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_STATIC(uint32_t, newreno_beta) = 50; 90 VNET_DEFINE_STATIC(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 struct newreno { 105 uint32_t beta; 106 uint32_t beta_ecn; 107 }; 108 109 static inline struct newreno * 110 newreno_malloc(struct cc_var *ccv) 111 { 112 struct newreno *nreno; 113 114 nreno = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT); 115 if (nreno != NULL) { 116 /* NB: nreno is not zeroed, so initialise all fields. */ 117 nreno->beta = V_newreno_beta; 118 nreno->beta_ecn = V_newreno_beta_ecn; 119 ccv->cc_data = nreno; 120 } 121 122 return (nreno); 123 } 124 125 static void 126 newreno_cb_destroy(struct cc_var *ccv) 127 { 128 free(ccv->cc_data, M_NEWRENO); 129 } 130 131 static void 132 newreno_ack_received(struct cc_var *ccv, uint16_t type) 133 { 134 if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && 135 (ccv->flags & CCF_CWND_LIMITED)) { 136 u_int cw = CCV(ccv, snd_cwnd); 137 u_int incr = CCV(ccv, t_maxseg); 138 139 /* 140 * Regular in-order ACK, open the congestion window. 141 * Method depends on which congestion control state we're 142 * in (slow start or cong avoid) and if ABC (RFC 3465) is 143 * enabled. 144 * 145 * slow start: cwnd <= ssthresh 146 * cong avoid: cwnd > ssthresh 147 * 148 * slow start and ABC (RFC 3465): 149 * Grow cwnd exponentially by the amount of data 150 * ACKed capping the max increment per ACK to 151 * (abc_l_var * maxseg) bytes. 152 * 153 * slow start without ABC (RFC 5681): 154 * Grow cwnd exponentially by maxseg per ACK. 155 * 156 * cong avoid and ABC (RFC 3465): 157 * Grow cwnd linearly by maxseg per RTT for each 158 * cwnd worth of ACKed data. 159 * 160 * cong avoid without ABC (RFC 5681): 161 * Grow cwnd linearly by approximately maxseg per RTT using 162 * maxseg^2 / cwnd per ACK as the increment. 163 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to 164 * avoid capping cwnd. 165 */ 166 if (cw > CCV(ccv, snd_ssthresh)) { 167 if (V_tcp_do_rfc3465) { 168 if (ccv->flags & CCF_ABC_SENTAWND) 169 ccv->flags &= ~CCF_ABC_SENTAWND; 170 else 171 incr = 0; 172 } else 173 incr = max((incr * incr / cw), 1); 174 } else if (V_tcp_do_rfc3465) { 175 /* 176 * In slow-start with ABC enabled and no RTO in sight? 177 * (Must not use abc_l_var > 1 if slow starting after 178 * an RTO. On RTO, snd_nxt = snd_una, so the 179 * snd_nxt == snd_max check is sufficient to 180 * handle this). 181 * 182 * XXXLAS: Find a way to signal SS after RTO that 183 * doesn't rely on tcpcb vars. 184 */ 185 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) 186 incr = min(ccv->bytes_this_ack, 187 ccv->nsegs * V_tcp_abc_l_var * 188 CCV(ccv, t_maxseg)); 189 else 190 incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); 191 } 192 /* ABC is on by default, so incr equals 0 frequently. */ 193 if (incr > 0) 194 CCV(ccv, snd_cwnd) = min(cw + incr, 195 TCP_MAXWIN << CCV(ccv, snd_scale)); 196 } 197 } 198 199 static void 200 newreno_after_idle(struct cc_var *ccv) 201 { 202 uint32_t rw; 203 204 /* 205 * If we've been idle for more than one retransmit timeout the old 206 * congestion window is no longer current and we have to reduce it to 207 * the restart window before we can transmit again. 208 * 209 * The restart window is the initial window or the last CWND, whichever 210 * is smaller. 211 * 212 * This is done to prevent us from flooding the path with a full CWND at 213 * wirespeed, overloading router and switch buffers along the way. 214 * 215 * See RFC5681 Section 4.1. "Restarting Idle Connections". 216 * 217 * In addition, per RFC2861 Section 2, the ssthresh is set to the 218 * maximum of the former ssthresh or 3/4 of the old cwnd, to 219 * not exit slow-start prematurely. 220 */ 221 rw = tcp_compute_initwnd(tcp_maxseg(ccv->ccvc.tcp)); 222 223 CCV(ccv, snd_ssthresh) = max(CCV(ccv, snd_ssthresh), 224 CCV(ccv, snd_cwnd)-(CCV(ccv, snd_cwnd)>>2)); 225 226 CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd)); 227 } 228 229 /* 230 * Perform any necessary tasks before we enter congestion recovery. 231 */ 232 static void 233 newreno_cong_signal(struct cc_var *ccv, uint32_t type) 234 { 235 struct newreno *nreno; 236 uint32_t beta, beta_ecn, cwin, factor; 237 u_int mss; 238 239 cwin = CCV(ccv, snd_cwnd); 240 mss = tcp_maxseg(ccv->ccvc.tcp); 241 nreno = ccv->cc_data; 242 beta = (nreno == NULL) ? V_newreno_beta : nreno->beta; 243 beta_ecn = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn; 244 if (V_cc_do_abe && type == CC_ECN) 245 factor = beta_ecn; 246 else 247 factor = beta; 248 249 /* Catch algos which mistakenly leak private signal types. */ 250 KASSERT((type & CC_SIGPRIVMASK) == 0, 251 ("%s: congestion signal type 0x%08x is private\n", __func__, type)); 252 253 cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss), 254 2) * mss; 255 256 switch (type) { 257 case CC_NDUPACK: 258 if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { 259 if (IN_CONGRECOVERY(CCV(ccv, t_flags) && 260 V_cc_do_abe && V_cc_abe_frlossreduce)) { 261 CCV(ccv, snd_ssthresh) = 262 ((uint64_t)CCV(ccv, snd_ssthresh) * 263 (uint64_t)beta) / 264 (100ULL * (uint64_t)beta_ecn); 265 } 266 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) 267 CCV(ccv, snd_ssthresh) = cwin; 268 ENTER_RECOVERY(CCV(ccv, t_flags)); 269 } 270 break; 271 case CC_ECN: 272 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { 273 CCV(ccv, snd_ssthresh) = cwin; 274 CCV(ccv, snd_cwnd) = cwin; 275 ENTER_CONGRECOVERY(CCV(ccv, t_flags)); 276 } 277 break; 278 case CC_RTO: 279 CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd), 280 CCV(ccv, snd_cwnd)) / 2 / mss, 281 2) * mss; 282 CCV(ccv, snd_cwnd) = mss; 283 break; 284 } 285 } 286 287 /* 288 * Perform any necessary tasks before we exit congestion recovery. 289 */ 290 static void 291 newreno_post_recovery(struct cc_var *ccv) 292 { 293 int pipe; 294 295 if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { 296 /* 297 * Fast recovery will conclude after returning from this 298 * function. Window inflation should have left us with 299 * approximately snd_ssthresh outstanding data. But in case we 300 * would be inclined to send a burst, better to do it via the 301 * slow start mechanism. 302 * 303 * XXXLAS: Find a way to do this without needing curack 304 */ 305 if (V_tcp_do_rfc6675_pipe) 306 pipe = tcp_compute_pipe(ccv->ccvc.tcp); 307 else 308 pipe = CCV(ccv, snd_max) - ccv->curack; 309 310 if (pipe < CCV(ccv, snd_ssthresh)) 311 /* 312 * Ensure that cwnd does not collapse to 1 MSS under 313 * adverse conditons. Implements RFC6582 314 */ 315 CCV(ccv, snd_cwnd) = max(pipe, CCV(ccv, t_maxseg)) + 316 CCV(ccv, t_maxseg); 317 else 318 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 319 } 320 } 321 322 static int 323 newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf) 324 { 325 struct newreno *nreno; 326 struct cc_newreno_opts *opt; 327 328 if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts)) 329 return (EMSGSIZE); 330 331 nreno = ccv->cc_data; 332 opt = buf; 333 334 switch (sopt->sopt_dir) { 335 case SOPT_SET: 336 /* We cannot set without cc_data memory. */ 337 if (nreno == NULL) { 338 nreno = newreno_malloc(ccv); 339 if (nreno == NULL) 340 return (ENOMEM); 341 } 342 switch (opt->name) { 343 case CC_NEWRENO_BETA: 344 nreno->beta = opt->val; 345 break; 346 case CC_NEWRENO_BETA_ECN: 347 if (!V_cc_do_abe) 348 return (EACCES); 349 nreno->beta_ecn = opt->val; 350 break; 351 default: 352 return (ENOPROTOOPT); 353 } 354 break; 355 case SOPT_GET: 356 switch (opt->name) { 357 case CC_NEWRENO_BETA: 358 opt->val = (nreno == NULL) ? 359 V_newreno_beta : nreno->beta; 360 break; 361 case CC_NEWRENO_BETA_ECN: 362 opt->val = (nreno == NULL) ? 363 V_newreno_beta_ecn : nreno->beta_ecn; 364 break; 365 default: 366 return (ENOPROTOOPT); 367 } 368 break; 369 default: 370 return (EINVAL); 371 } 372 373 return (0); 374 } 375 376 static int 377 newreno_beta_handler(SYSCTL_HANDLER_ARGS) 378 { 379 int error; 380 uint32_t new; 381 382 new = *(uint32_t *)arg1; 383 error = sysctl_handle_int(oidp, &new, 0, req); 384 if (error == 0 && req->newptr != NULL ) { 385 if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe) 386 error = EACCES; 387 else if (new == 0 || new > 100) 388 error = EINVAL; 389 else 390 *(uint32_t *)arg1 = new; 391 } 392 393 return (error); 394 } 395 396 SYSCTL_DECL(_net_inet_tcp_cc_newreno); 397 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, newreno, 398 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 399 "New Reno related settings"); 400 401 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta, 402 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 403 &VNET_NAME(newreno_beta), 3, &newreno_beta_handler, "IU", 404 "New Reno beta, specified as number between 1 and 100"); 405 406 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn, 407 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 408 &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU", 409 "New Reno beta ecn, specified as number between 1 and 100"); 410 411 DECLARE_CC_MODULE(newreno, &newreno_cc_algo); 412 MODULE_VERSION(newreno, 1); 413