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 = q->mq.head; 332 333 if (m == NULL) 334 return m; 335 q->mq.head = m->m_nextpkt; 336 337 /* Update stats */ 338 update_stats(q, -m->m_pkthdr.len, 0); 339 340 if (q->ni.length == 0) /* queue is now idle */ 341 q->q_time = dn_cfg.curr_time; 342 343 if (getts) { 344 /* extract packet TS*/ 345 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL); 346 if (mtag == NULL) { 347 D("PIE timestamp mtag not found!"); 348 *pkt_ts = 0; 349 } else { 350 *pkt_ts = *(aqm_time_t *)(mtag + 1); 351 m_tag_delete(m,mtag); 352 } 353 } 354 return m; 355 } 356 357 /* 358 * Initiate PIE variable and optionally activate it 359 */ 360 __inline static void 361 init_activate_pie(struct pie_status *pst, int resettimer) 362 { 363 struct dn_aqm_pie_parms *pprms; 364 365 mtx_lock(&pst->lock_mtx); 366 pprms = pst->parms; 367 pst->drop_prob = 0; 368 pst->qdelay_old = 0; 369 pst->burst_allowance = pprms->max_burst; 370 pst->accu_prob = 0; 371 pst->dq_count = 0; 372 pst->avg_dq_time = 0; 373 pst->sflags = PIE_INMEASUREMENT; 374 pst->measurement_start = AQM_UNOW; 375 376 if (resettimer) { 377 pst->sflags |= PIE_ACTIVE; 378 callout_reset_sbt(&pst->aqm_pie_callout, 379 (uint64_t)pprms->tupdate * SBT_1US, 380 0, calculate_drop_prob, pst, 0); 381 } 382 //DX(2, "PIE Activated"); 383 mtx_unlock(&pst->lock_mtx); 384 } 385 386 /* 387 * Deactivate PIE and stop probe update callout 388 */ 389 __inline static void 390 deactivate_pie(struct pie_status *pst) 391 { 392 mtx_lock(&pst->lock_mtx); 393 pst->sflags &= ~(PIE_ACTIVE | PIE_INMEASUREMENT); 394 callout_stop(&pst->aqm_pie_callout); 395 //D("PIE Deactivated"); 396 mtx_unlock(&pst->lock_mtx); 397 } 398 399 /* 400 * Dequeue and return a pcaket from queue 'q' or NULL if 'q' is empty. 401 * Also, caculate depature time or queue delay using timestamp 402 */ 403 static struct mbuf * 404 aqm_pie_dequeue(struct dn_queue *q) 405 { 406 struct mbuf *m; 407 struct dn_flow *ni; /* stats for scheduler instance */ 408 struct dn_aqm_pie_parms *pprms; 409 struct pie_status *pst; 410 aqm_time_t now; 411 aqm_time_t pkt_ts, dq_time; 412 int32_t w; 413 414 pst = q->aqm_status; 415 pprms = pst->parms; 416 ni = &q->_si->ni; 417 418 /*we extarct packet ts only when Departure Rate Estimation dis not used*/ 419 m = pie_extract_head(q, &pkt_ts, !(pprms->flags & PIE_DEPRATEEST_ENABLED)); 420 421 if (!m || !(pst->sflags & PIE_ACTIVE)) 422 return m; 423 424 now = AQM_UNOW; 425 if (pprms->flags & PIE_DEPRATEEST_ENABLED) { 426 /* calculate average depature time */ 427 if(pst->sflags & PIE_INMEASUREMENT) { 428 pst->dq_count += m->m_pkthdr.len; 429 430 if (pst->dq_count >= PIE_DQ_THRESHOLD) { 431 dq_time = now - pst->measurement_start; 432 433 /* 434 * if we don't have old avg dq_time i.e PIE is (re)initialized, 435 * don't use weight to calculate new avg_dq_time 436 */ 437 if(pst->avg_dq_time == 0) 438 pst->avg_dq_time = dq_time; 439 else { 440 /* 441 * weight = PIE_DQ_THRESHOLD/2^6, but we scaled 442 * weight by 2^8. Thus, scaled 443 * weight = PIE_DQ_THRESHOLD /2^8 444 * */ 445 w = PIE_DQ_THRESHOLD >> 8; 446 pst->avg_dq_time = (dq_time* w 447 + (pst->avg_dq_time * ((1L << 8) - w))) >> 8; 448 pst->sflags &= ~PIE_INMEASUREMENT; 449 } 450 } 451 } 452 453 /* 454 * Start new measurment cycle when the queue has 455 * PIE_DQ_THRESHOLD worth of bytes. 456 */ 457 if(!(pst->sflags & PIE_INMEASUREMENT) && 458 q->ni.len_bytes >= PIE_DQ_THRESHOLD) { 459 pst->sflags |= PIE_INMEASUREMENT; 460 pst->measurement_start = now; 461 pst->dq_count = 0; 462 } 463 } 464 /* Optionally, use packet timestamp to estimate queue delay */ 465 else 466 pst->current_qdelay = now - pkt_ts; 467 468 return m; 469 } 470 471 /* 472 * Enqueue a packet in q, subject to space and PIE queue management policy 473 * (whose parameters are in q->fs). 474 * Update stats for the queue and the scheduler. 475 * Return 0 on success, 1 on drop. The packet is consumed anyways. 476 */ 477 static int 478 aqm_pie_enqueue(struct dn_queue *q, struct mbuf* m) 479 { 480 struct dn_fs *f; 481 uint64_t len; 482 uint32_t qlen; 483 struct pie_status *pst; 484 struct dn_aqm_pie_parms *pprms; 485 int t; 486 487 len = m->m_pkthdr.len; 488 pst = q->aqm_status; 489 if(!pst) { 490 DX(2, "PIE queue is not initialized\n"); 491 update_stats(q, 0, 1); 492 FREE_PKT(m); 493 return 1; 494 } 495 496 f = &(q->fs->fs); 497 pprms = pst->parms; 498 t = ENQUE; 499 500 /* get current queue length in bytes or packets*/ 501 qlen = (f->flags & DN_QSIZE_BYTES) ? 502 q->ni.len_bytes : q->ni.length; 503 504 /* check for queue size and drop the tail if exceed queue limit*/ 505 if (qlen >= f->qsize) 506 t = DROP; 507 /* drop/mark the packet when PIE is active and burst time elapsed */ 508 else if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance==0 509 && drop_early(pst, q->ni.len_bytes) == DROP) { 510 /* 511 * if drop_prob over ECN threshold, drop the packet 512 * otherwise mark and enqueue it. 513 */ 514 if ((pprms->flags & PIE_ECN_ENABLED) && pst->drop_prob < 515 (pprms->max_ecnth << (PIE_PROB_BITS - PIE_FIX_POINT_BITS)) 516 && ecn_mark(m)) 517 t = ENQUE; 518 else 519 t = DROP; 520 } 521 522 /* Turn PIE on when 1/3 of the queue is full */ 523 if (!(pst->sflags & PIE_ACTIVE) && qlen >= pst->one_third_q_size) { 524 init_activate_pie(pst, 1); 525 } 526 527 /* Reset burst tolerance and optinally turn PIE off*/ 528 if ((pst->sflags & PIE_ACTIVE) && pst->drop_prob == 0 && 529 pst->current_qdelay < (pprms->qdelay_ref >> 1) && 530 pst->qdelay_old < (pprms->qdelay_ref >> 1)) { 531 pst->burst_allowance = pprms->max_burst; 532 if ((pprms->flags & PIE_ON_OFF_MODE_ENABLED) && qlen<=0) 533 deactivate_pie(pst); 534 } 535 536 /* Timestamp the packet if Departure Rate Estimation is disabled */ 537 if (t != DROP && !(pprms->flags & PIE_DEPRATEEST_ENABLED)) { 538 /* Add TS to mbuf as a TAG */ 539 struct m_tag *mtag; 540 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL); 541 if (mtag == NULL) 542 mtag = m_tag_alloc(MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, 543 sizeof(aqm_time_t), M_NOWAIT); 544 if (mtag == NULL) { 545 m_freem(m); 546 t = DROP; 547 } 548 *(aqm_time_t *)(mtag + 1) = AQM_UNOW; 549 m_tag_prepend(m, mtag); 550 } 551 552 if (t != DROP) { 553 mq_append(&q->mq, m); 554 update_stats(q, len, 0); 555 return (0); 556 } else { 557 update_stats(q, 0, 1); 558 559 /* reset accu_prob after packet drop */ 560 pst->accu_prob = 0; 561 FREE_PKT(m); 562 return 1; 563 } 564 return 0; 565 } 566 567 /* 568 * initialize PIE for queue 'q' 569 * First allocate memory for PIE status. 570 */ 571 static int 572 aqm_pie_init(struct dn_queue *q) 573 { 574 struct pie_status *pst; 575 struct dn_aqm_pie_parms *pprms; 576 int err = 0; 577 578 pprms = q->fs->aqmcfg; 579 580 do { /* exit with break when error occurs*/ 581 if (!pprms){ 582 DX(2, "AQM_PIE is not configured"); 583 err = EINVAL; 584 break; 585 } 586 587 q->aqm_status = malloc(sizeof(struct pie_status), 588 M_DUMMYNET, M_NOWAIT | M_ZERO); 589 if (q->aqm_status == NULL) { 590 D("cannot allocate PIE private data"); 591 err = ENOMEM ; 592 break; 593 } 594 595 pst = q->aqm_status; 596 /* increase reference count for PIE module */ 597 pie_desc.ref_count++; 598 599 pst->pq = q; 600 pst->parms = pprms; 601 602 /* For speed optimization, we caculate 1/3 queue size once here */ 603 // we can use x/3 = (x >>2) + (x >>4) + (x >>7) 604 pst->one_third_q_size = q->fs->fs.qsize/3; 605 606 mtx_init(&pst->lock_mtx, "mtx_pie", NULL, MTX_DEF); 607 callout_init_mtx(&pst->aqm_pie_callout, &pst->lock_mtx, 608 CALLOUT_RETURNUNLOCKED); 609 610 pst->current_qdelay = 0; 611 init_activate_pie(pst, !(pprms->flags & PIE_ON_OFF_MODE_ENABLED)); 612 613 //DX(2, "aqm_PIE_init"); 614 615 } while(0); 616 617 return err; 618 } 619 620 /* 621 * Callout function to destroy pie mtx and free PIE status memory 622 */ 623 static void 624 pie_callout_cleanup(void *x) 625 { 626 struct pie_status *pst = (struct pie_status *) x; 627 628 mtx_unlock(&pst->lock_mtx); 629 mtx_destroy(&pst->lock_mtx); 630 free(x, M_DUMMYNET); 631 DN_BH_WLOCK(); 632 pie_desc.ref_count--; 633 DN_BH_WUNLOCK(); 634 } 635 636 /* 637 * Clean up PIE status for queue 'q' 638 * Destroy memory allocated for PIE status. 639 */ 640 static int 641 aqm_pie_cleanup(struct dn_queue *q) 642 { 643 644 if(!q) { 645 D("q is null"); 646 return 0; 647 } 648 struct pie_status *pst = q->aqm_status; 649 if(!pst) { 650 //D("queue is already cleaned up"); 651 return 0; 652 } 653 if(!q->fs || !q->fs->aqmcfg) { 654 D("fs is null or no cfg"); 655 return 1; 656 } 657 if (q->fs->aqmfp && q->fs->aqmfp->type !=DN_AQM_PIE) { 658 D("Not PIE fs (%d)", q->fs->fs.fs_nr); 659 return 1; 660 } 661 662 /* 663 * Free PIE status allocated memory using pie_callout_cleanup() callout 664 * function to avoid any potential race. 665 * We reset aqm_pie_callout to call pie_callout_cleanup() in next 1um. This 666 * stops the scheduled calculate_drop_prob() callout and call pie_callout_cleanup() 667 * which does memory freeing. 668 */ 669 mtx_lock(&pst->lock_mtx); 670 callout_reset_sbt(&pst->aqm_pie_callout, 671 SBT_1US, 0, pie_callout_cleanup, pst, 0); 672 q->aqm_status = NULL; 673 mtx_unlock(&pst->lock_mtx); 674 675 return 0; 676 } 677 678 /* 679 * Config PIE parameters 680 * also allocate memory for PIE configurations 681 */ 682 static int 683 aqm_pie_config(struct dn_fsk* fs, struct dn_extra_parms *ep, int len) 684 { 685 struct dn_aqm_pie_parms *pcfg; 686 687 int l = sizeof(struct dn_extra_parms); 688 if (len < l) { 689 D("invalid sched parms length got %d need %d", len, l); 690 return EINVAL; 691 } 692 /* we free the old cfg because maybe the orignal allocation 693 * was used for diffirent AQM type. 694 */ 695 if (fs->aqmcfg) { 696 free(fs->aqmcfg, M_DUMMYNET); 697 fs->aqmcfg = NULL; 698 } 699 700 fs->aqmcfg = malloc(sizeof(struct dn_aqm_pie_parms), 701 M_DUMMYNET, M_NOWAIT | M_ZERO); 702 if (fs->aqmcfg== NULL) { 703 D("cannot allocate PIE configuration parameters"); 704 return ENOMEM; 705 } 706 707 /* par array contains pie configuration as follow 708 * 0- qdelay_ref,1- tupdate, 2- max_burst 709 * 3- max_ecnth, 4- alpha, 5- beta, 6- flags 710 */ 711 712 /* configure PIE parameters */ 713 pcfg = fs->aqmcfg; 714 715 if (ep->par[0] < 0) 716 pcfg->qdelay_ref = pie_sysctl.qdelay_ref * AQM_TIME_1US; 717 else 718 pcfg->qdelay_ref = ep->par[0]; 719 if (ep->par[1] < 0) 720 pcfg->tupdate = pie_sysctl.tupdate * AQM_TIME_1US; 721 else 722 pcfg->tupdate = ep->par[1]; 723 if (ep->par[2] < 0) 724 pcfg->max_burst = pie_sysctl.max_burst * AQM_TIME_1US; 725 else 726 pcfg->max_burst = ep->par[2]; 727 if (ep->par[3] < 0) 728 pcfg->max_ecnth = pie_sysctl.max_ecnth; 729 else 730 pcfg->max_ecnth = ep->par[3]; 731 if (ep->par[4] < 0) 732 pcfg->alpha = pie_sysctl.alpha; 733 else 734 pcfg->alpha = ep->par[4]; 735 if (ep->par[5] < 0) 736 pcfg->beta = pie_sysctl.beta; 737 else 738 pcfg->beta = ep->par[5]; 739 if (ep->par[6] < 0) 740 pcfg->flags = pie_sysctl.flags; 741 else 742 pcfg->flags = ep->par[6]; 743 744 /* bound PIE configurations */ 745 pcfg->qdelay_ref = BOUND_VAR(pcfg->qdelay_ref, 1, 10 * AQM_TIME_1S); 746 pcfg->tupdate = BOUND_VAR(pcfg->tupdate, 1, 10 * AQM_TIME_1S); 747 pcfg->max_burst = BOUND_VAR(pcfg->max_burst, 0, 10 * AQM_TIME_1S); 748 pcfg->max_ecnth = BOUND_VAR(pcfg->max_ecnth, 0, PIE_SCALE); 749 pcfg->alpha = BOUND_VAR(pcfg->alpha, 0, 7 * PIE_SCALE); 750 pcfg->beta = BOUND_VAR(pcfg->beta, 0 , 7 * PIE_SCALE); 751 752 pie_desc.cfg_ref_count++; 753 //D("pie cfg_ref_count=%d", pie_desc.cfg_ref_count); 754 return 0; 755 } 756 757 /* 758 * Deconfigure PIE and free memory allocation 759 */ 760 static int 761 aqm_pie_deconfig(struct dn_fsk* fs) 762 { 763 if (fs && fs->aqmcfg) { 764 free(fs->aqmcfg, M_DUMMYNET); 765 fs->aqmcfg = NULL; 766 pie_desc.cfg_ref_count--; 767 } 768 return 0; 769 } 770 771 /* 772 * Retrieve PIE configuration parameters. 773 */ 774 static int 775 aqm_pie_getconfig (struct dn_fsk *fs, struct dn_extra_parms * ep) 776 { 777 struct dn_aqm_pie_parms *pcfg; 778 if (fs->aqmcfg) { 779 strlcpy(ep->name, pie_desc.name, sizeof(ep->name)); 780 pcfg = fs->aqmcfg; 781 ep->par[0] = pcfg->qdelay_ref / AQM_TIME_1US; 782 ep->par[1] = pcfg->tupdate / AQM_TIME_1US; 783 ep->par[2] = pcfg->max_burst / AQM_TIME_1US; 784 ep->par[3] = pcfg->max_ecnth; 785 ep->par[4] = pcfg->alpha; 786 ep->par[5] = pcfg->beta; 787 ep->par[6] = pcfg->flags; 788 789 return 0; 790 } 791 return 1; 792 } 793 794 static struct dn_aqm pie_desc = { 795 _SI( .type = ) DN_AQM_PIE, 796 _SI( .name = ) "PIE", 797 _SI( .ref_count = ) 0, 798 _SI( .cfg_ref_count = ) 0, 799 _SI( .enqueue = ) aqm_pie_enqueue, 800 _SI( .dequeue = ) aqm_pie_dequeue, 801 _SI( .config = ) aqm_pie_config, 802 _SI( .deconfig = ) aqm_pie_deconfig, 803 _SI( .getconfig = ) aqm_pie_getconfig, 804 _SI( .init = ) aqm_pie_init, 805 _SI( .cleanup = ) aqm_pie_cleanup, 806 }; 807 808 DECLARE_DNAQM_MODULE(dn_aqm_pie, &pie_desc); 809 #endif 810