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