1 /*- 2 * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions are met: 6 * 7 * a) Redistributions of source code must retain the above copyright notice, 8 * this list of conditions and the following disclaimer. 9 * 10 * b) Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in 12 * the documentation and/or other materials provided with the distribution. 13 * 14 * c) Neither the name of Cisco Systems, Inc. nor the names of its 15 * contributors may be used to endorse or promote products derived 16 * from this software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 20 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 28 * THE POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31 #include <netinet/sctp_os.h> 32 #include <netinet/sctp_var.h> 33 #include <netinet/sctp_sysctl.h> 34 #include <netinet/sctp_pcb.h> 35 #include <netinet/sctp_header.h> 36 #include <netinet/sctputil.h> 37 #include <netinet/sctp_output.h> 38 #include <netinet/sctp_input.h> 39 #include <netinet/sctp_indata.h> 40 #include <netinet/sctp_uio.h> 41 #include <netinet/sctp_timer.h> 42 #include <netinet/sctp_auth.h> 43 #include <netinet/sctp_asconf.h> 44 #include <netinet/sctp_cc_functions.h> 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 void 48 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 49 { 50 /* 51 * We take the max of the burst limit times a MTU or the 52 * INITIAL_CWND. We then limit this to 4 MTU's of sending. cwnd must 53 * be at least 2 MTU. 54 */ 55 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 56 net->ssthresh = stcb->asoc.peers_rwnd; 57 58 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 59 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 60 } 61 } 62 63 void 64 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb, 65 struct sctp_association *asoc) 66 { 67 struct sctp_nets *net; 68 69 /*- 70 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 71 * (net->fast_retran_loss_recovery == 0))) 72 */ 73 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 74 if ((asoc->fast_retran_loss_recovery == 0) || 75 (asoc->sctp_cmt_on_off == 1)) { 76 /* out of a RFC2582 Fast recovery window? */ 77 if (net->net_ack > 0) { 78 /* 79 * per section 7.2.3, are there any 80 * destinations that had a fast retransmit 81 * to them. If so what we need to do is 82 * adjust ssthresh and cwnd. 83 */ 84 struct sctp_tmit_chunk *lchk; 85 int old_cwnd = net->cwnd; 86 87 net->ssthresh = net->cwnd / 2; 88 if (net->ssthresh < (net->mtu * 2)) { 89 net->ssthresh = 2 * net->mtu; 90 } 91 net->cwnd = net->ssthresh; 92 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 93 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 94 SCTP_CWND_LOG_FROM_FR); 95 } 96 lchk = TAILQ_FIRST(&asoc->send_queue); 97 98 net->partial_bytes_acked = 0; 99 /* Turn on fast recovery window */ 100 asoc->fast_retran_loss_recovery = 1; 101 if (lchk == NULL) { 102 /* Mark end of the window */ 103 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 104 } else { 105 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 106 } 107 108 /* 109 * CMT fast recovery -- per destination 110 * recovery variable. 111 */ 112 net->fast_retran_loss_recovery = 1; 113 114 if (lchk == NULL) { 115 /* Mark end of the window */ 116 net->fast_recovery_tsn = asoc->sending_seq - 1; 117 } else { 118 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 119 } 120 121 /* 122 * Disable Nonce Sum Checking and store the 123 * resync tsn 124 */ 125 asoc->nonce_sum_check = 0; 126 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 127 128 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 129 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 130 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 131 stcb->sctp_ep, stcb, net); 132 } 133 } else if (net->net_ack > 0) { 134 /* 135 * Mark a peg that we WOULD have done a cwnd 136 * reduction but RFC2582 prevented this action. 137 */ 138 SCTP_STAT_INCR(sctps_fastretransinrtt); 139 } 140 } 141 } 142 143 void 144 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 145 struct sctp_association *asoc, 146 int accum_moved, int reneged_all, int will_exit) 147 { 148 struct sctp_nets *net; 149 150 /******************************/ 151 /* update cwnd and Early FR */ 152 /******************************/ 153 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 154 155 #ifdef JANA_CMT_FAST_RECOVERY 156 /* 157 * CMT fast recovery code. Need to debug. 158 */ 159 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 160 if (compare_with_wrap(asoc->last_acked_seq, 161 net->fast_recovery_tsn, MAX_TSN) || 162 (asoc->last_acked_seq == net->fast_recovery_tsn) || 163 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 164 (net->pseudo_cumack == net->fast_recovery_tsn)) { 165 net->will_exit_fast_recovery = 1; 166 } 167 } 168 #endif 169 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 170 /* 171 * So, first of all do we need to have a Early FR 172 * timer running? 173 */ 174 if (((TAILQ_FIRST(&asoc->sent_queue)) && 175 (net->ref_count > 1) && 176 (net->flight_size < net->cwnd)) || 177 (reneged_all)) { 178 /* 179 * yes, so in this case stop it if its 180 * running, and then restart it. Reneging 181 * all is a special case where we want to 182 * run the Early FR timer and then force the 183 * last few unacked to be sent, causing us 184 * to illicit a sack with gaps to force out 185 * the others. 186 */ 187 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 188 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 189 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 190 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 191 } 192 SCTP_STAT_INCR(sctps_earlyfrstrid); 193 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 194 } else { 195 /* No, stop it if its running */ 196 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 197 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 198 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 199 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 200 } 201 } 202 } 203 /* if nothing was acked on this destination skip it */ 204 if (net->net_ack == 0) { 205 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 206 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 207 } 208 continue; 209 } 210 if (net->net_ack2 > 0) { 211 /* 212 * Karn's rule applies to clearing error count, this 213 * is optional. 214 */ 215 net->error_count = 0; 216 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 217 SCTP_ADDR_NOT_REACHABLE) { 218 /* addr came good */ 219 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 220 net->dest_state |= SCTP_ADDR_REACHABLE; 221 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 222 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 223 /* now was it the primary? if so restore */ 224 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 225 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 226 } 227 } 228 /* 229 * JRS 5/14/07 - If CMT PF is on and the destination 230 * is in PF state, set the destination to active 231 * state and set the cwnd to one or two MTU's based 232 * on whether PF1 or PF2 is being used. 233 * 234 * Should we stop any running T3 timer here? 235 */ 236 if ((asoc->sctp_cmt_on_off == 1) && 237 (asoc->sctp_cmt_pf > 0) && 238 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 239 net->dest_state &= ~SCTP_ADDR_PF; 240 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 241 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 242 net, net->cwnd); 243 /* 244 * Since the cwnd value is explicitly set, 245 * skip the code that updates the cwnd 246 * value. 247 */ 248 goto skip_cwnd_update; 249 } 250 } 251 #ifdef JANA_CMT_FAST_RECOVERY 252 /* 253 * CMT fast recovery code 254 */ 255 /* 256 * if (sctp_cmt_on_off == 1 && 257 * net->fast_retran_loss_recovery && 258 * net->will_exit_fast_recovery == 0) { @@@ Do something } 259 * else if (sctp_cmt_on_off == 0 && 260 * asoc->fast_retran_loss_recovery && will_exit == 0) { 261 */ 262 #endif 263 264 if (asoc->fast_retran_loss_recovery && 265 (will_exit == 0) && 266 (asoc->sctp_cmt_on_off == 0)) { 267 /* 268 * If we are in loss recovery we skip any cwnd 269 * update 270 */ 271 goto skip_cwnd_update; 272 } 273 /* 274 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 275 * moved. 276 */ 277 if (accum_moved || 278 ((asoc->sctp_cmt_on_off == 1) && net->new_pseudo_cumack)) { 279 /* If the cumulative ack moved we can proceed */ 280 if (net->cwnd <= net->ssthresh) { 281 /* We are in slow start */ 282 if (net->flight_size + net->net_ack >= net->cwnd) { 283 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 284 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 285 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 286 sctp_log_cwnd(stcb, net, net->mtu, 287 SCTP_CWND_LOG_FROM_SS); 288 } 289 } else { 290 net->cwnd += net->net_ack; 291 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 292 sctp_log_cwnd(stcb, net, net->net_ack, 293 SCTP_CWND_LOG_FROM_SS); 294 } 295 } 296 } else { 297 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 298 sctp_log_cwnd(stcb, net, net->net_ack, 299 SCTP_CWND_LOG_NOADV_SS); 300 } 301 } 302 } else { 303 /* We are in congestion avoidance */ 304 /* 305 * Add to pba 306 */ 307 net->partial_bytes_acked += net->net_ack; 308 309 if ((net->flight_size + net->net_ack >= net->cwnd) && 310 (net->partial_bytes_acked >= net->cwnd)) { 311 net->partial_bytes_acked -= net->cwnd; 312 net->cwnd += net->mtu; 313 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 314 sctp_log_cwnd(stcb, net, net->mtu, 315 SCTP_CWND_LOG_FROM_CA); 316 } 317 } else { 318 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 319 sctp_log_cwnd(stcb, net, net->net_ack, 320 SCTP_CWND_LOG_NOADV_CA); 321 } 322 } 323 } 324 } else { 325 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 326 sctp_log_cwnd(stcb, net, net->mtu, 327 SCTP_CWND_LOG_NO_CUMACK); 328 } 329 } 330 skip_cwnd_update: 331 /* 332 * NOW, according to Karn's rule do we need to restore the 333 * RTO timer back? Check our net_ack2. If not set then we 334 * have a ambiguity.. i.e. all data ack'd was sent to more 335 * than one place. 336 */ 337 if (net->net_ack2) { 338 /* restore any doubled timers */ 339 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 340 if (net->RTO < stcb->asoc.minrto) { 341 net->RTO = stcb->asoc.minrto; 342 } 343 if (net->RTO > stcb->asoc.maxrto) { 344 net->RTO = stcb->asoc.maxrto; 345 } 346 } 347 } 348 } 349 350 void 351 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net) 352 { 353 int old_cwnd = net->cwnd; 354 355 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu); 356 net->cwnd = net->mtu; 357 net->partial_bytes_acked = 0; 358 359 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 360 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 361 } 362 } 363 364 void 365 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net) 366 { 367 int old_cwnd = net->cwnd; 368 369 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 370 net->ssthresh = net->cwnd / 2; 371 if (net->ssthresh < net->mtu) { 372 net->ssthresh = net->mtu; 373 /* here back off the timer as well, to slow us down */ 374 net->RTO <<= 1; 375 } 376 net->cwnd = net->ssthresh; 377 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 378 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 379 } 380 } 381 382 void 383 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 384 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 385 uint32_t * bottle_bw, uint32_t * on_queue) 386 { 387 uint32_t bw_avail; 388 int rtt, incr; 389 int old_cwnd = net->cwnd; 390 391 /* need real RTT for this calc */ 392 rtt = ((net->lastsa >> 2) + net->lastsv) >> 1; 393 /* get bottle neck bw */ 394 *bottle_bw = ntohl(cp->bottle_bw); 395 /* and whats on queue */ 396 *on_queue = ntohl(cp->current_onq); 397 /* 398 * adjust the on-queue if our flight is more it could be that the 399 * router has not yet gotten data "in-flight" to it 400 */ 401 if (*on_queue < net->flight_size) 402 *on_queue = net->flight_size; 403 /* calculate the available space */ 404 bw_avail = (*bottle_bw * rtt) / 1000; 405 if (bw_avail > *bottle_bw) { 406 /* 407 * Cap the growth to no more than the bottle neck. This can 408 * happen as RTT slides up due to queues. It also means if 409 * you have more than a 1 second RTT with a empty queue you 410 * will be limited to the bottle_bw per second no matter if 411 * other points have 1/2 the RTT and you could get more 412 * out... 413 */ 414 bw_avail = *bottle_bw; 415 } 416 if (*on_queue > bw_avail) { 417 /* 418 * No room for anything else don't allow anything else to be 419 * "added to the fire". 420 */ 421 int seg_inflight, seg_onqueue, my_portion; 422 423 net->partial_bytes_acked = 0; 424 425 /* how much are we over queue size? */ 426 incr = *on_queue - bw_avail; 427 if (stcb->asoc.seen_a_sack_this_pkt) { 428 /* 429 * undo any cwnd adjustment that the sack might have 430 * made 431 */ 432 net->cwnd = net->prev_cwnd; 433 } 434 /* Now how much of that is mine? */ 435 seg_inflight = net->flight_size / net->mtu; 436 seg_onqueue = *on_queue / net->mtu; 437 my_portion = (incr * seg_inflight) / seg_onqueue; 438 439 /* Have I made an adjustment already */ 440 if (net->cwnd > net->flight_size) { 441 /* 442 * for this flight I made an adjustment we need to 443 * decrease the portion by a share our previous 444 * adjustment. 445 */ 446 int diff_adj; 447 448 diff_adj = net->cwnd - net->flight_size; 449 if (diff_adj > my_portion) 450 my_portion = 0; 451 else 452 my_portion -= diff_adj; 453 } 454 /* 455 * back down to the previous cwnd (assume we have had a sack 456 * before this packet). minus what ever portion of the 457 * overage is my fault. 458 */ 459 net->cwnd -= my_portion; 460 461 /* we will NOT back down more than 1 MTU */ 462 if (net->cwnd <= net->mtu) { 463 net->cwnd = net->mtu; 464 } 465 /* force into CA */ 466 net->ssthresh = net->cwnd - 1; 467 } else { 468 /* 469 * Take 1/4 of the space left or max burst up .. whichever 470 * is less. 471 */ 472 incr = min((bw_avail - *on_queue) >> 2, 473 stcb->asoc.max_burst * net->mtu); 474 net->cwnd += incr; 475 } 476 if (net->cwnd > bw_avail) { 477 /* We can't exceed the pipe size */ 478 net->cwnd = bw_avail; 479 } 480 if (net->cwnd < net->mtu) { 481 /* We always have 1 MTU */ 482 net->cwnd = net->mtu; 483 } 484 if (net->cwnd - old_cwnd != 0) { 485 /* log only changes */ 486 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 487 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 488 SCTP_CWND_LOG_FROM_SAT); 489 } 490 } 491 } 492 493 void 494 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 495 struct sctp_nets *net, int burst_limit) 496 { 497 int old_cwnd = net->cwnd; 498 499 if (net->ssthresh < net->cwnd) 500 net->ssthresh = net->cwnd; 501 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 502 503 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 504 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 505 } 506 } 507 508 void 509 sctp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 510 struct sctp_tcb *stcb, struct sctp_nets *net) 511 { 512 int old_cwnd = net->cwnd; 513 514 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 515 /* 516 * make a small adjustment to cwnd and force to CA. 517 */ 518 if (net->cwnd > net->mtu) 519 /* drop down one MTU after sending */ 520 net->cwnd -= net->mtu; 521 if (net->cwnd < net->ssthresh) 522 /* still in SS move to CA */ 523 net->ssthresh = net->cwnd - 1; 524 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 525 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 526 } 527 } 528 529 struct sctp_hs_raise_drop { 530 int32_t cwnd; 531 int32_t increase; 532 int32_t drop_percent; 533 }; 534 535 #define SCTP_HS_TABLE_SIZE 73 536 537 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 538 {38, 1, 50}, /* 0 */ 539 {118, 2, 44}, /* 1 */ 540 {221, 3, 41}, /* 2 */ 541 {347, 4, 38}, /* 3 */ 542 {495, 5, 37}, /* 4 */ 543 {663, 6, 35}, /* 5 */ 544 {851, 7, 34}, /* 6 */ 545 {1058, 8, 33}, /* 7 */ 546 {1284, 9, 32}, /* 8 */ 547 {1529, 10, 31}, /* 9 */ 548 {1793, 11, 30}, /* 10 */ 549 {2076, 12, 29}, /* 11 */ 550 {2378, 13, 28}, /* 12 */ 551 {2699, 14, 28}, /* 13 */ 552 {3039, 15, 27}, /* 14 */ 553 {3399, 16, 27}, /* 15 */ 554 {3778, 17, 26}, /* 16 */ 555 {4177, 18, 26}, /* 17 */ 556 {4596, 19, 25}, /* 18 */ 557 {5036, 20, 25}, /* 19 */ 558 {5497, 21, 24}, /* 20 */ 559 {5979, 22, 24}, /* 21 */ 560 {6483, 23, 23}, /* 22 */ 561 {7009, 24, 23}, /* 23 */ 562 {7558, 25, 22}, /* 24 */ 563 {8130, 26, 22}, /* 25 */ 564 {8726, 27, 22}, /* 26 */ 565 {9346, 28, 21}, /* 27 */ 566 {9991, 29, 21}, /* 28 */ 567 {10661, 30, 21}, /* 29 */ 568 {11358, 31, 20}, /* 30 */ 569 {12082, 32, 20}, /* 31 */ 570 {12834, 33, 20}, /* 32 */ 571 {13614, 34, 19}, /* 33 */ 572 {14424, 35, 19}, /* 34 */ 573 {15265, 36, 19}, /* 35 */ 574 {16137, 37, 19}, /* 36 */ 575 {17042, 38, 18}, /* 37 */ 576 {17981, 39, 18}, /* 38 */ 577 {18955, 40, 18}, /* 39 */ 578 {19965, 41, 17}, /* 40 */ 579 {21013, 42, 17}, /* 41 */ 580 {22101, 43, 17}, /* 42 */ 581 {23230, 44, 17}, /* 43 */ 582 {24402, 45, 16}, /* 44 */ 583 {25618, 46, 16}, /* 45 */ 584 {26881, 47, 16}, /* 46 */ 585 {28193, 48, 16}, /* 47 */ 586 {29557, 49, 15}, /* 48 */ 587 {30975, 50, 15}, /* 49 */ 588 {32450, 51, 15}, /* 50 */ 589 {33986, 52, 15}, /* 51 */ 590 {35586, 53, 14}, /* 52 */ 591 {37253, 54, 14}, /* 53 */ 592 {38992, 55, 14}, /* 54 */ 593 {40808, 56, 14}, /* 55 */ 594 {42707, 57, 13}, /* 56 */ 595 {44694, 58, 13}, /* 57 */ 596 {46776, 59, 13}, /* 58 */ 597 {48961, 60, 13}, /* 59 */ 598 {51258, 61, 13}, /* 60 */ 599 {53677, 62, 12}, /* 61 */ 600 {56230, 63, 12}, /* 62 */ 601 {58932, 64, 12}, /* 63 */ 602 {61799, 65, 12}, /* 64 */ 603 {64851, 66, 11}, /* 65 */ 604 {68113, 67, 11}, /* 66 */ 605 {71617, 68, 11}, /* 67 */ 606 {75401, 69, 10}, /* 68 */ 607 {79517, 70, 10}, /* 69 */ 608 {84035, 71, 10}, /* 70 */ 609 {89053, 72, 10}, /* 71 */ 610 {94717, 73, 9} /* 72 */ 611 }; 612 613 static void 614 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 615 { 616 int cur_val, i, indx, incr; 617 618 cur_val = net->cwnd >> 10; 619 indx = SCTP_HS_TABLE_SIZE - 1; 620 #ifdef SCTP_DEBUG 621 printf("HS CC CAlled.\n"); 622 #endif 623 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 624 /* normal mode */ 625 if (net->net_ack > net->mtu) { 626 net->cwnd += net->mtu; 627 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 628 sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS); 629 } 630 } else { 631 net->cwnd += net->net_ack; 632 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 633 sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS); 634 } 635 } 636 } else { 637 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 638 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 639 indx = i; 640 break; 641 } 642 } 643 net->last_hs_used = indx; 644 incr = ((sctp_cwnd_adjust[indx].increase) << 10); 645 net->cwnd += incr; 646 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 647 sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS); 648 } 649 } 650 } 651 652 static void 653 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 654 { 655 int cur_val, i, indx; 656 int old_cwnd = net->cwnd; 657 658 cur_val = net->cwnd >> 10; 659 indx = net->last_hs_used; 660 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 661 /* normal mode */ 662 net->ssthresh = net->cwnd / 2; 663 if (net->ssthresh < (net->mtu * 2)) { 664 net->ssthresh = 2 * net->mtu; 665 } 666 net->cwnd = net->ssthresh; 667 } else { 668 /* drop by the proper amount */ 669 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 670 sctp_cwnd_adjust[net->last_hs_used].drop_percent); 671 net->cwnd = net->ssthresh; 672 /* now where are we */ 673 indx = net->last_hs_used; 674 cur_val = net->cwnd >> 10; 675 /* reset where we are in the table */ 676 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 677 /* feel out of hs */ 678 net->last_hs_used = 0; 679 } else { 680 for (i = indx; i >= 1; i--) { 681 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 682 break; 683 } 684 } 685 net->last_hs_used = indx; 686 } 687 } 688 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 689 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 690 } 691 } 692 693 void 694 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 695 struct sctp_association *asoc) 696 { 697 struct sctp_nets *net; 698 699 /* 700 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 701 * (net->fast_retran_loss_recovery == 0))) 702 */ 703 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 704 if ((asoc->fast_retran_loss_recovery == 0) || 705 (asoc->sctp_cmt_on_off == 1)) { 706 /* out of a RFC2582 Fast recovery window? */ 707 if (net->net_ack > 0) { 708 /* 709 * per section 7.2.3, are there any 710 * destinations that had a fast retransmit 711 * to them. If so what we need to do is 712 * adjust ssthresh and cwnd. 713 */ 714 struct sctp_tmit_chunk *lchk; 715 716 sctp_hs_cwnd_decrease(stcb, net); 717 718 lchk = TAILQ_FIRST(&asoc->send_queue); 719 720 net->partial_bytes_acked = 0; 721 /* Turn on fast recovery window */ 722 asoc->fast_retran_loss_recovery = 1; 723 if (lchk == NULL) { 724 /* Mark end of the window */ 725 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 726 } else { 727 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 728 } 729 730 /* 731 * CMT fast recovery -- per destination 732 * recovery variable. 733 */ 734 net->fast_retran_loss_recovery = 1; 735 736 if (lchk == NULL) { 737 /* Mark end of the window */ 738 net->fast_recovery_tsn = asoc->sending_seq - 1; 739 } else { 740 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 741 } 742 743 /* 744 * Disable Nonce Sum Checking and store the 745 * resync tsn 746 */ 747 asoc->nonce_sum_check = 0; 748 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 749 750 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 751 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 752 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 753 stcb->sctp_ep, stcb, net); 754 } 755 } else if (net->net_ack > 0) { 756 /* 757 * Mark a peg that we WOULD have done a cwnd 758 * reduction but RFC2582 prevented this action. 759 */ 760 SCTP_STAT_INCR(sctps_fastretransinrtt); 761 } 762 } 763 } 764 765 void 766 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 767 struct sctp_association *asoc, 768 int accum_moved, int reneged_all, int will_exit) 769 { 770 struct sctp_nets *net; 771 772 /******************************/ 773 /* update cwnd and Early FR */ 774 /******************************/ 775 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 776 777 #ifdef JANA_CMT_FAST_RECOVERY 778 /* 779 * CMT fast recovery code. Need to debug. 780 */ 781 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 782 if (compare_with_wrap(asoc->last_acked_seq, 783 net->fast_recovery_tsn, MAX_TSN) || 784 (asoc->last_acked_seq == net->fast_recovery_tsn) || 785 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 786 (net->pseudo_cumack == net->fast_recovery_tsn)) { 787 net->will_exit_fast_recovery = 1; 788 } 789 } 790 #endif 791 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 792 /* 793 * So, first of all do we need to have a Early FR 794 * timer running? 795 */ 796 if (((TAILQ_FIRST(&asoc->sent_queue)) && 797 (net->ref_count > 1) && 798 (net->flight_size < net->cwnd)) || 799 (reneged_all)) { 800 /* 801 * yes, so in this case stop it if its 802 * running, and then restart it. Reneging 803 * all is a special case where we want to 804 * run the Early FR timer and then force the 805 * last few unacked to be sent, causing us 806 * to illicit a sack with gaps to force out 807 * the others. 808 */ 809 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 810 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 811 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 812 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 813 } 814 SCTP_STAT_INCR(sctps_earlyfrstrid); 815 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 816 } else { 817 /* No, stop it if its running */ 818 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 819 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 820 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 821 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 822 } 823 } 824 } 825 /* if nothing was acked on this destination skip it */ 826 if (net->net_ack == 0) { 827 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 828 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 829 } 830 continue; 831 } 832 if (net->net_ack2 > 0) { 833 /* 834 * Karn's rule applies to clearing error count, this 835 * is optional. 836 */ 837 net->error_count = 0; 838 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 839 SCTP_ADDR_NOT_REACHABLE) { 840 /* addr came good */ 841 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 842 net->dest_state |= SCTP_ADDR_REACHABLE; 843 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 844 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 845 /* now was it the primary? if so restore */ 846 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 847 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 848 } 849 } 850 /* 851 * JRS 5/14/07 - If CMT PF is on and the destination 852 * is in PF state, set the destination to active 853 * state and set the cwnd to one or two MTU's based 854 * on whether PF1 or PF2 is being used. 855 * 856 * Should we stop any running T3 timer here? 857 */ 858 if ((asoc->sctp_cmt_on_off == 1) && 859 (asoc->sctp_cmt_pf > 0) && 860 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 861 net->dest_state &= ~SCTP_ADDR_PF; 862 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 863 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 864 net, net->cwnd); 865 /* 866 * Since the cwnd value is explicitly set, 867 * skip the code that updates the cwnd 868 * value. 869 */ 870 goto skip_cwnd_update; 871 } 872 } 873 #ifdef JANA_CMT_FAST_RECOVERY 874 /* 875 * CMT fast recovery code 876 */ 877 /* 878 * if (sctp_cmt_on_off == 1 && 879 * net->fast_retran_loss_recovery && 880 * net->will_exit_fast_recovery == 0) { @@@ Do something } 881 * else if (sctp_cmt_on_off == 0 && 882 * asoc->fast_retran_loss_recovery && will_exit == 0) { 883 */ 884 #endif 885 886 if (asoc->fast_retran_loss_recovery && 887 (will_exit == 0) && 888 (asoc->sctp_cmt_on_off == 0)) { 889 /* 890 * If we are in loss recovery we skip any cwnd 891 * update 892 */ 893 goto skip_cwnd_update; 894 } 895 /* 896 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 897 * moved. 898 */ 899 if (accum_moved || 900 ((asoc->sctp_cmt_on_off == 1) && net->new_pseudo_cumack)) { 901 /* If the cumulative ack moved we can proceed */ 902 if (net->cwnd <= net->ssthresh) { 903 /* We are in slow start */ 904 if (net->flight_size + net->net_ack >= net->cwnd) { 905 906 sctp_hs_cwnd_increase(stcb, net); 907 908 } else { 909 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 910 sctp_log_cwnd(stcb, net, net->net_ack, 911 SCTP_CWND_LOG_NOADV_SS); 912 } 913 } 914 } else { 915 /* We are in congestion avoidance */ 916 net->partial_bytes_acked += net->net_ack; 917 if ((net->flight_size + net->net_ack >= net->cwnd) && 918 (net->partial_bytes_acked >= net->cwnd)) { 919 net->partial_bytes_acked -= net->cwnd; 920 net->cwnd += net->mtu; 921 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 922 sctp_log_cwnd(stcb, net, net->mtu, 923 SCTP_CWND_LOG_FROM_CA); 924 } 925 } else { 926 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 927 sctp_log_cwnd(stcb, net, net->net_ack, 928 SCTP_CWND_LOG_NOADV_CA); 929 } 930 } 931 } 932 } else { 933 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 934 sctp_log_cwnd(stcb, net, net->mtu, 935 SCTP_CWND_LOG_NO_CUMACK); 936 } 937 } 938 skip_cwnd_update: 939 /* 940 * NOW, according to Karn's rule do we need to restore the 941 * RTO timer back? Check our net_ack2. If not set then we 942 * have a ambiguity.. i.e. all data ack'd was sent to more 943 * than one place. 944 */ 945 if (net->net_ack2) { 946 /* restore any doubled timers */ 947 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 948 if (net->RTO < stcb->asoc.minrto) { 949 net->RTO = stcb->asoc.minrto; 950 } 951 if (net->RTO > stcb->asoc.maxrto) { 952 net->RTO = stcb->asoc.maxrto; 953 } 954 } 955 } 956 } 957 958 959 /* 960 * H-TCP congestion control. The algorithm is detailed in: 961 * R.N.Shorten, D.J.Leith: 962 * "H-TCP: TCP for high-speed and long-distance networks" 963 * Proc. PFLDnet, Argonne, 2004. 964 * http://www.hamilton.ie/net/htcp3.pdf 965 */ 966 967 968 static int use_rtt_scaling = 1; 969 static int use_bandwidth_switch = 1; 970 971 static inline int 972 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 973 { 974 return seq3 - seq2 >= seq1 - seq2; 975 } 976 977 static inline uint32_t 978 htcp_cong_time(struct htcp *ca) 979 { 980 return sctp_get_tick_count() - ca->last_cong; 981 } 982 983 static inline uint32_t 984 htcp_ccount(struct htcp *ca) 985 { 986 return htcp_cong_time(ca) / ca->minRTT; 987 } 988 989 static inline void 990 htcp_reset(struct htcp *ca) 991 { 992 ca->undo_last_cong = ca->last_cong; 993 ca->undo_maxRTT = ca->maxRTT; 994 ca->undo_old_maxB = ca->old_maxB; 995 ca->last_cong = sctp_get_tick_count(); 996 } 997 998 #ifdef SCTP_NOT_USED 999 1000 static uint32_t 1001 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1002 { 1003 net->htcp_ca.last_cong = net->htcp_ca.undo_last_cong; 1004 net->htcp_ca.maxRTT = net->htcp_ca.undo_maxRTT; 1005 net->htcp_ca.old_maxB = net->htcp_ca.undo_old_maxB; 1006 return max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->htcp_ca.beta) * net->mtu); 1007 } 1008 1009 #endif 1010 1011 static inline void 1012 measure_rtt(struct sctp_tcb *stcb, struct sctp_nets *net) 1013 { 1014 uint32_t srtt = net->lastsa >> 3; 1015 1016 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1017 if (net->htcp_ca.minRTT > srtt || !net->htcp_ca.minRTT) 1018 net->htcp_ca.minRTT = srtt; 1019 1020 /* max RTT */ 1021 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->htcp_ca) > 3) { 1022 if (net->htcp_ca.maxRTT < net->htcp_ca.minRTT) 1023 net->htcp_ca.maxRTT = net->htcp_ca.minRTT; 1024 if (net->htcp_ca.maxRTT < srtt && srtt <= net->htcp_ca.maxRTT + MSEC_TO_TICKS(20)) 1025 net->htcp_ca.maxRTT = srtt; 1026 } 1027 } 1028 1029 static void 1030 measure_achieved_throughput(struct sctp_tcb *stcb, struct sctp_nets *net) 1031 { 1032 uint32_t now = sctp_get_tick_count(); 1033 1034 if (net->fast_retran_ip == 0) 1035 net->htcp_ca.bytes_acked = net->net_ack; 1036 1037 if (!use_bandwidth_switch) 1038 return; 1039 1040 /* achieved throughput calculations */ 1041 /* JRS - not 100% sure of this statement */ 1042 if (net->fast_retran_ip == 1) { 1043 net->htcp_ca.bytecount = 0; 1044 net->htcp_ca.lasttime = now; 1045 return; 1046 } 1047 net->htcp_ca.bytecount += net->net_ack; 1048 1049 if (net->htcp_ca.bytecount >= net->cwnd - ((net->htcp_ca.alpha >> 7 ? : 1) * net->mtu) 1050 && now - net->htcp_ca.lasttime >= net->htcp_ca.minRTT 1051 && net->htcp_ca.minRTT > 0) { 1052 uint32_t cur_Bi = net->htcp_ca.bytecount / net->mtu * hz / (now - net->htcp_ca.lasttime); 1053 1054 if (htcp_ccount(&net->htcp_ca) <= 3) { 1055 /* just after backoff */ 1056 net->htcp_ca.minB = net->htcp_ca.maxB = net->htcp_ca.Bi = cur_Bi; 1057 } else { 1058 net->htcp_ca.Bi = (3 * net->htcp_ca.Bi + cur_Bi) / 4; 1059 if (net->htcp_ca.Bi > net->htcp_ca.maxB) 1060 net->htcp_ca.maxB = net->htcp_ca.Bi; 1061 if (net->htcp_ca.minB > net->htcp_ca.maxB) 1062 net->htcp_ca.minB = net->htcp_ca.maxB; 1063 } 1064 net->htcp_ca.bytecount = 0; 1065 net->htcp_ca.lasttime = now; 1066 } 1067 } 1068 1069 static inline void 1070 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1071 { 1072 if (use_bandwidth_switch) { 1073 uint32_t maxB = ca->maxB; 1074 uint32_t old_maxB = ca->old_maxB; 1075 1076 ca->old_maxB = ca->maxB; 1077 1078 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1079 ca->beta = BETA_MIN; 1080 ca->modeswitch = 0; 1081 return; 1082 } 1083 } 1084 if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) { 1085 ca->beta = (minRTT << 7) / maxRTT; 1086 if (ca->beta < BETA_MIN) 1087 ca->beta = BETA_MIN; 1088 else if (ca->beta > BETA_MAX) 1089 ca->beta = BETA_MAX; 1090 } else { 1091 ca->beta = BETA_MIN; 1092 ca->modeswitch = 1; 1093 } 1094 } 1095 1096 static inline void 1097 htcp_alpha_update(struct htcp *ca) 1098 { 1099 uint32_t minRTT = ca->minRTT; 1100 uint32_t factor = 1; 1101 uint32_t diff = htcp_cong_time(ca); 1102 1103 if (diff > (uint32_t) hz) { 1104 diff -= hz; 1105 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1106 } 1107 if (use_rtt_scaling && minRTT) { 1108 uint32_t scale = (hz << 3) / (10 * minRTT); 1109 1110 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 1111 * interval [0.5,10]<<3 */ 1112 factor = (factor << 3) / scale; 1113 if (!factor) 1114 factor = 1; 1115 } 1116 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 1117 if (!ca->alpha) 1118 ca->alpha = ALPHA_BASE; 1119 } 1120 1121 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 1122 * rtt before we adjust our beta to ensure we are working from a consistent 1123 * data. 1124 * 1125 * This function should be called when we hit a congestion event since only at 1126 * that point do we really have a real sense of maxRTT (the queues en route 1127 * were getting just too full now). 1128 */ 1129 static void 1130 htcp_param_update(struct sctp_tcb *stcb, struct sctp_nets *net) 1131 { 1132 uint32_t minRTT = net->htcp_ca.minRTT; 1133 uint32_t maxRTT = net->htcp_ca.maxRTT; 1134 1135 htcp_beta_update(&net->htcp_ca, minRTT, maxRTT); 1136 htcp_alpha_update(&net->htcp_ca); 1137 1138 /* 1139 * add slowly fading memory for maxRTT to accommodate routing 1140 * changes etc 1141 */ 1142 if (minRTT > 0 && maxRTT > minRTT) 1143 net->htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 1144 } 1145 1146 static uint32_t 1147 htcp_recalc_ssthresh(struct sctp_tcb *stcb, struct sctp_nets *net) 1148 { 1149 htcp_param_update(stcb, net); 1150 return max(((net->cwnd / net->mtu * net->htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu); 1151 } 1152 1153 static void 1154 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 1155 { 1156 /*- 1157 * How to handle these functions? 1158 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 1159 * return; 1160 */ 1161 if (net->cwnd <= net->ssthresh) { 1162 /* We are in slow start */ 1163 if (net->flight_size + net->net_ack >= net->cwnd) { 1164 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) { 1165 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)); 1166 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1167 sctp_log_cwnd(stcb, net, net->mtu, 1168 SCTP_CWND_LOG_FROM_SS); 1169 } 1170 } else { 1171 net->cwnd += net->net_ack; 1172 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1173 sctp_log_cwnd(stcb, net, net->net_ack, 1174 SCTP_CWND_LOG_FROM_SS); 1175 } 1176 } 1177 } else { 1178 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1179 sctp_log_cwnd(stcb, net, net->net_ack, 1180 SCTP_CWND_LOG_NOADV_SS); 1181 } 1182 } 1183 } else { 1184 measure_rtt(stcb, net); 1185 1186 /* 1187 * In dangerous area, increase slowly. In theory this is 1188 * net->cwnd += alpha / net->cwnd 1189 */ 1190 /* What is snd_cwnd_cnt?? */ 1191 if (((net->partial_bytes_acked / net->mtu * net->htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 1192 /*- 1193 * Does SCTP have a cwnd clamp? 1194 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 1195 */ 1196 net->cwnd += net->mtu; 1197 net->partial_bytes_acked = 0; 1198 htcp_alpha_update(&net->htcp_ca); 1199 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1200 sctp_log_cwnd(stcb, net, net->mtu, 1201 SCTP_CWND_LOG_FROM_CA); 1202 } 1203 } else { 1204 net->partial_bytes_acked += net->net_ack; 1205 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1206 sctp_log_cwnd(stcb, net, net->net_ack, 1207 SCTP_CWND_LOG_NOADV_CA); 1208 } 1209 } 1210 1211 net->htcp_ca.bytes_acked = net->mtu; 1212 } 1213 } 1214 1215 #ifdef SCTP_NOT_USED 1216 /* Lower bound on congestion window. */ 1217 static uint32_t 1218 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 1219 { 1220 return net->ssthresh; 1221 } 1222 1223 #endif 1224 1225 static void 1226 htcp_init(struct sctp_tcb *stcb, struct sctp_nets *net) 1227 { 1228 memset(&net->htcp_ca, 0, sizeof(struct htcp)); 1229 net->htcp_ca.alpha = ALPHA_BASE; 1230 net->htcp_ca.beta = BETA_MIN; 1231 net->htcp_ca.bytes_acked = net->mtu; 1232 net->htcp_ca.last_cong = sctp_get_tick_count(); 1233 } 1234 1235 void 1236 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 1237 { 1238 /* 1239 * We take the max of the burst limit times a MTU or the 1240 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 1241 */ 1242 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1243 net->ssthresh = stcb->asoc.peers_rwnd; 1244 htcp_init(stcb, net); 1245 1246 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 1247 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 1248 } 1249 } 1250 1251 void 1252 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1253 struct sctp_association *asoc, 1254 int accum_moved, int reneged_all, int will_exit) 1255 { 1256 struct sctp_nets *net; 1257 1258 /******************************/ 1259 /* update cwnd and Early FR */ 1260 /******************************/ 1261 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1262 1263 #ifdef JANA_CMT_FAST_RECOVERY 1264 /* 1265 * CMT fast recovery code. Need to debug. 1266 */ 1267 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1268 if (compare_with_wrap(asoc->last_acked_seq, 1269 net->fast_recovery_tsn, MAX_TSN) || 1270 (asoc->last_acked_seq == net->fast_recovery_tsn) || 1271 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 1272 (net->pseudo_cumack == net->fast_recovery_tsn)) { 1273 net->will_exit_fast_recovery = 1; 1274 } 1275 } 1276 #endif 1277 if (SCTP_BASE_SYSCTL(sctp_early_fr)) { 1278 /* 1279 * So, first of all do we need to have a Early FR 1280 * timer running? 1281 */ 1282 if (((TAILQ_FIRST(&asoc->sent_queue)) && 1283 (net->ref_count > 1) && 1284 (net->flight_size < net->cwnd)) || 1285 (reneged_all)) { 1286 /* 1287 * yes, so in this case stop it if its 1288 * running, and then restart it. Reneging 1289 * all is a special case where we want to 1290 * run the Early FR timer and then force the 1291 * last few unacked to be sent, causing us 1292 * to illicit a sack with gaps to force out 1293 * the others. 1294 */ 1295 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1296 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 1297 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1298 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 1299 } 1300 SCTP_STAT_INCR(sctps_earlyfrstrid); 1301 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 1302 } else { 1303 /* No, stop it if its running */ 1304 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1305 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 1306 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1307 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 1308 } 1309 } 1310 } 1311 /* if nothing was acked on this destination skip it */ 1312 if (net->net_ack == 0) { 1313 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1314 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1315 } 1316 continue; 1317 } 1318 if (net->net_ack2 > 0) { 1319 /* 1320 * Karn's rule applies to clearing error count, this 1321 * is optional. 1322 */ 1323 net->error_count = 0; 1324 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 1325 SCTP_ADDR_NOT_REACHABLE) { 1326 /* addr came good */ 1327 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 1328 net->dest_state |= SCTP_ADDR_REACHABLE; 1329 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 1330 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 1331 /* now was it the primary? if so restore */ 1332 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 1333 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 1334 } 1335 } 1336 /* 1337 * JRS 5/14/07 - If CMT PF is on and the destination 1338 * is in PF state, set the destination to active 1339 * state and set the cwnd to one or two MTU's based 1340 * on whether PF1 or PF2 is being used. 1341 * 1342 * Should we stop any running T3 timer here? 1343 */ 1344 if ((asoc->sctp_cmt_on_off == 1) && 1345 (asoc->sctp_cmt_pf > 0) && 1346 ((net->dest_state & SCTP_ADDR_PF) == SCTP_ADDR_PF)) { 1347 net->dest_state &= ~SCTP_ADDR_PF; 1348 net->cwnd = net->mtu * asoc->sctp_cmt_pf; 1349 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1350 net, net->cwnd); 1351 /* 1352 * Since the cwnd value is explicitly set, 1353 * skip the code that updates the cwnd 1354 * value. 1355 */ 1356 goto skip_cwnd_update; 1357 } 1358 } 1359 #ifdef JANA_CMT_FAST_RECOVERY 1360 /* 1361 * CMT fast recovery code 1362 */ 1363 /* 1364 * if (sctp_cmt_on_off == 1 && 1365 * net->fast_retran_loss_recovery && 1366 * net->will_exit_fast_recovery == 0) { @@@ Do something } 1367 * else if (sctp_cmt_on_off == 0 && 1368 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1369 */ 1370 #endif 1371 1372 if (asoc->fast_retran_loss_recovery && 1373 will_exit == 0 && 1374 (asoc->sctp_cmt_on_off == 0)) { 1375 /* 1376 * If we are in loss recovery we skip any cwnd 1377 * update 1378 */ 1379 goto skip_cwnd_update; 1380 } 1381 /* 1382 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1383 * moved. 1384 */ 1385 if (accum_moved || 1386 ((asoc->sctp_cmt_on_off == 1) && net->new_pseudo_cumack)) { 1387 htcp_cong_avoid(stcb, net); 1388 measure_achieved_throughput(stcb, net); 1389 } else { 1390 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { 1391 sctp_log_cwnd(stcb, net, net->mtu, 1392 SCTP_CWND_LOG_NO_CUMACK); 1393 } 1394 } 1395 skip_cwnd_update: 1396 /* 1397 * NOW, according to Karn's rule do we need to restore the 1398 * RTO timer back? Check our net_ack2. If not set then we 1399 * have a ambiguity.. i.e. all data ack'd was sent to more 1400 * than one place. 1401 */ 1402 if (net->net_ack2) { 1403 /* restore any doubled timers */ 1404 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 1405 if (net->RTO < stcb->asoc.minrto) { 1406 net->RTO = stcb->asoc.minrto; 1407 } 1408 if (net->RTO > stcb->asoc.maxrto) { 1409 net->RTO = stcb->asoc.maxrto; 1410 } 1411 } 1412 } 1413 } 1414 1415 void 1416 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 1417 struct sctp_association *asoc) 1418 { 1419 struct sctp_nets *net; 1420 1421 /* 1422 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 1423 * (net->fast_retran_loss_recovery == 0))) 1424 */ 1425 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1426 if ((asoc->fast_retran_loss_recovery == 0) || 1427 (asoc->sctp_cmt_on_off == 1)) { 1428 /* out of a RFC2582 Fast recovery window? */ 1429 if (net->net_ack > 0) { 1430 /* 1431 * per section 7.2.3, are there any 1432 * destinations that had a fast retransmit 1433 * to them. If so what we need to do is 1434 * adjust ssthresh and cwnd. 1435 */ 1436 struct sctp_tmit_chunk *lchk; 1437 int old_cwnd = net->cwnd; 1438 1439 /* JRS - reset as if state were changed */ 1440 htcp_reset(&net->htcp_ca); 1441 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1442 net->cwnd = net->ssthresh; 1443 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1444 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1445 SCTP_CWND_LOG_FROM_FR); 1446 } 1447 lchk = TAILQ_FIRST(&asoc->send_queue); 1448 1449 net->partial_bytes_acked = 0; 1450 /* Turn on fast recovery window */ 1451 asoc->fast_retran_loss_recovery = 1; 1452 if (lchk == NULL) { 1453 /* Mark end of the window */ 1454 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1455 } else { 1456 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1457 } 1458 1459 /* 1460 * CMT fast recovery -- per destination 1461 * recovery variable. 1462 */ 1463 net->fast_retran_loss_recovery = 1; 1464 1465 if (lchk == NULL) { 1466 /* Mark end of the window */ 1467 net->fast_recovery_tsn = asoc->sending_seq - 1; 1468 } else { 1469 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1470 } 1471 1472 /* 1473 * Disable Nonce Sum Checking and store the 1474 * resync tsn 1475 */ 1476 asoc->nonce_sum_check = 0; 1477 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 1478 1479 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1480 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 1481 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1482 stcb->sctp_ep, stcb, net); 1483 } 1484 } else if (net->net_ack > 0) { 1485 /* 1486 * Mark a peg that we WOULD have done a cwnd 1487 * reduction but RFC2582 prevented this action. 1488 */ 1489 SCTP_STAT_INCR(sctps_fastretransinrtt); 1490 } 1491 } 1492 } 1493 1494 void 1495 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 1496 struct sctp_nets *net) 1497 { 1498 int old_cwnd = net->cwnd; 1499 1500 /* JRS - reset as if the state were being changed to timeout */ 1501 htcp_reset(&net->htcp_ca); 1502 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1503 net->cwnd = net->mtu; 1504 net->partial_bytes_acked = 0; 1505 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1506 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1507 } 1508 } 1509 1510 void 1511 sctp_htcp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 1512 struct sctp_tcb *stcb, struct sctp_nets *net) 1513 { 1514 int old_cwnd; 1515 1516 old_cwnd = net->cwnd; 1517 1518 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 1519 net->htcp_ca.last_cong = sctp_get_tick_count(); 1520 /* 1521 * make a small adjustment to cwnd and force to CA. 1522 */ 1523 if (net->cwnd > net->mtu) 1524 /* drop down one MTU after sending */ 1525 net->cwnd -= net->mtu; 1526 if (net->cwnd < net->ssthresh) 1527 /* still in SS move to CA */ 1528 net->ssthresh = net->cwnd - 1; 1529 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1530 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 1531 } 1532 } 1533 1534 void 1535 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 1536 struct sctp_nets *net) 1537 { 1538 int old_cwnd; 1539 1540 old_cwnd = net->cwnd; 1541 1542 /* JRS - reset hctp as if state changed */ 1543 htcp_reset(&net->htcp_ca); 1544 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1545 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1546 if (net->ssthresh < net->mtu) { 1547 net->ssthresh = net->mtu; 1548 /* here back off the timer as well, to slow us down */ 1549 net->RTO <<= 1; 1550 } 1551 net->cwnd = net->ssthresh; 1552 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { 1553 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1554 } 1555 } 1556