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 /* 200 * Callout function for drop probability calculation 201 * This function is called over tupdate ms and takes pointer of PIE 202 * status variables as an argument 203 */ 204 static void 205 calculate_drop_prob(void *x) 206 { 207 int64_t p, prob, oldprob; 208 struct dn_aqm_pie_parms *pprms; 209 struct pie_status *pst = (struct pie_status *) x; 210 int p_isneg; 211 212 pprms = pst->parms; 213 prob = pst->drop_prob; 214 215 /* calculate current qdelay using DRE method. 216 * If TS is used and no data in the queue, reset current_qdelay 217 * as it stays at last value during dequeue process. 218 */ 219 if (pprms->flags & PIE_DEPRATEEST_ENABLED) 220 pst->current_qdelay = ((uint64_t)pst->pq->ni.len_bytes * 221 pst->avg_dq_time) >> PIE_DQ_THRESHOLD_BITS; 222 else 223 if (!pst->pq->ni.len_bytes) 224 pst->current_qdelay = 0; 225 226 /* calculate drop probability */ 227 p = (int64_t)pprms->alpha * 228 ((int64_t)pst->current_qdelay - (int64_t)pprms->qdelay_ref); 229 p +=(int64_t) pprms->beta * 230 ((int64_t)pst->current_qdelay - (int64_t)pst->qdelay_old); 231 232 /* take absolute value so right shift result is well defined */ 233 p_isneg = p < 0; 234 if (p_isneg) { 235 p = -p; 236 } 237 238 /* We PIE_MAX_PROB shift by 12-bits to increase the division precision */ 239 p *= (PIE_MAX_PROB << 12) / AQM_TIME_1S; 240 241 /* auto-tune drop probability */ 242 if (prob < (PIE_MAX_PROB / 1000000)) /* 0.000001 */ 243 p >>= 11 + PIE_FIX_POINT_BITS + 12; 244 else if (prob < (PIE_MAX_PROB / 100000)) /* 0.00001 */ 245 p >>= 9 + PIE_FIX_POINT_BITS + 12; 246 else if (prob < (PIE_MAX_PROB / 10000)) /* 0.0001 */ 247 p >>= 7 + PIE_FIX_POINT_BITS + 12; 248 else if (prob < (PIE_MAX_PROB / 1000)) /* 0.001 */ 249 p >>= 5 + PIE_FIX_POINT_BITS + 12; 250 else if (prob < (PIE_MAX_PROB / 100)) /* 0.01 */ 251 p >>= 3 + PIE_FIX_POINT_BITS + 12; 252 else if (prob < (PIE_MAX_PROB / 10)) /* 0.1 */ 253 p >>= 1 + PIE_FIX_POINT_BITS + 12; 254 else 255 p >>= PIE_FIX_POINT_BITS + 12; 256 257 oldprob = prob; 258 259 if (p_isneg) { 260 prob = prob - p; 261 262 /* check for multiplication underflow */ 263 if (prob > oldprob) { 264 prob= 0; 265 D("underflow"); 266 } 267 } else { 268 /* Cap Drop adjustment */ 269 if ((pprms->flags & PIE_CAPDROP_ENABLED) && 270 prob >= PIE_MAX_PROB / 10 && 271 p > PIE_MAX_PROB / 50 ) { 272 p = PIE_MAX_PROB / 50; 273 } 274 275 prob = prob + p; 276 277 /* check for multiplication overflow */ 278 if (prob<oldprob) { 279 D("overflow"); 280 prob= PIE_MAX_PROB; 281 } 282 } 283 284 /* 285 * decay the drop probability exponentially 286 * and restrict it to range 0 to PIE_MAX_PROB 287 */ 288 if (prob < 0) { 289 prob = 0; 290 } else { 291 if (pst->current_qdelay == 0 && pst->qdelay_old == 0) { 292 /* 0.98 ~= 1- 1/64 */ 293 prob = prob - (prob >> 6); 294 } 295 296 if (prob > PIE_MAX_PROB) { 297 prob = PIE_MAX_PROB; 298 } 299 } 300 301 pst->drop_prob = prob; 302 303 /* store current queue delay value in old queue delay*/ 304 pst->qdelay_old = pst->current_qdelay; 305 306 /* update burst allowance */ 307 if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance>0) { 308 309 if (pst->burst_allowance > pprms->tupdate ) 310 pst->burst_allowance -= pprms->tupdate; 311 else 312 pst->burst_allowance = 0; 313 } 314 315 /* reschedule calculate_drop_prob function */ 316 if (pst->sflags & PIE_ACTIVE) 317 callout_reset_sbt(&pst->aqm_pie_callout, 318 (uint64_t)pprms->tupdate * SBT_1US, 0, calculate_drop_prob, pst, 0); 319 320 mtx_unlock(&pst->lock_mtx); 321 } 322 323 /* 324 * Extract a packet from the head of queue 'q' 325 * Return a packet or NULL if the queue is empty. 326 * If getts is set, also extract packet's timestamp from mtag. 327 */ 328 static struct mbuf * 329 pie_extract_head(struct dn_queue *q, aqm_time_t *pkt_ts, int getts) 330 { 331 struct m_tag *mtag; 332 struct mbuf *m = q->mq.head; 333 334 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 = 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 return m; 356 } 357 358 /* 359 * Initiate PIE variable and optionally activate it 360 */ 361 __inline static void 362 init_activate_pie(struct pie_status *pst, int resettimer) 363 { 364 struct dn_aqm_pie_parms *pprms; 365 366 mtx_lock(&pst->lock_mtx); 367 pprms = pst->parms; 368 pst->drop_prob = 0; 369 pst->qdelay_old = 0; 370 pst->burst_allowance = pprms->max_burst; 371 pst->accu_prob = 0; 372 pst->dq_count = 0; 373 pst->avg_dq_time = 0; 374 pst->sflags = PIE_INMEASUREMENT; 375 pst->measurement_start = AQM_UNOW; 376 377 if (resettimer) { 378 pst->sflags |= PIE_ACTIVE; 379 callout_reset_sbt(&pst->aqm_pie_callout, 380 (uint64_t)pprms->tupdate * SBT_1US, 381 0, calculate_drop_prob, pst, 0); 382 } 383 //DX(2, "PIE Activated"); 384 mtx_unlock(&pst->lock_mtx); 385 } 386 387 /* 388 * Deactivate PIE and stop probe update callout 389 */ 390 __inline static void 391 deactivate_pie(struct pie_status *pst) 392 { 393 mtx_lock(&pst->lock_mtx); 394 pst->sflags &= ~(PIE_ACTIVE | PIE_INMEASUREMENT); 395 callout_stop(&pst->aqm_pie_callout); 396 //D("PIE Deactivated"); 397 mtx_unlock(&pst->lock_mtx); 398 } 399 400 /* 401 * Dequeue and return a pcaket from queue 'q' or NULL if 'q' is empty. 402 * Also, caculate depature time or queue delay using timestamp 403 */ 404 static struct mbuf * 405 aqm_pie_dequeue(struct dn_queue *q) 406 { 407 struct mbuf *m; 408 struct dn_flow *ni; /* stats for scheduler instance */ 409 struct dn_aqm_pie_parms *pprms; 410 struct pie_status *pst; 411 aqm_time_t now; 412 aqm_time_t pkt_ts, dq_time; 413 int32_t w; 414 415 pst = q->aqm_status; 416 pprms = pst->parms; 417 ni = &q->_si->ni; 418 419 /*we extarct packet ts only when Departure Rate Estimation dis not used*/ 420 m = pie_extract_head(q, &pkt_ts, !(pprms->flags & PIE_DEPRATEEST_ENABLED)); 421 422 if (!m || !(pst->sflags & PIE_ACTIVE)) 423 return m; 424 425 now = AQM_UNOW; 426 if (pprms->flags & PIE_DEPRATEEST_ENABLED) { 427 /* calculate average depature time */ 428 if(pst->sflags & PIE_INMEASUREMENT) { 429 pst->dq_count += m->m_pkthdr.len; 430 431 if (pst->dq_count >= PIE_DQ_THRESHOLD) { 432 dq_time = now - pst->measurement_start; 433 434 /* 435 * if we don't have old avg dq_time i.e PIE is (re)initialized, 436 * don't use weight to calculate new avg_dq_time 437 */ 438 if(pst->avg_dq_time == 0) 439 pst->avg_dq_time = dq_time; 440 else { 441 /* 442 * weight = PIE_DQ_THRESHOLD/2^6, but we scaled 443 * weight by 2^8. Thus, scaled 444 * weight = PIE_DQ_THRESHOLD /2^8 445 * */ 446 w = PIE_DQ_THRESHOLD >> 8; 447 pst->avg_dq_time = (dq_time* w 448 + (pst->avg_dq_time * ((1L << 8) - w))) >> 8; 449 pst->sflags &= ~PIE_INMEASUREMENT; 450 } 451 } 452 } 453 454 /* 455 * Start new measurment cycle when the queue has 456 * PIE_DQ_THRESHOLD worth of bytes. 457 */ 458 if(!(pst->sflags & PIE_INMEASUREMENT) && 459 q->ni.len_bytes >= PIE_DQ_THRESHOLD) { 460 pst->sflags |= PIE_INMEASUREMENT; 461 pst->measurement_start = now; 462 pst->dq_count = 0; 463 } 464 } 465 /* Optionally, use packet timestamp to estimate queue delay */ 466 else 467 pst->current_qdelay = now - pkt_ts; 468 469 return m; 470 } 471 472 /* 473 * Enqueue a packet in q, subject to space and PIE queue management policy 474 * (whose parameters are in q->fs). 475 * Update stats for the queue and the scheduler. 476 * Return 0 on success, 1 on drop. The packet is consumed anyways. 477 */ 478 static int 479 aqm_pie_enqueue(struct dn_queue *q, struct mbuf* m) 480 { 481 struct dn_fs *f; 482 uint64_t len; 483 uint32_t qlen; 484 struct pie_status *pst; 485 struct dn_aqm_pie_parms *pprms; 486 int t; 487 488 len = m->m_pkthdr.len; 489 pst = q->aqm_status; 490 if(!pst) { 491 DX(2, "PIE queue is not initialized\n"); 492 update_stats(q, 0, 1); 493 FREE_PKT(m); 494 return 1; 495 } 496 497 f = &(q->fs->fs); 498 pprms = pst->parms; 499 t = ENQUE; 500 501 /* get current queue length in bytes or packets*/ 502 qlen = (f->flags & DN_QSIZE_BYTES) ? 503 q->ni.len_bytes : q->ni.length; 504 505 /* check for queue size and drop the tail if exceed queue limit*/ 506 if (qlen >= f->qsize) 507 t = DROP; 508 /* drop/mark the packet when PIE is active and burst time elapsed */ 509 else if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance==0 510 && drop_early(pst, q->ni.len_bytes) == DROP) { 511 /* 512 * if drop_prob over ECN threshold, drop the packet 513 * otherwise mark and enqueue it. 514 */ 515 if ((pprms->flags & PIE_ECN_ENABLED) && pst->drop_prob < 516 (pprms->max_ecnth << (PIE_PROB_BITS - PIE_FIX_POINT_BITS)) 517 && ecn_mark(m)) 518 t = ENQUE; 519 else 520 t = DROP; 521 } 522 523 /* Turn PIE on when 1/3 of the queue is full */ 524 if (!(pst->sflags & PIE_ACTIVE) && qlen >= pst->one_third_q_size) { 525 init_activate_pie(pst, 1); 526 } 527 528 /* Reset burst tolerance and optinally turn PIE off*/ 529 if ((pst->sflags & PIE_ACTIVE) && pst->drop_prob == 0 && 530 pst->current_qdelay < (pprms->qdelay_ref >> 1) && 531 pst->qdelay_old < (pprms->qdelay_ref >> 1)) { 532 533 pst->burst_allowance = pprms->max_burst; 534 if ((pprms->flags & PIE_ON_OFF_MODE_ENABLED) && qlen<=0) 535 deactivate_pie(pst); 536 } 537 538 /* Timestamp the packet if Departure Rate Estimation is disabled */ 539 if (t != DROP && !(pprms->flags & PIE_DEPRATEEST_ENABLED)) { 540 /* Add TS to mbuf as a TAG */ 541 struct m_tag *mtag; 542 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL); 543 if (mtag == NULL) 544 mtag = m_tag_alloc(MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, 545 sizeof(aqm_time_t), M_NOWAIT); 546 if (mtag == NULL) { 547 m_freem(m); 548 t = DROP; 549 } 550 *(aqm_time_t *)(mtag + 1) = AQM_UNOW; 551 m_tag_prepend(m, mtag); 552 } 553 554 if (t != DROP) { 555 mq_append(&q->mq, m); 556 update_stats(q, len, 0); 557 return (0); 558 } else { 559 update_stats(q, 0, 1); 560 561 /* reset accu_prob after packet drop */ 562 pst->accu_prob = 0; 563 FREE_PKT(m); 564 return 1; 565 } 566 return 0; 567 } 568 569 /* 570 * initialize PIE for queue 'q' 571 * First allocate memory for PIE status. 572 */ 573 static int 574 aqm_pie_init(struct dn_queue *q) 575 { 576 struct pie_status *pst; 577 struct dn_aqm_pie_parms *pprms; 578 int err = 0; 579 580 pprms = q->fs->aqmcfg; 581 582 do { /* exit with break when error occurs*/ 583 if (!pprms){ 584 DX(2, "AQM_PIE is not configured"); 585 err = EINVAL; 586 break; 587 } 588 589 q->aqm_status = malloc(sizeof(struct pie_status), 590 M_DUMMYNET, M_NOWAIT | M_ZERO); 591 if (q->aqm_status == NULL) { 592 D("cannot allocate PIE private data"); 593 err = ENOMEM ; 594 break; 595 } 596 597 pst = q->aqm_status; 598 /* increase reference count for PIE module */ 599 pie_desc.ref_count++; 600 601 pst->pq = q; 602 pst->parms = pprms; 603 604 /* For speed optimization, we caculate 1/3 queue size once here */ 605 // we can use x/3 = (x >>2) + (x >>4) + (x >>7) 606 pst->one_third_q_size = q->fs->fs.qsize/3; 607 608 mtx_init(&pst->lock_mtx, "mtx_pie", NULL, MTX_DEF); 609 callout_init_mtx(&pst->aqm_pie_callout, &pst->lock_mtx, 610 CALLOUT_RETURNUNLOCKED); 611 612 pst->current_qdelay = 0; 613 init_activate_pie(pst, !(pprms->flags & PIE_ON_OFF_MODE_ENABLED)); 614 615 //DX(2, "aqm_PIE_init"); 616 617 } while(0); 618 619 return err; 620 } 621 622 /* 623 * Callout function to destroy pie mtx and free PIE status memory 624 */ 625 static void 626 pie_callout_cleanup(void *x) 627 { 628 struct pie_status *pst = (struct pie_status *) x; 629 630 mtx_unlock(&pst->lock_mtx); 631 mtx_destroy(&pst->lock_mtx); 632 free(x, M_DUMMYNET); 633 DN_BH_WLOCK(); 634 pie_desc.ref_count--; 635 DN_BH_WUNLOCK(); 636 } 637 638 /* 639 * Clean up PIE status for queue 'q' 640 * Destroy memory allocated for PIE status. 641 */ 642 static int 643 aqm_pie_cleanup(struct dn_queue *q) 644 { 645 646 if(!q) { 647 D("q is null"); 648 return 0; 649 } 650 struct pie_status *pst = q->aqm_status; 651 if(!pst) { 652 //D("queue is already cleaned up"); 653 return 0; 654 } 655 if(!q->fs || !q->fs->aqmcfg) { 656 D("fs is null or no cfg"); 657 return 1; 658 } 659 if (q->fs->aqmfp && q->fs->aqmfp->type !=DN_AQM_PIE) { 660 D("Not PIE fs (%d)", q->fs->fs.fs_nr); 661 return 1; 662 } 663 664 /* 665 * Free PIE status allocated memory using pie_callout_cleanup() callout 666 * function to avoid any potential race. 667 * We reset aqm_pie_callout to call pie_callout_cleanup() in next 1um. This 668 * stops the scheduled calculate_drop_prob() callout and call pie_callout_cleanup() 669 * which does memory freeing. 670 */ 671 mtx_lock(&pst->lock_mtx); 672 callout_reset_sbt(&pst->aqm_pie_callout, 673 SBT_1US, 0, pie_callout_cleanup, pst, 0); 674 q->aqm_status = NULL; 675 mtx_unlock(&pst->lock_mtx); 676 677 return 0; 678 } 679 680 /* 681 * Config PIE parameters 682 * also allocate memory for PIE configurations 683 */ 684 static int 685 aqm_pie_config(struct dn_fsk* fs, struct dn_extra_parms *ep, int len) 686 { 687 struct dn_aqm_pie_parms *pcfg; 688 689 int l = sizeof(struct dn_extra_parms); 690 if (len < l) { 691 D("invalid sched parms length got %d need %d", len, l); 692 return EINVAL; 693 } 694 /* we free the old cfg because maybe the orignal allocation 695 * was used for diffirent AQM type. 696 */ 697 if (fs->aqmcfg) { 698 free(fs->aqmcfg, M_DUMMYNET); 699 fs->aqmcfg = NULL; 700 } 701 702 fs->aqmcfg = malloc(sizeof(struct dn_aqm_pie_parms), 703 M_DUMMYNET, M_NOWAIT | M_ZERO); 704 if (fs->aqmcfg== NULL) { 705 D("cannot allocate PIE configuration parameters"); 706 return ENOMEM; 707 } 708 709 /* par array contains pie configuration as follow 710 * 0- qdelay_ref,1- tupdate, 2- max_burst 711 * 3- max_ecnth, 4- alpha, 5- beta, 6- flags 712 */ 713 714 /* configure PIE parameters */ 715 pcfg = fs->aqmcfg; 716 717 if (ep->par[0] < 0) 718 pcfg->qdelay_ref = pie_sysctl.qdelay_ref * AQM_TIME_1US; 719 else 720 pcfg->qdelay_ref = ep->par[0]; 721 if (ep->par[1] < 0) 722 pcfg->tupdate = pie_sysctl.tupdate * AQM_TIME_1US; 723 else 724 pcfg->tupdate = ep->par[1]; 725 if (ep->par[2] < 0) 726 pcfg->max_burst = pie_sysctl.max_burst * AQM_TIME_1US; 727 else 728 pcfg->max_burst = ep->par[2]; 729 if (ep->par[3] < 0) 730 pcfg->max_ecnth = pie_sysctl.max_ecnth; 731 else 732 pcfg->max_ecnth = ep->par[3]; 733 if (ep->par[4] < 0) 734 pcfg->alpha = pie_sysctl.alpha; 735 else 736 pcfg->alpha = ep->par[4]; 737 if (ep->par[5] < 0) 738 pcfg->beta = pie_sysctl.beta; 739 else 740 pcfg->beta = ep->par[5]; 741 if (ep->par[6] < 0) 742 pcfg->flags = pie_sysctl.flags; 743 else 744 pcfg->flags = ep->par[6]; 745 746 /* bound PIE configurations */ 747 pcfg->qdelay_ref = BOUND_VAR(pcfg->qdelay_ref, 1, 10 * AQM_TIME_1S); 748 pcfg->tupdate = BOUND_VAR(pcfg->tupdate, 1, 10 * AQM_TIME_1S); 749 pcfg->max_burst = BOUND_VAR(pcfg->max_burst, 0, 10 * AQM_TIME_1S); 750 pcfg->max_ecnth = BOUND_VAR(pcfg->max_ecnth, 0, PIE_SCALE); 751 pcfg->alpha = BOUND_VAR(pcfg->alpha, 0, 7 * PIE_SCALE); 752 pcfg->beta = BOUND_VAR(pcfg->beta, 0 , 7 * PIE_SCALE); 753 754 pie_desc.cfg_ref_count++; 755 //D("pie cfg_ref_count=%d", pie_desc.cfg_ref_count); 756 return 0; 757 } 758 759 /* 760 * Deconfigure PIE and free memory allocation 761 */ 762 static int 763 aqm_pie_deconfig(struct dn_fsk* fs) 764 { 765 if (fs && fs->aqmcfg) { 766 free(fs->aqmcfg, M_DUMMYNET); 767 fs->aqmcfg = NULL; 768 pie_desc.cfg_ref_count--; 769 } 770 return 0; 771 } 772 773 /* 774 * Retrieve PIE configuration parameters. 775 */ 776 static int 777 aqm_pie_getconfig (struct dn_fsk *fs, struct dn_extra_parms * ep) 778 { 779 struct dn_aqm_pie_parms *pcfg; 780 if (fs->aqmcfg) { 781 strlcpy(ep->name, pie_desc.name, sizeof(ep->name)); 782 pcfg = fs->aqmcfg; 783 ep->par[0] = pcfg->qdelay_ref / AQM_TIME_1US; 784 ep->par[1] = pcfg->tupdate / AQM_TIME_1US; 785 ep->par[2] = pcfg->max_burst / AQM_TIME_1US; 786 ep->par[3] = pcfg->max_ecnth; 787 ep->par[4] = pcfg->alpha; 788 ep->par[5] = pcfg->beta; 789 ep->par[6] = pcfg->flags; 790 791 return 0; 792 } 793 return 1; 794 } 795 796 static struct dn_aqm pie_desc = { 797 _SI( .type = ) DN_AQM_PIE, 798 _SI( .name = ) "PIE", 799 _SI( .ref_count = ) 0, 800 _SI( .cfg_ref_count = ) 0, 801 _SI( .enqueue = ) aqm_pie_enqueue, 802 _SI( .dequeue = ) aqm_pie_dequeue, 803 _SI( .config = ) aqm_pie_config, 804 _SI( .deconfig = ) aqm_pie_deconfig, 805 _SI( .getconfig = ) aqm_pie_getconfig, 806 _SI( .init = ) aqm_pie_init, 807 _SI( .cleanup = ) aqm_pie_cleanup, 808 }; 809 810 DECLARE_DNAQM_MODULE(dn_aqm_pie, &pie_desc); 811 #endif 812