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/socketvar.h> 67 #include <sys/sysctl.h> 68 #include <sys/systm.h> 69 70 #include <net/vnet.h> 71 72 #include <netinet/tcp.h> 73 #include <netinet/tcp_seq.h> 74 #include <netinet/tcp_var.h> 75 #include <netinet/cc/cc.h> 76 #include <netinet/cc/cc_module.h> 77 #include <netinet/cc/cc_newreno.h> 78 79 static MALLOC_DEFINE(M_NEWRENO, "newreno data", 80 "newreno beta values"); 81 82 static void newreno_cb_destroy(struct cc_var *ccv); 83 static void newreno_ack_received(struct cc_var *ccv, uint16_t type); 84 static void newreno_after_idle(struct cc_var *ccv); 85 static void newreno_cong_signal(struct cc_var *ccv, uint32_t type); 86 static void newreno_post_recovery(struct cc_var *ccv); 87 static int newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf); 88 89 VNET_DEFINE(uint32_t, newreno_beta) = 50; 90 VNET_DEFINE(uint32_t, newreno_beta_ecn) = 80; 91 #define V_newreno_beta VNET(newreno_beta) 92 #define V_newreno_beta_ecn VNET(newreno_beta_ecn) 93 94 struct cc_algo newreno_cc_algo = { 95 .name = "newreno", 96 .cb_destroy = newreno_cb_destroy, 97 .ack_received = newreno_ack_received, 98 .after_idle = newreno_after_idle, 99 .cong_signal = newreno_cong_signal, 100 .post_recovery = newreno_post_recovery, 101 .ctl_output = newreno_ctl_output, 102 }; 103 104 static inline struct newreno * 105 newreno_malloc(struct cc_var *ccv) 106 { 107 struct newreno *nreno; 108 109 nreno = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT); 110 if (nreno != NULL) { 111 /* NB: nreno is not zeroed, so initialise all fields. */ 112 nreno->beta = V_newreno_beta; 113 nreno->beta_ecn = V_newreno_beta_ecn; 114 ccv->cc_data = nreno; 115 } 116 117 return (nreno); 118 } 119 120 static void 121 newreno_cb_destroy(struct cc_var *ccv) 122 { 123 free(ccv->cc_data, M_NEWRENO); 124 } 125 126 static void 127 newreno_ack_received(struct cc_var *ccv, uint16_t type) 128 { 129 if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && 130 (ccv->flags & CCF_CWND_LIMITED)) { 131 u_int cw = CCV(ccv, snd_cwnd); 132 u_int incr = CCV(ccv, t_maxseg); 133 134 /* 135 * Regular in-order ACK, open the congestion window. 136 * Method depends on which congestion control state we're 137 * in (slow start or cong avoid) and if ABC (RFC 3465) is 138 * enabled. 139 * 140 * slow start: cwnd <= ssthresh 141 * cong avoid: cwnd > ssthresh 142 * 143 * slow start and ABC (RFC 3465): 144 * Grow cwnd exponentially by the amount of data 145 * ACKed capping the max increment per ACK to 146 * (abc_l_var * maxseg) bytes. 147 * 148 * slow start without ABC (RFC 5681): 149 * Grow cwnd exponentially by maxseg per ACK. 150 * 151 * cong avoid and ABC (RFC 3465): 152 * Grow cwnd linearly by maxseg per RTT for each 153 * cwnd worth of ACKed data. 154 * 155 * cong avoid without ABC (RFC 5681): 156 * Grow cwnd linearly by approximately maxseg per RTT using 157 * maxseg^2 / cwnd per ACK as the increment. 158 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to 159 * avoid capping cwnd. 160 */ 161 if (cw > CCV(ccv, snd_ssthresh)) { 162 if (V_tcp_do_rfc3465) { 163 if (ccv->flags & CCF_ABC_SENTAWND) 164 ccv->flags &= ~CCF_ABC_SENTAWND; 165 else 166 incr = 0; 167 } else 168 incr = max((incr * incr / cw), 1); 169 } else if (V_tcp_do_rfc3465) { 170 /* 171 * In slow-start with ABC enabled and no RTO in sight? 172 * (Must not use abc_l_var > 1 if slow starting after 173 * an RTO. On RTO, snd_nxt = snd_una, so the 174 * snd_nxt == snd_max check is sufficient to 175 * handle this). 176 * 177 * XXXLAS: Find a way to signal SS after RTO that 178 * doesn't rely on tcpcb vars. 179 */ 180 uint16_t abc_val; 181 182 if (ccv->flags & CCF_USE_LOCAL_ABC) 183 abc_val = ccv->labc; 184 else 185 abc_val = V_tcp_abc_l_var; 186 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) 187 incr = min(ccv->bytes_this_ack, 188 ccv->nsegs * abc_val * 189 CCV(ccv, t_maxseg)); 190 else 191 incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); 192 } 193 /* ABC is on by default, so incr equals 0 frequently. */ 194 if (incr > 0) 195 CCV(ccv, snd_cwnd) = min(cw + incr, 196 TCP_MAXWIN << CCV(ccv, snd_scale)); 197 } 198 } 199 200 static void 201 newreno_after_idle(struct cc_var *ccv) 202 { 203 uint32_t rw; 204 205 /* 206 * If we've been idle for more than one retransmit timeout the old 207 * congestion window is no longer current and we have to reduce it to 208 * the restart window before we can transmit again. 209 * 210 * The restart window is the initial window or the last CWND, whichever 211 * is smaller. 212 * 213 * This is done to prevent us from flooding the path with a full CWND at 214 * wirespeed, overloading router and switch buffers along the way. 215 * 216 * See RFC5681 Section 4.1. "Restarting Idle Connections". 217 * 218 * In addition, per RFC2861 Section 2, the ssthresh is set to the 219 * maximum of the former ssthresh or 3/4 of the old cwnd, to 220 * not exit slow-start prematurely. 221 */ 222 rw = tcp_compute_initwnd(tcp_maxseg(ccv->ccvc.tcp)); 223 224 CCV(ccv, snd_ssthresh) = max(CCV(ccv, snd_ssthresh), 225 CCV(ccv, snd_cwnd)-(CCV(ccv, snd_cwnd)>>2)); 226 227 CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd)); 228 } 229 230 /* 231 * Perform any necessary tasks before we enter congestion recovery. 232 */ 233 static void 234 newreno_cong_signal(struct cc_var *ccv, uint32_t type) 235 { 236 struct newreno *nreno; 237 uint32_t beta, beta_ecn, cwin, factor; 238 u_int mss; 239 240 cwin = CCV(ccv, snd_cwnd); 241 mss = tcp_fixed_maxseg(ccv->ccvc.tcp); 242 /* 243 * Other TCP congestion controls use newreno_cong_signal(), but 244 * with their own private cc_data. Make sure the cc_data is used 245 * correctly. 246 */ 247 nreno = (CC_ALGO(ccv->ccvc.tcp) == &newreno_cc_algo) ? ccv->cc_data : NULL; 248 beta = (nreno == NULL) ? V_newreno_beta : nreno->beta; 249 beta_ecn = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn; 250 251 /* 252 * Note that we only change the backoff for ECN if the 253 * global sysctl V_cc_do_abe is set <or> the stack itself 254 * has set a flag in our newreno_flags (due to pacing) telling 255 * us to use the lower valued back-off. 256 */ 257 if (V_cc_do_abe || 258 (nreno && (nreno->newreno_flags & CC_NEWRENO_BETA_ECN) && (type == CC_ECN))) 259 factor = beta_ecn; 260 else 261 factor = beta; 262 263 /* Catch algos which mistakenly leak private signal types. */ 264 KASSERT((type & CC_SIGPRIVMASK) == 0, 265 ("%s: congestion signal type 0x%08x is private\n", __func__, type)); 266 267 cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss), 268 2) * mss; 269 270 switch (type) { 271 case CC_NDUPACK: 272 if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { 273 if (IN_CONGRECOVERY(CCV(ccv, t_flags) && 274 V_cc_do_abe && V_cc_abe_frlossreduce)) { 275 CCV(ccv, snd_ssthresh) = 276 ((uint64_t)CCV(ccv, snd_ssthresh) * 277 (uint64_t)beta) / (uint64_t)beta_ecn; 278 } 279 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) 280 CCV(ccv, snd_ssthresh) = cwin; 281 ENTER_RECOVERY(CCV(ccv, t_flags)); 282 } 283 break; 284 case CC_ECN: 285 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { 286 CCV(ccv, snd_ssthresh) = cwin; 287 CCV(ccv, snd_cwnd) = cwin; 288 ENTER_CONGRECOVERY(CCV(ccv, t_flags)); 289 } 290 break; 291 case CC_RTO: 292 CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd), 293 CCV(ccv, snd_cwnd)) / 2 / mss, 294 2) * mss; 295 CCV(ccv, snd_cwnd) = mss; 296 break; 297 } 298 } 299 300 /* 301 * Perform any necessary tasks before we exit congestion recovery. 302 */ 303 static void 304 newreno_post_recovery(struct cc_var *ccv) 305 { 306 int pipe; 307 308 if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { 309 /* 310 * Fast recovery will conclude after returning from this 311 * function. Window inflation should have left us with 312 * approximately snd_ssthresh outstanding data. But in case we 313 * would be inclined to send a burst, better to do it via the 314 * slow start mechanism. 315 * 316 * XXXLAS: Find a way to do this without needing curack 317 */ 318 if (V_tcp_do_newsack) 319 pipe = tcp_compute_pipe(ccv->ccvc.tcp); 320 else 321 pipe = CCV(ccv, snd_max) - ccv->curack; 322 323 if (pipe < CCV(ccv, snd_ssthresh)) 324 /* 325 * Ensure that cwnd does not collapse to 1 MSS under 326 * adverse conditons. Implements RFC6582 327 */ 328 CCV(ccv, snd_cwnd) = max(pipe, CCV(ccv, t_maxseg)) + 329 CCV(ccv, t_maxseg); 330 else 331 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 332 } 333 } 334 335 static int 336 newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf) 337 { 338 struct newreno *nreno; 339 struct cc_newreno_opts *opt; 340 341 if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts)) 342 return (EMSGSIZE); 343 344 if (CC_ALGO(ccv->ccvc.tcp) != &newreno_cc_algo) 345 return (ENOPROTOOPT); 346 347 nreno = ccv->cc_data; 348 opt = buf; 349 350 switch (sopt->sopt_dir) { 351 case SOPT_SET: 352 /* We cannot set without cc_data memory. */ 353 if (nreno == NULL) { 354 nreno = newreno_malloc(ccv); 355 if (nreno == NULL) 356 return (ENOMEM); 357 } 358 switch (opt->name) { 359 case CC_NEWRENO_BETA: 360 nreno->beta = opt->val; 361 break; 362 case CC_NEWRENO_BETA_ECN: 363 if ((!V_cc_do_abe) && ((nreno->newreno_flags & CC_NEWRENO_BETA_ECN) == 0)) 364 return (EACCES); 365 nreno->beta_ecn = opt->val; 366 break; 367 default: 368 return (ENOPROTOOPT); 369 } 370 break; 371 case SOPT_GET: 372 switch (opt->name) { 373 case CC_NEWRENO_BETA: 374 opt->val = (nreno == NULL) ? 375 V_newreno_beta : nreno->beta; 376 break; 377 case CC_NEWRENO_BETA_ECN: 378 opt->val = (nreno == NULL) ? 379 V_newreno_beta_ecn : nreno->beta_ecn; 380 break; 381 default: 382 return (ENOPROTOOPT); 383 } 384 break; 385 default: 386 return (EINVAL); 387 } 388 389 return (0); 390 } 391 392 static int 393 newreno_beta_handler(SYSCTL_HANDLER_ARGS) 394 { 395 int error; 396 uint32_t new; 397 398 new = *(uint32_t *)arg1; 399 error = sysctl_handle_int(oidp, &new, 0, req); 400 if (error == 0 && req->newptr != NULL ) { 401 if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe) 402 error = EACCES; 403 else if (new == 0 || new > 100) 404 error = EINVAL; 405 else 406 *(uint32_t *)arg1 = new; 407 } 408 409 return (error); 410 } 411 412 SYSCTL_DECL(_net_inet_tcp_cc_newreno); 413 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, newreno, 414 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 415 "New Reno related settings"); 416 417 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta, 418 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 419 &VNET_NAME(newreno_beta), 3, &newreno_beta_handler, "IU", 420 "New Reno beta, specified as number between 1 and 100"); 421 422 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn, 423 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 424 &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU", 425 "New Reno beta ecn, specified as number between 1 and 100"); 426 427 DECLARE_CC_MODULE(newreno, &newreno_cc_algo); 428 MODULE_VERSION(newreno, 1); 429