xref: /freebsd/sys/net/altq/altq_hfsc.c (revision 39ee7a7a6bdd1557b1c3532abf60d139798ac88b)
1 /*-
2  * Copyright (c) 1997-1999 Carnegie Mellon University. All Rights Reserved.
3  *
4  * Permission to use, copy, modify, and distribute this software and
5  * its documentation is hereby granted (including for commercial or
6  * for-profit use), provided that both the copyright notice and this
7  * permission notice appear in all copies of the software, derivative
8  * works, or modified versions, and any portions thereof.
9  *
10  * THIS SOFTWARE IS EXPERIMENTAL AND IS KNOWN TO HAVE BUGS, SOME OF
11  * WHICH MAY HAVE SERIOUS CONSEQUENCES.  CARNEGIE MELLON PROVIDES THIS
12  * SOFTWARE IN ITS ``AS IS'' CONDITION, AND ANY EXPRESS OR IMPLIED
13  * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
14  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
15  * DISCLAIMED.  IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY BE LIABLE
16  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
17  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
18  * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
19  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
20  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
21  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
22  * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
23  * DAMAGE.
24  *
25  * Carnegie Mellon encourages (but does not require) users of this
26  * software to return any improvements or extensions that they make,
27  * and to grant Carnegie Mellon the rights to redistribute these
28  * changes without encumbrance.
29  *
30  * $KAME: altq_hfsc.c,v 1.24 2003/12/05 05:40:46 kjc Exp $
31  * $FreeBSD$
32  */
33 /*
34  * H-FSC is described in Proceedings of SIGCOMM'97,
35  * "A Hierarchical Fair Service Curve Algorithm for Link-Sharing,
36  * Real-Time and Priority Service"
37  * by Ion Stoica, Hui Zhang, and T. S. Eugene Ng.
38  *
39  * Oleg Cherevko <olwi@aq.ml.com.ua> added the upperlimit for link-sharing.
40  * when a class has an upperlimit, the fit-time is computed from the
41  * upperlimit service curve.  the link-sharing scheduler does not schedule
42  * a class whose fit-time exceeds the current time.
43  */
44 
45 #include "opt_altq.h"
46 #include "opt_inet.h"
47 #include "opt_inet6.h"
48 
49 #ifdef ALTQ_HFSC  /* hfsc is enabled by ALTQ_HFSC option in opt_altq.h */
50 
51 #include <sys/param.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/socket.h>
55 #include <sys/systm.h>
56 #include <sys/errno.h>
57 #include <sys/queue.h>
58 #if 1 /* ALTQ3_COMPAT */
59 #include <sys/sockio.h>
60 #include <sys/proc.h>
61 #include <sys/kernel.h>
62 #endif /* ALTQ3_COMPAT */
63 
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <netinet/in.h>
67 
68 #include <netpfil/pf/pf.h>
69 #include <netpfil/pf/pf_altq.h>
70 #include <netpfil/pf/pf_mtag.h>
71 #include <net/altq/altq.h>
72 #include <net/altq/altq_hfsc.h>
73 #ifdef ALTQ3_COMPAT
74 #include <net/altq/altq_conf.h>
75 #endif
76 
77 /*
78  * function prototypes
79  */
80 static int			 hfsc_clear_interface(struct hfsc_if *);
81 static int			 hfsc_request(struct ifaltq *, int, void *);
82 static void			 hfsc_purge(struct hfsc_if *);
83 static struct hfsc_class	*hfsc_class_create(struct hfsc_if *,
84     struct service_curve *, struct service_curve *, struct service_curve *,
85     struct hfsc_class *, int, int, int);
86 static int			 hfsc_class_destroy(struct hfsc_class *);
87 static struct hfsc_class	*hfsc_nextclass(struct hfsc_class *);
88 static int			 hfsc_enqueue(struct ifaltq *, struct mbuf *,
89 				    struct altq_pktattr *);
90 static struct mbuf		*hfsc_dequeue(struct ifaltq *, int);
91 
92 static int		 hfsc_addq(struct hfsc_class *, struct mbuf *);
93 static struct mbuf	*hfsc_getq(struct hfsc_class *);
94 static struct mbuf	*hfsc_pollq(struct hfsc_class *);
95 static void		 hfsc_purgeq(struct hfsc_class *);
96 
97 static void		 update_cfmin(struct hfsc_class *);
98 static void		 set_active(struct hfsc_class *, int);
99 static void		 set_passive(struct hfsc_class *);
100 
101 static void		 init_ed(struct hfsc_class *, int);
102 static void		 update_ed(struct hfsc_class *, int);
103 static void		 update_d(struct hfsc_class *, int);
104 static void		 init_vf(struct hfsc_class *, int);
105 static void		 update_vf(struct hfsc_class *, int, u_int64_t);
106 static void		 ellist_insert(struct hfsc_class *);
107 static void		 ellist_remove(struct hfsc_class *);
108 static void		 ellist_update(struct hfsc_class *);
109 struct hfsc_class	*hfsc_get_mindl(struct hfsc_if *, u_int64_t);
110 static void		 actlist_insert(struct hfsc_class *);
111 static void		 actlist_remove(struct hfsc_class *);
112 static void		 actlist_update(struct hfsc_class *);
113 
114 static struct hfsc_class	*actlist_firstfit(struct hfsc_class *,
115 				    u_int64_t);
116 
117 static __inline u_int64_t	seg_x2y(u_int64_t, u_int64_t);
118 static __inline u_int64_t	seg_y2x(u_int64_t, u_int64_t);
119 static __inline u_int64_t	m2sm(u_int);
120 static __inline u_int64_t	m2ism(u_int);
121 static __inline u_int64_t	d2dx(u_int);
122 static u_int			sm2m(u_int64_t);
123 static u_int			dx2d(u_int64_t);
124 
125 static void		sc2isc(struct service_curve *, struct internal_sc *);
126 static void		rtsc_init(struct runtime_sc *, struct internal_sc *,
127 			    u_int64_t, u_int64_t);
128 static u_int64_t	rtsc_y2x(struct runtime_sc *, u_int64_t);
129 static u_int64_t	rtsc_x2y(struct runtime_sc *, u_int64_t);
130 static void		rtsc_min(struct runtime_sc *, struct internal_sc *,
131 			    u_int64_t, u_int64_t);
132 
133 static void			 get_class_stats(struct hfsc_classstats *,
134 				    struct hfsc_class *);
135 static struct hfsc_class	*clh_to_clp(struct hfsc_if *, u_int32_t);
136 
137 
138 #ifdef ALTQ3_COMPAT
139 static struct hfsc_if *hfsc_attach(struct ifaltq *, u_int);
140 static int hfsc_detach(struct hfsc_if *);
141 static int hfsc_class_modify(struct hfsc_class *, struct service_curve *,
142     struct service_curve *, struct service_curve *);
143 
144 static int hfsccmd_if_attach(struct hfsc_attach *);
145 static int hfsccmd_if_detach(struct hfsc_interface *);
146 static int hfsccmd_add_class(struct hfsc_add_class *);
147 static int hfsccmd_delete_class(struct hfsc_delete_class *);
148 static int hfsccmd_modify_class(struct hfsc_modify_class *);
149 static int hfsccmd_add_filter(struct hfsc_add_filter *);
150 static int hfsccmd_delete_filter(struct hfsc_delete_filter *);
151 static int hfsccmd_class_stats(struct hfsc_class_stats *);
152 
153 altqdev_decl(hfsc);
154 #endif /* ALTQ3_COMPAT */
155 
156 /*
157  * macros
158  */
159 #define	is_a_parent_class(cl)	((cl)->cl_children != NULL)
160 
161 #define	HT_INFINITY	0xffffffffffffffffLL	/* infinite time value */
162 
163 #ifdef ALTQ3_COMPAT
164 /* hif_list keeps all hfsc_if's allocated. */
165 static struct hfsc_if *hif_list = NULL;
166 #endif /* ALTQ3_COMPAT */
167 
168 int
169 hfsc_pfattach(struct pf_altq *a)
170 {
171 	struct ifnet *ifp;
172 	int s, error;
173 
174 	if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL)
175 		return (EINVAL);
176 	s = splnet();
177 	error = altq_attach(&ifp->if_snd, ALTQT_HFSC, a->altq_disc,
178 	    hfsc_enqueue, hfsc_dequeue, hfsc_request, NULL, NULL);
179 	splx(s);
180 	return (error);
181 }
182 
183 int
184 hfsc_add_altq(struct pf_altq *a)
185 {
186 	struct hfsc_if *hif;
187 	struct ifnet *ifp;
188 
189 	if ((ifp = ifunit(a->ifname)) == NULL)
190 		return (EINVAL);
191 	if (!ALTQ_IS_READY(&ifp->if_snd))
192 		return (ENODEV);
193 
194 	hif = malloc(sizeof(struct hfsc_if), M_DEVBUF, M_NOWAIT | M_ZERO);
195 	if (hif == NULL)
196 		return (ENOMEM);
197 
198 	TAILQ_INIT(&hif->hif_eligible);
199 	hif->hif_ifq = &ifp->if_snd;
200 
201 	/* keep the state in pf_altq */
202 	a->altq_disc = hif;
203 
204 	return (0);
205 }
206 
207 int
208 hfsc_remove_altq(struct pf_altq *a)
209 {
210 	struct hfsc_if *hif;
211 
212 	if ((hif = a->altq_disc) == NULL)
213 		return (EINVAL);
214 	a->altq_disc = NULL;
215 
216 	(void)hfsc_clear_interface(hif);
217 	(void)hfsc_class_destroy(hif->hif_rootclass);
218 
219 	free(hif, M_DEVBUF);
220 
221 	return (0);
222 }
223 
224 int
225 hfsc_add_queue(struct pf_altq *a)
226 {
227 	struct hfsc_if *hif;
228 	struct hfsc_class *cl, *parent;
229 	struct hfsc_opts *opts;
230 	struct service_curve rtsc, lssc, ulsc;
231 
232 	if ((hif = a->altq_disc) == NULL)
233 		return (EINVAL);
234 
235 	opts = &a->pq_u.hfsc_opts;
236 
237 	if (a->parent_qid == HFSC_NULLCLASS_HANDLE &&
238 	    hif->hif_rootclass == NULL)
239 		parent = NULL;
240 	else if ((parent = clh_to_clp(hif, a->parent_qid)) == NULL)
241 		return (EINVAL);
242 
243 	if (a->qid == 0)
244 		return (EINVAL);
245 
246 	if (clh_to_clp(hif, a->qid) != NULL)
247 		return (EBUSY);
248 
249 	rtsc.m1 = opts->rtsc_m1;
250 	rtsc.d  = opts->rtsc_d;
251 	rtsc.m2 = opts->rtsc_m2;
252 	lssc.m1 = opts->lssc_m1;
253 	lssc.d  = opts->lssc_d;
254 	lssc.m2 = opts->lssc_m2;
255 	ulsc.m1 = opts->ulsc_m1;
256 	ulsc.d  = opts->ulsc_d;
257 	ulsc.m2 = opts->ulsc_m2;
258 
259 	cl = hfsc_class_create(hif, &rtsc, &lssc, &ulsc,
260 	    parent, a->qlimit, opts->flags, a->qid);
261 	if (cl == NULL)
262 		return (ENOMEM);
263 
264 	return (0);
265 }
266 
267 int
268 hfsc_remove_queue(struct pf_altq *a)
269 {
270 	struct hfsc_if *hif;
271 	struct hfsc_class *cl;
272 
273 	if ((hif = a->altq_disc) == NULL)
274 		return (EINVAL);
275 
276 	if ((cl = clh_to_clp(hif, a->qid)) == NULL)
277 		return (EINVAL);
278 
279 	return (hfsc_class_destroy(cl));
280 }
281 
282 int
283 hfsc_getqstats(struct pf_altq *a, void *ubuf, int *nbytes)
284 {
285 	struct hfsc_if *hif;
286 	struct hfsc_class *cl;
287 	struct hfsc_classstats stats;
288 	int error = 0;
289 
290 	if ((hif = altq_lookup(a->ifname, ALTQT_HFSC)) == NULL)
291 		return (EBADF);
292 
293 	if ((cl = clh_to_clp(hif, a->qid)) == NULL)
294 		return (EINVAL);
295 
296 	if (*nbytes < sizeof(stats))
297 		return (EINVAL);
298 
299 	get_class_stats(&stats, cl);
300 
301 	if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0)
302 		return (error);
303 	*nbytes = sizeof(stats);
304 	return (0);
305 }
306 
307 /*
308  * bring the interface back to the initial state by discarding
309  * all the filters and classes except the root class.
310  */
311 static int
312 hfsc_clear_interface(struct hfsc_if *hif)
313 {
314 	struct hfsc_class	*cl;
315 
316 #ifdef ALTQ3_COMPAT
317 	/* free the filters for this interface */
318 	acc_discard_filters(&hif->hif_classifier, NULL, 1);
319 #endif
320 
321 	/* clear out the classes */
322 	while (hif->hif_rootclass != NULL &&
323 	    (cl = hif->hif_rootclass->cl_children) != NULL) {
324 		/*
325 		 * remove the first leaf class found in the hierarchy
326 		 * then start over
327 		 */
328 		for (; cl != NULL; cl = hfsc_nextclass(cl)) {
329 			if (!is_a_parent_class(cl)) {
330 				(void)hfsc_class_destroy(cl);
331 				break;
332 			}
333 		}
334 	}
335 
336 	return (0);
337 }
338 
339 static int
340 hfsc_request(struct ifaltq *ifq, int req, void *arg)
341 {
342 	struct hfsc_if	*hif = (struct hfsc_if *)ifq->altq_disc;
343 
344 	IFQ_LOCK_ASSERT(ifq);
345 
346 	switch (req) {
347 	case ALTRQ_PURGE:
348 		hfsc_purge(hif);
349 		break;
350 	}
351 	return (0);
352 }
353 
354 /* discard all the queued packets on the interface */
355 static void
356 hfsc_purge(struct hfsc_if *hif)
357 {
358 	struct hfsc_class *cl;
359 
360 	for (cl = hif->hif_rootclass; cl != NULL; cl = hfsc_nextclass(cl))
361 		if (!qempty(cl->cl_q))
362 			hfsc_purgeq(cl);
363 	if (ALTQ_IS_ENABLED(hif->hif_ifq))
364 		hif->hif_ifq->ifq_len = 0;
365 }
366 
367 struct hfsc_class *
368 hfsc_class_create(struct hfsc_if *hif, struct service_curve *rsc,
369     struct service_curve *fsc, struct service_curve *usc,
370     struct hfsc_class *parent, int qlimit, int flags, int qid)
371 {
372 	struct hfsc_class *cl, *p;
373 	int i, s;
374 
375 	if (hif->hif_classes >= HFSC_MAX_CLASSES)
376 		return (NULL);
377 
378 #ifndef ALTQ_RED
379 	if (flags & HFCF_RED) {
380 #ifdef ALTQ_DEBUG
381 		printf("hfsc_class_create: RED not configured for HFSC!\n");
382 #endif
383 		return (NULL);
384 	}
385 #endif
386 #ifndef ALTQ_CODEL
387 	if (flags & HFCF_CODEL) {
388 #ifdef ALTQ_DEBUG
389 		printf("hfsc_class_create: CODEL not configured for HFSC!\n");
390 #endif
391 		return (NULL);
392 	}
393 #endif
394 
395 	cl = malloc(sizeof(struct hfsc_class), M_DEVBUF, M_NOWAIT | M_ZERO);
396 	if (cl == NULL)
397 		return (NULL);
398 
399 	cl->cl_q = malloc(sizeof(class_queue_t), M_DEVBUF, M_NOWAIT | M_ZERO);
400 	if (cl->cl_q == NULL)
401 		goto err_ret;
402 
403 	TAILQ_INIT(&cl->cl_actc);
404 
405 	if (qlimit == 0)
406 		qlimit = 50;  /* use default */
407 	qlimit(cl->cl_q) = qlimit;
408 	qtype(cl->cl_q) = Q_DROPTAIL;
409 	qlen(cl->cl_q) = 0;
410 	qsize(cl->cl_q) = 0;
411 	cl->cl_flags = flags;
412 #ifdef ALTQ_RED
413 	if (flags & (HFCF_RED|HFCF_RIO)) {
414 		int red_flags, red_pkttime;
415 		u_int m2;
416 
417 		m2 = 0;
418 		if (rsc != NULL && rsc->m2 > m2)
419 			m2 = rsc->m2;
420 		if (fsc != NULL && fsc->m2 > m2)
421 			m2 = fsc->m2;
422 		if (usc != NULL && usc->m2 > m2)
423 			m2 = usc->m2;
424 
425 		red_flags = 0;
426 		if (flags & HFCF_ECN)
427 			red_flags |= REDF_ECN;
428 #ifdef ALTQ_RIO
429 		if (flags & HFCF_CLEARDSCP)
430 			red_flags |= RIOF_CLEARDSCP;
431 #endif
432 		if (m2 < 8)
433 			red_pkttime = 1000 * 1000 * 1000; /* 1 sec */
434 		else
435 			red_pkttime = (int64_t)hif->hif_ifq->altq_ifp->if_mtu
436 				* 1000 * 1000 * 1000 / (m2 / 8);
437 		if (flags & HFCF_RED) {
438 			cl->cl_red = red_alloc(0, 0,
439 			    qlimit(cl->cl_q) * 10/100,
440 			    qlimit(cl->cl_q) * 30/100,
441 			    red_flags, red_pkttime);
442 			if (cl->cl_red != NULL)
443 				qtype(cl->cl_q) = Q_RED;
444 		}
445 #ifdef ALTQ_RIO
446 		else {
447 			cl->cl_red = (red_t *)rio_alloc(0, NULL,
448 			    red_flags, red_pkttime);
449 			if (cl->cl_red != NULL)
450 				qtype(cl->cl_q) = Q_RIO;
451 		}
452 #endif
453 	}
454 #endif /* ALTQ_RED */
455 #ifdef ALTQ_CODEL
456 	if (flags & HFCF_CODEL) {
457 		cl->cl_codel = codel_alloc(5, 100, 0);
458 		if (cl->cl_codel != NULL)
459 			qtype(cl->cl_q) = Q_CODEL;
460 	}
461 #endif
462 
463 	if (rsc != NULL && (rsc->m1 != 0 || rsc->m2 != 0)) {
464 		cl->cl_rsc = malloc(sizeof(struct internal_sc),
465 		    M_DEVBUF, M_NOWAIT);
466 		if (cl->cl_rsc == NULL)
467 			goto err_ret;
468 		sc2isc(rsc, cl->cl_rsc);
469 		rtsc_init(&cl->cl_deadline, cl->cl_rsc, 0, 0);
470 		rtsc_init(&cl->cl_eligible, cl->cl_rsc, 0, 0);
471 	}
472 	if (fsc != NULL && (fsc->m1 != 0 || fsc->m2 != 0)) {
473 		cl->cl_fsc = malloc(sizeof(struct internal_sc),
474 		    M_DEVBUF, M_NOWAIT);
475 		if (cl->cl_fsc == NULL)
476 			goto err_ret;
477 		sc2isc(fsc, cl->cl_fsc);
478 		rtsc_init(&cl->cl_virtual, cl->cl_fsc, 0, 0);
479 	}
480 	if (usc != NULL && (usc->m1 != 0 || usc->m2 != 0)) {
481 		cl->cl_usc = malloc(sizeof(struct internal_sc),
482 		    M_DEVBUF, M_NOWAIT);
483 		if (cl->cl_usc == NULL)
484 			goto err_ret;
485 		sc2isc(usc, cl->cl_usc);
486 		rtsc_init(&cl->cl_ulimit, cl->cl_usc, 0, 0);
487 	}
488 
489 	cl->cl_id = hif->hif_classid++;
490 	cl->cl_handle = qid;
491 	cl->cl_hif = hif;
492 	cl->cl_parent = parent;
493 
494 	s = splnet();
495 	IFQ_LOCK(hif->hif_ifq);
496 	hif->hif_classes++;
497 
498 	/*
499 	 * find a free slot in the class table.  if the slot matching
500 	 * the lower bits of qid is free, use this slot.  otherwise,
501 	 * use the first free slot.
502 	 */
503 	i = qid % HFSC_MAX_CLASSES;
504 	if (hif->hif_class_tbl[i] == NULL)
505 		hif->hif_class_tbl[i] = cl;
506 	else {
507 		for (i = 0; i < HFSC_MAX_CLASSES; i++)
508 			if (hif->hif_class_tbl[i] == NULL) {
509 				hif->hif_class_tbl[i] = cl;
510 				break;
511 			}
512 		if (i == HFSC_MAX_CLASSES) {
513 			IFQ_UNLOCK(hif->hif_ifq);
514 			splx(s);
515 			goto err_ret;
516 		}
517 	}
518 
519 	if (flags & HFCF_DEFAULTCLASS)
520 		hif->hif_defaultclass = cl;
521 
522 	if (parent == NULL) {
523 		/* this is root class */
524 		hif->hif_rootclass = cl;
525 	} else {
526 		/* add this class to the children list of the parent */
527 		if ((p = parent->cl_children) == NULL)
528 			parent->cl_children = cl;
529 		else {
530 			while (p->cl_siblings != NULL)
531 				p = p->cl_siblings;
532 			p->cl_siblings = cl;
533 		}
534 	}
535 	IFQ_UNLOCK(hif->hif_ifq);
536 	splx(s);
537 
538 	return (cl);
539 
540  err_ret:
541 	if (cl->cl_red != NULL) {
542 #ifdef ALTQ_RIO
543 		if (q_is_rio(cl->cl_q))
544 			rio_destroy((rio_t *)cl->cl_red);
545 #endif
546 #ifdef ALTQ_RED
547 		if (q_is_red(cl->cl_q))
548 			red_destroy(cl->cl_red);
549 #endif
550 #ifdef ALTQ_CODEL
551 		if (q_is_codel(cl->cl_q))
552 			codel_destroy(cl->cl_codel);
553 #endif
554 	}
555 	if (cl->cl_fsc != NULL)
556 		free(cl->cl_fsc, M_DEVBUF);
557 	if (cl->cl_rsc != NULL)
558 		free(cl->cl_rsc, M_DEVBUF);
559 	if (cl->cl_usc != NULL)
560 		free(cl->cl_usc, M_DEVBUF);
561 	if (cl->cl_q != NULL)
562 		free(cl->cl_q, M_DEVBUF);
563 	free(cl, M_DEVBUF);
564 	return (NULL);
565 }
566 
567 static int
568 hfsc_class_destroy(struct hfsc_class *cl)
569 {
570 	int i, s;
571 
572 	if (cl == NULL)
573 		return (0);
574 
575 	if (is_a_parent_class(cl))
576 		return (EBUSY);
577 
578 	s = splnet();
579 	IFQ_LOCK(cl->cl_hif->hif_ifq);
580 
581 #ifdef ALTQ3_COMPAT
582 	/* delete filters referencing to this class */
583 	acc_discard_filters(&cl->cl_hif->hif_classifier, cl, 0);
584 #endif /* ALTQ3_COMPAT */
585 
586 	if (!qempty(cl->cl_q))
587 		hfsc_purgeq(cl);
588 
589 	if (cl->cl_parent == NULL) {
590 		/* this is root class */
591 	} else {
592 		struct hfsc_class *p = cl->cl_parent->cl_children;
593 
594 		if (p == cl)
595 			cl->cl_parent->cl_children = cl->cl_siblings;
596 		else do {
597 			if (p->cl_siblings == cl) {
598 				p->cl_siblings = cl->cl_siblings;
599 				break;
600 			}
601 		} while ((p = p->cl_siblings) != NULL);
602 		ASSERT(p != NULL);
603 	}
604 
605 	for (i = 0; i < HFSC_MAX_CLASSES; i++)
606 		if (cl->cl_hif->hif_class_tbl[i] == cl) {
607 			cl->cl_hif->hif_class_tbl[i] = NULL;
608 			break;
609 		}
610 
611 	cl->cl_hif->hif_classes--;
612 	IFQ_UNLOCK(cl->cl_hif->hif_ifq);
613 	splx(s);
614 
615 	if (cl->cl_red != NULL) {
616 #ifdef ALTQ_RIO
617 		if (q_is_rio(cl->cl_q))
618 			rio_destroy((rio_t *)cl->cl_red);
619 #endif
620 #ifdef ALTQ_RED
621 		if (q_is_red(cl->cl_q))
622 			red_destroy(cl->cl_red);
623 #endif
624 #ifdef ALTQ_CODEL
625 		if (q_is_codel(cl->cl_q))
626 			codel_destroy(cl->cl_codel);
627 #endif
628 	}
629 
630 	IFQ_LOCK(cl->cl_hif->hif_ifq);
631 	if (cl == cl->cl_hif->hif_rootclass)
632 		cl->cl_hif->hif_rootclass = NULL;
633 	if (cl == cl->cl_hif->hif_defaultclass)
634 		cl->cl_hif->hif_defaultclass = NULL;
635 	IFQ_UNLOCK(cl->cl_hif->hif_ifq);
636 
637 	if (cl->cl_usc != NULL)
638 		free(cl->cl_usc, M_DEVBUF);
639 	if (cl->cl_fsc != NULL)
640 		free(cl->cl_fsc, M_DEVBUF);
641 	if (cl->cl_rsc != NULL)
642 		free(cl->cl_rsc, M_DEVBUF);
643 	free(cl->cl_q, M_DEVBUF);
644 	free(cl, M_DEVBUF);
645 
646 	return (0);
647 }
648 
649 /*
650  * hfsc_nextclass returns the next class in the tree.
651  *   usage:
652  *	for (cl = hif->hif_rootclass; cl != NULL; cl = hfsc_nextclass(cl))
653  *		do_something;
654  */
655 static struct hfsc_class *
656 hfsc_nextclass(struct hfsc_class *cl)
657 {
658 	if (cl->cl_children != NULL)
659 		cl = cl->cl_children;
660 	else if (cl->cl_siblings != NULL)
661 		cl = cl->cl_siblings;
662 	else {
663 		while ((cl = cl->cl_parent) != NULL)
664 			if (cl->cl_siblings) {
665 				cl = cl->cl_siblings;
666 				break;
667 			}
668 	}
669 
670 	return (cl);
671 }
672 
673 /*
674  * hfsc_enqueue is an enqueue function to be registered to
675  * (*altq_enqueue) in struct ifaltq.
676  */
677 static int
678 hfsc_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr)
679 {
680 	struct hfsc_if	*hif = (struct hfsc_if *)ifq->altq_disc;
681 	struct hfsc_class *cl;
682 	struct pf_mtag *t;
683 	int len;
684 
685 	IFQ_LOCK_ASSERT(ifq);
686 
687 	/* grab class set by classifier */
688 	if ((m->m_flags & M_PKTHDR) == 0) {
689 		/* should not happen */
690 		printf("altq: packet for %s does not have pkthdr\n",
691 		    ifq->altq_ifp->if_xname);
692 		m_freem(m);
693 		return (ENOBUFS);
694 	}
695 	cl = NULL;
696 	if ((t = pf_find_mtag(m)) != NULL)
697 		cl = clh_to_clp(hif, t->qid);
698 #ifdef ALTQ3_COMPAT
699 	else if ((ifq->altq_flags & ALTQF_CLASSIFY) && pktattr != NULL)
700 		cl = pktattr->pattr_class;
701 #endif
702 	if (cl == NULL || is_a_parent_class(cl)) {
703 		cl = hif->hif_defaultclass;
704 		if (cl == NULL) {
705 			m_freem(m);
706 			return (ENOBUFS);
707 		}
708 	}
709 #ifdef ALTQ3_COMPAT
710 	if (pktattr != NULL)
711 		cl->cl_pktattr = pktattr;  /* save proto hdr used by ECN */
712 	else
713 #endif
714 		cl->cl_pktattr = NULL;
715 	len = m_pktlen(m);
716 	if (hfsc_addq(cl, m) != 0) {
717 		/* drop occurred.  mbuf was freed in hfsc_addq. */
718 		PKTCNTR_ADD(&cl->cl_stats.drop_cnt, len);
719 		return (ENOBUFS);
720 	}
721 	IFQ_INC_LEN(ifq);
722 	cl->cl_hif->hif_packets++;
723 
724 	/* successfully queued. */
725 	if (qlen(cl->cl_q) == 1)
726 		set_active(cl, m_pktlen(m));
727 
728 	return (0);
729 }
730 
731 /*
732  * hfsc_dequeue is a dequeue function to be registered to
733  * (*altq_dequeue) in struct ifaltq.
734  *
735  * note: ALTDQ_POLL returns the next packet without removing the packet
736  *	from the queue.  ALTDQ_REMOVE is a normal dequeue operation.
737  *	ALTDQ_REMOVE must return the same packet if called immediately
738  *	after ALTDQ_POLL.
739  */
740 static struct mbuf *
741 hfsc_dequeue(struct ifaltq *ifq, int op)
742 {
743 	struct hfsc_if	*hif = (struct hfsc_if *)ifq->altq_disc;
744 	struct hfsc_class *cl;
745 	struct mbuf *m;
746 	int len, next_len;
747 	int realtime = 0;
748 	u_int64_t cur_time;
749 
750 	IFQ_LOCK_ASSERT(ifq);
751 
752 	if (hif->hif_packets == 0)
753 		/* no packet in the tree */
754 		return (NULL);
755 
756 	cur_time = read_machclk();
757 
758 	if (op == ALTDQ_REMOVE && hif->hif_pollcache != NULL) {
759 
760 		cl = hif->hif_pollcache;
761 		hif->hif_pollcache = NULL;
762 		/* check if the class was scheduled by real-time criteria */
763 		if (cl->cl_rsc != NULL)
764 			realtime = (cl->cl_e <= cur_time);
765 	} else {
766 		/*
767 		 * if there are eligible classes, use real-time criteria.
768 		 * find the class with the minimum deadline among
769 		 * the eligible classes.
770 		 */
771 		if ((cl = hfsc_get_mindl(hif, cur_time))
772 		    != NULL) {
773 			realtime = 1;
774 		} else {
775 #ifdef ALTQ_DEBUG
776 			int fits = 0;
777 #endif
778 			/*
779 			 * use link-sharing criteria
780 			 * get the class with the minimum vt in the hierarchy
781 			 */
782 			cl = hif->hif_rootclass;
783 			while (is_a_parent_class(cl)) {
784 
785 				cl = actlist_firstfit(cl, cur_time);
786 				if (cl == NULL) {
787 #ifdef ALTQ_DEBUG
788 					if (fits > 0)
789 						printf("%d fit but none found\n",fits);
790 #endif
791 					return (NULL);
792 				}
793 				/*
794 				 * update parent's cl_cvtmin.
795 				 * don't update if the new vt is smaller.
796 				 */
797 				if (cl->cl_parent->cl_cvtmin < cl->cl_vt)
798 					cl->cl_parent->cl_cvtmin = cl->cl_vt;
799 #ifdef ALTQ_DEBUG
800 				fits++;
801 #endif
802 			}
803 		}
804 
805 		if (op == ALTDQ_POLL) {
806 			hif->hif_pollcache = cl;
807 			m = hfsc_pollq(cl);
808 			return (m);
809 		}
810 	}
811 
812 	m = hfsc_getq(cl);
813 	if (m == NULL)
814 		panic("hfsc_dequeue:");
815 	len = m_pktlen(m);
816 	cl->cl_hif->hif_packets--;
817 	IFQ_DEC_LEN(ifq);
818 	PKTCNTR_ADD(&cl->cl_stats.xmit_cnt, len);
819 
820 	update_vf(cl, len, cur_time);
821 	if (realtime)
822 		cl->cl_cumul += len;
823 
824 	if (!qempty(cl->cl_q)) {
825 		if (cl->cl_rsc != NULL) {
826 			/* update ed */
827 			next_len = m_pktlen(qhead(cl->cl_q));
828 
829 			if (realtime)
830 				update_ed(cl, next_len);
831 			else
832 				update_d(cl, next_len);
833 		}
834 	} else {
835 		/* the class becomes passive */
836 		set_passive(cl);
837 	}
838 
839 	return (m);
840 }
841 
842 static int
843 hfsc_addq(struct hfsc_class *cl, struct mbuf *m)
844 {
845 
846 #ifdef ALTQ_RIO
847 	if (q_is_rio(cl->cl_q))
848 		return rio_addq((rio_t *)cl->cl_red, cl->cl_q,
849 				m, cl->cl_pktattr);
850 #endif
851 #ifdef ALTQ_RED
852 	if (q_is_red(cl->cl_q))
853 		return red_addq(cl->cl_red, cl->cl_q, m, cl->cl_pktattr);
854 #endif
855 #ifdef ALTQ_CODEL
856 	if (q_is_codel(cl->cl_q))
857 		return codel_addq(cl->cl_codel, cl->cl_q, m);
858 #endif
859 	if (qlen(cl->cl_q) >= qlimit(cl->cl_q)) {
860 		m_freem(m);
861 		return (-1);
862 	}
863 
864 	if (cl->cl_flags & HFCF_CLEARDSCP)
865 		write_dsfield(m, cl->cl_pktattr, 0);
866 
867 	_addq(cl->cl_q, m);
868 
869 	return (0);
870 }
871 
872 static struct mbuf *
873 hfsc_getq(struct hfsc_class *cl)
874 {
875 #ifdef ALTQ_RIO
876 	if (q_is_rio(cl->cl_q))
877 		return rio_getq((rio_t *)cl->cl_red, cl->cl_q);
878 #endif
879 #ifdef ALTQ_RED
880 	if (q_is_red(cl->cl_q))
881 		return red_getq(cl->cl_red, cl->cl_q);
882 #endif
883 #ifdef ALTQ_CODEL
884 	if (q_is_codel(cl->cl_q))
885 		return codel_getq(cl->cl_codel, cl->cl_q);
886 #endif
887 	return _getq(cl->cl_q);
888 }
889 
890 static struct mbuf *
891 hfsc_pollq(struct hfsc_class *cl)
892 {
893 	return qhead(cl->cl_q);
894 }
895 
896 static void
897 hfsc_purgeq(struct hfsc_class *cl)
898 {
899 	struct mbuf *m;
900 
901 	if (qempty(cl->cl_q))
902 		return;
903 
904 	while ((m = _getq(cl->cl_q)) != NULL) {
905 		PKTCNTR_ADD(&cl->cl_stats.drop_cnt, m_pktlen(m));
906 		m_freem(m);
907 		cl->cl_hif->hif_packets--;
908 		IFQ_DEC_LEN(cl->cl_hif->hif_ifq);
909 	}
910 	ASSERT(qlen(cl->cl_q) == 0);
911 
912 	update_vf(cl, 0, 0);	/* remove cl from the actlist */
913 	set_passive(cl);
914 }
915 
916 static void
917 set_active(struct hfsc_class *cl, int len)
918 {
919 	if (cl->cl_rsc != NULL)
920 		init_ed(cl, len);
921 	if (cl->cl_fsc != NULL)
922 		init_vf(cl, len);
923 
924 	cl->cl_stats.period++;
925 }
926 
927 static void
928 set_passive(struct hfsc_class *cl)
929 {
930 	if (cl->cl_rsc != NULL)
931 		ellist_remove(cl);
932 
933 	/*
934 	 * actlist is now handled in update_vf() so that update_vf(cl, 0, 0)
935 	 * needs to be called explicitly to remove a class from actlist
936 	 */
937 }
938 
939 static void
940 init_ed(struct hfsc_class *cl, int next_len)
941 {
942 	u_int64_t cur_time;
943 
944 	cur_time = read_machclk();
945 
946 	/* update the deadline curve */
947 	rtsc_min(&cl->cl_deadline, cl->cl_rsc, cur_time, cl->cl_cumul);
948 
949 	/*
950 	 * update the eligible curve.
951 	 * for concave, it is equal to the deadline curve.
952 	 * for convex, it is a linear curve with slope m2.
953 	 */
954 	cl->cl_eligible = cl->cl_deadline;
955 	if (cl->cl_rsc->sm1 <= cl->cl_rsc->sm2) {
956 		cl->cl_eligible.dx = 0;
957 		cl->cl_eligible.dy = 0;
958 	}
959 
960 	/* compute e and d */
961 	cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
962 	cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
963 
964 	ellist_insert(cl);
965 }
966 
967 static void
968 update_ed(struct hfsc_class *cl, int next_len)
969 {
970 	cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
971 	cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
972 
973 	ellist_update(cl);
974 }
975 
976 static void
977 update_d(struct hfsc_class *cl, int next_len)
978 {
979 	cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
980 }
981 
982 static void
983 init_vf(struct hfsc_class *cl, int len)
984 {
985 	struct hfsc_class *max_cl, *p;
986 	u_int64_t vt, f, cur_time;
987 	int go_active;
988 
989 	cur_time = 0;
990 	go_active = 1;
991 	for ( ; cl->cl_parent != NULL; cl = cl->cl_parent) {
992 
993 		if (go_active && cl->cl_nactive++ == 0)
994 			go_active = 1;
995 		else
996 			go_active = 0;
997 
998 		if (go_active) {
999 			max_cl = TAILQ_LAST(&cl->cl_parent->cl_actc, acthead);
1000 			if (max_cl != NULL) {
1001 				/*
1002 				 * set vt to the average of the min and max
1003 				 * classes.  if the parent's period didn't
1004 				 * change, don't decrease vt of the class.
1005 				 */
1006 				vt = max_cl->cl_vt;
1007 				if (cl->cl_parent->cl_cvtmin != 0)
1008 					vt = (cl->cl_parent->cl_cvtmin + vt)/2;
1009 
1010 				if (cl->cl_parent->cl_vtperiod !=
1011 				    cl->cl_parentperiod || vt > cl->cl_vt)
1012 					cl->cl_vt = vt;
1013 			} else {
1014 				/*
1015 				 * first child for a new parent backlog period.
1016 				 * add parent's cvtmax to vtoff of children
1017 				 * to make a new vt (vtoff + vt) larger than
1018 				 * the vt in the last period for all children.
1019 				 */
1020 				vt = cl->cl_parent->cl_cvtmax;
1021 				for (p = cl->cl_parent->cl_children; p != NULL;
1022 				     p = p->cl_siblings)
1023 					p->cl_vtoff += vt;
1024 				cl->cl_vt = 0;
1025 				cl->cl_parent->cl_cvtmax = 0;
1026 				cl->cl_parent->cl_cvtmin = 0;
1027 			}
1028 			cl->cl_initvt = cl->cl_vt;
1029 
1030 			/* update the virtual curve */
1031 			vt = cl->cl_vt + cl->cl_vtoff;
1032 			rtsc_min(&cl->cl_virtual, cl->cl_fsc, vt, cl->cl_total);
1033 			if (cl->cl_virtual.x == vt) {
1034 				cl->cl_virtual.x -= cl->cl_vtoff;
1035 				cl->cl_vtoff = 0;
1036 			}
1037 			cl->cl_vtadj = 0;
1038 
1039 			cl->cl_vtperiod++;  /* increment vt period */
1040 			cl->cl_parentperiod = cl->cl_parent->cl_vtperiod;
1041 			if (cl->cl_parent->cl_nactive == 0)
1042 				cl->cl_parentperiod++;
1043 			cl->cl_f = 0;
1044 
1045 			actlist_insert(cl);
1046 
1047 			if (cl->cl_usc != NULL) {
1048 				/* class has upper limit curve */
1049 				if (cur_time == 0)
1050 					cur_time = read_machclk();
1051 
1052 				/* update the ulimit curve */
1053 				rtsc_min(&cl->cl_ulimit, cl->cl_usc, cur_time,
1054 				    cl->cl_total);
1055 				/* compute myf */
1056 				cl->cl_myf = rtsc_y2x(&cl->cl_ulimit,
1057 				    cl->cl_total);
1058 				cl->cl_myfadj = 0;
1059 			}
1060 		}
1061 
1062 		if (cl->cl_myf > cl->cl_cfmin)
1063 			f = cl->cl_myf;
1064 		else
1065 			f = cl->cl_cfmin;
1066 		if (f != cl->cl_f) {
1067 			cl->cl_f = f;
1068 			update_cfmin(cl->cl_parent);
1069 		}
1070 	}
1071 }
1072 
1073 static void
1074 update_vf(struct hfsc_class *cl, int len, u_int64_t cur_time)
1075 {
1076 	u_int64_t f, myf_bound, delta;
1077 	int go_passive;
1078 
1079 	go_passive = qempty(cl->cl_q);
1080 
1081 	for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
1082 
1083 		cl->cl_total += len;
1084 
1085 		if (cl->cl_fsc == NULL || cl->cl_nactive == 0)
1086 			continue;
1087 
1088 		if (go_passive && --cl->cl_nactive == 0)
1089 			go_passive = 1;
1090 		else
1091 			go_passive = 0;
1092 
1093 		if (go_passive) {
1094 			/* no more active child, going passive */
1095 
1096 			/* update cvtmax of the parent class */
1097 			if (cl->cl_vt > cl->cl_parent->cl_cvtmax)
1098 				cl->cl_parent->cl_cvtmax = cl->cl_vt;
1099 
1100 			/* remove this class from the vt list */
1101 			actlist_remove(cl);
1102 
1103 			update_cfmin(cl->cl_parent);
1104 
1105 			continue;
1106 		}
1107 
1108 		/*
1109 		 * update vt and f
1110 		 */
1111 		cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total)
1112 		    - cl->cl_vtoff + cl->cl_vtadj;
1113 
1114 		/*
1115 		 * if vt of the class is smaller than cvtmin,
1116 		 * the class was skipped in the past due to non-fit.
1117 		 * if so, we need to adjust vtadj.
1118 		 */
1119 		if (cl->cl_vt < cl->cl_parent->cl_cvtmin) {
1120 			cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt;
1121 			cl->cl_vt = cl->cl_parent->cl_cvtmin;
1122 		}
1123 
1124 		/* update the vt list */
1125 		actlist_update(cl);
1126 
1127 		if (cl->cl_usc != NULL) {
1128 			cl->cl_myf = cl->cl_myfadj
1129 			    + rtsc_y2x(&cl->cl_ulimit, cl->cl_total);
1130 
1131 			/*
1132 			 * if myf lags behind by more than one clock tick
1133 			 * from the current time, adjust myfadj to prevent
1134 			 * a rate-limited class from going greedy.
1135 			 * in a steady state under rate-limiting, myf
1136 			 * fluctuates within one clock tick.
1137 			 */
1138 			myf_bound = cur_time - machclk_per_tick;
1139 			if (cl->cl_myf < myf_bound) {
1140 				delta = cur_time - cl->cl_myf;
1141 				cl->cl_myfadj += delta;
1142 				cl->cl_myf += delta;
1143 			}
1144 		}
1145 
1146 		/* cl_f is max(cl_myf, cl_cfmin) */
1147 		if (cl->cl_myf > cl->cl_cfmin)
1148 			f = cl->cl_myf;
1149 		else
1150 			f = cl->cl_cfmin;
1151 		if (f != cl->cl_f) {
1152 			cl->cl_f = f;
1153 			update_cfmin(cl->cl_parent);
1154 		}
1155 	}
1156 }
1157 
1158 static void
1159 update_cfmin(struct hfsc_class *cl)
1160 {
1161 	struct hfsc_class *p;
1162 	u_int64_t cfmin;
1163 
1164 	if (TAILQ_EMPTY(&cl->cl_actc)) {
1165 		cl->cl_cfmin = 0;
1166 		return;
1167 	}
1168 	cfmin = HT_INFINITY;
1169 	TAILQ_FOREACH(p, &cl->cl_actc, cl_actlist) {
1170 		if (p->cl_f == 0) {
1171 			cl->cl_cfmin = 0;
1172 			return;
1173 		}
1174 		if (p->cl_f < cfmin)
1175 			cfmin = p->cl_f;
1176 	}
1177 	cl->cl_cfmin = cfmin;
1178 }
1179 
1180 /*
1181  * TAILQ based ellist and actlist implementation
1182  * (ion wanted to make a calendar queue based implementation)
1183  */
1184 /*
1185  * eligible list holds backlogged classes being sorted by their eligible times.
1186  * there is one eligible list per interface.
1187  */
1188 
1189 static void
1190 ellist_insert(struct hfsc_class *cl)
1191 {
1192 	struct hfsc_if	*hif = cl->cl_hif;
1193 	struct hfsc_class *p;
1194 
1195 	/* check the last entry first */
1196 	if ((p = TAILQ_LAST(&hif->hif_eligible, elighead)) == NULL ||
1197 	    p->cl_e <= cl->cl_e) {
1198 		TAILQ_INSERT_TAIL(&hif->hif_eligible, cl, cl_ellist);
1199 		return;
1200 	}
1201 
1202 	TAILQ_FOREACH(p, &hif->hif_eligible, cl_ellist) {
1203 		if (cl->cl_e < p->cl_e) {
1204 			TAILQ_INSERT_BEFORE(p, cl, cl_ellist);
1205 			return;
1206 		}
1207 	}
1208 	ASSERT(0); /* should not reach here */
1209 }
1210 
1211 static void
1212 ellist_remove(struct hfsc_class *cl)
1213 {
1214 	struct hfsc_if	*hif = cl->cl_hif;
1215 
1216 	TAILQ_REMOVE(&hif->hif_eligible, cl, cl_ellist);
1217 }
1218 
1219 static void
1220 ellist_update(struct hfsc_class *cl)
1221 {
1222 	struct hfsc_if	*hif = cl->cl_hif;
1223 	struct hfsc_class *p, *last;
1224 
1225 	/*
1226 	 * the eligible time of a class increases monotonically.
1227 	 * if the next entry has a larger eligible time, nothing to do.
1228 	 */
1229 	p = TAILQ_NEXT(cl, cl_ellist);
1230 	if (p == NULL || cl->cl_e <= p->cl_e)
1231 		return;
1232 
1233 	/* check the last entry */
1234 	last = TAILQ_LAST(&hif->hif_eligible, elighead);
1235 	ASSERT(last != NULL);
1236 	if (last->cl_e <= cl->cl_e) {
1237 		TAILQ_REMOVE(&hif->hif_eligible, cl, cl_ellist);
1238 		TAILQ_INSERT_TAIL(&hif->hif_eligible, cl, cl_ellist);
1239 		return;
1240 	}
1241 
1242 	/*
1243 	 * the new position must be between the next entry
1244 	 * and the last entry
1245 	 */
1246 	while ((p = TAILQ_NEXT(p, cl_ellist)) != NULL) {
1247 		if (cl->cl_e < p->cl_e) {
1248 			TAILQ_REMOVE(&hif->hif_eligible, cl, cl_ellist);
1249 			TAILQ_INSERT_BEFORE(p, cl, cl_ellist);
1250 			return;
1251 		}
1252 	}
1253 	ASSERT(0); /* should not reach here */
1254 }
1255 
1256 /* find the class with the minimum deadline among the eligible classes */
1257 struct hfsc_class *
1258 hfsc_get_mindl(struct hfsc_if *hif, u_int64_t cur_time)
1259 {
1260 	struct hfsc_class *p, *cl = NULL;
1261 
1262 	TAILQ_FOREACH(p, &hif->hif_eligible, cl_ellist) {
1263 		if (p->cl_e > cur_time)
1264 			break;
1265 		if (cl == NULL || p->cl_d < cl->cl_d)
1266 			cl = p;
1267 	}
1268 	return (cl);
1269 }
1270 
1271 /*
1272  * active children list holds backlogged child classes being sorted
1273  * by their virtual time.
1274  * each intermediate class has one active children list.
1275  */
1276 
1277 static void
1278 actlist_insert(struct hfsc_class *cl)
1279 {
1280 	struct hfsc_class *p;
1281 
1282 	/* check the last entry first */
1283 	if ((p = TAILQ_LAST(&cl->cl_parent->cl_actc, acthead)) == NULL
1284 	    || p->cl_vt <= cl->cl_vt) {
1285 		TAILQ_INSERT_TAIL(&cl->cl_parent->cl_actc, cl, cl_actlist);
1286 		return;
1287 	}
1288 
1289 	TAILQ_FOREACH(p, &cl->cl_parent->cl_actc, cl_actlist) {
1290 		if (cl->cl_vt < p->cl_vt) {
1291 			TAILQ_INSERT_BEFORE(p, cl, cl_actlist);
1292 			return;
1293 		}
1294 	}
1295 	ASSERT(0); /* should not reach here */
1296 }
1297 
1298 static void
1299 actlist_remove(struct hfsc_class *cl)
1300 {
1301 	TAILQ_REMOVE(&cl->cl_parent->cl_actc, cl, cl_actlist);
1302 }
1303 
1304 static void
1305 actlist_update(struct hfsc_class *cl)
1306 {
1307 	struct hfsc_class *p, *last;
1308 
1309 	/*
1310 	 * the virtual time of a class increases monotonically during its
1311 	 * backlogged period.
1312 	 * if the next entry has a larger virtual time, nothing to do.
1313 	 */
1314 	p = TAILQ_NEXT(cl, cl_actlist);
1315 	if (p == NULL || cl->cl_vt < p->cl_vt)
1316 		return;
1317 
1318 	/* check the last entry */
1319 	last = TAILQ_LAST(&cl->cl_parent->cl_actc, acthead);
1320 	ASSERT(last != NULL);
1321 	if (last->cl_vt <= cl->cl_vt) {
1322 		TAILQ_REMOVE(&cl->cl_parent->cl_actc, cl, cl_actlist);
1323 		TAILQ_INSERT_TAIL(&cl->cl_parent->cl_actc, cl, cl_actlist);
1324 		return;
1325 	}
1326 
1327 	/*
1328 	 * the new position must be between the next entry
1329 	 * and the last entry
1330 	 */
1331 	while ((p = TAILQ_NEXT(p, cl_actlist)) != NULL) {
1332 		if (cl->cl_vt < p->cl_vt) {
1333 			TAILQ_REMOVE(&cl->cl_parent->cl_actc, cl, cl_actlist);
1334 			TAILQ_INSERT_BEFORE(p, cl, cl_actlist);
1335 			return;
1336 		}
1337 	}
1338 	ASSERT(0); /* should not reach here */
1339 }
1340 
1341 static struct hfsc_class *
1342 actlist_firstfit(struct hfsc_class *cl, u_int64_t cur_time)
1343 {
1344 	struct hfsc_class *p;
1345 
1346 	TAILQ_FOREACH(p, &cl->cl_actc, cl_actlist) {
1347 		if (p->cl_f <= cur_time)
1348 			return (p);
1349 	}
1350 	return (NULL);
1351 }
1352 
1353 /*
1354  * service curve support functions
1355  *
1356  *  external service curve parameters
1357  *	m: bits/sec
1358  *	d: msec
1359  *  internal service curve parameters
1360  *	sm: (bytes/tsc_interval) << SM_SHIFT
1361  *	ism: (tsc_count/byte) << ISM_SHIFT
1362  *	dx: tsc_count
1363  *
1364  * SM_SHIFT and ISM_SHIFT are scaled in order to keep effective digits.
1365  * we should be able to handle 100K-1Gbps linkspeed with 200Hz-1GHz CPU
1366  * speed.  SM_SHIFT and ISM_SHIFT are selected to have at least 3 effective
1367  * digits in decimal using the following table.
1368  *
1369  *  bits/sec    100Kbps     1Mbps     10Mbps     100Mbps    1Gbps
1370  *  ----------+-------------------------------------------------------
1371  *  bytes/nsec  12.5e-6    125e-6     1250e-6    12500e-6   125000e-6
1372  *  sm(500MHz)  25.0e-6    250e-6     2500e-6    25000e-6   250000e-6
1373  *  sm(200MHz)  62.5e-6    625e-6     6250e-6    62500e-6   625000e-6
1374  *
1375  *  nsec/byte   80000      8000       800        80         8
1376  *  ism(500MHz) 40000      4000       400        40         4
1377  *  ism(200MHz) 16000      1600       160        16         1.6
1378  */
1379 #define	SM_SHIFT	24
1380 #define	ISM_SHIFT	10
1381 
1382 #define	SM_MASK		((1LL << SM_SHIFT) - 1)
1383 #define	ISM_MASK	((1LL << ISM_SHIFT) - 1)
1384 
1385 static __inline u_int64_t
1386 seg_x2y(u_int64_t x, u_int64_t sm)
1387 {
1388 	u_int64_t y;
1389 
1390 	/*
1391 	 * compute
1392 	 *	y = x * sm >> SM_SHIFT
1393 	 * but divide it for the upper and lower bits to avoid overflow
1394 	 */
1395 	y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT);
1396 	return (y);
1397 }
1398 
1399 static __inline u_int64_t
1400 seg_y2x(u_int64_t y, u_int64_t ism)
1401 {
1402 	u_int64_t x;
1403 
1404 	if (y == 0)
1405 		x = 0;
1406 	else if (ism == HT_INFINITY)
1407 		x = HT_INFINITY;
1408 	else {
1409 		x = (y >> ISM_SHIFT) * ism
1410 		    + (((y & ISM_MASK) * ism) >> ISM_SHIFT);
1411 	}
1412 	return (x);
1413 }
1414 
1415 static __inline u_int64_t
1416 m2sm(u_int m)
1417 {
1418 	u_int64_t sm;
1419 
1420 	sm = ((u_int64_t)m << SM_SHIFT) / 8 / machclk_freq;
1421 	return (sm);
1422 }
1423 
1424 static __inline u_int64_t
1425 m2ism(u_int m)
1426 {
1427 	u_int64_t ism;
1428 
1429 	if (m == 0)
1430 		ism = HT_INFINITY;
1431 	else
1432 		ism = ((u_int64_t)machclk_freq << ISM_SHIFT) * 8 / m;
1433 	return (ism);
1434 }
1435 
1436 static __inline u_int64_t
1437 d2dx(u_int d)
1438 {
1439 	u_int64_t dx;
1440 
1441 	dx = ((u_int64_t)d * machclk_freq) / 1000;
1442 	return (dx);
1443 }
1444 
1445 static u_int
1446 sm2m(u_int64_t sm)
1447 {
1448 	u_int64_t m;
1449 
1450 	m = (sm * 8 * machclk_freq) >> SM_SHIFT;
1451 	return ((u_int)m);
1452 }
1453 
1454 static u_int
1455 dx2d(u_int64_t dx)
1456 {
1457 	u_int64_t d;
1458 
1459 	d = dx * 1000 / machclk_freq;
1460 	return ((u_int)d);
1461 }
1462 
1463 static void
1464 sc2isc(struct service_curve *sc, struct internal_sc *isc)
1465 {
1466 	isc->sm1 = m2sm(sc->m1);
1467 	isc->ism1 = m2ism(sc->m1);
1468 	isc->dx = d2dx(sc->d);
1469 	isc->dy = seg_x2y(isc->dx, isc->sm1);
1470 	isc->sm2 = m2sm(sc->m2);
1471 	isc->ism2 = m2ism(sc->m2);
1472 }
1473 
1474 /*
1475  * initialize the runtime service curve with the given internal
1476  * service curve starting at (x, y).
1477  */
1478 static void
1479 rtsc_init(struct runtime_sc *rtsc, struct internal_sc * isc, u_int64_t x,
1480     u_int64_t y)
1481 {
1482 	rtsc->x =	x;
1483 	rtsc->y =	y;
1484 	rtsc->sm1 =	isc->sm1;
1485 	rtsc->ism1 =	isc->ism1;
1486 	rtsc->dx =	isc->dx;
1487 	rtsc->dy =	isc->dy;
1488 	rtsc->sm2 =	isc->sm2;
1489 	rtsc->ism2 =	isc->ism2;
1490 }
1491 
1492 /*
1493  * calculate the y-projection of the runtime service curve by the
1494  * given x-projection value
1495  */
1496 static u_int64_t
1497 rtsc_y2x(struct runtime_sc *rtsc, u_int64_t y)
1498 {
1499 	u_int64_t	x;
1500 
1501 	if (y < rtsc->y)
1502 		x = rtsc->x;
1503 	else if (y <= rtsc->y + rtsc->dy) {
1504 		/* x belongs to the 1st segment */
1505 		if (rtsc->dy == 0)
1506 			x = rtsc->x + rtsc->dx;
1507 		else
1508 			x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1);
1509 	} else {
1510 		/* x belongs to the 2nd segment */
1511 		x = rtsc->x + rtsc->dx
1512 		    + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2);
1513 	}
1514 	return (x);
1515 }
1516 
1517 static u_int64_t
1518 rtsc_x2y(struct runtime_sc *rtsc, u_int64_t x)
1519 {
1520 	u_int64_t	y;
1521 
1522 	if (x <= rtsc->x)
1523 		y = rtsc->y;
1524 	else if (x <= rtsc->x + rtsc->dx)
1525 		/* y belongs to the 1st segment */
1526 		y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1);
1527 	else
1528 		/* y belongs to the 2nd segment */
1529 		y = rtsc->y + rtsc->dy
1530 		    + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2);
1531 	return (y);
1532 }
1533 
1534 /*
1535  * update the runtime service curve by taking the minimum of the current
1536  * runtime service curve and the service curve starting at (x, y).
1537  */
1538 static void
1539 rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u_int64_t x,
1540     u_int64_t y)
1541 {
1542 	u_int64_t	y1, y2, dx, dy;
1543 
1544 	if (isc->sm1 <= isc->sm2) {
1545 		/* service curve is convex */
1546 		y1 = rtsc_x2y(rtsc, x);
1547 		if (y1 < y)
1548 			/* the current rtsc is smaller */
1549 			return;
1550 		rtsc->x = x;
1551 		rtsc->y = y;
1552 		return;
1553 	}
1554 
1555 	/*
1556 	 * service curve is concave
1557 	 * compute the two y values of the current rtsc
1558 	 *	y1: at x
1559 	 *	y2: at (x + dx)
1560 	 */
1561 	y1 = rtsc_x2y(rtsc, x);
1562 	if (y1 <= y) {
1563 		/* rtsc is below isc, no change to rtsc */
1564 		return;
1565 	}
1566 
1567 	y2 = rtsc_x2y(rtsc, x + isc->dx);
1568 	if (y2 >= y + isc->dy) {
1569 		/* rtsc is above isc, replace rtsc by isc */
1570 		rtsc->x = x;
1571 		rtsc->y = y;
1572 		rtsc->dx = isc->dx;
1573 		rtsc->dy = isc->dy;
1574 		return;
1575 	}
1576 
1577 	/*
1578 	 * the two curves intersect
1579 	 * compute the offsets (dx, dy) using the reverse
1580 	 * function of seg_x2y()
1581 	 *	seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y)
1582 	 */
1583 	dx = ((y1 - y) << SM_SHIFT) / (isc->sm1 - isc->sm2);
1584 	/*
1585 	 * check if (x, y1) belongs to the 1st segment of rtsc.
1586 	 * if so, add the offset.
1587 	 */
1588 	if (rtsc->x + rtsc->dx > x)
1589 		dx += rtsc->x + rtsc->dx - x;
1590 	dy = seg_x2y(dx, isc->sm1);
1591 
1592 	rtsc->x = x;
1593 	rtsc->y = y;
1594 	rtsc->dx = dx;
1595 	rtsc->dy = dy;
1596 	return;
1597 }
1598 
1599 static void
1600 get_class_stats(struct hfsc_classstats *sp, struct hfsc_class *cl)
1601 {
1602 	sp->class_id = cl->cl_id;
1603 	sp->class_handle = cl->cl_handle;
1604 
1605 	if (cl->cl_rsc != NULL) {
1606 		sp->rsc.m1 = sm2m(cl->cl_rsc->sm1);
1607 		sp->rsc.d = dx2d(cl->cl_rsc->dx);
1608 		sp->rsc.m2 = sm2m(cl->cl_rsc->sm2);
1609 	} else {
1610 		sp->rsc.m1 = 0;
1611 		sp->rsc.d = 0;
1612 		sp->rsc.m2 = 0;
1613 	}
1614 	if (cl->cl_fsc != NULL) {
1615 		sp->fsc.m1 = sm2m(cl->cl_fsc->sm1);
1616 		sp->fsc.d = dx2d(cl->cl_fsc->dx);
1617 		sp->fsc.m2 = sm2m(cl->cl_fsc->sm2);
1618 	} else {
1619 		sp->fsc.m1 = 0;
1620 		sp->fsc.d = 0;
1621 		sp->fsc.m2 = 0;
1622 	}
1623 	if (cl->cl_usc != NULL) {
1624 		sp->usc.m1 = sm2m(cl->cl_usc->sm1);
1625 		sp->usc.d = dx2d(cl->cl_usc->dx);
1626 		sp->usc.m2 = sm2m(cl->cl_usc->sm2);
1627 	} else {
1628 		sp->usc.m1 = 0;
1629 		sp->usc.d = 0;
1630 		sp->usc.m2 = 0;
1631 	}
1632 
1633 	sp->total = cl->cl_total;
1634 	sp->cumul = cl->cl_cumul;
1635 
1636 	sp->d = cl->cl_d;
1637 	sp->e = cl->cl_e;
1638 	sp->vt = cl->cl_vt;
1639 	sp->f = cl->cl_f;
1640 
1641 	sp->initvt = cl->cl_initvt;
1642 	sp->vtperiod = cl->cl_vtperiod;
1643 	sp->parentperiod = cl->cl_parentperiod;
1644 	sp->nactive = cl->cl_nactive;
1645 	sp->vtoff = cl->cl_vtoff;
1646 	sp->cvtmax = cl->cl_cvtmax;
1647 	sp->myf = cl->cl_myf;
1648 	sp->cfmin = cl->cl_cfmin;
1649 	sp->cvtmin = cl->cl_cvtmin;
1650 	sp->myfadj = cl->cl_myfadj;
1651 	sp->vtadj = cl->cl_vtadj;
1652 
1653 	sp->cur_time = read_machclk();
1654 	sp->machclk_freq = machclk_freq;
1655 
1656 	sp->qlength = qlen(cl->cl_q);
1657 	sp->qlimit = qlimit(cl->cl_q);
1658 	sp->xmit_cnt = cl->cl_stats.xmit_cnt;
1659 	sp->drop_cnt = cl->cl_stats.drop_cnt;
1660 	sp->period = cl->cl_stats.period;
1661 
1662 	sp->qtype = qtype(cl->cl_q);
1663 #ifdef ALTQ_RED
1664 	if (q_is_red(cl->cl_q))
1665 		red_getstats(cl->cl_red, &sp->red[0]);
1666 #endif
1667 #ifdef ALTQ_RIO
1668 	if (q_is_rio(cl->cl_q))
1669 		rio_getstats((rio_t *)cl->cl_red, &sp->red[0]);
1670 #endif
1671 #ifdef ALTQ_CODEL
1672 	if (q_is_codel(cl->cl_q))
1673 		codel_getstats(cl->cl_codel, &sp->codel);
1674 #endif
1675 }
1676 
1677 /* convert a class handle to the corresponding class pointer */
1678 static struct hfsc_class *
1679 clh_to_clp(struct hfsc_if *hif, u_int32_t chandle)
1680 {
1681 	int i;
1682 	struct hfsc_class *cl;
1683 
1684 	if (chandle == 0)
1685 		return (NULL);
1686 	/*
1687 	 * first, try optimistically the slot matching the lower bits of
1688 	 * the handle.  if it fails, do the linear table search.
1689 	 */
1690 	i = chandle % HFSC_MAX_CLASSES;
1691 	if ((cl = hif->hif_class_tbl[i]) != NULL && cl->cl_handle == chandle)
1692 		return (cl);
1693 	for (i = 0; i < HFSC_MAX_CLASSES; i++)
1694 		if ((cl = hif->hif_class_tbl[i]) != NULL &&
1695 		    cl->cl_handle == chandle)
1696 			return (cl);
1697 	return (NULL);
1698 }
1699 
1700 #ifdef ALTQ3_COMPAT
1701 static struct hfsc_if *
1702 hfsc_attach(ifq, bandwidth)
1703 	struct ifaltq *ifq;
1704 	u_int bandwidth;
1705 {
1706 	struct hfsc_if *hif;
1707 
1708 	hif = malloc(sizeof(struct hfsc_if), M_DEVBUF, M_WAITOK);
1709 	if (hif == NULL)
1710 		return (NULL);
1711 	bzero(hif, sizeof(struct hfsc_if));
1712 
1713 	hif->hif_eligible = ellist_alloc();
1714 	if (hif->hif_eligible == NULL) {
1715 		free(hif, M_DEVBUF);
1716 		return NULL;
1717 	}
1718 
1719 	hif->hif_ifq = ifq;
1720 
1721 	/* add this state to the hfsc list */
1722 	hif->hif_next = hif_list;
1723 	hif_list = hif;
1724 
1725 	return (hif);
1726 }
1727 
1728 static int
1729 hfsc_detach(hif)
1730 	struct hfsc_if *hif;
1731 {
1732 	(void)hfsc_clear_interface(hif);
1733 	(void)hfsc_class_destroy(hif->hif_rootclass);
1734 
1735 	/* remove this interface from the hif list */
1736 	if (hif_list == hif)
1737 		hif_list = hif->hif_next;
1738 	else {
1739 		struct hfsc_if *h;
1740 
1741 		for (h = hif_list; h != NULL; h = h->hif_next)
1742 			if (h->hif_next == hif) {
1743 				h->hif_next = hif->hif_next;
1744 				break;
1745 			}
1746 		ASSERT(h != NULL);
1747 	}
1748 
1749 	ellist_destroy(hif->hif_eligible);
1750 
1751 	free(hif, M_DEVBUF);
1752 
1753 	return (0);
1754 }
1755 
1756 static int
1757 hfsc_class_modify(cl, rsc, fsc, usc)
1758 	struct hfsc_class *cl;
1759 	struct service_curve *rsc, *fsc, *usc;
1760 {
1761 	struct internal_sc *rsc_tmp, *fsc_tmp, *usc_tmp;
1762 	u_int64_t cur_time;
1763 	int s;
1764 
1765 	rsc_tmp = fsc_tmp = usc_tmp = NULL;
1766 	if (rsc != NULL && (rsc->m1 != 0 || rsc->m2 != 0) &&
1767 	    cl->cl_rsc == NULL) {
1768 		rsc_tmp = malloc(sizeof(struct internal_sc),
1769 		    M_DEVBUF, M_WAITOK);
1770 		if (rsc_tmp == NULL)
1771 			return (ENOMEM);
1772 	}
1773 	if (fsc != NULL && (fsc->m1 != 0 || fsc->m2 != 0) &&
1774 	    cl->cl_fsc == NULL) {
1775 		fsc_tmp = malloc(sizeof(struct internal_sc),
1776 		    M_DEVBUF, M_WAITOK);
1777 		if (fsc_tmp == NULL) {
1778 			free(rsc_tmp);
1779 			return (ENOMEM);
1780 		}
1781 	}
1782 	if (usc != NULL && (usc->m1 != 0 || usc->m2 != 0) &&
1783 	    cl->cl_usc == NULL) {
1784 		usc_tmp = malloc(sizeof(struct internal_sc),
1785 		    M_DEVBUF, M_WAITOK);
1786 		if (usc_tmp == NULL) {
1787 			free(rsc_tmp);
1788 			free(fsc_tmp);
1789 			return (ENOMEM);
1790 		}
1791 	}
1792 
1793 	cur_time = read_machclk();
1794 	s = splnet();
1795 	IFQ_LOCK(cl->cl_hif->hif_ifq);
1796 
1797 	if (rsc != NULL) {
1798 		if (rsc->m1 == 0 && rsc->m2 == 0) {
1799 			if (cl->cl_rsc != NULL) {
1800 				if (!qempty(cl->cl_q))
1801 					hfsc_purgeq(cl);
1802 				free(cl->cl_rsc, M_DEVBUF);
1803 				cl->cl_rsc = NULL;
1804 			}
1805 		} else {
1806 			if (cl->cl_rsc == NULL)
1807 				cl->cl_rsc = rsc_tmp;
1808 			sc2isc(rsc, cl->cl_rsc);
1809 			rtsc_init(&cl->cl_deadline, cl->cl_rsc, cur_time,
1810 			    cl->cl_cumul);
1811 			cl->cl_eligible = cl->cl_deadline;
1812 			if (cl->cl_rsc->sm1 <= cl->cl_rsc->sm2) {
1813 				cl->cl_eligible.dx = 0;
1814 				cl->cl_eligible.dy = 0;
1815 			}
1816 		}
1817 	}
1818 
1819 	if (fsc != NULL) {
1820 		if (fsc->m1 == 0 && fsc->m2 == 0) {
1821 			if (cl->cl_fsc != NULL) {
1822 				if (!qempty(cl->cl_q))
1823 					hfsc_purgeq(cl);
1824 				free(cl->cl_fsc, M_DEVBUF);
1825 				cl->cl_fsc = NULL;
1826 			}
1827 		} else {
1828 			if (cl->cl_fsc == NULL)
1829 				cl->cl_fsc = fsc_tmp;
1830 			sc2isc(fsc, cl->cl_fsc);
1831 			rtsc_init(&cl->cl_virtual, cl->cl_fsc, cl->cl_vt,
1832 			    cl->cl_total);
1833 		}
1834 	}
1835 
1836 	if (usc != NULL) {
1837 		if (usc->m1 == 0 && usc->m2 == 0) {
1838 			if (cl->cl_usc != NULL) {
1839 				free(cl->cl_usc, M_DEVBUF);
1840 				cl->cl_usc = NULL;
1841 				cl->cl_myf = 0;
1842 			}
1843 		} else {
1844 			if (cl->cl_usc == NULL)
1845 				cl->cl_usc = usc_tmp;
1846 			sc2isc(usc, cl->cl_usc);
1847 			rtsc_init(&cl->cl_ulimit, cl->cl_usc, cur_time,
1848 			    cl->cl_total);
1849 		}
1850 	}
1851 
1852 	if (!qempty(cl->cl_q)) {
1853 		if (cl->cl_rsc != NULL)
1854 			update_ed(cl, m_pktlen(qhead(cl->cl_q)));
1855 		if (cl->cl_fsc != NULL)
1856 			update_vf(cl, 0, cur_time);
1857 		/* is this enough? */
1858 	}
1859 
1860 	IFQ_UNLOCK(cl->cl_hif->hif_ifq);
1861 	splx(s);
1862 
1863 	return (0);
1864 }
1865 
1866 /*
1867  * hfsc device interface
1868  */
1869 int
1870 hfscopen(dev, flag, fmt, p)
1871 	dev_t dev;
1872 	int flag, fmt;
1873 #if (__FreeBSD_version > 500000)
1874 	struct thread *p;
1875 #else
1876 	struct proc *p;
1877 #endif
1878 {
1879 	if (machclk_freq == 0)
1880 		init_machclk();
1881 
1882 	if (machclk_freq == 0) {
1883 		printf("hfsc: no cpu clock available!\n");
1884 		return (ENXIO);
1885 	}
1886 
1887 	/* everything will be done when the queueing scheme is attached. */
1888 	return 0;
1889 }
1890 
1891 int
1892 hfscclose(dev, flag, fmt, p)
1893 	dev_t dev;
1894 	int flag, fmt;
1895 #if (__FreeBSD_version > 500000)
1896 	struct thread *p;
1897 #else
1898 	struct proc *p;
1899 #endif
1900 {
1901 	struct hfsc_if *hif;
1902 	int err, error = 0;
1903 
1904 	while ((hif = hif_list) != NULL) {
1905 		/* destroy all */
1906 		if (ALTQ_IS_ENABLED(hif->hif_ifq))
1907 			altq_disable(hif->hif_ifq);
1908 
1909 		err = altq_detach(hif->hif_ifq);
1910 		if (err == 0)
1911 			err = hfsc_detach(hif);
1912 		if (err != 0 && error == 0)
1913 			error = err;
1914 	}
1915 
1916 	return error;
1917 }
1918 
1919 int
1920 hfscioctl(dev, cmd, addr, flag, p)
1921 	dev_t dev;
1922 	ioctlcmd_t cmd;
1923 	caddr_t addr;
1924 	int flag;
1925 #if (__FreeBSD_version > 500000)
1926 	struct thread *p;
1927 #else
1928 	struct proc *p;
1929 #endif
1930 {
1931 	struct hfsc_if *hif;
1932 	struct hfsc_interface *ifacep;
1933 	int	error = 0;
1934 
1935 	/* check super-user privilege */
1936 	switch (cmd) {
1937 	case HFSC_GETSTATS:
1938 		break;
1939 	default:
1940 #if (__FreeBSD_version > 700000)
1941 		if ((error = priv_check(p, PRIV_ALTQ_MANAGE)) != 0)
1942 			return (error);
1943 #elsif (__FreeBSD_version > 400000)
1944 		if ((error = suser(p)) != 0)
1945 			return (error);
1946 #else
1947 		if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
1948 			return (error);
1949 #endif
1950 		break;
1951 	}
1952 
1953 	switch (cmd) {
1954 
1955 	case HFSC_IF_ATTACH:
1956 		error = hfsccmd_if_attach((struct hfsc_attach *)addr);
1957 		break;
1958 
1959 	case HFSC_IF_DETACH:
1960 		error = hfsccmd_if_detach((struct hfsc_interface *)addr);
1961 		break;
1962 
1963 	case HFSC_ENABLE:
1964 	case HFSC_DISABLE:
1965 	case HFSC_CLEAR_HIERARCHY:
1966 		ifacep = (struct hfsc_interface *)addr;
1967 		if ((hif = altq_lookup(ifacep->hfsc_ifname,
1968 				       ALTQT_HFSC)) == NULL) {
1969 			error = EBADF;
1970 			break;
1971 		}
1972 
1973 		switch (cmd) {
1974 
1975 		case HFSC_ENABLE:
1976 			if (hif->hif_defaultclass == NULL) {
1977 #ifdef ALTQ_DEBUG
1978 				printf("hfsc: no default class\n");
1979 #endif
1980 				error = EINVAL;
1981 				break;
1982 			}
1983 			error = altq_enable(hif->hif_ifq);
1984 			break;
1985 
1986 		case HFSC_DISABLE:
1987 			error = altq_disable(hif->hif_ifq);
1988 			break;
1989 
1990 		case HFSC_CLEAR_HIERARCHY:
1991 			hfsc_clear_interface(hif);
1992 			break;
1993 		}
1994 		break;
1995 
1996 	case HFSC_ADD_CLASS:
1997 		error = hfsccmd_add_class((struct hfsc_add_class *)addr);
1998 		break;
1999 
2000 	case HFSC_DEL_CLASS:
2001 		error = hfsccmd_delete_class((struct hfsc_delete_class *)addr);
2002 		break;
2003 
2004 	case HFSC_MOD_CLASS:
2005 		error = hfsccmd_modify_class((struct hfsc_modify_class *)addr);
2006 		break;
2007 
2008 	case HFSC_ADD_FILTER:
2009 		error = hfsccmd_add_filter((struct hfsc_add_filter *)addr);
2010 		break;
2011 
2012 	case HFSC_DEL_FILTER:
2013 		error = hfsccmd_delete_filter((struct hfsc_delete_filter *)addr);
2014 		break;
2015 
2016 	case HFSC_GETSTATS:
2017 		error = hfsccmd_class_stats((struct hfsc_class_stats *)addr);
2018 		break;
2019 
2020 	default:
2021 		error = EINVAL;
2022 		break;
2023 	}
2024 	return error;
2025 }
2026 
2027 static int
2028 hfsccmd_if_attach(ap)
2029 	struct hfsc_attach *ap;
2030 {
2031 	struct hfsc_if *hif;
2032 	struct ifnet *ifp;
2033 	int error;
2034 
2035 	if ((ifp = ifunit(ap->iface.hfsc_ifname)) == NULL)
2036 		return (ENXIO);
2037 
2038 	if ((hif = hfsc_attach(&ifp->if_snd, ap->bandwidth)) == NULL)
2039 		return (ENOMEM);
2040 
2041 	/*
2042 	 * set HFSC to this ifnet structure.
2043 	 */
2044 	if ((error = altq_attach(&ifp->if_snd, ALTQT_HFSC, hif,
2045 				 hfsc_enqueue, hfsc_dequeue, hfsc_request,
2046 				 &hif->hif_classifier, acc_classify)) != 0)
2047 		(void)hfsc_detach(hif);
2048 
2049 	return (error);
2050 }
2051 
2052 static int
2053 hfsccmd_if_detach(ap)
2054 	struct hfsc_interface *ap;
2055 {
2056 	struct hfsc_if *hif;
2057 	int error;
2058 
2059 	if ((hif = altq_lookup(ap->hfsc_ifname, ALTQT_HFSC)) == NULL)
2060 		return (EBADF);
2061 
2062 	if (ALTQ_IS_ENABLED(hif->hif_ifq))
2063 		altq_disable(hif->hif_ifq);
2064 
2065 	if ((error = altq_detach(hif->hif_ifq)))
2066 		return (error);
2067 
2068 	return hfsc_detach(hif);
2069 }
2070 
2071 static int
2072 hfsccmd_add_class(ap)
2073 	struct hfsc_add_class *ap;
2074 {
2075 	struct hfsc_if *hif;
2076 	struct hfsc_class *cl, *parent;
2077 	int	i;
2078 
2079 	if ((hif = altq_lookup(ap->iface.hfsc_ifname, ALTQT_HFSC)) == NULL)
2080 		return (EBADF);
2081 
2082 	if (ap->parent_handle == HFSC_NULLCLASS_HANDLE &&
2083 	    hif->hif_rootclass == NULL)
2084 		parent = NULL;
2085 	else if ((parent = clh_to_clp(hif, ap->parent_handle)) == NULL)
2086 		return (EINVAL);
2087 
2088 	/* assign a class handle (use a free slot number for now) */
2089 	for (i = 1; i < HFSC_MAX_CLASSES; i++)
2090 		if (hif->hif_class_tbl[i] == NULL)
2091 			break;
2092 	if (i == HFSC_MAX_CLASSES)
2093 		return (EBUSY);
2094 
2095 	if ((cl = hfsc_class_create(hif, &ap->service_curve, NULL, NULL,
2096 	    parent, ap->qlimit, ap->flags, i)) == NULL)
2097 		return (ENOMEM);
2098 
2099 	/* return a class handle to the user */
2100 	ap->class_handle = i;
2101 
2102 	return (0);
2103 }
2104 
2105 static int
2106 hfsccmd_delete_class(ap)
2107 	struct hfsc_delete_class *ap;
2108 {
2109 	struct hfsc_if *hif;
2110 	struct hfsc_class *cl;
2111 
2112 	if ((hif = altq_lookup(ap->iface.hfsc_ifname, ALTQT_HFSC)) == NULL)
2113 		return (EBADF);
2114 
2115 	if ((cl = clh_to_clp(hif, ap->class_handle)) == NULL)
2116 		return (EINVAL);
2117 
2118 	return hfsc_class_destroy(cl);
2119 }
2120 
2121 static int
2122 hfsccmd_modify_class(ap)
2123 	struct hfsc_modify_class *ap;
2124 {
2125 	struct hfsc_if *hif;
2126 	struct hfsc_class *cl;
2127 	struct service_curve *rsc = NULL;
2128 	struct service_curve *fsc = NULL;
2129 	struct service_curve *usc = NULL;
2130 
2131 	if ((hif = altq_lookup(ap->iface.hfsc_ifname, ALTQT_HFSC)) == NULL)
2132 		return (EBADF);
2133 
2134 	if ((cl = clh_to_clp(hif, ap->class_handle)) == NULL)
2135 		return (EINVAL);
2136 
2137 	if (ap->sctype & HFSC_REALTIMESC)
2138 		rsc = &ap->service_curve;
2139 	if (ap->sctype & HFSC_LINKSHARINGSC)
2140 		fsc = &ap->service_curve;
2141 	if (ap->sctype & HFSC_UPPERLIMITSC)
2142 		usc = &ap->service_curve;
2143 
2144 	return hfsc_class_modify(cl, rsc, fsc, usc);
2145 }
2146 
2147 static int
2148 hfsccmd_add_filter(ap)
2149 	struct hfsc_add_filter *ap;
2150 {
2151 	struct hfsc_if *hif;
2152 	struct hfsc_class *cl;
2153 
2154 	if ((hif = altq_lookup(ap->iface.hfsc_ifname, ALTQT_HFSC)) == NULL)
2155 		return (EBADF);
2156 
2157 	if ((cl = clh_to_clp(hif, ap->class_handle)) == NULL)
2158 		return (EINVAL);
2159 
2160 	if (is_a_parent_class(cl)) {
2161 #ifdef ALTQ_DEBUG
2162 		printf("hfsccmd_add_filter: not a leaf class!\n");
2163 #endif
2164 		return (EINVAL);
2165 	}
2166 
2167 	return acc_add_filter(&hif->hif_classifier, &ap->filter,
2168 			      cl, &ap->filter_handle);
2169 }
2170 
2171 static int
2172 hfsccmd_delete_filter(ap)
2173 	struct hfsc_delete_filter *ap;
2174 {
2175 	struct hfsc_if *hif;
2176 
2177 	if ((hif = altq_lookup(ap->iface.hfsc_ifname, ALTQT_HFSC)) == NULL)
2178 		return (EBADF);
2179 
2180 	return acc_delete_filter(&hif->hif_classifier,
2181 				 ap->filter_handle);
2182 }
2183 
2184 static int
2185 hfsccmd_class_stats(ap)
2186 	struct hfsc_class_stats *ap;
2187 {
2188 	struct hfsc_if *hif;
2189 	struct hfsc_class *cl;
2190 	struct hfsc_classstats stats, *usp;
2191 	int	n, nclasses, error;
2192 
2193 	if ((hif = altq_lookup(ap->iface.hfsc_ifname, ALTQT_HFSC)) == NULL)
2194 		return (EBADF);
2195 
2196 	ap->cur_time = read_machclk();
2197 	ap->machclk_freq = machclk_freq;
2198 	ap->hif_classes = hif->hif_classes;
2199 	ap->hif_packets = hif->hif_packets;
2200 
2201 	/* skip the first N classes in the tree */
2202 	nclasses = ap->nskip;
2203 	for (cl = hif->hif_rootclass, n = 0; cl != NULL && n < nclasses;
2204 	     cl = hfsc_nextclass(cl), n++)
2205 		;
2206 	if (n != nclasses)
2207 		return (EINVAL);
2208 
2209 	/* then, read the next N classes in the tree */
2210 	nclasses = ap->nclasses;
2211 	usp = ap->stats;
2212 	for (n = 0; cl != NULL && n < nclasses; cl = hfsc_nextclass(cl), n++) {
2213 
2214 		get_class_stats(&stats, cl);
2215 
2216 		if ((error = copyout((caddr_t)&stats, (caddr_t)usp++,
2217 				     sizeof(stats))) != 0)
2218 			return (error);
2219 	}
2220 
2221 	ap->nclasses = n;
2222 
2223 	return (0);
2224 }
2225 
2226 #ifdef KLD_MODULE
2227 
2228 static struct altqsw hfsc_sw =
2229 	{"hfsc", hfscopen, hfscclose, hfscioctl};
2230 
2231 ALTQ_MODULE(altq_hfsc, ALTQT_HFSC, &hfsc_sw);
2232 MODULE_DEPEND(altq_hfsc, altq_red, 1, 1, 1);
2233 MODULE_DEPEND(altq_hfsc, altq_rio, 1, 1, 1);
2234 
2235 #endif /* KLD_MODULE */
2236 #endif /* ALTQ3_COMPAT */
2237 
2238 #endif /* ALTQ_HFSC */
2239