1 /*- 2 * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved. 3 * Copyright (c) 2008-2011, by Randall Stewart. All rights reserved. 4 * Copyright (c) 2008-2011, 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 <netinet/sctp_os.h> 34 #include <netinet/sctp_var.h> 35 #include <netinet/sctp_sysctl.h> 36 #include <netinet/sctp_pcb.h> 37 #include <netinet/sctp_header.h> 38 #include <netinet/sctputil.h> 39 #include <netinet/sctp_output.h> 40 #include <netinet/sctp_input.h> 41 #include <netinet/sctp_indata.h> 42 #include <netinet/sctp_uio.h> 43 #include <netinet/sctp_timer.h> 44 #include <netinet/sctp_auth.h> 45 #include <netinet/sctp_asconf.h> 46 #include <netinet/sctp_dtrace_declare.h> 47 #include <sys/cdefs.h> 48 __FBSDID("$FreeBSD$"); 49 50 static void 51 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 52 { 53 struct sctp_association *assoc; 54 uint32_t cwnd_in_mtu; 55 56 assoc = &stcb->asoc; 57 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 58 if (cwnd_in_mtu == 0) { 59 /* Using 0 means that the value of RFC 4960 is used. */ 60 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 61 } else { 62 /* 63 * We take the minimum of the burst limit and the initial 64 * congestion window. 65 */ 66 if ((assoc->max_burst > 0) && (cwnd_in_mtu > assoc->max_burst)) 67 cwnd_in_mtu = assoc->max_burst; 68 net->cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 69 } 70 if (stcb->asoc.sctp_cmt_on_off == 2) { 71 /* In case of resource pooling initialize appropriately */ 72 net->cwnd /= assoc->numnets; 73 if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) { 74 net->cwnd = net->mtu - sizeof(struct sctphdr); 75 } 76 } 77 net->ssthresh = assoc->peers_rwnd; 78 79 SDT_PROBE(sctp, cwnd, net, init, 80 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 81 0, net->cwnd); 82 if (SCTP_BASE_SYSCTL(sctp_logging_level) & 83 (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 84 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 85 } 86 } 87 88 static void 89 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb, 90 struct sctp_association *asoc) 91 { 92 struct sctp_nets *net; 93 uint32_t t_ssthresh, t_cwnd; 94 95 /* MT FIXME: Don't compute this over and over again */ 96 t_ssthresh = 0; 97 t_cwnd = 0; 98 if (asoc->sctp_cmt_on_off == 2) { 99 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 100 t_ssthresh += net->ssthresh; 101 t_cwnd += net->cwnd; 102 } 103 } 104 /*- 105 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 106 * (net->fast_retran_loss_recovery == 0))) 107 */ 108 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 109 if ((asoc->fast_retran_loss_recovery == 0) || 110 (asoc->sctp_cmt_on_off > 0)) { 111 /* out of a RFC2582 Fast recovery window? */ 112 if (net->net_ack > 0) { 113 /* 114 * per section 7.2.3, are there any 115 * destinations that had a fast retransmit 116 * to them. If so what we need to do is 117 * adjust ssthresh and cwnd. 118 */ 119 struct sctp_tmit_chunk *lchk; 120 int old_cwnd = net->cwnd; 121 122 if (asoc->sctp_cmt_on_off == 2) { 123 net->ssthresh = (uint32_t) (((uint64_t) 4 * 124 (uint64_t) net->mtu * 125 (uint64_t) net->ssthresh) / 126 (uint64_t) t_ssthresh); 127 if ((net->cwnd > t_cwnd / 2) && 128 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 129 net->ssthresh = net->cwnd - t_cwnd / 2; 130 } 131 if (net->ssthresh < net->mtu) { 132 net->ssthresh = net->mtu; 133 } 134 } else { 135 net->ssthresh = net->cwnd / 2; 136 if (net->ssthresh < (net->mtu * 2)) { 137 net->ssthresh = 2 * net->mtu; 138 } 139 } 140 net->cwnd = net->ssthresh; 141 SDT_PROBE(sctp, cwnd, net, fr, 142 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 143 old_cwnd, net->cwnd); 144 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 145 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 146 SCTP_CWND_LOG_FROM_FR); 147 } 148 lchk = TAILQ_FIRST(&asoc->send_queue); 149 150 net->partial_bytes_acked = 0; 151 /* Turn on fast recovery window */ 152 asoc->fast_retran_loss_recovery = 1; 153 if (lchk == NULL) { 154 /* Mark end of the window */ 155 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 156 } else { 157 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 158 } 159 160 /* 161 * CMT fast recovery -- per destination 162 * recovery variable. 163 */ 164 net->fast_retran_loss_recovery = 1; 165 166 if (lchk == NULL) { 167 /* Mark end of the window */ 168 net->fast_recovery_tsn = asoc->sending_seq - 1; 169 } else { 170 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 171 } 172 173 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 174 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 175 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 176 stcb->sctp_ep, stcb, net); 177 } 178 } else if (net->net_ack > 0) { 179 /* 180 * Mark a peg that we WOULD have done a cwnd 181 * reduction but RFC2582 prevented this action. 182 */ 183 SCTP_STAT_INCR(sctps_fastretransinrtt); 184 } 185 } 186 } 187 188 /* Defines for instantaneous bw decisions */ 189 #define SCTP_INST_LOOSING 1 /* Loosing to other flows */ 190 #define SCTP_INST_NEUTRAL 2 /* Neutral, no indication */ 191 #define SCTP_INST_GAINING 3 /* Gaining, step down possible */ 192 193 194 static int 195 cc_bw_same(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, 196 uint64_t rtt_offset, uint64_t vtag, uint8_t inst_ind) 197 { 198 uint64_t oth, probepoint; 199 200 probepoint = (((uint64_t) net->cwnd) << 32); 201 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 202 /* 203 * rtt increased we don't update bw.. so we don't update the 204 * rtt either. 205 */ 206 /* Probe point 5 */ 207 probepoint |= ((5 << 16) | 1); 208 SDT_PROBE(sctp, cwnd, net, rttvar, 209 vtag, 210 ((net->cc_mod.rtcc.lbw << 32) | nbw), 211 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 212 net->flight_size, 213 probepoint); 214 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 215 if (net->cc_mod.rtcc.last_step_state == 5) 216 net->cc_mod.rtcc.step_cnt++; 217 else 218 net->cc_mod.rtcc.step_cnt = 1; 219 net->cc_mod.rtcc.last_step_state = 5; 220 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 221 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 222 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 223 /* Try a step down */ 224 oth = net->cc_mod.rtcc.vol_reduce; 225 oth <<= 16; 226 oth |= net->cc_mod.rtcc.step_cnt; 227 oth <<= 16; 228 oth |= net->cc_mod.rtcc.last_step_state; 229 SDT_PROBE(sctp, cwnd, net, rttstep, 230 vtag, 231 ((net->cc_mod.rtcc.lbw << 32) | nbw), 232 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 233 oth, 234 probepoint); 235 if (net->cwnd > (4 * net->mtu)) { 236 net->cwnd -= net->mtu; 237 net->cc_mod.rtcc.vol_reduce++; 238 } else { 239 net->cc_mod.rtcc.step_cnt = 0; 240 } 241 } 242 } 243 return (1); 244 } 245 if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 246 /* 247 * rtt decreased, there could be more room. we update both 248 * the bw and the rtt here to lock this in as a good step 249 * down. 250 */ 251 /* Probe point 6 */ 252 probepoint |= ((6 << 16) | 0); 253 SDT_PROBE(sctp, cwnd, net, rttvar, 254 vtag, 255 ((net->cc_mod.rtcc.lbw << 32) | nbw), 256 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 257 net->flight_size, 258 probepoint); 259 if (net->cc_mod.rtcc.steady_step) { 260 oth = net->cc_mod.rtcc.vol_reduce; 261 oth <<= 16; 262 oth |= net->cc_mod.rtcc.step_cnt; 263 oth <<= 16; 264 oth |= net->cc_mod.rtcc.last_step_state; 265 SDT_PROBE(sctp, cwnd, net, rttstep, 266 vtag, 267 ((net->cc_mod.rtcc.lbw << 32) | nbw), 268 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 269 oth, 270 probepoint); 271 if ((net->cc_mod.rtcc.last_step_state == 5) && 272 (net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step)) { 273 /* Step down worked */ 274 net->cc_mod.rtcc.step_cnt = 0; 275 return (1); 276 } else { 277 net->cc_mod.rtcc.last_step_state = 6; 278 net->cc_mod.rtcc.step_cnt = 0; 279 } 280 } 281 net->cc_mod.rtcc.lbw = nbw; 282 net->cc_mod.rtcc.lbw_rtt = net->rtt; 283 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 284 if (inst_ind == SCTP_INST_GAINING) 285 return (1); 286 else if (inst_ind == SCTP_INST_NEUTRAL) 287 return (1); 288 else 289 return (0); 290 } 291 /* 292 * Ok bw and rtt remained the same .. no update to any 293 */ 294 /* Probe point 7 */ 295 probepoint |= ((7 << 16) | net->cc_mod.rtcc.ret_from_eq); 296 SDT_PROBE(sctp, cwnd, net, rttvar, 297 vtag, 298 ((net->cc_mod.rtcc.lbw << 32) | nbw), 299 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 300 net->flight_size, 301 probepoint); 302 303 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) { 304 if (net->cc_mod.rtcc.last_step_state == 5) 305 net->cc_mod.rtcc.step_cnt++; 306 else 307 net->cc_mod.rtcc.step_cnt = 1; 308 net->cc_mod.rtcc.last_step_state = 5; 309 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) || 310 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) && 311 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) { 312 /* Try a step down */ 313 if (net->cwnd > (4 * net->mtu)) { 314 net->cwnd -= net->mtu; 315 net->cc_mod.rtcc.vol_reduce++; 316 return (1); 317 } else { 318 net->cc_mod.rtcc.step_cnt = 0; 319 } 320 } 321 } 322 if (inst_ind == SCTP_INST_GAINING) 323 return (1); 324 else if (inst_ind == SCTP_INST_NEUTRAL) 325 return (1); 326 else 327 return ((int)net->cc_mod.rtcc.ret_from_eq); 328 } 329 330 static int 331 cc_bw_decrease(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset, 332 uint64_t vtag, uint8_t inst_ind) 333 { 334 uint64_t oth, probepoint; 335 336 /* Bandwidth decreased. */ 337 probepoint = (((uint64_t) net->cwnd) << 32); 338 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) { 339 /* rtt increased */ 340 /* Did we add more */ 341 if ((net->cwnd > net->cc_mod.rtcc.cwnd_at_bw_set) && 342 (inst_ind != SCTP_INST_LOOSING)) { 343 /* We caused it maybe.. back off? */ 344 /* PROBE POINT 1 */ 345 probepoint |= ((1 << 16) | 1); 346 SDT_PROBE(sctp, cwnd, net, rttvar, 347 vtag, 348 ((net->cc_mod.rtcc.lbw << 32) | nbw), 349 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 350 net->flight_size, 351 probepoint); 352 353 if (net->cc_mod.rtcc.ret_from_eq) { 354 /* 355 * Switch over to CA if we are less 356 * aggressive 357 */ 358 net->ssthresh = net->cwnd - 1; 359 net->partial_bytes_acked = 0; 360 } 361 return (1); 362 } 363 /* Probe point 2 */ 364 probepoint |= ((2 << 16) | 0); 365 SDT_PROBE(sctp, cwnd, net, rttvar, 366 vtag, 367 ((net->cc_mod.rtcc.lbw << 32) | nbw), 368 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 369 net->flight_size, 370 probepoint); 371 372 /* Someone else - fight for more? */ 373 if (net->cc_mod.rtcc.steady_step) { 374 oth = net->cc_mod.rtcc.vol_reduce; 375 oth <<= 16; 376 oth |= net->cc_mod.rtcc.step_cnt; 377 oth <<= 16; 378 oth |= net->cc_mod.rtcc.last_step_state; 379 SDT_PROBE(sctp, cwnd, net, rttstep, 380 vtag, 381 ((net->cc_mod.rtcc.lbw << 32) | nbw), 382 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 383 oth, 384 probepoint); 385 /* 386 * Did we voluntarily give up some? if so take one 387 * back please 388 */ 389 if ((net->cc_mod.rtcc.vol_reduce) && 390 (inst_ind != SCTP_INST_GAINING)) { 391 net->cwnd += net->mtu; 392 net->cc_mod.rtcc.vol_reduce--; 393 } 394 net->cc_mod.rtcc.last_step_state = 2; 395 net->cc_mod.rtcc.step_cnt = 0; 396 } 397 goto out_decision; 398 } else if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) { 399 /* bw & rtt decreased */ 400 /* Probe point 3 */ 401 probepoint |= ((3 << 16) | 0); 402 SDT_PROBE(sctp, cwnd, net, rttvar, 403 vtag, 404 ((net->cc_mod.rtcc.lbw << 32) | nbw), 405 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 406 net->flight_size, 407 probepoint); 408 if (net->cc_mod.rtcc.steady_step) { 409 oth = net->cc_mod.rtcc.vol_reduce; 410 oth <<= 16; 411 oth |= net->cc_mod.rtcc.step_cnt; 412 oth <<= 16; 413 oth |= net->cc_mod.rtcc.last_step_state; 414 SDT_PROBE(sctp, cwnd, net, rttstep, 415 vtag, 416 ((net->cc_mod.rtcc.lbw << 32) | nbw), 417 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 418 oth, 419 probepoint); 420 if ((net->cc_mod.rtcc.vol_reduce) && 421 (inst_ind != SCTP_INST_GAINING)) { 422 net->cwnd += net->mtu; 423 net->cc_mod.rtcc.vol_reduce--; 424 } 425 net->cc_mod.rtcc.last_step_state = 3; 426 net->cc_mod.rtcc.step_cnt = 0; 427 } 428 goto out_decision; 429 } 430 /* The bw decreased but rtt stayed the same */ 431 /* Probe point 4 */ 432 probepoint |= ((4 << 16) | 0); 433 SDT_PROBE(sctp, cwnd, net, rttvar, 434 vtag, 435 ((net->cc_mod.rtcc.lbw << 32) | nbw), 436 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 437 net->flight_size, 438 probepoint); 439 if (net->cc_mod.rtcc.steady_step) { 440 oth = net->cc_mod.rtcc.vol_reduce; 441 oth <<= 16; 442 oth |= net->cc_mod.rtcc.step_cnt; 443 oth <<= 16; 444 oth |= net->cc_mod.rtcc.last_step_state; 445 SDT_PROBE(sctp, cwnd, net, rttstep, 446 vtag, 447 ((net->cc_mod.rtcc.lbw << 32) | nbw), 448 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 449 oth, 450 probepoint); 451 if ((net->cc_mod.rtcc.vol_reduce) && 452 (inst_ind != SCTP_INST_GAINING)) { 453 net->cwnd += net->mtu; 454 net->cc_mod.rtcc.vol_reduce--; 455 } 456 net->cc_mod.rtcc.last_step_state = 4; 457 net->cc_mod.rtcc.step_cnt = 0; 458 } 459 out_decision: 460 net->cc_mod.rtcc.lbw = nbw; 461 net->cc_mod.rtcc.lbw_rtt = net->rtt; 462 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 463 if (inst_ind == SCTP_INST_GAINING) { 464 return (1); 465 } else { 466 return (0); 467 } 468 } 469 470 static int 471 cc_bw_increase(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, 472 uint64_t vtag, uint8_t inst_ind) 473 { 474 uint64_t oth, probepoint; 475 476 /* 477 * BW increased, so update and return 0, since all actions in our 478 * table say to do the normal CC update. Note that we pay no 479 * attention to the inst_ind since our overall sum is increasing. 480 */ 481 /* PROBE POINT 0 */ 482 probepoint = (((uint64_t) net->cwnd) << 32); 483 SDT_PROBE(sctp, cwnd, net, rttvar, 484 vtag, 485 ((net->cc_mod.rtcc.lbw << 32) | nbw), 486 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 487 net->flight_size, 488 probepoint); 489 if (net->cc_mod.rtcc.steady_step) { 490 oth = net->cc_mod.rtcc.vol_reduce; 491 oth <<= 16; 492 oth |= net->cc_mod.rtcc.step_cnt; 493 oth <<= 16; 494 oth |= net->cc_mod.rtcc.last_step_state; 495 SDT_PROBE(sctp, cwnd, net, rttstep, 496 vtag, 497 ((net->cc_mod.rtcc.lbw << 32) | nbw), 498 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 499 oth, 500 probepoint); 501 net->cc_mod.rtcc.last_step_state = 0; 502 net->cc_mod.rtcc.step_cnt = 0; 503 net->cc_mod.rtcc.vol_reduce = 0; 504 } 505 net->cc_mod.rtcc.lbw = nbw; 506 net->cc_mod.rtcc.lbw_rtt = net->rtt; 507 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd; 508 return (0); 509 } 510 511 /* RTCC Algoritm to limit growth of cwnd, return 512 * true if you want to NOT allow cwnd growth 513 */ 514 static int 515 cc_bw_limit(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw) 516 { 517 uint64_t bw_offset, rtt_offset, rtt, vtag, probepoint; 518 uint64_t bytes_for_this_rtt, inst_bw; 519 uint64_t div, inst_off; 520 int bw_shift; 521 uint8_t inst_ind; 522 int ret; 523 524 /*- 525 * Here we need to see if we want 526 * to limit cwnd growth due to increase 527 * in overall rtt but no increase in bw. 528 * We use the following table to figure 529 * out what we should do. When we return 530 * 0, cc update goes on as planned. If we 531 * return 1, then no cc update happens and cwnd 532 * stays where it is at. 533 * ---------------------------------- 534 * BW | RTT | Action 535 * ********************************* 536 * INC | INC | return 0 537 * ---------------------------------- 538 * INC | SAME | return 0 539 * ---------------------------------- 540 * INC | DECR | return 0 541 * ---------------------------------- 542 * SAME | INC | return 1 543 * ---------------------------------- 544 * SAME | SAME | return 1 545 * ---------------------------------- 546 * SAME | DECR | return 0 547 * ---------------------------------- 548 * DECR | INC | return 0 or 1 based on if we caused. 549 * ---------------------------------- 550 * DECR | SAME | return 0 551 * ---------------------------------- 552 * DECR | DECR | return 0 553 * ---------------------------------- 554 * 555 * We are a bit fuzz on what an increase or 556 * decrease is. For BW it is the same if 557 * it did not change within 1/64th. For 558 * RTT it stayed the same if it did not 559 * change within 1/32nd 560 */ 561 bw_shift = SCTP_BASE_SYSCTL(sctp_rttvar_bw); 562 rtt = stcb->asoc.my_vtag; 563 vtag = (rtt << 32) | (((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) | (stcb->rport); 564 probepoint = (((uint64_t) net->cwnd) << 32); 565 rtt = net->rtt; 566 if (net->cc_mod.rtcc.rtt_set_this_sack) { 567 net->cc_mod.rtcc.rtt_set_this_sack = 0; 568 bytes_for_this_rtt = net->cc_mod.rtcc.bw_bytes - net->cc_mod.rtcc.bw_bytes_at_last_rttc; 569 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 570 if (net->rtt) { 571 div = net->rtt / 1000; 572 if (div) { 573 inst_bw = bytes_for_this_rtt / div; 574 inst_off = inst_bw >> bw_shift; 575 if (inst_bw > nbw) 576 inst_ind = SCTP_INST_GAINING; 577 else if ((inst_bw + inst_off) < nbw) 578 inst_ind = SCTP_INST_LOOSING; 579 else 580 inst_ind = SCTP_INST_NEUTRAL; 581 probepoint |= ((0xb << 16) | inst_ind); 582 } else { 583 inst_bw = bytes_for_this_rtt / (uint64_t) (net->rtt); 584 /* Can't determine do not change */ 585 inst_ind = net->cc_mod.rtcc.last_inst_ind; 586 probepoint |= ((0xc << 16) | inst_ind); 587 } 588 } else { 589 inst_bw = bytes_for_this_rtt; 590 /* Can't determine do not change */ 591 inst_ind = net->cc_mod.rtcc.last_inst_ind; 592 probepoint |= ((0xd << 16) | inst_ind); 593 } 594 SDT_PROBE(sctp, cwnd, net, rttvar, 595 vtag, 596 ((nbw << 32) | inst_bw), 597 ((net->cc_mod.rtcc.lbw_rtt << 32) | rtt), 598 net->flight_size, 599 probepoint); 600 } else { 601 /* No rtt measurement, use last one */ 602 inst_ind = net->cc_mod.rtcc.last_inst_ind; 603 } 604 bw_offset = net->cc_mod.rtcc.lbw >> bw_shift; 605 if (nbw > net->cc_mod.rtcc.lbw + bw_offset) { 606 ret = cc_bw_increase(stcb, net, nbw, vtag, inst_ind); 607 goto out; 608 } 609 rtt_offset = net->cc_mod.rtcc.lbw_rtt >> SCTP_BASE_SYSCTL(sctp_rttvar_rtt); 610 if (nbw < net->cc_mod.rtcc.lbw - bw_offset) { 611 ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, vtag, inst_ind); 612 goto out; 613 } 614 /* 615 * If we reach here then we are in a situation where the bw stayed 616 * the same. 617 */ 618 ret = cc_bw_same(stcb, net, nbw, rtt_offset, vtag, inst_ind); 619 out: 620 net->cc_mod.rtcc.last_inst_ind = inst_ind; 621 return (ret); 622 } 623 624 static void 625 sctp_cwnd_update_after_sack_common(struct sctp_tcb *stcb, 626 struct sctp_association *asoc, 627 int accum_moved, int reneged_all, int will_exit, int use_rtcc) 628 { 629 struct sctp_nets *net; 630 int old_cwnd; 631 uint32_t t_ssthresh, t_cwnd, incr; 632 633 /* MT FIXME: Don't compute this over and over again */ 634 t_ssthresh = 0; 635 t_cwnd = 0; 636 if (stcb->asoc.sctp_cmt_on_off == 2) { 637 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 638 t_ssthresh += net->ssthresh; 639 t_cwnd += net->cwnd; 640 } 641 } 642 /******************************/ 643 /* update cwnd and Early FR */ 644 /******************************/ 645 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 646 647 #ifdef JANA_CMT_FAST_RECOVERY 648 /* 649 * CMT fast recovery code. Need to debug. 650 */ 651 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 652 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 653 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 654 net->will_exit_fast_recovery = 1; 655 } 656 } 657 #endif 658 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 659 /* 660 * So, first of all do we need to have a Early FR 661 * timer running? 662 */ 663 if ((!TAILQ_EMPTY(&asoc->sent_queue) && 664 (net->ref_count > 1) && 665 (net->flight_size < net->cwnd)) || 666 (reneged_all)) { 667 /* 668 * yes, so in this case stop it if its 669 * running, and then restart it. Reneging 670 * all is a special case where we want to 671 * run the Early FR timer and then force the 672 * last few unacked to be sent, causing us 673 * to illicit a sack with gaps to force out 674 * the others. 675 */ 676 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 677 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 678 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 679 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 680 } 681 SCTP_STAT_INCR(sctps_earlyfrstrid); 682 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 683 } else { 684 /* No, stop it if its running */ 685 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 686 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 687 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 688 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 689 } 690 } 691 } 692 /* if nothing was acked on this destination skip it */ 693 if (net->net_ack == 0) { 694 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 695 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 696 } 697 continue; 698 } 699 if (net->net_ack2 > 0) { 700 /* 701 * Karn's rule applies to clearing error count, this 702 * is optional. 703 */ 704 net->error_count = 0; 705 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 706 SCTP_ADDR_NOT_REACHABLE) { 707 /* addr came good */ 708 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 709 net->dest_state |= SCTP_ADDR_REACHABLE; 710 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 711 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 712 /* now was it the primary? if so restore */ 713 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 714 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 715 } 716 } 717 /* 718 * JRS 5/14/07 - If CMT PF is on and the destination 719 * is in PF state, set the destination to active 720 * state and set the cwnd to one or two MTU's based 721 * on whether PF1 or PF2 is being used. 722 * 723 * Should we stop any running T3 timer here? 724 */ 725 if ((asoc->sctp_cmt_on_off > 0) && 726 (asoc->sctp_cmt_pf > 0) && 727 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 728 net->dest_state &= ~SCTP_ADDR_PF; 729 old_cwnd = net->cwnd; 730 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 731 SDT_PROBE(sctp, cwnd, net, ack, 732 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net, 733 old_cwnd, net->cwnd); 734 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 735 net, net->cwnd); 736 /* 737 * Since the cwnd value is explicitly set, 738 * skip the code that updates the cwnd 739 * value. 740 */ 741 goto skip_cwnd_update; 742 } 743 } 744 #ifdef JANA_CMT_FAST_RECOVERY 745 /* 746 * CMT fast recovery code 747 */ 748 /* 749 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 750 * && net->will_exit_fast_recovery == 0) { @@@ Do something 751 * } else if (sctp_cmt_on_off == 0 && 752 * asoc->fast_retran_loss_recovery && will_exit == 0) { 753 */ 754 #endif 755 756 if (asoc->fast_retran_loss_recovery && 757 (will_exit == 0) && 758 (asoc->sctp_cmt_on_off == 0)) { 759 /* 760 * If we are in loss recovery we skip any cwnd 761 * update 762 */ 763 goto skip_cwnd_update; 764 } 765 /* 766 * Did any measurements go on for this network? 767 */ 768 if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) { 769 uint64_t nbw; 770 771 /* 772 * At this point our bw_bytes has been updated by 773 * incoming sack information. 774 * 775 * But our bw may not yet be set. 776 * 777 */ 778 if ((net->cc_mod.rtcc.new_tot_time / 1000) > 0) { 779 nbw = net->cc_mod.rtcc.bw_bytes / (net->cc_mod.rtcc.new_tot_time / 1000); 780 } else { 781 nbw = net->cc_mod.rtcc.bw_bytes; 782 } 783 if (net->cc_mod.rtcc.lbw) { 784 if (cc_bw_limit(stcb, net, nbw)) { 785 /* Hold here, no update */ 786 goto skip_cwnd_update; 787 } 788 } else { 789 uint64_t vtag, probepoint; 790 791 probepoint = (((uint64_t) net->cwnd) << 32); 792 probepoint |= ((0xa << 16) | 0); 793 vtag = (net->rtt << 32) | 794 (((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) | 795 (stcb->rport); 796 797 SDT_PROBE(sctp, cwnd, net, rttvar, 798 vtag, 799 nbw, 800 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 801 net->flight_size, 802 probepoint); 803 net->cc_mod.rtcc.lbw = nbw; 804 net->cc_mod.rtcc.lbw_rtt = net->rtt; 805 if (net->cc_mod.rtcc.rtt_set_this_sack) { 806 net->cc_mod.rtcc.rtt_set_this_sack = 0; 807 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes; 808 } 809 } 810 } 811 /* 812 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 813 * moved. 814 */ 815 if (accum_moved || 816 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 817 /* If the cumulative ack moved we can proceed */ 818 if (net->cwnd <= net->ssthresh) { 819 /* We are in slow start */ 820 if (net->flight_size + net->net_ack >= net->cwnd) { 821 old_cwnd = net->cwnd; 822 if (stcb->asoc.sctp_cmt_on_off == 2) { 823 uint32_t limit; 824 825 limit = (uint32_t) (((uint64_t) net->mtu * 826 (uint64_t) SCTP_BASE_SYSCTL(sctp_L2_abc_variable) * 827 (uint64_t) net->ssthresh) / 828 (uint64_t) t_ssthresh); 829 incr = (uint32_t) (((uint64_t) net->net_ack * 830 (uint64_t) net->ssthresh) / 831 (uint64_t) t_ssthresh); 832 if (incr > limit) { 833 incr = limit; 834 } 835 if (incr == 0) { 836 incr = 1; 837 } 838 } else { 839 incr = net->net_ack; 840 if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) { 841 incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable); 842 } 843 } 844 net->cwnd += incr; 845 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 846 sctp_log_cwnd(stcb, net, incr, 847 SCTP_CWND_LOG_FROM_SS); 848 } 849 SDT_PROBE(sctp, cwnd, net, ack, 850 stcb->asoc.my_vtag, 851 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 852 net, 853 old_cwnd, net->cwnd); 854 } else { 855 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 856 sctp_log_cwnd(stcb, net, net->net_ack, 857 SCTP_CWND_LOG_NOADV_SS); 858 } 859 } 860 } else { 861 /* We are in congestion avoidance */ 862 /* 863 * Add to pba 864 */ 865 net->partial_bytes_acked += net->net_ack; 866 867 if ((net->flight_size + net->net_ack >= net->cwnd) && 868 (net->partial_bytes_acked >= net->cwnd)) { 869 net->partial_bytes_acked -= net->cwnd; 870 old_cwnd = net->cwnd; 871 if (asoc->sctp_cmt_on_off == 2) { 872 incr = (uint32_t) (((uint64_t) net->mtu * 873 (uint64_t) net->ssthresh) / 874 (uint64_t) t_ssthresh); 875 if (incr == 0) { 876 incr = 1; 877 } 878 } else { 879 incr = net->mtu; 880 } 881 net->cwnd += incr; 882 SDT_PROBE(sctp, cwnd, net, ack, 883 stcb->asoc.my_vtag, 884 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 885 net, 886 old_cwnd, net->cwnd); 887 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 888 sctp_log_cwnd(stcb, net, net->mtu, 889 SCTP_CWND_LOG_FROM_CA); 890 } 891 } else { 892 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 893 sctp_log_cwnd(stcb, net, net->net_ack, 894 SCTP_CWND_LOG_NOADV_CA); 895 } 896 } 897 } 898 } else { 899 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 900 sctp_log_cwnd(stcb, net, net->mtu, 901 SCTP_CWND_LOG_NO_CUMACK); 902 } 903 } 904 skip_cwnd_update: 905 /* 906 * NOW, according to Karn's rule do we need to restore the 907 * RTO timer back? Check our net_ack2. If not set then we 908 * have a ambiguity.. i.e. all data ack'd was sent to more 909 * than one place. 910 */ 911 if (net->net_ack2) { 912 /* restore any doubled timers */ 913 net->RTO = (net->lastsa >> SCTP_RTT_SHIFT) + net->lastsv; 914 if (net->RTO < stcb->asoc.minrto) { 915 net->RTO = stcb->asoc.minrto; 916 } 917 if (net->RTO > stcb->asoc.maxrto) { 918 net->RTO = stcb->asoc.maxrto; 919 } 920 } 921 } 922 } 923 924 static void 925 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 926 { 927 int old_cwnd = net->cwnd; 928 uint32_t t_ssthresh, t_cwnd; 929 930 /* MT FIXME: Don't compute this over and over again */ 931 t_ssthresh = 0; 932 t_cwnd = 0; 933 if (stcb->asoc.sctp_cmt_on_off == 2) { 934 struct sctp_nets *lnet; 935 936 TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) { 937 t_ssthresh += lnet->ssthresh; 938 t_cwnd += lnet->cwnd; 939 } 940 net->ssthresh = (uint32_t) (((uint64_t) 4 * 941 (uint64_t) net->mtu * 942 (uint64_t) net->ssthresh) / 943 (uint64_t) t_ssthresh); 944 if ((net->cwnd > t_cwnd / 2) && 945 (net->ssthresh < net->cwnd - t_cwnd / 2)) { 946 net->ssthresh = net->cwnd - t_cwnd / 2; 947 } 948 if (net->ssthresh < net->mtu) { 949 net->ssthresh = net->mtu; 950 } 951 } else { 952 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 953 } 954 net->cwnd = net->mtu; 955 net->partial_bytes_acked = 0; 956 SDT_PROBE(sctp, cwnd, net, to, 957 stcb->asoc.my_vtag, 958 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 959 net, 960 old_cwnd, net->cwnd); 961 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 962 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 963 } 964 } 965 966 static void 967 sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net, 968 int in_window, int num_pkt_lost, int use_rtcc) 969 { 970 int old_cwnd = net->cwnd; 971 972 if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) { 973 /* Data center Congestion Control */ 974 if (in_window == 0) { 975 /* 976 * Go to CA with the cwnd at the point we sent the 977 * TSN that was marked with a CE. 978 */ 979 if (net->ecn_prev_cwnd < net->cwnd) { 980 /* Restore to prev cwnd */ 981 net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost); 982 } else { 983 /* Just cut in 1/2 */ 984 net->cwnd /= 2; 985 } 986 /* Drop to CA */ 987 net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu); 988 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 989 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 990 } 991 } else { 992 /* 993 * Further tuning down required over the drastic 994 * orginal cut 995 */ 996 net->ssthresh -= (net->mtu * num_pkt_lost); 997 net->cwnd -= (net->mtu * num_pkt_lost); 998 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 999 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1000 } 1001 } 1002 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1003 } else { 1004 if (in_window == 0) { 1005 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1006 net->ssthresh = net->cwnd / 2; 1007 if (net->ssthresh < net->mtu) { 1008 net->ssthresh = net->mtu; 1009 /* 1010 * here back off the timer as well, to slow 1011 * us down 1012 */ 1013 net->RTO <<= 1; 1014 } 1015 net->cwnd = net->ssthresh; 1016 SDT_PROBE(sctp, cwnd, net, ecn, 1017 stcb->asoc.my_vtag, 1018 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1019 net, 1020 old_cwnd, net->cwnd); 1021 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1022 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1023 } 1024 } 1025 } 1026 1027 } 1028 1029 static void 1030 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 1031 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 1032 uint32_t * bottle_bw, uint32_t * on_queue) 1033 { 1034 uint32_t bw_avail; 1035 int rtt; 1036 unsigned int incr; 1037 int old_cwnd = net->cwnd; 1038 1039 /* need real RTT in msd for this calc */ 1040 rtt = net->rtt / 1000; 1041 /* get bottle neck bw */ 1042 *bottle_bw = ntohl(cp->bottle_bw); 1043 /* and whats on queue */ 1044 *on_queue = ntohl(cp->current_onq); 1045 /* 1046 * adjust the on-queue if our flight is more it could be that the 1047 * router has not yet gotten data "in-flight" to it 1048 */ 1049 if (*on_queue < net->flight_size) 1050 *on_queue = net->flight_size; 1051 /* calculate the available space */ 1052 bw_avail = (*bottle_bw * rtt) / 1000; 1053 if (bw_avail > *bottle_bw) { 1054 /* 1055 * Cap the growth to no more than the bottle neck. This can 1056 * happen as RTT slides up due to queues. It also means if 1057 * you have more than a 1 second RTT with a empty queue you 1058 * will be limited to the bottle_bw per second no matter if 1059 * other points have 1/2 the RTT and you could get more 1060 * out... 1061 */ 1062 bw_avail = *bottle_bw; 1063 } 1064 if (*on_queue > bw_avail) { 1065 /* 1066 * No room for anything else don't allow anything else to be 1067 * "added to the fire". 1068 */ 1069 int seg_inflight, seg_onqueue, my_portion; 1070 1071 net->partial_bytes_acked = 0; 1072 1073 /* how much are we over queue size? */ 1074 incr = *on_queue - bw_avail; 1075 if (stcb->asoc.seen_a_sack_this_pkt) { 1076 /* 1077 * undo any cwnd adjustment that the sack might have 1078 * made 1079 */ 1080 net->cwnd = net->prev_cwnd; 1081 } 1082 /* Now how much of that is mine? */ 1083 seg_inflight = net->flight_size / net->mtu; 1084 seg_onqueue = *on_queue / net->mtu; 1085 my_portion = (incr * seg_inflight) / seg_onqueue; 1086 1087 /* Have I made an adjustment already */ 1088 if (net->cwnd > net->flight_size) { 1089 /* 1090 * for this flight I made an adjustment we need to 1091 * decrease the portion by a share our previous 1092 * adjustment. 1093 */ 1094 int diff_adj; 1095 1096 diff_adj = net->cwnd - net->flight_size; 1097 if (diff_adj > my_portion) 1098 my_portion = 0; 1099 else 1100 my_portion -= diff_adj; 1101 } 1102 /* 1103 * back down to the previous cwnd (assume we have had a sack 1104 * before this packet). minus what ever portion of the 1105 * overage is my fault. 1106 */ 1107 net->cwnd -= my_portion; 1108 1109 /* we will NOT back down more than 1 MTU */ 1110 if (net->cwnd <= net->mtu) { 1111 net->cwnd = net->mtu; 1112 } 1113 /* force into CA */ 1114 net->ssthresh = net->cwnd - 1; 1115 } else { 1116 /* 1117 * Take 1/4 of the space left or max burst up .. whichever 1118 * is less. 1119 */ 1120 incr = (bw_avail - *on_queue) >> 2; 1121 if ((stcb->asoc.max_burst > 0) && 1122 (stcb->asoc.max_burst * net->mtu < incr)) { 1123 incr = stcb->asoc.max_burst * net->mtu; 1124 } 1125 net->cwnd += incr; 1126 } 1127 if (net->cwnd > bw_avail) { 1128 /* We can't exceed the pipe size */ 1129 net->cwnd = bw_avail; 1130 } 1131 if (net->cwnd < net->mtu) { 1132 /* We always have 1 MTU */ 1133 net->cwnd = net->mtu; 1134 } 1135 if (net->cwnd - old_cwnd != 0) { 1136 /* log only changes */ 1137 SDT_PROBE(sctp, cwnd, net, pd, 1138 stcb->asoc.my_vtag, 1139 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1140 net, 1141 old_cwnd, net->cwnd); 1142 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1143 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1144 SCTP_CWND_LOG_FROM_SAT); 1145 } 1146 } 1147 } 1148 1149 static void 1150 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 1151 struct sctp_nets *net, int burst_limit) 1152 { 1153 int old_cwnd = net->cwnd; 1154 1155 if (net->ssthresh < net->cwnd) 1156 net->ssthresh = net->cwnd; 1157 if (burst_limit) { 1158 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 1159 SDT_PROBE(sctp, cwnd, net, bl, 1160 stcb->asoc.my_vtag, 1161 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1162 net, 1163 old_cwnd, net->cwnd); 1164 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1165 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 1166 } 1167 } 1168 } 1169 1170 static void 1171 sctp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 1172 struct sctp_tcb *stcb, struct sctp_nets *net) 1173 { 1174 int old_cwnd = net->cwnd; 1175 1176 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 1177 /* 1178 * make a small adjustment to cwnd and force to CA. 1179 */ 1180 if (net->cwnd > net->mtu) 1181 /* drop down one MTU after sending */ 1182 net->cwnd -= net->mtu; 1183 if (net->cwnd < net->ssthresh) 1184 /* still in SS move to CA */ 1185 net->ssthresh = net->cwnd - 1; 1186 SDT_PROBE(sctp, cwnd, net, fr, 1187 stcb->asoc.my_vtag, 1188 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), 1189 net, 1190 old_cwnd, net->cwnd); 1191 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1192 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 1193 } 1194 } 1195 1196 static void 1197 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1198 struct sctp_association *asoc, 1199 int accum_moved, int reneged_all, int will_exit) 1200 { 1201 /* Passing a zero argument in last disables the rtcc algoritm */ 1202 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0); 1203 } 1204 1205 static void 1206 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1207 int in_window, int num_pkt_lost) 1208 { 1209 /* Passing a zero argument in last disables the rtcc algoritm */ 1210 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0); 1211 } 1212 1213 /* Here starts the RTCCVAR type CC invented by RRS which 1214 * is a slight mod to RFC2581. We reuse a common routine or 1215 * two since these algoritms are so close and need to 1216 * remain the same. 1217 */ 1218 static void 1219 sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, 1220 int in_window, int num_pkt_lost) 1221 { 1222 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1); 1223 } 1224 1225 1226 static 1227 void 1228 sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net, 1229 struct sctp_tmit_chunk *tp1) 1230 { 1231 net->cc_mod.rtcc.bw_bytes += tp1->send_size; 1232 } 1233 1234 static void 1235 sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb, 1236 struct sctp_nets *net) 1237 { 1238 if (net->cc_mod.rtcc.tls_needs_set > 0) { 1239 /* We had a bw measurment going on */ 1240 struct timeval ltls; 1241 1242 SCTP_GETPTIME_TIMEVAL(<ls); 1243 timevalsub(<ls, &net->cc_mod.rtcc.tls); 1244 net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec; 1245 } 1246 } 1247 1248 static void 1249 sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb, 1250 struct sctp_nets *net) 1251 { 1252 uint64_t vtag, probepoint; 1253 1254 if (net->cc_mod.rtcc.lbw) { 1255 /* Clear the old bw.. we went to 0 in-flight */ 1256 vtag = (net->rtt << 32) | (((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) | 1257 (stcb->rport); 1258 probepoint = (((uint64_t) net->cwnd) << 32); 1259 /* Probe point 8 */ 1260 probepoint |= ((8 << 16) | 0); 1261 SDT_PROBE(sctp, cwnd, net, rttvar, 1262 vtag, 1263 ((net->cc_mod.rtcc.lbw << 32) | 0), 1264 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt), 1265 net->flight_size, 1266 probepoint); 1267 net->cc_mod.rtcc.lbw_rtt = 0; 1268 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1269 net->cc_mod.rtcc.lbw = 0; 1270 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1271 net->cc_mod.rtcc.vol_reduce = 0; 1272 net->cc_mod.rtcc.bw_tot_time = 0; 1273 net->cc_mod.rtcc.bw_bytes = 0; 1274 net->cc_mod.rtcc.tls_needs_set = 0; 1275 if (net->cc_mod.rtcc.steady_step) { 1276 net->cc_mod.rtcc.vol_reduce = 0; 1277 net->cc_mod.rtcc.step_cnt = 0; 1278 net->cc_mod.rtcc.last_step_state = 0; 1279 } 1280 if (net->cc_mod.rtcc.ret_from_eq) { 1281 /* less aggressive one - reset cwnd too */ 1282 uint32_t cwnd_in_mtu, cwnd; 1283 1284 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd); 1285 if (cwnd_in_mtu == 0) { 1286 /* 1287 * Using 0 means that the value of RFC 4960 1288 * is used. 1289 */ 1290 cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1291 } else { 1292 /* 1293 * We take the minimum of the burst limit 1294 * and the initial congestion window. 1295 */ 1296 if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst)) 1297 cwnd_in_mtu = stcb->asoc.max_burst; 1298 cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu; 1299 } 1300 if (net->cwnd > cwnd) { 1301 /* 1302 * Only set if we are not a timeout (i.e. 1303 * down to 1 mtu) 1304 */ 1305 net->cwnd = cwnd; 1306 } 1307 } 1308 } 1309 } 1310 1311 static void 1312 sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb, 1313 struct sctp_nets *net) 1314 { 1315 uint64_t vtag, probepoint; 1316 1317 sctp_set_initial_cc_param(stcb, net); 1318 stcb->asoc.use_precise_time = 1; 1319 probepoint = (((uint64_t) net->cwnd) << 32); 1320 probepoint |= ((9 << 16) | 0); 1321 vtag = (net->rtt << 32) | 1322 (((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) | 1323 (stcb->rport); 1324 SDT_PROBE(sctp, cwnd, net, rttvar, 1325 vtag, 1326 0, 1327 0, 1328 0, 1329 probepoint); 1330 net->cc_mod.rtcc.lbw_rtt = 0; 1331 net->cc_mod.rtcc.cwnd_at_bw_set = 0; 1332 net->cc_mod.rtcc.vol_reduce = 0; 1333 net->cc_mod.rtcc.lbw = 0; 1334 net->cc_mod.rtcc.vol_reduce = 0; 1335 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0; 1336 net->cc_mod.rtcc.bw_tot_time = 0; 1337 net->cc_mod.rtcc.bw_bytes = 0; 1338 net->cc_mod.rtcc.tls_needs_set = 0; 1339 net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret); 1340 net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step); 1341 net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn); 1342 net->cc_mod.rtcc.step_cnt = 0; 1343 net->cc_mod.rtcc.last_step_state = 0; 1344 1345 1346 } 1347 1348 static int 1349 sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget, 1350 struct sctp_cc_option *cc_opt) 1351 { 1352 struct sctp_nets *net; 1353 1354 if (setorget == 1) { 1355 /* a set */ 1356 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1357 if ((cc_opt->aid_value.assoc_value != 0) && 1358 (cc_opt->aid_value.assoc_value != 1)) { 1359 return (EINVAL); 1360 } 1361 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1362 net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value; 1363 } 1364 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1365 if ((cc_opt->aid_value.assoc_value != 0) && 1366 (cc_opt->aid_value.assoc_value != 1)) { 1367 return (EINVAL); 1368 } 1369 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1370 net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value; 1371 } 1372 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1373 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { 1374 net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value; 1375 } 1376 } else { 1377 return (EINVAL); 1378 } 1379 } else { 1380 /* a get */ 1381 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) { 1382 net = TAILQ_FIRST(&stcb->asoc.nets); 1383 if (net == NULL) { 1384 return (EFAULT); 1385 } 1386 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.ret_from_eq; 1387 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) { 1388 net = TAILQ_FIRST(&stcb->asoc.nets); 1389 if (net == NULL) { 1390 return (EFAULT); 1391 } 1392 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.use_dccc_ecn; 1393 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) { 1394 net = TAILQ_FIRST(&stcb->asoc.nets); 1395 if (net == NULL) { 1396 return (EFAULT); 1397 } 1398 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step; 1399 } else { 1400 return (EINVAL); 1401 } 1402 } 1403 return (0); 1404 } 1405 1406 static void 1407 sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb, 1408 struct sctp_nets *net) 1409 { 1410 if (net->cc_mod.rtcc.tls_needs_set == 0) { 1411 SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls); 1412 net->cc_mod.rtcc.tls_needs_set = 2; 1413 } 1414 } 1415 1416 static void 1417 sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb, 1418 struct sctp_association *asoc, 1419 int accum_moved, int reneged_all, int will_exit) 1420 { 1421 /* Passing a one argument at the last enables the rtcc algoritm */ 1422 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1); 1423 } 1424 1425 static void 1426 sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb, 1427 struct sctp_nets *net, struct timeval *now) 1428 { 1429 net->cc_mod.rtcc.rtt_set_this_sack = 1; 1430 } 1431 1432 /* Here starts Sally Floyds HS-TCP */ 1433 1434 struct sctp_hs_raise_drop { 1435 int32_t cwnd; 1436 int32_t increase; 1437 int32_t drop_percent; 1438 }; 1439 1440 #define SCTP_HS_TABLE_SIZE 73 1441 1442 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 1443 {38, 1, 50}, /* 0 */ 1444 {118, 2, 44}, /* 1 */ 1445 {221, 3, 41}, /* 2 */ 1446 {347, 4, 38}, /* 3 */ 1447 {495, 5, 37}, /* 4 */ 1448 {663, 6, 35}, /* 5 */ 1449 {851, 7, 34}, /* 6 */ 1450 {1058, 8, 33}, /* 7 */ 1451 {1284, 9, 32}, /* 8 */ 1452 {1529, 10, 31}, /* 9 */ 1453 {1793, 11, 30}, /* 10 */ 1454 {2076, 12, 29}, /* 11 */ 1455 {2378, 13, 28}, /* 12 */ 1456 {2699, 14, 28}, /* 13 */ 1457 {3039, 15, 27}, /* 14 */ 1458 {3399, 16, 27}, /* 15 */ 1459 {3778, 17, 26}, /* 16 */ 1460 {4177, 18, 26}, /* 17 */ 1461 {4596, 19, 25}, /* 18 */ 1462 {5036, 20, 25}, /* 19 */ 1463 {5497, 21, 24}, /* 20 */ 1464 {5979, 22, 24}, /* 21 */ 1465 {6483, 23, 23}, /* 22 */ 1466 {7009, 24, 23}, /* 23 */ 1467 {7558, 25, 22}, /* 24 */ 1468 {8130, 26, 22}, /* 25 */ 1469 {8726, 27, 22}, /* 26 */ 1470 {9346, 28, 21}, /* 27 */ 1471 {9991, 29, 21}, /* 28 */ 1472 {10661, 30, 21}, /* 29 */ 1473 {11358, 31, 20}, /* 30 */ 1474 {12082, 32, 20}, /* 31 */ 1475 {12834, 33, 20}, /* 32 */ 1476 {13614, 34, 19}, /* 33 */ 1477 {14424, 35, 19}, /* 34 */ 1478 {15265, 36, 19}, /* 35 */ 1479 {16137, 37, 19}, /* 36 */ 1480 {17042, 38, 18}, /* 37 */ 1481 {17981, 39, 18}, /* 38 */ 1482 {18955, 40, 18}, /* 39 */ 1483 {19965, 41, 17}, /* 40 */ 1484 {21013, 42, 17}, /* 41 */ 1485 {22101, 43, 17}, /* 42 */ 1486 {23230, 44, 17}, /* 43 */ 1487 {24402, 45, 16}, /* 44 */ 1488 {25618, 46, 16}, /* 45 */ 1489 {26881, 47, 16}, /* 46 */ 1490 {28193, 48, 16}, /* 47 */ 1491 {29557, 49, 15}, /* 48 */ 1492 {30975, 50, 15}, /* 49 */ 1493 {32450, 51, 15}, /* 50 */ 1494 {33986, 52, 15}, /* 51 */ 1495 {35586, 53, 14}, /* 52 */ 1496 {37253, 54, 14}, /* 53 */ 1497 {38992, 55, 14}, /* 54 */ 1498 {40808, 56, 14}, /* 55 */ 1499 {42707, 57, 13}, /* 56 */ 1500 {44694, 58, 13}, /* 57 */ 1501 {46776, 59, 13}, /* 58 */ 1502 {48961, 60, 13}, /* 59 */ 1503 {51258, 61, 13}, /* 60 */ 1504 {53677, 62, 12}, /* 61 */ 1505 {56230, 63, 12}, /* 62 */ 1506 {58932, 64, 12}, /* 63 */ 1507 {61799, 65, 12}, /* 64 */ 1508 {64851, 66, 11}, /* 65 */ 1509 {68113, 67, 11}, /* 66 */ 1510 {71617, 68, 11}, /* 67 */ 1511 {75401, 69, 10}, /* 68 */ 1512 {79517, 70, 10}, /* 69 */ 1513 {84035, 71, 10}, /* 70 */ 1514 {89053, 72, 10}, /* 71 */ 1515 {94717, 73, 9} /* 72 */ 1516 }; 1517 1518 static void 1519 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 1520 { 1521 int cur_val, i, indx, incr; 1522 1523 cur_val = net->cwnd >> 10; 1524 indx = SCTP_HS_TABLE_SIZE - 1; 1525 #ifdef SCTP_DEBUG 1526 printf("HS CC CAlled.\n"); 1527 #endif 1528 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1529 /* normal mode */ 1530 if (net->net_ack > net->mtu) { 1531 net->cwnd += net->mtu; 1532 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1533 sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS); 1534 } 1535 } else { 1536 net->cwnd += net->net_ack; 1537 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1538 sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS); 1539 } 1540 } 1541 } else { 1542 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 1543 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 1544 indx = i; 1545 break; 1546 } 1547 } 1548 net->last_hs_used = indx; 1549 incr = ((sctp_cwnd_adjust[indx].increase) << 10); 1550 net->cwnd += incr; 1551 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1552 sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS); 1553 } 1554 } 1555 } 1556 1557 static void 1558 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 1559 { 1560 int cur_val, i, indx; 1561 int old_cwnd = net->cwnd; 1562 1563 cur_val = net->cwnd >> 10; 1564 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1565 /* normal mode */ 1566 net->ssthresh = net->cwnd / 2; 1567 if (net->ssthresh < (net->mtu * 2)) { 1568 net->ssthresh = 2 * net->mtu; 1569 } 1570 net->cwnd = net->ssthresh; 1571 } else { 1572 /* drop by the proper amount */ 1573 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 1574 sctp_cwnd_adjust[net->last_hs_used].drop_percent); 1575 net->cwnd = net->ssthresh; 1576 /* now where are we */ 1577 indx = net->last_hs_used; 1578 cur_val = net->cwnd >> 10; 1579 /* reset where we are in the table */ 1580 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 1581 /* feel out of hs */ 1582 net->last_hs_used = 0; 1583 } else { 1584 for (i = indx; i >= 1; i--) { 1585 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 1586 break; 1587 } 1588 } 1589 net->last_hs_used = indx; 1590 } 1591 } 1592 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1593 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 1594 } 1595 } 1596 1597 static void 1598 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 1599 struct sctp_association *asoc) 1600 { 1601 struct sctp_nets *net; 1602 1603 /* 1604 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 1605 * (net->fast_retran_loss_recovery == 0))) 1606 */ 1607 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1608 if ((asoc->fast_retran_loss_recovery == 0) || 1609 (asoc->sctp_cmt_on_off > 0)) { 1610 /* out of a RFC2582 Fast recovery window? */ 1611 if (net->net_ack > 0) { 1612 /* 1613 * per section 7.2.3, are there any 1614 * destinations that had a fast retransmit 1615 * to them. If so what we need to do is 1616 * adjust ssthresh and cwnd. 1617 */ 1618 struct sctp_tmit_chunk *lchk; 1619 1620 sctp_hs_cwnd_decrease(stcb, net); 1621 1622 lchk = TAILQ_FIRST(&asoc->send_queue); 1623 1624 net->partial_bytes_acked = 0; 1625 /* Turn on fast recovery window */ 1626 asoc->fast_retran_loss_recovery = 1; 1627 if (lchk == NULL) { 1628 /* Mark end of the window */ 1629 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1630 } else { 1631 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1632 } 1633 1634 /* 1635 * CMT fast recovery -- per destination 1636 * recovery variable. 1637 */ 1638 net->fast_retran_loss_recovery = 1; 1639 1640 if (lchk == NULL) { 1641 /* Mark end of the window */ 1642 net->fast_recovery_tsn = asoc->sending_seq - 1; 1643 } else { 1644 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1645 } 1646 1647 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1648 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 1649 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1650 stcb->sctp_ep, stcb, net); 1651 } 1652 } else if (net->net_ack > 0) { 1653 /* 1654 * Mark a peg that we WOULD have done a cwnd 1655 * reduction but RFC2582 prevented this action. 1656 */ 1657 SCTP_STAT_INCR(sctps_fastretransinrtt); 1658 } 1659 } 1660 } 1661 1662 static void 1663 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 1664 struct sctp_association *asoc, 1665 int accum_moved, int reneged_all, int will_exit) 1666 { 1667 struct sctp_nets *net; 1668 1669 /******************************/ 1670 /* update cwnd and Early FR */ 1671 /******************************/ 1672 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1673 1674 #ifdef JANA_CMT_FAST_RECOVERY 1675 /* 1676 * CMT fast recovery code. Need to debug. 1677 */ 1678 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1679 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 1680 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 1681 net->will_exit_fast_recovery = 1; 1682 } 1683 } 1684 #endif 1685 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 1686 /* 1687 * So, first of all do we need to have a Early FR 1688 * timer running? 1689 */ 1690 if ((!TAILQ_EMPTY(&asoc->sent_queue) && 1691 (net->ref_count > 1) && 1692 (net->flight_size < net->cwnd)) || 1693 (reneged_all)) { 1694 /* 1695 * yes, so in this case stop it if its 1696 * running, and then restart it. Reneging 1697 * all is a special case where we want to 1698 * run the Early FR timer and then force the 1699 * last few unacked to be sent, causing us 1700 * to illicit a sack with gaps to force out 1701 * the others. 1702 */ 1703 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1704 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 1705 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1706 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 1707 } 1708 SCTP_STAT_INCR(sctps_earlyfrstrid); 1709 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 1710 } else { 1711 /* No, stop it if its running */ 1712 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1713 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 1714 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1715 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 1716 } 1717 } 1718 } 1719 /* if nothing was acked on this destination skip it */ 1720 if (net->net_ack == 0) { 1721 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1722 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1723 } 1724 continue; 1725 } 1726 if (net->net_ack2 > 0) { 1727 /* 1728 * Karn's rule applies to clearing error count, this 1729 * is optional. 1730 */ 1731 net->error_count = 0; 1732 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 1733 SCTP_ADDR_NOT_REACHABLE) { 1734 /* addr came good */ 1735 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 1736 net->dest_state |= SCTP_ADDR_REACHABLE; 1737 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 1738 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 1739 /* now was it the primary? if so restore */ 1740 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 1741 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 1742 } 1743 } 1744 /* 1745 * JRS 5/14/07 - If CMT PF is on and the destination 1746 * is in PF state, set the destination to active 1747 * state and set the cwnd to one or two MTU's based 1748 * on whether PF1 or PF2 is being used. 1749 * 1750 * Should we stop any running T3 timer here? 1751 */ 1752 if ((asoc->sctp_cmt_on_off > 0) && 1753 (asoc->sctp_cmt_pf > 0) && 1754 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 1755 net->dest_state &= ~SCTP_ADDR_PF; 1756 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 1757 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1758 net, net->cwnd); 1759 /* 1760 * Since the cwnd value is explicitly set, 1761 * skip the code that updates the cwnd 1762 * value. 1763 */ 1764 goto skip_cwnd_update; 1765 } 1766 } 1767 #ifdef JANA_CMT_FAST_RECOVERY 1768 /* 1769 * CMT fast recovery code 1770 */ 1771 /* 1772 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 1773 * && net->will_exit_fast_recovery == 0) { @@@ Do something 1774 * } else if (sctp_cmt_on_off == 0 && 1775 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1776 */ 1777 #endif 1778 1779 if (asoc->fast_retran_loss_recovery && 1780 (will_exit == 0) && 1781 (asoc->sctp_cmt_on_off == 0)) { 1782 /* 1783 * If we are in loss recovery we skip any cwnd 1784 * update 1785 */ 1786 goto skip_cwnd_update; 1787 } 1788 /* 1789 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1790 * moved. 1791 */ 1792 if (accum_moved || 1793 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 1794 /* If the cumulative ack moved we can proceed */ 1795 if (net->cwnd <= net->ssthresh) { 1796 /* We are in slow start */ 1797 if (net->flight_size + net->net_ack >= net->cwnd) { 1798 1799 sctp_hs_cwnd_increase(stcb, net); 1800 1801 } else { 1802 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1803 sctp_log_cwnd(stcb, net, net->net_ack, 1804 SCTP_CWND_LOG_NOADV_SS); 1805 } 1806 } 1807 } else { 1808 /* We are in congestion avoidance */ 1809 net->partial_bytes_acked += net->net_ack; 1810 if ((net->flight_size + net->net_ack >= net->cwnd) && 1811 (net->partial_bytes_acked >= net->cwnd)) { 1812 net->partial_bytes_acked -= net->cwnd; 1813 net->cwnd += net->mtu; 1814 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1815 sctp_log_cwnd(stcb, net, net->mtu, 1816 SCTP_CWND_LOG_FROM_CA); 1817 } 1818 } else { 1819 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1820 sctp_log_cwnd(stcb, net, net->net_ack, 1821 SCTP_CWND_LOG_NOADV_CA); 1822 } 1823 } 1824 } 1825 } else { 1826 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1827 sctp_log_cwnd(stcb, net, net->mtu, 1828 SCTP_CWND_LOG_NO_CUMACK); 1829 } 1830 } 1831 skip_cwnd_update: 1832 /* 1833 * NOW, according to Karn's rule do we need to restore the 1834 * RTO timer back? Check our net_ack2. If not set then we 1835 * have a ambiguity.. i.e. all data ack'd was sent to more 1836 * than one place. 1837 */ 1838 if (net->net_ack2) { 1839 /* restore any doubled timers */ 1840 net->RTO = (net->lastsa >> SCTP_RTT_SHIFT) + net->lastsv; 1841 if (net->RTO < stcb->asoc.minrto) { 1842 net->RTO = stcb->asoc.minrto; 1843 } 1844 if (net->RTO > stcb->asoc.maxrto) { 1845 net->RTO = stcb->asoc.maxrto; 1846 } 1847 } 1848 } 1849 } 1850 1851 1852 /* 1853 * H-TCP congestion control. The algorithm is detailed in: 1854 * R.N.Shorten, D.J.Leith: 1855 * "H-TCP: TCP for high-speed and long-distance networks" 1856 * Proc. PFLDnet, Argonne, 2004. 1857 * http://www.hamilton.ie/net/htcp3.pdf 1858 */ 1859 1860 1861 static int use_rtt_scaling = 1; 1862 static int use_bandwidth_switch = 1; 1863 1864 static inline int 1865 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1866 { 1867 return seq3 - seq2 >= seq1 - seq2; 1868 } 1869 1870 static inline uint32_t 1871 htcp_cong_time(struct htcp *ca) 1872 { 1873 return sctp_get_tick_count() - ca->last_cong; 1874 } 1875 1876 static inline uint32_t 1877 htcp_ccount(struct htcp *ca) 1878 { 1879 return htcp_cong_time(ca) / ca->minRTT; 1880 } 1881 1882 static inline void 1883 htcp_reset(struct htcp *ca) 1884 { 1885 ca->undo_last_cong = ca->last_cong; 1886 ca->undo_maxRTT = ca->maxRTT; 1887 ca->undo_old_maxB = ca->old_maxB; 1888 ca->last_cong = sctp_get_tick_count(); 1889 } 1890 1891 #ifdef SCTP_NOT_USED 1892 1893 static uint32_t 1894 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1895 { 1896 net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong; 1897 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT; 1898 net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB; 1899 return max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->cc_mod.htcp_ca.beta) * net->mtu); 1900 } 1901 1902 #endif 1903 1904 static inline void 1905 measure_rtt(struct sctp_tcb *stcb, struct sctp_nets *net) 1906 { 1907 uint32_t srtt = net->lastsa >> SCTP_RTT_SHIFT; 1908 1909 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1910 if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT) 1911 net->cc_mod.htcp_ca.minRTT = srtt; 1912 1913 /* max RTT */ 1914 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) { 1915 if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT) 1916 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT; 1917 if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT + MSEC_TO_TICKS(20)) 1918 net->cc_mod.htcp_ca.maxRTT = srtt; 1919 } 1920 } 1921 1922 static void 1923 measure_achieved_throughput(struct sctp_tcb *stcb, struct sctp_nets *net) 1924 { 1925 uint32_t now = sctp_get_tick_count(); 1926 1927 if (net->fast_retran_ip == 0) 1928 net->cc_mod.htcp_ca.bytes_acked = net->net_ack; 1929 1930 if (!use_bandwidth_switch) 1931 return; 1932 1933 /* achieved throughput calculations */ 1934 /* JRS - not 100% sure of this statement */ 1935 if (net->fast_retran_ip == 1) { 1936 net->cc_mod.htcp_ca.bytecount = 0; 1937 net->cc_mod.htcp_ca.lasttime = now; 1938 return; 1939 } 1940 net->cc_mod.htcp_ca.bytecount += net->net_ack; 1941 1942 if (net->cc_mod.htcp_ca.bytecount >= net->cwnd - ((net->cc_mod.htcp_ca.alpha >> 7 ? : 1) * net->mtu) 1943 && now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT 1944 && net->cc_mod.htcp_ca.minRTT > 0) { 1945 uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount / net->mtu * hz / (now - net->cc_mod.htcp_ca.lasttime); 1946 1947 if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) { 1948 /* just after backoff */ 1949 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi; 1950 } else { 1951 net->cc_mod.htcp_ca.Bi = (3 * net->cc_mod.htcp_ca.Bi + cur_Bi) / 4; 1952 if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB) 1953 net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi; 1954 if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB) 1955 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB; 1956 } 1957 net->cc_mod.htcp_ca.bytecount = 0; 1958 net->cc_mod.htcp_ca.lasttime = now; 1959 } 1960 } 1961 1962 static inline void 1963 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1964 { 1965 if (use_bandwidth_switch) { 1966 uint32_t maxB = ca->maxB; 1967 uint32_t old_maxB = ca->old_maxB; 1968 1969 ca->old_maxB = ca->maxB; 1970 1971 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1972 ca->beta = BETA_MIN; 1973 ca->modeswitch = 0; 1974 return; 1975 } 1976 } 1977 if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) { 1978 ca->beta = (minRTT << 7) / maxRTT; 1979 if (ca->beta < BETA_MIN) 1980 ca->beta = BETA_MIN; 1981 else if (ca->beta > BETA_MAX) 1982 ca->beta = BETA_MAX; 1983 } else { 1984 ca->beta = BETA_MIN; 1985 ca->modeswitch = 1; 1986 } 1987 } 1988 1989 static inline void 1990 htcp_alpha_update(struct htcp *ca) 1991 { 1992 uint32_t minRTT = ca->minRTT; 1993 uint32_t factor = 1; 1994 uint32_t diff = htcp_cong_time(ca); 1995 1996 if (diff > (uint32_t) hz) { 1997 diff -= hz; 1998 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1999 } 2000 if (use_rtt_scaling && minRTT) { 2001 uint32_t scale = (hz << 3) / (10 * minRTT); 2002 2003 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 2004 * interval [0.5,10]<<3 */ 2005 factor = (factor << 3) / scale; 2006 if (!factor) 2007 factor = 1; 2008 } 2009 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 2010 if (!ca->alpha) 2011 ca->alpha = ALPHA_BASE; 2012 } 2013 2014 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 2015 * rtt before we adjust our beta to ensure we are working from a consistent 2016 * data. 2017 * 2018 * This function should be called when we hit a congestion event since only at 2019 * that point do we really have a real sense of maxRTT (the queues en route 2020 * were getting just too full now). 2021 */ 2022 static void 2023 htcp_param_update(struct sctp_tcb *stcb, struct sctp_nets *net) 2024 { 2025 uint32_t minRTT = net->cc_mod.htcp_ca.minRTT; 2026 uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT; 2027 2028 htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT); 2029 htcp_alpha_update(&net->cc_mod.htcp_ca); 2030 2031 /* 2032 * add slowly fading memory for maxRTT to accommodate routing 2033 * changes etc 2034 */ 2035 if (minRTT > 0 && maxRTT > minRTT) 2036 net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 2037 } 2038 2039 static uint32_t 2040 htcp_recalc_ssthresh(struct sctp_tcb *stcb, struct sctp_nets *net) 2041 { 2042 htcp_param_update(stcb, net); 2043 return max(((net->cwnd / net->mtu * net->cc_mod.htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu); 2044 } 2045 2046 static void 2047 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 2048 { 2049 /*- 2050 * How to handle these functions? 2051 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 2052 * return; 2053 */ 2054 if (net->cwnd <= net->ssthresh) { 2055 /* We are in slow start */ 2056 if (net->flight_size + net->net_ack >= net->cwnd) { 2057 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 2058 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 2059 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2060 sctp_log_cwnd(stcb, net, net->mtu, 2061 SCTP_CWND_LOG_FROM_SS); 2062 } 2063 } else { 2064 net->cwnd += net->net_ack; 2065 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2066 sctp_log_cwnd(stcb, net, net->net_ack, 2067 SCTP_CWND_LOG_FROM_SS); 2068 } 2069 } 2070 } else { 2071 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2072 sctp_log_cwnd(stcb, net, net->net_ack, 2073 SCTP_CWND_LOG_NOADV_SS); 2074 } 2075 } 2076 } else { 2077 measure_rtt(stcb, net); 2078 2079 /* 2080 * In dangerous area, increase slowly. In theory this is 2081 * net->cwnd += alpha / net->cwnd 2082 */ 2083 /* What is snd_cwnd_cnt?? */ 2084 if (((net->partial_bytes_acked / net->mtu * net->cc_mod.htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 2085 /*- 2086 * Does SCTP have a cwnd clamp? 2087 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 2088 */ 2089 net->cwnd += net->mtu; 2090 net->partial_bytes_acked = 0; 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 2116 #endif 2117 2118 static void 2119 htcp_init(struct sctp_tcb *stcb, struct sctp_nets *net) 2120 { 2121 memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp)); 2122 net->cc_mod.htcp_ca.alpha = ALPHA_BASE; 2123 net->cc_mod.htcp_ca.beta = BETA_MIN; 2124 net->cc_mod.htcp_ca.bytes_acked = net->mtu; 2125 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count(); 2126 } 2127 2128 static void 2129 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 2130 { 2131 /* 2132 * We take the max of the burst limit times a MTU or the 2133 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 2134 */ 2135 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 2136 net->ssthresh = stcb->asoc.peers_rwnd; 2137 htcp_init(stcb, 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, 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 2156 #ifdef JANA_CMT_FAST_RECOVERY 2157 /* 2158 * CMT fast recovery code. Need to debug. 2159 */ 2160 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 2161 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) || 2162 SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) { 2163 net->will_exit_fast_recovery = 1; 2164 } 2165 } 2166 #endif 2167 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 2168 /* 2169 * So, first of all do we need to have a Early FR 2170 * timer running? 2171 */ 2172 if ((!TAILQ_EMPTY(&asoc->sent_queue) && 2173 (net->ref_count > 1) && 2174 (net->flight_size < net->cwnd)) || 2175 (reneged_all)) { 2176 /* 2177 * yes, so in this case stop it if its 2178 * running, and then restart it. Reneging 2179 * all is a special case where we want to 2180 * run the Early FR timer and then force the 2181 * last few unacked to be sent, causing us 2182 * to illicit a sack with gaps to force out 2183 * the others. 2184 */ 2185 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 2186 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 2187 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 2188 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 2189 } 2190 SCTP_STAT_INCR(sctps_earlyfrstrid); 2191 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 2192 } else { 2193 /* No, stop it if its running */ 2194 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 2195 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 2196 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 2197 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 2198 } 2199 } 2200 } 2201 /* if nothing was acked on this destination skip it */ 2202 if (net->net_ack == 0) { 2203 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2204 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 2205 } 2206 continue; 2207 } 2208 if (net->net_ack2 > 0) { 2209 /* 2210 * Karn's rule applies to clearing error count, this 2211 * is optional. 2212 */ 2213 net->error_count = 0; 2214 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 2215 SCTP_ADDR_NOT_REACHABLE) { 2216 /* addr came good */ 2217 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 2218 net->dest_state |= SCTP_ADDR_REACHABLE; 2219 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 2220 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 2221 /* now was it the primary? if so restore */ 2222 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 2223 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 2224 } 2225 } 2226 /* 2227 * JRS 5/14/07 - If CMT PF is on and the destination 2228 * is in PF state, set the destination to active 2229 * state and set the cwnd to one or two MTU's based 2230 * on whether PF1 or PF2 is being used. 2231 * 2232 * Should we stop any running T3 timer here? 2233 */ 2234 if ((asoc->sctp_cmt_on_off > 0) && 2235 (asoc->sctp_cmt_pf > 0) && 2236 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 2237 net->dest_state &= ~SCTP_ADDR_PF; 2238 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 2239 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 2240 net, net->cwnd); 2241 /* 2242 * Since the cwnd value is explicitly set, 2243 * skip the code that updates the cwnd 2244 * value. 2245 */ 2246 goto skip_cwnd_update; 2247 } 2248 } 2249 #ifdef JANA_CMT_FAST_RECOVERY 2250 /* 2251 * CMT fast recovery code 2252 */ 2253 /* 2254 * if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery 2255 * && net->will_exit_fast_recovery == 0) { @@@ Do something 2256 * } else if (sctp_cmt_on_off == 0 && 2257 * asoc->fast_retran_loss_recovery && will_exit == 0) { 2258 */ 2259 #endif 2260 2261 if (asoc->fast_retran_loss_recovery && 2262 will_exit == 0 && 2263 (asoc->sctp_cmt_on_off == 0)) { 2264 /* 2265 * If we are in loss recovery we skip any cwnd 2266 * update 2267 */ 2268 goto skip_cwnd_update; 2269 } 2270 /* 2271 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 2272 * moved. 2273 */ 2274 if (accum_moved || 2275 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) { 2276 htcp_cong_avoid(stcb, net); 2277 measure_achieved_throughput(stcb, net); 2278 } else { 2279 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 2280 sctp_log_cwnd(stcb, net, net->mtu, 2281 SCTP_CWND_LOG_NO_CUMACK); 2282 } 2283 } 2284 skip_cwnd_update: 2285 /* 2286 * NOW, according to Karn's rule do we need to restore the 2287 * RTO timer back? Check our net_ack2. If not set then we 2288 * have a ambiguity.. i.e. all data ack'd was sent to more 2289 * than one place. 2290 */ 2291 if (net->net_ack2) { 2292 /* restore any doubled timers */ 2293 net->RTO = (net->lastsa >> SCTP_RTT_SHIFT) + net->lastsv; 2294 if (net->RTO < stcb->asoc.minrto) { 2295 net->RTO = stcb->asoc.minrto; 2296 } 2297 if (net->RTO > stcb->asoc.maxrto) { 2298 net->RTO = stcb->asoc.maxrto; 2299 } 2300 } 2301 } 2302 } 2303 2304 static void 2305 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 2306 struct sctp_association *asoc) 2307 { 2308 struct sctp_nets *net; 2309 2310 /* 2311 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) && 2312 * (net->fast_retran_loss_recovery == 0))) 2313 */ 2314 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 2315 if ((asoc->fast_retran_loss_recovery == 0) || 2316 (asoc->sctp_cmt_on_off > 0)) { 2317 /* out of a RFC2582 Fast recovery window? */ 2318 if (net->net_ack > 0) { 2319 /* 2320 * per section 7.2.3, are there any 2321 * destinations that had a fast retransmit 2322 * to them. If so what we need to do is 2323 * adjust ssthresh and cwnd. 2324 */ 2325 struct sctp_tmit_chunk *lchk; 2326 int old_cwnd = net->cwnd; 2327 2328 /* JRS - reset as if state were changed */ 2329 htcp_reset(&net->cc_mod.htcp_ca); 2330 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 2331 net->cwnd = net->ssthresh; 2332 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2333 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 2334 SCTP_CWND_LOG_FROM_FR); 2335 } 2336 lchk = TAILQ_FIRST(&asoc->send_queue); 2337 2338 net->partial_bytes_acked = 0; 2339 /* Turn on fast recovery window */ 2340 asoc->fast_retran_loss_recovery = 1; 2341 if (lchk == NULL) { 2342 /* Mark end of the window */ 2343 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 2344 } else { 2345 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 2346 } 2347 2348 /* 2349 * CMT fast recovery -- per destination 2350 * recovery variable. 2351 */ 2352 net->fast_retran_loss_recovery = 1; 2353 2354 if (lchk == NULL) { 2355 /* Mark end of the window */ 2356 net->fast_recovery_tsn = asoc->sending_seq - 1; 2357 } else { 2358 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 2359 } 2360 2361 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 2362 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 2363 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 2364 stcb->sctp_ep, stcb, net); 2365 } 2366 } else if (net->net_ack > 0) { 2367 /* 2368 * Mark a peg that we WOULD have done a cwnd 2369 * reduction but RFC2582 prevented this action. 2370 */ 2371 SCTP_STAT_INCR(sctps_fastretransinrtt); 2372 } 2373 } 2374 } 2375 2376 static void 2377 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 2378 struct sctp_nets *net) 2379 { 2380 int old_cwnd = net->cwnd; 2381 2382 /* JRS - reset as if the state were being changed to timeout */ 2383 htcp_reset(&net->cc_mod.htcp_ca); 2384 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 2385 net->cwnd = net->mtu; 2386 net->partial_bytes_acked = 0; 2387 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2388 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 2389 } 2390 } 2391 2392 static void 2393 sctp_htcp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 2394 struct sctp_tcb *stcb, struct sctp_nets *net) 2395 { 2396 int old_cwnd; 2397 2398 old_cwnd = net->cwnd; 2399 2400 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 2401 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count(); 2402 /* 2403 * make a small adjustment to cwnd and force to CA. 2404 */ 2405 if (net->cwnd > net->mtu) 2406 /* drop down one MTU after sending */ 2407 net->cwnd -= net->mtu; 2408 if (net->cwnd < net->ssthresh) 2409 /* still in SS move to CA */ 2410 net->ssthresh = net->cwnd - 1; 2411 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2412 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 2413 } 2414 } 2415 2416 static void 2417 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 2418 struct sctp_nets *net, int in_window, int num_pkt_lost) 2419 { 2420 int old_cwnd; 2421 2422 old_cwnd = net->cwnd; 2423 2424 /* JRS - reset hctp as if state changed */ 2425 if (in_window == 0) { 2426 htcp_reset(&net->cc_mod.htcp_ca); 2427 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 2428 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 2429 if (net->ssthresh < net->mtu) { 2430 net->ssthresh = net->mtu; 2431 /* here back off the timer as well, to slow us down */ 2432 net->RTO <<= 1; 2433 } 2434 net->cwnd = net->ssthresh; 2435 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 2436 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 2437 } 2438 } 2439 } 2440 2441 struct sctp_cc_functions sctp_cc_functions[] = { 2442 { 2443 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2444 .sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack, 2445 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2446 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2447 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2448 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2449 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2450 .sctp_cwnd_update_after_fr_timer = sctp_cwnd_update_after_fr_timer 2451 }, 2452 { 2453 .sctp_set_initial_cc_param = sctp_set_initial_cc_param, 2454 .sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack, 2455 .sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr, 2456 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2457 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo, 2458 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2459 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2460 .sctp_cwnd_update_after_fr_timer = sctp_cwnd_update_after_fr_timer 2461 }, 2462 { 2463 .sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param, 2464 .sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack, 2465 .sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr, 2466 .sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout, 2467 .sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo, 2468 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2469 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2470 .sctp_cwnd_update_after_fr_timer = sctp_htcp_cwnd_update_after_fr_timer 2471 }, 2472 { 2473 .sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param, 2474 .sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack, 2475 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr, 2476 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout, 2477 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo, 2478 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped, 2479 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output, 2480 .sctp_cwnd_update_after_fr_timer = sctp_cwnd_update_after_fr_timer, 2481 .sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted, 2482 .sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged, 2483 .sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins, 2484 .sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack, 2485 .sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option, 2486 .sctp_rtt_calculated = sctp_rtt_rtcc_calculated 2487 } 2488 }; 2489