2 * FQ_PIE - The FlowQueue-PIE scheduler/AQM
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>
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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
35 * As there is no an office document for FQ-PIE specification, we used
36 * FQ-CoDel algorithm with some modifications to implement FQ-PIE.
37 * This FQ-PIE implementation is a beta version and have not been tested
38 * extensively. Our FQ-PIE uses stand-alone PIE AQM per sub-queue. By
39 * default, timestamp is used to calculate queue delay instead of departure
40 * rate estimation method. Although departure rate estimation is available
41 * as testing option, the results could be incorrect. Moreover, turning PIE on
42 * and off option is available but it does not work properly in this version.
47 #include <sys/malloc.h>
48 #include <sys/socket.h>
49 #include <sys/kernel.h>
52 #include <sys/module.h>
53 #include <sys/mutex.h>
54 #include <net/if.h> /* IFNAMSIZ */
55 #include <netinet/in.h>
56 #include <netinet/ip_var.h> /* ipfw_rule_ref */
57 #include <netinet/ip_fw.h> /* flow_id */
58 #include <netinet/ip_dummynet.h>
61 #include <sys/rwlock.h>
63 #include <netpfil/ipfw/ip_fw_private.h>
64 #include <sys/sysctl.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip6.h>
67 #include <netinet/ip_icmp.h>
68 #include <netinet/tcp.h>
69 #include <netinet/udp.h>
70 #include <sys/queue.h>
73 #include <netpfil/ipfw/dn_heap.h>
74 #include <netpfil/ipfw/ip_dn_private.h>
76 #include <netpfil/ipfw/dn_aqm.h>
77 #include <netpfil/ipfw/dn_aqm_pie.h>
78 #include <netpfil/ipfw/dn_sched.h>
84 #define DN_SCHED_FQ_PIE 7
87 STAILQ_HEAD(fq_pie_list, fq_pie_flow) ;
89 /* FQ_PIE parameters including PIE */
90 struct dn_sch_fq_pie_parms {
91 struct dn_aqm_pie_parms pcfg; /* PIE configuration Parameters */
92 /* FQ_PIE Parameters */
93 uint32_t flows_cnt; /* number of flows */
94 uint32_t limit; /* hard limit of FQ_PIE queue size*/
98 /* flow (sub-queue) stats */
100 uint64_t tot_pkts; /* statistics counters */
102 uint32_t length; /* Queue length, in packets */
103 uint32_t len_bytes; /* Queue length, in bytes */
107 /* A flow of packets (sub-queue)*/
109 struct mq mq; /* list of packets */
110 struct flow_stats stats; /* statistics */
112 int active; /* 1: flow is active (in a list) */
113 struct pie_status pst; /* pie status variables */
114 struct fq_pie_si *psi; /* parent scheduler instance */
115 STAILQ_ENTRY(fq_pie_flow) flowchain;
118 /* extra fq_pie scheduler configurations */
120 struct dn_sch_fq_pie_parms cfg;
123 /* fq_pie scheduler instance */
125 struct dn_sch_inst _si; /* standard scheduler instance */
126 struct dn_queue main_q; /* main queue is after si directly */
127 uint32_t nr_active_q;
128 struct fq_pie_flow *flows; /* array of flows (queues) */
129 uint32_t perturbation; /* random value */
130 struct fq_pie_list newflows; /* list of new queues */
131 struct fq_pie_list oldflows; /* list of old queues */
139 static struct mtx freemem_mtx;
140 static struct dn_alg fq_pie_desc;
142 /* Default FQ-PIE parameters including PIE */
144 * target=15ms, max_burst=150ms, max_ecnth=0.1,
145 * alpha=0.125, beta=1.25, tupdate=15ms
147 * flows=1024, limit=10240, quantum =1514
149 struct dn_sch_fq_pie_parms
150 fq_pie_sysctl = {{15000 * AQM_TIME_1US, 15000 * AQM_TIME_1US,
151 150000 * AQM_TIME_1US, PIE_SCALE * 0.1, PIE_SCALE * 0.125,
152 PIE_SCALE * 1.25, PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED},
156 fqpie_sysctl_alpha_beta_handler(SYSCTL_HANDLER_ARGS)
161 if (!strcmp(oidp->oid_name,"alpha"))
162 value = fq_pie_sysctl.pcfg.alpha;
164 value = fq_pie_sysctl.pcfg.beta;
166 value = value * 1000 / PIE_SCALE;
167 error = sysctl_handle_long(oidp, &value, 0, req);
168 if (error != 0 || req->newptr == NULL)
170 if (value < 1 || value > 7 * PIE_SCALE)
172 value = (value * PIE_SCALE) / 1000;
173 if (!strcmp(oidp->oid_name,"alpha"))
174 fq_pie_sysctl.pcfg.alpha = value;
176 fq_pie_sysctl.pcfg.beta = value;
181 fqpie_sysctl_target_tupdate_maxb_handler(SYSCTL_HANDLER_ARGS)
186 if (!strcmp(oidp->oid_name,"target"))
187 value = fq_pie_sysctl.pcfg.qdelay_ref;
188 else if (!strcmp(oidp->oid_name,"tupdate"))
189 value = fq_pie_sysctl.pcfg.tupdate;
191 value = fq_pie_sysctl.pcfg.max_burst;
193 value = value / AQM_TIME_1US;
194 error = sysctl_handle_long(oidp, &value, 0, req);
195 if (error != 0 || req->newptr == NULL)
197 if (value < 1 || value > 10 * AQM_TIME_1S)
199 value = value * AQM_TIME_1US;
201 if (!strcmp(oidp->oid_name,"target"))
202 fq_pie_sysctl.pcfg.qdelay_ref = value;
203 else if (!strcmp(oidp->oid_name,"tupdate"))
204 fq_pie_sysctl.pcfg.tupdate = value;
206 fq_pie_sysctl.pcfg.max_burst = value;
211 fqpie_sysctl_max_ecnth_handler(SYSCTL_HANDLER_ARGS)
216 value = fq_pie_sysctl.pcfg.max_ecnth;
217 value = value * 1000 / PIE_SCALE;
218 error = sysctl_handle_long(oidp, &value, 0, req);
219 if (error != 0 || req->newptr == NULL)
221 if (value < 1 || value > PIE_SCALE)
223 value = (value * PIE_SCALE) / 1000;
224 fq_pie_sysctl.pcfg.max_ecnth = value;
228 /* define FQ- PIE sysctl variables */
230 SYSCTL_DECL(_net_inet);
231 SYSCTL_DECL(_net_inet_ip);
232 SYSCTL_DECL(_net_inet_ip_dummynet);
233 static SYSCTL_NODE(_net_inet_ip_dummynet, OID_AUTO, fqpie,
234 CTLFLAG_RW, 0, "FQ_PIE");
238 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, target,
239 CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
240 fqpie_sysctl_target_tupdate_maxb_handler, "L",
241 "queue target in microsecond");
243 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, tupdate,
244 CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
245 fqpie_sysctl_target_tupdate_maxb_handler, "L",
246 "the frequency of drop probability calculation in microsecond");
248 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, max_burst,
249 CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
250 fqpie_sysctl_target_tupdate_maxb_handler, "L",
251 "Burst allowance interval in microsecond");
253 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, max_ecnth,
254 CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
255 fqpie_sysctl_max_ecnth_handler, "L",
256 "ECN safeguard threshold scaled by 1000");
258 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, alpha,
259 CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
260 fqpie_sysctl_alpha_beta_handler, "L", "PIE alpha scaled by 1000");
262 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, beta,
263 CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
264 fqpie_sysctl_alpha_beta_handler, "L", "beta scaled by 1000");
266 SYSCTL_UINT(_net_inet_ip_dummynet_fqpie, OID_AUTO, quantum,
267 CTLFLAG_RW, &fq_pie_sysctl.quantum, 1514, "quantum for FQ_PIE");
268 SYSCTL_UINT(_net_inet_ip_dummynet_fqpie, OID_AUTO, flows,
269 CTLFLAG_RW, &fq_pie_sysctl.flows_cnt, 1024, "Number of queues for FQ_PIE");
270 SYSCTL_UINT(_net_inet_ip_dummynet_fqpie, OID_AUTO, limit,
271 CTLFLAG_RW, &fq_pie_sysctl.limit, 10240, "limit for FQ_PIE");
274 /* Helper function to update queue&main-queue and scheduler statistics.
275 * negative len & drop -> drop
276 * negative len -> dequeue
277 * positive len -> enqueue
278 * positive len + drop -> drop during enqueue
281 fq_update_stats(struct fq_pie_flow *q, struct fq_pie_si *si, int len,
292 si->main_q.ni.drops ++;
298 if (!drop || (drop && len < 0)) {
299 /* Update stats for the main queue */
300 si->main_q.ni.length += inc;
301 si->main_q.ni.len_bytes += len;
303 /*update sub-queue stats */
304 q->stats.length += inc;
305 q->stats.len_bytes += len;
307 /*update scheduler instance stats */
308 si->_si.ni.length += inc;
309 si->_si.ni.len_bytes += len;
313 si->main_q.ni.tot_bytes += len;
314 si->main_q.ni.tot_pkts ++;
316 q->stats.tot_bytes +=len;
319 si->_si.ni.tot_bytes +=len;
320 si->_si.ni.tot_pkts ++;
326 * Extract a packet from the head of sub-queue 'q'
327 * Return a packet or NULL if the queue is empty.
328 * If getts is set, also extract packet's timestamp from mtag.
330 __inline static struct mbuf *
331 fq_pie_extract_head(struct fq_pie_flow *q, aqm_time_t *pkt_ts,
332 struct fq_pie_si *si, int getts)
334 struct mbuf *m = q->mq.head;
338 q->mq.head = m->m_nextpkt;
340 fq_update_stats(q, si, -m->m_pkthdr.len, 0);
342 if (si->main_q.ni.length == 0) /* queue is now idle */
343 si->main_q.q_time = dn_cfg.curr_time;
346 /* extract packet timestamp*/
348 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL);
350 D("PIE timestamp mtag not found!");
353 *pkt_ts = *(aqm_time_t *)(mtag + 1);
354 m_tag_delete(m,mtag);
361 * Callout function for drop probability calculation
362 * This function is called over tupdate ms and takes pointer of FQ-PIE
363 * flow as an argument
366 fq_calculate_drop_prob(void *x)
368 struct fq_pie_flow *q = (struct fq_pie_flow *) x;
369 struct pie_status *pst = &q->pst;
370 struct dn_aqm_pie_parms *pprms;
371 int64_t p, prob, oldprob;
374 /* dealing with race condition */
375 if (callout_pending(&pst->aqm_pie_callout)) {
376 /* callout was reset */
377 mtx_unlock(&pst->lock_mtx);
381 if (!callout_active(&pst->aqm_pie_callout)) {
382 /* callout was stopped */
383 mtx_unlock(&pst->lock_mtx);
384 mtx_destroy(&pst->lock_mtx);
385 q->psi->nr_active_q--;
388 callout_deactivate(&pst->aqm_pie_callout);
392 prob = pst->drop_prob;
394 /* calculate current qdelay */
395 if (pprms->flags & PIE_DEPRATEEST_ENABLED) {
396 pst->current_qdelay = ((uint64_t)q->stats.len_bytes * pst->avg_dq_time)
397 >> PIE_DQ_THRESHOLD_BITS;
400 /* calculate drop probability */
401 p = (int64_t)pprms->alpha *
402 ((int64_t)pst->current_qdelay - (int64_t)pprms->qdelay_ref);
403 p +=(int64_t) pprms->beta *
404 ((int64_t)pst->current_qdelay - (int64_t)pst->qdelay_old);
406 /* We PIE_MAX_PROB shift by 12-bits to increase the division precision */
407 p *= (PIE_MAX_PROB << 12) / AQM_TIME_1S;
409 /* auto-tune drop probability */
410 if (prob < (PIE_MAX_PROB / 1000000)) /* 0.000001 */
411 p >>= 11 + PIE_FIX_POINT_BITS + 12;
412 else if (prob < (PIE_MAX_PROB / 100000)) /* 0.00001 */
413 p >>= 9 + PIE_FIX_POINT_BITS + 12;
414 else if (prob < (PIE_MAX_PROB / 10000)) /* 0.0001 */
415 p >>= 7 + PIE_FIX_POINT_BITS + 12;
416 else if (prob < (PIE_MAX_PROB / 1000)) /* 0.001 */
417 p >>= 5 + PIE_FIX_POINT_BITS + 12;
418 else if (prob < (PIE_MAX_PROB / 100)) /* 0.01 */
419 p >>= 3 + PIE_FIX_POINT_BITS + 12;
420 else if (prob < (PIE_MAX_PROB / 10)) /* 0.1 */
421 p >>= 1 + PIE_FIX_POINT_BITS + 12;
423 p >>= PIE_FIX_POINT_BITS + 12;
427 /* Cap Drop adjustment */
428 if ((pprms->flags & PIE_CAPDROP_ENABLED) && prob >= PIE_MAX_PROB / 10
429 && p > PIE_MAX_PROB / 50 )
430 p = PIE_MAX_PROB / 50;
434 /* decay the drop probability exponentially */
435 if (pst->current_qdelay == 0 && pst->qdelay_old == 0)
436 /* 0.98 ~= 1- 1/64 */
437 prob = prob - (prob >> 6);
440 /* check for multiplication over/under flow */
453 /* make drop probability between 0 and PIE_MAX_PROB*/
456 else if (prob > PIE_MAX_PROB)
459 pst->drop_prob = prob;
461 /* store current delay value */
462 pst->qdelay_old = pst->current_qdelay;
464 /* update burst allowance */
465 if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance) {
466 if (pst->burst_allowance > pprms->tupdate)
467 pst->burst_allowance -= pprms->tupdate;
469 pst->burst_allowance = 0;
472 if (pst->sflags & PIE_ACTIVE)
473 callout_reset_sbt(&pst->aqm_pie_callout,
474 (uint64_t)pprms->tupdate * SBT_1US,
475 0, fq_calculate_drop_prob, q, 0);
477 mtx_unlock(&pst->lock_mtx);
481 * Reset PIE variables & activate the queue
484 fq_activate_pie(struct fq_pie_flow *q)
486 struct pie_status *pst = &q->pst;
487 struct dn_aqm_pie_parms *pprms;
489 mtx_lock(&pst->lock_mtx);
495 pst->burst_allowance = pprms->max_burst;
498 pst->avg_dq_time = 0;
499 pst->sflags = PIE_INMEASUREMENT | PIE_ACTIVE;
500 pst->measurement_start = AQM_UNOW;
502 callout_reset_sbt(&pst->aqm_pie_callout,
503 (uint64_t)pprms->tupdate * SBT_1US,
504 0, fq_calculate_drop_prob, q, 0);
506 mtx_unlock(&pst->lock_mtx);
511 * Deactivate PIE and stop probe update callout
514 fq_deactivate_pie(struct pie_status *pst)
516 mtx_lock(&pst->lock_mtx);
517 pst->sflags &= ~(PIE_ACTIVE | PIE_INMEASUREMENT);
518 callout_stop(&pst->aqm_pie_callout);
519 //D("PIE Deactivated");
520 mtx_unlock(&pst->lock_mtx);
524 * Initialize PIE for sub-queue 'q'
527 pie_init(struct fq_pie_flow *q)
529 struct pie_status *pst=&q->pst;
530 struct dn_aqm_pie_parms *pprms = pst->parms;
531 struct fq_pie_schk *fqpie_schk;
533 fqpie_schk = (struct fq_pie_schk *)(q->psi->_si.sched+1);
537 D("AQM_PIE is not configured");
540 q->psi->nr_active_q++;
542 /* For speed optimization, we caculate 1/3 queue size once here */
543 // XXX limit divided by number of queues divided by 3 ???
544 pst->one_third_q_size = (fqpie_schk->cfg.limit /
545 fqpie_schk->cfg.flows_cnt) / 3;
547 mtx_init(&pst->lock_mtx, "mtx_pie", NULL, MTX_DEF);
548 callout_init_mtx(&pst->aqm_pie_callout, &pst->lock_mtx,
549 CALLOUT_RETURNUNLOCKED);
556 * Clean up PIE status for sub-queue 'q'
557 * Stop callout timer and destroy mtx
560 pie_cleanup(struct fq_pie_flow *q)
562 struct pie_status *pst = &q->pst;
564 mtx_lock(&pst->lock_mtx);
565 if (callout_stop(&pst->aqm_pie_callout) || !(pst->sflags & PIE_ACTIVE)) {
566 mtx_unlock(&pst->lock_mtx);
567 mtx_destroy(&pst->lock_mtx);
568 q->psi->nr_active_q--;
570 mtx_unlock(&pst->lock_mtx);
577 * Dequeue and return a pcaket from sub-queue 'q' or NULL if 'q' is empty.
578 * Also, caculate depature time or queue delay using timestamp
581 pie_dequeue(struct fq_pie_flow *q, struct fq_pie_si *si)
584 struct dn_aqm_pie_parms *pprms;
585 struct pie_status *pst;
587 aqm_time_t pkt_ts, dq_time;
591 pprms = q->pst.parms;
593 /*we extarct packet ts only when Departure Rate Estimation dis not used*/
594 m = fq_pie_extract_head(q, &pkt_ts, si,
595 !(pprms->flags & PIE_DEPRATEEST_ENABLED));
597 if (!m || !(pst->sflags & PIE_ACTIVE))
601 if (pprms->flags & PIE_DEPRATEEST_ENABLED) {
602 /* calculate average depature time */
603 if(pst->sflags & PIE_INMEASUREMENT) {
604 pst->dq_count += m->m_pkthdr.len;
606 if (pst->dq_count >= PIE_DQ_THRESHOLD) {
607 dq_time = now - pst->measurement_start;
610 * if we don't have old avg dq_time i.e PIE is (re)initialized,
611 * don't use weight to calculate new avg_dq_time
613 if(pst->avg_dq_time == 0)
614 pst->avg_dq_time = dq_time;
617 * weight = PIE_DQ_THRESHOLD/2^6, but we scaled
618 * weight by 2^8. Thus, scaled
619 * weight = PIE_DQ_THRESHOLD /2^8
621 w = PIE_DQ_THRESHOLD >> 8;
622 pst->avg_dq_time = (dq_time* w
623 + (pst->avg_dq_time * ((1L << 8) - w))) >> 8;
624 pst->sflags &= ~PIE_INMEASUREMENT;
630 * Start new measurment cycle when the queue has
631 * PIE_DQ_THRESHOLD worth of bytes.
633 if(!(pst->sflags & PIE_INMEASUREMENT) &&
634 q->stats.len_bytes >= PIE_DQ_THRESHOLD) {
635 pst->sflags |= PIE_INMEASUREMENT;
636 pst->measurement_start = now;
640 /* Optionally, use packet timestamp to estimate queue delay */
642 pst->current_qdelay = now - pkt_ts;
649 * Enqueue a packet in q, subject to space and FQ-PIE queue management policy
650 * (whose parameters are in q->fs).
651 * Update stats for the queue and the scheduler.
652 * Return 0 on success, 1 on drop. The packet is consumed anyways.
655 pie_enqueue(struct fq_pie_flow *q, struct mbuf* m, struct fq_pie_si *si)
658 struct pie_status *pst;
659 struct dn_aqm_pie_parms *pprms;
662 len = m->m_pkthdr.len;
667 /* drop/mark the packet when PIE is active and burst time elapsed */
668 if (pst->sflags & PIE_ACTIVE && pst->burst_allowance == 0
669 && drop_early(pst, q->stats.len_bytes) == DROP) {
671 * if drop_prob over ECN threshold, drop the packet
672 * otherwise mark and enqueue it.
674 if (pprms->flags & PIE_ECN_ENABLED && pst->drop_prob <
675 (pprms->max_ecnth << (PIE_PROB_BITS - PIE_FIX_POINT_BITS))
682 /* Turn PIE on when 1/3 of the queue is full */
683 if (!(pst->sflags & PIE_ACTIVE) && q->stats.len_bytes >=
684 pst->one_third_q_size) {
688 /* reset burst tolerance and optinally turn PIE off*/
689 if (pst->drop_prob == 0 && pst->current_qdelay < (pprms->qdelay_ref >> 1)
690 && pst->qdelay_old < (pprms->qdelay_ref >> 1)) {
692 pst->burst_allowance = pprms->max_burst;
693 if (pprms->flags & PIE_ON_OFF_MODE_ENABLED && q->stats.len_bytes<=0)
694 fq_deactivate_pie(pst);
697 /* Use timestamp if Departure Rate Estimation mode is disabled */
698 if (t != DROP && !(pprms->flags & PIE_DEPRATEEST_ENABLED)) {
699 /* Add TS to mbuf as a TAG */
701 mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL);
703 mtag = m_tag_alloc(MTAG_ABI_COMPAT, DN_AQM_MTAG_TS,
704 sizeof(aqm_time_t), M_NOWAIT);
709 *(aqm_time_t *)(mtag + 1) = AQM_UNOW;
710 m_tag_prepend(m, mtag);
714 mq_append(&q->mq, m);
715 fq_update_stats(q, si, len, 0);
718 fq_update_stats(q, si, len, 1);
727 /* Drop a packet form the head of FQ-PIE sub-queue */
729 pie_drop_head(struct fq_pie_flow *q, struct fq_pie_si *si)
731 struct mbuf *m = q->mq.head;
735 q->mq.head = m->m_nextpkt;
737 fq_update_stats(q, si, -m->m_pkthdr.len, 1);
739 if (si->main_q.ni.length == 0) /* queue is now idle */
740 si->main_q.q_time = dn_cfg.curr_time;
741 /* reset accu_prob after packet drop */
742 q->pst.accu_prob = 0;
748 * Classify a packet to queue number using Jenkins hash function.
749 * Return: queue number
750 * the input of the hash are protocol no, perturbation, src IP, dst IP,
751 * src port, dst port,
754 fq_pie_classify_flow(struct mbuf *m, uint16_t fcount, struct fq_pie_si *si)
765 isip6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
768 ip6 = mtod(m, struct ip6_hdr *);
769 *((uint8_t *) &tuple[0]) = ip6->ip6_nxt;
770 *((uint32_t *) &tuple[1]) = si->perturbation;
771 memcpy(&tuple[5], ip6->ip6_src.s6_addr, 16);
772 memcpy(&tuple[21], ip6->ip6_dst.s6_addr, 16);
774 switch (ip6->ip6_nxt) {
776 th = (struct tcphdr *)(ip6 + 1);
777 *((uint16_t *) &tuple[37]) = th->th_dport;
778 *((uint16_t *) &tuple[39]) = th->th_sport;
782 uh = (struct udphdr *)(ip6 + 1);
783 *((uint16_t *) &tuple[37]) = uh->uh_dport;
784 *((uint16_t *) &tuple[39]) = uh->uh_sport;
787 memset(&tuple[37], 0, 4);
790 hash = jenkins_hash(tuple, 41, HASHINIT) % fcount;
796 ip = mtod(m, struct ip *);
797 *((uint8_t *) &tuple[0]) = ip->ip_p;
798 *((uint32_t *) &tuple[1]) = si->perturbation;
799 *((uint32_t *) &tuple[5]) = ip->ip_src.s_addr;
800 *((uint32_t *) &tuple[9]) = ip->ip_dst.s_addr;
804 th = (struct tcphdr *)(ip + 1);
805 *((uint16_t *) &tuple[13]) = th->th_dport;
806 *((uint16_t *) &tuple[15]) = th->th_sport;
810 uh = (struct udphdr *)(ip + 1);
811 *((uint16_t *) &tuple[13]) = uh->uh_dport;
812 *((uint16_t *) &tuple[15]) = uh->uh_sport;
815 memset(&tuple[13], 0, 4);
817 hash = jenkins_hash(tuple, 17, HASHINIT) % fcount;
823 * Enqueue a packet into an appropriate queue according to
824 * FQ-CoDe; algorithm.
827 fq_pie_enqueue(struct dn_sch_inst *_si, struct dn_queue *_q,
830 struct fq_pie_si *si;
831 struct fq_pie_schk *schk;
832 struct dn_sch_fq_pie_parms *param;
833 struct dn_queue *mainq;
834 int idx, drop, i, maxidx;
836 mainq = (struct dn_queue *)(_si + 1);
837 si = (struct fq_pie_si *)_si;
838 schk = (struct fq_pie_schk *)(si->_si.sched+1);
841 /* classify a packet to queue number*/
842 idx = fq_pie_classify_flow(m, param->flows_cnt, si);
844 /* enqueue packet into appropriate queue using PIE AQM.
845 * Note: 'pie_enqueue' function returns 1 only when it unable to
846 * add timestamp to packet (no limit check)*/
847 drop = pie_enqueue(&si->flows[idx], m, si);
849 /* pie unable to timestamp a packet */
853 /* If the flow (sub-queue) is not active ,then add it to tail of
854 * new flows list, initialize and activate it.
856 if (!si->flows[idx].active) {
857 STAILQ_INSERT_TAIL(&si->newflows, &si->flows[idx], flowchain);
858 si->flows[idx].deficit = param->quantum;
859 fq_activate_pie(&si->flows[idx]);
860 si->flows[idx].active = 1;
863 /* check the limit for all queues and remove a packet from the
866 if (mainq->ni.length > schk->cfg.limit) {
867 /* find first active flow */
868 for (maxidx = 0; maxidx < schk->cfg.flows_cnt; maxidx++)
869 if (si->flows[maxidx].active)
871 if (maxidx < schk->cfg.flows_cnt) {
872 /* find the largest sub- queue */
873 for (i = maxidx + 1; i < schk->cfg.flows_cnt; i++)
874 if (si->flows[i].active && si->flows[i].stats.length >
875 si->flows[maxidx].stats.length)
877 pie_drop_head(&si->flows[maxidx], si);
886 * Dequeue a packet from an appropriate queue according to
887 * FQ-CoDel algorithm.
890 fq_pie_dequeue(struct dn_sch_inst *_si)
892 struct fq_pie_si *si;
893 struct fq_pie_schk *schk;
894 struct dn_sch_fq_pie_parms *param;
895 struct fq_pie_flow *f;
897 struct fq_pie_list *fq_pie_flowlist;
899 si = (struct fq_pie_si *)_si;
900 schk = (struct fq_pie_schk *)(si->_si.sched+1);
904 /* select a list to start with */
905 if (STAILQ_EMPTY(&si->newflows))
906 fq_pie_flowlist = &si->oldflows;
908 fq_pie_flowlist = &si->newflows;
910 /* Both new and old queue lists are empty, return NULL */
911 if (STAILQ_EMPTY(fq_pie_flowlist))
914 f = STAILQ_FIRST(fq_pie_flowlist);
916 /* if there is no flow(sub-queue) deficit, increase deficit
917 * by quantum, move the flow to the tail of old flows list
918 * and try another flow.
919 * Otherwise, the flow will be used for dequeue.
921 if (f->deficit < 0) {
922 f->deficit += param->quantum;
923 STAILQ_REMOVE_HEAD(fq_pie_flowlist, flowchain);
924 STAILQ_INSERT_TAIL(&si->oldflows, f, flowchain);
928 f = STAILQ_FIRST(fq_pie_flowlist);
931 /* the new flows list is empty, try old flows list */
932 if (STAILQ_EMPTY(fq_pie_flowlist))
935 /* Dequeue a packet from the selected flow */
936 mbuf = pie_dequeue(f, si);
938 /* pie did not return a packet */
940 /* If the selected flow belongs to new flows list, then move
941 * it to the tail of old flows list. Otherwise, deactivate it and
942 * remove it from the old list and
944 if (fq_pie_flowlist == &si->newflows) {
945 STAILQ_REMOVE_HEAD(fq_pie_flowlist, flowchain);
946 STAILQ_INSERT_TAIL(&si->oldflows, f, flowchain);
949 fq_deactivate_pie(&f->pst);
950 STAILQ_REMOVE_HEAD(fq_pie_flowlist, flowchain);
956 /* we have a packet to return,
957 * update flow deficit and return the packet*/
958 f->deficit -= mbuf->m_pkthdr.len;
963 /* unreachable point */
968 * Initialize fq_pie scheduler instance.
969 * also, allocate memory for flows array.
972 fq_pie_new_sched(struct dn_sch_inst *_si)
974 struct fq_pie_si *si;
976 struct fq_pie_schk *schk;
979 si = (struct fq_pie_si *)_si;
980 schk = (struct fq_pie_schk *)(_si->sched+1);
983 D("si already configured!");
987 /* init the main queue */
989 set_oid(&q->ni.oid, DN_QUEUE, sizeof(*q));
991 q->fs = _si->sched->fs;
993 /* allocate memory for flows array */
994 si->flows = malloc(schk->cfg.flows_cnt * sizeof(struct fq_pie_flow),
995 M_DUMMYNET, M_NOWAIT | M_ZERO);
996 if (si->flows == NULL) {
997 D("cannot allocate memory for fq_pie configuration parameters");
1001 /* init perturbation for this si */
1002 si->perturbation = random();
1003 si->nr_active_q = 0;
1005 /* init the old and new flows lists */
1006 STAILQ_INIT(&si->newflows);
1007 STAILQ_INIT(&si->oldflows);
1009 /* init the flows (sub-queues) */
1010 for (i = 0; i < schk->cfg.flows_cnt; i++) {
1011 si->flows[i].pst.parms = &schk->cfg.pcfg;
1012 si->flows[i].psi = si;
1013 pie_init(&si->flows[i]);
1016 /* init mtx lock and callout function for free memory */
1017 if (!fq_pie_desc.ref_count) {
1018 mtx_init(&freemem_mtx, "mtx_pie", NULL, MTX_DEF);
1021 mtx_lock(&freemem_mtx);
1022 fq_pie_desc.ref_count++;
1023 mtx_unlock(&freemem_mtx);
1029 * Free FQ-PIE flows memory callout function.
1030 * This function is scheduled when a flow or more still active and
1031 * the scheduer is about to be destroyed, to prevent memory leak.
1034 free_flows(void *_mem)
1036 struct mem_to_free *mem = _mem;
1038 free(mem->mem_flows, M_DUMMYNET);
1039 free(mem->mem_callout, M_DUMMYNET);
1040 free(_mem, M_DUMMYNET);
1042 fq_pie_desc.ref_count--;
1043 if (!fq_pie_desc.ref_count) {
1044 mtx_unlock(&freemem_mtx);
1045 mtx_destroy(&freemem_mtx);
1047 mtx_unlock(&freemem_mtx);
1048 //D("mem freed ok!");
1052 * Free fq_pie scheduler instance.
1055 fq_pie_free_sched(struct dn_sch_inst *_si)
1057 struct fq_pie_si *si;
1058 struct fq_pie_schk *schk;
1061 si = (struct fq_pie_si *)_si;
1062 schk = (struct fq_pie_schk *)(_si->sched+1);
1064 for (i = 0; i < schk->cfg.flows_cnt; i++) {
1065 pie_cleanup(&si->flows[i]);
1068 /* if there are still some queues have a callout going to start,
1069 * we cannot free flows memory. If we do so, a panic can happen
1070 * as prob calculate callout function uses flows memory.
1072 if (!si->nr_active_q) {
1073 /* free the flows array */
1074 free(si->flows , M_DUMMYNET);
1076 mtx_lock(&freemem_mtx);
1077 fq_pie_desc.ref_count--;
1078 if (!fq_pie_desc.ref_count) {
1079 mtx_unlock(&freemem_mtx);
1080 mtx_destroy(&freemem_mtx);
1082 mtx_unlock(&freemem_mtx);
1086 /* memory leak happens here. So, we register a callout function to free
1087 * flows memory later.
1089 D("unable to stop all fq_pie sub-queues!");
1090 mtx_lock(&freemem_mtx);
1092 struct callout *mem_callout;
1093 struct mem_to_free *mem;
1095 mem = malloc(sizeof(*mem), M_DUMMYNET,
1097 mem_callout = malloc(sizeof(*mem_callout), M_DUMMYNET,
1100 callout_init_mtx(mem_callout, &freemem_mtx,
1101 CALLOUT_RETURNUNLOCKED);
1103 mem->mem_flows = si->flows;
1104 mem->mem_callout = mem_callout;
1105 callout_reset_sbt(mem_callout,
1106 (uint64_t)(si->flows[0].pst.parms->tupdate + 1000) * SBT_1US,
1107 0, free_flows, mem, 0);
1110 mtx_unlock(&freemem_mtx);
1117 * Configure FQ-PIE scheduler.
1118 * the configurations for the scheduler is passed fromipfw userland.
1121 fq_pie_config(struct dn_schk *_schk)
1123 struct fq_pie_schk *schk;
1124 struct dn_extra_parms *ep;
1125 struct dn_sch_fq_pie_parms *fqp_cfg;
1127 schk = (struct fq_pie_schk *)(_schk+1);
1128 ep = (struct dn_extra_parms *) _schk->cfg;
1130 /* par array contains fq_pie configuration as follow
1131 * PIE: 0- qdelay_ref,1- tupdate, 2- max_burst
1132 * 3- max_ecnth, 4- alpha, 5- beta, 6- flags
1133 * FQ_PIE: 7- quantum, 8- limit, 9- flows
1135 if (ep && ep->oid.len ==sizeof(*ep) &&
1136 ep->oid.subtype == DN_SCH_PARAMS) {
1138 fqp_cfg = &schk->cfg;
1140 fqp_cfg->pcfg.qdelay_ref = fq_pie_sysctl.pcfg.qdelay_ref;
1142 fqp_cfg->pcfg.qdelay_ref = ep->par[0];
1144 fqp_cfg->pcfg.tupdate = fq_pie_sysctl.pcfg.tupdate;
1146 fqp_cfg->pcfg.tupdate = ep->par[1];
1148 fqp_cfg->pcfg.max_burst = fq_pie_sysctl.pcfg.max_burst;
1150 fqp_cfg->pcfg.max_burst = ep->par[2];
1152 fqp_cfg->pcfg.max_ecnth = fq_pie_sysctl.pcfg.max_ecnth;
1154 fqp_cfg->pcfg.max_ecnth = ep->par[3];
1156 fqp_cfg->pcfg.alpha = fq_pie_sysctl.pcfg.alpha;
1158 fqp_cfg->pcfg.alpha = ep->par[4];
1160 fqp_cfg->pcfg.beta = fq_pie_sysctl.pcfg.beta;
1162 fqp_cfg->pcfg.beta = ep->par[5];
1164 fqp_cfg->pcfg.flags = 0;
1166 fqp_cfg->pcfg.flags = ep->par[6];
1168 /* FQ configurations */
1170 fqp_cfg->quantum = fq_pie_sysctl.quantum;
1172 fqp_cfg->quantum = ep->par[7];
1174 fqp_cfg->limit = fq_pie_sysctl.limit;
1176 fqp_cfg->limit = ep->par[8];
1178 fqp_cfg->flows_cnt = fq_pie_sysctl.flows_cnt;
1180 fqp_cfg->flows_cnt = ep->par[9];
1182 /* Bound the configurations */
1183 fqp_cfg->pcfg.qdelay_ref = BOUND_VAR(fqp_cfg->pcfg.qdelay_ref,
1184 1, 5 * AQM_TIME_1S);
1185 fqp_cfg->pcfg.tupdate = BOUND_VAR(fqp_cfg->pcfg.tupdate,
1186 1, 5 * AQM_TIME_1S);
1187 fqp_cfg->pcfg.max_burst = BOUND_VAR(fqp_cfg->pcfg.max_burst,
1188 0, 5 * AQM_TIME_1S);
1189 fqp_cfg->pcfg.max_ecnth = BOUND_VAR(fqp_cfg->pcfg.max_ecnth,
1191 fqp_cfg->pcfg.alpha = BOUND_VAR(fqp_cfg->pcfg.alpha, 0, 7 * PIE_SCALE);
1192 fqp_cfg->pcfg.beta = BOUND_VAR(fqp_cfg->pcfg.beta, 0, 7 * PIE_SCALE);
1194 fqp_cfg->quantum = BOUND_VAR(fqp_cfg->quantum,1,9000);
1195 fqp_cfg->limit= BOUND_VAR(fqp_cfg->limit,1,20480);
1196 fqp_cfg->flows_cnt= BOUND_VAR(fqp_cfg->flows_cnt,1,65536);
1199 D("Wrong parameters for fq_pie scheduler");
1207 * Return FQ-PIE scheduler configurations
1208 * the configurations for the scheduler is passed to userland.
1211 fq_pie_getconfig (struct dn_schk *_schk, struct dn_extra_parms *ep) {
1213 struct fq_pie_schk *schk = (struct fq_pie_schk *)(_schk+1);
1214 struct dn_sch_fq_pie_parms *fqp_cfg;
1216 fqp_cfg = &schk->cfg;
1218 strcpy(ep->name, fq_pie_desc.name);
1219 ep->par[0] = fqp_cfg->pcfg.qdelay_ref;
1220 ep->par[1] = fqp_cfg->pcfg.tupdate;
1221 ep->par[2] = fqp_cfg->pcfg.max_burst;
1222 ep->par[3] = fqp_cfg->pcfg.max_ecnth;
1223 ep->par[4] = fqp_cfg->pcfg.alpha;
1224 ep->par[5] = fqp_cfg->pcfg.beta;
1225 ep->par[6] = fqp_cfg->pcfg.flags;
1227 ep->par[7] = fqp_cfg->quantum;
1228 ep->par[8] = fqp_cfg->limit;
1229 ep->par[9] = fqp_cfg->flows_cnt;
1235 * FQ-PIE scheduler descriptor
1236 * contains the type of the scheduler, the name, the size of extra
1237 * data structures, and function pointers.
1239 static struct dn_alg fq_pie_desc = {
1240 _SI( .type = ) DN_SCHED_FQ_PIE,
1241 _SI( .name = ) "FQ_PIE",
1244 _SI( .schk_datalen = ) sizeof(struct fq_pie_schk),
1245 _SI( .si_datalen = ) sizeof(struct fq_pie_si) - sizeof(struct dn_sch_inst),
1246 _SI( .q_datalen = ) 0,
1248 _SI( .enqueue = ) fq_pie_enqueue,
1249 _SI( .dequeue = ) fq_pie_dequeue,
1250 _SI( .config = ) fq_pie_config, /* new sched i.e. sched X config ...*/
1251 _SI( .destroy = ) NULL, /*sched x delete */
1252 _SI( .new_sched = ) fq_pie_new_sched, /* new schd instance */
1253 _SI( .free_sched = ) fq_pie_free_sched, /* delete schd instance */
1254 _SI( .new_fsk = ) NULL,
1255 _SI( .free_fsk = ) NULL,
1256 _SI( .new_queue = ) NULL,
1257 _SI( .free_queue = ) NULL,
1258 _SI( .getconfig = ) fq_pie_getconfig,
1259 _SI( .ref_count = ) 0
1262 DECLARE_DNSCHED_MODULE(dn_fq_pie, &fq_pie_desc);