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 #include <netinet/sctp_os.h> 37 #include <netinet/sctp_var.h> 38 #include <netinet/sctp_sysctl.h> 39 #include <netinet/sctp_pcb.h> 40 #include <netinet/sctp_header.h> 41 #include <netinet/sctputil.h> 42 #include <netinet/sctp_output.h> 43 #include <netinet/sctp_input.h> 44 #include <netinet/sctp_indata.h> 45 #include <netinet/sctp_uio.h> 46 #include <netinet/sctp_timer.h> 47 #include <netinet/sctp_auth.h> 48 #include <netinet/sctp_asconf.h> 49 #include <netinet/sctp_kdtrace.h> 50 51 #define SHIFT_MPTCP_MULTI_N 40 52 #define SHIFT_MPTCP_MULTI_Z 16 53 #define SHIFT_MPTCP_MULTI 8 54 55 #ifdef KDTRACE_HOOOKS 56 #define __dtrace 57 #else 58 #define __dtrace __unused 59 #endif 60 61 static void 62 sctp_enforce_cwnd_limit(struct sctp_association *assoc, struct sctp_nets *net) 63 { 64 if ((assoc->max_cwnd > 0) && 65 (net->cwnd > assoc->max_cwnd) && 66 (net->cwnd > (net->mtu - sizeof(struct sctphdr)))) { 67 net->cwnd = assoc->max_cwnd; 68 if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) { 69 net->cwnd = net->mtu - sizeof(struct sctphdr); 70 } 71 } 72 } 73 74 static void 75 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 76 { 77 struct sctp_association *assoc; 78 uint32_t cwnd_in_mtu; 79 80 assoc = &stcb->asoc; 81 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 82 if (cwnd_in_mtu == 0) { 83 /* Using 0 means that the value of RFC 4960 is used. */ 84 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 85 } else { 86 /* 87 * We take the minimum of the burst limit and the initial 88 * congestion window. 89 */ 90 if ((assoc->max_burst > 0) && (cwnd_in_mtu > assoc->max_burst)) 91 cwnd_in_mtu = assoc->max_burst; 92 net->cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 93 } 94 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 95 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) { 96 /* In case of resource pooling initialize appropriately */ 97 net->cwnd /= assoc->numnets; 98 if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) { 99 net->cwnd = net->mtu - sizeof(struct sctphdr); 100 } 101 } 102 sctp_enforce_cwnd_limit(assoc, net); 103 net->ssthresh = assoc->peers_rwnd; 104 SDT_PROBE5(sctp, cwnd, net, init, 105 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 106 0, net->cwnd); 107 if (SCTP_BASE_SYSCTL(sctp_logging_level) & 108 (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 109 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 110 } 111 } 112 113 static void 114 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb, 115 struct sctp_association *asoc) 116 { 117 struct sctp_nets *net; 118 uint32_t t_ssthresh, t_cwnd; 119 uint64_t t_ucwnd_sbw; 120 121 /* MT FIXME: Don't compute this over and over again */ 122 t_ssthresh = 0; 123 t_cwnd = 0; 124 t_ucwnd_sbw = 0; 125 if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) || 126 (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) { 127 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 128 t_ssthresh += net->ssthresh; 129 t_cwnd += net->cwnd; 130 if (net->lastsa > 0) { 131 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)net->lastsa; 132 } 133 } 134 if (t_ucwnd_sbw == 0) { 135 t_ucwnd_sbw = 1; 136 } 137 } 138 139 /*- 140 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 141 * (net->fast_retran_loss_recovery == 0))) 142 */ 143 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 144 if ((asoc->fast_retran_loss_recovery == 0) || 145 (asoc->sctp_cmt_on_off > 0)) { 146 /* out of a RFC2582 Fast recovery window? */ 147 if (net->net_ack > 0) { 148 /* 149 * per section 7.2.3, are there any 150 * destinations that had a fast retransmit 151 * to them. If so what we need to do is 152 * adjust ssthresh and cwnd. 153 */ 154 struct sctp_tmit_chunk *lchk; 155 int old_cwnd = net->cwnd; 156 157 if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) || 158 (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) { 159 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) { 160 net->ssthresh = (uint32_t)(((uint64_t)4 * 161 (uint64_t)net->mtu * 162 (uint64_t)net->ssthresh) / 163 (uint64_t)t_ssthresh); 164 } 165 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2) { 166 uint32_t srtt; 167 168 srtt = net->lastsa; 169 /* 170 * lastsa>>3; we don't need 171 * to devide ... 172 */ 173 if (srtt == 0) { 174 srtt = 1; 175 } 176 /* 177 * Short Version => Equal to 178 * Contel Version MBe 179 */ 180 net->ssthresh = (uint32_t)(((uint64_t)4 * 181 (uint64_t)net->mtu * 182 (uint64_t)net->cwnd) / 183 ((uint64_t)srtt * 184 t_ucwnd_sbw)); 185 /* INCREASE FACTOR */ ; 186 } 187 if ((net->cwnd > t_cwnd / 2) && 188 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 189 net->ssthresh = net->cwnd - t_cwnd / 2; 190 } 191 if (net->ssthresh < net->mtu) { 192 net->ssthresh = net->mtu; 193 } 194 } else { 195 net->ssthresh = net->cwnd / 2; 196 if (net->ssthresh < (net->mtu * 2)) { 197 net->ssthresh = 2 * net->mtu; 198 } 199 } 200 net->cwnd = net->ssthresh; 201 sctp_enforce_cwnd_limit(asoc, net); 202 SDT_PROBE5(sctp, cwnd, net, fr, 203 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 204 old_cwnd, net->cwnd); 205 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 206 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 207 SCTP_CWND_LOG_FROM_FR); 208 } 209 lchk = TAILQ_FIRST(&asoc->send_queue); 210 211 net->partial_bytes_acked = 0; 212 /* Turn on fast recovery window */ 213 asoc->fast_retran_loss_recovery = 1; 214 if (lchk == NULL) { 215 /* Mark end of the window */ 216 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 217 } else { 218 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 219 } 220 221 /* 222 * CMT fast recovery -- per destination 223 * recovery variable. 224 */ 225 net->fast_retran_loss_recovery = 1; 226 227 if (lchk == NULL) { 228 /* Mark end of the window */ 229 net->fast_recovery_tsn = asoc->sending_seq - 1; 230 } else { 231 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 232 } 233 234 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 235 stcb->sctp_ep, stcb, net, 236 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_1); 237 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 238 stcb->sctp_ep, stcb, net); 239 } 240 } else if (net->net_ack > 0) { 241 /* 242 * Mark a peg that we WOULD have done a cwnd 243 * reduction but RFC2582 prevented this action. 244 */ 245 SCTP_STAT_INCR(sctps_fastretransinrtt); 246 } 247 } 248 } 249 250 /* Defines for instantaneous bw decisions */ 251 #define SCTP_INST_LOOSING 1 /* Losing to other flows */ 252 #define SCTP_INST_NEUTRAL 2 /* Neutral, no indication */ 253 #define SCTP_INST_GAINING 3 /* Gaining, step down possible */ 254 255 static int 256 cc_bw_same(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, 257 uint64_t rtt_offset, uint64_t vtag, uint8_t inst_ind) 258 { 259 uint64_t oth __dtrace, probepoint __dtrace; 260 261 probepoint = (((uint64_t)net->cwnd) << 32); 262 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 263 /* 264 * rtt increased we don't update bw.. so we don't update the 265 * rtt either. 266 */ 267 /* Probe point 5 */ 268 probepoint |= ((5 << 16) | 1); 269 SDT_PROBE5(sctp, cwnd, net, rttvar, 270 vtag, 271 ((net->cc_mod.rtcc.lbw << 32) | nbw), 272 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 273 net->flight_size, 274 probepoint); 275 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 276 if (net->cc_mod.rtcc.last_step_state == 5) 277 net->cc_mod.rtcc.step_cnt++; 278 else 279 net->cc_mod.rtcc.step_cnt = 1; 280 net->cc_mod.rtcc.last_step_state = 5; 281 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 282 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 283 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 284 /* Try a step down */ 285 oth = net->cc_mod.rtcc.vol_reduce; 286 oth <<= 16; 287 oth |= net->cc_mod.rtcc.step_cnt; 288 oth <<= 16; 289 oth |= net->cc_mod.rtcc.last_step_state; 290 SDT_PROBE5(sctp, cwnd, net, rttstep, 291 vtag, 292 ((net->cc_mod.rtcc.lbw << 32) | nbw), 293 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 294 oth, 295 probepoint); 296 if (net->cwnd > (4 * net->mtu)) { 297 net->cwnd -= net->mtu; 298 net->cc_mod.rtcc.vol_reduce++; 299 } else { 300 net->cc_mod.rtcc.step_cnt = 0; 301 } 302 } 303 } 304 return (1); 305 } 306 if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 307 /* 308 * rtt decreased, there could be more room. we update both 309 * the bw and the rtt here to lock this in as a good step 310 * down. 311 */ 312 /* Probe point 6 */ 313 probepoint |= ((6 << 16) | 0); 314 SDT_PROBE5(sctp, cwnd, net, rttvar, 315 vtag, 316 ((net->cc_mod.rtcc.lbw << 32) | nbw), 317 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 318 net->flight_size, 319 probepoint); 320 if (net->cc_mod.rtcc.steady_step) { 321 oth = net->cc_mod.rtcc.vol_reduce; 322 oth <<= 16; 323 oth |= net->cc_mod.rtcc.step_cnt; 324 oth <<= 16; 325 oth |= net->cc_mod.rtcc.last_step_state; 326 SDT_PROBE5(sctp, cwnd, net, rttstep, 327 vtag, 328 ((net->cc_mod.rtcc.lbw << 32) | nbw), 329 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 330 oth, 331 probepoint); 332 if ((net->cc_mod.rtcc.last_step_state == 5) && 333 (net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step)) { 334 /* Step down worked */ 335 net->cc_mod.rtcc.step_cnt = 0; 336 return (1); 337 } else { 338 net->cc_mod.rtcc.last_step_state = 6; 339 net->cc_mod.rtcc.step_cnt = 0; 340 } 341 } 342 net->cc_mod.rtcc.lbw = nbw; 343 net->cc_mod.rtcc.lbw_rtt = net->rtt; 344 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 345 if (inst_ind == SCTP_INST_GAINING) 346 return (1); 347 else if (inst_ind == SCTP_INST_NEUTRAL) 348 return (1); 349 else 350 return (0); 351 } 352 /* 353 * Ok bw and rtt remained the same .. no update to any 354 */ 355 /* Probe point 7 */ 356 probepoint |= ((7 << 16) | net->cc_mod.rtcc.ret_from_eq); 357 SDT_PROBE5(sctp, cwnd, net, rttvar, 358 vtag, 359 ((net->cc_mod.rtcc.lbw << 32) | nbw), 360 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 361 net->flight_size, 362 probepoint); 363 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 364 if (net->cc_mod.rtcc.last_step_state == 5) 365 net->cc_mod.rtcc.step_cnt++; 366 else 367 net->cc_mod.rtcc.step_cnt = 1; 368 net->cc_mod.rtcc.last_step_state = 5; 369 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 370 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 371 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 372 /* Try a step down */ 373 if (net->cwnd > (4 * net->mtu)) { 374 net->cwnd -= net->mtu; 375 net->cc_mod.rtcc.vol_reduce++; 376 return (1); 377 } else { 378 net->cc_mod.rtcc.step_cnt = 0; 379 } 380 } 381 } 382 if (inst_ind == SCTP_INST_GAINING) 383 return (1); 384 else if (inst_ind == SCTP_INST_NEUTRAL) 385 return (1); 386 else 387 return ((int)net->cc_mod.rtcc.ret_from_eq); 388 } 389 390 static int 391 cc_bw_decrease(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset, 392 uint64_t vtag, uint8_t inst_ind) 393 { 394 uint64_t oth __dtrace, probepoint __dtrace; 395 396 /* Bandwidth decreased. */ 397 probepoint = (((uint64_t)net->cwnd) << 32); 398 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 399 /* rtt increased */ 400 /* Did we add more */ 401 if ((net->cwnd > net->cc_mod.rtcc.cwnd_at_bw_set) && 402 (inst_ind != SCTP_INST_LOOSING)) { 403 /* We caused it maybe.. back off? */ 404 /* PROBE POINT 1 */ 405 probepoint |= ((1 << 16) | 1); 406 SDT_PROBE5(sctp, cwnd, net, rttvar, 407 vtag, 408 ((net->cc_mod.rtcc.lbw << 32) | nbw), 409 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 410 net->flight_size, 411 probepoint); 412 if (net->cc_mod.rtcc.ret_from_eq) { 413 /* 414 * Switch over to CA if we are less 415 * aggressive 416 */ 417 net->ssthresh = net->cwnd - 1; 418 net->partial_bytes_acked = 0; 419 } 420 return (1); 421 } 422 /* Probe point 2 */ 423 probepoint |= ((2 << 16) | 0); 424 SDT_PROBE5(sctp, cwnd, net, rttvar, 425 vtag, 426 ((net->cc_mod.rtcc.lbw << 32) | nbw), 427 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 428 net->flight_size, 429 probepoint); 430 /* Someone else - fight for more? */ 431 if (net->cc_mod.rtcc.steady_step) { 432 oth = net->cc_mod.rtcc.vol_reduce; 433 oth <<= 16; 434 oth |= net->cc_mod.rtcc.step_cnt; 435 oth <<= 16; 436 oth |= net->cc_mod.rtcc.last_step_state; 437 SDT_PROBE5(sctp, cwnd, net, rttstep, 438 vtag, 439 ((net->cc_mod.rtcc.lbw << 32) | nbw), 440 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 441 oth, 442 probepoint); 443 /* 444 * Did we voluntarily give up some? if so take one 445 * back please 446 */ 447 if ((net->cc_mod.rtcc.vol_reduce) && 448 (inst_ind != SCTP_INST_GAINING)) { 449 net->cwnd += net->mtu; 450 sctp_enforce_cwnd_limit(&stcb->asoc, net); 451 net->cc_mod.rtcc.vol_reduce--; 452 } 453 net->cc_mod.rtcc.last_step_state = 2; 454 net->cc_mod.rtcc.step_cnt = 0; 455 } 456 goto out_decision; 457 } else if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 458 /* bw & rtt decreased */ 459 /* Probe point 3 */ 460 probepoint |= ((3 << 16) | 0); 461 SDT_PROBE5(sctp, cwnd, net, rttvar, 462 vtag, 463 ((net->cc_mod.rtcc.lbw << 32) | nbw), 464 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 465 net->flight_size, 466 probepoint); 467 if (net->cc_mod.rtcc.steady_step) { 468 oth = net->cc_mod.rtcc.vol_reduce; 469 oth <<= 16; 470 oth |= net->cc_mod.rtcc.step_cnt; 471 oth <<= 16; 472 oth |= net->cc_mod.rtcc.last_step_state; 473 SDT_PROBE5(sctp, cwnd, net, rttstep, 474 vtag, 475 ((net->cc_mod.rtcc.lbw << 32) | nbw), 476 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 477 oth, 478 probepoint); 479 if ((net->cc_mod.rtcc.vol_reduce) && 480 (inst_ind != SCTP_INST_GAINING)) { 481 net->cwnd += net->mtu; 482 sctp_enforce_cwnd_limit(&stcb->asoc, net); 483 net->cc_mod.rtcc.vol_reduce--; 484 } 485 net->cc_mod.rtcc.last_step_state = 3; 486 net->cc_mod.rtcc.step_cnt = 0; 487 } 488 goto out_decision; 489 } 490 /* The bw decreased but rtt stayed the same */ 491 /* Probe point 4 */ 492 probepoint |= ((4 << 16) | 0); 493 SDT_PROBE5(sctp, cwnd, net, rttvar, 494 vtag, 495 ((net->cc_mod.rtcc.lbw << 32) | nbw), 496 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 497 net->flight_size, 498 probepoint); 499 if (net->cc_mod.rtcc.steady_step) { 500 oth = net->cc_mod.rtcc.vol_reduce; 501 oth <<= 16; 502 oth |= net->cc_mod.rtcc.step_cnt; 503 oth <<= 16; 504 oth |= net->cc_mod.rtcc.last_step_state; 505 SDT_PROBE5(sctp, cwnd, net, rttstep, 506 vtag, 507 ((net->cc_mod.rtcc.lbw << 32) | nbw), 508 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 509 oth, 510 probepoint); 511 if ((net->cc_mod.rtcc.vol_reduce) && 512 (inst_ind != SCTP_INST_GAINING)) { 513 net->cwnd += net->mtu; 514 sctp_enforce_cwnd_limit(&stcb->asoc, net); 515 net->cc_mod.rtcc.vol_reduce--; 516 } 517 net->cc_mod.rtcc.last_step_state = 4; 518 net->cc_mod.rtcc.step_cnt = 0; 519 } 520 out_decision: 521 net->cc_mod.rtcc.lbw = nbw; 522 net->cc_mod.rtcc.lbw_rtt = net->rtt; 523 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 524 if (inst_ind == SCTP_INST_GAINING) { 525 return (1); 526 } else { 527 return (0); 528 } 529 } 530 531 static int 532 cc_bw_increase(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t vtag) 533 { 534 uint64_t oth __dtrace, probepoint __dtrace; 535 536 /* 537 * BW increased, so update and return 0, since all actions in our 538 * table say to do the normal CC update. Note that we pay no 539 * attention to the inst_ind since our overall sum is increasing. 540 */ 541 /* PROBE POINT 0 */ 542 probepoint = (((uint64_t)net->cwnd) << 32); 543 SDT_PROBE5(sctp, cwnd, net, rttvar, 544 vtag, 545 ((net->cc_mod.rtcc.lbw << 32) | nbw), 546 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 547 net->flight_size, 548 probepoint); 549 if (net->cc_mod.rtcc.steady_step) { 550 oth = net->cc_mod.rtcc.vol_reduce; 551 oth <<= 16; 552 oth |= net->cc_mod.rtcc.step_cnt; 553 oth <<= 16; 554 oth |= net->cc_mod.rtcc.last_step_state; 555 SDT_PROBE5(sctp, cwnd, net, rttstep, 556 vtag, 557 ((net->cc_mod.rtcc.lbw << 32) | nbw), 558 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 559 oth, 560 probepoint); 561 net->cc_mod.rtcc.last_step_state = 0; 562 net->cc_mod.rtcc.step_cnt = 0; 563 net->cc_mod.rtcc.vol_reduce = 0; 564 } 565 net->cc_mod.rtcc.lbw = nbw; 566 net->cc_mod.rtcc.lbw_rtt = net->rtt; 567 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 568 return (0); 569 } 570 571 /* RTCC Algorithm to limit growth of cwnd, return 572 * true if you want to NOT allow cwnd growth 573 */ 574 static int 575 cc_bw_limit(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw) 576 { 577 uint64_t bw_offset, rtt_offset; 578 uint64_t probepoint __dtrace, rtt, vtag; 579 uint64_t bytes_for_this_rtt, inst_bw; 580 uint64_t div, inst_off; 581 int bw_shift; 582 uint8_t inst_ind; 583 int ret; 584 585 /*- 586 * Here we need to see if we want 587 * to limit cwnd growth due to increase 588 * in overall rtt but no increase in bw. 589 * We use the following table to figure 590 * out what we should do. When we return 591 * 0, cc update goes on as planned. If we 592 * return 1, then no cc update happens and cwnd 593 * stays where it is at. 594 * ---------------------------------- 595 * BW | RTT | Action 596 * ********************************* 597 * INC | INC | return 0 598 * ---------------------------------- 599 * INC | SAME | return 0 600 * ---------------------------------- 601 * INC | DECR | return 0 602 * ---------------------------------- 603 * SAME | INC | return 1 604 * ---------------------------------- 605 * SAME | SAME | return 1 606 * ---------------------------------- 607 * SAME | DECR | return 0 608 * ---------------------------------- 609 * DECR | INC | return 0 or 1 based on if we caused. 610 * ---------------------------------- 611 * DECR | SAME | return 0 612 * ---------------------------------- 613 * DECR | DECR | return 0 614 * ---------------------------------- 615 * 616 * We are a bit fuzz on what an increase or 617 * decrease is. For BW it is the same if 618 * it did not change within 1/64th. For 619 * RTT it stayed the same if it did not 620 * change within 1/32nd 621 */ 622 bw_shift = SCTP_BASE_SYSCTL(sctp_rttvar_bw); 623 rtt = stcb->asoc.my_vtag; 624 vtag = (rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | (stcb->rport); 625 probepoint = (((uint64_t)net->cwnd) << 32); 626 rtt = net->rtt; 627 if (net->cc_mod.rtcc.rtt_set_this_sack) { 628 net->cc_mod.rtcc.rtt_set_this_sack = 0; 629 bytes_for_this_rtt = net->cc_mod.rtcc.bw_bytes - net->cc_mod.rtcc.bw_bytes_at_last_rttc; 630 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 631 if (net->rtt) { 632 div = net->rtt / 1000; 633 if (div) { 634 inst_bw = bytes_for_this_rtt / div; 635 inst_off = inst_bw >> bw_shift; 636 if (inst_bw > nbw) 637 inst_ind = SCTP_INST_GAINING; 638 else if ((inst_bw + inst_off) < nbw) 639 inst_ind = SCTP_INST_LOOSING; 640 else 641 inst_ind = SCTP_INST_NEUTRAL; 642 probepoint |= ((0xb << 16) | inst_ind); 643 } else { 644 inst_ind = net->cc_mod.rtcc.last_inst_ind; 645 inst_bw = bytes_for_this_rtt / (uint64_t)(net->rtt); 646 /* Can't determine do not change */ 647 probepoint |= ((0xc << 16) | inst_ind); 648 } 649 } else { 650 inst_ind = net->cc_mod.rtcc.last_inst_ind; 651 inst_bw = bytes_for_this_rtt; 652 /* Can't determine do not change */ 653 probepoint |= ((0xd << 16) | inst_ind); 654 } 655 SDT_PROBE5(sctp, cwnd, net, rttvar, 656 vtag, 657 ((nbw << 32) | inst_bw), 658 ((net->cc_mod.rtcc.lbw_rtt << 32) | rtt), 659 net->flight_size, 660 probepoint); 661 } else { 662 /* No rtt measurement, use last one */ 663 inst_ind = net->cc_mod.rtcc.last_inst_ind; 664 } 665 bw_offset = net->cc_mod.rtcc.lbw >> bw_shift; 666 if (nbw > net->cc_mod.rtcc.lbw + bw_offset) { 667 ret = cc_bw_increase(stcb, net, nbw, vtag); 668 goto out; 669 } 670 rtt_offset = net->cc_mod.rtcc.lbw_rtt >> SCTP_BASE_SYSCTL(sctp_rttvar_rtt); 671 if (nbw < net->cc_mod.rtcc.lbw - bw_offset) { 672 ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, vtag, inst_ind); 673 goto out; 674 } 675 /* 676 * If we reach here then we are in a situation where the bw stayed 677 * the same. 678 */ 679 ret = cc_bw_same(stcb, net, nbw, rtt_offset, vtag, inst_ind); 680 out: 681 net->cc_mod.rtcc.last_inst_ind = inst_ind; 682 return (ret); 683 } 684 685 static void 686 sctp_cwnd_update_after_sack_common(struct sctp_tcb *stcb, 687 struct sctp_association *asoc, 688 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit, int use_rtcc) 689 { 690 struct sctp_nets *net; 691 int old_cwnd __dtrace; 692 uint32_t t_ssthresh, incr; 693 uint64_t t_ucwnd_sbw; 694 uint64_t t_path_mptcp; 695 uint64_t mptcp_like_alpha; 696 uint32_t srtt; 697 uint64_t max_path; 698 699 /* MT FIXME: Don't compute this over and over again */ 700 t_ssthresh = 0; 701 t_ucwnd_sbw = 0; 702 t_path_mptcp = 0; 703 mptcp_like_alpha = 1; 704 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 705 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2) || 706 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_MPTCP)) { 707 max_path = 0; 708 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 709 t_ssthresh += net->ssthresh; 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 #ifdef JANA_CMT_FAST_RECOVERY 742 /* 743 * CMT fast recovery code. Need to debug. 744 */ 745 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 746 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 747 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 748 net->will_exit_fast_recovery = 1; 749 } 750 } 751 #endif 752 /* if nothing was acked on this destination skip it */ 753 if (net->net_ack == 0) { 754 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 755 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 756 } 757 continue; 758 } 759 #ifdef JANA_CMT_FAST_RECOVERY 760 /* 761 * CMT fast recovery code 762 */ 763 /* 764 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 765 * && net->will_exit_fast_recovery == 0) { @@@ Do something 766 * } else if (sctp_cmt_on_off == 0 && 767 * asoc->fast_retran_loss_recovery && will_exit == 0) { 768 */ 769 #endif 770 771 if (asoc->fast_retran_loss_recovery && 772 (will_exit == 0) && 773 (asoc->sctp_cmt_on_off == 0)) { 774 /* 775 * If we are in loss recovery we skip any cwnd 776 * update 777 */ 778 return; 779 } 780 /* 781 * Did any measurements go on for this network? 782 */ 783 if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) { 784 uint64_t nbw; 785 786 /* 787 * At this point our bw_bytes has been updated by 788 * incoming sack information. 789 * 790 * But our bw may not yet be set. 791 * 792 */ 793 if ((net->cc_mod.rtcc.new_tot_time / 1000) > 0) { 794 nbw = net->cc_mod.rtcc.bw_bytes / (net->cc_mod.rtcc.new_tot_time / 1000); 795 } else { 796 nbw = net->cc_mod.rtcc.bw_bytes; 797 } 798 if (net->cc_mod.rtcc.lbw) { 799 if (cc_bw_limit(stcb, net, nbw)) { 800 /* Hold here, no update */ 801 continue; 802 } 803 } else { 804 uint64_t vtag __dtrace, probepoint __dtrace; 805 806 probepoint = (((uint64_t)net->cwnd) << 32); 807 probepoint |= ((0xa << 16) | 0); 808 vtag = (net->rtt << 32) | 809 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 810 (stcb->rport); 811 812 SDT_PROBE5(sctp, cwnd, net, rttvar, 813 vtag, 814 nbw, 815 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 816 net->flight_size, 817 probepoint); 818 net->cc_mod.rtcc.lbw = nbw; 819 net->cc_mod.rtcc.lbw_rtt = net->rtt; 820 if (net->cc_mod.rtcc.rtt_set_this_sack) { 821 net->cc_mod.rtcc.rtt_set_this_sack = 0; 822 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 823 } 824 } 825 } 826 /* 827 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 828 * moved. 829 */ 830 if (accum_moved || 831 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 832 /* If the cumulative ack moved we can proceed */ 833 if (net->cwnd <= net->ssthresh) { 834 /* We are in slow start */ 835 if (net->flight_size + net->net_ack >= net->cwnd) { 836 uint32_t limit; 837 838 old_cwnd = net->cwnd; 839 switch (asoc->sctp_cmt_on_off) { 840 case SCTP_CMT_RPV1: 841 limit = (uint32_t)(((uint64_t)net->mtu * 842 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 843 (uint64_t)net->ssthresh) / 844 (uint64_t)t_ssthresh); 845 incr = (uint32_t)(((uint64_t)net->net_ack * 846 (uint64_t)net->ssthresh) / 847 (uint64_t)t_ssthresh); 848 if (incr > limit) { 849 incr = limit; 850 } 851 if (incr == 0) { 852 incr = 1; 853 } 854 break; 855 case SCTP_CMT_RPV2: 856 /* 857 * lastsa>>3; we don't need 858 * to divide ... 859 */ 860 srtt = net->lastsa; 861 if (srtt == 0) { 862 srtt = 1; 863 } 864 limit = (uint32_t)(((uint64_t)net->mtu * 865 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 866 (uint64_t)net->cwnd) / 867 ((uint64_t)srtt * t_ucwnd_sbw)); 868 /* INCREASE FACTOR */ 869 incr = (uint32_t)(((uint64_t)net->net_ack * 870 (uint64_t)net->cwnd) / 871 ((uint64_t)srtt * t_ucwnd_sbw)); 872 /* INCREASE FACTOR */ 873 if (incr > limit) { 874 incr = limit; 875 } 876 if (incr == 0) { 877 incr = 1; 878 } 879 break; 880 case SCTP_CMT_MPTCP: 881 limit = (uint32_t)(((uint64_t)net->mtu * 882 mptcp_like_alpha * 883 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) >> 884 SHIFT_MPTCP_MULTI); 885 incr = (uint32_t)(((uint64_t)net->net_ack * 886 mptcp_like_alpha) >> 887 SHIFT_MPTCP_MULTI); 888 if (incr > limit) { 889 incr = limit; 890 } 891 if (incr > net->net_ack) { 892 incr = net->net_ack; 893 } 894 if (incr > net->mtu) { 895 incr = net->mtu; 896 } 897 break; 898 default: 899 incr = net->net_ack; 900 if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) { 901 incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable); 902 } 903 break; 904 } 905 net->cwnd += incr; 906 sctp_enforce_cwnd_limit(asoc, net); 907 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 908 sctp_log_cwnd(stcb, net, incr, 909 SCTP_CWND_LOG_FROM_SS); 910 } 911 SDT_PROBE5(sctp, cwnd, net, ack, 912 stcb->asoc.my_vtag, 913 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 914 net, 915 old_cwnd, net->cwnd); 916 } else { 917 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 918 sctp_log_cwnd(stcb, net, net->net_ack, 919 SCTP_CWND_LOG_NOADV_SS); 920 } 921 } 922 } else { 923 /* We are in congestion avoidance */ 924 /* 925 * Add to pba 926 */ 927 net->partial_bytes_acked += net->net_ack; 928 929 if ((net->flight_size + net->net_ack >= net->cwnd) && 930 (net->partial_bytes_acked >= net->cwnd)) { 931 net->partial_bytes_acked -= net->cwnd; 932 old_cwnd = net->cwnd; 933 switch (asoc->sctp_cmt_on_off) { 934 case SCTP_CMT_RPV1: 935 incr = (uint32_t)(((uint64_t)net->mtu * 936 (uint64_t)net->ssthresh) / 937 (uint64_t)t_ssthresh); 938 if (incr == 0) { 939 incr = 1; 940 } 941 break; 942 case SCTP_CMT_RPV2: 943 /* 944 * lastsa>>3; we don't need 945 * to divide ... 946 */ 947 srtt = net->lastsa; 948 if (srtt == 0) { 949 srtt = 1; 950 } 951 incr = (uint32_t)((uint64_t)net->mtu * 952 (uint64_t)net->cwnd / 953 ((uint64_t)srtt * 954 t_ucwnd_sbw)); 955 /* INCREASE FACTOR */ 956 if (incr == 0) { 957 incr = 1; 958 } 959 break; 960 case SCTP_CMT_MPTCP: 961 incr = (uint32_t)((mptcp_like_alpha * 962 (uint64_t)net->cwnd) >> 963 SHIFT_MPTCP_MULTI); 964 if (incr > net->mtu) { 965 incr = net->mtu; 966 } 967 break; 968 default: 969 incr = net->mtu; 970 break; 971 } 972 net->cwnd += incr; 973 sctp_enforce_cwnd_limit(asoc, net); 974 SDT_PROBE5(sctp, cwnd, net, ack, 975 stcb->asoc.my_vtag, 976 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 977 net, 978 old_cwnd, net->cwnd); 979 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 980 sctp_log_cwnd(stcb, net, net->mtu, 981 SCTP_CWND_LOG_FROM_CA); 982 } 983 } else { 984 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 985 sctp_log_cwnd(stcb, net, net->net_ack, 986 SCTP_CWND_LOG_NOADV_CA); 987 } 988 } 989 } 990 } else { 991 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 992 sctp_log_cwnd(stcb, net, net->mtu, 993 SCTP_CWND_LOG_NO_CUMACK); 994 } 995 } 996 } 997 } 998 999 static void 1000 sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb, struct sctp_nets *net) 1001 { 1002 int old_cwnd __dtrace; 1003 1004 old_cwnd = net->cwnd; 1005 net->cwnd = net->mtu; 1006 SDT_PROBE5(sctp, cwnd, net, ack, 1007 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 1008 old_cwnd, net->cwnd); 1009 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1010 (void *)net, net->cwnd); 1011 } 1012 1013 static void 1014 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 1015 { 1016 int old_cwnd = net->cwnd; 1017 uint32_t t_ssthresh, t_cwnd; 1018 uint64_t t_ucwnd_sbw; 1019 1020 /* MT FIXME: Don't compute this over and over again */ 1021 t_ssthresh = 0; 1022 t_cwnd = 0; 1023 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) || 1024 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) { 1025 struct sctp_nets *lnet; 1026 uint32_t srtt; 1027 1028 t_ucwnd_sbw = 0; 1029 TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) { 1030 t_ssthresh += lnet->ssthresh; 1031 t_cwnd += lnet->cwnd; 1032 srtt = lnet->lastsa; 1033 /* lastsa>>3; we don't need to divide ... */ 1034 if (srtt > 0) { 1035 t_ucwnd_sbw += (uint64_t)lnet->cwnd / (uint64_t)srtt; 1036 } 1037 } 1038 if (t_ssthresh < 1) { 1039 t_ssthresh = 1; 1040 } 1041 if (t_ucwnd_sbw < 1) { 1042 t_ucwnd_sbw = 1; 1043 } 1044 if (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) { 1045 net->ssthresh = (uint32_t)(((uint64_t)4 * 1046 (uint64_t)net->mtu * 1047 (uint64_t)net->ssthresh) / 1048 (uint64_t)t_ssthresh); 1049 } else { 1050 uint64_t cc_delta; 1051 1052 srtt = net->lastsa; 1053 /* lastsa>>3; we don't need to divide ... */ 1054 if (srtt == 0) { 1055 srtt = 1; 1056 } 1057 cc_delta = t_ucwnd_sbw * (uint64_t)srtt / 2; 1058 if (cc_delta < t_cwnd) { 1059 net->ssthresh = (uint32_t)((uint64_t)t_cwnd - cc_delta); 1060 } else { 1061 net->ssthresh = net->mtu; 1062 } 1063 } 1064 if ((net->cwnd > t_cwnd / 2) && 1065 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 1066 net->ssthresh = net->cwnd - t_cwnd / 2; 1067 } 1068 if (net->ssthresh < net->mtu) { 1069 net->ssthresh = net->mtu; 1070 } 1071 } else { 1072 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 1073 } 1074 net->cwnd = net->mtu; 1075 net->partial_bytes_acked = 0; 1076 SDT_PROBE5(sctp, cwnd, net, to, 1077 stcb->asoc.my_vtag, 1078 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1079 net, 1080 old_cwnd, net->cwnd); 1081 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1082 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1083 } 1084 } 1085 1086 static void 1087 sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net, 1088 int in_window, int num_pkt_lost, int use_rtcc) 1089 { 1090 int old_cwnd = net->cwnd; 1091 1092 if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) { 1093 /* Data center Congestion Control */ 1094 if (in_window == 0) { 1095 /* 1096 * Go to CA with the cwnd at the point we sent the 1097 * TSN that was marked with a CE. 1098 */ 1099 if (net->ecn_prev_cwnd < net->cwnd) { 1100 /* Restore to prev cwnd */ 1101 net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost); 1102 } else { 1103 /* Just cut in 1/2 */ 1104 net->cwnd /= 2; 1105 } 1106 /* Drop to CA */ 1107 net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu); 1108 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1109 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1110 } 1111 } else { 1112 /* 1113 * Further tuning down required over the drastic 1114 * original cut 1115 */ 1116 net->ssthresh -= (net->mtu * num_pkt_lost); 1117 net->cwnd -= (net->mtu * num_pkt_lost); 1118 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1119 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1120 } 1121 } 1122 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1123 } else { 1124 if (in_window == 0) { 1125 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1126 net->ssthresh = net->cwnd / 2; 1127 if (net->ssthresh < net->mtu) { 1128 net->ssthresh = net->mtu; 1129 /* 1130 * here back off the timer as well, to slow 1131 * us down 1132 */ 1133 net->RTO <<= 1; 1134 } 1135 net->cwnd = net->ssthresh; 1136 SDT_PROBE5(sctp, cwnd, net, ecn, 1137 stcb->asoc.my_vtag, 1138 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1139 net, 1140 old_cwnd, net->cwnd); 1141 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1142 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1143 } 1144 } 1145 } 1146 1147 } 1148 1149 static void 1150 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 1151 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 1152 uint32_t *bottle_bw, uint32_t *on_queue) 1153 { 1154 uint32_t bw_avail; 1155 unsigned int incr; 1156 int old_cwnd = net->cwnd; 1157 1158 /* get bottle neck bw */ 1159 *bottle_bw = ntohl(cp->bottle_bw); 1160 /* and whats on queue */ 1161 *on_queue = ntohl(cp->current_onq); 1162 /* 1163 * adjust the on-queue if our flight is more it could be that the 1164 * router has not yet gotten data "in-flight" to it 1165 */ 1166 if (*on_queue < net->flight_size) { 1167 *on_queue = net->flight_size; 1168 } 1169 /* rtt is measured in micro seconds, bottle_bw in bytes per second */ 1170 bw_avail = (uint32_t)(((uint64_t)(*bottle_bw) * net->rtt) / (uint64_t)1000000); 1171 if (bw_avail > *bottle_bw) { 1172 /* 1173 * Cap the growth to no more than the bottle neck. This can 1174 * happen as RTT slides up due to queues. It also means if 1175 * you have more than a 1 second RTT with a empty queue you 1176 * will be limited to the bottle_bw per second no matter if 1177 * other points have 1/2 the RTT and you could get more 1178 * out... 1179 */ 1180 bw_avail = *bottle_bw; 1181 } 1182 if (*on_queue > bw_avail) { 1183 /* 1184 * No room for anything else don't allow anything else to be 1185 * "added to the fire". 1186 */ 1187 int seg_inflight, seg_onqueue, my_portion; 1188 1189 net->partial_bytes_acked = 0; 1190 /* how much are we over queue size? */ 1191 incr = *on_queue - bw_avail; 1192 if (stcb->asoc.seen_a_sack_this_pkt) { 1193 /* 1194 * undo any cwnd adjustment that the sack might have 1195 * made 1196 */ 1197 net->cwnd = net->prev_cwnd; 1198 } 1199 /* Now how much of that is mine? */ 1200 seg_inflight = net->flight_size / net->mtu; 1201 seg_onqueue = *on_queue / net->mtu; 1202 my_portion = (incr * seg_inflight) / seg_onqueue; 1203 1204 /* Have I made an adjustment already */ 1205 if (net->cwnd > net->flight_size) { 1206 /* 1207 * for this flight I made an adjustment we need to 1208 * decrease the portion by a share our previous 1209 * adjustment. 1210 */ 1211 int diff_adj; 1212 1213 diff_adj = net->cwnd - net->flight_size; 1214 if (diff_adj > my_portion) 1215 my_portion = 0; 1216 else 1217 my_portion -= diff_adj; 1218 } 1219 /* 1220 * back down to the previous cwnd (assume we have had a sack 1221 * before this packet). minus what ever portion of the 1222 * overage is my fault. 1223 */ 1224 net->cwnd -= my_portion; 1225 1226 /* we will NOT back down more than 1 MTU */ 1227 if (net->cwnd <= net->mtu) { 1228 net->cwnd = net->mtu; 1229 } 1230 /* force into CA */ 1231 net->ssthresh = net->cwnd - 1; 1232 } else { 1233 /* 1234 * Take 1/4 of the space left or max burst up .. whichever 1235 * is less. 1236 */ 1237 incr = (bw_avail - *on_queue) >> 2; 1238 if ((stcb->asoc.max_burst > 0) && 1239 (stcb->asoc.max_burst * net->mtu < incr)) { 1240 incr = stcb->asoc.max_burst * net->mtu; 1241 } 1242 net->cwnd += incr; 1243 } 1244 if (net->cwnd > bw_avail) { 1245 /* We can't exceed the pipe size */ 1246 net->cwnd = bw_avail; 1247 } 1248 if (net->cwnd < net->mtu) { 1249 /* We always have 1 MTU */ 1250 net->cwnd = net->mtu; 1251 } 1252 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1253 if (net->cwnd - old_cwnd != 0) { 1254 /* log only changes */ 1255 SDT_PROBE5(sctp, cwnd, net, pd, 1256 stcb->asoc.my_vtag, 1257 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1258 net, 1259 old_cwnd, net->cwnd); 1260 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1261 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1262 SCTP_CWND_LOG_FROM_SAT); 1263 } 1264 } 1265 } 1266 1267 static void 1268 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 1269 struct sctp_nets *net, int burst_limit) 1270 { 1271 int old_cwnd = net->cwnd; 1272 1273 if (net->ssthresh < net->cwnd) 1274 net->ssthresh = net->cwnd; 1275 if (burst_limit) { 1276 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 1277 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1278 SDT_PROBE5(sctp, cwnd, net, bl, 1279 stcb->asoc.my_vtag, 1280 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1281 net, 1282 old_cwnd, net->cwnd); 1283 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1284 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 1285 } 1286 } 1287 } 1288 1289 static void 1290 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1291 struct sctp_association *asoc, 1292 int accum_moved, int reneged_all, int will_exit) 1293 { 1294 /* Passing a zero argument in last disables the rtcc algorithm */ 1295 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0); 1296 } 1297 1298 static void 1299 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1300 int in_window, int num_pkt_lost) 1301 { 1302 /* Passing a zero argument in last disables the rtcc algorithm */ 1303 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0); 1304 } 1305 1306 /* Here starts the RTCCVAR type CC invented by RRS which 1307 * is a slight mod to RFC2581. We reuse a common routine or 1308 * two since these algorithms are so close and need to 1309 * remain the same. 1310 */ 1311 static void 1312 sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1313 int in_window, int num_pkt_lost) 1314 { 1315 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1); 1316 } 1317 1318 static void 1319 sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net, 1320 struct sctp_tmit_chunk *tp1) 1321 { 1322 net->cc_mod.rtcc.bw_bytes += tp1->send_size; 1323 } 1324 1325 static void 1326 sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb SCTP_UNUSED, 1327 struct sctp_nets *net) 1328 { 1329 if (net->cc_mod.rtcc.tls_needs_set > 0) { 1330 /* We had a bw measurement going on */ 1331 struct timeval ltls; 1332 1333 SCTP_GETPTIME_TIMEVAL(<ls); 1334 timevalsub(<ls, &net->cc_mod.rtcc.tls); 1335 net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec; 1336 } 1337 } 1338 1339 static void 1340 sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb, 1341 struct sctp_nets *net) 1342 { 1343 uint64_t vtag __dtrace, probepoint __dtrace; 1344 1345 if (net->cc_mod.rtcc.lbw) { 1346 /* Clear the old bw.. we went to 0 in-flight */ 1347 vtag = (net->rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1348 (stcb->rport); 1349 probepoint = (((uint64_t)net->cwnd) << 32); 1350 /* Probe point 8 */ 1351 probepoint |= ((8 << 16) | 0); 1352 SDT_PROBE5(sctp, cwnd, net, rttvar, 1353 vtag, 1354 ((net->cc_mod.rtcc.lbw << 32) | 0), 1355 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 1356 net->flight_size, 1357 probepoint); 1358 net->cc_mod.rtcc.lbw_rtt = 0; 1359 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1360 net->cc_mod.rtcc.lbw = 0; 1361 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1362 net->cc_mod.rtcc.vol_reduce = 0; 1363 net->cc_mod.rtcc.bw_tot_time = 0; 1364 net->cc_mod.rtcc.bw_bytes = 0; 1365 net->cc_mod.rtcc.tls_needs_set = 0; 1366 if (net->cc_mod.rtcc.steady_step) { 1367 net->cc_mod.rtcc.vol_reduce = 0; 1368 net->cc_mod.rtcc.step_cnt = 0; 1369 net->cc_mod.rtcc.last_step_state = 0; 1370 } 1371 if (net->cc_mod.rtcc.ret_from_eq) { 1372 /* less aggressive one - reset cwnd too */ 1373 uint32_t cwnd_in_mtu, cwnd; 1374 1375 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 1376 if (cwnd_in_mtu == 0) { 1377 /* 1378 * Using 0 means that the value of RFC 4960 1379 * is used. 1380 */ 1381 cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1382 } else { 1383 /* 1384 * We take the minimum of the burst limit 1385 * and the initial congestion window. 1386 */ 1387 if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst)) 1388 cwnd_in_mtu = stcb->asoc.max_burst; 1389 cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 1390 } 1391 if (net->cwnd > cwnd) { 1392 /* 1393 * Only set if we are not a timeout (i.e. 1394 * down to 1 mtu) 1395 */ 1396 net->cwnd = cwnd; 1397 } 1398 } 1399 } 1400 } 1401 1402 static void 1403 sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb, 1404 struct sctp_nets *net) 1405 { 1406 #ifdef KDTRACE_HOOOKS 1407 uint64_t vtag, probepoint; 1408 #else 1409 uint64_t vtag __unused, probepoint __unused; 1410 #endif 1411 1412 sctp_set_initial_cc_param(stcb, net); 1413 stcb->asoc.use_precise_time = 1; 1414 probepoint = (((uint64_t)net->cwnd) << 32); 1415 probepoint |= ((9 << 16) | 0); 1416 vtag = (net->rtt << 32) | 1417 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | 1418 (stcb->rport); 1419 SDT_PROBE5(sctp, cwnd, net, rttvar, 1420 vtag, 1421 0, 1422 0, 1423 0, 1424 probepoint); 1425 net->cc_mod.rtcc.lbw_rtt = 0; 1426 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1427 net->cc_mod.rtcc.vol_reduce = 0; 1428 net->cc_mod.rtcc.lbw = 0; 1429 net->cc_mod.rtcc.vol_reduce = 0; 1430 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1431 net->cc_mod.rtcc.bw_tot_time = 0; 1432 net->cc_mod.rtcc.bw_bytes = 0; 1433 net->cc_mod.rtcc.tls_needs_set = 0; 1434 net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret); 1435 net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step); 1436 net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn); 1437 net->cc_mod.rtcc.step_cnt = 0; 1438 net->cc_mod.rtcc.last_step_state = 0; 1439 } 1440 1441 static int 1442 sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget, 1443 struct sctp_cc_option *cc_opt) 1444 { 1445 struct sctp_nets *net; 1446 1447 if (setorget == 1) { 1448 /* a set */ 1449 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1450 if ((cc_opt->aid_value.assoc_value != 0) && 1451 (cc_opt->aid_value.assoc_value != 1)) { 1452 return (EINVAL); 1453 } 1454 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1455 net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value; 1456 } 1457 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1458 if ((cc_opt->aid_value.assoc_value != 0) && 1459 (cc_opt->aid_value.assoc_value != 1)) { 1460 return (EINVAL); 1461 } 1462 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1463 net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value; 1464 } 1465 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1466 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1467 net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value; 1468 } 1469 } else { 1470 return (EINVAL); 1471 } 1472 } else { 1473 /* a get */ 1474 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 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.ret_from_eq; 1480 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 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.use_dccc_ecn; 1486 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1487 net = TAILQ_FIRST(&stcb->asoc.nets); 1488 if (net == NULL) { 1489 return (EFAULT); 1490 } 1491 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step; 1492 } else { 1493 return (EINVAL); 1494 } 1495 } 1496 return (0); 1497 } 1498 1499 static void 1500 sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb SCTP_UNUSED, 1501 struct sctp_nets *net) 1502 { 1503 if (net->cc_mod.rtcc.tls_needs_set == 0) { 1504 SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls); 1505 net->cc_mod.rtcc.tls_needs_set = 2; 1506 } 1507 } 1508 1509 static void 1510 sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb, 1511 struct sctp_association *asoc, 1512 int accum_moved, int reneged_all, int will_exit) 1513 { 1514 /* Passing a one argument at the last enables the rtcc algorithm */ 1515 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1); 1516 } 1517 1518 static void 1519 sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb SCTP_UNUSED, 1520 struct sctp_nets *net, 1521 struct timeval *now SCTP_UNUSED) 1522 { 1523 net->cc_mod.rtcc.rtt_set_this_sack = 1; 1524 } 1525 1526 /* Here starts Sally Floyds HS-TCP */ 1527 1528 struct sctp_hs_raise_drop { 1529 int32_t cwnd; 1530 int8_t increase; 1531 int8_t drop_percent; 1532 }; 1533 1534 #define SCTP_HS_TABLE_SIZE 73 1535 1536 static const struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 1537 {38, 1, 50}, /* 0 */ 1538 {118, 2, 44}, /* 1 */ 1539 {221, 3, 41}, /* 2 */ 1540 {347, 4, 38}, /* 3 */ 1541 {495, 5, 37}, /* 4 */ 1542 {663, 6, 35}, /* 5 */ 1543 {851, 7, 34}, /* 6 */ 1544 {1058, 8, 33}, /* 7 */ 1545 {1284, 9, 32}, /* 8 */ 1546 {1529, 10, 31}, /* 9 */ 1547 {1793, 11, 30}, /* 10 */ 1548 {2076, 12, 29}, /* 11 */ 1549 {2378, 13, 28}, /* 12 */ 1550 {2699, 14, 28}, /* 13 */ 1551 {3039, 15, 27}, /* 14 */ 1552 {3399, 16, 27}, /* 15 */ 1553 {3778, 17, 26}, /* 16 */ 1554 {4177, 18, 26}, /* 17 */ 1555 {4596, 19, 25}, /* 18 */ 1556 {5036, 20, 25}, /* 19 */ 1557 {5497, 21, 24}, /* 20 */ 1558 {5979, 22, 24}, /* 21 */ 1559 {6483, 23, 23}, /* 22 */ 1560 {7009, 24, 23}, /* 23 */ 1561 {7558, 25, 22}, /* 24 */ 1562 {8130, 26, 22}, /* 25 */ 1563 {8726, 27, 22}, /* 26 */ 1564 {9346, 28, 21}, /* 27 */ 1565 {9991, 29, 21}, /* 28 */ 1566 {10661, 30, 21}, /* 29 */ 1567 {11358, 31, 20}, /* 30 */ 1568 {12082, 32, 20}, /* 31 */ 1569 {12834, 33, 20}, /* 32 */ 1570 {13614, 34, 19}, /* 33 */ 1571 {14424, 35, 19}, /* 34 */ 1572 {15265, 36, 19}, /* 35 */ 1573 {16137, 37, 19}, /* 36 */ 1574 {17042, 38, 18}, /* 37 */ 1575 {17981, 39, 18}, /* 38 */ 1576 {18955, 40, 18}, /* 39 */ 1577 {19965, 41, 17}, /* 40 */ 1578 {21013, 42, 17}, /* 41 */ 1579 {22101, 43, 17}, /* 42 */ 1580 {23230, 44, 17}, /* 43 */ 1581 {24402, 45, 16}, /* 44 */ 1582 {25618, 46, 16}, /* 45 */ 1583 {26881, 47, 16}, /* 46 */ 1584 {28193, 48, 16}, /* 47 */ 1585 {29557, 49, 15}, /* 48 */ 1586 {30975, 50, 15}, /* 49 */ 1587 {32450, 51, 15}, /* 50 */ 1588 {33986, 52, 15}, /* 51 */ 1589 {35586, 53, 14}, /* 52 */ 1590 {37253, 54, 14}, /* 53 */ 1591 {38992, 55, 14}, /* 54 */ 1592 {40808, 56, 14}, /* 55 */ 1593 {42707, 57, 13}, /* 56 */ 1594 {44694, 58, 13}, /* 57 */ 1595 {46776, 59, 13}, /* 58 */ 1596 {48961, 60, 13}, /* 59 */ 1597 {51258, 61, 13}, /* 60 */ 1598 {53677, 62, 12}, /* 61 */ 1599 {56230, 63, 12}, /* 62 */ 1600 {58932, 64, 12}, /* 63 */ 1601 {61799, 65, 12}, /* 64 */ 1602 {64851, 66, 11}, /* 65 */ 1603 {68113, 67, 11}, /* 66 */ 1604 {71617, 68, 11}, /* 67 */ 1605 {75401, 69, 10}, /* 68 */ 1606 {79517, 70, 10}, /* 69 */ 1607 {84035, 71, 10}, /* 70 */ 1608 {89053, 72, 10}, /* 71 */ 1609 {94717, 73, 9} /* 72 */ 1610 }; 1611 1612 static void 1613 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 1614 { 1615 int cur_val, i, indx, incr; 1616 int old_cwnd = net->cwnd; 1617 1618 cur_val = net->cwnd >> 10; 1619 indx = SCTP_HS_TABLE_SIZE - 1; 1620 1621 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1622 /* normal mode */ 1623 if (net->net_ack > net->mtu) { 1624 net->cwnd += net->mtu; 1625 } else { 1626 net->cwnd += net->net_ack; 1627 } 1628 } else { 1629 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 1630 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 1631 indx = i; 1632 break; 1633 } 1634 } 1635 net->last_hs_used = indx; 1636 incr = (((int32_t)sctp_cwnd_adjust[indx].increase) << 10); 1637 net->cwnd += incr; 1638 } 1639 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1640 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1641 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SS); 1642 } 1643 } 1644 1645 static void 1646 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 1647 { 1648 int cur_val, i, indx; 1649 int old_cwnd = net->cwnd; 1650 1651 cur_val = net->cwnd >> 10; 1652 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1653 /* normal mode */ 1654 net->ssthresh = net->cwnd / 2; 1655 if (net->ssthresh < (net->mtu * 2)) { 1656 net->ssthresh = 2 * net->mtu; 1657 } 1658 net->cwnd = net->ssthresh; 1659 } else { 1660 /* drop by the proper amount */ 1661 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 1662 (int32_t)sctp_cwnd_adjust[net->last_hs_used].drop_percent); 1663 net->cwnd = net->ssthresh; 1664 /* now where are we */ 1665 indx = net->last_hs_used; 1666 cur_val = net->cwnd >> 10; 1667 /* reset where we are in the table */ 1668 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1669 /* feel out of hs */ 1670 net->last_hs_used = 0; 1671 } else { 1672 for (i = indx; i >= 1; i--) { 1673 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 1674 break; 1675 } 1676 } 1677 net->last_hs_used = indx; 1678 } 1679 } 1680 sctp_enforce_cwnd_limit(&stcb->asoc, net); 1681 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1682 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 1683 } 1684 } 1685 1686 static void 1687 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 1688 struct sctp_association *asoc) 1689 { 1690 struct sctp_nets *net; 1691 1692 /* 1693 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 1694 * (net->fast_retran_loss_recovery == 0))) 1695 */ 1696 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1697 if ((asoc->fast_retran_loss_recovery == 0) || 1698 (asoc->sctp_cmt_on_off > 0)) { 1699 /* out of a RFC2582 Fast recovery window? */ 1700 if (net->net_ack > 0) { 1701 /* 1702 * per section 7.2.3, are there any 1703 * destinations that had a fast retransmit 1704 * to them. If so what we need to do is 1705 * adjust ssthresh and cwnd. 1706 */ 1707 struct sctp_tmit_chunk *lchk; 1708 1709 sctp_hs_cwnd_decrease(stcb, net); 1710 1711 lchk = TAILQ_FIRST(&asoc->send_queue); 1712 1713 net->partial_bytes_acked = 0; 1714 /* Turn on fast recovery window */ 1715 asoc->fast_retran_loss_recovery = 1; 1716 if (lchk == NULL) { 1717 /* Mark end of the window */ 1718 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1719 } else { 1720 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1721 } 1722 1723 /* 1724 * CMT fast recovery -- per destination 1725 * recovery variable. 1726 */ 1727 net->fast_retran_loss_recovery = 1; 1728 1729 if (lchk == NULL) { 1730 /* Mark end of the window */ 1731 net->fast_recovery_tsn = asoc->sending_seq - 1; 1732 } else { 1733 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 1734 } 1735 1736 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1737 stcb->sctp_ep, stcb, net, 1738 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_2); 1739 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1740 stcb->sctp_ep, stcb, net); 1741 } 1742 } else if (net->net_ack > 0) { 1743 /* 1744 * Mark a peg that we WOULD have done a cwnd 1745 * reduction but RFC2582 prevented this action. 1746 */ 1747 SCTP_STAT_INCR(sctps_fastretransinrtt); 1748 } 1749 } 1750 } 1751 1752 static void 1753 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 1754 struct sctp_association *asoc, 1755 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 1756 { 1757 struct sctp_nets *net; 1758 1759 /******************************/ 1760 /* update cwnd and Early FR */ 1761 /******************************/ 1762 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1763 #ifdef JANA_CMT_FAST_RECOVERY 1764 /* 1765 * CMT fast recovery code. Need to debug. 1766 */ 1767 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1768 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 1769 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 1770 net->will_exit_fast_recovery = 1; 1771 } 1772 } 1773 #endif 1774 /* if nothing was acked on this destination skip it */ 1775 if (net->net_ack == 0) { 1776 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1777 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1778 } 1779 continue; 1780 } 1781 #ifdef JANA_CMT_FAST_RECOVERY 1782 /* 1783 * CMT fast recovery code 1784 */ 1785 /* 1786 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 1787 * && net->will_exit_fast_recovery == 0) { @@@ Do something 1788 * } else if (sctp_cmt_on_off == 0 && 1789 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1790 */ 1791 #endif 1792 1793 if (asoc->fast_retran_loss_recovery && 1794 (will_exit == 0) && 1795 (asoc->sctp_cmt_on_off == 0)) { 1796 /* 1797 * If we are in loss recovery we skip any cwnd 1798 * update 1799 */ 1800 return; 1801 } 1802 /* 1803 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1804 * moved. 1805 */ 1806 if (accum_moved || 1807 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 1808 /* If the cumulative ack moved we can proceed */ 1809 if (net->cwnd <= net->ssthresh) { 1810 /* We are in slow start */ 1811 if (net->flight_size + net->net_ack >= net->cwnd) { 1812 sctp_hs_cwnd_increase(stcb, net); 1813 } else { 1814 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1815 sctp_log_cwnd(stcb, net, net->net_ack, 1816 SCTP_CWND_LOG_NOADV_SS); 1817 } 1818 } 1819 } else { 1820 /* We are in congestion avoidance */ 1821 net->partial_bytes_acked += net->net_ack; 1822 if ((net->flight_size + net->net_ack >= net->cwnd) && 1823 (net->partial_bytes_acked >= net->cwnd)) { 1824 net->partial_bytes_acked -= net->cwnd; 1825 net->cwnd += net->mtu; 1826 sctp_enforce_cwnd_limit(asoc, net); 1827 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1828 sctp_log_cwnd(stcb, net, net->mtu, 1829 SCTP_CWND_LOG_FROM_CA); 1830 } 1831 } else { 1832 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1833 sctp_log_cwnd(stcb, net, net->net_ack, 1834 SCTP_CWND_LOG_NOADV_CA); 1835 } 1836 } 1837 } 1838 } else { 1839 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1840 sctp_log_cwnd(stcb, net, net->mtu, 1841 SCTP_CWND_LOG_NO_CUMACK); 1842 } 1843 } 1844 } 1845 } 1846 1847 /* 1848 * H-TCP congestion control. The algorithm is detailed in: 1849 * R.N.Shorten, D.J.Leith: 1850 * "H-TCP: TCP for high-speed and long-distance networks" 1851 * Proc. PFLDnet, Argonne, 2004. 1852 * http://www.hamilton.ie/net/htcp3.pdf 1853 */ 1854 1855 static int use_rtt_scaling = 1; 1856 static int use_bandwidth_switch = 1; 1857 1858 static inline int 1859 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1860 { 1861 return (seq3 - seq2 >= seq1 - seq2); 1862 } 1863 1864 static inline uint32_t 1865 htcp_cong_time(struct htcp *ca) 1866 { 1867 return (sctp_get_tick_count() - ca->last_cong); 1868 } 1869 1870 static inline uint32_t 1871 htcp_ccount(struct htcp *ca) 1872 { 1873 return (ca->minRTT == 0 ? htcp_cong_time(ca) : htcp_cong_time(ca) / ca->minRTT); 1874 } 1875 1876 static inline void 1877 htcp_reset(struct htcp *ca) 1878 { 1879 ca->undo_last_cong = ca->last_cong; 1880 ca->undo_maxRTT = ca->maxRTT; 1881 ca->undo_old_maxB = ca->old_maxB; 1882 ca->last_cong = sctp_get_tick_count(); 1883 } 1884 1885 #ifdef SCTP_NOT_USED 1886 1887 static uint32_t 1888 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1889 { 1890 net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong; 1891 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT; 1892 net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB; 1893 return (max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->cc_mod.htcp_ca.beta) * net->mtu)); 1894 } 1895 1896 #endif 1897 1898 static inline void 1899 measure_rtt(struct sctp_nets *net) 1900 { 1901 uint32_t srtt = net->lastsa >> SCTP_RTT_SHIFT; 1902 1903 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1904 if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT) 1905 net->cc_mod.htcp_ca.minRTT = srtt; 1906 1907 /* max RTT */ 1908 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) { 1909 if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT) 1910 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT; 1911 if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT + sctp_msecs_to_ticks(20)) 1912 net->cc_mod.htcp_ca.maxRTT = srtt; 1913 } 1914 } 1915 1916 static void 1917 measure_achieved_throughput(struct sctp_nets *net) 1918 { 1919 uint32_t now = sctp_get_tick_count(); 1920 1921 if (net->fast_retran_ip == 0) 1922 net->cc_mod.htcp_ca.bytes_acked = net->net_ack; 1923 1924 if (!use_bandwidth_switch) 1925 return; 1926 1927 /* achieved throughput calculations */ 1928 /* JRS - not 100% sure of this statement */ 1929 if (net->fast_retran_ip == 1) { 1930 net->cc_mod.htcp_ca.bytecount = 0; 1931 net->cc_mod.htcp_ca.lasttime = now; 1932 return; 1933 } 1934 1935 net->cc_mod.htcp_ca.bytecount += net->net_ack; 1936 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)) && 1937 (now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT) && 1938 (net->cc_mod.htcp_ca.minRTT > 0)) { 1939 uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount / net->mtu * hz / (now - net->cc_mod.htcp_ca.lasttime); 1940 1941 if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) { 1942 /* just after backoff */ 1943 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi; 1944 } else { 1945 net->cc_mod.htcp_ca.Bi = (3 * net->cc_mod.htcp_ca.Bi + cur_Bi) / 4; 1946 if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB) 1947 net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi; 1948 if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB) 1949 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB; 1950 } 1951 net->cc_mod.htcp_ca.bytecount = 0; 1952 net->cc_mod.htcp_ca.lasttime = now; 1953 } 1954 } 1955 1956 static inline void 1957 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1958 { 1959 if (use_bandwidth_switch) { 1960 uint32_t maxB = ca->maxB; 1961 uint32_t old_maxB = ca->old_maxB; 1962 1963 ca->old_maxB = ca->maxB; 1964 1965 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1966 ca->beta = BETA_MIN; 1967 ca->modeswitch = 0; 1968 return; 1969 } 1970 } 1971 1972 if (ca->modeswitch && minRTT > sctp_msecs_to_ticks(10) && maxRTT) { 1973 ca->beta = (minRTT << 7) / maxRTT; 1974 if (ca->beta < BETA_MIN) 1975 ca->beta = BETA_MIN; 1976 else if (ca->beta > BETA_MAX) 1977 ca->beta = BETA_MAX; 1978 } else { 1979 ca->beta = BETA_MIN; 1980 ca->modeswitch = 1; 1981 } 1982 } 1983 1984 static inline void 1985 htcp_alpha_update(struct htcp *ca) 1986 { 1987 uint32_t minRTT = ca->minRTT; 1988 uint32_t factor = 1; 1989 uint32_t diff = htcp_cong_time(ca); 1990 1991 if (diff > (uint32_t)hz) { 1992 diff -= hz; 1993 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1994 } 1995 1996 if (use_rtt_scaling && minRTT) { 1997 uint32_t scale = (hz << 3) / (10 * minRTT); 1998 1999 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 2000 * interval [0.5,10]<<3 */ 2001 factor = (factor << 3) / scale; 2002 if (factor != 0) 2003 factor = 1; 2004 } 2005 2006 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 2007 if (ca->alpha != 0) 2008 ca->alpha = ALPHA_BASE; 2009 } 2010 2011 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 2012 * rtt before we adjust our beta to ensure we are working from a consistent 2013 * data. 2014 * 2015 * This function should be called when we hit a congestion event since only at 2016 * that point do we really have a real sense of maxRTT (the queues en route 2017 * were getting just too full now). 2018 */ 2019 static void 2020 htcp_param_update(struct sctp_nets *net) 2021 { 2022 uint32_t minRTT = net->cc_mod.htcp_ca.minRTT; 2023 uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT; 2024 2025 htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT); 2026 htcp_alpha_update(&net->cc_mod.htcp_ca); 2027 2028 /* 2029 * add slowly fading memory for maxRTT to accommodate routing 2030 * changes etc 2031 */ 2032 if (minRTT > 0 && maxRTT > minRTT) 2033 net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 2034 } 2035 2036 static uint32_t 2037 htcp_recalc_ssthresh(struct sctp_nets *net) 2038 { 2039 htcp_param_update(net); 2040 return (max(((net->cwnd / net->mtu * net->cc_mod.htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu)); 2041 } 2042 2043 static void 2044 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 2045 { 2046 /*- 2047 * How to handle these functions? 2048 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 2049 * return; 2050 */ 2051 if (net->cwnd <= net->ssthresh) { 2052 /* We are in slow start */ 2053 if (net->flight_size + net->net_ack >= net->cwnd) { 2054 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 2055 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 2056 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2057 sctp_log_cwnd(stcb, net, net->mtu, 2058 SCTP_CWND_LOG_FROM_SS); 2059 } 2060 2061 } else { 2062 net->cwnd += net->net_ack; 2063 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2064 sctp_log_cwnd(stcb, net, net->net_ack, 2065 SCTP_CWND_LOG_FROM_SS); 2066 } 2067 } 2068 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2069 } else { 2070 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2071 sctp_log_cwnd(stcb, net, net->net_ack, 2072 SCTP_CWND_LOG_NOADV_SS); 2073 } 2074 } 2075 } else { 2076 measure_rtt(net); 2077 2078 /* 2079 * In dangerous area, increase slowly. In theory this is 2080 * net->cwnd += alpha / net->cwnd 2081 */ 2082 /* What is snd_cwnd_cnt?? */ 2083 if (((net->partial_bytes_acked / net->mtu * net->cc_mod.htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 2084 /*- 2085 * Does SCTP have a cwnd clamp? 2086 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 2087 */ 2088 net->cwnd += net->mtu; 2089 net->partial_bytes_acked = 0; 2090 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2091 htcp_alpha_update(&net->cc_mod.htcp_ca); 2092 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2093 sctp_log_cwnd(stcb, net, net->mtu, 2094 SCTP_CWND_LOG_FROM_CA); 2095 } 2096 } else { 2097 net->partial_bytes_acked += net->net_ack; 2098 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2099 sctp_log_cwnd(stcb, net, net->net_ack, 2100 SCTP_CWND_LOG_NOADV_CA); 2101 } 2102 } 2103 2104 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2105 } 2106 } 2107 2108 #ifdef SCTP_NOT_USED 2109 /* Lower bound on congestion window. */ 2110 static uint32_t 2111 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 2112 { 2113 return (net->ssthresh); 2114 } 2115 #endif 2116 2117 static void 2118 htcp_init(struct sctp_nets *net) 2119 { 2120 memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp)); 2121 net->cc_mod.htcp_ca.alpha = ALPHA_BASE; 2122 net->cc_mod.htcp_ca.beta = BETA_MIN; 2123 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2124 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count(); 2125 } 2126 2127 static void 2128 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 2129 { 2130 /* 2131 * We take the max of the burst limit times a MTU or the 2132 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 2133 */ 2134 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 2135 net->ssthresh = stcb->asoc.peers_rwnd; 2136 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2137 htcp_init(net); 2138 2139 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 2140 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 2141 } 2142 } 2143 2144 static void 2145 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 2146 struct sctp_association *asoc, 2147 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit) 2148 { 2149 struct sctp_nets *net; 2150 2151 /******************************/ 2152 /* update cwnd and Early FR */ 2153 /******************************/ 2154 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2155 #ifdef JANA_CMT_FAST_RECOVERY 2156 /* 2157 * CMT fast recovery code. Need to debug. 2158 */ 2159 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 2160 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 2161 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 2162 net->will_exit_fast_recovery = 1; 2163 } 2164 } 2165 #endif 2166 /* if nothing was acked on this destination skip it */ 2167 if (net->net_ack == 0) { 2168 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2169 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 2170 } 2171 continue; 2172 } 2173 #ifdef JANA_CMT_FAST_RECOVERY 2174 /* 2175 * CMT fast recovery code 2176 */ 2177 /* 2178 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 2179 * && net->will_exit_fast_recovery == 0) { @@@ Do something 2180 * } else if (sctp_cmt_on_off == 0 && 2181 * asoc->fast_retran_loss_recovery && will_exit == 0) { 2182 */ 2183 #endif 2184 2185 if (asoc->fast_retran_loss_recovery && 2186 will_exit == 0 && 2187 (asoc->sctp_cmt_on_off == 0)) { 2188 /* 2189 * If we are in loss recovery we skip any cwnd 2190 * update 2191 */ 2192 return; 2193 } 2194 /* 2195 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 2196 * moved. 2197 */ 2198 if (accum_moved || 2199 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 2200 htcp_cong_avoid(stcb, net); 2201 measure_achieved_throughput(net); 2202 } else { 2203 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2204 sctp_log_cwnd(stcb, net, net->mtu, 2205 SCTP_CWND_LOG_NO_CUMACK); 2206 } 2207 } 2208 } 2209 } 2210 2211 static void 2212 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 2213 struct sctp_association *asoc) 2214 { 2215 struct sctp_nets *net; 2216 2217 /* 2218 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 2219 * (net->fast_retran_loss_recovery == 0))) 2220 */ 2221 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2222 if ((asoc->fast_retran_loss_recovery == 0) || 2223 (asoc->sctp_cmt_on_off > 0)) { 2224 /* out of a RFC2582 Fast recovery window? */ 2225 if (net->net_ack > 0) { 2226 /* 2227 * per section 7.2.3, are there any 2228 * destinations that had a fast retransmit 2229 * to them. If so what we need to do is 2230 * adjust ssthresh and cwnd. 2231 */ 2232 struct sctp_tmit_chunk *lchk; 2233 int old_cwnd = net->cwnd; 2234 2235 /* JRS - reset as if state were changed */ 2236 htcp_reset(&net->cc_mod.htcp_ca); 2237 net->ssthresh = htcp_recalc_ssthresh(net); 2238 net->cwnd = net->ssthresh; 2239 sctp_enforce_cwnd_limit(asoc, net); 2240 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2241 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 2242 SCTP_CWND_LOG_FROM_FR); 2243 } 2244 lchk = TAILQ_FIRST(&asoc->send_queue); 2245 2246 net->partial_bytes_acked = 0; 2247 /* Turn on fast recovery window */ 2248 asoc->fast_retran_loss_recovery = 1; 2249 if (lchk == NULL) { 2250 /* Mark end of the window */ 2251 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 2252 } else { 2253 asoc->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2254 } 2255 2256 /* 2257 * CMT fast recovery -- per destination 2258 * recovery variable. 2259 */ 2260 net->fast_retran_loss_recovery = 1; 2261 2262 if (lchk == NULL) { 2263 /* Mark end of the window */ 2264 net->fast_recovery_tsn = asoc->sending_seq - 1; 2265 } else { 2266 net->fast_recovery_tsn = lchk->rec.data.tsn - 1; 2267 } 2268 2269 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 2270 stcb->sctp_ep, stcb, net, 2271 SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_3); 2272 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 2273 stcb->sctp_ep, stcb, net); 2274 } 2275 } else if (net->net_ack > 0) { 2276 /* 2277 * Mark a peg that we WOULD have done a cwnd 2278 * reduction but RFC2582 prevented this action. 2279 */ 2280 SCTP_STAT_INCR(sctps_fastretransinrtt); 2281 } 2282 } 2283 } 2284 2285 static void 2286 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 2287 struct sctp_nets *net) 2288 { 2289 int old_cwnd = net->cwnd; 2290 2291 /* JRS - reset as if the state were being changed to timeout */ 2292 htcp_reset(&net->cc_mod.htcp_ca); 2293 net->ssthresh = htcp_recalc_ssthresh(net); 2294 net->cwnd = net->mtu; 2295 net->partial_bytes_acked = 0; 2296 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2297 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 2298 } 2299 } 2300 2301 static void 2302 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 2303 struct sctp_nets *net, int in_window, int num_pkt_lost SCTP_UNUSED) 2304 { 2305 int old_cwnd; 2306 2307 old_cwnd = net->cwnd; 2308 2309 /* JRS - reset hctp as if state changed */ 2310 if (in_window == 0) { 2311 htcp_reset(&net->cc_mod.htcp_ca); 2312 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 2313 net->ssthresh = htcp_recalc_ssthresh(net); 2314 if (net->ssthresh < net->mtu) { 2315 net->ssthresh = net->mtu; 2316 /* here back off the timer as well, to slow us down */ 2317 net->RTO <<= 1; 2318 } 2319 net->cwnd = net->ssthresh; 2320 sctp_enforce_cwnd_limit(&stcb->asoc, net); 2321 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2322 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 2323 } 2324 } 2325 } 2326 2327 const struct sctp_cc_functions sctp_cc_functions[] = { 2328 { 2329 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2330 .sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack, 2331 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2332 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2333 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2334 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2335 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2336 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2337 }, 2338 { 2339 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2340 .sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack, 2341 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2342 .sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr, 2343 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2344 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2345 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2346 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2347 }, 2348 { 2349 .sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param, 2350 .sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack, 2351 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2352 .sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr, 2353 .sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout, 2354 .sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo, 2355 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2356 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2357 }, 2358 { 2359 .sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param, 2360 .sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack, 2361 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common, 2362 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2363 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2364 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo, 2365 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2366 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2367 .sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted, 2368 .sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged, 2369 .sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins, 2370 .sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack, 2371 .sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option, 2372 .sctp_rtt_calculated = sctp_rtt_rtcc_calculated 2373 } 2374 }; 2375