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