Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netinet/sctp_cc_functions.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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. 53 */ 54 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 55 /* we always get at LEAST 2 MTU's */ 56 if (net->cwnd < (2 * net->mtu)) { 57 net->cwnd = 2 * net->mtu; 58 } 59 net->ssthresh = stcb->asoc.peers_rwnd; 60 61 if (sctp_logging_level & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 62 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 63 } 64 } 65 66 void 67 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb, 68 struct sctp_association *asoc) 69 { 70 struct sctp_nets *net; 71 72 /*- 73 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 74 * (net->fast_retran_loss_recovery == 0))) 75 */ 76 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 77 if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) { 78 /* out of a RFC2582 Fast recovery window? */ 79 if (net->net_ack > 0) { 80 /* 81 * per section 7.2.3, are there any 82 * destinations that had a fast retransmit 83 * to them. If so what we need to do is 84 * adjust ssthresh and cwnd. 85 */ 86 struct sctp_tmit_chunk *lchk; 87 int old_cwnd = net->cwnd; 88 89 net->ssthresh = net->cwnd / 2; 90 if (net->ssthresh < (net->mtu * 2)) { 91 net->ssthresh = 2 * net->mtu; 92 } 93 net->cwnd = net->ssthresh; 94 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 95 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 96 SCTP_CWND_LOG_FROM_FR); 97 } 98 lchk = TAILQ_FIRST(&asoc->send_queue); 99 100 net->partial_bytes_acked = 0; 101 /* Turn on fast recovery window */ 102 asoc->fast_retran_loss_recovery = 1; 103 if (lchk == NULL) { 104 /* Mark end of the window */ 105 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 106 } else { 107 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 108 } 109 110 /* 111 * CMT fast recovery -- per destination 112 * recovery variable. 113 */ 114 net->fast_retran_loss_recovery = 1; 115 116 if (lchk == NULL) { 117 /* Mark end of the window */ 118 net->fast_recovery_tsn = asoc->sending_seq - 1; 119 } else { 120 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 121 } 122 123 /* 124 * Disable Nonce Sum Checking and store the 125 * resync tsn 126 */ 127 asoc->nonce_sum_check = 0; 128 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 129 130 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 131 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 132 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 133 stcb->sctp_ep, stcb, net); 134 } 135 } else if (net->net_ack > 0) { 136 /* 137 * Mark a peg that we WOULD have done a cwnd 138 * reduction but RFC2582 prevented this action. 139 */ 140 SCTP_STAT_INCR(sctps_fastretransinrtt); 141 } 142 } 143 } 144 145 void 146 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb, 147 struct sctp_association *asoc, 148 int accum_moved, int reneged_all, int will_exit) 149 { 150 struct sctp_nets *net; 151 152 /******************************/ 153 /* update cwnd and Early FR */ 154 /******************************/ 155 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 156 157 #ifdef JANA_CMT_FAST_RECOVERY 158 /* 159 * CMT fast recovery code. Need to debug. 160 */ 161 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 162 if (compare_with_wrap(asoc->last_acked_seq, 163 net->fast_recovery_tsn, MAX_TSN) || 164 (asoc->last_acked_seq == net->fast_recovery_tsn) || 165 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 166 (net->pseudo_cumack == net->fast_recovery_tsn)) { 167 net->will_exit_fast_recovery = 1; 168 } 169 } 170 #endif 171 if (sctp_early_fr) { 172 /* 173 * So, first of all do we need to have a Early FR 174 * timer running? 175 */ 176 if (((TAILQ_FIRST(&asoc->sent_queue)) && 177 (net->ref_count > 1) && 178 (net->flight_size < net->cwnd)) || 179 (reneged_all)) { 180 /* 181 * yes, so in this case stop it if its 182 * running, and then restart it. Reneging 183 * all is a special case where we want to 184 * run the Early FR timer and then force the 185 * last few unacked to be sent, causing us 186 * to illicit a sack with gaps to force out 187 * the others. 188 */ 189 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 190 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 191 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 192 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 193 } 194 SCTP_STAT_INCR(sctps_earlyfrstrid); 195 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 196 } else { 197 /* No, stop it if its running */ 198 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 199 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 200 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 201 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 202 } 203 } 204 } 205 /* if nothing was acked on this destination skip it */ 206 if (net->net_ack == 0) { 207 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 208 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 209 } 210 continue; 211 } 212 if (net->net_ack2 > 0) { 213 /* 214 * Karn's rule applies to clearing error count, this 215 * is optional. 216 */ 217 net->error_count = 0; 218 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 219 SCTP_ADDR_NOT_REACHABLE) { 220 /* addr came good */ 221 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 222 net->dest_state |= SCTP_ADDR_REACHABLE; 223 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 224 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 225 /* now was it the primary? if so restore */ 226 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 227 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 228 } 229 } 230 /* 231 * JRS 5/14/07 - If CMT PF is on and the destination 232 * is in PF state, set the destination to active 233 * state and set the cwnd to one or two MTU's based 234 * on whether PF1 or PF2 is being used. 235 * 236 * Should we stop any running T3 timer here? 237 */ 238 if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) == 239 SCTP_ADDR_PF)) { 240 net->dest_state &= ~SCTP_ADDR_PF; 241 net->cwnd = net->mtu * sctp_cmt_pf; 242 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 243 net, net->cwnd); 244 /* 245 * Since the cwnd value is explicitly set, 246 * skip the code that updates the cwnd 247 * value. 248 */ 249 goto skip_cwnd_update; 250 } 251 } 252 #ifdef JANA_CMT_FAST_RECOVERY 253 /* 254 * CMT fast recovery code 255 */ 256 /* 257 * if (sctp_cmt_on_off == 1 && 258 * net->fast_retran_loss_recovery && 259 * net->will_exit_fast_recovery == 0) { @@@ Do something } 260 * else if (sctp_cmt_on_off == 0 && 261 * asoc->fast_retran_loss_recovery && will_exit == 0) { 262 */ 263 #endif 264 265 if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) { 266 /* 267 * If we are in loss recovery we skip any cwnd 268 * update 269 */ 270 goto skip_cwnd_update; 271 } 272 /* 273 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 274 * moved. 275 */ 276 if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) { 277 /* If the cumulative ack moved we can proceed */ 278 if (net->cwnd <= net->ssthresh) { 279 /* We are in slow start */ 280 if (net->flight_size + net->net_ack >= 281 net->cwnd) { 282 if (net->net_ack > (net->mtu * sctp_L2_abc_variable)) { 283 net->cwnd += (net->mtu * sctp_L2_abc_variable); 284 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 285 sctp_log_cwnd(stcb, net, net->mtu, 286 SCTP_CWND_LOG_FROM_SS); 287 } 288 } else { 289 net->cwnd += net->net_ack; 290 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 291 sctp_log_cwnd(stcb, net, net->net_ack, 292 SCTP_CWND_LOG_FROM_SS); 293 } 294 } 295 } else { 296 unsigned int dif; 297 298 dif = net->cwnd - (net->flight_size + 299 net->net_ack); 300 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 301 sctp_log_cwnd(stcb, net, net->net_ack, 302 SCTP_CWND_LOG_NOADV_SS); 303 } 304 } 305 } else { 306 /* We are in congestion avoidance */ 307 if (net->flight_size + net->net_ack >= 308 net->cwnd) { 309 /* 310 * add to pba only if we had a 311 * cwnd's worth (or so) in flight OR 312 * the burst limit was applied. 313 */ 314 net->partial_bytes_acked += 315 net->net_ack; 316 317 /* 318 * Do we need to increase (if pba is 319 * > cwnd)? 320 */ 321 if (net->partial_bytes_acked >= 322 net->cwnd) { 323 if (net->cwnd < 324 net->partial_bytes_acked) { 325 net->partial_bytes_acked -= 326 net->cwnd; 327 } else { 328 net->partial_bytes_acked = 329 0; 330 } 331 net->cwnd += net->mtu; 332 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 333 sctp_log_cwnd(stcb, net, net->mtu, 334 SCTP_CWND_LOG_FROM_CA); 335 } 336 } else { 337 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 338 sctp_log_cwnd(stcb, net, net->net_ack, 339 SCTP_CWND_LOG_NOADV_CA); 340 } 341 } 342 } else { 343 unsigned int dif; 344 345 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 346 sctp_log_cwnd(stcb, net, net->net_ack, 347 SCTP_CWND_LOG_NOADV_CA); 348 } 349 dif = net->cwnd - (net->flight_size + 350 net->net_ack); 351 } 352 } 353 } else { 354 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 355 sctp_log_cwnd(stcb, net, net->mtu, 356 SCTP_CWND_LOG_NO_CUMACK); 357 } 358 } 359 skip_cwnd_update: 360 /* 361 * NOW, according to Karn's rule do we need to restore the 362 * RTO timer back? Check our net_ack2. If not set then we 363 * have a ambiguity.. i.e. all data ack'd was sent to more 364 * than one place. 365 */ 366 if (net->net_ack2) { 367 /* restore any doubled timers */ 368 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 369 if (net->RTO < stcb->asoc.minrto) { 370 net->RTO = stcb->asoc.minrto; 371 } 372 if (net->RTO > stcb->asoc.maxrto) { 373 net->RTO = stcb->asoc.maxrto; 374 } 375 } 376 } 377 } 378 379 void 380 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 381 struct sctp_nets *net) 382 { 383 int old_cwnd = net->cwnd; 384 385 net->ssthresh = net->cwnd >> 1; 386 if (net->ssthresh < (net->mtu << 1)) { 387 net->ssthresh = (net->mtu << 1); 388 } 389 net->cwnd = net->mtu; 390 /* floor of 1 mtu */ 391 if (net->cwnd < net->mtu) 392 net->cwnd = net->mtu; 393 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 394 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 395 } 396 net->partial_bytes_acked = 0; 397 } 398 399 struct sctp_hs_raise_drop { 400 int32_t cwnd; 401 int32_t increase; 402 int32_t drop_percent; 403 }; 404 405 #define SCTP_HS_TABLE_SIZE 73 406 407 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = { 408 {38, 1, 50}, /* 0 */ 409 {118, 2, 44}, /* 1 */ 410 {221, 3, 41}, /* 2 */ 411 {347, 4, 38}, /* 3 */ 412 {495, 5, 37}, /* 4 */ 413 {663, 6, 35}, /* 5 */ 414 {851, 7, 34}, /* 6 */ 415 {1058, 8, 33}, /* 7 */ 416 {1284, 9, 32}, /* 8 */ 417 {1529, 10, 31}, /* 9 */ 418 {1793, 11, 30}, /* 10 */ 419 {2076, 12, 29}, /* 11 */ 420 {2378, 13, 28}, /* 12 */ 421 {2699, 14, 28}, /* 13 */ 422 {3039, 15, 27}, /* 14 */ 423 {3399, 16, 27}, /* 15 */ 424 {3778, 17, 26}, /* 16 */ 425 {4177, 18, 26}, /* 17 */ 426 {4596, 19, 25}, /* 18 */ 427 {5036, 20, 25}, /* 19 */ 428 {5497, 21, 24}, /* 20 */ 429 {5979, 22, 24}, /* 21 */ 430 {6483, 23, 23}, /* 22 */ 431 {7009, 24, 23}, /* 23 */ 432 {7558, 25, 22}, /* 24 */ 433 {8130, 26, 22}, /* 25 */ 434 {8726, 27, 22}, /* 26 */ 435 {9346, 28, 21}, /* 27 */ 436 {9991, 29, 21}, /* 28 */ 437 {10661, 30, 21}, /* 29 */ 438 {11358, 31, 20}, /* 30 */ 439 {12082, 32, 20}, /* 31 */ 440 {12834, 33, 20}, /* 32 */ 441 {13614, 34, 19}, /* 33 */ 442 {14424, 35, 19}, /* 34 */ 443 {15265, 36, 19}, /* 35 */ 444 {16137, 37, 19}, /* 36 */ 445 {17042, 38, 18}, /* 37 */ 446 {17981, 39, 18}, /* 38 */ 447 {18955, 40, 18}, /* 39 */ 448 {19965, 41, 17}, /* 40 */ 449 {21013, 42, 17}, /* 41 */ 450 {22101, 43, 17}, /* 42 */ 451 {23230, 44, 17}, /* 43 */ 452 {24402, 45, 16}, /* 44 */ 453 {25618, 46, 16}, /* 45 */ 454 {26881, 47, 16}, /* 46 */ 455 {28193, 48, 16}, /* 47 */ 456 {29557, 49, 15}, /* 48 */ 457 {30975, 50, 15}, /* 49 */ 458 {32450, 51, 15}, /* 50 */ 459 {33986, 52, 15}, /* 51 */ 460 {35586, 53, 14}, /* 52 */ 461 {37253, 54, 14}, /* 53 */ 462 {38992, 55, 14}, /* 54 */ 463 {40808, 56, 14}, /* 55 */ 464 {42707, 57, 13}, /* 56 */ 465 {44694, 58, 13}, /* 57 */ 466 {46776, 59, 13}, /* 58 */ 467 {48961, 60, 13}, /* 59 */ 468 {51258, 61, 13}, /* 60 */ 469 {53677, 62, 12}, /* 61 */ 470 {56230, 63, 12}, /* 62 */ 471 {58932, 64, 12}, /* 63 */ 472 {61799, 65, 12}, /* 64 */ 473 {64851, 66, 11}, /* 65 */ 474 {68113, 67, 11}, /* 66 */ 475 {71617, 68, 11}, /* 67 */ 476 {75401, 69, 10}, /* 68 */ 477 {79517, 70, 10}, /* 69 */ 478 {84035, 71, 10}, /* 70 */ 479 {89053, 72, 10}, /* 71 */ 480 {94717, 73, 9} /* 72 */ 481 }; 482 483 static void 484 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net) 485 { 486 int cur_val, i, indx, incr; 487 488 cur_val = net->cwnd >> 10; 489 indx = SCTP_HS_TABLE_SIZE - 1; 490 #ifdef SCTP_DEBUG 491 printf("HS CC CAlled.\n"); 492 #endif 493 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 494 /* normal mode */ 495 if (net->net_ack > net->mtu) { 496 net->cwnd += net->mtu; 497 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 498 sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS); 499 } 500 } else { 501 net->cwnd += net->net_ack; 502 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 503 sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS); 504 } 505 } 506 } else { 507 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) { 508 if (cur_val < sctp_cwnd_adjust[i].cwnd) { 509 indx = i; 510 break; 511 } 512 } 513 net->last_hs_used = indx; 514 incr = ((sctp_cwnd_adjust[indx].increase) << 10); 515 net->cwnd += incr; 516 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 517 sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS); 518 } 519 } 520 } 521 522 static void 523 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net) 524 { 525 int cur_val, i, indx; 526 int old_cwnd = net->cwnd; 527 528 cur_val = net->cwnd >> 10; 529 indx = net->last_hs_used; 530 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 531 /* normal mode */ 532 net->ssthresh = net->cwnd / 2; 533 if (net->ssthresh < (net->mtu * 2)) { 534 net->ssthresh = 2 * net->mtu; 535 } 536 net->cwnd = net->ssthresh; 537 } else { 538 /* drop by the proper amount */ 539 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) * 540 sctp_cwnd_adjust[net->last_hs_used].drop_percent); 541 net->cwnd = net->ssthresh; 542 /* now where are we */ 543 indx = net->last_hs_used; 544 cur_val = net->cwnd >> 10; 545 /* reset where we are in the table */ 546 if (cur_val < sctp_cwnd_adjust[0].cwnd) { 547 /* feel out of hs */ 548 net->last_hs_used = 0; 549 } else { 550 for (i = indx; i >= 1; i--) { 551 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) { 552 break; 553 } 554 } 555 net->last_hs_used = indx; 556 } 557 } 558 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 559 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR); 560 } 561 } 562 563 void 564 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb, 565 struct sctp_association *asoc) 566 { 567 struct sctp_nets *net; 568 569 /* 570 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 571 * (net->fast_retran_loss_recovery == 0))) 572 */ 573 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 574 if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) { 575 /* out of a RFC2582 Fast recovery window? */ 576 if (net->net_ack > 0) { 577 /* 578 * per section 7.2.3, are there any 579 * destinations that had a fast retransmit 580 * to them. If so what we need to do is 581 * adjust ssthresh and cwnd. 582 */ 583 struct sctp_tmit_chunk *lchk; 584 585 sctp_hs_cwnd_decrease(stcb, net); 586 587 lchk = TAILQ_FIRST(&asoc->send_queue); 588 589 net->partial_bytes_acked = 0; 590 /* Turn on fast recovery window */ 591 asoc->fast_retran_loss_recovery = 1; 592 if (lchk == NULL) { 593 /* Mark end of the window */ 594 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 595 } else { 596 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 597 } 598 599 /* 600 * CMT fast recovery -- per destination 601 * recovery variable. 602 */ 603 net->fast_retran_loss_recovery = 1; 604 605 if (lchk == NULL) { 606 /* Mark end of the window */ 607 net->fast_recovery_tsn = asoc->sending_seq - 1; 608 } else { 609 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 610 } 611 612 /* 613 * Disable Nonce Sum Checking and store the 614 * resync tsn 615 */ 616 asoc->nonce_sum_check = 0; 617 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 618 619 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 620 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 621 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 622 stcb->sctp_ep, stcb, net); 623 } 624 } else if (net->net_ack > 0) { 625 /* 626 * Mark a peg that we WOULD have done a cwnd 627 * reduction but RFC2582 prevented this action. 628 */ 629 SCTP_STAT_INCR(sctps_fastretransinrtt); 630 } 631 } 632 } 633 634 void 635 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb, 636 struct sctp_association *asoc, 637 int accum_moved, int reneged_all, int will_exit) 638 { 639 struct sctp_nets *net; 640 641 /******************************/ 642 /* update cwnd and Early FR */ 643 /******************************/ 644 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 645 646 #ifdef JANA_CMT_FAST_RECOVERY 647 /* 648 * CMT fast recovery code. Need to debug. 649 */ 650 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 651 if (compare_with_wrap(asoc->last_acked_seq, 652 net->fast_recovery_tsn, MAX_TSN) || 653 (asoc->last_acked_seq == net->fast_recovery_tsn) || 654 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 655 (net->pseudo_cumack == net->fast_recovery_tsn)) { 656 net->will_exit_fast_recovery = 1; 657 } 658 } 659 #endif 660 if (sctp_early_fr) { 661 /* 662 * So, first of all do we need to have a Early FR 663 * timer running? 664 */ 665 if (((TAILQ_FIRST(&asoc->sent_queue)) && 666 (net->ref_count > 1) && 667 (net->flight_size < net->cwnd)) || 668 (reneged_all)) { 669 /* 670 * yes, so in this case stop it if its 671 * running, and then restart it. Reneging 672 * all is a special case where we want to 673 * run the Early FR timer and then force the 674 * last few unacked to be sent, causing us 675 * to illicit a sack with gaps to force out 676 * the others. 677 */ 678 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 679 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 680 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 681 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 682 } 683 SCTP_STAT_INCR(sctps_earlyfrstrid); 684 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 685 } else { 686 /* No, stop it if its running */ 687 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 688 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 689 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 690 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 691 } 692 } 693 } 694 /* if nothing was acked on this destination skip it */ 695 if (net->net_ack == 0) { 696 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 697 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 698 } 699 continue; 700 } 701 if (net->net_ack2 > 0) { 702 /* 703 * Karn's rule applies to clearing error count, this 704 * is optional. 705 */ 706 net->error_count = 0; 707 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 708 SCTP_ADDR_NOT_REACHABLE) { 709 /* addr came good */ 710 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 711 net->dest_state |= SCTP_ADDR_REACHABLE; 712 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 713 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 714 /* now was it the primary? if so restore */ 715 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 716 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 717 } 718 } 719 /* 720 * JRS 5/14/07 - If CMT PF is on and the destination 721 * is in PF state, set the destination to active 722 * state and set the cwnd to one or two MTU's based 723 * on whether PF1 or PF2 is being used. 724 * 725 * Should we stop any running T3 timer here? 726 */ 727 if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) == 728 SCTP_ADDR_PF)) { 729 net->dest_state &= ~SCTP_ADDR_PF; 730 net->cwnd = net->mtu * sctp_cmt_pf; 731 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 732 net, net->cwnd); 733 /* 734 * Since the cwnd value is explicitly set, 735 * skip the code that updates the cwnd 736 * value. 737 */ 738 goto skip_cwnd_update; 739 } 740 } 741 #ifdef JANA_CMT_FAST_RECOVERY 742 /* 743 * CMT fast recovery code 744 */ 745 /* 746 * if (sctp_cmt_on_off == 1 && 747 * net->fast_retran_loss_recovery && 748 * net->will_exit_fast_recovery == 0) { @@@ Do something } 749 * else if (sctp_cmt_on_off == 0 && 750 * asoc->fast_retran_loss_recovery && will_exit == 0) { 751 */ 752 #endif 753 754 if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) { 755 /* 756 * If we are in loss recovery we skip any cwnd 757 * update 758 */ 759 goto skip_cwnd_update; 760 } 761 /* 762 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 763 * moved. 764 */ 765 if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) { 766 /* If the cumulative ack moved we can proceed */ 767 if (net->cwnd <= net->ssthresh) { 768 /* We are in slow start */ 769 if (net->flight_size + net->net_ack >= 770 net->cwnd) { 771 772 sctp_hs_cwnd_increase(stcb, net); 773 774 } else { 775 unsigned int dif; 776 777 dif = net->cwnd - (net->flight_size + 778 net->net_ack); 779 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 780 sctp_log_cwnd(stcb, net, net->net_ack, 781 SCTP_CWND_LOG_NOADV_SS); 782 } 783 } 784 } else { 785 /* We are in congestion avoidance */ 786 if (net->flight_size + net->net_ack >= 787 net->cwnd) { 788 /* 789 * add to pba only if we had a 790 * cwnd's worth (or so) in flight OR 791 * the burst limit was applied. 792 */ 793 net->partial_bytes_acked += 794 net->net_ack; 795 796 /* 797 * Do we need to increase (if pba is 798 * > cwnd)? 799 */ 800 if (net->partial_bytes_acked >= 801 net->cwnd) { 802 if (net->cwnd < 803 net->partial_bytes_acked) { 804 net->partial_bytes_acked -= 805 net->cwnd; 806 } else { 807 net->partial_bytes_acked = 808 0; 809 } 810 net->cwnd += net->mtu; 811 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 812 sctp_log_cwnd(stcb, net, net->mtu, 813 SCTP_CWND_LOG_FROM_CA); 814 } 815 } else { 816 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 817 sctp_log_cwnd(stcb, net, net->net_ack, 818 SCTP_CWND_LOG_NOADV_CA); 819 } 820 } 821 } else { 822 unsigned int dif; 823 824 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 825 sctp_log_cwnd(stcb, net, net->net_ack, 826 SCTP_CWND_LOG_NOADV_CA); 827 } 828 dif = net->cwnd - (net->flight_size + 829 net->net_ack); 830 } 831 } 832 } else { 833 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 834 sctp_log_cwnd(stcb, net, net->mtu, 835 SCTP_CWND_LOG_NO_CUMACK); 836 } 837 } 838 skip_cwnd_update: 839 /* 840 * NOW, according to Karn's rule do we need to restore the 841 * RTO timer back? Check our net_ack2. If not set then we 842 * have a ambiguity.. i.e. all data ack'd was sent to more 843 * than one place. 844 */ 845 if (net->net_ack2) { 846 /* restore any doubled timers */ 847 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 848 if (net->RTO < stcb->asoc.minrto) { 849 net->RTO = stcb->asoc.minrto; 850 } 851 if (net->RTO > stcb->asoc.maxrto) { 852 net->RTO = stcb->asoc.maxrto; 853 } 854 } 855 } 856 } 857 858 void 859 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 860 struct sctp_nets *net) 861 { 862 int old_cwnd; 863 864 old_cwnd = net->cwnd; 865 866 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 867 net->ssthresh = net->cwnd / 2; 868 if (net->ssthresh < net->mtu) { 869 net->ssthresh = net->mtu; 870 /* here back off the timer as well, to slow us down */ 871 net->RTO <<= 1; 872 } 873 net->cwnd = net->ssthresh; 874 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 875 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 876 } 877 } 878 879 void 880 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb, 881 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, 882 uint32_t * bottle_bw, uint32_t * on_queue) 883 { 884 uint32_t bw_avail; 885 int rtt, incr; 886 int old_cwnd = net->cwnd; 887 888 /* need real RTT for this calc */ 889 rtt = ((net->lastsa >> 2) + net->lastsv) >> 1; 890 /* get bottle neck bw */ 891 *bottle_bw = ntohl(cp->bottle_bw); 892 /* and whats on queue */ 893 *on_queue = ntohl(cp->current_onq); 894 /* 895 * adjust the on-queue if our flight is more it could be that the 896 * router has not yet gotten data "in-flight" to it 897 */ 898 if (*on_queue < net->flight_size) 899 *on_queue = net->flight_size; 900 /* calculate the available space */ 901 bw_avail = (*bottle_bw * rtt) / 1000; 902 if (bw_avail > *bottle_bw) { 903 /* 904 * Cap the growth to no more than the bottle neck. This can 905 * happen as RTT slides up due to queues. It also means if 906 * you have more than a 1 second RTT with a empty queue you 907 * will be limited to the bottle_bw per second no matter if 908 * other points have 1/2 the RTT and you could get more 909 * out... 910 */ 911 bw_avail = *bottle_bw; 912 } 913 if (*on_queue > bw_avail) { 914 /* 915 * No room for anything else don't allow anything else to be 916 * "added to the fire". 917 */ 918 int seg_inflight, seg_onqueue, my_portion; 919 920 net->partial_bytes_acked = 0; 921 922 /* how much are we over queue size? */ 923 incr = *on_queue - bw_avail; 924 if (stcb->asoc.seen_a_sack_this_pkt) { 925 /* 926 * undo any cwnd adjustment that the sack might have 927 * made 928 */ 929 net->cwnd = net->prev_cwnd; 930 } 931 /* Now how much of that is mine? */ 932 seg_inflight = net->flight_size / net->mtu; 933 seg_onqueue = *on_queue / net->mtu; 934 my_portion = (incr * seg_inflight) / seg_onqueue; 935 936 /* Have I made an adjustment already */ 937 if (net->cwnd > net->flight_size) { 938 /* 939 * for this flight I made an adjustment we need to 940 * decrease the portion by a share our previous 941 * adjustment. 942 */ 943 int diff_adj; 944 945 diff_adj = net->cwnd - net->flight_size; 946 if (diff_adj > my_portion) 947 my_portion = 0; 948 else 949 my_portion -= diff_adj; 950 } 951 /* 952 * back down to the previous cwnd (assume we have had a sack 953 * before this packet). minus what ever portion of the 954 * overage is my fault. 955 */ 956 net->cwnd -= my_portion; 957 958 /* we will NOT back down more than 1 MTU */ 959 if (net->cwnd <= net->mtu) { 960 net->cwnd = net->mtu; 961 } 962 /* force into CA */ 963 net->ssthresh = net->cwnd - 1; 964 } else { 965 /* 966 * Take 1/4 of the space left or max burst up .. whichever 967 * is less. 968 */ 969 incr = min((bw_avail - *on_queue) >> 2, 970 stcb->asoc.max_burst * net->mtu); 971 net->cwnd += incr; 972 } 973 if (net->cwnd > bw_avail) { 974 /* We can't exceed the pipe size */ 975 net->cwnd = bw_avail; 976 } 977 if (net->cwnd < net->mtu) { 978 /* We always have 1 MTU */ 979 net->cwnd = net->mtu; 980 } 981 if (net->cwnd - old_cwnd != 0) { 982 /* log only changes */ 983 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 984 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 985 SCTP_CWND_LOG_FROM_SAT); 986 } 987 } 988 } 989 990 void 991 sctp_cwnd_update_after_output(struct sctp_tcb *stcb, 992 struct sctp_nets *net, int burst_limit) 993 { 994 int old_cwnd; 995 996 if (net->ssthresh < net->cwnd) 997 net->ssthresh = net->cwnd; 998 old_cwnd = net->cwnd; 999 net->cwnd = (net->flight_size + (burst_limit * net->mtu)); 1000 1001 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1002 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST); 1003 } 1004 } 1005 1006 void 1007 sctp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 1008 struct sctp_tcb *stcb, struct sctp_nets *net) 1009 { 1010 int old_cwnd; 1011 1012 old_cwnd = net->cwnd; 1013 1014 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 1015 /* 1016 * make a small adjustment to cwnd and force to CA. 1017 */ 1018 if (net->cwnd > net->mtu) 1019 /* drop down one MTU after sending */ 1020 net->cwnd -= net->mtu; 1021 if (net->cwnd < net->ssthresh) 1022 /* still in SS move to CA */ 1023 net->ssthresh = net->cwnd - 1; 1024 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1025 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 1026 } 1027 } 1028 1029 /* 1030 * H-TCP congestion control. The algorithm is detailed in: 1031 * R.N.Shorten, D.J.Leith: 1032 * "H-TCP: TCP for high-speed and long-distance networks" 1033 * Proc. PFLDnet, Argonne, 2004. 1034 * http://www.hamilton.ie/net/htcp3.pdf 1035 */ 1036 1037 1038 static int use_rtt_scaling = 1; 1039 static int use_bandwidth_switch = 1; 1040 1041 static inline int 1042 between(uint32_t seq1, uint32_t seq2, uint32_t seq3) 1043 { 1044 return seq3 - seq2 >= seq1 - seq2; 1045 } 1046 1047 static inline uint32_t 1048 htcp_cong_time(struct htcp *ca) 1049 { 1050 return sctp_get_tick_count() - ca->last_cong; 1051 } 1052 1053 static inline uint32_t 1054 htcp_ccount(struct htcp *ca) 1055 { 1056 return htcp_cong_time(ca) / ca->minRTT; 1057 } 1058 1059 static inline void 1060 htcp_reset(struct htcp *ca) 1061 { 1062 ca->undo_last_cong = ca->last_cong; 1063 ca->undo_maxRTT = ca->maxRTT; 1064 ca->undo_old_maxB = ca->old_maxB; 1065 ca->last_cong = sctp_get_tick_count(); 1066 } 1067 1068 #ifdef SCTP_NOT_USED 1069 1070 static uint32_t 1071 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net) 1072 { 1073 net->htcp_ca.last_cong = net->htcp_ca.undo_last_cong; 1074 net->htcp_ca.maxRTT = net->htcp_ca.undo_maxRTT; 1075 net->htcp_ca.old_maxB = net->htcp_ca.undo_old_maxB; 1076 return max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->htcp_ca.beta) * net->mtu); 1077 } 1078 1079 #endif 1080 1081 static inline void 1082 measure_rtt(struct sctp_tcb *stcb, struct sctp_nets *net) 1083 { 1084 uint32_t srtt = net->lastsa >> 3; 1085 1086 /* keep track of minimum RTT seen so far, minRTT is zero at first */ 1087 if (net->htcp_ca.minRTT > srtt || !net->htcp_ca.minRTT) 1088 net->htcp_ca.minRTT = srtt; 1089 1090 /* max RTT */ 1091 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->htcp_ca) > 3) { 1092 if (net->htcp_ca.maxRTT < net->htcp_ca.minRTT) 1093 net->htcp_ca.maxRTT = net->htcp_ca.minRTT; 1094 if (net->htcp_ca.maxRTT < srtt && srtt <= net->htcp_ca.maxRTT + MSEC_TO_TICKS(20)) 1095 net->htcp_ca.maxRTT = srtt; 1096 } 1097 } 1098 1099 static void 1100 measure_achieved_throughput(struct sctp_tcb *stcb, struct sctp_nets *net) 1101 { 1102 uint32_t now = sctp_get_tick_count(); 1103 1104 if (net->fast_retran_ip == 0) 1105 net->htcp_ca.bytes_acked = net->net_ack; 1106 1107 if (!use_bandwidth_switch) 1108 return; 1109 1110 /* achieved throughput calculations */ 1111 /* JRS - not 100% sure of this statement */ 1112 if (net->fast_retran_ip == 1) { 1113 net->htcp_ca.bytecount = 0; 1114 net->htcp_ca.lasttime = now; 1115 return; 1116 } 1117 net->htcp_ca.bytecount += net->net_ack; 1118 1119 if (net->htcp_ca.bytecount >= net->cwnd - ((net->htcp_ca.alpha >> 7 ? : 1) * net->mtu) 1120 && now - net->htcp_ca.lasttime >= net->htcp_ca.minRTT 1121 && net->htcp_ca.minRTT > 0) { 1122 uint32_t cur_Bi = net->htcp_ca.bytecount / net->mtu * hz / (now - net->htcp_ca.lasttime); 1123 1124 if (htcp_ccount(&net->htcp_ca) <= 3) { 1125 /* just after backoff */ 1126 net->htcp_ca.minB = net->htcp_ca.maxB = net->htcp_ca.Bi = cur_Bi; 1127 } else { 1128 net->htcp_ca.Bi = (3 * net->htcp_ca.Bi + cur_Bi) / 4; 1129 if (net->htcp_ca.Bi > net->htcp_ca.maxB) 1130 net->htcp_ca.maxB = net->htcp_ca.Bi; 1131 if (net->htcp_ca.minB > net->htcp_ca.maxB) 1132 net->htcp_ca.minB = net->htcp_ca.maxB; 1133 } 1134 net->htcp_ca.bytecount = 0; 1135 net->htcp_ca.lasttime = now; 1136 } 1137 } 1138 1139 static inline void 1140 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT) 1141 { 1142 if (use_bandwidth_switch) { 1143 uint32_t maxB = ca->maxB; 1144 uint32_t old_maxB = ca->old_maxB; 1145 1146 ca->old_maxB = ca->maxB; 1147 1148 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) { 1149 ca->beta = BETA_MIN; 1150 ca->modeswitch = 0; 1151 return; 1152 } 1153 } 1154 if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) { 1155 ca->beta = (minRTT << 7) / maxRTT; 1156 if (ca->beta < BETA_MIN) 1157 ca->beta = BETA_MIN; 1158 else if (ca->beta > BETA_MAX) 1159 ca->beta = BETA_MAX; 1160 } else { 1161 ca->beta = BETA_MIN; 1162 ca->modeswitch = 1; 1163 } 1164 } 1165 1166 static inline void 1167 htcp_alpha_update(struct htcp *ca) 1168 { 1169 uint32_t minRTT = ca->minRTT; 1170 uint32_t factor = 1; 1171 uint32_t diff = htcp_cong_time(ca); 1172 1173 if (diff > (uint32_t) hz) { 1174 diff -= hz; 1175 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz; 1176 } 1177 if (use_rtt_scaling && minRTT) { 1178 uint32_t scale = (hz << 3) / (10 * minRTT); 1179 1180 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to 1181 * interval [0.5,10]<<3 */ 1182 factor = (factor << 3) / scale; 1183 if (!factor) 1184 factor = 1; 1185 } 1186 ca->alpha = 2 * factor * ((1 << 7) - ca->beta); 1187 if (!ca->alpha) 1188 ca->alpha = ALPHA_BASE; 1189 } 1190 1191 /* After we have the rtt data to calculate beta, we'd still prefer to wait one 1192 * rtt before we adjust our beta to ensure we are working from a consistent 1193 * data. 1194 * 1195 * This function should be called when we hit a congestion event since only at 1196 * that point do we really have a real sense of maxRTT (the queues en route 1197 * were getting just too full now). 1198 */ 1199 static void 1200 htcp_param_update(struct sctp_tcb *stcb, struct sctp_nets *net) 1201 { 1202 uint32_t minRTT = net->htcp_ca.minRTT; 1203 uint32_t maxRTT = net->htcp_ca.maxRTT; 1204 1205 htcp_beta_update(&net->htcp_ca, minRTT, maxRTT); 1206 htcp_alpha_update(&net->htcp_ca); 1207 1208 /* 1209 * add slowly fading memory for maxRTT to accommodate routing 1210 * changes etc 1211 */ 1212 if (minRTT > 0 && maxRTT > minRTT) 1213 net->htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100; 1214 } 1215 1216 static uint32_t 1217 htcp_recalc_ssthresh(struct sctp_tcb *stcb, struct sctp_nets *net) 1218 { 1219 htcp_param_update(stcb, net); 1220 return max(((net->cwnd / net->mtu * net->htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu); 1221 } 1222 1223 static void 1224 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net) 1225 { 1226 /*- 1227 * How to handle these functions? 1228 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question. 1229 * return; 1230 */ 1231 if (net->cwnd <= net->ssthresh) { 1232 /* We are in slow start */ 1233 if (net->flight_size + net->net_ack >= net->cwnd) { 1234 if (net->net_ack > (net->mtu * sctp_L2_abc_variable)) { 1235 net->cwnd += (net->mtu * sctp_L2_abc_variable); 1236 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1237 sctp_log_cwnd(stcb, net, net->mtu, 1238 SCTP_CWND_LOG_FROM_SS); 1239 } 1240 } else { 1241 net->cwnd += net->net_ack; 1242 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1243 sctp_log_cwnd(stcb, net, net->net_ack, 1244 SCTP_CWND_LOG_FROM_SS); 1245 } 1246 } 1247 } else { 1248 unsigned int dif; 1249 1250 dif = net->cwnd - (net->flight_size + 1251 net->net_ack); 1252 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 1253 sctp_log_cwnd(stcb, net, net->net_ack, 1254 SCTP_CWND_LOG_NOADV_SS); 1255 } 1256 } 1257 } else { 1258 measure_rtt(stcb, net); 1259 1260 /* 1261 * In dangerous area, increase slowly. In theory this is 1262 * net->cwnd += alpha / net->cwnd 1263 */ 1264 /* What is snd_cwnd_cnt?? */ 1265 if (((net->partial_bytes_acked / net->mtu * net->htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) { 1266 /*- 1267 * Does SCTP have a cwnd clamp? 1268 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS). 1269 */ 1270 net->cwnd += net->mtu; 1271 net->partial_bytes_acked = 0; 1272 htcp_alpha_update(&net->htcp_ca); 1273 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1274 sctp_log_cwnd(stcb, net, net->mtu, 1275 SCTP_CWND_LOG_FROM_CA); 1276 } 1277 } else { 1278 net->partial_bytes_acked += net->net_ack; 1279 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 1280 sctp_log_cwnd(stcb, net, net->net_ack, 1281 SCTP_CWND_LOG_NOADV_CA); 1282 } 1283 } 1284 1285 net->htcp_ca.bytes_acked = net->mtu; 1286 } 1287 } 1288 1289 #ifdef SCTP_NOT_USED 1290 /* Lower bound on congestion window. */ 1291 static uint32_t 1292 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net) 1293 { 1294 return net->ssthresh; 1295 } 1296 1297 #endif 1298 1299 static void 1300 htcp_init(struct sctp_tcb *stcb, struct sctp_nets *net) 1301 { 1302 memset(&net->htcp_ca, 0, sizeof(struct htcp)); 1303 net->htcp_ca.alpha = ALPHA_BASE; 1304 net->htcp_ca.beta = BETA_MIN; 1305 net->htcp_ca.bytes_acked = net->mtu; 1306 net->htcp_ca.last_cong = sctp_get_tick_count(); 1307 } 1308 1309 void 1310 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net) 1311 { 1312 /* 1313 * We take the max of the burst limit times a MTU or the 1314 * INITIAL_CWND. We then limit this to 4 MTU's of sending. 1315 */ 1316 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND)); 1317 /* we always get at LEAST 2 MTU's */ 1318 if (net->cwnd < (2 * net->mtu)) { 1319 net->cwnd = 2 * net->mtu; 1320 } 1321 net->ssthresh = stcb->asoc.peers_rwnd; 1322 htcp_init(stcb, net); 1323 1324 if (sctp_logging_level & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) { 1325 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION); 1326 } 1327 } 1328 1329 void 1330 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb, 1331 struct sctp_association *asoc, 1332 int accum_moved, int reneged_all, int will_exit) 1333 { 1334 struct sctp_nets *net; 1335 1336 /******************************/ 1337 /* update cwnd and Early FR */ 1338 /******************************/ 1339 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1340 1341 #ifdef JANA_CMT_FAST_RECOVERY 1342 /* 1343 * CMT fast recovery code. Need to debug. 1344 */ 1345 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) { 1346 if (compare_with_wrap(asoc->last_acked_seq, 1347 net->fast_recovery_tsn, MAX_TSN) || 1348 (asoc->last_acked_seq == net->fast_recovery_tsn) || 1349 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) || 1350 (net->pseudo_cumack == net->fast_recovery_tsn)) { 1351 net->will_exit_fast_recovery = 1; 1352 } 1353 } 1354 #endif 1355 if (sctp_early_fr) { 1356 /* 1357 * So, first of all do we need to have a Early FR 1358 * timer running? 1359 */ 1360 if (((TAILQ_FIRST(&asoc->sent_queue)) && 1361 (net->ref_count > 1) && 1362 (net->flight_size < net->cwnd)) || 1363 (reneged_all)) { 1364 /* 1365 * yes, so in this case stop it if its 1366 * running, and then restart it. Reneging 1367 * all is a special case where we want to 1368 * run the Early FR timer and then force the 1369 * last few unacked to be sent, causing us 1370 * to illicit a sack with gaps to force out 1371 * the others. 1372 */ 1373 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1374 SCTP_STAT_INCR(sctps_earlyfrstpidsck2); 1375 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1376 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20); 1377 } 1378 SCTP_STAT_INCR(sctps_earlyfrstrid); 1379 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net); 1380 } else { 1381 /* No, stop it if its running */ 1382 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) { 1383 SCTP_STAT_INCR(sctps_earlyfrstpidsck3); 1384 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net, 1385 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21); 1386 } 1387 } 1388 } 1389 /* if nothing was acked on this destination skip it */ 1390 if (net->net_ack == 0) { 1391 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 1392 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK); 1393 } 1394 continue; 1395 } 1396 if (net->net_ack2 > 0) { 1397 /* 1398 * Karn's rule applies to clearing error count, this 1399 * is optional. 1400 */ 1401 net->error_count = 0; 1402 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) == 1403 SCTP_ADDR_NOT_REACHABLE) { 1404 /* addr came good */ 1405 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE; 1406 net->dest_state |= SCTP_ADDR_REACHABLE; 1407 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb, 1408 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED); 1409 /* now was it the primary? if so restore */ 1410 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) { 1411 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net); 1412 } 1413 } 1414 /* 1415 * JRS 5/14/07 - If CMT PF is on and the destination 1416 * is in PF state, set the destination to active 1417 * state and set the cwnd to one or two MTU's based 1418 * on whether PF1 or PF2 is being used. 1419 * 1420 * Should we stop any running T3 timer here? 1421 */ 1422 if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) == 1423 SCTP_ADDR_PF)) { 1424 net->dest_state &= ~SCTP_ADDR_PF; 1425 net->cwnd = net->mtu * sctp_cmt_pf; 1426 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n", 1427 net, net->cwnd); 1428 /* 1429 * Since the cwnd value is explicitly set, 1430 * skip the code that updates the cwnd 1431 * value. 1432 */ 1433 goto skip_cwnd_update; 1434 } 1435 } 1436 #ifdef JANA_CMT_FAST_RECOVERY 1437 /* 1438 * CMT fast recovery code 1439 */ 1440 /* 1441 * if (sctp_cmt_on_off == 1 && 1442 * net->fast_retran_loss_recovery && 1443 * net->will_exit_fast_recovery == 0) { @@@ Do something } 1444 * else if (sctp_cmt_on_off == 0 && 1445 * asoc->fast_retran_loss_recovery && will_exit == 0) { 1446 */ 1447 #endif 1448 1449 if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) { 1450 /* 1451 * If we are in loss recovery we skip any cwnd 1452 * update 1453 */ 1454 goto skip_cwnd_update; 1455 } 1456 /* 1457 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has 1458 * moved. 1459 */ 1460 if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) { 1461 htcp_cong_avoid(stcb, net); 1462 measure_achieved_throughput(stcb, net); 1463 } else { 1464 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) { 1465 sctp_log_cwnd(stcb, net, net->mtu, 1466 SCTP_CWND_LOG_NO_CUMACK); 1467 } 1468 } 1469 skip_cwnd_update: 1470 /* 1471 * NOW, according to Karn's rule do we need to restore the 1472 * RTO timer back? Check our net_ack2. If not set then we 1473 * have a ambiguity.. i.e. all data ack'd was sent to more 1474 * than one place. 1475 */ 1476 if (net->net_ack2) { 1477 /* restore any doubled timers */ 1478 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1; 1479 if (net->RTO < stcb->asoc.minrto) { 1480 net->RTO = stcb->asoc.minrto; 1481 } 1482 if (net->RTO > stcb->asoc.maxrto) { 1483 net->RTO = stcb->asoc.maxrto; 1484 } 1485 } 1486 } 1487 } 1488 1489 void 1490 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb, 1491 struct sctp_association *asoc) 1492 { 1493 struct sctp_nets *net; 1494 1495 /* 1496 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) && 1497 * (net->fast_retran_loss_recovery == 0))) 1498 */ 1499 TAILQ_FOREACH(net, &asoc->nets, sctp_next) { 1500 if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) { 1501 /* out of a RFC2582 Fast recovery window? */ 1502 if (net->net_ack > 0) { 1503 /* 1504 * per section 7.2.3, are there any 1505 * destinations that had a fast retransmit 1506 * to them. If so what we need to do is 1507 * adjust ssthresh and cwnd. 1508 */ 1509 struct sctp_tmit_chunk *lchk; 1510 int old_cwnd = net->cwnd; 1511 1512 /* JRS - reset as if state were changed */ 1513 htcp_reset(&net->htcp_ca); 1514 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1515 net->cwnd = net->ssthresh; 1516 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1517 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), 1518 SCTP_CWND_LOG_FROM_FR); 1519 } 1520 lchk = TAILQ_FIRST(&asoc->send_queue); 1521 1522 net->partial_bytes_acked = 0; 1523 /* Turn on fast recovery window */ 1524 asoc->fast_retran_loss_recovery = 1; 1525 if (lchk == NULL) { 1526 /* Mark end of the window */ 1527 asoc->fast_recovery_tsn = asoc->sending_seq - 1; 1528 } else { 1529 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1530 } 1531 1532 /* 1533 * CMT fast recovery -- per destination 1534 * recovery variable. 1535 */ 1536 net->fast_retran_loss_recovery = 1; 1537 1538 if (lchk == NULL) { 1539 /* Mark end of the window */ 1540 net->fast_recovery_tsn = asoc->sending_seq - 1; 1541 } else { 1542 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1; 1543 } 1544 1545 /* 1546 * Disable Nonce Sum Checking and store the 1547 * resync tsn 1548 */ 1549 asoc->nonce_sum_check = 0; 1550 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1; 1551 1552 sctp_timer_stop(SCTP_TIMER_TYPE_SEND, 1553 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32); 1554 sctp_timer_start(SCTP_TIMER_TYPE_SEND, 1555 stcb->sctp_ep, stcb, net); 1556 } 1557 } else if (net->net_ack > 0) { 1558 /* 1559 * Mark a peg that we WOULD have done a cwnd 1560 * reduction but RFC2582 prevented this action. 1561 */ 1562 SCTP_STAT_INCR(sctps_fastretransinrtt); 1563 } 1564 } 1565 } 1566 1567 void 1568 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb, 1569 struct sctp_nets *net) 1570 { 1571 int old_cwnd = net->cwnd; 1572 1573 /* JRS - reset as if the state were being changed to timeout */ 1574 htcp_reset(&net->htcp_ca); 1575 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1576 net->cwnd = net->mtu; 1577 /* floor of 1 mtu */ 1578 if (net->cwnd < net->mtu) 1579 net->cwnd = net->mtu; 1580 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1581 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX); 1582 } 1583 net->partial_bytes_acked = 0; 1584 } 1585 1586 void 1587 sctp_htcp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp, 1588 struct sctp_tcb *stcb, struct sctp_nets *net) 1589 { 1590 int old_cwnd; 1591 1592 old_cwnd = net->cwnd; 1593 1594 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED); 1595 net->htcp_ca.last_cong = sctp_get_tick_count(); 1596 /* 1597 * make a small adjustment to cwnd and force to CA. 1598 */ 1599 if (net->cwnd > net->mtu) 1600 /* drop down one MTU after sending */ 1601 net->cwnd -= net->mtu; 1602 if (net->cwnd < net->ssthresh) 1603 /* still in SS move to CA */ 1604 net->ssthresh = net->cwnd - 1; 1605 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1606 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR); 1607 } 1608 } 1609 1610 void 1611 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, 1612 struct sctp_nets *net) 1613 { 1614 int old_cwnd; 1615 1616 old_cwnd = net->cwnd; 1617 1618 /* JRS - reset hctp as if state changed */ 1619 htcp_reset(&net->htcp_ca); 1620 SCTP_STAT_INCR(sctps_ecnereducedcwnd); 1621 net->ssthresh = htcp_recalc_ssthresh(stcb, net); 1622 if (net->ssthresh < net->mtu) { 1623 net->ssthresh = net->mtu; 1624 /* here back off the timer as well, to slow us down */ 1625 net->RTO <<= 1; 1626 } 1627 net->cwnd = net->ssthresh; 1628 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) { 1629 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT); 1630 } 1631 }
Cache object: 98e136e998455870ba22eeedad2c0772
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]