xref: /freebsd/sys/netpfil/ipfw/dn_sched_fq_pie.c (revision 076ad2f8)
1 /*
2  * FQ_PIE - The FlowQueue-PIE scheduler/AQM
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 /* Important note:
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.
43  */
44 
45 
46 #ifdef _KERNEL
47 #include <sys/malloc.h>
48 #include <sys/socket.h>
49 #include <sys/kernel.h>
50 #include <sys/mbuf.h>
51 #include <sys/lock.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>
59 
60 #include <sys/proc.h>
61 #include <sys/rwlock.h>
62 
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>
71 #include <sys/hash.h>
72 
73 #include <netpfil/ipfw/dn_heap.h>
74 #include <netpfil/ipfw/ip_dn_private.h>
75 
76 #include <netpfil/ipfw/dn_aqm.h>
77 #include <netpfil/ipfw/dn_aqm_pie.h>
78 #include <netpfil/ipfw/dn_sched.h>
79 
80 #else
81 #include <dn_test.h>
82 #endif
83 
84 #define DN_SCHED_FQ_PIE 7
85 
86 /* list of queues */
87 STAILQ_HEAD(fq_pie_list, fq_pie_flow) ;
88 
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*/
95 	uint32_t quantum;
96 };
97 
98 /* flow (sub-queue) stats */
99 struct flow_stats {
100 	uint64_t tot_pkts;	/* statistics counters  */
101 	uint64_t tot_bytes;
102 	uint32_t length;		/* Queue length, in packets */
103 	uint32_t len_bytes;	/* Queue length, in bytes */
104 	uint32_t drops;
105 };
106 
107 /* A flow of packets (sub-queue)*/
108 struct fq_pie_flow {
109 	struct mq	mq;	/* list of packets */
110 	struct flow_stats stats;	/* statistics */
111 	int deficit;
112 	int active;		/* 1: flow is active (in a list) */
113 	struct pie_status pst;	/* pie status variables */
114 	struct fq_pie_si_extra *psi_extra;
115 	STAILQ_ENTRY(fq_pie_flow) flowchain;
116 };
117 
118 /* extra fq_pie scheduler configurations */
119 struct fq_pie_schk {
120 	struct dn_sch_fq_pie_parms cfg;
121 };
122 
123 
124 /* fq_pie scheduler instance extra state vars.
125  * The purpose of separation this structure is to preserve number of active
126  * sub-queues and the flows array pointer even after the scheduler instance
127  * is destroyed.
128  * Preserving these varaiables allows freeing the allocated memory by
129  * fqpie_callout_cleanup() independently from fq_pie_free_sched().
130  */
131 struct fq_pie_si_extra {
132 	uint32_t nr_active_q;	/* number of active queues */
133 	struct fq_pie_flow *flows;	/* array of flows (queues) */
134 	};
135 
136 /* fq_pie scheduler instance */
137 struct fq_pie_si {
138 	struct dn_sch_inst _si;	/* standard scheduler instance. SHOULD BE FIRST */
139 	struct dn_queue main_q; /* main queue is after si directly */
140 	uint32_t perturbation; 	/* random value */
141 	struct fq_pie_list newflows;	/* list of new queues */
142 	struct fq_pie_list oldflows;	/* list of old queues */
143 	struct fq_pie_si_extra *si_extra; /* extra state vars*/
144 };
145 
146 
147 static struct dn_alg fq_pie_desc;
148 
149 /*  Default FQ-PIE parameters including PIE */
150 /*  PIE defaults
151  * target=15ms, max_burst=150ms, max_ecnth=0.1,
152  * alpha=0.125, beta=1.25, tupdate=15ms
153  * FQ-
154  * flows=1024, limit=10240, quantum =1514
155  */
156 struct dn_sch_fq_pie_parms
157  fq_pie_sysctl = {{15000 * AQM_TIME_1US, 15000 * AQM_TIME_1US,
158 	150000 * AQM_TIME_1US, PIE_SCALE * 0.1, PIE_SCALE * 0.125,
159 	PIE_SCALE * 1.25,	PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED},
160 	1024, 10240, 1514};
161 
162 static int
163 fqpie_sysctl_alpha_beta_handler(SYSCTL_HANDLER_ARGS)
164 {
165 	int error;
166 	long  value;
167 
168 	if (!strcmp(oidp->oid_name,"alpha"))
169 		value = fq_pie_sysctl.pcfg.alpha;
170 	else
171 		value = fq_pie_sysctl.pcfg.beta;
172 
173 	value = value * 1000 / PIE_SCALE;
174 	error = sysctl_handle_long(oidp, &value, 0, req);
175 	if (error != 0 || req->newptr == NULL)
176 		return (error);
177 	if (value < 1 || value > 7 * PIE_SCALE)
178 		return (EINVAL);
179 	value = (value * PIE_SCALE) / 1000;
180 	if (!strcmp(oidp->oid_name,"alpha"))
181 			fq_pie_sysctl.pcfg.alpha = value;
182 	else
183 		fq_pie_sysctl.pcfg.beta = value;
184 	return (0);
185 }
186 
187 static int
188 fqpie_sysctl_target_tupdate_maxb_handler(SYSCTL_HANDLER_ARGS)
189 {
190 	int error;
191 	long  value;
192 
193 	if (!strcmp(oidp->oid_name,"target"))
194 		value = fq_pie_sysctl.pcfg.qdelay_ref;
195 	else if (!strcmp(oidp->oid_name,"tupdate"))
196 		value = fq_pie_sysctl.pcfg.tupdate;
197 	else
198 		value = fq_pie_sysctl.pcfg.max_burst;
199 
200 	value = value / AQM_TIME_1US;
201 	error = sysctl_handle_long(oidp, &value, 0, req);
202 	if (error != 0 || req->newptr == NULL)
203 		return (error);
204 	if (value < 1 || value > 10 * AQM_TIME_1S)
205 		return (EINVAL);
206 	value = value * AQM_TIME_1US;
207 
208 	if (!strcmp(oidp->oid_name,"target"))
209 		fq_pie_sysctl.pcfg.qdelay_ref  = value;
210 	else if (!strcmp(oidp->oid_name,"tupdate"))
211 		fq_pie_sysctl.pcfg.tupdate  = value;
212 	else
213 		fq_pie_sysctl.pcfg.max_burst = value;
214 	return (0);
215 }
216 
217 static int
218 fqpie_sysctl_max_ecnth_handler(SYSCTL_HANDLER_ARGS)
219 {
220 	int error;
221 	long  value;
222 
223 	value = fq_pie_sysctl.pcfg.max_ecnth;
224 	value = value * 1000 / PIE_SCALE;
225 	error = sysctl_handle_long(oidp, &value, 0, req);
226 	if (error != 0 || req->newptr == NULL)
227 		return (error);
228 	if (value < 1 || value > PIE_SCALE)
229 		return (EINVAL);
230 	value = (value * PIE_SCALE) / 1000;
231 	fq_pie_sysctl.pcfg.max_ecnth = value;
232 	return (0);
233 }
234 
235 /* define FQ- PIE sysctl variables */
236 SYSBEGIN(f4)
237 SYSCTL_DECL(_net_inet);
238 SYSCTL_DECL(_net_inet_ip);
239 SYSCTL_DECL(_net_inet_ip_dummynet);
240 static SYSCTL_NODE(_net_inet_ip_dummynet, OID_AUTO, fqpie,
241 	CTLFLAG_RW, 0, "FQ_PIE");
242 
243 #ifdef SYSCTL_NODE
244 
245 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, target,
246 	CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
247 	fqpie_sysctl_target_tupdate_maxb_handler, "L",
248 	"queue target in microsecond");
249 
250 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, tupdate,
251 	CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
252 	fqpie_sysctl_target_tupdate_maxb_handler, "L",
253 	"the frequency of drop probability calculation in microsecond");
254 
255 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, max_burst,
256 	CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
257 	fqpie_sysctl_target_tupdate_maxb_handler, "L",
258 	"Burst allowance interval in microsecond");
259 
260 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, max_ecnth,
261 	CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
262 	fqpie_sysctl_max_ecnth_handler, "L",
263 	"ECN safeguard threshold scaled by 1000");
264 
265 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, alpha,
266 	CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
267 	fqpie_sysctl_alpha_beta_handler, "L", "PIE alpha scaled by 1000");
268 
269 SYSCTL_PROC(_net_inet_ip_dummynet_fqpie, OID_AUTO, beta,
270 	CTLTYPE_LONG | CTLFLAG_RW, NULL, 0,
271 	fqpie_sysctl_alpha_beta_handler, "L", "beta scaled by 1000");
272 
273 SYSCTL_UINT(_net_inet_ip_dummynet_fqpie, OID_AUTO, quantum,
274 	CTLFLAG_RW, &fq_pie_sysctl.quantum, 1514, "quantum for FQ_PIE");
275 SYSCTL_UINT(_net_inet_ip_dummynet_fqpie, OID_AUTO, flows,
276 	CTLFLAG_RW, &fq_pie_sysctl.flows_cnt, 1024, "Number of queues for FQ_PIE");
277 SYSCTL_UINT(_net_inet_ip_dummynet_fqpie, OID_AUTO, limit,
278 	CTLFLAG_RW, &fq_pie_sysctl.limit, 10240, "limit for FQ_PIE");
279 #endif
280 
281 /* Helper function to update queue&main-queue and scheduler statistics.
282  * negative len & drop -> drop
283  * negative len -> dequeue
284  * positive len -> enqueue
285  * positive len + drop -> drop during enqueue
286  */
287 __inline static void
288 fq_update_stats(struct fq_pie_flow *q, struct fq_pie_si *si, int len,
289 	int drop)
290 {
291 	int inc = 0;
292 
293 	if (len < 0)
294 		inc = -1;
295 	else if (len > 0)
296 		inc = 1;
297 
298 	if (drop) {
299 		si->main_q.ni.drops ++;
300 		q->stats.drops ++;
301 		si->_si.ni.drops ++;
302 		io_pkt_drop ++;
303 	}
304 
305 	if (!drop || (drop && len < 0)) {
306 		/* Update stats for the main queue */
307 		si->main_q.ni.length += inc;
308 		si->main_q.ni.len_bytes += len;
309 
310 		/*update sub-queue stats */
311 		q->stats.length += inc;
312 		q->stats.len_bytes += len;
313 
314 		/*update scheduler instance stats */
315 		si->_si.ni.length += inc;
316 		si->_si.ni.len_bytes += len;
317 	}
318 
319 	if (inc > 0) {
320 		si->main_q.ni.tot_bytes += len;
321 		si->main_q.ni.tot_pkts ++;
322 
323 		q->stats.tot_bytes +=len;
324 		q->stats.tot_pkts++;
325 
326 		si->_si.ni.tot_bytes +=len;
327 		si->_si.ni.tot_pkts ++;
328 	}
329 
330 }
331 
332 /*
333  * Extract a packet from the head of sub-queue 'q'
334  * Return a packet or NULL if the queue is empty.
335  * If getts is set, also extract packet's timestamp from mtag.
336  */
337 __inline static struct mbuf *
338 fq_pie_extract_head(struct fq_pie_flow *q, aqm_time_t *pkt_ts,
339 	struct fq_pie_si *si, int getts)
340 {
341 	struct mbuf *m = q->mq.head;
342 
343 	if (m == NULL)
344 		return m;
345 	q->mq.head = m->m_nextpkt;
346 
347 	fq_update_stats(q, si, -m->m_pkthdr.len, 0);
348 
349 	if (si->main_q.ni.length == 0) /* queue is now idle */
350 			si->main_q.q_time = dn_cfg.curr_time;
351 
352 	if (getts) {
353 		/* extract packet timestamp*/
354 		struct m_tag *mtag;
355 		mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL);
356 		if (mtag == NULL){
357 			D("PIE timestamp mtag not found!");
358 			*pkt_ts = 0;
359 		} else {
360 			*pkt_ts = *(aqm_time_t *)(mtag + 1);
361 			m_tag_delete(m,mtag);
362 		}
363 	}
364 	return m;
365 }
366 
367 /*
368  * Callout function for drop probability calculation
369  * This function is called over tupdate ms and takes pointer of FQ-PIE
370  * flow as an argument
371   */
372 static void
373 fq_calculate_drop_prob(void *x)
374 {
375 	struct fq_pie_flow *q = (struct fq_pie_flow *) x;
376 	struct pie_status *pst = &q->pst;
377 	struct dn_aqm_pie_parms *pprms;
378 	int64_t p, prob, oldprob;
379 	aqm_time_t now;
380 
381 	now = AQM_UNOW;
382 	pprms = pst->parms;
383 	prob = pst->drop_prob;
384 
385 	/* calculate current qdelay */
386 	if (pprms->flags & PIE_DEPRATEEST_ENABLED) {
387 		pst->current_qdelay = ((uint64_t)q->stats.len_bytes  * pst->avg_dq_time)
388 			>> PIE_DQ_THRESHOLD_BITS;
389 	}
390 
391 	/* calculate drop probability */
392 	p = (int64_t)pprms->alpha *
393 		((int64_t)pst->current_qdelay - (int64_t)pprms->qdelay_ref);
394 	p +=(int64_t) pprms->beta *
395 		((int64_t)pst->current_qdelay - (int64_t)pst->qdelay_old);
396 
397 	/* We PIE_MAX_PROB shift by 12-bits to increase the division precision  */
398 	p *= (PIE_MAX_PROB << 12) / AQM_TIME_1S;
399 
400 	/* auto-tune drop probability */
401 	if (prob < (PIE_MAX_PROB / 1000000)) /* 0.000001 */
402 		p >>= 11 + PIE_FIX_POINT_BITS + 12;
403 	else if (prob < (PIE_MAX_PROB / 100000)) /* 0.00001 */
404 		p >>= 9 + PIE_FIX_POINT_BITS + 12;
405 	else if (prob < (PIE_MAX_PROB / 10000)) /* 0.0001 */
406 		p >>= 7 + PIE_FIX_POINT_BITS + 12;
407 	else if (prob < (PIE_MAX_PROB / 1000)) /* 0.001 */
408 		p >>= 5 + PIE_FIX_POINT_BITS + 12;
409 	else if (prob < (PIE_MAX_PROB / 100)) /* 0.01 */
410 		p >>= 3 + PIE_FIX_POINT_BITS + 12;
411 	else if (prob < (PIE_MAX_PROB / 10)) /* 0.1 */
412 		p >>= 1 + PIE_FIX_POINT_BITS + 12;
413 	else
414 		p >>= PIE_FIX_POINT_BITS + 12;
415 
416 	oldprob = prob;
417 
418 	/* Cap Drop adjustment */
419 	if ((pprms->flags & PIE_CAPDROP_ENABLED) && prob >= PIE_MAX_PROB / 10
420 		&& p > PIE_MAX_PROB / 50 )
421 			p = PIE_MAX_PROB / 50;
422 
423 	prob = prob + p;
424 
425 	/* decay the drop probability exponentially */
426 	if (pst->current_qdelay == 0 && pst->qdelay_old == 0)
427 		/* 0.98 ~= 1- 1/64 */
428 		prob = prob - (prob >> 6);
429 
430 
431 	/* check for multiplication over/under flow */
432 	if (p>0) {
433 		if (prob<oldprob) {
434 			D("overflow");
435 			prob= PIE_MAX_PROB;
436 		}
437 	}
438 	else
439 		if (prob>oldprob) {
440 			prob= 0;
441 			D("underflow");
442 		}
443 
444 	/* make drop probability between 0 and PIE_MAX_PROB*/
445 	if (prob < 0)
446 		prob = 0;
447 	else if (prob > PIE_MAX_PROB)
448 		prob = PIE_MAX_PROB;
449 
450 	pst->drop_prob = prob;
451 
452 	/* store current delay value */
453 	pst->qdelay_old = pst->current_qdelay;
454 
455 	/* update burst allowance */
456 	if ((pst->sflags & PIE_ACTIVE) && pst->burst_allowance) {
457 		if (pst->burst_allowance > pprms->tupdate)
458 			pst->burst_allowance -= pprms->tupdate;
459 		else
460 			pst->burst_allowance = 0;
461 	}
462 
463 	if (pst->sflags & PIE_ACTIVE)
464 	callout_reset_sbt(&pst->aqm_pie_callout,
465 		(uint64_t)pprms->tupdate * SBT_1US,
466 		0, fq_calculate_drop_prob, q, 0);
467 
468 	mtx_unlock(&pst->lock_mtx);
469 }
470 
471 /*
472  * Reset PIE variables & activate the queue
473  */
474 __inline static void
475 fq_activate_pie(struct fq_pie_flow *q)
476 {
477 	struct pie_status *pst = &q->pst;
478 	struct dn_aqm_pie_parms *pprms;
479 
480 	mtx_lock(&pst->lock_mtx);
481 	pprms = pst->parms;
482 
483 	pprms = pst->parms;
484 	pst->drop_prob = 0;
485 	pst->qdelay_old = 0;
486 	pst->burst_allowance = pprms->max_burst;
487 	pst->accu_prob = 0;
488 	pst->dq_count = 0;
489 	pst->avg_dq_time = 0;
490 	pst->sflags = PIE_INMEASUREMENT | PIE_ACTIVE;
491 	pst->measurement_start = AQM_UNOW;
492 
493 	callout_reset_sbt(&pst->aqm_pie_callout,
494 		(uint64_t)pprms->tupdate * SBT_1US,
495 		0, fq_calculate_drop_prob, q, 0);
496 
497 	mtx_unlock(&pst->lock_mtx);
498 }
499 
500 
501  /*
502   * Deactivate PIE and stop probe update callout
503   */
504 __inline static void
505 fq_deactivate_pie(struct pie_status *pst)
506 {
507 	mtx_lock(&pst->lock_mtx);
508 	pst->sflags &= ~(PIE_ACTIVE | PIE_INMEASUREMENT);
509 	callout_stop(&pst->aqm_pie_callout);
510 	//D("PIE Deactivated");
511 	mtx_unlock(&pst->lock_mtx);
512 }
513 
514  /*
515   * Initialize PIE for sub-queue 'q'
516   */
517 static int
518 pie_init(struct fq_pie_flow *q, struct fq_pie_schk *fqpie_schk)
519 {
520 	struct pie_status *pst=&q->pst;
521 	struct dn_aqm_pie_parms *pprms = pst->parms;
522 
523 	int err = 0;
524 	if (!pprms){
525 		D("AQM_PIE is not configured");
526 		err = EINVAL;
527 	} else {
528 		q->psi_extra->nr_active_q++;
529 
530 		/* For speed optimization, we caculate 1/3 queue size once here */
531 		// XXX limit divided by number of queues divided by 3 ???
532 		pst->one_third_q_size = (fqpie_schk->cfg.limit /
533 			fqpie_schk->cfg.flows_cnt) / 3;
534 
535 		mtx_init(&pst->lock_mtx, "mtx_pie", NULL, MTX_DEF);
536 		callout_init_mtx(&pst->aqm_pie_callout, &pst->lock_mtx,
537 			CALLOUT_RETURNUNLOCKED);
538 	}
539 
540 	return err;
541 }
542 
543 /*
544  * callout function to destroy PIE lock, and free fq_pie flows and fq_pie si
545  * extra memory when number of active sub-queues reaches zero.
546  * 'x' is a fq_pie_flow to be destroyed
547  */
548 static void
549 fqpie_callout_cleanup(void *x)
550 {
551 	struct fq_pie_flow *q = x;
552 	struct pie_status *pst = &q->pst;
553 	struct fq_pie_si_extra *psi_extra;
554 
555 	mtx_unlock(&pst->lock_mtx);
556 	mtx_destroy(&pst->lock_mtx);
557 	psi_extra = q->psi_extra;
558 
559 	DN_BH_WLOCK();
560 	psi_extra->nr_active_q--;
561 
562 	/* when all sub-queues are destroyed, free flows fq_pie extra vars memory */
563 	if (!psi_extra->nr_active_q) {
564 		free(psi_extra->flows, M_DUMMYNET);
565 		free(psi_extra, M_DUMMYNET);
566 		fq_pie_desc.ref_count--;
567 	}
568 	DN_BH_WUNLOCK();
569 }
570 
571 /*
572  * Clean up PIE status for sub-queue 'q'
573  * Stop callout timer and destroy mtx using fqpie_callout_cleanup() callout.
574  */
575 static int
576 pie_cleanup(struct fq_pie_flow *q)
577 {
578 	struct pie_status *pst  = &q->pst;
579 
580 	mtx_lock(&pst->lock_mtx);
581 	callout_reset_sbt(&pst->aqm_pie_callout,
582 		SBT_1US, 0, fqpie_callout_cleanup, q, 0);
583 	mtx_unlock(&pst->lock_mtx);
584 	return 0;
585 }
586 
587 /*
588  * Dequeue and return a pcaket from sub-queue 'q' or NULL if 'q' is empty.
589  * Also, caculate depature time or queue delay using timestamp
590  */
591  static struct mbuf *
592 pie_dequeue(struct fq_pie_flow *q, struct fq_pie_si *si)
593 {
594 	struct mbuf *m;
595 	struct dn_aqm_pie_parms *pprms;
596 	struct pie_status *pst;
597 	aqm_time_t now;
598 	aqm_time_t pkt_ts, dq_time;
599 	int32_t w;
600 
601 	pst  = &q->pst;
602 	pprms = q->pst.parms;
603 
604 	/*we extarct packet ts only when Departure Rate Estimation dis not used*/
605 	m = fq_pie_extract_head(q, &pkt_ts, si,
606 		!(pprms->flags & PIE_DEPRATEEST_ENABLED));
607 
608 	if (!m || !(pst->sflags & PIE_ACTIVE))
609 		return m;
610 
611 	now = AQM_UNOW;
612 	if (pprms->flags & PIE_DEPRATEEST_ENABLED) {
613 		/* calculate average depature time */
614 		if(pst->sflags & PIE_INMEASUREMENT) {
615 			pst->dq_count += m->m_pkthdr.len;
616 
617 			if (pst->dq_count >= PIE_DQ_THRESHOLD) {
618 				dq_time = now - pst->measurement_start;
619 
620 				/*
621 				 * if we don't have old avg dq_time i.e PIE is (re)initialized,
622 				 * don't use weight to calculate new avg_dq_time
623 				 */
624 				if(pst->avg_dq_time == 0)
625 					pst->avg_dq_time = dq_time;
626 				else {
627 					/*
628 					 * weight = PIE_DQ_THRESHOLD/2^6, but we scaled
629 					 * weight by 2^8. Thus, scaled
630 					 * weight = PIE_DQ_THRESHOLD /2^8
631 					 * */
632 					w = PIE_DQ_THRESHOLD >> 8;
633 					pst->avg_dq_time = (dq_time* w
634 						+ (pst->avg_dq_time * ((1L << 8) - w))) >> 8;
635 					pst->sflags &= ~PIE_INMEASUREMENT;
636 				}
637 			}
638 		}
639 
640 		/*
641 		 * Start new measurment cycle when the queue has
642 		 *  PIE_DQ_THRESHOLD worth of bytes.
643 		 */
644 		if(!(pst->sflags & PIE_INMEASUREMENT) &&
645 			q->stats.len_bytes >= PIE_DQ_THRESHOLD) {
646 			pst->sflags |= PIE_INMEASUREMENT;
647 			pst->measurement_start = now;
648 			pst->dq_count = 0;
649 		}
650 	}
651 	/* Optionally, use packet timestamp to estimate queue delay */
652 	else
653 		pst->current_qdelay = now - pkt_ts;
654 
655 	return m;
656 }
657 
658 
659  /*
660  * Enqueue a packet in q, subject to space and FQ-PIE queue management policy
661  * (whose parameters are in q->fs).
662  * Update stats for the queue and the scheduler.
663  * Return 0 on success, 1 on drop. The packet is consumed anyways.
664  */
665 static int
666 pie_enqueue(struct fq_pie_flow *q, struct mbuf* m, struct fq_pie_si *si)
667 {
668 	uint64_t len;
669 	struct pie_status *pst;
670 	struct dn_aqm_pie_parms *pprms;
671 	int t;
672 
673 	len = m->m_pkthdr.len;
674 	pst  = &q->pst;
675 	pprms = pst->parms;
676 	t = ENQUE;
677 
678 	/* drop/mark the packet when PIE is active and burst time elapsed */
679 	if (pst->sflags & PIE_ACTIVE && pst->burst_allowance == 0
680 		&& drop_early(pst, q->stats.len_bytes) == DROP) {
681 			/*
682 			 * if drop_prob over ECN threshold, drop the packet
683 			 * otherwise mark and enqueue it.
684 			 */
685 			if (pprms->flags & PIE_ECN_ENABLED && pst->drop_prob <
686 				(pprms->max_ecnth << (PIE_PROB_BITS - PIE_FIX_POINT_BITS))
687 				&& ecn_mark(m))
688 				t = ENQUE;
689 			else
690 				t = DROP;
691 		}
692 
693 	/* Turn PIE on when 1/3 of the queue is full */
694 	if (!(pst->sflags & PIE_ACTIVE) && q->stats.len_bytes >=
695 		pst->one_third_q_size) {
696 		fq_activate_pie(q);
697 	}
698 
699 	/*  reset burst tolerance and optinally turn PIE off*/
700 	if (pst->drop_prob == 0 && pst->current_qdelay < (pprms->qdelay_ref >> 1)
701 		&& pst->qdelay_old < (pprms->qdelay_ref >> 1)) {
702 
703 			pst->burst_allowance = pprms->max_burst;
704 		if (pprms->flags & PIE_ON_OFF_MODE_ENABLED && q->stats.len_bytes<=0)
705 			fq_deactivate_pie(pst);
706 	}
707 
708 	/* Use timestamp if Departure Rate Estimation mode is disabled */
709 	if (t != DROP && !(pprms->flags & PIE_DEPRATEEST_ENABLED)) {
710 		/* Add TS to mbuf as a TAG */
711 		struct m_tag *mtag;
712 		mtag = m_tag_locate(m, MTAG_ABI_COMPAT, DN_AQM_MTAG_TS, NULL);
713 		if (mtag == NULL)
714 			mtag = m_tag_alloc(MTAG_ABI_COMPAT, DN_AQM_MTAG_TS,
715 				sizeof(aqm_time_t), M_NOWAIT);
716 		if (mtag == NULL) {
717 			m_freem(m);
718 			t = DROP;
719 		}
720 		*(aqm_time_t *)(mtag + 1) = AQM_UNOW;
721 		m_tag_prepend(m, mtag);
722 	}
723 
724 	if (t != DROP) {
725 		mq_append(&q->mq, m);
726 		fq_update_stats(q, si, len, 0);
727 		return 0;
728 	} else {
729 		fq_update_stats(q, si, len, 1);
730 		pst->accu_prob = 0;
731 		FREE_PKT(m);
732 		return 1;
733 	}
734 
735 	return 0;
736 }
737 
738 /* Drop a packet form the head of FQ-PIE sub-queue */
739 static void
740 pie_drop_head(struct fq_pie_flow *q, struct fq_pie_si *si)
741 {
742 	struct mbuf *m = q->mq.head;
743 
744 	if (m == NULL)
745 		return;
746 	q->mq.head = m->m_nextpkt;
747 
748 	fq_update_stats(q, si, -m->m_pkthdr.len, 1);
749 
750 	if (si->main_q.ni.length == 0) /* queue is now idle */
751 			si->main_q.q_time = dn_cfg.curr_time;
752 	/* reset accu_prob after packet drop */
753 	q->pst.accu_prob = 0;
754 
755 	FREE_PKT(m);
756 }
757 
758 /*
759  * Classify a packet to queue number using Jenkins hash function.
760  * Return: queue number
761  * the input of the hash are protocol no, perturbation, src IP, dst IP,
762  * src port, dst port,
763  */
764 static inline int
765 fq_pie_classify_flow(struct mbuf *m, uint16_t fcount, struct fq_pie_si *si)
766 {
767 	struct ip *ip;
768 	struct tcphdr *th;
769 	struct udphdr *uh;
770 	uint8_t tuple[41];
771 	uint16_t hash=0;
772 
773 //#ifdef INET6
774 	struct ip6_hdr *ip6;
775 	int isip6;
776 	isip6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
777 
778 	if(isip6) {
779 		ip6 = mtod(m, struct ip6_hdr *);
780 		*((uint8_t *) &tuple[0]) = ip6->ip6_nxt;
781 		*((uint32_t *) &tuple[1]) = si->perturbation;
782 		memcpy(&tuple[5], ip6->ip6_src.s6_addr, 16);
783 		memcpy(&tuple[21], ip6->ip6_dst.s6_addr, 16);
784 
785 		switch (ip6->ip6_nxt) {
786 		case IPPROTO_TCP:
787 			th = (struct tcphdr *)(ip6 + 1);
788 			*((uint16_t *) &tuple[37]) = th->th_dport;
789 			*((uint16_t *) &tuple[39]) = th->th_sport;
790 			break;
791 
792 		case IPPROTO_UDP:
793 			uh = (struct udphdr *)(ip6 + 1);
794 			*((uint16_t *) &tuple[37]) = uh->uh_dport;
795 			*((uint16_t *) &tuple[39]) = uh->uh_sport;
796 			break;
797 		default:
798 			memset(&tuple[37], 0, 4);
799 		}
800 
801 		hash = jenkins_hash(tuple, 41, HASHINIT) %  fcount;
802 		return hash;
803 	}
804 //#endif
805 
806 	/* IPv4 */
807 	ip = mtod(m, struct ip *);
808 	*((uint8_t *) &tuple[0]) = ip->ip_p;
809 	*((uint32_t *) &tuple[1]) = si->perturbation;
810 	*((uint32_t *) &tuple[5]) = ip->ip_src.s_addr;
811 	*((uint32_t *) &tuple[9]) = ip->ip_dst.s_addr;
812 
813 	switch (ip->ip_p) {
814 		case IPPROTO_TCP:
815 			th = (struct tcphdr *)(ip + 1);
816 			*((uint16_t *) &tuple[13]) = th->th_dport;
817 			*((uint16_t *) &tuple[15]) = th->th_sport;
818 			break;
819 
820 		case IPPROTO_UDP:
821 			uh = (struct udphdr *)(ip + 1);
822 			*((uint16_t *) &tuple[13]) = uh->uh_dport;
823 			*((uint16_t *) &tuple[15]) = uh->uh_sport;
824 			break;
825 		default:
826 			memset(&tuple[13], 0, 4);
827 	}
828 	hash = jenkins_hash(tuple, 17, HASHINIT) % fcount;
829 
830 	return hash;
831 }
832 
833 /*
834  * Enqueue a packet into an appropriate queue according to
835  * FQ-CoDe; algorithm.
836  */
837 static int
838 fq_pie_enqueue(struct dn_sch_inst *_si, struct dn_queue *_q,
839 	struct mbuf *m)
840 {
841 	struct fq_pie_si *si;
842 	struct fq_pie_schk *schk;
843 	struct dn_sch_fq_pie_parms *param;
844 	struct dn_queue *mainq;
845 	struct fq_pie_flow *flows;
846 	int idx, drop, i, maxidx;
847 
848 	mainq = (struct dn_queue *)(_si + 1);
849 	si = (struct fq_pie_si *)_si;
850 	flows = si->si_extra->flows;
851 	schk = (struct fq_pie_schk *)(si->_si.sched+1);
852 	param = &schk->cfg;
853 
854 	 /* classify a packet to queue number*/
855 	idx = fq_pie_classify_flow(m, param->flows_cnt, si);
856 
857 	/* enqueue packet into appropriate queue using PIE AQM.
858 	 * Note: 'pie_enqueue' function returns 1 only when it unable to
859 	 * add timestamp to packet (no limit check)*/
860 	drop = pie_enqueue(&flows[idx], m, si);
861 
862 	/* pie unable to timestamp a packet */
863 	if (drop)
864 		return 1;
865 
866 	/* If the flow (sub-queue) is not active ,then add it to tail of
867 	 * new flows list, initialize and activate it.
868 	 */
869 	if (!flows[idx].active) {
870 		STAILQ_INSERT_TAIL(&si->newflows, &flows[idx], flowchain);
871 		flows[idx].deficit = param->quantum;
872 		fq_activate_pie(&flows[idx]);
873 		flows[idx].active = 1;
874 	}
875 
876 	/* check the limit for all queues and remove a packet from the
877 	 * largest one
878 	 */
879 	if (mainq->ni.length > schk->cfg.limit) {
880 		/* find first active flow */
881 		for (maxidx = 0; maxidx < schk->cfg.flows_cnt; maxidx++)
882 			if (flows[maxidx].active)
883 				break;
884 		if (maxidx < schk->cfg.flows_cnt) {
885 			/* find the largest sub- queue */
886 			for (i = maxidx + 1; i < schk->cfg.flows_cnt; i++)
887 				if (flows[i].active && flows[i].stats.length >
888 					flows[maxidx].stats.length)
889 					maxidx = i;
890 			pie_drop_head(&flows[maxidx], si);
891 			drop = 1;
892 		}
893 	}
894 
895 	return drop;
896 }
897 
898 /*
899  * Dequeue a packet from an appropriate queue according to
900  * FQ-CoDel algorithm.
901  */
902 static struct mbuf *
903 fq_pie_dequeue(struct dn_sch_inst *_si)
904 {
905 	struct fq_pie_si *si;
906 	struct fq_pie_schk *schk;
907 	struct dn_sch_fq_pie_parms *param;
908 	struct fq_pie_flow *f;
909 	struct mbuf *mbuf;
910 	struct fq_pie_list *fq_pie_flowlist;
911 
912 	si = (struct fq_pie_si *)_si;
913 	schk = (struct fq_pie_schk *)(si->_si.sched+1);
914 	param = &schk->cfg;
915 
916 	do {
917 		/* select a list to start with */
918 		if (STAILQ_EMPTY(&si->newflows))
919 			fq_pie_flowlist = &si->oldflows;
920 		else
921 			fq_pie_flowlist = &si->newflows;
922 
923 		/* Both new and old queue lists are empty, return NULL */
924 		if (STAILQ_EMPTY(fq_pie_flowlist))
925 			return NULL;
926 
927 		f = STAILQ_FIRST(fq_pie_flowlist);
928 		while (f != NULL)	{
929 			/* if there is no flow(sub-queue) deficit, increase deficit
930 			 * by quantum, move the flow to the tail of old flows list
931 			 * and try another flow.
932 			 * Otherwise, the flow will be used for dequeue.
933 			 */
934 			if (f->deficit < 0) {
935 				 f->deficit += param->quantum;
936 				 STAILQ_REMOVE_HEAD(fq_pie_flowlist, flowchain);
937 				 STAILQ_INSERT_TAIL(&si->oldflows, f, flowchain);
938 			 } else
939 				 break;
940 
941 			f = STAILQ_FIRST(fq_pie_flowlist);
942 		}
943 
944 		/* the new flows list is empty, try old flows list */
945 		if (STAILQ_EMPTY(fq_pie_flowlist))
946 			continue;
947 
948 		/* Dequeue a packet from the selected flow */
949 		mbuf = pie_dequeue(f, si);
950 
951 		/* pie did not return a packet */
952 		if (!mbuf) {
953 			/* If the selected flow belongs to new flows list, then move
954 			 * it to the tail of old flows list. Otherwise, deactivate it and
955 			 * remove it from the old list and
956 			 */
957 			if (fq_pie_flowlist == &si->newflows) {
958 				STAILQ_REMOVE_HEAD(fq_pie_flowlist, flowchain);
959 				STAILQ_INSERT_TAIL(&si->oldflows, f, flowchain);
960 			}	else {
961 				f->active = 0;
962 				fq_deactivate_pie(&f->pst);
963 				STAILQ_REMOVE_HEAD(fq_pie_flowlist, flowchain);
964 			}
965 			/* start again */
966 			continue;
967 		}
968 
969 		/* we have a packet to return,
970 		 * update flow deficit and return the packet*/
971 		f->deficit -= mbuf->m_pkthdr.len;
972 		return mbuf;
973 
974 	} while (1);
975 
976 	/* unreachable point */
977 	return NULL;
978 }
979 
980 /*
981  * Initialize fq_pie scheduler instance.
982  * also, allocate memory for flows array.
983  */
984 static int
985 fq_pie_new_sched(struct dn_sch_inst *_si)
986 {
987 	struct fq_pie_si *si;
988 	struct dn_queue *q;
989 	struct fq_pie_schk *schk;
990 	struct fq_pie_flow *flows;
991 	int i;
992 
993 	si = (struct fq_pie_si *)_si;
994 	schk = (struct fq_pie_schk *)(_si->sched+1);
995 
996 	if(si->si_extra) {
997 		D("si already configured!");
998 		return 0;
999 	}
1000 
1001 	/* init the main queue */
1002 	q = &si->main_q;
1003 	set_oid(&q->ni.oid, DN_QUEUE, sizeof(*q));
1004 	q->_si = _si;
1005 	q->fs = _si->sched->fs;
1006 
1007 	/* allocate memory for scheduler instance extra vars */
1008 	si->si_extra = malloc(sizeof(struct fq_pie_si_extra),
1009 		 M_DUMMYNET, M_NOWAIT | M_ZERO);
1010 	if (si->si_extra == NULL) {
1011 		D("cannot allocate memory for fq_pie si extra vars");
1012 		return ENOMEM ;
1013 	}
1014 	/* allocate memory for flows array */
1015 	si->si_extra->flows = malloc(schk->cfg.flows_cnt * sizeof(struct fq_pie_flow),
1016 		 M_DUMMYNET, M_NOWAIT | M_ZERO);
1017 	flows = si->si_extra->flows;
1018 	if (flows == NULL) {
1019 		free(si->si_extra, M_DUMMYNET);
1020 		si->si_extra = NULL;
1021 		D("cannot allocate memory for fq_pie flows");
1022 		return ENOMEM ;
1023 	}
1024 
1025 	/* init perturbation for this si */
1026 	si->perturbation = random();
1027 	si->si_extra->nr_active_q = 0;
1028 
1029 	/* init the old and new flows lists */
1030 	STAILQ_INIT(&si->newflows);
1031 	STAILQ_INIT(&si->oldflows);
1032 
1033 	/* init the flows (sub-queues) */
1034 	for (i = 0; i < schk->cfg.flows_cnt; i++) {
1035 		flows[i].pst.parms = &schk->cfg.pcfg;
1036 		flows[i].psi_extra = si->si_extra;
1037 		pie_init(&flows[i], schk);
1038 	}
1039 
1040 	fq_pie_desc.ref_count++;
1041 
1042 	return 0;
1043 }
1044 
1045 
1046 /*
1047  * Free fq_pie scheduler instance.
1048  */
1049 static int
1050 fq_pie_free_sched(struct dn_sch_inst *_si)
1051 {
1052 	struct fq_pie_si *si;
1053 	struct fq_pie_schk *schk;
1054 	struct fq_pie_flow *flows;
1055 	int i;
1056 
1057 	si = (struct fq_pie_si *)_si;
1058 	schk = (struct fq_pie_schk *)(_si->sched+1);
1059 	flows = si->si_extra->flows;
1060 	for (i = 0; i < schk->cfg.flows_cnt; i++) {
1061 		pie_cleanup(&flows[i]);
1062 	}
1063 	si->si_extra = NULL;
1064 	return 0;
1065 }
1066 
1067 /*
1068  * Configure FQ-PIE scheduler.
1069  * the configurations for the scheduler is passed fromipfw  userland.
1070  */
1071 static int
1072 fq_pie_config(struct dn_schk *_schk)
1073 {
1074 	struct fq_pie_schk *schk;
1075 	struct dn_extra_parms *ep;
1076 	struct dn_sch_fq_pie_parms *fqp_cfg;
1077 
1078 	schk = (struct fq_pie_schk *)(_schk+1);
1079 	ep = (struct dn_extra_parms *) _schk->cfg;
1080 
1081 	/* par array contains fq_pie configuration as follow
1082 	 * PIE: 0- qdelay_ref,1- tupdate, 2- max_burst
1083 	 * 3- max_ecnth, 4- alpha, 5- beta, 6- flags
1084 	 * FQ_PIE: 7- quantum, 8- limit, 9- flows
1085 	 */
1086 	if (ep && ep->oid.len ==sizeof(*ep) &&
1087 		ep->oid.subtype == DN_SCH_PARAMS) {
1088 
1089 		fqp_cfg = &schk->cfg;
1090 		if (ep->par[0] < 0)
1091 			fqp_cfg->pcfg.qdelay_ref = fq_pie_sysctl.pcfg.qdelay_ref;
1092 		else
1093 			fqp_cfg->pcfg.qdelay_ref = ep->par[0];
1094 		if (ep->par[1] < 0)
1095 			fqp_cfg->pcfg.tupdate = fq_pie_sysctl.pcfg.tupdate;
1096 		else
1097 			fqp_cfg->pcfg.tupdate = ep->par[1];
1098 		if (ep->par[2] < 0)
1099 			fqp_cfg->pcfg.max_burst = fq_pie_sysctl.pcfg.max_burst;
1100 		else
1101 			fqp_cfg->pcfg.max_burst = ep->par[2];
1102 		if (ep->par[3] < 0)
1103 			fqp_cfg->pcfg.max_ecnth = fq_pie_sysctl.pcfg.max_ecnth;
1104 		else
1105 			fqp_cfg->pcfg.max_ecnth = ep->par[3];
1106 		if (ep->par[4] < 0)
1107 			fqp_cfg->pcfg.alpha = fq_pie_sysctl.pcfg.alpha;
1108 		else
1109 			fqp_cfg->pcfg.alpha = ep->par[4];
1110 		if (ep->par[5] < 0)
1111 			fqp_cfg->pcfg.beta = fq_pie_sysctl.pcfg.beta;
1112 		else
1113 			fqp_cfg->pcfg.beta = ep->par[5];
1114 		if (ep->par[6] < 0)
1115 			fqp_cfg->pcfg.flags = 0;
1116 		else
1117 			fqp_cfg->pcfg.flags = ep->par[6];
1118 
1119 		/* FQ configurations */
1120 		if (ep->par[7] < 0)
1121 			fqp_cfg->quantum = fq_pie_sysctl.quantum;
1122 		else
1123 			fqp_cfg->quantum = ep->par[7];
1124 		if (ep->par[8] < 0)
1125 			fqp_cfg->limit = fq_pie_sysctl.limit;
1126 		else
1127 			fqp_cfg->limit = ep->par[8];
1128 		if (ep->par[9] < 0)
1129 			fqp_cfg->flows_cnt = fq_pie_sysctl.flows_cnt;
1130 		else
1131 			fqp_cfg->flows_cnt = ep->par[9];
1132 
1133 		/* Bound the configurations */
1134 		fqp_cfg->pcfg.qdelay_ref = BOUND_VAR(fqp_cfg->pcfg.qdelay_ref,
1135 			1, 5 * AQM_TIME_1S);
1136 		fqp_cfg->pcfg.tupdate = BOUND_VAR(fqp_cfg->pcfg.tupdate,
1137 			1, 5 * AQM_TIME_1S);
1138 		fqp_cfg->pcfg.max_burst = BOUND_VAR(fqp_cfg->pcfg.max_burst,
1139 			0, 5 * AQM_TIME_1S);
1140 		fqp_cfg->pcfg.max_ecnth = BOUND_VAR(fqp_cfg->pcfg.max_ecnth,
1141 			0, PIE_SCALE);
1142 		fqp_cfg->pcfg.alpha = BOUND_VAR(fqp_cfg->pcfg.alpha, 0, 7 * PIE_SCALE);
1143 		fqp_cfg->pcfg.beta = BOUND_VAR(fqp_cfg->pcfg.beta, 0, 7 * PIE_SCALE);
1144 
1145 		fqp_cfg->quantum = BOUND_VAR(fqp_cfg->quantum,1,9000);
1146 		fqp_cfg->limit= BOUND_VAR(fqp_cfg->limit,1,20480);
1147 		fqp_cfg->flows_cnt= BOUND_VAR(fqp_cfg->flows_cnt,1,65536);
1148 	}
1149 	else {
1150 		D("Wrong parameters for fq_pie scheduler");
1151 		return 1;
1152 	}
1153 
1154 	return 0;
1155 }
1156 
1157 /*
1158  * Return FQ-PIE scheduler configurations
1159  * the configurations for the scheduler is passed to userland.
1160  */
1161 static int
1162 fq_pie_getconfig (struct dn_schk *_schk, struct dn_extra_parms *ep) {
1163 
1164 	struct fq_pie_schk *schk = (struct fq_pie_schk *)(_schk+1);
1165 	struct dn_sch_fq_pie_parms *fqp_cfg;
1166 
1167 	fqp_cfg = &schk->cfg;
1168 
1169 	strcpy(ep->name, fq_pie_desc.name);
1170 	ep->par[0] = fqp_cfg->pcfg.qdelay_ref;
1171 	ep->par[1] = fqp_cfg->pcfg.tupdate;
1172 	ep->par[2] = fqp_cfg->pcfg.max_burst;
1173 	ep->par[3] = fqp_cfg->pcfg.max_ecnth;
1174 	ep->par[4] = fqp_cfg->pcfg.alpha;
1175 	ep->par[5] = fqp_cfg->pcfg.beta;
1176 	ep->par[6] = fqp_cfg->pcfg.flags;
1177 
1178 	ep->par[7] = fqp_cfg->quantum;
1179 	ep->par[8] = fqp_cfg->limit;
1180 	ep->par[9] = fqp_cfg->flows_cnt;
1181 
1182 	return 0;
1183 }
1184 
1185 /*
1186  *  FQ-PIE scheduler descriptor
1187  * contains the type of the scheduler, the name, the size of extra
1188  * data structures, and function pointers.
1189  */
1190 static struct dn_alg fq_pie_desc = {
1191 	_SI( .type = )  DN_SCHED_FQ_PIE,
1192 	_SI( .name = ) "FQ_PIE",
1193 	_SI( .flags = ) 0,
1194 
1195 	_SI( .schk_datalen = ) sizeof(struct fq_pie_schk),
1196 	_SI( .si_datalen = ) sizeof(struct fq_pie_si) - sizeof(struct dn_sch_inst),
1197 	_SI( .q_datalen = ) 0,
1198 
1199 	_SI( .enqueue = ) fq_pie_enqueue,
1200 	_SI( .dequeue = ) fq_pie_dequeue,
1201 	_SI( .config = ) fq_pie_config, /* new sched i.e. sched X config ...*/
1202 	_SI( .destroy = ) NULL,  /*sched x delete */
1203 	_SI( .new_sched = ) fq_pie_new_sched, /* new schd instance */
1204 	_SI( .free_sched = ) fq_pie_free_sched,	/* delete schd instance */
1205 	_SI( .new_fsk = ) NULL,
1206 	_SI( .free_fsk = ) NULL,
1207 	_SI( .new_queue = ) NULL,
1208 	_SI( .free_queue = ) NULL,
1209 	_SI( .getconfig = )  fq_pie_getconfig,
1210 	_SI( .ref_count = ) 0
1211 };
1212 
1213 DECLARE_DNSCHED_MODULE(dn_fq_pie, &fq_pie_desc);
1214