1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. 6 * Copyright (c) 2007-2008,2010,2014 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * All rights reserved. 11 * 12 * This software was developed at the Centre for Advanced Internet 13 * Architectures, Swinburne University of Technology, by Lawrence Stewart, James 14 * Healy and David Hayes, made possible in part by a grant from the Cisco 15 * University Research Program Fund at Community Foundation Silicon Valley. 16 * 17 * Portions of this software were developed at the Centre for Advanced 18 * Internet Architectures, Swinburne University of Technology, Melbourne, 19 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 20 * 21 * Redistribution and use in source and binary forms, with or without 22 * modification, are permitted provided that the following conditions 23 * are met: 24 * 1. Redistributions of source code must retain the above copyright 25 * notice, this list of conditions and the following disclaimer. 26 * 2. Redistributions in binary form must reproduce the above copyright 27 * notice, this list of conditions and the following disclaimer in the 28 * documentation and/or other materials provided with the distribution. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 */ 42 43 /* 44 * This software was first released in 2007 by James Healy and Lawrence Stewart 45 * whilst working on the NewTCP research project at Swinburne University of 46 * Technology's Centre for Advanced Internet Architectures, Melbourne, 47 * Australia, which was made possible in part by a grant from the Cisco 48 * University Research Program Fund at Community Foundation Silicon Valley. 49 * More details are available at: 50 * http://caia.swin.edu.au/urp/newtcp/ 51 * 52 * Dec 2014 garmitage@swin.edu.au 53 * Borrowed code fragments from cc_cdg.c to add modifiable beta 54 * via sysctls. 55 * 56 */ 57 58 #include <sys/cdefs.h> 59 __FBSDID("$FreeBSD$"); 60 61 #include <sys/param.h> 62 #include <sys/kernel.h> 63 #include <sys/malloc.h> 64 #include <sys/module.h> 65 #include <sys/socket.h> 66 #include <sys/lock.h> 67 #include <sys/mutex.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/systm.h> 71 72 #include <net/vnet.h> 73 74 #include <netinet/in.h> 75 #include <netinet/in_pcb.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_seq.h> 78 #include <netinet/tcp_var.h> 79 #include <netinet/tcp_log_buf.h> 80 #include <netinet/tcp_hpts.h> 81 #include <netinet/cc/cc.h> 82 #include <netinet/cc/cc_module.h> 83 #include <netinet/cc/cc_newreno.h> 84 85 static MALLOC_DEFINE(M_NEWRENO, "newreno data", 86 "newreno beta values"); 87 88 static void newreno_cb_destroy(struct cc_var *ccv); 89 static void newreno_ack_received(struct cc_var *ccv, uint16_t type); 90 static void newreno_after_idle(struct cc_var *ccv); 91 static void newreno_cong_signal(struct cc_var *ccv, uint32_t type); 92 static void newreno_post_recovery(struct cc_var *ccv); 93 static int newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf); 94 static void newreno_newround(struct cc_var *ccv, uint32_t round_cnt); 95 static void newreno_rttsample(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas); 96 static int newreno_cb_init(struct cc_var *ccv); 97 98 VNET_DEFINE(uint32_t, newreno_beta) = 50; 99 VNET_DEFINE(uint32_t, newreno_beta_ecn) = 80; 100 #define V_newreno_beta VNET(newreno_beta) 101 #define V_newreno_beta_ecn VNET(newreno_beta_ecn) 102 103 struct cc_algo newreno_cc_algo = { 104 .name = "newreno", 105 .cb_destroy = newreno_cb_destroy, 106 .ack_received = newreno_ack_received, 107 .after_idle = newreno_after_idle, 108 .cong_signal = newreno_cong_signal, 109 .post_recovery = newreno_post_recovery, 110 .ctl_output = newreno_ctl_output, 111 .newround = newreno_newround, 112 .rttsample = newreno_rttsample, 113 .cb_init = newreno_cb_init, 114 }; 115 116 static uint32_t hystart_lowcwnd = 16; 117 static uint32_t hystart_minrtt_thresh = 4000; 118 static uint32_t hystart_maxrtt_thresh = 16000; 119 static uint32_t hystart_n_rttsamples = 8; 120 static uint32_t hystart_css_growth_div = 4; 121 static uint32_t hystart_css_rounds = 5; 122 static uint32_t hystart_bblogs = 0; 123 124 static void 125 newreno_log_hystart_event(struct cc_var *ccv, struct newreno *nreno, uint8_t mod, uint32_t flex1) 126 { 127 /* 128 * Types of logs (mod value) 129 * 1 - rtt_thresh in flex1, checking to see if RTT is to great. 130 * 2 - rtt is too great, rtt_thresh in flex1. 131 * 3 - CSS is active incr in flex1 132 * 4 - A new round is beginning flex1 is round count 133 * 5 - A new RTT measurement flex1 is the new measurement. 134 * 6 - We enter CA ssthresh is also in flex1. 135 * 7 - Socket option to change hystart executed opt.val in flex1. 136 * 8 - Back out of CSS into SS, flex1 is the css_baseline_minrtt 137 */ 138 struct tcpcb *tp; 139 140 if (hystart_bblogs == 0) 141 return; 142 tp = ccv->ccvc.tcp; 143 if (tp->t_logstate != TCP_LOG_STATE_OFF) { 144 union tcp_log_stackspecific log; 145 struct timeval tv; 146 147 memset(&log, 0, sizeof(log)); 148 log.u_bbr.flex1 = flex1; 149 log.u_bbr.flex2 = nreno->css_current_round_minrtt; 150 log.u_bbr.flex3 = nreno->css_lastround_minrtt; 151 log.u_bbr.flex4 = nreno->css_rttsample_count; 152 log.u_bbr.flex5 = nreno->css_entered_at_round; 153 log.u_bbr.flex6 = nreno->css_baseline_minrtt; 154 /* We only need bottom 16 bits of flags */ 155 log.u_bbr.flex7 = nreno->newreno_flags & 0x0000ffff; 156 log.u_bbr.flex8 = mod; 157 log.u_bbr.epoch = nreno->css_current_round; 158 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 159 log.u_bbr.lt_epoch = nreno->css_fas_at_css_entry; 160 log.u_bbr.pkts_out = nreno->css_last_fas; 161 log.u_bbr.delivered = nreno->css_lowrtt_fas; 162 TCP_LOG_EVENTP(tp, NULL, 163 &tp->t_inpcb->inp_socket->so_rcv, 164 &tp->t_inpcb->inp_socket->so_snd, 165 TCP_HYSTART, 0, 166 0, &log, false, &tv); 167 } 168 } 169 170 static int 171 newreno_cb_init(struct cc_var *ccv) 172 { 173 struct newreno *nreno; 174 175 ccv->cc_data = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT); 176 if (ccv->cc_data == NULL) 177 return (ENOMEM); 178 nreno = (struct newreno *)ccv->cc_data; 179 /* NB: nreno is not zeroed, so initialise all fields. */ 180 nreno->beta = V_newreno_beta; 181 nreno->beta_ecn = V_newreno_beta_ecn; 182 /* 183 * We set the enabled flag so that if 184 * the socket option gets strobed and 185 * we have not hit a loss 186 */ 187 nreno->newreno_flags = CC_NEWRENO_HYSTART_ENABLED; 188 /* At init set both to infinity */ 189 nreno->css_lastround_minrtt = 0xffffffff; 190 nreno->css_current_round_minrtt = 0xffffffff; 191 nreno->css_current_round = 0; 192 nreno->css_baseline_minrtt = 0xffffffff; 193 nreno->css_rttsample_count = 0; 194 nreno->css_entered_at_round = 0; 195 nreno->css_fas_at_css_entry = 0; 196 nreno->css_lowrtt_fas = 0; 197 nreno->css_last_fas = 0; 198 return (0); 199 } 200 201 static void 202 newreno_cb_destroy(struct cc_var *ccv) 203 { 204 free(ccv->cc_data, M_NEWRENO); 205 } 206 207 static void 208 newreno_ack_received(struct cc_var *ccv, uint16_t type) 209 { 210 struct newreno *nreno; 211 212 /* 213 * Other TCP congestion controls use newreno_ack_received(), but 214 * with their own private cc_data. Make sure the cc_data is used 215 * correctly. 216 */ 217 nreno = (CC_ALGO(ccv->ccvc.tcp) == &newreno_cc_algo) ? ccv->cc_data : NULL; 218 219 if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && 220 (ccv->flags & CCF_CWND_LIMITED)) { 221 u_int cw = CCV(ccv, snd_cwnd); 222 u_int incr = CCV(ccv, t_maxseg); 223 224 /* 225 * Regular in-order ACK, open the congestion window. 226 * Method depends on which congestion control state we're 227 * in (slow start or cong avoid) and if ABC (RFC 3465) is 228 * enabled. 229 * 230 * slow start: cwnd <= ssthresh 231 * cong avoid: cwnd > ssthresh 232 * 233 * slow start and ABC (RFC 3465): 234 * Grow cwnd exponentially by the amount of data 235 * ACKed capping the max increment per ACK to 236 * (abc_l_var * maxseg) bytes. 237 * 238 * slow start without ABC (RFC 5681): 239 * Grow cwnd exponentially by maxseg per ACK. 240 * 241 * cong avoid and ABC (RFC 3465): 242 * Grow cwnd linearly by maxseg per RTT for each 243 * cwnd worth of ACKed data. 244 * 245 * cong avoid without ABC (RFC 5681): 246 * Grow cwnd linearly by approximately maxseg per RTT using 247 * maxseg^2 / cwnd per ACK as the increment. 248 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to 249 * avoid capping cwnd. 250 */ 251 if (cw > CCV(ccv, snd_ssthresh)) { 252 if ((nreno != NULL) && 253 (nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS)) { 254 /* 255 * We have slipped into CA with 256 * CSS active. Deactivate all. 257 */ 258 /* Turn off the CSS flag */ 259 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 260 /* Disable use of CSS in the future except long idle */ 261 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 262 } 263 if (V_tcp_do_rfc3465) { 264 if (ccv->flags & CCF_ABC_SENTAWND) 265 ccv->flags &= ~CCF_ABC_SENTAWND; 266 else 267 incr = 0; 268 } else 269 incr = max((incr * incr / cw), 1); 270 } else if (V_tcp_do_rfc3465) { 271 /* 272 * In slow-start with ABC enabled and no RTO in sight? 273 * (Must not use abc_l_var > 1 if slow starting after 274 * an RTO. On RTO, snd_nxt = snd_una, so the 275 * snd_nxt == snd_max check is sufficient to 276 * handle this). 277 * 278 * XXXLAS: Find a way to signal SS after RTO that 279 * doesn't rely on tcpcb vars. 280 */ 281 uint16_t abc_val; 282 283 if (ccv->flags & CCF_USE_LOCAL_ABC) 284 abc_val = ccv->labc; 285 else 286 abc_val = V_tcp_abc_l_var; 287 if ((nreno != NULL) && 288 (nreno->newreno_flags & CC_NEWRENO_HYSTART_ALLOWED) && 289 (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) && 290 ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) == 0)) { 291 /* 292 * Hystart is allowed and still enabled and we are not yet 293 * in CSS. Lets check to see if we can make a decision on 294 * if we need to go into CSS. 295 */ 296 if ((nreno->css_rttsample_count >= hystart_n_rttsamples) && 297 (CCV(ccv, snd_cwnd) > 298 (hystart_lowcwnd * tcp_fixed_maxseg(ccv->ccvc.tcp)))) { 299 uint32_t rtt_thresh; 300 301 /* Clamp (minrtt_thresh, lastround/8, maxrtt_thresh) */ 302 rtt_thresh = (nreno->css_lastround_minrtt >> 3); 303 if (rtt_thresh < hystart_minrtt_thresh) 304 rtt_thresh = hystart_minrtt_thresh; 305 if (rtt_thresh > hystart_maxrtt_thresh) 306 rtt_thresh = hystart_maxrtt_thresh; 307 newreno_log_hystart_event(ccv, nreno, 1, rtt_thresh); 308 if (nreno->css_current_round_minrtt >= (nreno->css_lastround_minrtt + rtt_thresh)) { 309 /* Enter CSS */ 310 nreno->newreno_flags |= CC_NEWRENO_HYSTART_IN_CSS; 311 nreno->css_fas_at_css_entry = nreno->css_lowrtt_fas; 312 nreno->css_baseline_minrtt = nreno->css_current_round_minrtt; 313 nreno->css_entered_at_round = nreno->css_current_round; 314 newreno_log_hystart_event(ccv, nreno, 2, rtt_thresh); 315 } 316 } 317 } 318 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) 319 incr = min(ccv->bytes_this_ack, 320 ccv->nsegs * abc_val * 321 CCV(ccv, t_maxseg)); 322 else 323 incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); 324 325 /* Only if Hystart is enabled will the flag get set */ 326 if ((nreno != NULL) && 327 (nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS)) { 328 incr /= hystart_css_growth_div; 329 newreno_log_hystart_event(ccv, nreno, 3, incr); 330 } 331 } 332 /* ABC is on by default, so incr equals 0 frequently. */ 333 if (incr > 0) 334 CCV(ccv, snd_cwnd) = min(cw + incr, 335 TCP_MAXWIN << CCV(ccv, snd_scale)); 336 } 337 } 338 339 static void 340 newreno_after_idle(struct cc_var *ccv) 341 { 342 struct newreno *nreno; 343 uint32_t rw; 344 345 /* 346 * Other TCP congestion controls use newreno_after_idle(), but 347 * with their own private cc_data. Make sure the cc_data is used 348 * correctly. 349 */ 350 nreno = (CC_ALGO(ccv->ccvc.tcp) == &newreno_cc_algo) ? ccv->cc_data : NULL; 351 /* 352 * If we've been idle for more than one retransmit timeout the old 353 * congestion window is no longer current and we have to reduce it to 354 * the restart window before we can transmit again. 355 * 356 * The restart window is the initial window or the last CWND, whichever 357 * is smaller. 358 * 359 * This is done to prevent us from flooding the path with a full CWND at 360 * wirespeed, overloading router and switch buffers along the way. 361 * 362 * See RFC5681 Section 4.1. "Restarting Idle Connections". 363 * 364 * In addition, per RFC2861 Section 2, the ssthresh is set to the 365 * maximum of the former ssthresh or 3/4 of the old cwnd, to 366 * not exit slow-start prematurely. 367 */ 368 rw = tcp_compute_initwnd(tcp_maxseg(ccv->ccvc.tcp)); 369 370 CCV(ccv, snd_ssthresh) = max(CCV(ccv, snd_ssthresh), 371 CCV(ccv, snd_cwnd)-(CCV(ccv, snd_cwnd)>>2)); 372 373 CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd)); 374 if ((nreno != NULL) && 375 (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) == 0) { 376 if (CCV(ccv, snd_cwnd) <= (hystart_lowcwnd * tcp_fixed_maxseg(ccv->ccvc.tcp))) { 377 /* 378 * Re-enable hystart if our cwnd has fallen below 379 * the hystart lowcwnd point. 380 */ 381 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 382 nreno->newreno_flags |= CC_NEWRENO_HYSTART_ENABLED; 383 } 384 } 385 } 386 387 /* 388 * Perform any necessary tasks before we enter congestion recovery. 389 */ 390 static void 391 newreno_cong_signal(struct cc_var *ccv, uint32_t type) 392 { 393 struct newreno *nreno; 394 uint32_t beta, beta_ecn, cwin, factor; 395 u_int mss; 396 397 cwin = CCV(ccv, snd_cwnd); 398 mss = tcp_fixed_maxseg(ccv->ccvc.tcp); 399 /* 400 * Other TCP congestion controls use newreno_cong_signal(), but 401 * with their own private cc_data. Make sure the cc_data is used 402 * correctly. 403 */ 404 nreno = (CC_ALGO(ccv->ccvc.tcp) == &newreno_cc_algo) ? ccv->cc_data : NULL; 405 beta = (nreno == NULL) ? V_newreno_beta : nreno->beta;; 406 beta_ecn = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn; 407 /* 408 * Note that we only change the backoff for ECN if the 409 * global sysctl V_cc_do_abe is set <or> the stack itself 410 * has set a flag in our newreno_flags (due to pacing) telling 411 * us to use the lower valued back-off. 412 */ 413 if ((type == CC_ECN) && 414 (V_cc_do_abe || 415 ((nreno != NULL) && (nreno->newreno_flags & CC_NEWRENO_BETA_ECN_ENABLED)))) 416 factor = beta_ecn; 417 else 418 factor = beta; 419 420 /* Catch algos which mistakenly leak private signal types. */ 421 KASSERT((type & CC_SIGPRIVMASK) == 0, 422 ("%s: congestion signal type 0x%08x is private\n", __func__, type)); 423 424 cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss), 425 2) * mss; 426 427 switch (type) { 428 case CC_NDUPACK: 429 if ((nreno != NULL) && 430 (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED)) { 431 /* Make sure the flags are all off we had a loss */ 432 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 433 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 434 } 435 if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { 436 if (IN_CONGRECOVERY(CCV(ccv, t_flags) && 437 V_cc_do_abe && V_cc_abe_frlossreduce)) { 438 CCV(ccv, snd_ssthresh) = 439 ((uint64_t)CCV(ccv, snd_ssthresh) * 440 (uint64_t)beta) / (uint64_t)beta_ecn; 441 } 442 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) 443 CCV(ccv, snd_ssthresh) = cwin; 444 ENTER_RECOVERY(CCV(ccv, t_flags)); 445 } 446 break; 447 case CC_ECN: 448 if ((nreno != NULL) && 449 (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED)) { 450 /* Make sure the flags are all off we had a loss */ 451 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 452 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 453 } 454 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { 455 CCV(ccv, snd_ssthresh) = cwin; 456 CCV(ccv, snd_cwnd) = cwin; 457 ENTER_CONGRECOVERY(CCV(ccv, t_flags)); 458 } 459 break; 460 case CC_RTO: 461 CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd), 462 CCV(ccv, snd_cwnd)) / 2 / mss, 463 2) * mss; 464 CCV(ccv, snd_cwnd) = mss; 465 break; 466 } 467 } 468 469 /* 470 * Perform any necessary tasks before we exit congestion recovery. 471 */ 472 static void 473 newreno_post_recovery(struct cc_var *ccv) 474 { 475 int pipe; 476 477 if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { 478 /* 479 * Fast recovery will conclude after returning from this 480 * function. Window inflation should have left us with 481 * approximately snd_ssthresh outstanding data. But in case we 482 * would be inclined to send a burst, better to do it via the 483 * slow start mechanism. 484 * 485 * XXXLAS: Find a way to do this without needing curack 486 */ 487 if (V_tcp_do_newsack) 488 pipe = tcp_compute_pipe(ccv->ccvc.tcp); 489 else 490 pipe = CCV(ccv, snd_max) - ccv->curack; 491 492 if (pipe < CCV(ccv, snd_ssthresh)) 493 /* 494 * Ensure that cwnd does not collapse to 1 MSS under 495 * adverse conditons. Implements RFC6582 496 */ 497 CCV(ccv, snd_cwnd) = max(pipe, CCV(ccv, t_maxseg)) + 498 CCV(ccv, t_maxseg); 499 else 500 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 501 } 502 } 503 504 static int 505 newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf) 506 { 507 struct newreno *nreno; 508 struct cc_newreno_opts *opt; 509 510 if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts)) 511 return (EMSGSIZE); 512 513 if (CC_ALGO(ccv->ccvc.tcp) != &newreno_cc_algo) 514 return (ENOPROTOOPT); 515 516 nreno = (struct newreno *)ccv->cc_data; 517 opt = buf; 518 switch (sopt->sopt_dir) { 519 case SOPT_SET: 520 switch (opt->name) { 521 case CC_NEWRENO_BETA: 522 nreno->beta = opt->val; 523 break; 524 case CC_NEWRENO_BETA_ECN: 525 if ((!V_cc_do_abe) && ((nreno->newreno_flags & CC_NEWRENO_BETA_ECN) == 0)) 526 return (EACCES); 527 nreno->beta_ecn = opt->val; 528 nreno->newreno_flags |= CC_NEWRENO_BETA_ECN_ENABLED; 529 break; 530 case CC_NEWRENO_ENABLE_HYSTART: 531 /* Allow hystart on this connection */ 532 if (opt->val != 0) { 533 nreno->newreno_flags |= CC_NEWRENO_HYSTART_ALLOWED; 534 if (opt->val > 1) 535 nreno->newreno_flags |= CC_NEWRENO_HYSTART_CAN_SH_CWND; 536 if (opt->val > 2) 537 nreno->newreno_flags |= CC_NEWRENO_HYSTART_CONS_SSTH; 538 } else 539 nreno->newreno_flags &= ~(CC_NEWRENO_HYSTART_ALLOWED|CC_NEWRENO_HYSTART_CAN_SH_CWND|CC_NEWRENO_HYSTART_CONS_SSTH); 540 newreno_log_hystart_event(ccv, nreno, 7, opt->val); 541 break; 542 default: 543 return (ENOPROTOOPT); 544 } 545 break; 546 case SOPT_GET: 547 switch (opt->name) { 548 case CC_NEWRENO_BETA: 549 opt->val = (nreno == NULL) ? 550 V_newreno_beta : nreno->beta; 551 break; 552 case CC_NEWRENO_BETA_ECN: 553 opt->val = (nreno == NULL) ? 554 V_newreno_beta_ecn : nreno->beta_ecn; 555 break; 556 case CC_NEWRENO_ENABLE_HYSTART: 557 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ALLOWED) { 558 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CONS_SSTH) 559 opt->val = 3; 560 else if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CAN_SH_CWND) 561 opt->val = 2; 562 else 563 opt->val = 1; 564 } else 565 opt->val = 0; 566 break; 567 default: 568 return (ENOPROTOOPT); 569 } 570 break; 571 default: 572 return (EINVAL); 573 } 574 575 return (0); 576 } 577 578 static int 579 newreno_beta_handler(SYSCTL_HANDLER_ARGS) 580 { 581 int error; 582 uint32_t new; 583 584 new = *(uint32_t *)arg1; 585 error = sysctl_handle_int(oidp, &new, 0, req); 586 if (error == 0 && req->newptr != NULL ) { 587 if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe) 588 error = EACCES; 589 else if (new == 0 || new > 100) 590 error = EINVAL; 591 else 592 *(uint32_t *)arg1 = new; 593 } 594 595 return (error); 596 } 597 598 static void 599 newreno_newround(struct cc_var *ccv, uint32_t round_cnt) 600 { 601 struct newreno *nreno; 602 603 nreno = (struct newreno *)ccv->cc_data; 604 /* We have entered a new round */ 605 nreno->css_lastround_minrtt = nreno->css_current_round_minrtt; 606 nreno->css_current_round_minrtt = 0xffffffff; 607 nreno->css_rttsample_count = 0; 608 nreno->css_current_round = round_cnt; 609 if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) && 610 ((round_cnt - nreno->css_entered_at_round) >= hystart_css_rounds)) { 611 /* Enter CA */ 612 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CAN_SH_CWND) { 613 /* 614 * We engage more than snd_ssthresh, engage 615 * the brakes!! Though we will stay in SS to 616 * creep back up again, so lets leave CSS active 617 * and give us hystart_css_rounds more rounds. 618 */ 619 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CONS_SSTH) { 620 CCV(ccv, snd_ssthresh) = ((nreno->css_lowrtt_fas + nreno->css_fas_at_css_entry) / 2); 621 } else { 622 CCV(ccv, snd_ssthresh) = nreno->css_lowrtt_fas; 623 } 624 CCV(ccv, snd_cwnd) = nreno->css_fas_at_css_entry; 625 nreno->css_entered_at_round = round_cnt; 626 } else { 627 CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); 628 /* Turn off the CSS flag */ 629 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 630 /* Disable use of CSS in the future except long idle */ 631 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 632 } 633 newreno_log_hystart_event(ccv, nreno, 6, CCV(ccv, snd_ssthresh)); 634 } 635 newreno_log_hystart_event(ccv, nreno, 4, round_cnt); 636 } 637 638 static void 639 newreno_rttsample(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas) 640 { 641 struct newreno *nreno; 642 643 nreno = (struct newreno *)ccv->cc_data; 644 if (rxtcnt > 1) { 645 /* 646 * Only look at RTT's that are non-ambiguous. 647 */ 648 return; 649 } 650 nreno->css_rttsample_count++; 651 nreno->css_last_fas = fas; 652 if (nreno->css_current_round_minrtt > usec_rtt) { 653 nreno->css_current_round_minrtt = usec_rtt; 654 nreno->css_lowrtt_fas = nreno->css_last_fas; 655 } 656 if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) && 657 (nreno->css_rttsample_count >= hystart_n_rttsamples) && 658 (nreno->css_baseline_minrtt > nreno->css_current_round_minrtt)) { 659 /* 660 * We were in CSS and the RTT is now less, we 661 * entered CSS erroneously. 662 */ 663 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 664 newreno_log_hystart_event(ccv, nreno, 8, nreno->css_baseline_minrtt); 665 nreno->css_baseline_minrtt = 0xffffffff; 666 } 667 newreno_log_hystart_event(ccv, nreno, 5, usec_rtt); 668 } 669 670 SYSCTL_DECL(_net_inet_tcp_cc_newreno); 671 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, newreno, 672 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 673 "New Reno related settings"); 674 675 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta, 676 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 677 &VNET_NAME(newreno_beta), 3, &newreno_beta_handler, "IU", 678 "New Reno beta, specified as number between 1 and 100"); 679 680 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn, 681 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 682 &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU", 683 "New Reno beta ecn, specified as number between 1 and 100"); 684 685 SYSCTL_NODE(_net_inet_tcp_cc_newreno, OID_AUTO, hystartplusplus, 686 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 687 "New Reno related HyStart++ settings"); 688 689 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, lowcwnd, 690 CTLFLAG_RW, 691 &hystart_lowcwnd, 16, 692 "The number of MSS in the CWND before HyStart++ is active"); 693 694 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, minrtt_thresh, 695 CTLFLAG_RW, 696 &hystart_minrtt_thresh, 4000, 697 "HyStarts++ minimum RTT thresh used in clamp (in microseconds)"); 698 699 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, maxrtt_thresh, 700 CTLFLAG_RW, 701 &hystart_maxrtt_thresh, 16000, 702 "HyStarts++ maximum RTT thresh used in clamp (in microseconds)"); 703 704 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, n_rttsamples, 705 CTLFLAG_RW, 706 &hystart_n_rttsamples, 8, 707 "The number of RTT samples that must be seen to consider HyStart++"); 708 709 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, css_growth_div, 710 CTLFLAG_RW, 711 &hystart_css_growth_div, 4, 712 "The divisor to the growth when in Hystart++ CSS"); 713 714 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, css_rounds, 715 CTLFLAG_RW, 716 &hystart_css_rounds, 5, 717 "The number of rounds HyStart++ lasts in CSS before falling to CA"); 718 719 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, bblogs, 720 CTLFLAG_RW, 721 &hystart_bblogs, 0, 722 "Do we enable HyStart++ Black Box logs to be generated if BB logging is on"); 723 724 725 DECLARE_CC_MODULE(newreno, &newreno_cc_algo); 726 MODULE_VERSION(newreno, 1); 727