1 /*- 2 * Copyright (c) 2009-2010 3 * Swinburne University of Technology, Melbourne, Australia 4 * Copyright (c) 2010-2011 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * This software was developed at the Centre for Advanced Internet 8 * Architectures, Swinburne University of Technology, by David Hayes and 9 * Lawrence Stewart, made possible in part by a grant from the Cisco University 10 * Research Program Fund at Community Foundation Silicon Valley. 11 * 12 * Portions of this software were developed at the Centre for Advanced Internet 13 * Architectures, Swinburne University of Technology, Melbourne, Australia by 14 * David Hayes under sponsorship from the FreeBSD Foundation. 15 * 16 * Redistribution and use in source and binary forms, with or without 17 * modification, are permitted provided that the following conditions 18 * are met: 19 * 1. Redistributions of source code must retain the above copyright 20 * notice, this list of conditions and the following disclaimer. 21 * 2. Redistributions in binary form must reproduce the above copyright 22 * notice, this list of conditions and the following disclaimer in the 23 * documentation and/or other materials provided with the distribution. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38 /* 39 * An implementation of the CAIA-Hamilton delay based congestion control 40 * algorithm, based on "Improved coexistence and loss tolerance for delay based 41 * TCP congestion control" by D. A. Hayes and G. Armitage., in 35th Annual IEEE 42 * Conference on Local Computer Networks (LCN 2010), Denver, Colorado, USA, 43 * 11-14 October 2010. 44 * 45 * Originally released as part of the NewTCP research project at Swinburne 46 * University of Technology's Centre for Advanced Internet Architectures, 47 * Melbourne, Australia, which was made possible in part by a grant from the 48 * Cisco University Research Program Fund at Community Foundation Silicon 49 * Valley. More details are available at: 50 * http://caia.swin.edu.au/urp/newtcp/ 51 */ 52 53 #include <sys/cdefs.h> 54 __FBSDID("$FreeBSD$"); 55 56 #include <sys/param.h> 57 #include <sys/kernel.h> 58 #include <sys/khelp.h> 59 #include <sys/limits.h> 60 #include <sys/malloc.h> 61 #include <sys/module.h> 62 #include <sys/queue.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/sysctl.h> 66 #include <sys/systm.h> 67 68 #include <net/if.h> 69 #include <net/vnet.h> 70 71 #include <netinet/cc.h> 72 #include <netinet/tcp_seq.h> 73 #include <netinet/tcp_timer.h> 74 #include <netinet/tcp_var.h> 75 76 #include <netinet/cc/cc_module.h> 77 78 #include <netinet/khelp/h_ertt.h> 79 80 #define CAST_PTR_INT(X) (*((int*)(X))) 81 82 /* 83 * Private signal type for rate based congestion signal. 84 * See <netinet/cc.h> for appropriate bit-range to use for private signals. 85 */ 86 #define CC_CHD_DELAY 0x02000000 87 88 /* Largest possible number returned by random(). */ 89 #define RANDOM_MAX INT_MAX 90 91 static void chd_ack_received(struct cc_var *ccv, uint16_t ack_type); 92 static void chd_cb_destroy(struct cc_var *ccv); 93 static int chd_cb_init(struct cc_var *ccv); 94 static void chd_cong_signal(struct cc_var *ccv, uint32_t signal_type); 95 static void chd_conn_init(struct cc_var *ccv); 96 static int chd_mod_init(void); 97 98 struct chd { 99 /* 100 * Shadow window - keeps track of what the NewReno congestion window 101 * would have been if delay-based cwnd backoffs had not been made. This 102 * functionality aids coexistence with loss-based TCP flows which may be 103 * sharing links along the path. 104 */ 105 unsigned long shadow_w; 106 /* 107 * Loss-based TCP compatibility flag - When set, it turns on the shadow 108 * window functionality. 109 */ 110 int loss_compete; 111 /* The maximum round trip time seen within a measured rtt period. */ 112 int maxrtt_in_rtt; 113 /* The previous qdly that caused cwnd to backoff. */ 114 int prev_backoff_qdly; 115 }; 116 117 static int ertt_id; 118 119 static VNET_DEFINE(uint32_t, chd_qmin) = 5; 120 static VNET_DEFINE(uint32_t, chd_pmax) = 50; 121 static VNET_DEFINE(uint32_t, chd_loss_fair) = 1; 122 static VNET_DEFINE(uint32_t, chd_use_max) = 1; 123 static VNET_DEFINE(uint32_t, chd_qthresh) = 20; 124 #define V_chd_qthresh VNET(chd_qthresh) 125 #define V_chd_qmin VNET(chd_qmin) 126 #define V_chd_pmax VNET(chd_pmax) 127 #define V_chd_loss_fair VNET(chd_loss_fair) 128 #define V_chd_use_max VNET(chd_use_max) 129 130 static MALLOC_DEFINE(M_CHD, "chd data", 131 "Per connection data required for the CHD congestion control algorithm"); 132 133 struct cc_algo chd_cc_algo = { 134 .name = "chd", 135 .ack_received = chd_ack_received, 136 .cb_destroy = chd_cb_destroy, 137 .cb_init = chd_cb_init, 138 .cong_signal = chd_cong_signal, 139 .conn_init = chd_conn_init, 140 .mod_init = chd_mod_init 141 }; 142 143 static __inline void 144 chd_window_decrease(struct cc_var *ccv) 145 { 146 unsigned long win; 147 148 win = min(CCV(ccv, snd_wnd), CCV(ccv, snd_cwnd)) / CCV(ccv, t_maxseg); 149 win -= max((win / 2), 1); 150 CCV(ccv, snd_ssthresh) = max(win, 2) * CCV(ccv, t_maxseg); 151 } 152 153 /* 154 * Probabilistic backoff function. Returns 1 if we should backoff or 0 155 * otherwise. The calculation of p is similar to the calculation of p in cc_hd. 156 */ 157 static __inline int 158 should_backoff(int qdly, int maxqdly, struct chd *chd_data) 159 { 160 unsigned long p, rand; 161 162 rand = random(); 163 164 if (qdly < V_chd_qthresh) { 165 chd_data->loss_compete = 0; 166 p = (((RANDOM_MAX / 100) * V_chd_pmax) / 167 (V_chd_qthresh - V_chd_qmin)) * 168 (qdly - V_chd_qmin); 169 } else { 170 if (qdly > V_chd_qthresh) { 171 p = (((RANDOM_MAX / 100) * V_chd_pmax) / 172 (maxqdly - V_chd_qthresh)) * 173 (maxqdly - qdly); 174 if (V_chd_loss_fair && rand < p) 175 chd_data->loss_compete = 1; 176 } else { 177 p = (RANDOM_MAX / 100) * V_chd_pmax; 178 chd_data->loss_compete = 0; 179 } 180 } 181 182 return (rand < p); 183 } 184 185 static __inline void 186 chd_window_increase(struct cc_var *ccv, int new_measurement) 187 { 188 struct chd *chd_data; 189 int incr; 190 191 chd_data = ccv->cc_data; 192 incr = 0; 193 194 if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { 195 /* Adapted from NewReno slow start. */ 196 if (V_tcp_do_rfc3465) { 197 /* In slow-start with ABC enabled. */ 198 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) { 199 /* Not due to RTO. */ 200 incr = min(ccv->bytes_this_ack, 201 V_tcp_abc_l_var * CCV(ccv, t_maxseg)); 202 } else { 203 /* Due to RTO. */ 204 incr = min(ccv->bytes_this_ack, 205 CCV(ccv, t_maxseg)); 206 } 207 } else 208 incr = CCV(ccv, t_maxseg); 209 210 } else { /* Congestion avoidance. */ 211 if (V_tcp_do_rfc3465) { 212 if (ccv->flags & CCF_ABC_SENTAWND) { 213 ccv->flags &= ~CCF_ABC_SENTAWND; 214 incr = CCV(ccv, t_maxseg); 215 } 216 } else if (new_measurement) 217 incr = CCV(ccv, t_maxseg); 218 } 219 220 if (chd_data->shadow_w > 0) { 221 /* Track NewReno window. */ 222 chd_data->shadow_w = min(chd_data->shadow_w + incr, 223 TCP_MAXWIN << CCV(ccv, snd_scale)); 224 } 225 226 CCV(ccv,snd_cwnd) = min(CCV(ccv, snd_cwnd) + incr, 227 TCP_MAXWIN << CCV(ccv, snd_scale)); 228 } 229 230 /* 231 * All ACK signals are used for timing measurements to determine delay-based 232 * congestion. However, window increases are only performed when 233 * ack_type == CC_ACK. 234 */ 235 static void 236 chd_ack_received(struct cc_var *ccv, uint16_t ack_type) 237 { 238 struct chd *chd_data; 239 struct ertt *e_t; 240 int backoff, new_measurement, qdly, rtt; 241 242 e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); 243 chd_data = ccv->cc_data; 244 new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT; 245 backoff = qdly = 0; 246 247 chd_data->maxrtt_in_rtt = imax(e_t->rtt, chd_data->maxrtt_in_rtt); 248 249 if (new_measurement) { 250 /* 251 * There is a new per RTT measurement, so check to see if there 252 * is delay based congestion. 253 */ 254 rtt = V_chd_use_max ? chd_data->maxrtt_in_rtt : e_t->rtt; 255 chd_data->maxrtt_in_rtt = 0; 256 257 if (rtt && e_t->minrtt && !IN_RECOVERY(CCV(ccv, t_flags))) { 258 qdly = rtt - e_t->minrtt; 259 if (qdly > V_chd_qmin) { 260 /* 261 * Probabilistic delay based congestion 262 * indication. 263 */ 264 backoff = should_backoff(qdly, 265 e_t->maxrtt - e_t->minrtt, chd_data); 266 } else 267 chd_data->loss_compete = 0; 268 } 269 /* Reset per RTT measurement flag to start a new measurement. */ 270 e_t->flags &= ~ERTT_NEW_MEASUREMENT; 271 } 272 273 if (backoff) { 274 /* 275 * Update shadow_w before delay based backoff. 276 */ 277 if (chd_data->loss_compete || 278 qdly > chd_data->prev_backoff_qdly) { 279 /* 280 * Delay is higher than when we backed off previously, 281 * so it is possible that this flow is competing with 282 * loss based flows. 283 */ 284 chd_data->shadow_w = max(CCV(ccv, snd_cwnd), 285 chd_data->shadow_w); 286 } else { 287 /* 288 * Reset shadow_w, as it is probable that this flow is 289 * not competing with loss based flows at the moment. 290 */ 291 chd_data->shadow_w = 0; 292 } 293 294 chd_data->prev_backoff_qdly = qdly; 295 /* 296 * Send delay-based congestion signal to the congestion signal 297 * handler. 298 */ 299 chd_cong_signal(ccv, CC_CHD_DELAY); 300 301 } else if (ack_type == CC_ACK) 302 chd_window_increase(ccv, new_measurement); 303 } 304 305 static void 306 chd_cb_destroy(struct cc_var *ccv) 307 { 308 309 if (ccv->cc_data != NULL) 310 free(ccv->cc_data, M_CHD); 311 } 312 313 static int 314 chd_cb_init(struct cc_var *ccv) 315 { 316 struct chd *chd_data; 317 318 chd_data = malloc(sizeof(struct chd), M_CHD, M_NOWAIT); 319 if (chd_data == NULL) 320 return (ENOMEM); 321 322 chd_data->shadow_w = 0; 323 ccv->cc_data = chd_data; 324 325 return (0); 326 } 327 328 static void 329 chd_cong_signal(struct cc_var *ccv, uint32_t signal_type) 330 { 331 struct ertt *e_t; 332 struct chd *chd_data; 333 int qdly; 334 335 e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); 336 chd_data = ccv->cc_data; 337 qdly = imax(e_t->rtt, chd_data->maxrtt_in_rtt) - e_t->minrtt; 338 339 switch(signal_type) { 340 case CC_CHD_DELAY: 341 chd_window_decrease(ccv); /* Set new ssthresh. */ 342 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 343 CCV(ccv, snd_recover) = CCV(ccv, snd_max); 344 ENTER_CONGRECOVERY(CCV(ccv, t_flags)); 345 break; 346 347 case CC_NDUPACK: /* Packet loss. */ 348 /* 349 * Only react to loss as a congestion signal if qdly > 350 * V_chd_qthresh. If qdly is less than qthresh, presume that 351 * this is a non congestion related loss. If qdly is greater 352 * than qthresh, assume that we are competing with loss based 353 * tcp flows and restore window from any unnecessary backoffs, 354 * before the decrease. 355 */ 356 if (!IN_RECOVERY(CCV(ccv, t_flags)) && qdly > V_chd_qthresh) { 357 if (chd_data->loss_compete) { 358 CCV(ccv, snd_cwnd) = max(CCV(ccv, snd_cwnd), 359 chd_data->shadow_w); 360 } 361 chd_window_decrease(ccv); 362 } else { 363 /* 364 * This loss isn't congestion related, or already 365 * recovering from congestion. 366 */ 367 CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); 368 CCV(ccv, snd_recover) = CCV(ccv, snd_max); 369 } 370 371 if (chd_data->shadow_w > 0) { 372 chd_data->shadow_w = max(chd_data->shadow_w / 373 CCV(ccv, t_maxseg) / 2, 2) * CCV(ccv, t_maxseg); 374 } 375 ENTER_FASTRECOVERY(CCV(ccv, t_flags)); 376 break; 377 378 default: 379 newreno_cc_algo.cong_signal(ccv, signal_type); 380 } 381 } 382 383 static void 384 chd_conn_init(struct cc_var *ccv) 385 { 386 struct chd *chd_data; 387 388 chd_data = ccv->cc_data; 389 chd_data->prev_backoff_qdly = 0; 390 chd_data->maxrtt_in_rtt = 0; 391 chd_data->loss_compete = 0; 392 /* 393 * Initialise the shadow_cwnd to be equal to snd_cwnd in case we are 394 * competing with loss based flows from the start. 395 */ 396 chd_data->shadow_w = CCV(ccv, snd_cwnd); 397 } 398 399 static int 400 chd_mod_init(void) 401 { 402 403 ertt_id = khelp_get_id("ertt"); 404 if (ertt_id <= 0) { 405 printf("%s: h_ertt module not found\n", __func__); 406 return (ENOENT); 407 } 408 409 chd_cc_algo.after_idle = newreno_cc_algo.after_idle; 410 chd_cc_algo.post_recovery = newreno_cc_algo.post_recovery; 411 412 return (0); 413 } 414 415 static int 416 chd_loss_fair_handler(SYSCTL_HANDLER_ARGS) 417 { 418 int error; 419 uint32_t new; 420 421 new = V_chd_loss_fair; 422 error = sysctl_handle_int(oidp, &new, 0, req); 423 if (error == 0 && req->newptr != NULL) { 424 if (CAST_PTR_INT(req->newptr) > 1) 425 error = EINVAL; 426 else 427 V_chd_loss_fair = new; 428 } 429 430 return (error); 431 } 432 433 static int 434 chd_pmax_handler(SYSCTL_HANDLER_ARGS) 435 { 436 int error; 437 uint32_t new; 438 439 new = V_chd_pmax; 440 error = sysctl_handle_int(oidp, &new, 0, req); 441 if (error == 0 && req->newptr != NULL) { 442 if (CAST_PTR_INT(req->newptr) == 0 || 443 CAST_PTR_INT(req->newptr) > 100) 444 error = EINVAL; 445 else 446 V_chd_pmax = new; 447 } 448 449 return (error); 450 } 451 452 static int 453 chd_qthresh_handler(SYSCTL_HANDLER_ARGS) 454 { 455 int error; 456 uint32_t new; 457 458 new = V_chd_qthresh; 459 error = sysctl_handle_int(oidp, &new, 0, req); 460 if (error == 0 && req->newptr != NULL) { 461 if (CAST_PTR_INT(req->newptr) <= V_chd_qmin) 462 error = EINVAL; 463 else 464 V_chd_qthresh = new; 465 } 466 467 return (error); 468 } 469 470 SYSCTL_DECL(_net_inet_tcp_cc_chd); 471 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, chd, CTLFLAG_RW, NULL, 472 "CAIA Hamilton delay-based congestion control related settings"); 473 474 SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, loss_fair, 475 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 476 &VNET_NAME(chd_loss_fair), 1, &chd_loss_fair_handler, 477 "IU", "Flag to enable shadow window functionality."); 478 479 SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, pmax, 480 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 481 &VNET_NAME(chd_pmax), 5, &chd_pmax_handler, 482 "IU", "Per RTT maximum backoff probability as a percentage"); 483 484 SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, queue_threshold, 485 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 486 &VNET_NAME(chd_qthresh), 20, &chd_qthresh_handler, 487 "IU", "Queueing congestion threshold in ticks"); 488 489 SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, queue_min, 490 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_qmin), 5, 491 "Minimum queueing delay threshold in ticks"); 492 493 SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, use_max, 494 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_use_max), 1, 495 "Use the maximum RTT seen within the measurement period (RTT) " 496 "as the basic delay measurement for the algorithm."); 497 498 DECLARE_CC_MODULE(chd, &chd_cc_algo); 499 MODULE_DEPEND(chd, ertt, 1, 1, 1); 500