xref: /freebsd/sys/net/altq/altq_fairq.c (revision 2e3507c25e42292b45a5482e116d278f5515d04d)
1 /*
2  * Copyright (c) 2008 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sys/net/altq/altq_fairq.c,v 1.1 2008/04/06 18:58:15 dillon Exp $
35  */
36 /*
37  * Matt: I gutted altq_priq.c and used it as a skeleton on which to build
38  * fairq.  The fairq algorithm is completely different then priq, of course,
39  * but because I used priq's skeleton I believe I should include priq's
40  * copyright.
41  *
42  * Copyright (C) 2000-2003
43  *	Sony Computer Science Laboratories Inc.  All rights reserved.
44  *
45  * Redistribution and use in source and binary forms, with or without
46  * modification, are permitted provided that the following conditions
47  * are met:
48  * 1. Redistributions of source code must retain the above copyright
49  *    notice, this list of conditions and the following disclaimer.
50  * 2. Redistributions in binary form must reproduce the above copyright
51  *    notice, this list of conditions and the following disclaimer in the
52  *    documentation and/or other materials provided with the distribution.
53  *
54  * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  */
66 
67 /*
68  * FAIRQ - take traffic classified by keep state (hashed into
69  * mbuf->m_pkthdr.altq_state_hash) and bucketize it.  Fairly extract
70  * the first packet from each bucket in a round-robin fashion.
71  *
72  * TODO - better overall qlimit support (right now it is per-bucket).
73  *	- NOTE: red etc is per bucket, not overall.
74  *	- better service curve support.
75  *
76  * EXAMPLE:
77  *
78  *  altq on em0 fairq bandwidth 650Kb queue { std, bulk }
79  *  queue std  priority 3 bandwidth 400Kb \
80  *	fairq (buckets 64, default, hogs 1Kb) qlimit 50
81  *  queue bulk priority 2 bandwidth 100Kb \
82  *	fairq (buckets 64, hogs 1Kb) qlimit 50
83  *
84  *  pass out on em0 from any to any keep state queue std
85  *  pass out on em0 inet proto tcp ..... port ... keep state queue bulk
86  */
87 #include "opt_altq.h"
88 #include "opt_inet.h"
89 #include "opt_inet6.h"
90 
91 #ifdef ALTQ_FAIRQ  /* fairq is enabled in the kernel conf */
92 
93 #include <sys/param.h>
94 #include <sys/malloc.h>
95 #include <sys/mbuf.h>
96 #include <sys/socket.h>
97 #include <sys/sockio.h>
98 #include <sys/systm.h>
99 #include <sys/proc.h>
100 #include <sys/errno.h>
101 #include <sys/kernel.h>
102 #include <sys/queue.h>
103 
104 #include <net/if.h>
105 #include <net/if_var.h>
106 #include <net/if_private.h>
107 #include <netinet/in.h>
108 
109 #include <netpfil/pf/pf.h>
110 #include <netpfil/pf/pf_altq.h>
111 #include <netpfil/pf/pf_mtag.h>
112 #include <net/altq/altq.h>
113 #include <net/altq/altq_fairq.h>
114 
115 /*
116  * function prototypes
117  */
118 static int	fairq_clear_interface(struct fairq_if *);
119 static int	fairq_request(struct ifaltq *, int, void *);
120 static void	fairq_purge(struct fairq_if *);
121 static struct fairq_class *fairq_class_create(struct fairq_if *, int, int, u_int, struct fairq_opts *, int);
122 static int	fairq_class_destroy(struct fairq_class *);
123 static int	fairq_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
124 static struct mbuf *fairq_dequeue(struct ifaltq *, int);
125 
126 static int	fairq_addq(struct fairq_class *, struct mbuf *, u_int32_t);
127 static struct mbuf *fairq_getq(struct fairq_class *, uint64_t);
128 static struct mbuf *fairq_pollq(struct fairq_class *, uint64_t, int *);
129 static fairq_bucket_t *fairq_selectq(struct fairq_class *, int);
130 static void	fairq_purgeq(struct fairq_class *);
131 
132 static void	get_class_stats(struct fairq_classstats *, struct fairq_class *);
133 static struct fairq_class *clh_to_clp(struct fairq_if *, uint32_t);
134 
135 int
136 fairq_pfattach(struct pf_altq *a)
137 {
138 	struct ifnet *ifp;
139 	int error;
140 
141 	if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL)
142 		return (EINVAL);
143 
144 	error = altq_attach(&ifp->if_snd, ALTQT_FAIRQ, a->altq_disc,
145 	    fairq_enqueue, fairq_dequeue, fairq_request);
146 
147 	return (error);
148 }
149 
150 int
151 fairq_add_altq(struct ifnet *ifp, struct pf_altq *a)
152 {
153 	struct fairq_if *pif;
154 
155 	if (ifp == NULL)
156 		return (EINVAL);
157 	if (!ALTQ_IS_READY(&ifp->if_snd))
158 		return (ENODEV);
159 
160 	pif = malloc(sizeof(struct fairq_if),
161 			M_DEVBUF, M_WAITOK | M_ZERO);
162 	pif->pif_bandwidth = a->ifbandwidth;
163 	pif->pif_maxpri = -1;
164 	pif->pif_ifq = &ifp->if_snd;
165 
166 	/* keep the state in pf_altq */
167 	a->altq_disc = pif;
168 
169 	return (0);
170 }
171 
172 int
173 fairq_remove_altq(struct pf_altq *a)
174 {
175 	struct fairq_if *pif;
176 
177 	if ((pif = a->altq_disc) == NULL)
178 		return (EINVAL);
179 	a->altq_disc = NULL;
180 
181 	fairq_clear_interface(pif);
182 
183 	free(pif, M_DEVBUF);
184 	return (0);
185 }
186 
187 int
188 fairq_add_queue(struct pf_altq *a)
189 {
190 	struct fairq_if *pif;
191 	struct fairq_class *cl;
192 
193 	if ((pif = a->altq_disc) == NULL)
194 		return (EINVAL);
195 
196 	/* check parameters */
197 	if (a->priority >= FAIRQ_MAXPRI)
198 		return (EINVAL);
199 	if (a->qid == 0)
200 		return (EINVAL);
201 	if (pif->pif_classes[a->priority] != NULL)
202 		return (EBUSY);
203 	if (clh_to_clp(pif, a->qid) != NULL)
204 		return (EBUSY);
205 
206 	cl = fairq_class_create(pif, a->priority, a->qlimit, a->bandwidth,
207 			       &a->pq_u.fairq_opts, a->qid);
208 	if (cl == NULL)
209 		return (ENOMEM);
210 
211 	return (0);
212 }
213 
214 int
215 fairq_remove_queue(struct pf_altq *a)
216 {
217 	struct fairq_if *pif;
218 	struct fairq_class *cl;
219 
220 	if ((pif = a->altq_disc) == NULL)
221 		return (EINVAL);
222 
223 	if ((cl = clh_to_clp(pif, a->qid)) == NULL)
224 		return (EINVAL);
225 
226 	return (fairq_class_destroy(cl));
227 }
228 
229 int
230 fairq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
231 {
232 	struct fairq_if *pif;
233 	struct fairq_class *cl;
234 	struct fairq_classstats stats;
235 	int error = 0;
236 
237 	if ((pif = altq_lookup(a->ifname, ALTQT_FAIRQ)) == NULL)
238 		return (EBADF);
239 
240 	if ((cl = clh_to_clp(pif, a->qid)) == NULL)
241 		return (EINVAL);
242 
243 	if (*nbytes < sizeof(stats))
244 		return (EINVAL);
245 
246 	get_class_stats(&stats, cl);
247 
248 	if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0)
249 		return (error);
250 	*nbytes = sizeof(stats);
251 	return (0);
252 }
253 
254 /*
255  * bring the interface back to the initial state by discarding
256  * all the filters and classes.
257  */
258 static int
259 fairq_clear_interface(struct fairq_if *pif)
260 {
261 	struct fairq_class *cl;
262 	int pri;
263 
264 	/* clear out the classes */
265 	for (pri = 0; pri <= pif->pif_maxpri; pri++) {
266 		if ((cl = pif->pif_classes[pri]) != NULL)
267 			fairq_class_destroy(cl);
268 	}
269 
270 	return (0);
271 }
272 
273 static int
274 fairq_request(struct ifaltq *ifq, int req, void *arg)
275 {
276 	struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc;
277 
278 	IFQ_LOCK_ASSERT(ifq);
279 
280 	switch (req) {
281 	case ALTRQ_PURGE:
282 		fairq_purge(pif);
283 		break;
284 	}
285 	return (0);
286 }
287 
288 /* discard all the queued packets on the interface */
289 static void
290 fairq_purge(struct fairq_if *pif)
291 {
292 	struct fairq_class *cl;
293 	int pri;
294 
295 	for (pri = 0; pri <= pif->pif_maxpri; pri++) {
296 		if ((cl = pif->pif_classes[pri]) != NULL && cl->cl_head)
297 			fairq_purgeq(cl);
298 	}
299 	if (ALTQ_IS_ENABLED(pif->pif_ifq))
300 		pif->pif_ifq->ifq_len = 0;
301 }
302 
303 static struct fairq_class *
304 fairq_class_create(struct fairq_if *pif, int pri, int qlimit,
305 		   u_int bandwidth, struct fairq_opts *opts, int qid)
306 {
307 	struct fairq_class *cl;
308 	int flags = opts->flags;
309 	u_int nbuckets = opts->nbuckets;
310 	int i;
311 
312 #ifndef ALTQ_RED
313 	if (flags & FARF_RED) {
314 #ifdef ALTQ_DEBUG
315 		printf("fairq_class_create: RED not configured for FAIRQ!\n");
316 #endif
317 		return (NULL);
318 	}
319 #endif
320 #ifndef ALTQ_CODEL
321 	if (flags & FARF_CODEL) {
322 #ifdef ALTQ_DEBUG
323 		printf("fairq_class_create: CODEL not configured for FAIRQ!\n");
324 #endif
325 		return (NULL);
326 	}
327 #endif
328 	if (nbuckets == 0)
329 		nbuckets = 256;
330 	if (nbuckets > FAIRQ_MAX_BUCKETS)
331 		nbuckets = FAIRQ_MAX_BUCKETS;
332 	/* enforce power-of-2 size */
333 	while ((nbuckets ^ (nbuckets - 1)) != ((nbuckets << 1) - 1))
334 		++nbuckets;
335 
336 	if ((cl = pif->pif_classes[pri]) != NULL) {
337 		/* modify the class instead of creating a new one */
338 		IFQ_LOCK(cl->cl_pif->pif_ifq);
339 		if (cl->cl_head)
340 			fairq_purgeq(cl);
341 		IFQ_UNLOCK(cl->cl_pif->pif_ifq);
342 #ifdef ALTQ_RIO
343 		if (cl->cl_qtype == Q_RIO)
344 			rio_destroy((rio_t *)cl->cl_red);
345 #endif
346 #ifdef ALTQ_RED
347 		if (cl->cl_qtype == Q_RED)
348 			red_destroy(cl->cl_red);
349 #endif
350 #ifdef ALTQ_CODEL
351 		if (cl->cl_qtype == Q_CODEL)
352 			codel_destroy(cl->cl_codel);
353 #endif
354 	} else {
355 		cl = malloc(sizeof(struct fairq_class),
356 				M_DEVBUF, M_WAITOK | M_ZERO);
357 		cl->cl_nbuckets = nbuckets;
358 		cl->cl_nbucket_mask = nbuckets - 1;
359 
360 		cl->cl_buckets = malloc(
361 			sizeof(struct fairq_bucket) * cl->cl_nbuckets,
362 			M_DEVBUF, M_WAITOK | M_ZERO);
363 		cl->cl_head = NULL;
364 	}
365 
366 	pif->pif_classes[pri] = cl;
367 	if (flags & FARF_DEFAULTCLASS)
368 		pif->pif_default = cl;
369 	if (qlimit == 0)
370 		qlimit = 50;  /* use default */
371 	cl->cl_qlimit = qlimit;
372 	for (i = 0; i < cl->cl_nbuckets; ++i) {
373 		qlimit(&cl->cl_buckets[i].queue) = qlimit;
374 	}
375 	cl->cl_bandwidth = bandwidth / 8;
376 	cl->cl_qtype = Q_DROPTAIL;
377 	cl->cl_flags = flags & FARF_USERFLAGS;
378 	cl->cl_pri = pri;
379 	if (pri > pif->pif_maxpri)
380 		pif->pif_maxpri = pri;
381 	cl->cl_pif = pif;
382 	cl->cl_handle = qid;
383 	cl->cl_hogs_m1 = opts->hogs_m1 / 8;
384 	cl->cl_lssc_m1 = opts->lssc_m1 / 8;	/* NOT YET USED */
385 
386 #ifdef ALTQ_RED
387 	if (flags & (FARF_RED|FARF_RIO)) {
388 		int red_flags, red_pkttime;
389 
390 		red_flags = 0;
391 		if (flags & FARF_ECN)
392 			red_flags |= REDF_ECN;
393 #ifdef ALTQ_RIO
394 		if (flags & FARF_CLEARDSCP)
395 			red_flags |= RIOF_CLEARDSCP;
396 #endif
397 		if (pif->pif_bandwidth < 8)
398 			red_pkttime = 1000 * 1000 * 1000; /* 1 sec */
399 		else
400 			red_pkttime = (int64_t)pif->pif_ifq->altq_ifp->if_mtu
401 			  * 1000 * 1000 * 1000 / (pif->pif_bandwidth / 8);
402 #ifdef ALTQ_RIO
403 		if (flags & FARF_RIO) {
404 			cl->cl_red = (red_t *)rio_alloc(0, NULL,
405 						red_flags, red_pkttime);
406 			if (cl->cl_red != NULL)
407 				cl->cl_qtype = Q_RIO;
408 		} else
409 #endif
410 		if (flags & FARF_RED) {
411 			cl->cl_red = red_alloc(0, 0,
412 			    cl->cl_qlimit * 10/100,
413 			    cl->cl_qlimit * 30/100,
414 			    red_flags, red_pkttime);
415 			if (cl->cl_red != NULL)
416 				cl->cl_qtype = Q_RED;
417 		}
418 	}
419 #endif /* ALTQ_RED */
420 #ifdef ALTQ_CODEL
421 	if (flags & FARF_CODEL) {
422 		cl->cl_codel = codel_alloc(5, 100, 0);
423 		if (cl->cl_codel != NULL)
424 			cl->cl_qtype = Q_CODEL;
425 	}
426 #endif
427 
428 	return (cl);
429 }
430 
431 static int
432 fairq_class_destroy(struct fairq_class *cl)
433 {
434 	struct fairq_if *pif;
435 	int pri;
436 
437 	IFQ_LOCK(cl->cl_pif->pif_ifq);
438 
439 	if (cl->cl_head)
440 		fairq_purgeq(cl);
441 
442 	pif = cl->cl_pif;
443 	pif->pif_classes[cl->cl_pri] = NULL;
444 	if (pif->pif_poll_cache == cl)
445 		pif->pif_poll_cache = NULL;
446 	if (pif->pif_maxpri == cl->cl_pri) {
447 		for (pri = cl->cl_pri; pri >= 0; pri--)
448 			if (pif->pif_classes[pri] != NULL) {
449 				pif->pif_maxpri = pri;
450 				break;
451 			}
452 		if (pri < 0)
453 			pif->pif_maxpri = -1;
454 	}
455 	IFQ_UNLOCK(cl->cl_pif->pif_ifq);
456 
457 	if (cl->cl_red != NULL) {
458 #ifdef ALTQ_RIO
459 		if (cl->cl_qtype == Q_RIO)
460 			rio_destroy((rio_t *)cl->cl_red);
461 #endif
462 #ifdef ALTQ_RED
463 		if (cl->cl_qtype == Q_RED)
464 			red_destroy(cl->cl_red);
465 #endif
466 #ifdef ALTQ_CODEL
467 		if (cl->cl_qtype == Q_CODEL)
468 			codel_destroy(cl->cl_codel);
469 #endif
470 	}
471 	free(cl->cl_buckets, M_DEVBUF);
472 	free(cl, M_DEVBUF);
473 
474 	return (0);
475 }
476 
477 /*
478  * fairq_enqueue is an enqueue function to be registered to
479  * (*altq_enqueue) in struct ifaltq.
480  */
481 static int
482 fairq_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr)
483 {
484 	struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc;
485 	struct fairq_class *cl = NULL; /* Make compiler happy */
486 	struct pf_mtag *t;
487 	u_int32_t qid_hash = 0;
488 	int len;
489 
490 	IFQ_LOCK_ASSERT(ifq);
491 
492 	/* grab class set by classifier */
493 	if ((m->m_flags & M_PKTHDR) == 0) {
494 		/* should not happen */
495 		printf("altq: packet for %s does not have pkthdr\n",
496 			ifq->altq_ifp->if_xname);
497 		m_freem(m);
498 		return (ENOBUFS);
499 	}
500 
501 	if ((t = pf_find_mtag(m)) != NULL) {
502 		cl = clh_to_clp(pif, t->qid);
503 		qid_hash = t->qid_hash;
504 	}
505 	if (cl == NULL) {
506 		cl = pif->pif_default;
507 		if (cl == NULL) {
508 			m_freem(m);
509 			return (ENOBUFS);
510 		}
511 	}
512 	cl->cl_flags |= FARF_HAS_PACKETS;
513 	cl->cl_pktattr = NULL;
514 	len = m_pktlen(m);
515 	if (fairq_addq(cl, m, qid_hash) != 0) {
516 		/* drop occurred.  mbuf was freed in fairq_addq. */
517 		PKTCNTR_ADD(&cl->cl_dropcnt, len);
518 		return (ENOBUFS);
519 	}
520 	IFQ_INC_LEN(ifq);
521 
522 	return (0);
523 }
524 
525 /*
526  * fairq_dequeue is a dequeue function to be registered to
527  * (*altq_dequeue) in struct ifaltq.
528  *
529  * note: ALTDQ_POLL returns the next packet without removing the packet
530  *	from the queue.  ALTDQ_REMOVE is a normal dequeue operation.
531  *	ALTDQ_REMOVE must return the same packet if called immediately
532  *	after ALTDQ_POLL.
533  */
534 static struct mbuf *
535 fairq_dequeue(struct ifaltq *ifq, int op)
536 {
537 	struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc;
538 	struct fairq_class *cl;
539 	struct fairq_class *best_cl;
540 	struct mbuf *best_m;
541 	struct mbuf *m = NULL;
542 	uint64_t cur_time = read_machclk();
543 	int pri;
544 	int hit_limit;
545 
546 	IFQ_LOCK_ASSERT(ifq);
547 
548 	if (IFQ_IS_EMPTY(ifq)) {
549 		return (NULL);
550 	}
551 
552 	if (pif->pif_poll_cache && op == ALTDQ_REMOVE) {
553 		best_cl = pif->pif_poll_cache;
554 		m = fairq_getq(best_cl, cur_time);
555 		pif->pif_poll_cache = NULL;
556 		if (m) {
557 			IFQ_DEC_LEN(ifq);
558 			PKTCNTR_ADD(&best_cl->cl_xmitcnt, m_pktlen(m));
559 			return (m);
560 		}
561 	} else {
562 		best_cl = NULL;
563 		best_m = NULL;
564 
565 		for (pri = pif->pif_maxpri;  pri >= 0; pri--) {
566 			if ((cl = pif->pif_classes[pri]) == NULL)
567 				continue;
568 			if ((cl->cl_flags & FARF_HAS_PACKETS) == 0)
569 				continue;
570 			m = fairq_pollq(cl, cur_time, &hit_limit);
571 			if (m == NULL) {
572 				cl->cl_flags &= ~FARF_HAS_PACKETS;
573 				continue;
574 			}
575 
576 			/*
577 			 * Only override the best choice if we are under
578 			 * the BW limit.
579 			 */
580 			if (hit_limit == 0 || best_cl == NULL) {
581 				best_cl = cl;
582 				best_m = m;
583 			}
584 
585 			/*
586 			 * Remember the highest priority mbuf in case we
587 			 * do not find any lower priority mbufs.
588 			 */
589 			if (hit_limit)
590 				continue;
591 			break;
592 		}
593 		if (op == ALTDQ_POLL) {
594 			pif->pif_poll_cache = best_cl;
595 			m = best_m;
596 		} else if (best_cl) {
597 			m = fairq_getq(best_cl, cur_time);
598 			if (m != NULL) {
599 				IFQ_DEC_LEN(ifq);
600 				PKTCNTR_ADD(&best_cl->cl_xmitcnt, m_pktlen(m));
601 			}
602 		}
603 		return (m);
604 	}
605 	return (NULL);
606 }
607 
608 static int
609 fairq_addq(struct fairq_class *cl, struct mbuf *m, u_int32_t bucketid)
610 {
611 	fairq_bucket_t *b;
612 	u_int hindex;
613 	uint64_t bw;
614 
615 	/*
616 	 * If the packet doesn't have any keep state put it on the end of
617 	 * our queue.  XXX this can result in out of order delivery.
618 	 */
619 	if (bucketid == 0) {
620 		if (cl->cl_head)
621 			b = cl->cl_head->prev;
622 		else
623 			b = &cl->cl_buckets[0];
624 	} else {
625 		hindex = bucketid & cl->cl_nbucket_mask;
626 		b = &cl->cl_buckets[hindex];
627 	}
628 
629 	/*
630 	 * Add the bucket to the end of the circular list of active buckets.
631 	 *
632 	 * As a special case we add the bucket to the beginning of the list
633 	 * instead of the end if it was not previously on the list and if
634 	 * its traffic is less then the hog level.
635 	 */
636 	if (b->in_use == 0) {
637 		b->in_use = 1;
638 		if (cl->cl_head == NULL) {
639 			cl->cl_head = b;
640 			b->next = b;
641 			b->prev = b;
642 		} else {
643 			b->next = cl->cl_head;
644 			b->prev = cl->cl_head->prev;
645 			b->prev->next = b;
646 			b->next->prev = b;
647 
648 			if (b->bw_delta && cl->cl_hogs_m1) {
649 				bw = b->bw_bytes * machclk_freq / b->bw_delta;
650 				if (bw < cl->cl_hogs_m1)
651 					cl->cl_head = b;
652 			}
653 		}
654 	}
655 
656 #ifdef ALTQ_RIO
657 	if (cl->cl_qtype == Q_RIO)
658 		return rio_addq((rio_t *)cl->cl_red, &b->queue, m, cl->cl_pktattr);
659 #endif
660 #ifdef ALTQ_RED
661 	if (cl->cl_qtype == Q_RED)
662 		return red_addq(cl->cl_red, &b->queue, m, cl->cl_pktattr);
663 #endif
664 #ifdef ALTQ_CODEL
665 	if (cl->cl_qtype == Q_CODEL)
666 		return codel_addq(cl->cl_codel, &b->queue, m);
667 #endif
668 	if (qlen(&b->queue) >= qlimit(&b->queue)) {
669 		m_freem(m);
670 		return (-1);
671 	}
672 
673 	if (cl->cl_flags & FARF_CLEARDSCP)
674 		write_dsfield(m, cl->cl_pktattr, 0);
675 
676 	_addq(&b->queue, m);
677 
678 	return (0);
679 }
680 
681 static struct mbuf *
682 fairq_getq(struct fairq_class *cl, uint64_t cur_time)
683 {
684 	fairq_bucket_t *b;
685 	struct mbuf *m;
686 
687 	b = fairq_selectq(cl, 0);
688 	if (b == NULL)
689 		m = NULL;
690 #ifdef ALTQ_RIO
691 	else if (cl->cl_qtype == Q_RIO)
692 		m = rio_getq((rio_t *)cl->cl_red, &b->queue);
693 #endif
694 #ifdef ALTQ_RED
695 	else if (cl->cl_qtype == Q_RED)
696 		m = red_getq(cl->cl_red, &b->queue);
697 #endif
698 #ifdef ALTQ_CODEL
699 	else if (cl->cl_qtype == Q_CODEL)
700 		m = codel_getq(cl->cl_codel, &b->queue);
701 #endif
702 	else
703 		m = _getq(&b->queue);
704 
705 	/*
706 	 * Calculate the BW change
707 	 */
708 	if (m != NULL) {
709 		uint64_t delta;
710 
711 		/*
712 		 * Per-class bandwidth calculation
713 		 */
714 		delta = (cur_time - cl->cl_last_time);
715 		if (delta > machclk_freq * 8)
716 			delta = machclk_freq * 8;
717 		cl->cl_bw_delta += delta;
718 		cl->cl_bw_bytes += m->m_pkthdr.len;
719 		cl->cl_last_time = cur_time;
720 		cl->cl_bw_delta -= cl->cl_bw_delta >> 3;
721 		cl->cl_bw_bytes -= cl->cl_bw_bytes >> 3;
722 
723 		/*
724 		 * Per-bucket bandwidth calculation
725 		 */
726 		delta = (cur_time - b->last_time);
727 		if (delta > machclk_freq * 8)
728 			delta = machclk_freq * 8;
729 		b->bw_delta += delta;
730 		b->bw_bytes += m->m_pkthdr.len;
731 		b->last_time = cur_time;
732 		b->bw_delta -= b->bw_delta >> 3;
733 		b->bw_bytes -= b->bw_bytes >> 3;
734 	}
735 	return(m);
736 }
737 
738 /*
739  * Figure out what the next packet would be if there were no limits.  If
740  * this class hits its bandwidth limit *hit_limit is set to no-zero, otherwise
741  * it is set to 0.  A non-NULL mbuf is returned either way.
742  */
743 static struct mbuf *
744 fairq_pollq(struct fairq_class *cl, uint64_t cur_time, int *hit_limit)
745 {
746 	fairq_bucket_t *b;
747 	struct mbuf *m;
748 	uint64_t delta;
749 	uint64_t bw;
750 
751 	*hit_limit = 0;
752 	b = fairq_selectq(cl, 1);
753 	if (b == NULL)
754 		return(NULL);
755 	m = qhead(&b->queue);
756 
757 	/*
758 	 * Did this packet exceed the class bandwidth?  Calculate the
759 	 * bandwidth component of the packet.
760 	 *
761 	 * - Calculate bytes per second
762 	 */
763 	delta = cur_time - cl->cl_last_time;
764 	if (delta > machclk_freq * 8)
765 		delta = machclk_freq * 8;
766 	cl->cl_bw_delta += delta;
767 	cl->cl_last_time = cur_time;
768 	if (cl->cl_bw_delta) {
769 		bw = cl->cl_bw_bytes * machclk_freq / cl->cl_bw_delta;
770 
771 		if (bw > cl->cl_bandwidth)
772 			*hit_limit = 1;
773 #ifdef ALTQ_DEBUG
774 		printf("BW %6ju relative to %6u %d queue %p\n",
775 			(uintmax_t)bw, cl->cl_bandwidth, *hit_limit, b);
776 #endif
777 	}
778 	return(m);
779 }
780 
781 /*
782  * Locate the next queue we want to pull a packet out of.  This code
783  * is also responsible for removing empty buckets from the circular list.
784  */
785 static
786 fairq_bucket_t *
787 fairq_selectq(struct fairq_class *cl, int ispoll)
788 {
789 	fairq_bucket_t *b;
790 	uint64_t bw;
791 
792 	if (ispoll == 0 && cl->cl_polled) {
793 		b = cl->cl_polled;
794 		cl->cl_polled = NULL;
795 		return(b);
796 	}
797 
798 	while ((b = cl->cl_head) != NULL) {
799 		/*
800 		 * Remove empty queues from consideration
801 		 */
802 		if (qempty(&b->queue)) {
803 			b->in_use = 0;
804 			cl->cl_head = b->next;
805 			if (cl->cl_head == b) {
806 				cl->cl_head = NULL;
807 			} else {
808 				b->next->prev = b->prev;
809 				b->prev->next = b->next;
810 			}
811 			continue;
812 		}
813 
814 		/*
815 		 * Advance the round robin.  Queues with bandwidths less
816 		 * then the hog bandwidth are allowed to burst.
817 		 */
818 		if (cl->cl_hogs_m1 == 0) {
819 			cl->cl_head = b->next;
820 		} else if (b->bw_delta) {
821 			bw = b->bw_bytes * machclk_freq / b->bw_delta;
822 			if (bw >= cl->cl_hogs_m1) {
823 				cl->cl_head = b->next;
824 			}
825 			/*
826 			 * XXX TODO -
827 			 */
828 		}
829 
830 		/*
831 		 * Return bucket b.
832 		 */
833 		break;
834 	}
835 	if (ispoll)
836 		cl->cl_polled = b;
837 	return(b);
838 }
839 
840 static void
841 fairq_purgeq(struct fairq_class *cl)
842 {
843 	fairq_bucket_t *b;
844 	struct mbuf *m;
845 
846 	while ((b = fairq_selectq(cl, 0)) != NULL) {
847 		while ((m = _getq(&b->queue)) != NULL) {
848 			PKTCNTR_ADD(&cl->cl_dropcnt, m_pktlen(m));
849 			m_freem(m);
850 		}
851 		ASSERT(qlen(&b->queue) == 0);
852 	}
853 }
854 
855 static void
856 get_class_stats(struct fairq_classstats *sp, struct fairq_class *cl)
857 {
858 	fairq_bucket_t *b;
859 
860 	sp->class_handle = cl->cl_handle;
861 	sp->qlimit = cl->cl_qlimit;
862 	sp->xmit_cnt = cl->cl_xmitcnt;
863 	sp->drop_cnt = cl->cl_dropcnt;
864 	sp->qtype = cl->cl_qtype;
865 	sp->qlength = 0;
866 
867 	if (cl->cl_head) {
868 		b = cl->cl_head;
869 		do {
870 			sp->qlength += qlen(&b->queue);
871 			b = b->next;
872 		} while (b != cl->cl_head);
873 	}
874 
875 #ifdef ALTQ_RED
876 	if (cl->cl_qtype == Q_RED)
877 		red_getstats(cl->cl_red, &sp->red[0]);
878 #endif
879 #ifdef ALTQ_RIO
880 	if (cl->cl_qtype == Q_RIO)
881 		rio_getstats((rio_t *)cl->cl_red, &sp->red[0]);
882 #endif
883 #ifdef ALTQ_CODEL
884 	if (cl->cl_qtype == Q_CODEL)
885 		codel_getstats(cl->cl_codel, &sp->codel);
886 #endif
887 }
888 
889 /* convert a class handle to the corresponding class pointer */
890 static struct fairq_class *
891 clh_to_clp(struct fairq_if *pif, uint32_t chandle)
892 {
893 	struct fairq_class *cl;
894 	int idx;
895 
896 	if (chandle == 0)
897 		return (NULL);
898 
899 	for (idx = pif->pif_maxpri; idx >= 0; idx--)
900 		if ((cl = pif->pif_classes[idx]) != NULL &&
901 		    cl->cl_handle == chandle)
902 			return (cl);
903 
904 	return (NULL);
905 }
906 
907 #endif /* ALTQ_FAIRQ */
908