1 /* 2 * PIE - Proportional Integral controller Enhanced AQM algorithm. 3 * 4 * $FreeBSD$ 5 * 6 * Copyright (C) 2016 Centre for Advanced Internet Architectures, 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Portions of this code were made possible in part by a gift from 9 * The Comcast Innovation Fund. 10 * Implemented by Rasool Al-Saadi <ralsaadi@swin.edu.au> 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 #include "opt_inet6.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/malloc.h> 40 #include <sys/mbuf.h> 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/module.h> 44 #include <sys/mutex.h> 45 #include <sys/priv.h> 46 #include <sys/proc.h> 47 #include <sys/rwlock.h> 48 #include <sys/socket.h> 49 #include <sys/time.h> 50 #include <sys/sysctl.h> 51 52 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */ 53 #include <net/netisr.h> 54 #include <net/vnet.h> 55 56 #include <netinet/in.h> 57 #include <netinet/ip.h> /* ip_len, ip_off */ 58 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */ 59 #include <netinet/ip_fw.h> 60 #include <netinet/ip_dummynet.h> 61 #include <netinet/if_ether.h> /* various ether_* routines */ 62 #include <netinet/ip6.h> /* for ip6_input, ip6_output prototypes */ 63 #include <netinet6/ip6_var.h> 64 #include <netpfil/ipfw/dn_heap.h> 65 66 #ifdef NEW_AQM 67 #include <netpfil/ipfw/ip_fw_private.h> 68 #include <netpfil/ipfw/ip_dn_private.h> 69 #include <netpfil/ipfw/dn_aqm.h> 70 #include <netpfil/ipfw/dn_aqm_pie.h> 71 #include <netpfil/ipfw/dn_sched.h> 72 73 /* for debugging */ 74 #include <sys/syslog.h> 75 76 static struct dn_aqm pie_desc; 77 78 /* PIE defaults 79 * target=15ms, tupdate=15ms, max_burst=150ms, 80 * max_ecnth=0.1, alpha=0.125, beta=1.25, 81 */ 82 struct dn_aqm_pie_parms pie_sysctl = 83 { 15 * AQM_TIME_1MS, 15 * AQM_TIME_1MS, 150 * AQM_TIME_1MS, 84 PIE_SCALE/10 , PIE_SCALE * 0.125, PIE_SCALE * 1.25 , 85 PIE_CAPDROP_ENABLED | PIE_DEPRATEEST_ENABLED | PIE_DERAND_ENABLED }; 86 87 static int 88 pie_sysctl_alpha_beta_handler(SYSCTL_HANDLER_ARGS) 89 { 90 int error; 91 long value; 92 93 if (!strcmp(oidp->oid_name,"alpha")) 94 value = pie_sysctl.alpha; 95 else 96 value = pie_sysctl.beta; 97 98 value = value * 1000 / PIE_SCALE; 99 error = sysctl_handle_long(oidp, &value, 0, req); 100 if (error != 0 || req->newptr == NULL) 101 return (error); 102 if (value < 1 || value > 7 * PIE_SCALE) 103 return (EINVAL); 104 value = (value * PIE_SCALE) / 1000; 105 if (!strcmp(oidp->oid_name,"alpha")) 106 pie_sysctl.alpha = value; 107 else 108 pie_sysctl.beta = value; 109 return (0); 110 } 111 112 static int 113 pie_sysctl_target_tupdate_maxb_handler(SYSCTL_HANDLER_ARGS) 114 { 115 int error; 116 long value; 117 118 if (!strcmp(oidp->oid_name,"target")) 119 value = pie_sysctl.qdelay_ref; 120 else if (!strcmp(oidp->oid_name,"tupdate")) 121 value = pie_sysctl.tupdate; 122 else 123 value = pie_sysctl.max_burst; 124 125 value = value / AQM_TIME_1US; 126 error = sysctl_handle_long(oidp, &value, 0, req); 127 if (error != 0 || req->newptr == NULL) 128 return (error); 129 if (value < 1 || value > 10 * AQM_TIME_1S) 130 return (EINVAL); 131 value = value * AQM_TIME_1US; 132 133 if (!strcmp(oidp->oid_name,"target")) 134 pie_sysctl.qdelay_ref = value; 135 else if (!strcmp(oidp->oid_name,"tupdate")) 136 pie_sysctl.tupdate = value; 137 else 138 pie_sysctl.max_burst = value; 139 return (0); 140 } 141 142 static int 143 pie_sysctl_max_ecnth_handler(SYSCTL_HANDLER_ARGS) 144 { 145 int error; 146 long value; 147 148 value = pie_sysctl.max_ecnth; 149 value = value * 1000 / PIE_SCALE; 150 error = sysctl_handle_long(oidp, &value, 0, req); 151 if (error != 0 || req->newptr == NULL) 152 return (error); 153 if (value < 1 || value > PIE_SCALE) 154 return (EINVAL); 155 value = (value * PIE_SCALE) / 1000; 156 pie_sysctl.max_ecnth = value; 157 return (0); 158 } 159 160 /* define PIE sysctl variables */ 161 SYSBEGIN(f4) 162 SYSCTL_DECL(_net_inet); 163 SYSCTL_DECL(_net_inet_ip); 164 SYSCTL_DECL(_net_inet_ip_dummynet); 165 static SYSCTL_NODE(_net_inet_ip_dummynet, OID_AUTO, pie, 166 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 167 "PIE"); 168 169 #ifdef SYSCTL_NODE 170 SYSCTL_PROC(_net_inet_ip_dummynet_pie, OID_AUTO, target, 171 CTLTYPE_LONG | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, 172 pie_sysctl_target_tupdate_maxb_handler, "L", 173 "queue target in microsecond"); 174 SYSCTL_PROC(_net_inet_ip_dummynet_pie, OID_AUTO, tupdate, 175 CTLTYPE_LONG | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, 176 pie_sysctl_target_tupdate_maxb_handler, "L", 177 "the frequency of drop probability calculation in microsecond"); 178 SYSCTL_PROC(_net_inet_ip_dummynet_pie, OID_AUTO, max_burst, 179 CTLTYPE_LONG | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, 180 pie_sysctl_target_tupdate_maxb_handler, "L", 181 "Burst allowance interval in microsecond"); 182 183 SYSCTL_PROC(_net_inet_ip_dummynet_pie, OID_AUTO, max_ecnth, 184 CTLTYPE_LONG | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, 185 pie_sysctl_max_ecnth_handler, "L", 186 "ECN safeguard threshold scaled by 1000"); 187 188 SYSCTL_PROC(_net_inet_ip_dummynet_pie, OID_AUTO, alpha, 189 CTLTYPE_LONG | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, 190 pie_sysctl_alpha_beta_handler, "L", 191 "PIE alpha scaled by 1000"); 192 SYSCTL_PROC(_net_inet_ip_dummynet_pie, OID_AUTO, beta, 193 CTLTYPE_LONG | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, 194 pie_sysctl_alpha_beta_handler, "L", 195 "beta scaled by 1000"); 196 #endif 197 198 /* 199 * Callout function for drop probability calculation 200 * This function is called over tupdate ms and takes pointer of PIE 201 * status variables as an argument 202 */ 203 static void 204 calculate_drop_prob(void *x) 205 { 206 int64_t p, prob, oldprob; 207 struct dn_aqm_pie_parms *pprms; 208 struct pie_status *pst = (struct pie_status *) x; 209 int p_isneg; 210 211 pprms = pst->parms; 212 prob = pst->drop_prob; 213 214 /* calculate current qdelay using DRE method. 215 * If TS is used and no data in the queue, reset current_qdelay 216 * as it stays at last value during dequeue process. 217 */ 218 if (pprms->flags & PIE_DEPRATEEST_ENABLED) 219 pst->current_qdelay = ((uint64_t)pst->pq->ni.len_bytes * 220 pst->avg_dq_time) >> PIE_DQ_THRESHOLD_BITS; 221 else 222 if (!pst->pq->ni.len_bytes) 223 pst->current_qdelay = 0; 224 225 /* calculate drop probability */ 226 p = (int64_t)pprms->alpha * 227 ((int64_t)pst->current_qdelay - (int64_t)pprms->qdelay_ref); 228 p +=(int64_t) pprms->beta * 229 ((int64_t)pst->current_qdelay - (int64_t)pst->qdelay_old); 230 231 /* take absolute value so right shift result is well defined */ 232 p_isneg = p < 0; 233 if (p_isneg) { 234 p = -p; 235 } 236 237 /* We PIE_MAX_PROB shift by 12-bits to increase the division precision */ 238 p *= (PIE_MAX_PROB << 12) / AQM_TIME_1S; 239 240 /* auto-tune drop probability */ 241 if (prob < (PIE_MAX_PROB / 1000000)) /* 0.000001 */ 242 p >>= 11 + PIE_FIX_POINT_BITS + 12; 243 else if (prob < (PIE_MAX_PROB / 100000)) /* 0.00001 */ 244 p >>= 9 + PIE_FIX_POINT_BITS + 12; 245 else if (prob < (PIE_MAX_PROB / 10000)) /* 0.0001 */ 246 p >>= 7 + PIE_FIX_POINT_BITS + 12; 247 else if (prob < (PIE_MAX_PROB / 1000)) /* 0.001 */ 248 p >>= 5 + PIE_FIX_POINT_BITS + 12; 249 else if (prob < (PIE_MAX_PROB / 100)) /* 0.01 */ 250 p >>= 3 + PIE_FIX_POINT_BITS + 12; 251 else if (prob < (PIE_MAX_PROB / 10)) /* 0.1 */ 252 p >>= 1 + PIE_FIX_POINT_BITS + 12; 253 else 254 p >>= PIE_FIX_POINT_BITS + 12; 255 256 oldprob = prob; 257 258 if (p_isneg) { 259 prob = prob - p; 260 261 /* check for multiplication underflow */ 262 if (prob > oldprob) { 263 prob= 0; 264 D("underflow"); 265 } 266 } else { 267 /* Cap Drop adjustment */ 268 if ((pprms->flags & PIE_CAPDROP_ENABLED) && 269 prob >= PIE_MAX_PROB / 10 && 270 p > PIE_MAX_PROB / 50 ) { 271 p = PIE_MAX_PROB / 50; 272 } 273 274 prob = prob + p; 275 276 /* check for multiplication overflow */ 277 if (prob<oldprob) { 278 D("overflow"); 279 prob= PIE_MAX_PROB; 280 } 281 } 282 283 /* 284 * decay the drop probability exponentially 285 * and restrict it to range 0 to PIE_MAX_PROB 286 */ 287 if (prob < 0) { 288 prob = 0; 289 } else { 290 if (pst->current_qdelay == 0 && pst->qdelay_old == 0) { 291 /* 0.98 ~= 1- 1/64 */ 292 prob = prob - (prob >> 6); 293 } 294 295 if (prob > PIE_MAX_PROB) { 296 prob = PIE_MAX_PROB; 297 } 298 } 299 300 pst->drop_prob = prob; 301 302 /* store current queue delay value in old queue delay*/ 303 pst->qdelay_old = pst->current_qdelay; 304 305 /* update burst allowance */ 306 if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance>0) { 307 308 if (pst->burst_allowance > pprms->tupdate ) 309 pst->burst_allowance -= pprms->tupdate; 310 else 311 pst->burst_allowance = 0; 312 } 313 314 /* reschedule calculate_drop_prob function */ 315 if (pst->sflags & PIE_ACTIVE) 316 callout_reset_sbt(&pst->aqm_pie_callout, 317 (uint64_t)pprms->tupdate * SBT_1US, 0, calculate_drop_prob, pst, 0); 318 319 mtx_unlock(&pst->lock_mtx); 320 } 321 322 /* 323 * Extract a packet from the head of queue 'q' 324 * Return a packet or NULL if the queue is empty. 325 * If getts is set, also extract packet's timestamp from mtag. 326 */ 327 static struct mbuf * 328 pie_extract_head(struct dn_queue *q, aqm_time_t *pkt_ts, int getts) 329 { 330 struct m_tag *mtag; 331 struct mbuf *m; 332 333 next: m = q->mq.head; 334 if (m == NULL) 335 return m; 336 q->mq.head = m->m_nextpkt; 337 338 /* Update stats */ 339 update_stats(q, -m->m_pkthdr.len, 0); 340 341 if (q->ni.length == 0) /* queue is now idle */ 342 q->q_time = V_dn_cfg.curr_time; 343 344 if (getts) { 345 /* extract packet TS*/ 346 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL); 347 if (mtag == NULL) { 348 D("PIE timestamp mtag not found!"); 349 *pkt_ts = 0; 350 } else { 351 *pkt_ts = *(aqm_time_t *)(mtag + 1); 352 m_tag_delete(m,mtag); 353 } 354 } 355 if (m->m_pkthdr.rcvif != NULL && 356 __predict_false(m_rcvif_restore(m) == NULL)) { 357 m_freem(m); 358 goto next; 359 } 360 return m; 361 } 362 363 /* 364 * Initiate PIE variable and optionally activate it 365 */ 366 __inline static void 367 init_activate_pie(struct pie_status *pst, int resettimer) 368 { 369 struct dn_aqm_pie_parms *pprms; 370 371 mtx_lock(&pst->lock_mtx); 372 pprms = pst->parms; 373 pst->drop_prob = 0; 374 pst->qdelay_old = 0; 375 pst->burst_allowance = pprms->max_burst; 376 pst->accu_prob = 0; 377 pst->dq_count = 0; 378 pst->avg_dq_time = 0; 379 pst->sflags = PIE_INMEASUREMENT; 380 pst->measurement_start = AQM_UNOW; 381 382 if (resettimer) { 383 pst->sflags |= PIE_ACTIVE; 384 callout_reset_sbt(&pst->aqm_pie_callout, 385 (uint64_t)pprms->tupdate * SBT_1US, 386 0, calculate_drop_prob, pst, 0); 387 } 388 //DX(2, "PIE Activated"); 389 mtx_unlock(&pst->lock_mtx); 390 } 391 392 /* 393 * Deactivate PIE and stop probe update callout 394 */ 395 __inline static void 396 deactivate_pie(struct pie_status *pst) 397 { 398 mtx_lock(&pst->lock_mtx); 399 pst->sflags &= ~(PIE_ACTIVE | PIE_INMEASUREMENT); 400 callout_stop(&pst->aqm_pie_callout); 401 //D("PIE Deactivated"); 402 mtx_unlock(&pst->lock_mtx); 403 } 404 405 /* 406 * Dequeue and return a pcaket from queue 'q' or NULL if 'q' is empty. 407 * Also, caculate depature time or queue delay using timestamp 408 */ 409 static struct mbuf * 410 aqm_pie_dequeue(struct dn_queue *q) 411 { 412 struct mbuf *m; 413 struct dn_aqm_pie_parms *pprms; 414 struct pie_status *pst; 415 aqm_time_t now; 416 aqm_time_t pkt_ts, dq_time; 417 int32_t w; 418 419 pst = q->aqm_status; 420 pprms = pst->parms; 421 422 /*we extarct packet ts only when Departure Rate Estimation dis not used*/ 423 m = pie_extract_head(q, &pkt_ts, !(pprms->flags & PIE_DEPRATEEST_ENABLED)); 424 425 if (!m || !(pst->sflags & PIE_ACTIVE)) 426 return m; 427 428 now = AQM_UNOW; 429 if (pprms->flags & PIE_DEPRATEEST_ENABLED) { 430 /* calculate average depature time */ 431 if(pst->sflags & PIE_INMEASUREMENT) { 432 pst->dq_count += m->m_pkthdr.len; 433 434 if (pst->dq_count >= PIE_DQ_THRESHOLD) { 435 dq_time = now - pst->measurement_start; 436 437 /* 438 * if we don't have old avg dq_time i.e PIE is (re)initialized, 439 * don't use weight to calculate new avg_dq_time 440 */ 441 if(pst->avg_dq_time == 0) 442 pst->avg_dq_time = dq_time; 443 else { 444 /* 445 * weight = PIE_DQ_THRESHOLD/2^6, but we scaled 446 * weight by 2^8. Thus, scaled 447 * weight = PIE_DQ_THRESHOLD /2^8 448 * */ 449 w = PIE_DQ_THRESHOLD >> 8; 450 pst->avg_dq_time = (dq_time* w 451 + (pst->avg_dq_time * ((1L << 8) - w))) >> 8; 452 pst->sflags &= ~PIE_INMEASUREMENT; 453 } 454 } 455 } 456 457 /* 458 * Start new measurment cycle when the queue has 459 * PIE_DQ_THRESHOLD worth of bytes. 460 */ 461 if(!(pst->sflags & PIE_INMEASUREMENT) && 462 q->ni.len_bytes >= PIE_DQ_THRESHOLD) { 463 pst->sflags |= PIE_INMEASUREMENT; 464 pst->measurement_start = now; 465 pst->dq_count = 0; 466 } 467 } 468 /* Optionally, use packet timestamp to estimate queue delay */ 469 else 470 pst->current_qdelay = now - pkt_ts; 471 472 return m; 473 } 474 475 /* 476 * Enqueue a packet in q, subject to space and PIE queue management policy 477 * (whose parameters are in q->fs). 478 * Update stats for the queue and the scheduler. 479 * Return 0 on success, 1 on drop. The packet is consumed anyways. 480 */ 481 static int 482 aqm_pie_enqueue(struct dn_queue *q, struct mbuf* m) 483 { 484 struct dn_fs *f; 485 uint64_t len; 486 uint32_t qlen; 487 struct pie_status *pst; 488 struct dn_aqm_pie_parms *pprms; 489 int t; 490 491 len = m->m_pkthdr.len; 492 pst = q->aqm_status; 493 if(!pst) { 494 DX(2, "PIE queue is not initialized\n"); 495 update_stats(q, 0, 1); 496 FREE_PKT(m); 497 return 1; 498 } 499 500 f = &(q->fs->fs); 501 pprms = pst->parms; 502 t = ENQUE; 503 504 /* get current queue length in bytes or packets*/ 505 qlen = (f->flags & DN_QSIZE_BYTES) ? 506 q->ni.len_bytes : q->ni.length; 507 508 /* check for queue size and drop the tail if exceed queue limit*/ 509 if (qlen >= f->qsize) 510 t = DROP; 511 /* drop/mark the packet when PIE is active and burst time elapsed */ 512 else if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance==0 513 && drop_early(pst, q->ni.len_bytes) == DROP) { 514 /* 515 * if drop_prob over ECN threshold, drop the packet 516 * otherwise mark and enqueue it. 517 */ 518 if ((pprms->flags & PIE_ECN_ENABLED) && pst->drop_prob < 519 (pprms->max_ecnth << (PIE_PROB_BITS - PIE_FIX_POINT_BITS)) 520 && ecn_mark(m)) 521 t = ENQUE; 522 else 523 t = DROP; 524 } 525 526 /* Turn PIE on when 1/3 of the queue is full */ 527 if (!(pst->sflags & PIE_ACTIVE) && qlen >= pst->one_third_q_size) { 528 init_activate_pie(pst, 1); 529 } 530 531 /* Reset burst tolerance and optinally turn PIE off*/ 532 if ((pst->sflags & PIE_ACTIVE) && pst->drop_prob == 0 && 533 pst->current_qdelay < (pprms->qdelay_ref >> 1) && 534 pst->qdelay_old < (pprms->qdelay_ref >> 1)) { 535 pst->burst_allowance = pprms->max_burst; 536 if ((pprms->flags & PIE_ON_OFF_MODE_ENABLED) && qlen<=0) 537 deactivate_pie(pst); 538 } 539 540 /* Timestamp the packet if Departure Rate Estimation is disabled */ 541 if (t != DROP && !(pprms->flags & PIE_DEPRATEEST_ENABLED)) { 542 /* Add TS to mbuf as a TAG */ 543 struct m_tag *mtag; 544 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL); 545 if (mtag == NULL) 546 mtag = m_tag_alloc(MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, 547 sizeof(aqm_time_t), M_NOWAIT); 548 if (mtag == NULL) { 549 t = DROP; 550 } else { 551 *(aqm_time_t *)(mtag + 1) = AQM_UNOW; 552 m_tag_prepend(m, mtag); 553 } 554 } 555 556 if (t != DROP) { 557 mq_append(&q->mq, m); 558 update_stats(q, len, 0); 559 return (0); 560 } else { 561 update_stats(q, 0, 1); 562 563 /* reset accu_prob after packet drop */ 564 pst->accu_prob = 0; 565 FREE_PKT(m); 566 return 1; 567 } 568 return 0; 569 } 570 571 /* 572 * initialize PIE for queue 'q' 573 * First allocate memory for PIE status. 574 */ 575 static int 576 aqm_pie_init(struct dn_queue *q) 577 { 578 struct pie_status *pst; 579 struct dn_aqm_pie_parms *pprms; 580 int err = 0; 581 582 pprms = q->fs->aqmcfg; 583 584 do { /* exit with break when error occurs*/ 585 if (!pprms){ 586 DX(2, "AQM_PIE is not configured"); 587 err = EINVAL; 588 break; 589 } 590 591 q->aqm_status = malloc(sizeof(struct pie_status), 592 M_DUMMYNET, M_NOWAIT | M_ZERO); 593 if (q->aqm_status == NULL) { 594 D("cannot allocate PIE private data"); 595 err = ENOMEM ; 596 break; 597 } 598 599 pst = q->aqm_status; 600 dummynet_sched_lock(); 601 /* increase reference count for PIE module */ 602 pie_desc.ref_count++; 603 dummynet_sched_unlock(); 604 605 pst->pq = q; 606 pst->parms = pprms; 607 608 /* For speed optimization, we caculate 1/3 queue size once here */ 609 // we can use x/3 = (x >>2) + (x >>4) + (x >>7) 610 pst->one_third_q_size = q->fs->fs.qsize/3; 611 612 mtx_init(&pst->lock_mtx, "mtx_pie", NULL, MTX_DEF); 613 callout_init_mtx(&pst->aqm_pie_callout, &pst->lock_mtx, 614 CALLOUT_RETURNUNLOCKED); 615 616 pst->current_qdelay = 0; 617 init_activate_pie(pst, !(pprms->flags & PIE_ON_OFF_MODE_ENABLED)); 618 619 //DX(2, "aqm_PIE_init"); 620 621 } while(0); 622 623 return err; 624 } 625 626 /* 627 * Callout function to destroy pie mtx and free PIE status memory 628 */ 629 static void 630 pie_callout_cleanup(void *x) 631 { 632 struct pie_status *pst = (struct pie_status *) x; 633 634 mtx_unlock(&pst->lock_mtx); 635 mtx_destroy(&pst->lock_mtx); 636 free(x, M_DUMMYNET); 637 dummynet_sched_lock(); 638 pie_desc.ref_count--; 639 dummynet_sched_unlock(); 640 } 641 642 /* 643 * Clean up PIE status for queue 'q' 644 * Destroy memory allocated for PIE status. 645 */ 646 static int 647 aqm_pie_cleanup(struct dn_queue *q) 648 { 649 650 if(!q) { 651 D("q is null"); 652 return 0; 653 } 654 struct pie_status *pst = q->aqm_status; 655 if(!pst) { 656 //D("queue is already cleaned up"); 657 return 0; 658 } 659 if(!q->fs || !q->fs->aqmcfg) { 660 D("fs is null or no cfg"); 661 return 1; 662 } 663 if (q->fs->aqmfp && q->fs->aqmfp->type !=DN_AQM_PIE) { 664 D("Not PIE fs (%d)", q->fs->fs.fs_nr); 665 return 1; 666 } 667 668 /* 669 * Free PIE status allocated memory using pie_callout_cleanup() callout 670 * function to avoid any potential race. 671 * We reset aqm_pie_callout to call pie_callout_cleanup() in next 1um. This 672 * stops the scheduled calculate_drop_prob() callout and call pie_callout_cleanup() 673 * which does memory freeing. 674 */ 675 mtx_lock(&pst->lock_mtx); 676 callout_reset_sbt(&pst->aqm_pie_callout, 677 SBT_1US, 0, pie_callout_cleanup, pst, 0); 678 q->aqm_status = NULL; 679 mtx_unlock(&pst->lock_mtx); 680 681 return 0; 682 } 683 684 /* 685 * Config PIE parameters 686 * also allocate memory for PIE configurations 687 */ 688 static int 689 aqm_pie_config(struct dn_fsk* fs, struct dn_extra_parms *ep, int len) 690 { 691 struct dn_aqm_pie_parms *pcfg; 692 693 int l = sizeof(struct dn_extra_parms); 694 if (len < l) { 695 D("invalid sched parms length got %d need %d", len, l); 696 return EINVAL; 697 } 698 /* we free the old cfg because maybe the orignal allocation 699 * was used for diffirent AQM type. 700 */ 701 if (fs->aqmcfg) { 702 free(fs->aqmcfg, M_DUMMYNET); 703 fs->aqmcfg = NULL; 704 } 705 706 fs->aqmcfg = malloc(sizeof(struct dn_aqm_pie_parms), 707 M_DUMMYNET, M_NOWAIT | M_ZERO); 708 if (fs->aqmcfg== NULL) { 709 D("cannot allocate PIE configuration parameters"); 710 return ENOMEM; 711 } 712 713 /* par array contains pie configuration as follow 714 * 0- qdelay_ref,1- tupdate, 2- max_burst 715 * 3- max_ecnth, 4- alpha, 5- beta, 6- flags 716 */ 717 718 /* configure PIE parameters */ 719 pcfg = fs->aqmcfg; 720 721 if (ep->par[0] < 0) 722 pcfg->qdelay_ref = pie_sysctl.qdelay_ref * AQM_TIME_1US; 723 else 724 pcfg->qdelay_ref = ep->par[0]; 725 if (ep->par[1] < 0) 726 pcfg->tupdate = pie_sysctl.tupdate * AQM_TIME_1US; 727 else 728 pcfg->tupdate = ep->par[1]; 729 if (ep->par[2] < 0) 730 pcfg->max_burst = pie_sysctl.max_burst * AQM_TIME_1US; 731 else 732 pcfg->max_burst = ep->par[2]; 733 if (ep->par[3] < 0) 734 pcfg->max_ecnth = pie_sysctl.max_ecnth; 735 else 736 pcfg->max_ecnth = ep->par[3]; 737 if (ep->par[4] < 0) 738 pcfg->alpha = pie_sysctl.alpha; 739 else 740 pcfg->alpha = ep->par[4]; 741 if (ep->par[5] < 0) 742 pcfg->beta = pie_sysctl.beta; 743 else 744 pcfg->beta = ep->par[5]; 745 if (ep->par[6] < 0) 746 pcfg->flags = pie_sysctl.flags; 747 else 748 pcfg->flags = ep->par[6]; 749 750 /* bound PIE configurations */ 751 pcfg->qdelay_ref = BOUND_VAR(pcfg->qdelay_ref, 1, 10 * AQM_TIME_1S); 752 pcfg->tupdate = BOUND_VAR(pcfg->tupdate, 1, 10 * AQM_TIME_1S); 753 pcfg->max_burst = BOUND_VAR(pcfg->max_burst, 0, 10 * AQM_TIME_1S); 754 pcfg->max_ecnth = BOUND_VAR(pcfg->max_ecnth, 0, PIE_SCALE); 755 pcfg->alpha = BOUND_VAR(pcfg->alpha, 0, 7 * PIE_SCALE); 756 pcfg->beta = BOUND_VAR(pcfg->beta, 0 , 7 * PIE_SCALE); 757 758 pie_desc.cfg_ref_count++; 759 //D("pie cfg_ref_count=%d", pie_desc.cfg_ref_count); 760 return 0; 761 } 762 763 /* 764 * Deconfigure PIE and free memory allocation 765 */ 766 static int 767 aqm_pie_deconfig(struct dn_fsk* fs) 768 { 769 if (fs && fs->aqmcfg) { 770 free(fs->aqmcfg, M_DUMMYNET); 771 fs->aqmcfg = NULL; 772 pie_desc.cfg_ref_count--; 773 } 774 return 0; 775 } 776 777 /* 778 * Retrieve PIE configuration parameters. 779 */ 780 static int 781 aqm_pie_getconfig (struct dn_fsk *fs, struct dn_extra_parms * ep) 782 { 783 struct dn_aqm_pie_parms *pcfg; 784 if (fs->aqmcfg) { 785 strlcpy(ep->name, pie_desc.name, sizeof(ep->name)); 786 pcfg = fs->aqmcfg; 787 ep->par[0] = pcfg->qdelay_ref / AQM_TIME_1US; 788 ep->par[1] = pcfg->tupdate / AQM_TIME_1US; 789 ep->par[2] = pcfg->max_burst / AQM_TIME_1US; 790 ep->par[3] = pcfg->max_ecnth; 791 ep->par[4] = pcfg->alpha; 792 ep->par[5] = pcfg->beta; 793 ep->par[6] = pcfg->flags; 794 795 return 0; 796 } 797 return 1; 798 } 799 800 static struct dn_aqm pie_desc = { 801 _SI( .type = ) DN_AQM_PIE, 802 _SI( .name = ) "PIE", 803 _SI( .ref_count = ) 0, 804 _SI( .cfg_ref_count = ) 0, 805 _SI( .enqueue = ) aqm_pie_enqueue, 806 _SI( .dequeue = ) aqm_pie_dequeue, 807 _SI( .config = ) aqm_pie_config, 808 _SI( .deconfig = ) aqm_pie_deconfig, 809 _SI( .getconfig = ) aqm_pie_getconfig, 810 _SI( .init = ) aqm_pie_init, 811 _SI( .cleanup = ) aqm_pie_cleanup, 812 }; 813 814 DECLARE_DNAQM_MODULE(dn_aqm_pie, &pie_desc); 815 #endif 816