTT > 0 && maxRTT > minRTT) 2025 net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 2026 } 2027 2028 static uint32_t 2029 htcp_recalc_ssthresh(struct sctp_nets *net) 2030 { 2031 htcp_param_update(net); 2032 return (max(((net->cwnd / net->mtu * net->cc_mod.htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu)); 2033 } 2034 2035 static void 2036 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 2037 { 2038 /*- 2039 * How to handle these functions? 2040 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 2041 * return; 2042 */ 2043 if (net->cwnd <= net->ssthresh) { 2044 /* We are in slow start */ 2045 if (net->flight_size + net->net_ack >= net->cwnd) { 2046 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 2047 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 2048 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2049 sctp_log_cwnd(stcb, net, net->mtu, 2050 SCTP_CWND_LOG_FROM_SS); 2051 } 2052 2053 } else { 2054 net->cwnd += net->net_ack; 2055 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2056 sctp_log_cwnd(stcb, net, net->net_ack, 2057 SCTP_CWND_LOG_FROM_SS); 2058 } 2059 } 2060 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2061 } else { 2062 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2063 sctp_log_cwnd(stcb, net, net->net_ack, 2064 SCTP_CWND_LOG_NOADV_SS); 2065 } 2066 } 2067 } else { 2068 measure_rtt(net); 2069 2070 /* 2071 * In dangerous area, increase slowly. In theory this is 2072 * net->cwnd += alpha / net->cwnd 2073 */ 2074 /* What is snd_cwnd_cnt?? */ 2075 if (((net->partial_bytes_acked / net->mtu * net->cc_mod.htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 2076 /*- 2077 * Does SCTP have a cwnd clamp? 2078 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 2079 */ 2080 net->cwnd += net->mtu; 2081 net->partial_bytes_acked = 0; 2082 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2083 htcp_alpha_update(&net->cc_mod.htcp_ca); 2084 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2085 sctp_log_cwnd(stcb, net, net->mtu, 2086 SCTP_CWND_LOG_FROM_CA); 2087 } 2088 } else { 2089 net->partial_bytes_acked += net->net_ack; 2090 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2091 sctp_log_cwnd(stcb, net, net->net_ack, 2092 SCTP_CWND_LOG_NOADV_CA); 2093 } 2094 } 2095 2096 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2097 } 2098 } 2099 2100 #ifdef SCTP_NOT_USED 2101 /* Lower bound on congestion window. */ 2102 static uint32_t 2103 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 2104 { 2105 return (net->ssthresh); 2106 } 2107 #endif 2108 2109 static void 2110 htcp_init(struct sctp_nets *net) 2111 { 2112 memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp)); 2113 net->cc_mod.htcp_ca.alpha = ALPHA_BASE; 2114 net->cc_mod.htcp_ca.beta = BETA_MIN; 2115 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2116 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count(); 2117 } 2118 2119 static void 2120 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 2121 { 2122 /* 2123 * We take the max of the burst limit times a MTU or the 2124 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 2125 */ 2126 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 2127 net->ssthresh = stcb->asoc.peers_rwnd; 2128 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2129 htcp_init(net); 2130 2131 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 2132 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 2133 } 2134 } 2135 2136 static void 2137 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 2138 struct sctp_association *asoc, 2139 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 2140 { 2141 struct sctp_nets *net; 2142 2143 /******************************/ 2144 /* update cwnd and Early FR */ 2145 /******************************/ 2146 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2147 #ifdef JANA_CMT_FAST_RECOVERY 2148 /* 2149 * CMT fast recovery code. Need to debug. 2150 */ 2151 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 2152 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 2153 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 2154 net->will_exit_fast_recovery = 1; 2155 } 2156 } 2157 #endif 2158 /* if nothing was acked on this destination skip it */ 2159 if (net->net_ack == 0) { 2160 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2161 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 2162 } 2163 continue; 2164 } 2165 #ifdef JANA_CMT_FAST_RECOVERY 2166 /* 2167 * CMT fast recovery code 2168 */ 2169 /* 2170 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 2171 * && net->will_exit_fast_recovery == 0) { @@@ Do something 2172 * } else if (sctp_cmt_on_off == 0 && 2173 * asoc->fast_retran_loss_recovery && will_exit == 0) { 2174 */ 2175 #endif 2176 2177 if (asoc->fast_retran_loss_recovery && 2178 will_exit == 0 && 2179 (asoc->sctp_cmt_on_off == 0)) { 2180 /* 2181 * If we are in loss recovery we skip any cwnd 2182 * update 2183 */ 2184 return; 2185 } 2186 /* 2187 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 2188 * moved. 2189 */ 2190 if (accum_moved || 2191 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 2192 htcp_cong_avoid(stcb, net); 2193 measure_achieved_throughput(net); 2194 } else { 2195 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2196 sctp_log_cwnd(stcb, net, net->mtu, 2197 SCTP_CWND_LOG_NO_CUMACK); 2198 } 2199 } 2200 } 2201 } 2202 2203 static void 2204 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 2205 struct sctp_association *asoc) 2206 { 2207 struct sctp_nets *net; 2208 2209 /* 2210 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 2211 * (net->fast_retran_loss_recovery == 0))) 2212 */ 2213 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2214 if ((asoc->fast_retran_loss_recovery == 0) || 2215 (asoc->sctp_cmt_on_off > 0)) { 2216 /* out of a RFC2582 Fast recovery window? */ 2217 if (net->net_ack > 0) { 2218 /* 2219 * per section 7.2.3, are there any 2220 * destinations that had a fast retransmit 2221 * to them. If so what we need to do is 2222 * adjust ssthresh and cwnd. 2223 */ 2224 struct sctp_tmit_chunk *lchk; 2225 int old_cwnd = net->cwnd; 2226 2227 /* JRS - reset as if state were changed */ 2228 htcp_reset(&net->cc_mod.htcp_ca); 2229 net->ssthresh = htcp_recalc_ssthresh(net); 2230 net->cwnd = net->ssthresh; 2231 sctp_enforce_cwnd_limit(asoc, net); 2232 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2233 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 2234 SCTP_CWND_LOG_FROM_FR); 2235 } 2236 lchk = TAILQ_FIRST(&asoc->send_queue); 2237 2238 net->partial_bytes_acked = 0; 2239 /* Turn on fast recovery window */ 2240 asoc->fast_retran_loss_recovery = 1; 2241 if (lchk == NULL) { 2242 /* Mark end of the window */ 2243 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 2244 } else { 2245 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2246 } 2247 2248 /* 2249 * CMT fast recovery -- per destination 2250 * recovery variable. 2251 */ 2252 net->fast_retran_loss_recovery = 1; 2253 2254 if (lchk == NULL) { 2255 /* Mark end of the window */ 2256 net->fast_recovery_tsn = asoc->sending_seq - 1; 2257 } else { 2258 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2259 } 2260 2261 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 2262 stcb->sctp_ep, stcb, net, 2263 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_3); 2264 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 2265 stcb->sctp_ep, stcb, net); 2266 } 2267 } else if (net->net_ack > 0) { 2268 /* 2269 * Mark a peg that we WOULD have done a cwnd 2270 * reduction but RFC2582 prevented this action. 2271 */ 2272 SCTP_STAT_INCR(sctps_fastretransinrtt); 2273 } 2274 } 2275 } 2276 2277 static void 2278 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 2279 struct sctp_nets *net) 2280 { 2281 int old_cwnd = net->cwnd; 2282 2283 /* JRS - reset as if the state were being changed to timeout */ 2284 htcp_reset(&net->cc_mod.htcp_ca); 2285 net->ssthresh = htcp_recalc_ssthresh(net); 2286 net->cwnd = net->mtu; 2287 net->partial_bytes_acked = 0; 2288 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2289 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 2290 } 2291 } 2292 2293 static void 2294 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 2295 struct sctp_nets *net, int in_window, int num_pkt_lost SCTP_UNUSED) 2296 { 2297 int old_cwnd; 2298 2299 old_cwnd = net->cwnd; 2300 2301 /* JRS - reset hctp as if state changed */ 2302 if (in_window == 0) { 2303 htcp_reset(&net->cc_mod.htcp_ca); 2304 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 2305 net->ssthresh = htcp_recalc_ssthresh(net); 2306 if (net->ssthresh < net->mtu) { 2307 net->ssthresh = net->mtu; 2308 /* here back off the timer as well, to slow us down */ 2309 net->RTO <<= 1; 2310 } 2311 net->cwnd = net->ssthresh; 2312 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2313 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2314 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 2315 } 2316 } 2317 } 2318 2319 const struct sctp_cc_functions sctp_cc_functions[] = { 2320 { 2321 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2322 .sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack, 2323 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2324 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2325 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2326 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2327 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2328 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2329 }, 2330 { 2331 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2332 .sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack, 2333 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2334 .sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr, 2335 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2336 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2337 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2338 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2339 }, 2340 { 2341 .sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param, 2342 .sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack, 2343 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2344 .sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr, 2345 .sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout, 2346 .sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo, 2347 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2348 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2349 }, 2350 { 2351 .sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param, 2352 .sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack, 2353 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2354 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2355 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2356 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo, 2357 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2358 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2359 .sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted, 2360 .sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged, 2361 .sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins, 2362 .sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack, 2363 .sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option, 2364 .sctp_rtt_calculated = sctp_rtt_rtcc_calculated 2365 } 2366 }; 2367