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_dtrace_declare.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 } 162 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2) { 163 uint32_t srtt; 164 165 srtt = net->lastsa; 166 /* 167 * lastsa>>3; we don't need 168 * to devide ... 169 */ 170 if (srtt == 0) { 171 srtt = 1; 172 } 173 /* 174 * Short Version => Equal to 175 * Contel Version MBe 176 */ 177 net->ssthresh = (uint32_t)(((uint64_t)4 * 178 (uint64_t)net->mtu * 179 (uint64_t)net->cwnd) / 180 ((uint64_t)srtt * 181 t_ucwnd_sbw)); 182 /* INCREASE FACTOR */ ; 183 } 184 if ((net->cwnd > t_cwnd / 2) && 185 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 186 net->ssthresh = net->cwnd - t_cwnd / 2; 187 } 188 if (net->ssthresh < net->mtu) { 189 net->ssthresh = net->mtu; 190 } 191 } else { 192 net->ssthresh = net->cwnd / 2; 193 if (net->ssthresh < (net->mtu * 2)) { 194 net->ssthresh = 2 * net->mtu; 195 } 196 } 197 net->cwnd = net->ssthresh; 198 sctp_enforce_cwnd_limit(asoc, net); 199 SDT_PROBE5(sctp, cwnd, net, fr, 200 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 201 old_cwnd, net->cwnd); 202 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 203 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 204 SCTP_CWND_LOG_FROM_FR); 205 } 206 lchk = TAILQ_FIRST(&asoc->send_queue); 207 208 net->partial_bytes_acked = 0; 209 /* Turn on fast recovery window */ 210 asoc->fast_retran_loss_recovery = 1; 211 if (lchk == NULL) { 212 /* Mark end of the window */ 213 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 214 } else { 215 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 216 } 217 218 /* 219 * CMT fast recovery -- per destination 220 * recovery variable. 221 */ 222 net->fast_retran_loss_recovery = 1; 223 224 if (lchk == NULL) { 225 /* Mark end of the window */ 226 net->fast_recovery_tsn = asoc->sending_seq - 1; 227 } else { 228 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 229 } 230 231 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 232 stcb->sctp_ep, stcb, net, 233 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_1); 234 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 235 stcb->sctp_ep, stcb, net); 236 } 237 } else if (net->net_ack > 0) { 238 /* 239 * Mark a peg that we WOULD have done a cwnd 240 * reduction but RFC2582 prevented this action. 241 */ 242 SCTP_STAT_INCR(sctps_fastretransinrtt); 243 } 244 } 245 } 246 247 /* Defines for instantaneous bw decisions */ 248 #define SCTP_INST_LOOSING 1 /* Losing to other flows */ 249 #define SCTP_INST_NEUTRAL 2 /* Neutral, no indication */ 250 #define SCTP_INST_GAINING 3 /* Gaining, step down possible */ 251 252 253 static int 254 cc_bw_same(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, 255 uint64_t rtt_offset, uint64_t vtag, uint8_t inst_ind) 256 { 257 uint64_t oth, probepoint; 258 259 probepoint = (((uint64_t)net->cwnd) << 32); 260 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 261 /* 262 * rtt increased we don't update bw.. so we don't update the 263 * rtt either. 264 */ 265 /* Probe point 5 */ 266 probepoint |= ((5 << 16) | 1); 267 SDT_PROBE5(sctp, cwnd, net, rttvar, 268 vtag, 269 ((net->cc_mod.rtcc.lbw << 32) | nbw), 270 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 271 net->flight_size, 272 probepoint); 273 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 274 if (net->cc_mod.rtcc.last_step_state == 5) 275 net->cc_mod.rtcc.step_cnt++; 276 else 277 net->cc_mod.rtcc.step_cnt = 1; 278 net->cc_mod.rtcc.last_step_state = 5; 279 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 280 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 281 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 282 /* Try a step down */ 283 oth = net->cc_mod.rtcc.vol_reduce; 284 oth <<= 16; 285 oth |= net->cc_mod.rtcc.step_cnt; 286 oth <<= 16; 287 oth |= net->cc_mod.rtcc.last_step_state; 288 SDT_PROBE5(sctp, cwnd, net, rttstep, 289 vtag, 290 ((net->cc_mod.rtcc.lbw << 32) | nbw), 291 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 292 oth, 293 probepoint); 294 if (net->cwnd > (4 * net->mtu)) { 295 net->cwnd -= net->mtu; 296 net->cc_mod.rtcc.vol_reduce++; 297 } else { 298 net->cc_mod.rtcc.step_cnt = 0; 299 } 300 } 301 } 302 return (1); 303 } 304 if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 305 /* 306 * rtt decreased, there could be more room. we update both 307 * the bw and the rtt here to lock this in as a good step 308 * down. 309 */ 310 /* Probe point 6 */ 311 probepoint |= ((6 << 16) | 0); 312 SDT_PROBE5(sctp, cwnd, net, rttvar, 313 vtag, 314 ((net->cc_mod.rtcc.lbw << 32) | nbw), 315 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 316 net->flight_size, 317 probepoint); 318 if (net->cc_mod.rtcc.steady_step) { 319 oth = net->cc_mod.rtcc.vol_reduce; 320 oth <<= 16; 321 oth |= net->cc_mod.rtcc.step_cnt; 322 oth <<= 16; 323 oth |= net->cc_mod.rtcc.last_step_state; 324 SDT_PROBE5(sctp, cwnd, net, rttstep, 325 vtag, 326 ((net->cc_mod.rtcc.lbw << 32) | nbw), 327 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 328 oth, 329 probepoint); 330 if ((net->cc_mod.rtcc.last_step_state == 5) && 331 (net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step)) { 332 /* Step down worked */ 333 net->cc_mod.rtcc.step_cnt = 0; 334 return (1); 335 } else { 336 net->cc_mod.rtcc.last_step_state = 6; 337 net->cc_mod.rtcc.step_cnt = 0; 338 } 339 } 340 net->cc_mod.rtcc.lbw = nbw; 341 net->cc_mod.rtcc.lbw_rtt = net->rtt; 342 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 343 if (inst_ind == SCTP_INST_GAINING) 344 return (1); 345 else if (inst_ind == SCTP_INST_NEUTRAL) 346 return (1); 347 else 348 return (0); 349 } 350 /* 351 * Ok bw and rtt remained the same .. no update to any 352 */ 353 /* Probe point 7 */ 354 probepoint |= ((7 << 16) | net->cc_mod.rtcc.ret_from_eq); 355 SDT_PROBE5(sctp, cwnd, net, rttvar, 356 vtag, 357 ((net->cc_mod.rtcc.lbw << 32) | nbw), 358 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 359 net->flight_size, 360 probepoint); 361 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 362 if (net->cc_mod.rtcc.last_step_state == 5) 363 net->cc_mod.rtcc.step_cnt++; 364 else 365 net->cc_mod.rtcc.step_cnt = 1; 366 net->cc_mod.rtcc.last_step_state = 5; 367 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 368 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 369 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 370 /* Try a step down */ 371 if (net->cwnd > (4 * net->mtu)) { 372 net->cwnd -= net->mtu; 373 net->cc_mod.rtcc.vol_reduce++; 374 return (1); 375 } else { 376 net->cc_mod.rtcc.step_cnt = 0; 377 } 378 } 379 } 380 if (inst_ind == SCTP_INST_GAINING) 381 return (1); 382 else if (inst_ind == SCTP_INST_NEUTRAL) 383 return (1); 384 else 385 return ((int)net->cc_mod.rtcc.ret_from_eq); 386 } 387 388 static int 389 cc_bw_decrease(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset, 390 uint64_t vtag, uint8_t inst_ind) 391 { 392 uint64_t oth, probepoint; 393 394 /* Bandwidth decreased. */ 395 probepoint = (((uint64_t)net->cwnd) << 32); 396 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 397 /* rtt increased */ 398 /* Did we add more */ 399 if ((net->cwnd > net->cc_mod.rtcc.cwnd_at_bw_set) && 400 (inst_ind != SCTP_INST_LOOSING)) { 401 /* We caused it maybe.. back off? */ 402 /* PROBE POINT 1 */ 403 probepoint |= ((1 << 16) | 1); 404 SDT_PROBE5(sctp, cwnd, net, rttvar, 405 vtag, 406 ((net->cc_mod.rtcc.lbw << 32) | nbw), 407 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 408 net->flight_size, 409 probepoint); 410 if (net->cc_mod.rtcc.ret_from_eq) { 411 /* 412 * Switch over to CA if we are less 413 * aggressive 414 */ 415 net->ssthresh = net->cwnd - 1; 416 net->partial_bytes_acked = 0; 417 } 418 return (1); 419 } 420 /* Probe point 2 */ 421 probepoint |= ((2 << 16) | 0); 422 SDT_PROBE5(sctp, cwnd, net, rttvar, 423 vtag, 424 ((net->cc_mod.rtcc.lbw << 32) | nbw), 425 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 426 net->flight_size, 427 probepoint); 428 /* Someone else - fight for more? */ 429 if (net->cc_mod.rtcc.steady_step) { 430 oth = net->cc_mod.rtcc.vol_reduce; 431 oth <<= 16; 432 oth |= net->cc_mod.rtcc.step_cnt; 433 oth <<= 16; 434 oth |= net->cc_mod.rtcc.last_step_state; 435 SDT_PROBE5(sctp, cwnd, net, rttstep, 436 vtag, 437 ((net->cc_mod.rtcc.lbw << 32) | nbw), 438 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 439 oth, 440 probepoint); 441 /* 442 * Did we voluntarily give up some? if so take one 443 * back please 444 */ 445 if ((net->cc_mod.rtcc.vol_reduce) && 446 (inst_ind != SCTP_INST_GAINING)) { 447 net->cwnd += net->mtu; 448 sctp_enforce_cwnd_limit(&stcb->asoc, net); 449 net->cc_mod.rtcc.vol_reduce--; 450 } 451 net->cc_mod.rtcc.last_step_state = 2; 452 net->cc_mod.rtcc.step_cnt = 0; 453 } 454 goto out_decision; 455 } else if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 456 /* bw & rtt decreased */ 457 /* Probe point 3 */ 458 probepoint |= ((3 << 16) | 0); 459 SDT_PROBE5(sctp, cwnd, net, rttvar, 460 vtag, 461 ((net->cc_mod.rtcc.lbw << 32) | nbw), 462 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 463 net->flight_size, 464 probepoint); 465 if (net->cc_mod.rtcc.steady_step) { 466 oth = net->cc_mod.rtcc.vol_reduce; 467 oth <<= 16; 468 oth |= net->cc_mod.rtcc.step_cnt; 469 oth <<= 16; 470 oth |= net->cc_mod.rtcc.last_step_state; 471 SDT_PROBE5(sctp, cwnd, net, rttstep, 472 vtag, 473 ((net->cc_mod.rtcc.lbw << 32) | nbw), 474 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 475 oth, 476 probepoint); 477 if ((net->cc_mod.rtcc.vol_reduce) && 478 (inst_ind != SCTP_INST_GAINING)) { 479 net->cwnd += net->mtu; 480 sctp_enforce_cwnd_limit(&stcb->asoc, net); 481 net->cc_mod.rtcc.vol_reduce--; 482 } 483 net->cc_mod.rtcc.last_step_state = 3; 484 net->cc_mod.rtcc.step_cnt = 0; 485 } 486 goto out_decision; 487 } 488 /* The bw decreased but rtt stayed the same */ 489 /* Probe point 4 */ 490 probepoint |= ((4 << 16) | 0); 491 SDT_PROBE5(sctp, cwnd, net, rttvar, 492 vtag, 493 ((net->cc_mod.rtcc.lbw << 32) | nbw), 494 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 495 net->flight_size, 496 probepoint); 497 if (net->cc_mod.rtcc.steady_step) { 498 oth = net->cc_mod.rtcc.vol_reduce; 499 oth <<= 16; 500 oth |= net->cc_mod.rtcc.step_cnt; 501 oth <<= 16; 502 oth |= net->cc_mod.rtcc.last_step_state; 503 SDT_PROBE5(sctp, cwnd, net, rttstep, 504 vtag, 505 ((net->cc_mod.rtcc.lbw << 32) | nbw), 506 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 507 oth, 508 probepoint); 509 if ((net->cc_mod.rtcc.vol_reduce) && 510 (inst_ind != SCTP_INST_GAINING)) { 511 net->cwnd += net->mtu; 512 sctp_enforce_cwnd_limit(&stcb->asoc, net); 513 net->cc_mod.rtcc.vol_reduce--; 514 } 515 net->cc_mod.rtcc.last_step_state = 4; 516 net->cc_mod.rtcc.step_cnt = 0; 517 } 518 out_decision: 519 net->cc_mod.rtcc.lbw = nbw; 520 net->cc_mod.rtcc.lbw_rtt = net->rtt; 521 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 522 if (inst_ind == SCTP_INST_GAINING) { 523 return (1); 524 } else { 525 return (0); 526 } 527 } 528 529 static int 530 cc_bw_increase(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t vtag) 531 { 532 uint64_t oth, probepoint; 533 534 /* 535 * BW increased, so update and return 0, since all actions in our 536 * table say to do the normal CC update. Note that we pay no 537 * attention to the inst_ind since our overall sum is increasing. 538 */ 539 /* PROBE POINT 0 */ 540 probepoint = (((uint64_t)net->cwnd) << 32); 541 SDT_PROBE5(sctp, cwnd, net, rttvar, 542 vtag, 543 ((net->cc_mod.rtcc.lbw << 32) | nbw), 544 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 545 net->flight_size, 546 probepoint); 547 if (net->cc_mod.rtcc.steady_step) { 548 oth = net->cc_mod.rtcc.vol_reduce; 549 oth <<= 16; 550 oth |= net->cc_mod.rtcc.step_cnt; 551 oth <<= 16; 552 oth |= net->cc_mod.rtcc.last_step_state; 553 SDT_PROBE5(sctp, cwnd, net, rttstep, 554 vtag, 555 ((net->cc_mod.rtcc.lbw << 32) | nbw), 556 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 557 oth, 558 probepoint); 559 net->cc_mod.rtcc.last_step_state = 0; 560 net->cc_mod.rtcc.step_cnt = 0; 561 net->cc_mod.rtcc.vol_reduce = 0; 562 } 563 net->cc_mod.rtcc.lbw = nbw; 564 net->cc_mod.rtcc.lbw_rtt = net->rtt; 565 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 566 return (0); 567 } 568 569 /* RTCC Algorithm to limit growth of cwnd, return 570 * true if you want to NOT allow cwnd growth 571 */ 572 static int 573 cc_bw_limit(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw) 574 { 575 uint64_t bw_offset, rtt_offset; 576 uint64_t probepoint, rtt, vtag; 577 uint64_t bytes_for_this_rtt, inst_bw; 578 uint64_t div, inst_off; 579 int bw_shift; 580 uint8_t inst_ind; 581 int ret; 582 583 /*- 584 * Here we need to see if we want 585 * to limit cwnd growth due to increase 586 * in overall rtt but no increase in bw. 587 * We use the following table to figure 588 * out what we should do. When we return 589 * 0, cc update goes on as planned. If we 590 * return 1, then no cc update happens and cwnd 591 * stays where it is at. 592 * ---------------------------------- 593 * BW | RTT | Action 594 * ********************************* 595 * INC | INC | return 0 596 * ---------------------------------- 597 * INC | SAME | return 0 598 * ---------------------------------- 599 * INC | DECR | return 0 600 * ---------------------------------- 601 * SAME | INC | return 1 602 * ---------------------------------- 603 * SAME | SAME | return 1 604 * ---------------------------------- 605 * SAME | DECR | return 0 606 * ---------------------------------- 607 * DECR | INC | return 0 or 1 based on if we caused. 608 * ---------------------------------- 609 * DECR | SAME | return 0 610 * ---------------------------------- 611 * DECR | DECR | return 0 612 * ---------------------------------- 613 * 614 * We are a bit fuzz on what an increase or 615 * decrease is. For BW it is the same if 616 * it did not change within 1/64th. For 617 * RTT it stayed the same if it did not 618 * change within 1/32nd 619 */ 620 bw_shift = SCTP_BASE_SYSCTL(sctp_rttvar_bw); 621 rtt = stcb->asoc.my_vtag; 622 vtag = (rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | (stcb->rport); 623 probepoint = (((uint64_t)net->cwnd) << 32); 624 rtt = net->rtt; 625 if (net->cc_mod.rtcc.rtt_set_this_sack) { 626 net->cc_mod.rtcc.rtt_set_this_sack = 0; 627 bytes_for_this_rtt = net->cc_mod.rtcc.bw_bytes - net->cc_mod.rtcc.bw_bytes_at_last_rttc; 628 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 629 if (net->rtt) { 630 div = net->rtt / 1000; 631 if (div) { 632 inst_bw = bytes_for_this_rtt / div; 633 inst_off = inst_bw >> bw_shift; 634 if (inst_bw > nbw) 635 inst_ind = SCTP_INST_GAINING; 636 else if ((inst_bw + inst_off) < nbw) 637 inst_ind = SCTP_INST_LOOSING; 638 else 639 inst_ind = SCTP_INST_NEUTRAL; 640 probepoint |= ((0xb << 16) | inst_ind); 641 } else { 642 inst_ind = net->cc_mod.rtcc.last_inst_ind; 643 inst_bw = bytes_for_this_rtt / (uint64_t)(net->rtt); 644 /* Can't determine do not change */ 645 probepoint |= ((0xc << 16) | inst_ind); 646 } 647 } else { 648 inst_ind = net->cc_mod.rtcc.last_inst_ind; 649 inst_bw = bytes_for_this_rtt; 650 /* Can't determine do not change */ 651 probepoint |= ((0xd << 16) | inst_ind); 652 } 653 SDT_PROBE5(sctp, cwnd, net, rttvar, 654 vtag, 655 ((nbw << 32) | inst_bw), 656 ((net->cc_mod.rtcc.lbw_rtt << 32) | rtt), 657 net->flight_size, 658 probepoint); 659 } else { 660 /* No rtt measurement, use last one */ 661 inst_ind = net->cc_mod.rtcc.last_inst_ind; 662 } 663 bw_offset = net->cc_mod.rtcc.lbw >> bw_shift; 664 if (nbw > net->cc_mod.rtcc.lbw + bw_offset) { 665 ret = cc_bw_increase(stcb, net, nbw, vtag); 666 goto out; 667 } 668 rtt_offset = net->cc_mod.rtcc.lbw_rtt >> SCTP_BASE_SYSCTL(sctp_rttvar_rtt); 669 if (nbw < net->cc_mod.rtcc.lbw - bw_offset) { 670 ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, vtag, inst_ind); 671 goto out; 672 } 673 /* 674 * If we reach here then we are in a situation where the bw stayed 675 * the same. 676 */ 677 ret = cc_bw_same(stcb, net, nbw, rtt_offset, vtag, inst_ind); 678 out: 679 net->cc_mod.rtcc.last_inst_ind = inst_ind; 680 return (ret); 681 } 682 683 static void 684 sctp_cwnd_update_after_sack_common(struct sctp_tcb *stcb, 685 struct sctp_association *asoc, 686 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit, int use_rtcc) 687 { 688 struct sctp_nets *net; 689 int old_cwnd; 690 uint32_t t_ssthresh, t_cwnd, incr; 691 uint64_t t_ucwnd_sbw; 692 uint64_t t_path_mptcp; 693 uint64_t mptcp_like_alpha; 694 uint32_t srtt; 695 uint64_t max_path; 696 697 /* MT FIXME: Don't compute this over and over again */ 698 t_ssthresh = 0; 699 t_cwnd = 0; 700 t_ucwnd_sbw = 0; 701 t_path_mptcp = 0; 702 mptcp_like_alpha = 1; 703 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 704 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2) || 705 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_MPTCP)) { 706 max_path = 0; 707 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 708 t_ssthresh += net->ssthresh; 709 t_cwnd += net->cwnd; 710 /* lastsa>>3; we don't need to devide ... */ 711 srtt = net->lastsa; 712 if (srtt > 0) { 713 uint64_t tmp; 714 715 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)srtt; 716 t_path_mptcp += (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_Z) / 717 (((uint64_t)net->mtu) * (uint64_t)srtt); 718 tmp = (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_N) / 719 ((uint64_t)net->mtu * (uint64_t)(srtt * srtt)); 720 if (tmp > max_path) { 721 max_path = tmp; 722 } 723 } 724 } 725 if (t_path_mptcp > 0) { 726 mptcp_like_alpha = max_path / (t_path_mptcp * t_path_mptcp); 727 } else { 728 mptcp_like_alpha = 1; 729 } 730 } 731 if (t_ssthresh == 0) { 732 t_ssthresh = 1; 733 } 734 if (t_ucwnd_sbw == 0) { 735 t_ucwnd_sbw = 1; 736 } 737 /******************************/ 738 /* update cwnd and Early FR */ 739 /******************************/ 740 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 741 742 #ifdef JANA_CMT_FAST_RECOVERY 743 /* 744 * CMT fast recovery code. Need to debug. 745 */ 746 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 747 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 748 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 749 net->will_exit_fast_recovery = 1; 750 } 751 } 752 #endif 753 /* if nothing was acked on this destination skip it */ 754 if (net->net_ack == 0) { 755 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 756 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 757 } 758 continue; 759 } 760 #ifdef JANA_CMT_FAST_RECOVERY 761 /* 762 * CMT fast recovery code 763 */ 764 /* 765 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 766 * && net->will_exit_fast_recovery == 0) { @@@ Do something 767 * } else if (sctp_cmt_on_off == 0 && 768 * asoc->fast_retran_loss_recovery && will_exit == 0) { 769 */ 770 #endif 771 772 if (asoc->fast_retran_loss_recovery && 773 (will_exit == 0) && 774 (asoc->sctp_cmt_on_off == 0)) { 775 /* 776 * If we are in loss recovery we skip any cwnd 777 * update 778 */ 779 return; 780 } 781 /* 782 * Did any measurements go on for this network? 783 */ 784 if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) { 785 uint64_t nbw; 786 787 /* 788 * At this point our bw_bytes has been updated by 789 * incoming sack information. 790 * 791 * But our bw may not yet be set. 792 * 793 */ 794 if ((net->cc_mod.rtcc.new_tot_time / 1000) > 0) { 795 nbw = net->cc_mod.rtcc.bw_bytes / (net->cc_mod.rtcc.new_tot_time / 1000); 796 } else { 797 nbw = net->cc_mod.rtcc.bw_bytes; 798 } 799 if (net->cc_mod.rtcc.lbw) { 800 if (cc_bw_limit(stcb, net, nbw)) { 801 /* Hold here, no update */ 802 continue; 803 } 804 } else { 805 uint64_t vtag, probepoint; 806 807 probepoint = (((uint64_t)net->cwnd) << 32); 808 probepoint |= ((0xa << 16) | 0); 809 vtag = (net->rtt << 32) | 810 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 811 (stcb->rport); 812 813 SDT_PROBE5(sctp, cwnd, net, rttvar, 814 vtag, 815 nbw, 816 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 817 net->flight_size, 818 probepoint); 819 net->cc_mod.rtcc.lbw = nbw; 820 net->cc_mod.rtcc.lbw_rtt = net->rtt; 821 if (net->cc_mod.rtcc.rtt_set_this_sack) { 822 net->cc_mod.rtcc.rtt_set_this_sack = 0; 823 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 824 } 825 } 826 } 827 /* 828 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 829 * moved. 830 */ 831 if (accum_moved || 832 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 833 /* If the cumulative ack moved we can proceed */ 834 if (net->cwnd <= net->ssthresh) { 835 /* We are in slow start */ 836 if (net->flight_size + net->net_ack >= net->cwnd) { 837 uint32_t limit; 838 839 old_cwnd = net->cwnd; 840 switch (asoc->sctp_cmt_on_off) { 841 case SCTP_CMT_RPV1: 842 limit = (uint32_t)(((uint64_t)net->mtu * 843 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 844 (uint64_t)net->ssthresh) / 845 (uint64_t)t_ssthresh); 846 incr = (uint32_t)(((uint64_t)net->net_ack * 847 (uint64_t)net->ssthresh) / 848 (uint64_t)t_ssthresh); 849 if (incr > limit) { 850 incr = limit; 851 } 852 if (incr == 0) { 853 incr = 1; 854 } 855 break; 856 case SCTP_CMT_RPV2: 857 /* 858 * lastsa>>3; we don't need 859 * to divide ... 860 */ 861 srtt = net->lastsa; 862 if (srtt == 0) { 863 srtt = 1; 864 } 865 limit = (uint32_t)(((uint64_t)net->mtu * 866 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 867 (uint64_t)net->cwnd) / 868 ((uint64_t)srtt * t_ucwnd_sbw)); 869 /* INCREASE FACTOR */ 870 incr = (uint32_t)(((uint64_t)net->net_ack * 871 (uint64_t)net->cwnd) / 872 ((uint64_t)srtt * t_ucwnd_sbw)); 873 /* INCREASE FACTOR */ 874 if (incr > limit) { 875 incr = limit; 876 } 877 if (incr == 0) { 878 incr = 1; 879 } 880 break; 881 case SCTP_CMT_MPTCP: 882 limit = (uint32_t)(((uint64_t)net->mtu * 883 mptcp_like_alpha * 884 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) >> 885 SHIFT_MPTCP_MULTI); 886 incr = (uint32_t)(((uint64_t)net->net_ack * 887 mptcp_like_alpha) >> 888 SHIFT_MPTCP_MULTI); 889 if (incr > limit) { 890 incr = limit; 891 } 892 if (incr > net->net_ack) { 893 incr = net->net_ack; 894 } 895 if (incr > net->mtu) { 896 incr = net->mtu; 897 } 898 break; 899 default: 900 incr = net->net_ack; 901 if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) { 902 incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable); 903 } 904 break; 905 } 906 net->cwnd += incr; 907 sctp_enforce_cwnd_limit(asoc, net); 908 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 909 sctp_log_cwnd(stcb, net, incr, 910 SCTP_CWND_LOG_FROM_SS); 911 } 912 SDT_PROBE5(sctp, cwnd, net, ack, 913 stcb->asoc.my_vtag, 914 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 915 net, 916 old_cwnd, net->cwnd); 917 } else { 918 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 919 sctp_log_cwnd(stcb, net, net->net_ack, 920 SCTP_CWND_LOG_NOADV_SS); 921 } 922 } 923 } else { 924 /* We are in congestion avoidance */ 925 /* 926 * Add to pba 927 */ 928 net->partial_bytes_acked += net->net_ack; 929 930 if ((net->flight_size + net->net_ack >= net->cwnd) && 931 (net->partial_bytes_acked >= net->cwnd)) { 932 net->partial_bytes_acked -= net->cwnd; 933 old_cwnd = net->cwnd; 934 switch (asoc->sctp_cmt_on_off) { 935 case SCTP_CMT_RPV1: 936 incr = (uint32_t)(((uint64_t)net->mtu * 937 (uint64_t)net->ssthresh) / 938 (uint64_t)t_ssthresh); 939 if (incr == 0) { 940 incr = 1; 941 } 942 break; 943 case SCTP_CMT_RPV2: 944 /* 945 * lastsa>>3; we don't need 946 * to divide ... 947 */ 948 srtt = net->lastsa; 949 if (srtt == 0) { 950 srtt = 1; 951 } 952 incr = (uint32_t)((uint64_t)net->mtu * 953 (uint64_t)net->cwnd / 954 ((uint64_t)srtt * 955 t_ucwnd_sbw)); 956 /* INCREASE FACTOR */ 957 if (incr == 0) { 958 incr = 1; 959 } 960 break; 961 case SCTP_CMT_MPTCP: 962 incr = (uint32_t)((mptcp_like_alpha * 963 (uint64_t)net->cwnd) >> 964 SHIFT_MPTCP_MULTI); 965 if (incr > net->mtu) { 966 incr = net->mtu; 967 } 968 break; 969 default: 970 incr = net->mtu; 971 break; 972 } 973 net->cwnd += incr; 974 sctp_enforce_cwnd_limit(asoc, net); 975 SDT_PROBE5(sctp, cwnd, net, ack, 976 stcb->asoc.my_vtag, 977 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 978 net, 979 old_cwnd, net->cwnd); 980 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 981 sctp_log_cwnd(stcb, net, net->mtu, 982 SCTP_CWND_LOG_FROM_CA); 983 } 984 } else { 985 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 986 sctp_log_cwnd(stcb, net, net->net_ack, 987 SCTP_CWND_LOG_NOADV_CA); 988 } 989 } 990 } 991 } else { 992 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 993 sctp_log_cwnd(stcb, net, net->mtu, 994 SCTP_CWND_LOG_NO_CUMACK); 995 } 996 } 997 } 998 } 999 1000 static void 1001 sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb, struct sctp_nets *net) 1002 { 1003 int old_cwnd; 1004 1005 old_cwnd = net->cwnd; 1006 net->cwnd = net->mtu; 1007 SDT_PROBE5(sctp, cwnd, net, ack, 1008 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 1009 old_cwnd, net->cwnd); 1010 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1011 (void *)net, net->cwnd); 1012 } 1013 1014 1015 static void 1016 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 1017 { 1018 int old_cwnd = net->cwnd; 1019 uint32_t t_ssthresh, t_cwnd; 1020 uint64_t t_ucwnd_sbw; 1021 1022 /* MT FIXME: Don't compute this over and over again */ 1023 t_ssthresh = 0; 1024 t_cwnd = 0; 1025 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 1026 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) { 1027 struct sctp_nets *lnet; 1028 uint32_t srtt; 1029 1030 t_ucwnd_sbw = 0; 1031 TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) { 1032 t_ssthresh += lnet->ssthresh; 1033 t_cwnd += lnet->cwnd; 1034 srtt = lnet->lastsa; 1035 /* lastsa>>3; we don't need to divide ... */ 1036 if (srtt > 0) { 1037 t_ucwnd_sbw += (uint64_t)lnet->cwnd / (uint64_t)srtt; 1038 } 1039 } 1040 if (t_ssthresh < 1) { 1041 t_ssthresh = 1; 1042 } 1043 if (t_ucwnd_sbw < 1) { 1044 t_ucwnd_sbw = 1; 1045 } 1046 if (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) { 1047 net->ssthresh = (uint32_t)(((uint64_t)4 * 1048 (uint64_t)net->mtu * 1049 (uint64_t)net->ssthresh) / 1050 (uint64_t)t_ssthresh); 1051 } else { 1052 uint64_t cc_delta; 1053 1054 srtt = net->lastsa; 1055 /* lastsa>>3; we don't need to divide ... */ 1056 if (srtt == 0) { 1057 srtt = 1; 1058 } 1059 cc_delta = t_ucwnd_sbw * (uint64_t)srtt / 2; 1060 if (cc_delta < t_cwnd) { 1061 net->ssthresh = (uint32_t)((uint64_t)t_cwnd - cc_delta); 1062 } else { 1063 net->ssthresh = net->mtu; 1064 } 1065 } 1066 if ((net->cwnd > t_cwnd / 2) && 1067 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 1068 net->ssthresh = net->cwnd - t_cwnd / 2; 1069 } 1070 if (net->ssthresh < net->mtu) { 1071 net->ssthresh = net->mtu; 1072 } 1073 } else { 1074 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 1075 } 1076 net->cwnd = net->mtu; 1077 net->partial_bytes_acked = 0; 1078 SDT_PROBE5(sctp, cwnd, net, to, 1079 stcb->asoc.my_vtag, 1080 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1081 net, 1082 old_cwnd, net->cwnd); 1083 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1084 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1085 } 1086 } 1087 1088 static void 1089 sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net, 1090 int in_window, int num_pkt_lost, int use_rtcc) 1091 { 1092 int old_cwnd = net->cwnd; 1093 1094 if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) { 1095 /* Data center Congestion Control */ 1096 if (in_window == 0) { 1097 /* 1098 * Go to CA with the cwnd at the point we sent the 1099 * TSN that was marked with a CE. 1100 */ 1101 if (net->ecn_prev_cwnd < net->cwnd) { 1102 /* Restore to prev cwnd */ 1103 net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost); 1104 } else { 1105 /* Just cut in 1/2 */ 1106 net->cwnd /= 2; 1107 } 1108 /* Drop to CA */ 1109 net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu); 1110 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1111 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1112 } 1113 } else { 1114 /* 1115 * Further tuning down required over the drastic 1116 * original cut 1117 */ 1118 net->ssthresh -= (net->mtu * num_pkt_lost); 1119 net->cwnd -= (net->mtu * num_pkt_lost); 1120 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1121 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1122 } 1123 1124 } 1125 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1126 } else { 1127 if (in_window == 0) { 1128 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1129 net->ssthresh = net->cwnd / 2; 1130 if (net->ssthresh < net->mtu) { 1131 net->ssthresh = net->mtu; 1132 /* 1133 * here back off the timer as well, to slow 1134 * us down 1135 */ 1136 net->RTO <<= 1; 1137 } 1138 net->cwnd = net->ssthresh; 1139 SDT_PROBE5(sctp, cwnd, net, ecn, 1140 stcb->asoc.my_vtag, 1141 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1142 net, 1143 old_cwnd, net->cwnd); 1144 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1145 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1146 } 1147 } 1148 } 1149 1150 } 1151 1152 static void 1153 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 1154 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 1155 uint32_t *bottle_bw, uint32_t *on_queue) 1156 { 1157 uint32_t bw_avail; 1158 unsigned int incr; 1159 int old_cwnd = net->cwnd; 1160 1161 /* get bottle neck bw */ 1162 *bottle_bw = ntohl(cp->bottle_bw); 1163 /* and whats on queue */ 1164 *on_queue = ntohl(cp->current_onq); 1165 /* 1166 * adjust the on-queue if our flight is more it could be that the 1167 * router has not yet gotten data "in-flight" to it 1168 */ 1169 if (*on_queue < net->flight_size) { 1170 *on_queue = net->flight_size; 1171 } 1172 /* rtt is measured in micro seconds, bottle_bw in bytes per second */ 1173 bw_avail = (uint32_t)(((uint64_t)(*bottle_bw) * net->rtt) / (uint64_t)1000000); 1174 if (bw_avail > *bottle_bw) { 1175 /* 1176 * Cap the growth to no more than the bottle neck. This can 1177 * happen as RTT slides up due to queues. It also means if 1178 * you have more than a 1 second RTT with a empty queue you 1179 * will be limited to the bottle_bw per second no matter if 1180 * other points have 1/2 the RTT and you could get more 1181 * out... 1182 */ 1183 bw_avail = *bottle_bw; 1184 } 1185 if (*on_queue > bw_avail) { 1186 /* 1187 * No room for anything else don't allow anything else to be 1188 * "added to the fire". 1189 */ 1190 int seg_inflight, seg_onqueue, my_portion; 1191 1192 net->partial_bytes_acked = 0; 1193 /* how much are we over queue size? */ 1194 incr = *on_queue - bw_avail; 1195 if (stcb->asoc.seen_a_sack_this_pkt) { 1196 /* 1197 * undo any cwnd adjustment that the sack might have 1198 * made 1199 */ 1200 net->cwnd = net->prev_cwnd; 1201 } 1202 /* Now how much of that is mine? */ 1203 seg_inflight = net->flight_size / net->mtu; 1204 seg_onqueue = *on_queue / net->mtu; 1205 my_portion = (incr * seg_inflight) / seg_onqueue; 1206 1207 /* Have I made an adjustment already */ 1208 if (net->cwnd > net->flight_size) { 1209 /* 1210 * for this flight I made an adjustment we need to 1211 * decrease the portion by a share our previous 1212 * adjustment. 1213 */ 1214 int diff_adj; 1215 1216 diff_adj = net->cwnd - net->flight_size; 1217 if (diff_adj > my_portion) 1218 my_portion = 0; 1219 else 1220 my_portion -= diff_adj; 1221 } 1222 /* 1223 * back down to the previous cwnd (assume we have had a sack 1224 * before this packet). minus what ever portion of the 1225 * overage is my fault. 1226 */ 1227 net->cwnd -= my_portion; 1228 1229 /* we will NOT back down more than 1 MTU */ 1230 if (net->cwnd <= net->mtu) { 1231 net->cwnd = net->mtu; 1232 } 1233 /* force into CA */ 1234 net->ssthresh = net->cwnd - 1; 1235 } else { 1236 /* 1237 * Take 1/4 of the space left or max burst up .. whichever 1238 * is less. 1239 */ 1240 incr = (bw_avail - *on_queue) >> 2; 1241 if ((stcb->asoc.max_burst > 0) && 1242 (stcb->asoc.max_burst * net->mtu < incr)) { 1243 incr = stcb->asoc.max_burst * net->mtu; 1244 } 1245 net->cwnd += incr; 1246 } 1247 if (net->cwnd > bw_avail) { 1248 /* We can't exceed the pipe size */ 1249 net->cwnd = bw_avail; 1250 } 1251 if (net->cwnd < net->mtu) { 1252 /* We always have 1 MTU */ 1253 net->cwnd = net->mtu; 1254 } 1255 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1256 if (net->cwnd - old_cwnd != 0) { 1257 /* log only changes */ 1258 SDT_PROBE5(sctp, cwnd, net, pd, 1259 stcb->asoc.my_vtag, 1260 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1261 net, 1262 old_cwnd, net->cwnd); 1263 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1264 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1265 SCTP_CWND_LOG_FROM_SAT); 1266 } 1267 } 1268 } 1269 1270 static void 1271 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 1272 struct sctp_nets *net, int burst_limit) 1273 { 1274 int old_cwnd = net->cwnd; 1275 1276 if (net->ssthresh < net->cwnd) 1277 net->ssthresh = net->cwnd; 1278 if (burst_limit) { 1279 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 1280 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1281 SDT_PROBE5(sctp, cwnd, net, bl, 1282 stcb->asoc.my_vtag, 1283 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1284 net, 1285 old_cwnd, net->cwnd); 1286 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1287 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 1288 } 1289 } 1290 } 1291 1292 static void 1293 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1294 struct sctp_association *asoc, 1295 int accum_moved, int reneged_all, int will_exit) 1296 { 1297 /* Passing a zero argument in last disables the rtcc algorithm */ 1298 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0); 1299 } 1300 1301 static void 1302 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1303 int in_window, int num_pkt_lost) 1304 { 1305 /* Passing a zero argument in last disables the rtcc algorithm */ 1306 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0); 1307 } 1308 1309 /* Here starts the RTCCVAR type CC invented by RRS which 1310 * is a slight mod to RFC2581. We reuse a common routine or 1311 * two since these algorithms are so close and need to 1312 * remain the same. 1313 */ 1314 static void 1315 sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1316 int in_window, int num_pkt_lost) 1317 { 1318 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1); 1319 } 1320 1321 1322 static 1323 void 1324 sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net, 1325 struct sctp_tmit_chunk *tp1) 1326 { 1327 net->cc_mod.rtcc.bw_bytes += tp1->send_size; 1328 } 1329 1330 static void 1331 sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb SCTP_UNUSED, 1332 struct sctp_nets *net) 1333 { 1334 if (net->cc_mod.rtcc.tls_needs_set > 0) { 1335 /* We had a bw measurment going on */ 1336 struct timeval ltls; 1337 1338 SCTP_GETPTIME_TIMEVAL(<ls); 1339 timevalsub(<ls, &net->cc_mod.rtcc.tls); 1340 net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec; 1341 } 1342 } 1343 1344 static void 1345 sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb, 1346 struct sctp_nets *net) 1347 { 1348 uint64_t vtag, probepoint; 1349 1350 if (net->cc_mod.rtcc.lbw) { 1351 /* Clear the old bw.. we went to 0 in-flight */ 1352 vtag = (net->rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1353 (stcb->rport); 1354 probepoint = (((uint64_t)net->cwnd) << 32); 1355 /* Probe point 8 */ 1356 probepoint |= ((8 << 16) | 0); 1357 SDT_PROBE5(sctp, cwnd, net, rttvar, 1358 vtag, 1359 ((net->cc_mod.rtcc.lbw << 32) | 0), 1360 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 1361 net->flight_size, 1362 probepoint); 1363 net->cc_mod.rtcc.lbw_rtt = 0; 1364 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1365 net->cc_mod.rtcc.lbw = 0; 1366 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1367 net->cc_mod.rtcc.vol_reduce = 0; 1368 net->cc_mod.rtcc.bw_tot_time = 0; 1369 net->cc_mod.rtcc.bw_bytes = 0; 1370 net->cc_mod.rtcc.tls_needs_set = 0; 1371 if (net->cc_mod.rtcc.steady_step) { 1372 net->cc_mod.rtcc.vol_reduce = 0; 1373 net->cc_mod.rtcc.step_cnt = 0; 1374 net->cc_mod.rtcc.last_step_state = 0; 1375 } 1376 if (net->cc_mod.rtcc.ret_from_eq) { 1377 /* less aggressive one - reset cwnd too */ 1378 uint32_t cwnd_in_mtu, cwnd; 1379 1380 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 1381 if (cwnd_in_mtu == 0) { 1382 /* 1383 * Using 0 means that the value of RFC 4960 1384 * is used. 1385 */ 1386 cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1387 } else { 1388 /* 1389 * We take the minimum of the burst limit 1390 * and the initial congestion window. 1391 */ 1392 if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst)) 1393 cwnd_in_mtu = stcb->asoc.max_burst; 1394 cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 1395 } 1396 if (net->cwnd > cwnd) { 1397 /* 1398 * Only set if we are not a timeout (i.e. 1399 * down to 1 mtu) 1400 */ 1401 net->cwnd = cwnd; 1402 } 1403 } 1404 } 1405 } 1406 1407 static void 1408 sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb, 1409 struct sctp_nets *net) 1410 { 1411 uint64_t vtag, probepoint; 1412 1413 sctp_set_initial_cc_param(stcb, net); 1414 stcb->asoc.use_precise_time = 1; 1415 probepoint = (((uint64_t)net->cwnd) << 32); 1416 probepoint |= ((9 << 16) | 0); 1417 vtag = (net->rtt << 32) | 1418 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1419 (stcb->rport); 1420 SDT_PROBE5(sctp, cwnd, net, rttvar, 1421 vtag, 1422 0, 1423 0, 1424 0, 1425 probepoint); 1426 net->cc_mod.rtcc.lbw_rtt = 0; 1427 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1428 net->cc_mod.rtcc.vol_reduce = 0; 1429 net->cc_mod.rtcc.lbw = 0; 1430 net->cc_mod.rtcc.vol_reduce = 0; 1431 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1432 net->cc_mod.rtcc.bw_tot_time = 0; 1433 net->cc_mod.rtcc.bw_bytes = 0; 1434 net->cc_mod.rtcc.tls_needs_set = 0; 1435 net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret); 1436 net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step); 1437 net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn); 1438 net->cc_mod.rtcc.step_cnt = 0; 1439 net->cc_mod.rtcc.last_step_state = 0; 1440 1441 1442 } 1443 1444 static int 1445 sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget, 1446 struct sctp_cc_option *cc_opt) 1447 { 1448 struct sctp_nets *net; 1449 1450 if (setorget == 1) { 1451 /* a set */ 1452 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1453 if ((cc_opt->aid_value.assoc_value != 0) && 1454 (cc_opt->aid_value.assoc_value != 1)) { 1455 return (EINVAL); 1456 } 1457 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1458 net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value; 1459 } 1460 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1461 if ((cc_opt->aid_value.assoc_value != 0) && 1462 (cc_opt->aid_value.assoc_value != 1)) { 1463 return (EINVAL); 1464 } 1465 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1466 net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value; 1467 } 1468 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1469 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1470 net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value; 1471 } 1472 } else { 1473 return (EINVAL); 1474 } 1475 } else { 1476 /* a get */ 1477 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1478 net = TAILQ_FIRST(&stcb->asoc.nets); 1479 if (net == NULL) { 1480 return (EFAULT); 1481 } 1482 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.ret_from_eq; 1483 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1484 net = TAILQ_FIRST(&stcb->asoc.nets); 1485 if (net == NULL) { 1486 return (EFAULT); 1487 } 1488 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.use_dccc_ecn; 1489 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1490 net = TAILQ_FIRST(&stcb->asoc.nets); 1491 if (net == NULL) { 1492 return (EFAULT); 1493 } 1494 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step; 1495 } else { 1496 return (EINVAL); 1497 } 1498 } 1499 return (0); 1500 } 1501 1502 static void 1503 sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb SCTP_UNUSED, 1504 struct sctp_nets *net) 1505 { 1506 if (net->cc_mod.rtcc.tls_needs_set == 0) { 1507 SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls); 1508 net->cc_mod.rtcc.tls_needs_set = 2; 1509 } 1510 } 1511 1512 static void 1513 sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb, 1514 struct sctp_association *asoc, 1515 int accum_moved, int reneged_all, int will_exit) 1516 { 1517 /* Passing a one argument at the last enables the rtcc algorithm */ 1518 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1); 1519 } 1520 1521 static void 1522 sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb SCTP_UNUSED, 1523 struct sctp_nets *net, 1524 struct timeval *now SCTP_UNUSED) 1525 { 1526 net->cc_mod.rtcc.rtt_set_this_sack = 1; 1527 } 1528 1529 /* Here starts Sally Floyds HS-TCP */ 1530 1531 struct sctp_hs_raise_drop { 1532 int32_t cwnd; 1533 int8_t increase; 1534 int8_t drop_percent; 1535 }; 1536 1537 #define SCTP_HS_TABLE_SIZE 73 1538 1539 static const struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 1540 {38, 1, 50}, /* 0 */ 1541 {118, 2, 44}, /* 1 */ 1542 {221, 3, 41}, /* 2 */ 1543 {347, 4, 38}, /* 3 */ 1544 {495, 5, 37}, /* 4 */ 1545 {663, 6, 35}, /* 5 */ 1546 {851, 7, 34}, /* 6 */ 1547 {1058, 8, 33}, /* 7 */ 1548 {1284, 9, 32}, /* 8 */ 1549 {1529, 10, 31}, /* 9 */ 1550 {1793, 11, 30}, /* 10 */ 1551 {2076, 12, 29}, /* 11 */ 1552 {2378, 13, 28}, /* 12 */ 1553 {2699, 14, 28}, /* 13 */ 1554 {3039, 15, 27}, /* 14 */ 1555 {3399, 16, 27}, /* 15 */ 1556 {3778, 17, 26}, /* 16 */ 1557 {4177, 18, 26}, /* 17 */ 1558 {4596, 19, 25}, /* 18 */ 1559 {5036, 20, 25}, /* 19 */ 1560 {5497, 21, 24}, /* 20 */ 1561 {5979, 22, 24}, /* 21 */ 1562 {6483, 23, 23}, /* 22 */ 1563 {7009, 24, 23}, /* 23 */ 1564 {7558, 25, 22}, /* 24 */ 1565 {8130, 26, 22}, /* 25 */ 1566 {8726, 27, 22}, /* 26 */ 1567 {9346, 28, 21}, /* 27 */ 1568 {9991, 29, 21}, /* 28 */ 1569 {10661, 30, 21}, /* 29 */ 1570 {11358, 31, 20}, /* 30 */ 1571 {12082, 32, 20}, /* 31 */ 1572 {12834, 33, 20}, /* 32 */ 1573 {13614, 34, 19}, /* 33 */ 1574 {14424, 35, 19}, /* 34 */ 1575 {15265, 36, 19}, /* 35 */ 1576 {16137, 37, 19}, /* 36 */ 1577 {17042, 38, 18}, /* 37 */ 1578 {17981, 39, 18}, /* 38 */ 1579 {18955, 40, 18}, /* 39 */ 1580 {19965, 41, 17}, /* 40 */ 1581 {21013, 42, 17}, /* 41 */ 1582 {22101, 43, 17}, /* 42 */ 1583 {23230, 44, 17}, /* 43 */ 1584 {24402, 45, 16}, /* 44 */ 1585 {25618, 46, 16}, /* 45 */ 1586 {26881, 47, 16}, /* 46 */ 1587 {28193, 48, 16}, /* 47 */ 1588 {29557, 49, 15}, /* 48 */ 1589 {30975, 50, 15}, /* 49 */ 1590 {32450, 51, 15}, /* 50 */ 1591 {33986, 52, 15}, /* 51 */ 1592 {35586, 53, 14}, /* 52 */ 1593 {37253, 54, 14}, /* 53 */ 1594 {38992, 55, 14}, /* 54 */ 1595 {40808, 56, 14}, /* 55 */ 1596 {42707, 57, 13}, /* 56 */ 1597 {44694, 58, 13}, /* 57 */ 1598 {46776, 59, 13}, /* 58 */ 1599 {48961, 60, 13}, /* 59 */ 1600 {51258, 61, 13}, /* 60 */ 1601 {53677, 62, 12}, /* 61 */ 1602 {56230, 63, 12}, /* 62 */ 1603 {58932, 64, 12}, /* 63 */ 1604 {61799, 65, 12}, /* 64 */ 1605 {64851, 66, 11}, /* 65 */ 1606 {68113, 67, 11}, /* 66 */ 1607 {71617, 68, 11}, /* 67 */ 1608 {75401, 69, 10}, /* 68 */ 1609 {79517, 70, 10}, /* 69 */ 1610 {84035, 71, 10}, /* 70 */ 1611 {89053, 72, 10}, /* 71 */ 1612 {94717, 73, 9} /* 72 */ 1613 }; 1614 1615 static void 1616 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 1617 { 1618 int cur_val, i, indx, incr; 1619 int old_cwnd = net->cwnd; 1620 1621 cur_val = net->cwnd >> 10; 1622 indx = SCTP_HS_TABLE_SIZE - 1; 1623 1624 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1625 /* normal mode */ 1626 if (net->net_ack > net->mtu) { 1627 net->cwnd += net->mtu; 1628 } else { 1629 net->cwnd += net->net_ack; 1630 } 1631 } else { 1632 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 1633 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 1634 indx = i; 1635 break; 1636 } 1637 } 1638 net->last_hs_used = indx; 1639 incr = (((int32_t)sctp_cwnd_adjust[indx].increase) << 10); 1640 net->cwnd += incr; 1641 } 1642 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1643 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1644 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SS); 1645 } 1646 } 1647 1648 static void 1649 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 1650 { 1651 int cur_val, i, indx; 1652 int old_cwnd = net->cwnd; 1653 1654 cur_val = net->cwnd >> 10; 1655 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1656 /* normal mode */ 1657 net->ssthresh = net->cwnd / 2; 1658 if (net->ssthresh < (net->mtu * 2)) { 1659 net->ssthresh = 2 * net->mtu; 1660 } 1661 net->cwnd = net->ssthresh; 1662 } else { 1663 /* drop by the proper amount */ 1664 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 1665 (int32_t)sctp_cwnd_adjust[net->last_hs_used].drop_percent); 1666 net->cwnd = net->ssthresh; 1667 /* now where are we */ 1668 indx = net->last_hs_used; 1669 cur_val = net->cwnd >> 10; 1670 /* reset where we are in the table */ 1671 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1672 /* feel out of hs */ 1673 net->last_hs_used = 0; 1674 } else { 1675 for (i = indx; i >= 1; i--) { 1676 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 1677 break; 1678 } 1679 } 1680 net->last_hs_used = indx; 1681 } 1682 } 1683 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1684 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1685 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 1686 } 1687 } 1688 1689 static void 1690 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 1691 struct sctp_association *asoc) 1692 { 1693 struct sctp_nets *net; 1694 1695 /* 1696 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 1697 * (net->fast_retran_loss_recovery == 0))) 1698 */ 1699 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1700 if ((asoc->fast_retran_loss_recovery == 0) || 1701 (asoc->sctp_cmt_on_off > 0)) { 1702 /* out of a RFC2582 Fast recovery window? */ 1703 if (net->net_ack > 0) { 1704 /* 1705 * per section 7.2.3, are there any 1706 * destinations that had a fast retransmit 1707 * to them. If so what we need to do is 1708 * adjust ssthresh and cwnd. 1709 */ 1710 struct sctp_tmit_chunk *lchk; 1711 1712 sctp_hs_cwnd_decrease(stcb, net); 1713 1714 lchk = TAILQ_FIRST(&asoc->send_queue); 1715 1716 net->partial_bytes_acked = 0; 1717 /* Turn on fast recovery window */ 1718 asoc->fast_retran_loss_recovery = 1; 1719 if (lchk == NULL) { 1720 /* Mark end of the window */ 1721 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1722 } else { 1723 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1724 } 1725 1726 /* 1727 * CMT fast recovery -- per destination 1728 * recovery variable. 1729 */ 1730 net->fast_retran_loss_recovery = 1; 1731 1732 if (lchk == NULL) { 1733 /* Mark end of the window */ 1734 net->fast_recovery_tsn = asoc->sending_seq - 1; 1735 } else { 1736 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1737 } 1738 1739 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1740 stcb->sctp_ep, stcb, net, 1741 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_2); 1742 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1743 stcb->sctp_ep, stcb, net); 1744 } 1745 } else if (net->net_ack > 0) { 1746 /* 1747 * Mark a peg that we WOULD have done a cwnd 1748 * reduction but RFC2582 prevented this action. 1749 */ 1750 SCTP_STAT_INCR(sctps_fastretransinrtt); 1751 } 1752 } 1753 } 1754 1755 static void 1756 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 1757 struct sctp_association *asoc, 1758 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 1759 { 1760 struct sctp_nets *net; 1761 1762 /******************************/ 1763 /* update cwnd and Early FR */ 1764 /******************************/ 1765 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1766 1767 #ifdef JANA_CMT_FAST_RECOVERY 1768 /* 1769 * CMT fast recovery code. Need to debug. 1770 */ 1771 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1772 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 1773 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 1774 net->will_exit_fast_recovery = 1; 1775 } 1776 } 1777 #endif 1778 /* if nothing was acked on this destination skip it */ 1779 if (net->net_ack == 0) { 1780 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1781 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1782 } 1783 continue; 1784 } 1785 #ifdef JANA_CMT_FAST_RECOVERY 1786 /* 1787 * CMT fast recovery code 1788 */ 1789 /* 1790 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 1791 * && net->will_exit_fast_recovery == 0) { @@@ Do something 1792 * } else if (sctp_cmt_on_off == 0 && 1793 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1794 */ 1795 #endif 1796 1797 if (asoc->fast_retran_loss_recovery && 1798 (will_exit == 0) && 1799 (asoc->sctp_cmt_on_off == 0)) { 1800 /* 1801 * If we are in loss recovery we skip any cwnd 1802 * update 1803 */ 1804 return; 1805 } 1806 /* 1807 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1808 * moved. 1809 */ 1810 if (accum_moved || 1811 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 1812 /* If the cumulative ack moved we can proceed */ 1813 if (net->cwnd <= net->ssthresh) { 1814 /* We are in slow start */ 1815 if (net->flight_size + net->net_ack >= net->cwnd) { 1816 sctp_hs_cwnd_increase(stcb, net); 1817 } else { 1818 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1819 sctp_log_cwnd(stcb, net, net->net_ack, 1820 SCTP_CWND_LOG_NOADV_SS); 1821 } 1822 } 1823 } else { 1824 /* We are in congestion avoidance */ 1825 net->partial_bytes_acked += net->net_ack; 1826 if ((net->flight_size + net->net_ack >= net->cwnd) && 1827 (net->partial_bytes_acked >= net->cwnd)) { 1828 net->partial_bytes_acked -= net->cwnd; 1829 net->cwnd += net->mtu; 1830 sctp_enforce_cwnd_limit(asoc, net); 1831 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1832 sctp_log_cwnd(stcb, net, net->mtu, 1833 SCTP_CWND_LOG_FROM_CA); 1834 } 1835 } else { 1836 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1837 sctp_log_cwnd(stcb, net, net->net_ack, 1838 SCTP_CWND_LOG_NOADV_CA); 1839 } 1840 } 1841 } 1842 } else { 1843 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1844 sctp_log_cwnd(stcb, net, net->mtu, 1845 SCTP_CWND_LOG_NO_CUMACK); 1846 } 1847 } 1848 } 1849 } 1850 1851 1852 /* 1853 * H-TCP congestion control. The algorithm is detailed in: 1854 * R.N.Shorten, D.J.Leith: 1855 * "H-TCP: TCP for high-speed and long-distance networks" 1856 * Proc. PFLDnet, Argonne, 2004. 1857 * http://www.hamilton.ie/net/htcp3.pdf 1858 */ 1859 1860 1861 static int use_rtt_scaling = 1; 1862 static int use_bandwidth_switch = 1; 1863 1864 static inline int 1865 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1866 { 1867 return (seq3 - seq2 >= seq1 - seq2); 1868 } 1869 1870 static inline uint32_t 1871 htcp_cong_time(struct htcp *ca) 1872 { 1873 return (sctp_get_tick_count() - ca->last_cong); 1874 } 1875 1876 static inline uint32_t 1877 htcp_ccount(struct htcp *ca) 1878 { 1879 return (htcp_cong_time(ca) / ca->minRTT); 1880 } 1881 1882 static inline void 1883 htcp_reset(struct htcp *ca) 1884 { 1885 ca->undo_last_cong = ca->last_cong; 1886 ca->undo_maxRTT = ca->maxRTT; 1887 ca->undo_old_maxB = ca->old_maxB; 1888 ca->last_cong = sctp_get_tick_count(); 1889 } 1890 1891 #ifdef SCTP_NOT_USED 1892 1893 static uint32_t 1894 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1895 { 1896 net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong; 1897 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT; 1898 net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB; 1899 return (max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->cc_mod.htcp_ca.beta) * net->mtu)); 1900 } 1901 1902 #endif 1903 1904 static inline void 1905 measure_rtt(struct sctp_nets *net) 1906 { 1907 uint32_t srtt = net->lastsa >> SCTP_RTT_SHIFT; 1908 1909 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1910 if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT) 1911 net->cc_mod.htcp_ca.minRTT = srtt; 1912 1913 /* max RTT */ 1914 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) { 1915 if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT) 1916 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT; 1917 if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT + MSEC_TO_TICKS(20)) 1918 net->cc_mod.htcp_ca.maxRTT = srtt; 1919 } 1920 } 1921 1922 static void 1923 measure_achieved_throughput(struct sctp_nets *net) 1924 { 1925 uint32_t now = sctp_get_tick_count(); 1926 1927 if (net->fast_retran_ip == 0) 1928 net->cc_mod.htcp_ca.bytes_acked = net->net_ack; 1929 1930 if (!use_bandwidth_switch) 1931 return; 1932 1933 /* achieved throughput calculations */ 1934 /* JRS - not 100% sure of this statement */ 1935 if (net->fast_retran_ip == 1) { 1936 net->cc_mod.htcp_ca.bytecount = 0; 1937 net->cc_mod.htcp_ca.lasttime = now; 1938 return; 1939 } 1940 1941 net->cc_mod.htcp_ca.bytecount += net->net_ack; 1942 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)) && 1943 (now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT) && 1944 (net->cc_mod.htcp_ca.minRTT > 0)) { 1945 uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount / net->mtu * hz / (now - net->cc_mod.htcp_ca.lasttime); 1946 1947 if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) { 1948 /* just after backoff */ 1949 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi; 1950 } else { 1951 net->cc_mod.htcp_ca.Bi = (3 * net->cc_mod.htcp_ca.Bi + cur_Bi) / 4; 1952 if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB) 1953 net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi; 1954 if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB) 1955 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB; 1956 } 1957 net->cc_mod.htcp_ca.bytecount = 0; 1958 net->cc_mod.htcp_ca.lasttime = now; 1959 } 1960 } 1961 1962 static inline void 1963 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1964 { 1965 if (use_bandwidth_switch) { 1966 uint32_t maxB = ca->maxB; 1967 uint32_t old_maxB = ca->old_maxB; 1968 1969 ca->old_maxB = ca->maxB; 1970 1971 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1972 ca->beta = BETA_MIN; 1973 ca->modeswitch = 0; 1974 return; 1975 } 1976 } 1977 1978 if (ca->modeswitch && minRTT > (uint32_t)MSEC_TO_TICKS(10) && maxRTT) { 1979 ca->beta = (minRTT << 7) / maxRTT; 1980 if (ca->beta < BETA_MIN) 1981 ca->beta = BETA_MIN; 1982 else if (ca->beta > BETA_MAX) 1983 ca->beta = BETA_MAX; 1984 } else { 1985 ca->beta = BETA_MIN; 1986 ca->modeswitch = 1; 1987 } 1988 } 1989 1990 static inline void 1991 htcp_alpha_update(struct htcp *ca) 1992 { 1993 uint32_t minRTT = ca->minRTT; 1994 uint32_t factor = 1; 1995 uint32_t diff = htcp_cong_time(ca); 1996 1997 if (diff > (uint32_t)hz) { 1998 diff -= hz; 1999 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 2000 } 2001 2002 if (use_rtt_scaling && minRTT) { 2003 uint32_t scale = (hz << 3) / (10 * minRTT); 2004 2005 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 2006 * interval [0.5,10]<<3 */ 2007 factor = (factor << 3) / scale; 2008 if (!factor) 2009 factor = 1; 2010 } 2011 2012 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 2013 if (!ca->alpha) 2014 ca->alpha = ALPHA_BASE; 2015 } 2016 2017 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 2018 * rtt before we adjust our beta to ensure we are working from a consistent 2019 * data. 2020 * 2021 * This function should be called when we hit a congestion event since only at 2022 * that point do we really have a real sense of maxRTT (the queues en route 2023 * were getting just too full now). 2024 */ 2025 static void 2026 htcp_param_update(struct sctp_nets *net) 2027 { 2028 uint32_t minRTT = net->cc_mod.htcp_ca.minRTT; 2029 uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT; 2030 2031 htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT); 2032 htcp_alpha_update(&net->cc_mod.htcp_ca); 2033 2034 /* 2035 * add slowly fading memory for maxRTT to accommodate routing 2036 * changes etc 2037 */ 2038 if (minRTT > 0 && maxRTT > minRTT) 2039 net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 2040 } 2041 2042 static uint32_t 2043 htcp_recalc_ssthresh(struct sctp_nets *net) 2044 { 2045 htcp_param_update(net); 2046 return (max(((net->cwnd / net->mtu * net->cc_mod.htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu)); 2047 } 2048 2049 static void 2050 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 2051 { 2052 /*- 2053 * How to handle these functions? 2054 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 2055 * return; 2056 */ 2057 if (net->cwnd <= net->ssthresh) { 2058 /* We are in slow start */ 2059 if (net->flight_size + net->net_ack >= net->cwnd) { 2060 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 2061 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 2062 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2063 sctp_log_cwnd(stcb, net, net->mtu, 2064 SCTP_CWND_LOG_FROM_SS); 2065 } 2066 2067 } else { 2068 net->cwnd += net->net_ack; 2069 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2070 sctp_log_cwnd(stcb, net, net->net_ack, 2071 SCTP_CWND_LOG_FROM_SS); 2072 } 2073 2074 } 2075 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2076 } else { 2077 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2078 sctp_log_cwnd(stcb, net, net->net_ack, 2079 SCTP_CWND_LOG_NOADV_SS); 2080 } 2081 } 2082 } else { 2083 measure_rtt(net); 2084 2085 /* 2086 * In dangerous area, increase slowly. In theory this is 2087 * net->cwnd += alpha / net->cwnd 2088 */ 2089 /* What is snd_cwnd_cnt?? */ 2090 if (((net->partial_bytes_acked / net->mtu * net->cc_mod.htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 2091 /*- 2092 * Does SCTP have a cwnd clamp? 2093 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 2094 */ 2095 net->cwnd += net->mtu; 2096 net->partial_bytes_acked = 0; 2097 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2098 htcp_alpha_update(&net->cc_mod.htcp_ca); 2099 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2100 sctp_log_cwnd(stcb, net, net->mtu, 2101 SCTP_CWND_LOG_FROM_CA); 2102 } 2103 } else { 2104 net->partial_bytes_acked += net->net_ack; 2105 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2106 sctp_log_cwnd(stcb, net, net->net_ack, 2107 SCTP_CWND_LOG_NOADV_CA); 2108 } 2109 } 2110 2111 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2112 } 2113 } 2114 2115 #ifdef SCTP_NOT_USED 2116 /* Lower bound on congestion window. */ 2117 static uint32_t 2118 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 2119 { 2120 return (net->ssthresh); 2121 } 2122 #endif 2123 2124 static void 2125 htcp_init(struct sctp_nets *net) 2126 { 2127 memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp)); 2128 net->cc_mod.htcp_ca.alpha = ALPHA_BASE; 2129 net->cc_mod.htcp_ca.beta = BETA_MIN; 2130 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2131 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count(); 2132 } 2133 2134 static void 2135 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 2136 { 2137 /* 2138 * We take the max of the burst limit times a MTU or the 2139 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 2140 */ 2141 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 2142 net->ssthresh = stcb->asoc.peers_rwnd; 2143 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2144 htcp_init(net); 2145 2146 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 2147 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 2148 } 2149 } 2150 2151 static void 2152 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 2153 struct sctp_association *asoc, 2154 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 2155 { 2156 struct sctp_nets *net; 2157 2158 /******************************/ 2159 /* update cwnd and Early FR */ 2160 /******************************/ 2161 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2162 2163 #ifdef JANA_CMT_FAST_RECOVERY 2164 /* 2165 * CMT fast recovery code. Need to debug. 2166 */ 2167 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 2168 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 2169 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 2170 net->will_exit_fast_recovery = 1; 2171 } 2172 } 2173 #endif 2174 /* if nothing was acked on this destination skip it */ 2175 if (net->net_ack == 0) { 2176 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2177 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 2178 } 2179 continue; 2180 } 2181 #ifdef JANA_CMT_FAST_RECOVERY 2182 /* 2183 * CMT fast recovery code 2184 */ 2185 /* 2186 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 2187 * && net->will_exit_fast_recovery == 0) { @@@ Do something 2188 * } else if (sctp_cmt_on_off == 0 && 2189 * asoc->fast_retran_loss_recovery && will_exit == 0) { 2190 */ 2191 #endif 2192 2193 if (asoc->fast_retran_loss_recovery && 2194 will_exit == 0 && 2195 (asoc->sctp_cmt_on_off == 0)) { 2196 /* 2197 * If we are in loss recovery we skip any cwnd 2198 * update 2199 */ 2200 return; 2201 } 2202 /* 2203 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 2204 * moved. 2205 */ 2206 if (accum_moved || 2207 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 2208 htcp_cong_avoid(stcb, net); 2209 measure_achieved_throughput(net); 2210 } else { 2211 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2212 sctp_log_cwnd(stcb, net, net->mtu, 2213 SCTP_CWND_LOG_NO_CUMACK); 2214 } 2215 } 2216 } 2217 } 2218 2219 static void 2220 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 2221 struct sctp_association *asoc) 2222 { 2223 struct sctp_nets *net; 2224 2225 /* 2226 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 2227 * (net->fast_retran_loss_recovery == 0))) 2228 */ 2229 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2230 if ((asoc->fast_retran_loss_recovery == 0) || 2231 (asoc->sctp_cmt_on_off > 0)) { 2232 /* out of a RFC2582 Fast recovery window? */ 2233 if (net->net_ack > 0) { 2234 /* 2235 * per section 7.2.3, are there any 2236 * destinations that had a fast retransmit 2237 * to them. If so what we need to do is 2238 * adjust ssthresh and cwnd. 2239 */ 2240 struct sctp_tmit_chunk *lchk; 2241 int old_cwnd = net->cwnd; 2242 2243 /* JRS - reset as if state were changed */ 2244 htcp_reset(&net->cc_mod.htcp_ca); 2245 net->ssthresh = htcp_recalc_ssthresh(net); 2246 net->cwnd = net->ssthresh; 2247 sctp_enforce_cwnd_limit(asoc, net); 2248 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2249 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 2250 SCTP_CWND_LOG_FROM_FR); 2251 } 2252 lchk = TAILQ_FIRST(&asoc->send_queue); 2253 2254 net->partial_bytes_acked = 0; 2255 /* Turn on fast recovery window */ 2256 asoc->fast_retran_loss_recovery = 1; 2257 if (lchk == NULL) { 2258 /* Mark end of the window */ 2259 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 2260 } else { 2261 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2262 } 2263 2264 /* 2265 * CMT fast recovery -- per destination 2266 * recovery variable. 2267 */ 2268 net->fast_retran_loss_recovery = 1; 2269 2270 if (lchk == NULL) { 2271 /* Mark end of the window */ 2272 net->fast_recovery_tsn = asoc->sending_seq - 1; 2273 } else { 2274 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2275 } 2276 2277 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 2278 stcb->sctp_ep, stcb, net, 2279 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_3); 2280 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 2281 stcb->sctp_ep, stcb, net); 2282 } 2283 } else if (net->net_ack > 0) { 2284 /* 2285 * Mark a peg that we WOULD have done a cwnd 2286 * reduction but RFC2582 prevented this action. 2287 */ 2288 SCTP_STAT_INCR(sctps_fastretransinrtt); 2289 } 2290 } 2291 } 2292 2293 static void 2294 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 2295 struct sctp_nets *net) 2296 { 2297 int old_cwnd = net->cwnd; 2298 2299 /* JRS - reset as if the state were being changed to timeout */ 2300 htcp_reset(&net->cc_mod.htcp_ca); 2301 net->ssthresh = htcp_recalc_ssthresh(net); 2302 net->cwnd = net->mtu; 2303 net->partial_bytes_acked = 0; 2304 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2305 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 2306 } 2307 } 2308 2309 static void 2310 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 2311 struct sctp_nets *net, int in_window, int num_pkt_lost SCTP_UNUSED) 2312 { 2313 int old_cwnd; 2314 2315 old_cwnd = net->cwnd; 2316 2317 /* JRS - reset hctp as if state changed */ 2318 if (in_window == 0) { 2319 htcp_reset(&net->cc_mod.htcp_ca); 2320 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 2321 net->ssthresh = htcp_recalc_ssthresh(net); 2322 if (net->ssthresh < net->mtu) { 2323 net->ssthresh = net->mtu; 2324 /* here back off the timer as well, to slow us down */ 2325 net->RTO <<= 1; 2326 } 2327 net->cwnd = net->ssthresh; 2328 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2329 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2330 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 2331 } 2332 } 2333 } 2334 2335 const struct sctp_cc_functions sctp_cc_functions[] = { 2336 { 2337 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2338 .sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack, 2339 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2340 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2341 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2342 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2343 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2344 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2345 }, 2346 { 2347 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2348 .sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack, 2349 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2350 .sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr, 2351 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2352 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2353 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2354 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2355 }, 2356 { 2357 .sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param, 2358 .sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack, 2359 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2360 .sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr, 2361 .sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout, 2362 .sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo, 2363 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2364 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2365 }, 2366 { 2367 .sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param, 2368 .sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack, 2369 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2370 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2371 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2372 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo, 2373 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2374 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2375 .sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted, 2376 .sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged, 2377 .sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins, 2378 .sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack, 2379 .sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option, 2380 .sctp_rtt_calculated = sctp_rtt_rtcc_calculated 2381 } 2382 }; 2383