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, t_cwnd, 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_cwnd = 0; 698 t_ucwnd_sbw = 0; 699 t_path_mptcp = 0; 700 mptcp_like_alpha = 1; 701 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 702 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2) || 703 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_MPTCP)) { 704 max_path = 0; 705 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 706 t_ssthresh += net->ssthresh; 707 t_cwnd += net->cwnd; 708 /* lastsa>>3; we don't need to devide ... */ 709 srtt = net->lastsa; 710 if (srtt > 0) { 711 uint64_t tmp; 712 713 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)srtt; 714 t_path_mptcp += (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_Z) / 715 (((uint64_t)net->mtu) * (uint64_t)srtt); 716 tmp = (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_N) / 717 ((uint64_t)net->mtu * (uint64_t)(srtt * srtt)); 718 if (tmp > max_path) { 719 max_path = tmp; 720 } 721 } 722 } 723 if (t_path_mptcp > 0) { 724 mptcp_like_alpha = max_path / (t_path_mptcp * t_path_mptcp); 725 } else { 726 mptcp_like_alpha = 1; 727 } 728 } 729 if (t_ssthresh == 0) { 730 t_ssthresh = 1; 731 } 732 if (t_ucwnd_sbw == 0) { 733 t_ucwnd_sbw = 1; 734 } 735 /******************************/ 736 /* update cwnd and Early FR */ 737 /******************************/ 738 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 739 #ifdef JANA_CMT_FAST_RECOVERY 740 /* 741 * CMT fast recovery code. Need to debug. 742 */ 743 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 744 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 745 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 746 net->will_exit_fast_recovery = 1; 747 } 748 } 749 #endif 750 /* if nothing was acked on this destination skip it */ 751 if (net->net_ack == 0) { 752 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 753 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 754 } 755 continue; 756 } 757 #ifdef JANA_CMT_FAST_RECOVERY 758 /* 759 * CMT fast recovery code 760 */ 761 /* 762 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 763 * && net->will_exit_fast_recovery == 0) { @@@ Do something 764 * } else if (sctp_cmt_on_off == 0 && 765 * asoc->fast_retran_loss_recovery && will_exit == 0) { 766 */ 767 #endif 768 769 if (asoc->fast_retran_loss_recovery && 770 (will_exit == 0) && 771 (asoc->sctp_cmt_on_off == 0)) { 772 /* 773 * If we are in loss recovery we skip any cwnd 774 * update 775 */ 776 return; 777 } 778 /* 779 * Did any measurements go on for this network? 780 */ 781 if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) { 782 uint64_t nbw; 783 784 /* 785 * At this point our bw_bytes has been updated by 786 * incoming sack information. 787 * 788 * But our bw may not yet be set. 789 * 790 */ 791 if ((net->cc_mod.rtcc.new_tot_time / 1000) > 0) { 792 nbw = net->cc_mod.rtcc.bw_bytes / (net->cc_mod.rtcc.new_tot_time / 1000); 793 } else { 794 nbw = net->cc_mod.rtcc.bw_bytes; 795 } 796 if (net->cc_mod.rtcc.lbw) { 797 if (cc_bw_limit(stcb, net, nbw)) { 798 /* Hold here, no update */ 799 continue; 800 } 801 } else { 802 uint64_t vtag, probepoint; 803 804 probepoint = (((uint64_t)net->cwnd) << 32); 805 probepoint |= ((0xa << 16) | 0); 806 vtag = (net->rtt << 32) | 807 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 808 (stcb->rport); 809 810 SDT_PROBE5(sctp, cwnd, net, rttvar, 811 vtag, 812 nbw, 813 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 814 net->flight_size, 815 probepoint); 816 net->cc_mod.rtcc.lbw = nbw; 817 net->cc_mod.rtcc.lbw_rtt = net->rtt; 818 if (net->cc_mod.rtcc.rtt_set_this_sack) { 819 net->cc_mod.rtcc.rtt_set_this_sack = 0; 820 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 821 } 822 } 823 } 824 /* 825 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 826 * moved. 827 */ 828 if (accum_moved || 829 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 830 /* If the cumulative ack moved we can proceed */ 831 if (net->cwnd <= net->ssthresh) { 832 /* We are in slow start */ 833 if (net->flight_size + net->net_ack >= net->cwnd) { 834 uint32_t limit; 835 836 old_cwnd = net->cwnd; 837 switch (asoc->sctp_cmt_on_off) { 838 case SCTP_CMT_RPV1: 839 limit = (uint32_t)(((uint64_t)net->mtu * 840 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 841 (uint64_t)net->ssthresh) / 842 (uint64_t)t_ssthresh); 843 incr = (uint32_t)(((uint64_t)net->net_ack * 844 (uint64_t)net->ssthresh) / 845 (uint64_t)t_ssthresh); 846 if (incr > limit) { 847 incr = limit; 848 } 849 if (incr == 0) { 850 incr = 1; 851 } 852 break; 853 case SCTP_CMT_RPV2: 854 /* 855 * lastsa>>3; we don't need 856 * to divide ... 857 */ 858 srtt = net->lastsa; 859 if (srtt == 0) { 860 srtt = 1; 861 } 862 limit = (uint32_t)(((uint64_t)net->mtu * 863 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 864 (uint64_t)net->cwnd) / 865 ((uint64_t)srtt * t_ucwnd_sbw)); 866 /* INCREASE FACTOR */ 867 incr = (uint32_t)(((uint64_t)net->net_ack * 868 (uint64_t)net->cwnd) / 869 ((uint64_t)srtt * t_ucwnd_sbw)); 870 /* INCREASE FACTOR */ 871 if (incr > limit) { 872 incr = limit; 873 } 874 if (incr == 0) { 875 incr = 1; 876 } 877 break; 878 case SCTP_CMT_MPTCP: 879 limit = (uint32_t)(((uint64_t)net->mtu * 880 mptcp_like_alpha * 881 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) >> 882 SHIFT_MPTCP_MULTI); 883 incr = (uint32_t)(((uint64_t)net->net_ack * 884 mptcp_like_alpha) >> 885 SHIFT_MPTCP_MULTI); 886 if (incr > limit) { 887 incr = limit; 888 } 889 if (incr > net->net_ack) { 890 incr = net->net_ack; 891 } 892 if (incr > net->mtu) { 893 incr = net->mtu; 894 } 895 break; 896 default: 897 incr = net->net_ack; 898 if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) { 899 incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable); 900 } 901 break; 902 } 903 net->cwnd += incr; 904 sctp_enforce_cwnd_limit(asoc, net); 905 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 906 sctp_log_cwnd(stcb, net, incr, 907 SCTP_CWND_LOG_FROM_SS); 908 } 909 SDT_PROBE5(sctp, cwnd, net, ack, 910 stcb->asoc.my_vtag, 911 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 912 net, 913 old_cwnd, net->cwnd); 914 } else { 915 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 916 sctp_log_cwnd(stcb, net, net->net_ack, 917 SCTP_CWND_LOG_NOADV_SS); 918 } 919 } 920 } else { 921 /* We are in congestion avoidance */ 922 /* 923 * Add to pba 924 */ 925 net->partial_bytes_acked += net->net_ack; 926 927 if ((net->flight_size + net->net_ack >= net->cwnd) && 928 (net->partial_bytes_acked >= net->cwnd)) { 929 net->partial_bytes_acked -= net->cwnd; 930 old_cwnd = net->cwnd; 931 switch (asoc->sctp_cmt_on_off) { 932 case SCTP_CMT_RPV1: 933 incr = (uint32_t)(((uint64_t)net->mtu * 934 (uint64_t)net->ssthresh) / 935 (uint64_t)t_ssthresh); 936 if (incr == 0) { 937 incr = 1; 938 } 939 break; 940 case SCTP_CMT_RPV2: 941 /* 942 * lastsa>>3; we don't need 943 * to divide ... 944 */ 945 srtt = net->lastsa; 946 if (srtt == 0) { 947 srtt = 1; 948 } 949 incr = (uint32_t)((uint64_t)net->mtu * 950 (uint64_t)net->cwnd / 951 ((uint64_t)srtt * 952 t_ucwnd_sbw)); 953 /* INCREASE FACTOR */ 954 if (incr == 0) { 955 incr = 1; 956 } 957 break; 958 case SCTP_CMT_MPTCP: 959 incr = (uint32_t)((mptcp_like_alpha * 960 (uint64_t)net->cwnd) >> 961 SHIFT_MPTCP_MULTI); 962 if (incr > net->mtu) { 963 incr = net->mtu; 964 } 965 break; 966 default: 967 incr = net->mtu; 968 break; 969 } 970 net->cwnd += incr; 971 sctp_enforce_cwnd_limit(asoc, net); 972 SDT_PROBE5(sctp, cwnd, net, ack, 973 stcb->asoc.my_vtag, 974 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 975 net, 976 old_cwnd, net->cwnd); 977 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 978 sctp_log_cwnd(stcb, net, net->mtu, 979 SCTP_CWND_LOG_FROM_CA); 980 } 981 } else { 982 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 983 sctp_log_cwnd(stcb, net, net->net_ack, 984 SCTP_CWND_LOG_NOADV_CA); 985 } 986 } 987 } 988 } else { 989 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 990 sctp_log_cwnd(stcb, net, net->mtu, 991 SCTP_CWND_LOG_NO_CUMACK); 992 } 993 } 994 } 995 } 996 997 static void 998 sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb, struct sctp_nets *net) 999 { 1000 int old_cwnd; 1001 1002 old_cwnd = net->cwnd; 1003 net->cwnd = net->mtu; 1004 SDT_PROBE5(sctp, cwnd, net, ack, 1005 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 1006 old_cwnd, net->cwnd); 1007 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1008 (void *)net, net->cwnd); 1009 } 1010 1011 static void 1012 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 1013 { 1014 int old_cwnd = net->cwnd; 1015 uint32_t t_ssthresh, t_cwnd; 1016 uint64_t t_ucwnd_sbw; 1017 1018 /* MT FIXME: Don't compute this over and over again */ 1019 t_ssthresh = 0; 1020 t_cwnd = 0; 1021 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 1022 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) { 1023 struct sctp_nets *lnet; 1024 uint32_t srtt; 1025 1026 t_ucwnd_sbw = 0; 1027 TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) { 1028 t_ssthresh += lnet->ssthresh; 1029 t_cwnd += lnet->cwnd; 1030 srtt = lnet->lastsa; 1031 /* lastsa>>3; we don't need to divide ... */ 1032 if (srtt > 0) { 1033 t_ucwnd_sbw += (uint64_t)lnet->cwnd / (uint64_t)srtt; 1034 } 1035 } 1036 if (t_ssthresh < 1) { 1037 t_ssthresh = 1; 1038 } 1039 if (t_ucwnd_sbw < 1) { 1040 t_ucwnd_sbw = 1; 1041 } 1042 if (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) { 1043 net->ssthresh = (uint32_t)(((uint64_t)4 * 1044 (uint64_t)net->mtu * 1045 (uint64_t)net->ssthresh) / 1046 (uint64_t)t_ssthresh); 1047 } else { 1048 uint64_t cc_delta; 1049 1050 srtt = net->lastsa; 1051 /* lastsa>>3; we don't need to divide ... */ 1052 if (srtt == 0) { 1053 srtt = 1; 1054 } 1055 cc_delta = t_ucwnd_sbw * (uint64_t)srtt / 2; 1056 if (cc_delta < t_cwnd) { 1057 net->ssthresh = (uint32_t)((uint64_t)t_cwnd - cc_delta); 1058 } else { 1059 net->ssthresh = net->mtu; 1060 } 1061 } 1062 if ((net->cwnd > t_cwnd / 2) && 1063 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 1064 net->ssthresh = net->cwnd - t_cwnd / 2; 1065 } 1066 if (net->ssthresh < net->mtu) { 1067 net->ssthresh = net->mtu; 1068 } 1069 } else { 1070 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 1071 } 1072 net->cwnd = net->mtu; 1073 net->partial_bytes_acked = 0; 1074 SDT_PROBE5(sctp, cwnd, net, to, 1075 stcb->asoc.my_vtag, 1076 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1077 net, 1078 old_cwnd, net->cwnd); 1079 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1080 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1081 } 1082 } 1083 1084 static void 1085 sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net, 1086 int in_window, int num_pkt_lost, int use_rtcc) 1087 { 1088 int old_cwnd = net->cwnd; 1089 1090 if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) { 1091 /* Data center Congestion Control */ 1092 if (in_window == 0) { 1093 /* 1094 * Go to CA with the cwnd at the point we sent the 1095 * TSN that was marked with a CE. 1096 */ 1097 if (net->ecn_prev_cwnd < net->cwnd) { 1098 /* Restore to prev cwnd */ 1099 net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost); 1100 } else { 1101 /* Just cut in 1/2 */ 1102 net->cwnd /= 2; 1103 } 1104 /* Drop to CA */ 1105 net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu); 1106 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1107 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1108 } 1109 } else { 1110 /* 1111 * Further tuning down required over the drastic 1112 * original cut 1113 */ 1114 net->ssthresh -= (net->mtu * num_pkt_lost); 1115 net->cwnd -= (net->mtu * num_pkt_lost); 1116 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1117 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1118 } 1119 } 1120 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1121 } else { 1122 if (in_window == 0) { 1123 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1124 net->ssthresh = net->cwnd / 2; 1125 if (net->ssthresh < net->mtu) { 1126 net->ssthresh = net->mtu; 1127 /* 1128 * here back off the timer as well, to slow 1129 * us down 1130 */ 1131 net->RTO <<= 1; 1132 } 1133 net->cwnd = net->ssthresh; 1134 SDT_PROBE5(sctp, cwnd, net, ecn, 1135 stcb->asoc.my_vtag, 1136 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1137 net, 1138 old_cwnd, net->cwnd); 1139 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1140 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1141 } 1142 } 1143 } 1144 1145 } 1146 1147 static void 1148 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 1149 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 1150 uint32_t *bottle_bw, uint32_t *on_queue) 1151 { 1152 uint32_t bw_avail; 1153 unsigned int incr; 1154 int old_cwnd = net->cwnd; 1155 1156 /* get bottle neck bw */ 1157 *bottle_bw = ntohl(cp->bottle_bw); 1158 /* and whats on queue */ 1159 *on_queue = ntohl(cp->current_onq); 1160 /* 1161 * adjust the on-queue if our flight is more it could be that the 1162 * router has not yet gotten data "in-flight" to it 1163 */ 1164 if (*on_queue < net->flight_size) { 1165 *on_queue = net->flight_size; 1166 } 1167 /* rtt is measured in micro seconds, bottle_bw in bytes per second */ 1168 bw_avail = (uint32_t)(((uint64_t)(*bottle_bw) * net->rtt) / (uint64_t)1000000); 1169 if (bw_avail > *bottle_bw) { 1170 /* 1171 * Cap the growth to no more than the bottle neck. This can 1172 * happen as RTT slides up due to queues. It also means if 1173 * you have more than a 1 second RTT with a empty queue you 1174 * will be limited to the bottle_bw per second no matter if 1175 * other points have 1/2 the RTT and you could get more 1176 * out... 1177 */ 1178 bw_avail = *bottle_bw; 1179 } 1180 if (*on_queue > bw_avail) { 1181 /* 1182 * No room for anything else don't allow anything else to be 1183 * "added to the fire". 1184 */ 1185 int seg_inflight, seg_onqueue, my_portion; 1186 1187 net->partial_bytes_acked = 0; 1188 /* how much are we over queue size? */ 1189 incr = *on_queue - bw_avail; 1190 if (stcb->asoc.seen_a_sack_this_pkt) { 1191 /* 1192 * undo any cwnd adjustment that the sack might have 1193 * made 1194 */ 1195 net->cwnd = net->prev_cwnd; 1196 } 1197 /* Now how much of that is mine? */ 1198 seg_inflight = net->flight_size / net->mtu; 1199 seg_onqueue = *on_queue / net->mtu; 1200 my_portion = (incr * seg_inflight) / seg_onqueue; 1201 1202 /* Have I made an adjustment already */ 1203 if (net->cwnd > net->flight_size) { 1204 /* 1205 * for this flight I made an adjustment we need to 1206 * decrease the portion by a share our previous 1207 * adjustment. 1208 */ 1209 int diff_adj; 1210 1211 diff_adj = net->cwnd - net->flight_size; 1212 if (diff_adj > my_portion) 1213 my_portion = 0; 1214 else 1215 my_portion -= diff_adj; 1216 } 1217 /* 1218 * back down to the previous cwnd (assume we have had a sack 1219 * before this packet). minus what ever portion of the 1220 * overage is my fault. 1221 */ 1222 net->cwnd -= my_portion; 1223 1224 /* we will NOT back down more than 1 MTU */ 1225 if (net->cwnd <= net->mtu) { 1226 net->cwnd = net->mtu; 1227 } 1228 /* force into CA */ 1229 net->ssthresh = net->cwnd - 1; 1230 } else { 1231 /* 1232 * Take 1/4 of the space left or max burst up .. whichever 1233 * is less. 1234 */ 1235 incr = (bw_avail - *on_queue) >> 2; 1236 if ((stcb->asoc.max_burst > 0) && 1237 (stcb->asoc.max_burst * net->mtu < incr)) { 1238 incr = stcb->asoc.max_burst * net->mtu; 1239 } 1240 net->cwnd += incr; 1241 } 1242 if (net->cwnd > bw_avail) { 1243 /* We can't exceed the pipe size */ 1244 net->cwnd = bw_avail; 1245 } 1246 if (net->cwnd < net->mtu) { 1247 /* We always have 1 MTU */ 1248 net->cwnd = net->mtu; 1249 } 1250 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1251 if (net->cwnd - old_cwnd != 0) { 1252 /* log only changes */ 1253 SDT_PROBE5(sctp, cwnd, net, pd, 1254 stcb->asoc.my_vtag, 1255 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1256 net, 1257 old_cwnd, net->cwnd); 1258 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1259 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1260 SCTP_CWND_LOG_FROM_SAT); 1261 } 1262 } 1263 } 1264 1265 static void 1266 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 1267 struct sctp_nets *net, int burst_limit) 1268 { 1269 int old_cwnd = net->cwnd; 1270 1271 if (net->ssthresh < net->cwnd) 1272 net->ssthresh = net->cwnd; 1273 if (burst_limit) { 1274 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 1275 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1276 SDT_PROBE5(sctp, cwnd, net, bl, 1277 stcb->asoc.my_vtag, 1278 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1279 net, 1280 old_cwnd, net->cwnd); 1281 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1282 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 1283 } 1284 } 1285 } 1286 1287 static void 1288 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1289 struct sctp_association *asoc, 1290 int accum_moved, int reneged_all, int will_exit) 1291 { 1292 /* Passing a zero argument in last disables the rtcc algorithm */ 1293 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0); 1294 } 1295 1296 static void 1297 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1298 int in_window, int num_pkt_lost) 1299 { 1300 /* Passing a zero argument in last disables the rtcc algorithm */ 1301 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0); 1302 } 1303 1304 /* Here starts the RTCCVAR type CC invented by RRS which 1305 * is a slight mod to RFC2581. We reuse a common routine or 1306 * two since these algorithms are so close and need to 1307 * remain the same. 1308 */ 1309 static void 1310 sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1311 int in_window, int num_pkt_lost) 1312 { 1313 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1); 1314 } 1315 1316 static void 1317 sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net, 1318 struct sctp_tmit_chunk *tp1) 1319 { 1320 net->cc_mod.rtcc.bw_bytes += tp1->send_size; 1321 } 1322 1323 static void 1324 sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb SCTP_UNUSED, 1325 struct sctp_nets *net) 1326 { 1327 if (net->cc_mod.rtcc.tls_needs_set > 0) { 1328 /* We had a bw measurment going on */ 1329 struct timeval ltls; 1330 1331 SCTP_GETPTIME_TIMEVAL(<ls); 1332 timevalsub(<ls, &net->cc_mod.rtcc.tls); 1333 net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec; 1334 } 1335 } 1336 1337 static void 1338 sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb, 1339 struct sctp_nets *net) 1340 { 1341 uint64_t vtag, probepoint; 1342 1343 if (net->cc_mod.rtcc.lbw) { 1344 /* Clear the old bw.. we went to 0 in-flight */ 1345 vtag = (net->rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1346 (stcb->rport); 1347 probepoint = (((uint64_t)net->cwnd) << 32); 1348 /* Probe point 8 */ 1349 probepoint |= ((8 << 16) | 0); 1350 SDT_PROBE5(sctp, cwnd, net, rttvar, 1351 vtag, 1352 ((net->cc_mod.rtcc.lbw << 32) | 0), 1353 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 1354 net->flight_size, 1355 probepoint); 1356 net->cc_mod.rtcc.lbw_rtt = 0; 1357 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1358 net->cc_mod.rtcc.lbw = 0; 1359 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1360 net->cc_mod.rtcc.vol_reduce = 0; 1361 net->cc_mod.rtcc.bw_tot_time = 0; 1362 net->cc_mod.rtcc.bw_bytes = 0; 1363 net->cc_mod.rtcc.tls_needs_set = 0; 1364 if (net->cc_mod.rtcc.steady_step) { 1365 net->cc_mod.rtcc.vol_reduce = 0; 1366 net->cc_mod.rtcc.step_cnt = 0; 1367 net->cc_mod.rtcc.last_step_state = 0; 1368 } 1369 if (net->cc_mod.rtcc.ret_from_eq) { 1370 /* less aggressive one - reset cwnd too */ 1371 uint32_t cwnd_in_mtu, cwnd; 1372 1373 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 1374 if (cwnd_in_mtu == 0) { 1375 /* 1376 * Using 0 means that the value of RFC 4960 1377 * is used. 1378 */ 1379 cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1380 } else { 1381 /* 1382 * We take the minimum of the burst limit 1383 * and the initial congestion window. 1384 */ 1385 if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst)) 1386 cwnd_in_mtu = stcb->asoc.max_burst; 1387 cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 1388 } 1389 if (net->cwnd > cwnd) { 1390 /* 1391 * Only set if we are not a timeout (i.e. 1392 * down to 1 mtu) 1393 */ 1394 net->cwnd = cwnd; 1395 } 1396 } 1397 } 1398 } 1399 1400 static void 1401 sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb, 1402 struct sctp_nets *net) 1403 { 1404 uint64_t vtag, probepoint; 1405 1406 sctp_set_initial_cc_param(stcb, net); 1407 stcb->asoc.use_precise_time = 1; 1408 probepoint = (((uint64_t)net->cwnd) << 32); 1409 probepoint |= ((9 << 16) | 0); 1410 vtag = (net->rtt << 32) | 1411 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1412 (stcb->rport); 1413 SDT_PROBE5(sctp, cwnd, net, rttvar, 1414 vtag, 1415 0, 1416 0, 1417 0, 1418 probepoint); 1419 net->cc_mod.rtcc.lbw_rtt = 0; 1420 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1421 net->cc_mod.rtcc.vol_reduce = 0; 1422 net->cc_mod.rtcc.lbw = 0; 1423 net->cc_mod.rtcc.vol_reduce = 0; 1424 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1425 net->cc_mod.rtcc.bw_tot_time = 0; 1426 net->cc_mod.rtcc.bw_bytes = 0; 1427 net->cc_mod.rtcc.tls_needs_set = 0; 1428 net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret); 1429 net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step); 1430 net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn); 1431 net->cc_mod.rtcc.step_cnt = 0; 1432 net->cc_mod.rtcc.last_step_state = 0; 1433 } 1434 1435 static int 1436 sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget, 1437 struct sctp_cc_option *cc_opt) 1438 { 1439 struct sctp_nets *net; 1440 1441 if (setorget == 1) { 1442 /* a set */ 1443 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1444 if ((cc_opt->aid_value.assoc_value != 0) && 1445 (cc_opt->aid_value.assoc_value != 1)) { 1446 return (EINVAL); 1447 } 1448 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1449 net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value; 1450 } 1451 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1452 if ((cc_opt->aid_value.assoc_value != 0) && 1453 (cc_opt->aid_value.assoc_value != 1)) { 1454 return (EINVAL); 1455 } 1456 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1457 net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value; 1458 } 1459 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1460 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1461 net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value; 1462 } 1463 } else { 1464 return (EINVAL); 1465 } 1466 } else { 1467 /* a get */ 1468 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1469 net = TAILQ_FIRST(&stcb->asoc.nets); 1470 if (net == NULL) { 1471 return (EFAULT); 1472 } 1473 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.ret_from_eq; 1474 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1475 net = TAILQ_FIRST(&stcb->asoc.nets); 1476 if (net == NULL) { 1477 return (EFAULT); 1478 } 1479 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.use_dccc_ecn; 1480 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1481 net = TAILQ_FIRST(&stcb->asoc.nets); 1482 if (net == NULL) { 1483 return (EFAULT); 1484 } 1485 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step; 1486 } else { 1487 return (EINVAL); 1488 } 1489 } 1490 return (0); 1491 } 1492 1493 static void 1494 sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb SCTP_UNUSED, 1495 struct sctp_nets *net) 1496 { 1497 if (net->cc_mod.rtcc.tls_needs_set == 0) { 1498 SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls); 1499 net->cc_mod.rtcc.tls_needs_set = 2; 1500 } 1501 } 1502 1503 static void 1504 sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb, 1505 struct sctp_association *asoc, 1506 int accum_moved, int reneged_all, int will_exit) 1507 { 1508 /* Passing a one argument at the last enables the rtcc algorithm */ 1509 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1); 1510 } 1511 1512 static void 1513 sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb SCTP_UNUSED, 1514 struct sctp_nets *net, 1515 struct timeval *now SCTP_UNUSED) 1516 { 1517 net->cc_mod.rtcc.rtt_set_this_sack = 1; 1518 } 1519 1520 /* Here starts Sally Floyds HS-TCP */ 1521 1522 struct sctp_hs_raise_drop { 1523 int32_t cwnd; 1524 int8_t increase; 1525 int8_t drop_percent; 1526 }; 1527 1528 #define SCTP_HS_TABLE_SIZE 73 1529 1530 static const struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 1531 {38, 1, 50}, /* 0 */ 1532 {118, 2, 44}, /* 1 */ 1533 {221, 3, 41}, /* 2 */ 1534 {347, 4, 38}, /* 3 */ 1535 {495, 5, 37}, /* 4 */ 1536 {663, 6, 35}, /* 5 */ 1537 {851, 7, 34}, /* 6 */ 1538 {1058, 8, 33}, /* 7 */ 1539 {1284, 9, 32}, /* 8 */ 1540 {1529, 10, 31}, /* 9 */ 1541 {1793, 11, 30}, /* 10 */ 1542 {2076, 12, 29}, /* 11 */ 1543 {2378, 13, 28}, /* 12 */ 1544 {2699, 14, 28}, /* 13 */ 1545 {3039, 15, 27}, /* 14 */ 1546 {3399, 16, 27}, /* 15 */ 1547 {3778, 17, 26}, /* 16 */ 1548 {4177, 18, 26}, /* 17 */ 1549 {4596, 19, 25}, /* 18 */ 1550 {5036, 20, 25}, /* 19 */ 1551 {5497, 21, 24}, /* 20 */ 1552 {5979, 22, 24}, /* 21 */ 1553 {6483, 23, 23}, /* 22 */ 1554 {7009, 24, 23}, /* 23 */ 1555 {7558, 25, 22}, /* 24 */ 1556 {8130, 26, 22}, /* 25 */ 1557 {8726, 27, 22}, /* 26 */ 1558 {9346, 28, 21}, /* 27 */ 1559 {9991, 29, 21}, /* 28 */ 1560 {10661, 30, 21}, /* 29 */ 1561 {11358, 31, 20}, /* 30 */ 1562 {12082, 32, 20}, /* 31 */ 1563 {12834, 33, 20}, /* 32 */ 1564 {13614, 34, 19}, /* 33 */ 1565 {14424, 35, 19}, /* 34 */ 1566 {15265, 36, 19}, /* 35 */ 1567 {16137, 37, 19}, /* 36 */ 1568 {17042, 38, 18}, /* 37 */ 1569 {17981, 39, 18}, /* 38 */ 1570 {18955, 40, 18}, /* 39 */ 1571 {19965, 41, 17}, /* 40 */ 1572 {21013, 42, 17}, /* 41 */ 1573 {22101, 43, 17}, /* 42 */ 1574 {23230, 44, 17}, /* 43 */ 1575 {24402, 45, 16}, /* 44 */ 1576 {25618, 46, 16}, /* 45 */ 1577 {26881, 47, 16}, /* 46 */ 1578 {28193, 48, 16}, /* 47 */ 1579 {29557, 49, 15}, /* 48 */ 1580 {30975, 50, 15}, /* 49 */ 1581 {32450, 51, 15}, /* 50 */ 1582 {33986, 52, 15}, /* 51 */ 1583 {35586, 53, 14}, /* 52 */ 1584 {37253, 54, 14}, /* 53 */ 1585 {38992, 55, 14}, /* 54 */ 1586 {40808, 56, 14}, /* 55 */ 1587 {42707, 57, 13}, /* 56 */ 1588 {44694, 58, 13}, /* 57 */ 1589 {46776, 59, 13}, /* 58 */ 1590 {48961, 60, 13}, /* 59 */ 1591 {51258, 61, 13}, /* 60 */ 1592 {53677, 62, 12}, /* 61 */ 1593 {56230, 63, 12}, /* 62 */ 1594 {58932, 64, 12}, /* 63 */ 1595 {61799, 65, 12}, /* 64 */ 1596 {64851, 66, 11}, /* 65 */ 1597 {68113, 67, 11}, /* 66 */ 1598 {71617, 68, 11}, /* 67 */ 1599 {75401, 69, 10}, /* 68 */ 1600 {79517, 70, 10}, /* 69 */ 1601 {84035, 71, 10}, /* 70 */ 1602 {89053, 72, 10}, /* 71 */ 1603 {94717, 73, 9} /* 72 */ 1604 }; 1605 1606 static void 1607 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 1608 { 1609 int cur_val, i, indx, incr; 1610 int old_cwnd = net->cwnd; 1611 1612 cur_val = net->cwnd >> 10; 1613 indx = SCTP_HS_TABLE_SIZE - 1; 1614 1615 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1616 /* normal mode */ 1617 if (net->net_ack > net->mtu) { 1618 net->cwnd += net->mtu; 1619 } else { 1620 net->cwnd += net->net_ack; 1621 } 1622 } else { 1623 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 1624 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 1625 indx = i; 1626 break; 1627 } 1628 } 1629 net->last_hs_used = indx; 1630 incr = (((int32_t)sctp_cwnd_adjust[indx].increase) << 10); 1631 net->cwnd += incr; 1632 } 1633 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1634 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1635 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SS); 1636 } 1637 } 1638 1639 static void 1640 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 1641 { 1642 int cur_val, i, indx; 1643 int old_cwnd = net->cwnd; 1644 1645 cur_val = net->cwnd >> 10; 1646 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1647 /* normal mode */ 1648 net->ssthresh = net->cwnd / 2; 1649 if (net->ssthresh < (net->mtu * 2)) { 1650 net->ssthresh = 2 * net->mtu; 1651 } 1652 net->cwnd = net->ssthresh; 1653 } else { 1654 /* drop by the proper amount */ 1655 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 1656 (int32_t)sctp_cwnd_adjust[net->last_hs_used].drop_percent); 1657 net->cwnd = net->ssthresh; 1658 /* now where are we */ 1659 indx = net->last_hs_used; 1660 cur_val = net->cwnd >> 10; 1661 /* reset where we are in the table */ 1662 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1663 /* feel out of hs */ 1664 net->last_hs_used = 0; 1665 } else { 1666 for (i = indx; i >= 1; i--) { 1667 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 1668 break; 1669 } 1670 } 1671 net->last_hs_used = indx; 1672 } 1673 } 1674 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1675 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1676 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 1677 } 1678 } 1679 1680 static void 1681 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 1682 struct sctp_association *asoc) 1683 { 1684 struct sctp_nets *net; 1685 1686 /* 1687 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 1688 * (net->fast_retran_loss_recovery == 0))) 1689 */ 1690 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1691 if ((asoc->fast_retran_loss_recovery == 0) || 1692 (asoc->sctp_cmt_on_off > 0)) { 1693 /* out of a RFC2582 Fast recovery window? */ 1694 if (net->net_ack > 0) { 1695 /* 1696 * per section 7.2.3, are there any 1697 * destinations that had a fast retransmit 1698 * to them. If so what we need to do is 1699 * adjust ssthresh and cwnd. 1700 */ 1701 struct sctp_tmit_chunk *lchk; 1702 1703 sctp_hs_cwnd_decrease(stcb, net); 1704 1705 lchk = TAILQ_FIRST(&asoc->send_queue); 1706 1707 net->partial_bytes_acked = 0; 1708 /* Turn on fast recovery window */ 1709 asoc->fast_retran_loss_recovery = 1; 1710 if (lchk == NULL) { 1711 /* Mark end of the window */ 1712 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1713 } else { 1714 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1715 } 1716 1717 /* 1718 * CMT fast recovery -- per destination 1719 * recovery variable. 1720 */ 1721 net->fast_retran_loss_recovery = 1; 1722 1723 if (lchk == NULL) { 1724 /* Mark end of the window */ 1725 net->fast_recovery_tsn = asoc->sending_seq - 1; 1726 } else { 1727 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1728 } 1729 1730 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1731 stcb->sctp_ep, stcb, net, 1732 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_2); 1733 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1734 stcb->sctp_ep, stcb, net); 1735 } 1736 } else if (net->net_ack > 0) { 1737 /* 1738 * Mark a peg that we WOULD have done a cwnd 1739 * reduction but RFC2582 prevented this action. 1740 */ 1741 SCTP_STAT_INCR(sctps_fastretransinrtt); 1742 } 1743 } 1744 } 1745 1746 static void 1747 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 1748 struct sctp_association *asoc, 1749 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 1750 { 1751 struct sctp_nets *net; 1752 1753 /******************************/ 1754 /* update cwnd and Early FR */ 1755 /******************************/ 1756 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1757 #ifdef JANA_CMT_FAST_RECOVERY 1758 /* 1759 * CMT fast recovery code. Need to debug. 1760 */ 1761 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1762 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 1763 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 1764 net->will_exit_fast_recovery = 1; 1765 } 1766 } 1767 #endif 1768 /* if nothing was acked on this destination skip it */ 1769 if (net->net_ack == 0) { 1770 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1771 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1772 } 1773 continue; 1774 } 1775 #ifdef JANA_CMT_FAST_RECOVERY 1776 /* 1777 * CMT fast recovery code 1778 */ 1779 /* 1780 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 1781 * && net->will_exit_fast_recovery == 0) { @@@ Do something 1782 * } else if (sctp_cmt_on_off == 0 && 1783 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1784 */ 1785 #endif 1786 1787 if (asoc->fast_retran_loss_recovery && 1788 (will_exit == 0) && 1789 (asoc->sctp_cmt_on_off == 0)) { 1790 /* 1791 * If we are in loss recovery we skip any cwnd 1792 * update 1793 */ 1794 return; 1795 } 1796 /* 1797 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1798 * moved. 1799 */ 1800 if (accum_moved || 1801 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 1802 /* If the cumulative ack moved we can proceed */ 1803 if (net->cwnd <= net->ssthresh) { 1804 /* We are in slow start */ 1805 if (net->flight_size + net->net_ack >= net->cwnd) { 1806 sctp_hs_cwnd_increase(stcb, net); 1807 } else { 1808 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1809 sctp_log_cwnd(stcb, net, net->net_ack, 1810 SCTP_CWND_LOG_NOADV_SS); 1811 } 1812 } 1813 } else { 1814 /* We are in congestion avoidance */ 1815 net->partial_bytes_acked += net->net_ack; 1816 if ((net->flight_size + net->net_ack >= net->cwnd) && 1817 (net->partial_bytes_acked >= net->cwnd)) { 1818 net->partial_bytes_acked -= net->cwnd; 1819 net->cwnd += net->mtu; 1820 sctp_enforce_cwnd_limit(asoc, net); 1821 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1822 sctp_log_cwnd(stcb, net, net->mtu, 1823 SCTP_CWND_LOG_FROM_CA); 1824 } 1825 } else { 1826 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1827 sctp_log_cwnd(stcb, net, net->net_ack, 1828 SCTP_CWND_LOG_NOADV_CA); 1829 } 1830 } 1831 } 1832 } else { 1833 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1834 sctp_log_cwnd(stcb, net, net->mtu, 1835 SCTP_CWND_LOG_NO_CUMACK); 1836 } 1837 } 1838 } 1839 } 1840 1841 /* 1842 * H-TCP congestion control. The algorithm is detailed in: 1843 * R.N.Shorten, D.J.Leith: 1844 * "H-TCP: TCP for high-speed and long-distance networks" 1845 * Proc. PFLDnet, Argonne, 2004. 1846 * http://www.hamilton.ie/net/htcp3.pdf 1847 */ 1848 1849 static int use_rtt_scaling = 1; 1850 static int use_bandwidth_switch = 1; 1851 1852 static inline int 1853 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1854 { 1855 return (seq3 - seq2 >= seq1 - seq2); 1856 } 1857 1858 static inline uint32_t 1859 htcp_cong_time(struct htcp *ca) 1860 { 1861 return (sctp_get_tick_count() - ca->last_cong); 1862 } 1863 1864 static inline uint32_t 1865 htcp_ccount(struct htcp *ca) 1866 { 1867 return (ca->minRTT == 0 ? htcp_cong_time(ca) : htcp_cong_time(ca) / ca->minRTT); 1868 } 1869 1870 static inline void 1871 htcp_reset(struct htcp *ca) 1872 { 1873 ca->undo_last_cong = ca->last_cong; 1874 ca->undo_maxRTT = ca->maxRTT; 1875 ca->undo_old_maxB = ca->old_maxB; 1876 ca->last_cong = sctp_get_tick_count(); 1877 } 1878 1879 #ifdef SCTP_NOT_USED 1880 1881 static uint32_t 1882 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1883 { 1884 net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong; 1885 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT; 1886 net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB; 1887 return (max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->cc_mod.htcp_ca.beta) * net->mtu)); 1888 } 1889 1890 #endif 1891 1892 static inline void 1893 measure_rtt(struct sctp_nets *net) 1894 { 1895 uint32_t srtt = net->lastsa >> SCTP_RTT_SHIFT; 1896 1897 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1898 if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT) 1899 net->cc_mod.htcp_ca.minRTT = srtt; 1900 1901 /* max RTT */ 1902 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) { 1903 if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT) 1904 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT; 1905 if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT + sctp_msecs_to_ticks(20)) 1906 net->cc_mod.htcp_ca.maxRTT = srtt; 1907 } 1908 } 1909 1910 static void 1911 measure_achieved_throughput(struct sctp_nets *net) 1912 { 1913 uint32_t now = sctp_get_tick_count(); 1914 1915 if (net->fast_retran_ip == 0) 1916 net->cc_mod.htcp_ca.bytes_acked = net->net_ack; 1917 1918 if (!use_bandwidth_switch) 1919 return; 1920 1921 /* achieved throughput calculations */ 1922 /* JRS - not 100% sure of this statement */ 1923 if (net->fast_retran_ip == 1) { 1924 net->cc_mod.htcp_ca.bytecount = 0; 1925 net->cc_mod.htcp_ca.lasttime = now; 1926 return; 1927 } 1928 1929 net->cc_mod.htcp_ca.bytecount += net->net_ack; 1930 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)) && 1931 (now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT) && 1932 (net->cc_mod.htcp_ca.minRTT > 0)) { 1933 uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount / net->mtu * hz / (now - net->cc_mod.htcp_ca.lasttime); 1934 1935 if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) { 1936 /* just after backoff */ 1937 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi; 1938 } else { 1939 net->cc_mod.htcp_ca.Bi = (3 * net->cc_mod.htcp_ca.Bi + cur_Bi) / 4; 1940 if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB) 1941 net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi; 1942 if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB) 1943 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB; 1944 } 1945 net->cc_mod.htcp_ca.bytecount = 0; 1946 net->cc_mod.htcp_ca.lasttime = now; 1947 } 1948 } 1949 1950 static inline void 1951 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1952 { 1953 if (use_bandwidth_switch) { 1954 uint32_t maxB = ca->maxB; 1955 uint32_t old_maxB = ca->old_maxB; 1956 1957 ca->old_maxB = ca->maxB; 1958 1959 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1960 ca->beta = BETA_MIN; 1961 ca->modeswitch = 0; 1962 return; 1963 } 1964 } 1965 1966 if (ca->modeswitch && minRTT > sctp_msecs_to_ticks(10) && maxRTT) { 1967 ca->beta = (minRTT << 7) / maxRTT; 1968 if (ca->beta < BETA_MIN) 1969 ca->beta = BETA_MIN; 1970 else if (ca->beta > BETA_MAX) 1971 ca->beta = BETA_MAX; 1972 } else { 1973 ca->beta = BETA_MIN; 1974 ca->modeswitch = 1; 1975 } 1976 } 1977 1978 static inline void 1979 htcp_alpha_update(struct htcp *ca) 1980 { 1981 uint32_t minRTT = ca->minRTT; 1982 uint32_t factor = 1; 1983 uint32_t diff = htcp_cong_time(ca); 1984 1985 if (diff > (uint32_t)hz) { 1986 diff -= hz; 1987 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1988 } 1989 1990 if (use_rtt_scaling && minRTT) { 1991 uint32_t scale = (hz << 3) / (10 * minRTT); 1992 1993 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 1994 * interval [0.5,10]<<3 */ 1995 factor = (factor << 3) / scale; 1996 if (factor != 0) 1997 factor = 1; 1998 } 1999 2000 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 2001 if (ca->alpha != 0) 2002 ca->alpha = ALPHA_BASE; 2003 } 2004 2005 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 2006 * rtt before we adjust our beta to ensure we are working from a consistent 2007 * data. 2008 * 2009 * This function should be called when we hit a congestion event since only at 2010 * that point do we really have a real sense of maxRTT (the queues en route 2011 * were getting just too full now). 2012 */ 2013 static void 2014 htcp_param_update(struct sctp_nets *net) 2015 { 2016 uint32_t minRTT = net->cc_mod.htcp_ca.minRTT; 2017 uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT; 2018 2019 htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT); 2020 htcp_alpha_update(&net->cc_mod.htcp_ca); 2021 2022 /* 2023 * add slowly fading memory for maxRTT to accommodate routing 2024 * changes etc 2025 */ 2026 if (minRTT > 0 && maxRTT > minRTT) 2027 net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 2028 } 2029 2030 static uint32_t 2031 htcp_recalc_ssthresh(struct sctp_nets *net) 2032 { 2033 htcp_param_update(net); 2034 return (max(((net->cwnd / net->mtu * net->cc_mod.htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu)); 2035 } 2036 2037 static void 2038 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 2039 { 2040 /*- 2041 * How to handle these functions? 2042 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 2043 * return; 2044 */ 2045 if (net->cwnd <= net->ssthresh) { 2046 /* We are in slow start */ 2047 if (net->flight_size + net->net_ack >= net->cwnd) { 2048 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 2049 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 2050 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2051 sctp_log_cwnd(stcb, net, net->mtu, 2052 SCTP_CWND_LOG_FROM_SS); 2053 } 2054 2055 } else { 2056 net->cwnd += net->net_ack; 2057 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2058 sctp_log_cwnd(stcb, net, net->net_ack, 2059 SCTP_CWND_LOG_FROM_SS); 2060 } 2061 } 2062 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2063 } else { 2064 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2065 sctp_log_cwnd(stcb, net, net->net_ack, 2066 SCTP_CWND_LOG_NOADV_SS); 2067 } 2068 } 2069 } else { 2070 measure_rtt(net); 2071 2072 /* 2073 * In dangerous area, increase slowly. In theory this is 2074 * net->cwnd += alpha / net->cwnd 2075 */ 2076 /* What is snd_cwnd_cnt?? */ 2077 if (((net->partial_bytes_acked / net->mtu * net->cc_mod.htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 2078 /*- 2079 * Does SCTP have a cwnd clamp? 2080 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 2081 */ 2082 net->cwnd += net->mtu; 2083 net->partial_bytes_acked = 0; 2084 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2085 htcp_alpha_update(&net->cc_mod.htcp_ca); 2086 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2087 sctp_log_cwnd(stcb, net, net->mtu, 2088 SCTP_CWND_LOG_FROM_CA); 2089 } 2090 } else { 2091 net->partial_bytes_acked += net->net_ack; 2092 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2093 sctp_log_cwnd(stcb, net, net->net_ack, 2094 SCTP_CWND_LOG_NOADV_CA); 2095 } 2096 } 2097 2098 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2099 } 2100 } 2101 2102 #ifdef SCTP_NOT_USED 2103 /* Lower bound on congestion window. */ 2104 static uint32_t 2105 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 2106 { 2107 return (net->ssthresh); 2108 } 2109 #endif 2110 2111 static void 2112 htcp_init(struct sctp_nets *net) 2113 { 2114 memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp)); 2115 net->cc_mod.htcp_ca.alpha = ALPHA_BASE; 2116 net->cc_mod.htcp_ca.beta = BETA_MIN; 2117 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2118 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count(); 2119 } 2120 2121 static void 2122 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 2123 { 2124 /* 2125 * We take the max of the burst limit times a MTU or the 2126 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 2127 */ 2128 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 2129 net->ssthresh = stcb->asoc.peers_rwnd; 2130 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2131 htcp_init(net); 2132 2133 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 2134 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 2135 } 2136 } 2137 2138 static void 2139 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 2140 struct sctp_association *asoc, 2141 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 2142 { 2143 struct sctp_nets *net; 2144 2145 /******************************/ 2146 /* update cwnd and Early FR */ 2147 /******************************/ 2148 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2149 #ifdef JANA_CMT_FAST_RECOVERY 2150 /* 2151 * CMT fast recovery code. Need to debug. 2152 */ 2153 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 2154 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 2155 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 2156 net->will_exit_fast_recovery = 1; 2157 } 2158 } 2159 #endif 2160 /* if nothing was acked on this destination skip it */ 2161 if (net->net_ack == 0) { 2162 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2163 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 2164 } 2165 continue; 2166 } 2167 #ifdef JANA_CMT_FAST_RECOVERY 2168 /* 2169 * CMT fast recovery code 2170 */ 2171 /* 2172 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 2173 * && net->will_exit_fast_recovery == 0) { @@@ Do something 2174 * } else if (sctp_cmt_on_off == 0 && 2175 * asoc->fast_retran_loss_recovery && will_exit == 0) { 2176 */ 2177 #endif 2178 2179 if (asoc->fast_retran_loss_recovery && 2180 will_exit == 0 && 2181 (asoc->sctp_cmt_on_off == 0)) { 2182 /* 2183 * If we are in loss recovery we skip any cwnd 2184 * update 2185 */ 2186 return; 2187 } 2188 /* 2189 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 2190 * moved. 2191 */ 2192 if (accum_moved || 2193 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 2194 htcp_cong_avoid(stcb, net); 2195 measure_achieved_throughput(net); 2196 } else { 2197 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2198 sctp_log_cwnd(stcb, net, net->mtu, 2199 SCTP_CWND_LOG_NO_CUMACK); 2200 } 2201 } 2202 } 2203 } 2204 2205 static void 2206 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 2207 struct sctp_association *asoc) 2208 { 2209 struct sctp_nets *net; 2210 2211 /* 2212 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 2213 * (net->fast_retran_loss_recovery == 0))) 2214 */ 2215 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2216 if ((asoc->fast_retran_loss_recovery == 0) || 2217 (asoc->sctp_cmt_on_off > 0)) { 2218 /* out of a RFC2582 Fast recovery window? */ 2219 if (net->net_ack > 0) { 2220 /* 2221 * per section 7.2.3, are there any 2222 * destinations that had a fast retransmit 2223 * to them. If so what we need to do is 2224 * adjust ssthresh and cwnd. 2225 */ 2226 struct sctp_tmit_chunk *lchk; 2227 int old_cwnd = net->cwnd; 2228 2229 /* JRS - reset as if state were changed */ 2230 htcp_reset(&net->cc_mod.htcp_ca); 2231 net->ssthresh = htcp_recalc_ssthresh(net); 2232 net->cwnd = net->ssthresh; 2233 sctp_enforce_cwnd_limit(asoc, net); 2234 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2235 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 2236 SCTP_CWND_LOG_FROM_FR); 2237 } 2238 lchk = TAILQ_FIRST(&asoc->send_queue); 2239 2240 net->partial_bytes_acked = 0; 2241 /* Turn on fast recovery window */ 2242 asoc->fast_retran_loss_recovery = 1; 2243 if (lchk == NULL) { 2244 /* Mark end of the window */ 2245 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 2246 } else { 2247 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2248 } 2249 2250 /* 2251 * CMT fast recovery -- per destination 2252 * recovery variable. 2253 */ 2254 net->fast_retran_loss_recovery = 1; 2255 2256 if (lchk == NULL) { 2257 /* Mark end of the window */ 2258 net->fast_recovery_tsn = asoc->sending_seq - 1; 2259 } else { 2260 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2261 } 2262 2263 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 2264 stcb->sctp_ep, stcb, net, 2265 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_3); 2266 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 2267 stcb->sctp_ep, stcb, net); 2268 } 2269 } else if (net->net_ack > 0) { 2270 /* 2271 * Mark a peg that we WOULD have done a cwnd 2272 * reduction but RFC2582 prevented this action. 2273 */ 2274 SCTP_STAT_INCR(sctps_fastretransinrtt); 2275 } 2276 } 2277 } 2278 2279 static void 2280 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 2281 struct sctp_nets *net) 2282 { 2283 int old_cwnd = net->cwnd; 2284 2285 /* JRS - reset as if the state were being changed to timeout */ 2286 htcp_reset(&net->cc_mod.htcp_ca); 2287 net->ssthresh = htcp_recalc_ssthresh(net); 2288 net->cwnd = net->mtu; 2289 net->partial_bytes_acked = 0; 2290 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2291 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 2292 } 2293 } 2294 2295 static void 2296 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 2297 struct sctp_nets *net, int in_window, int num_pkt_lost SCTP_UNUSED) 2298 { 2299 int old_cwnd; 2300 2301 old_cwnd = net->cwnd; 2302 2303 /* JRS - reset hctp as if state changed */ 2304 if (in_window == 0) { 2305 htcp_reset(&net->cc_mod.htcp_ca); 2306 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 2307 net->ssthresh = htcp_recalc_ssthresh(net); 2308 if (net->ssthresh < net->mtu) { 2309 net->ssthresh = net->mtu; 2310 /* here back off the timer as well, to slow us down */ 2311 net->RTO <<= 1; 2312 } 2313 net->cwnd = net->ssthresh; 2314 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2315 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2316 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 2317 } 2318 } 2319 } 2320 2321 const struct sctp_cc_functions sctp_cc_functions[] = { 2322 { 2323 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2324 .sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack, 2325 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2326 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2327 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2328 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2329 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2330 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2331 }, 2332 { 2333 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2334 .sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack, 2335 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2336 .sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr, 2337 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2338 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2339 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2340 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2341 }, 2342 { 2343 .sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param, 2344 .sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack, 2345 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2346 .sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr, 2347 .sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout, 2348 .sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo, 2349 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2350 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2351 }, 2352 { 2353 .sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param, 2354 .sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack, 2355 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2356 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2357 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2358 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo, 2359 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2360 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2361 .sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted, 2362 .sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged, 2363 .sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins, 2364 .sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack, 2365 .sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option, 2366 .sctp_rtt_calculated = sctp_rtt_rtcc_calculated 2367 } 2368 }; 2369