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 = NULL; 176 ccv->cc_data = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT); 177 if (ccv->cc_data == NULL) 178 return (ENOMEM); 179 nreno = (struct newreno *)ccv->cc_data; 180 /* NB: nreno is not zeroed, so initialise all fields. */ 181 nreno->beta = V_newreno_beta; 182 nreno->beta_ecn = V_newreno_beta_ecn; 183 /* 184 * We set the enabled flag so that if 185 * the socket option gets strobed and 186 * we have not hit a loss 187 */ 188 nreno->newreno_flags = CC_NEWRENO_HYSTART_ENABLED; 189 /* At init set both to infinity */ 190 nreno->css_lastround_minrtt = 0xffffffff; 191 nreno->css_current_round_minrtt = 0xffffffff; 192 nreno->css_current_round = 0; 193 nreno->css_baseline_minrtt = 0xffffffff; 194 nreno->css_rttsample_count = 0; 195 nreno->css_entered_at_round = 0; 196 nreno->css_fas_at_css_entry = 0; 197 nreno->css_lowrtt_fas = 0; 198 nreno->css_last_fas = 0; 199 return (0); 200 } 201 202 static void 203 newreno_cb_destroy(struct cc_var *ccv) 204 { 205 free(ccv->cc_data, M_NEWRENO); 206 } 207 208 static void 209 newreno_ack_received(struct cc_var *ccv, uint16_t type) 210 { 211 struct newreno *nreno; 212 213 nreno = (struct newreno *)ccv->cc_data; 214 if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && 215 (ccv->flags & CCF_CWND_LIMITED)) { 216 u_int cw = CCV(ccv, snd_cwnd); 217 u_int incr = CCV(ccv, t_maxseg); 218 219 /* 220 * Regular in-order ACK, open the congestion window. 221 * Method depends on which congestion control state we're 222 * in (slow start or cong avoid) and if ABC (RFC 3465) is 223 * enabled. 224 * 225 * slow start: cwnd <= ssthresh 226 * cong avoid: cwnd > ssthresh 227 * 228 * slow start and ABC (RFC 3465): 229 * Grow cwnd exponentially by the amount of data 230 * ACKed capping the max increment per ACK to 231 * (abc_l_var * maxseg) bytes. 232 * 233 * slow start without ABC (RFC 5681): 234 * Grow cwnd exponentially by maxseg per ACK. 235 * 236 * cong avoid and ABC (RFC 3465): 237 * Grow cwnd linearly by maxseg per RTT for each 238 * cwnd worth of ACKed data. 239 * 240 * cong avoid without ABC (RFC 5681): 241 * Grow cwnd linearly by approximately maxseg per RTT using 242 * maxseg^2 / cwnd per ACK as the increment. 243 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to 244 * avoid capping cwnd. 245 */ 246 if (cw > CCV(ccv, snd_ssthresh)) { 247 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) { 248 /* 249 * We have slipped into CA with 250 * CSS active. Deactivate all. 251 */ 252 /* Turn off the CSS flag */ 253 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 254 /* Disable use of CSS in the future except long idle */ 255 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 256 } 257 if (V_tcp_do_rfc3465) { 258 if (ccv->flags & CCF_ABC_SENTAWND) 259 ccv->flags &= ~CCF_ABC_SENTAWND; 260 else 261 incr = 0; 262 } else 263 incr = max((incr * incr / cw), 1); 264 } else if (V_tcp_do_rfc3465) { 265 /* 266 * In slow-start with ABC enabled and no RTO in sight? 267 * (Must not use abc_l_var > 1 if slow starting after 268 * an RTO. On RTO, snd_nxt = snd_una, so the 269 * snd_nxt == snd_max check is sufficient to 270 * handle this). 271 * 272 * XXXLAS: Find a way to signal SS after RTO that 273 * doesn't rely on tcpcb vars. 274 */ 275 uint16_t abc_val; 276 277 if (ccv->flags & CCF_USE_LOCAL_ABC) 278 abc_val = ccv->labc; 279 else 280 abc_val = V_tcp_abc_l_var; 281 if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_ALLOWED) && 282 (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) && 283 ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) == 0)) { 284 /* 285 * Hystart is allowed and still enabled and we are not yet 286 * in CSS. Lets check to see if we can make a decision on 287 * if we need to go into CSS. 288 */ 289 if ((nreno->css_rttsample_count >= hystart_n_rttsamples) && 290 (CCV(ccv, snd_cwnd) > 291 (hystart_lowcwnd * tcp_fixed_maxseg(ccv->ccvc.tcp)))) { 292 uint32_t rtt_thresh; 293 294 /* Clamp (minrtt_thresh, lastround/8, maxrtt_thresh) */ 295 rtt_thresh = (nreno->css_lastround_minrtt >> 3); 296 if (rtt_thresh < hystart_minrtt_thresh) 297 rtt_thresh = hystart_minrtt_thresh; 298 if (rtt_thresh > hystart_maxrtt_thresh) 299 rtt_thresh = hystart_maxrtt_thresh; 300 newreno_log_hystart_event(ccv, nreno, 1, rtt_thresh); 301 if (nreno->css_current_round_minrtt >= (nreno->css_lastround_minrtt + rtt_thresh)) { 302 /* Enter CSS */ 303 nreno->newreno_flags |= CC_NEWRENO_HYSTART_IN_CSS; 304 nreno->css_fas_at_css_entry = nreno->css_lowrtt_fas; 305 nreno->css_baseline_minrtt = nreno->css_current_round_minrtt; 306 nreno->css_entered_at_round = nreno->css_current_round; 307 newreno_log_hystart_event(ccv, nreno, 2, rtt_thresh); 308 } 309 } 310 } 311 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) 312 incr = min(ccv->bytes_this_ack, 313 ccv->nsegs * abc_val * 314 CCV(ccv, t_maxseg)); 315 else 316 incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); 317 318 /* Only if Hystart is enabled will the flag get set */ 319 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) { 320 incr /= hystart_css_growth_div; 321 newreno_log_hystart_event(ccv, nreno, 3, incr); 322 } 323 } 324 /* ABC is on by default, so incr equals 0 frequently. */ 325 if (incr > 0) 326 CCV(ccv, snd_cwnd) = min(cw + incr, 327 TCP_MAXWIN << CCV(ccv, snd_scale)); 328 } 329 } 330 331 static void 332 newreno_after_idle(struct cc_var *ccv) 333 { 334 struct newreno *nreno; 335 uint32_t rw; 336 337 nreno = (struct newreno *)ccv->cc_data; 338 /* 339 * If we've been idle for more than one retransmit timeout the old 340 * congestion window is no longer current and we have to reduce it to 341 * the restart window before we can transmit again. 342 * 343 * The restart window is the initial window or the last CWND, whichever 344 * is smaller. 345 * 346 * This is done to prevent us from flooding the path with a full CWND at 347 * wirespeed, overloading router and switch buffers along the way. 348 * 349 * See RFC5681 Section 4.1. "Restarting Idle Connections". 350 * 351 * In addition, per RFC2861 Section 2, the ssthresh is set to the 352 * maximum of the former ssthresh or 3/4 of the old cwnd, to 353 * not exit slow-start prematurely. 354 */ 355 rw = tcp_compute_initwnd(tcp_maxseg(ccv->ccvc.tcp)); 356 357 CCV(ccv, snd_ssthresh) = max(CCV(ccv, snd_ssthresh), 358 CCV(ccv, snd_cwnd)-(CCV(ccv, snd_cwnd)>>2)); 359 360 CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd)); 361 if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) == 0) { 362 if (CCV(ccv, snd_cwnd) <= (hystart_lowcwnd * tcp_fixed_maxseg(ccv->ccvc.tcp))) { 363 /* 364 * Re-enable hystart if our cwnd has fallen below 365 * the hystart lowcwnd point. 366 */ 367 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 368 nreno->newreno_flags |= CC_NEWRENO_HYSTART_ENABLED; 369 } 370 } 371 } 372 373 /* 374 * Perform any necessary tasks before we enter congestion recovery. 375 */ 376 static void 377 newreno_cong_signal(struct cc_var *ccv, uint32_t type) 378 { 379 struct newreno *nreno; 380 uint32_t beta, beta_ecn, cwin, factor; 381 u_int mss; 382 383 cwin = CCV(ccv, snd_cwnd); 384 mss = tcp_fixed_maxseg(ccv->ccvc.tcp); 385 nreno = (struct newreno *) ccv->cc_data; 386 beta = nreno->beta; 387 beta_ecn = nreno->beta_ecn; 388 /* 389 * Note that we only change the backoff for ECN if the 390 * global sysctl V_cc_do_abe is set <or> the stack itself 391 * has set a flag in our newreno_flags (due to pacing) telling 392 * us to use the lower valued back-off. 393 */ 394 if ((type == CC_ECN) && 395 (V_cc_do_abe || 396 ((nreno != NULL) && (nreno->newreno_flags & CC_NEWRENO_BETA_ECN_ENABLED)))) 397 factor = beta_ecn; 398 else 399 factor = beta; 400 401 /* Catch algos which mistakenly leak private signal types. */ 402 KASSERT((type & CC_SIGPRIVMASK) == 0, 403 ("%s: congestion signal type 0x%08x is private\n", __func__, type)); 404 405 cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss), 406 2) * mss; 407 408 switch (type) { 409 case CC_NDUPACK: 410 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) { 411 /* Make sure the flags are all off we had a loss */ 412 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 413 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 414 } 415 if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { 416 if (IN_CONGRECOVERY(CCV(ccv, t_flags) && 417 V_cc_do_abe && V_cc_abe_frlossreduce)) { 418 CCV(ccv, snd_ssthresh) = 419 ((uint64_t)CCV(ccv, snd_ssthresh) * 420 (uint64_t)beta) / (uint64_t)beta_ecn; 421 } 422 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) 423 CCV(ccv, snd_ssthresh) = cwin; 424 ENTER_RECOVERY(CCV(ccv, t_flags)); 425 } 426 break; 427 case CC_ECN: 428 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) { 429 /* Make sure the flags are all off we had a loss */ 430 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 431 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 432 } 433 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { 434 CCV(ccv, snd_ssthresh) = cwin; 435 CCV(ccv, snd_cwnd) = cwin; 436 ENTER_CONGRECOVERY(CCV(ccv, t_flags)); 437 } 438 break; 439 case CC_RTO: 440 CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd), 441 CCV(ccv, snd_cwnd)) / 2 / mss, 442 2) * mss; 443 CCV(ccv, snd_cwnd) = mss; 444 break; 445 } 446 } 447 448 /* 449 * Perform any necessary tasks before we exit congestion recovery. 450 */ 451 static void 452 newreno_post_recovery(struct cc_var *ccv) 453 { 454 int pipe; 455 456 if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { 457 /* 458 * Fast recovery will conclude after returning from this 459 * function. Window inflation should have left us with 460 * approximately snd_ssthresh outstanding data. But in case we 461 * would be inclined to send a burst, better to do it via the 462 * slow start mechanism. 463 * 464 * XXXLAS: Find a way to do this without needing curack 465 */ 466 if (V_tcp_do_newsack) 467 pipe = tcp_compute_pipe(ccv->ccvc.tcp); 468 else 469 pipe = CCV(ccv, snd_max) - ccv->curack; 470 471 if (pipe < CCV(ccv, snd_ssthresh)) 472 /* 473 * Ensure that cwnd does not collapse to 1 MSS under 474 * adverse conditons. Implements RFC6582 475 */ 476 CCV(ccv, snd_cwnd) = max(pipe, CCV(ccv, t_maxseg)) + 477 CCV(ccv, t_maxseg); 478 else 479 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 480 } 481 } 482 483 static int 484 newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf) 485 { 486 struct newreno *nreno; 487 struct cc_newreno_opts *opt; 488 489 if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts)) 490 return (EMSGSIZE); 491 492 if (CC_ALGO(ccv->ccvc.tcp) != &newreno_cc_algo) 493 return (ENOPROTOOPT); 494 495 nreno = (struct newreno *)ccv->cc_data; 496 opt = buf; 497 switch (sopt->sopt_dir) { 498 case SOPT_SET: 499 switch (opt->name) { 500 case CC_NEWRENO_BETA: 501 nreno->beta = opt->val; 502 break; 503 case CC_NEWRENO_BETA_ECN: 504 if ((!V_cc_do_abe) && ((nreno->newreno_flags & CC_NEWRENO_BETA_ECN) == 0)) 505 return (EACCES); 506 nreno->beta_ecn = opt->val; 507 nreno->newreno_flags |= CC_NEWRENO_BETA_ECN_ENABLED; 508 break; 509 case CC_NEWRENO_ENABLE_HYSTART: 510 /* Allow hystart on this connection */ 511 if (opt->val != 0) { 512 nreno->newreno_flags |= CC_NEWRENO_HYSTART_ALLOWED; 513 if (opt->val > 1) 514 nreno->newreno_flags |= CC_NEWRENO_HYSTART_CAN_SH_CWND; 515 if (opt->val > 2) 516 nreno->newreno_flags |= CC_NEWRENO_HYSTART_CONS_SSTH; 517 } else 518 nreno->newreno_flags &= ~(CC_NEWRENO_HYSTART_ALLOWED|CC_NEWRENO_HYSTART_CAN_SH_CWND|CC_NEWRENO_HYSTART_CONS_SSTH); 519 newreno_log_hystart_event(ccv, nreno, 7, opt->val); 520 break; 521 default: 522 return (ENOPROTOOPT); 523 } 524 break; 525 case SOPT_GET: 526 switch (opt->name) { 527 case CC_NEWRENO_BETA: 528 opt->val = (nreno == NULL) ? 529 V_newreno_beta : nreno->beta; 530 break; 531 case CC_NEWRENO_BETA_ECN: 532 opt->val = (nreno == NULL) ? 533 V_newreno_beta_ecn : nreno->beta_ecn; 534 break; 535 case CC_NEWRENO_ENABLE_HYSTART: 536 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ALLOWED) { 537 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CONS_SSTH) 538 opt->val = 3; 539 else if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CAN_SH_CWND) 540 opt->val = 2; 541 else 542 opt->val = 1; 543 } else 544 opt->val = 0; 545 break; 546 default: 547 return (ENOPROTOOPT); 548 } 549 break; 550 default: 551 return (EINVAL); 552 } 553 554 return (0); 555 } 556 557 static int 558 newreno_beta_handler(SYSCTL_HANDLER_ARGS) 559 { 560 int error; 561 uint32_t new; 562 563 new = *(uint32_t *)arg1; 564 error = sysctl_handle_int(oidp, &new, 0, req); 565 if (error == 0 && req->newptr != NULL ) { 566 if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe) 567 error = EACCES; 568 else if (new == 0 || new > 100) 569 error = EINVAL; 570 else 571 *(uint32_t *)arg1 = new; 572 } 573 574 return (error); 575 } 576 577 static void 578 newreno_newround(struct cc_var *ccv, uint32_t round_cnt) 579 { 580 struct newreno *nreno; 581 582 nreno = (struct newreno *)ccv->cc_data; 583 /* We have entered a new round */ 584 nreno->css_lastround_minrtt = nreno->css_current_round_minrtt; 585 nreno->css_current_round_minrtt = 0xffffffff; 586 nreno->css_rttsample_count = 0; 587 nreno->css_current_round = round_cnt; 588 if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) && 589 ((round_cnt - nreno->css_entered_at_round) >= hystart_css_rounds)) { 590 /* Enter CA */ 591 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CAN_SH_CWND) { 592 /* 593 * We engage more than snd_ssthresh, engage 594 * the brakes!! Though we will stay in SS to 595 * creep back up again, so lets leave CSS active 596 * and give us hystart_css_rounds more rounds. 597 */ 598 if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CONS_SSTH) { 599 CCV(ccv, snd_ssthresh) = ((nreno->css_lowrtt_fas + nreno->css_fas_at_css_entry) / 2); 600 } else { 601 CCV(ccv, snd_ssthresh) = nreno->css_lowrtt_fas; 602 } 603 CCV(ccv, snd_cwnd) = nreno->css_fas_at_css_entry; 604 nreno->css_entered_at_round = round_cnt; 605 } else { 606 CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); 607 /* Turn off the CSS flag */ 608 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 609 /* Disable use of CSS in the future except long idle */ 610 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED; 611 } 612 newreno_log_hystart_event(ccv, nreno, 6, CCV(ccv, snd_ssthresh)); 613 } 614 newreno_log_hystart_event(ccv, nreno, 4, round_cnt); 615 } 616 617 static void 618 newreno_rttsample(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas) 619 { 620 struct newreno *nreno; 621 622 nreno = (struct newreno *)ccv->cc_data; 623 if (rxtcnt > 1) { 624 /* 625 * Only look at RTT's that are non-ambiguous. 626 */ 627 return; 628 } 629 nreno->css_rttsample_count++; 630 nreno->css_last_fas = fas; 631 if (nreno->css_current_round_minrtt > usec_rtt) { 632 nreno->css_current_round_minrtt = usec_rtt; 633 nreno->css_lowrtt_fas = nreno->css_last_fas; 634 } 635 if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) && 636 (nreno->css_rttsample_count >= hystart_n_rttsamples) && 637 (nreno->css_baseline_minrtt > nreno->css_current_round_minrtt)) { 638 /* 639 * We were in CSS and the RTT is now less, we 640 * entered CSS erroneously. 641 */ 642 nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS; 643 newreno_log_hystart_event(ccv, nreno, 8, nreno->css_baseline_minrtt); 644 nreno->css_baseline_minrtt = 0xffffffff; 645 } 646 newreno_log_hystart_event(ccv, nreno, 5, usec_rtt); 647 } 648 649 SYSCTL_DECL(_net_inet_tcp_cc_newreno); 650 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, newreno, 651 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 652 "New Reno related settings"); 653 654 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta, 655 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 656 &VNET_NAME(newreno_beta), 3, &newreno_beta_handler, "IU", 657 "New Reno beta, specified as number between 1 and 100"); 658 659 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn, 660 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 661 &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU", 662 "New Reno beta ecn, specified as number between 1 and 100"); 663 664 SYSCTL_NODE(_net_inet_tcp_cc_newreno, OID_AUTO, hystartplusplus, 665 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 666 "New Reno related HyStart++ settings"); 667 668 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, lowcwnd, 669 CTLFLAG_RW, 670 &hystart_lowcwnd, 16, 671 "The number of MSS in the CWND before HyStart++ is active"); 672 673 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, minrtt_thresh, 674 CTLFLAG_RW, 675 &hystart_minrtt_thresh, 4000, 676 "HyStarts++ minimum RTT thresh used in clamp (in microseconds)"); 677 678 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, maxrtt_thresh, 679 CTLFLAG_RW, 680 &hystart_maxrtt_thresh, 16000, 681 "HyStarts++ maximum RTT thresh used in clamp (in microseconds)"); 682 683 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, n_rttsamples, 684 CTLFLAG_RW, 685 &hystart_n_rttsamples, 8, 686 "The number of RTT samples that must be seen to consider HyStart++"); 687 688 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, css_growth_div, 689 CTLFLAG_RW, 690 &hystart_css_growth_div, 4, 691 "The divisor to the growth when in Hystart++ CSS"); 692 693 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, css_rounds, 694 CTLFLAG_RW, 695 &hystart_css_rounds, 5, 696 "The number of rounds HyStart++ lasts in CSS before falling to CA"); 697 698 SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, bblogs, 699 CTLFLAG_RW, 700 &hystart_bblogs, 0, 701 "Do we enable HyStart++ Black Box logs to be generated if BB logging is on"); 702 703 704 DECLARE_CC_MODULE(newreno, &newreno_cc_algo); 705 MODULE_VERSION(newreno, 1); 706