xref: /netbsd-src/sys/altq/altq_rio.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: altq_rio.c,v 1.23 2016/05/16 09:53:59 ozaki-r Exp $	*/
2 /*	$KAME: altq_rio.c,v 1.19 2005/04/13 03:44:25 suz Exp $	*/
3 
4 /*
5  * Copyright (C) 1998-2003
6  *	Sony Computer Science Laboratories Inc.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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 the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 /*
30  * Copyright (c) 1990-1994 Regents of the University of California.
31  * All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. All advertising materials mentioning features or use of this software
42  *    must display the following acknowledgement:
43  *	This product includes software developed by the Computer Systems
44  *	Engineering Group at Lawrence Berkeley Laboratory.
45  * 4. Neither the name of the University nor of the Laboratory may be used
46  *    to endorse or promote products derived from this software without
47  *    specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  */
61 
62 #include <sys/cdefs.h>
63 __KERNEL_RCSID(0, "$NetBSD: altq_rio.c,v 1.23 2016/05/16 09:53:59 ozaki-r Exp $");
64 
65 #ifdef _KERNEL_OPT
66 #include "opt_altq.h"
67 #include "opt_inet.h"
68 #include "pf.h"
69 #endif
70 
71 #ifdef ALTQ_RIO	/* rio is enabled by ALTQ_RIO option in opt_altq.h */
72 
73 #include <sys/param.h>
74 #include <sys/malloc.h>
75 #include <sys/mbuf.h>
76 #include <sys/socket.h>
77 #include <sys/systm.h>
78 #include <sys/errno.h>
79 #include <sys/kauth.h>
80 #if 1 /* ALTQ3_COMPAT */
81 #include <sys/proc.h>
82 #include <sys/sockio.h>
83 #include <sys/kernel.h>
84 #endif
85 
86 #include <net/if.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/ip.h>
91 #ifdef INET6
92 #include <netinet/ip6.h>
93 #endif
94 
95 #if NPF > 0
96 #include <net/pfvar.h>
97 #endif
98 #include <altq/altq.h>
99 #include <altq/altq_cdnr.h>
100 #include <altq/altq_red.h>
101 #include <altq/altq_rio.h>
102 #ifdef ALTQ3_COMPAT
103 #include <altq/altq_conf.h>
104 #endif
105 
106 /*
107  * RIO: RED with IN/OUT bit
108  *   described in
109  *	"Explicit Allocation of Best Effort Packet Delivery Service"
110  *	David D. Clark and Wenjia Fang, MIT Lab for Computer Science
111  *	http://diffserv.lcs.mit.edu/Papers/exp-alloc-ddc-wf.{ps,pdf}
112  *
113  * this implementation is extended to support more than 2 drop precedence
114  * values as described in RFC2597 (Assured Forwarding PHB Group).
115  *
116  */
117 /*
118  * AF DS (differentiated service) codepoints.
119  * (classes can be mapped to CBQ or H-FSC classes.)
120  *
121  *      0   1   2   3   4   5   6   7
122  *    +---+---+---+---+---+---+---+---+
123  *    |   CLASS   |DropPre| 0 |  CU   |
124  *    +---+---+---+---+---+---+---+---+
125  *
126  *    class 1: 001
127  *    class 2: 010
128  *    class 3: 011
129  *    class 4: 100
130  *
131  *    low drop prec:    01
132  *    medium drop prec: 10
133  *    high drop prec:   11
134  */
135 
136 /* normal red parameters */
137 #define	W_WEIGHT	512	/* inverse of weight of EWMA (511/512) */
138 				/* q_weight = 0.00195 */
139 
140 /* red parameters for a slow link */
141 #define	W_WEIGHT_1	128	/* inverse of weight of EWMA (127/128) */
142 				/* q_weight = 0.0078125 */
143 
144 /* red parameters for a very slow link (e.g., dialup) */
145 #define	W_WEIGHT_2	64	/* inverse of weight of EWMA (63/64) */
146 				/* q_weight = 0.015625 */
147 
148 /* fixed-point uses 12-bit decimal places */
149 #define	FP_SHIFT	12	/* fixed-point shift */
150 
151 /* red parameters for drop probability */
152 #define	INV_P_MAX	10	/* inverse of max drop probability */
153 #define	TH_MIN		 5	/* min threshold */
154 #define	TH_MAX		15	/* max threshold */
155 
156 #define	RIO_LIMIT	60	/* default max queue lenght */
157 #define	RIO_STATS		/* collect statistics */
158 
159 #define	TV_DELTA(a, b, delta) {					\
160 	register int	xxs;					\
161 								\
162 	delta = (a)->tv_usec - (b)->tv_usec; 			\
163 	if ((xxs = (a)->tv_sec - (b)->tv_sec) != 0) { 		\
164 		if (xxs < 0) { 					\
165 			delta = 60000000;			\
166 		} else if (xxs > 4)  {				\
167 			if (xxs > 60)				\
168 				delta = 60000000;		\
169 			else					\
170 				delta += xxs * 1000000;		\
171 		} else while (xxs > 0) {			\
172 			delta += 1000000;			\
173 			xxs--;					\
174 		}						\
175 	}							\
176 }
177 
178 #ifdef ALTQ3_COMPAT
179 /* rio_list keeps all rio_queue_t's allocated. */
180 static rio_queue_t *rio_list = NULL;
181 #endif
182 /* default rio parameter values */
183 static struct redparams default_rio_params[RIO_NDROPPREC] = {
184   /* th_min,		 th_max,     inv_pmax */
185   { TH_MAX * 2 + TH_MIN, TH_MAX * 3, INV_P_MAX }, /* low drop precedence */
186   { TH_MAX + TH_MIN,	 TH_MAX * 2, INV_P_MAX }, /* medium drop precedence */
187   { TH_MIN,		 TH_MAX,     INV_P_MAX }  /* high drop precedence */
188 };
189 
190 /* internal function prototypes */
191 static int dscp2index(u_int8_t);
192 #ifdef ALTQ3_COMPAT
193 static int rio_enqueue(struct ifaltq *, struct mbuf *);
194 static struct mbuf *rio_dequeue(struct ifaltq *, int);
195 static int rio_request(struct ifaltq *, int, void *);
196 static int rio_detach(rio_queue_t *);
197 
198 /*
199  * rio device interface
200  */
201 altqdev_decl(rio);
202 
203 #endif /* ALTQ3_COMPAT */
204 
205 rio_t *
206 rio_alloc(int weight, struct redparams *params, int flags, int pkttime)
207 {
208 	rio_t	*rp;
209 	int	 w, i;
210 	int	 npkts_per_sec;
211 
212 	rp = malloc(sizeof(rio_t), M_DEVBUF, M_WAITOK|M_ZERO);
213 	if (rp == NULL)
214 		return (NULL);
215 
216 	rp->rio_flags = flags;
217 	if (pkttime == 0)
218 		/* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */
219 		rp->rio_pkttime = 800;
220 	else
221 		rp->rio_pkttime = pkttime;
222 
223 	if (weight != 0)
224 		rp->rio_weight = weight;
225 	else {
226 		/* use default */
227 		rp->rio_weight = W_WEIGHT;
228 
229 		/* when the link is very slow, adjust red parameters */
230 		npkts_per_sec = 1000000 / rp->rio_pkttime;
231 		if (npkts_per_sec < 50) {
232 			/* up to about 400Kbps */
233 			rp->rio_weight = W_WEIGHT_2;
234 		} else if (npkts_per_sec < 300) {
235 			/* up to about 2.4Mbps */
236 			rp->rio_weight = W_WEIGHT_1;
237 		}
238 	}
239 
240 	/* calculate wshift.  weight must be power of 2 */
241 	w = rp->rio_weight;
242 	for (i = 0; w > 1; i++)
243 		w = w >> 1;
244 	rp->rio_wshift = i;
245 	w = 1 << rp->rio_wshift;
246 	if (w != rp->rio_weight) {
247 		printf("invalid weight value %d for red! use %d\n",
248 		       rp->rio_weight, w);
249 		rp->rio_weight = w;
250 	}
251 
252 	/* allocate weight table */
253 	rp->rio_wtab = wtab_alloc(rp->rio_weight);
254 
255 	for (i = 0; i < RIO_NDROPPREC; i++) {
256 		struct dropprec_state *prec = &rp->rio_precstate[i];
257 
258 		prec->avg = 0;
259 		prec->idle = 1;
260 
261 		if (params == NULL || params[i].inv_pmax == 0)
262 			prec->inv_pmax = default_rio_params[i].inv_pmax;
263 		else
264 			prec->inv_pmax = params[i].inv_pmax;
265 		if (params == NULL || params[i].th_min == 0)
266 			prec->th_min = default_rio_params[i].th_min;
267 		else
268 			prec->th_min = params[i].th_min;
269 		if (params == NULL || params[i].th_max == 0)
270 			prec->th_max = default_rio_params[i].th_max;
271 		else
272 			prec->th_max = params[i].th_max;
273 
274 		/*
275 		 * th_min_s and th_max_s are scaled versions of th_min
276 		 * and th_max to be compared with avg.
277 		 */
278 		prec->th_min_s = prec->th_min << (rp->rio_wshift + FP_SHIFT);
279 		prec->th_max_s = prec->th_max << (rp->rio_wshift + FP_SHIFT);
280 
281 		/*
282 		 * precompute probability denominator
283 		 *  probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point
284 		 */
285 		prec->probd = (2 * (prec->th_max - prec->th_min)
286 			       * prec->inv_pmax) << FP_SHIFT;
287 
288 		microtime(&prec->last);
289 	}
290 
291 	return (rp);
292 }
293 
294 void
295 rio_destroy(rio_t *rp)
296 {
297 	wtab_destroy(rp->rio_wtab);
298 	free(rp, M_DEVBUF);
299 }
300 
301 void
302 rio_getstats(rio_t *rp, struct redstats *sp)
303 {
304 	int	i;
305 
306 	for (i = 0; i < RIO_NDROPPREC; i++) {
307 		memcpy(sp, &rp->q_stats[i], sizeof(struct redstats));
308 		sp->q_avg = rp->rio_precstate[i].avg >> rp->rio_wshift;
309 		sp++;
310 	}
311 }
312 
313 #if (RIO_NDROPPREC == 3)
314 /*
315  * internally, a drop precedence value is converted to an index
316  * starting from 0.
317  */
318 static int
319 dscp2index(u_int8_t dscp)
320 {
321 	int	dpindex = dscp & AF_DROPPRECMASK;
322 
323 	if (dpindex == 0)
324 		return (0);
325 	return ((dpindex >> 3) - 1);
326 }
327 #endif
328 
329 int
330 rio_addq(rio_t *rp, class_queue_t *q, struct mbuf *m,
331     struct altq_pktattr *pktattr)
332 {
333 	int			 avg, droptype;
334 	u_int8_t		 dsfield, odsfield;
335 	int			 dpindex, i, n, t;
336 	struct timeval		 now;
337 	struct dropprec_state	*prec;
338 
339 	dsfield = odsfield = read_dsfield(m, pktattr);
340 	dpindex = dscp2index(dsfield);
341 
342 	/*
343 	 * update avg of the precedence states whose drop precedence
344 	 * is larger than or equal to the drop precedence of the packet
345 	 */
346 	now.tv_sec = 0;
347 	for (i = dpindex; i < RIO_NDROPPREC; i++) {
348 		prec = &rp->rio_precstate[i];
349 		avg = prec->avg;
350 		if (prec->idle) {
351 			prec->idle = 0;
352 			if (now.tv_sec == 0)
353 				microtime(&now);
354 			t = (now.tv_sec - prec->last.tv_sec);
355 			if (t > 60)
356 				avg = 0;
357 			else {
358 				t = t * 1000000 +
359 					(now.tv_usec - prec->last.tv_usec);
360 				n = t / rp->rio_pkttime;
361 				/* calculate (avg = (1 - Wq)^n * avg) */
362 				if (n > 0)
363 					avg = (avg >> FP_SHIFT) *
364 						pow_w(rp->rio_wtab, n);
365 			}
366 		}
367 
368 		/* run estimator. (avg is scaled by WEIGHT in fixed-point) */
369 		avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift);
370 		prec->avg = avg;		/* save the new value */
371 		/*
372 		 * count keeps a tally of arriving traffic that has not
373 		 * been dropped.
374 		 */
375 		prec->count++;
376 	}
377 
378 	prec = &rp->rio_precstate[dpindex];
379 	avg = prec->avg;
380 
381 	/* see if we drop early */
382 	droptype = DTYPE_NODROP;
383 	if (avg >= prec->th_min_s && prec->qlen > 1) {
384 		if (avg >= prec->th_max_s) {
385 			/* avg >= th_max: forced drop */
386 			droptype = DTYPE_FORCED;
387 		} else if (prec->old == 0) {
388 			/* first exceeds th_min */
389 			prec->count = 1;
390 			prec->old = 1;
391 		} else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift,
392 				      prec->probd, prec->count)) {
393 			/* unforced drop by red */
394 			droptype = DTYPE_EARLY;
395 		}
396 	} else {
397 		/* avg < th_min */
398 		prec->old = 0;
399 	}
400 
401 	/*
402 	 * if the queue length hits the hard limit, it's a forced drop.
403 	 */
404 	if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q))
405 		droptype = DTYPE_FORCED;
406 
407 	if (droptype != DTYPE_NODROP) {
408 		/* always drop incoming packet (as opposed to randomdrop) */
409 		for (i = dpindex; i < RIO_NDROPPREC; i++)
410 			rp->rio_precstate[i].count = 0;
411 #ifdef RIO_STATS
412 		if (droptype == DTYPE_EARLY)
413 			rp->q_stats[dpindex].drop_unforced++;
414 		else
415 			rp->q_stats[dpindex].drop_forced++;
416 		PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m));
417 #endif
418 		m_freem(m);
419 		return (-1);
420 	}
421 
422 	for (i = dpindex; i < RIO_NDROPPREC; i++)
423 		rp->rio_precstate[i].qlen++;
424 
425 	/* save drop precedence index in mbuf hdr */
426 	M_SETCTX(m, (intptr_t)dpindex);
427 
428 	if (rp->rio_flags & RIOF_CLEARDSCP)
429 		dsfield &= ~DSCP_MASK;
430 
431 	if (dsfield != odsfield)
432 		write_dsfield(m, pktattr, dsfield);
433 
434 	_addq(q, m);
435 
436 #ifdef RIO_STATS
437 	PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m));
438 #endif
439 	return (0);
440 }
441 
442 struct mbuf *
443 rio_getq(rio_t *rp, class_queue_t *q)
444 {
445 	struct mbuf	*m;
446 	int		 dpindex, i;
447 
448 	if ((m = _getq(q)) == NULL)
449 		return NULL;
450 
451 	dpindex = M_GETCTX(m, intptr_t);
452 	for (i = dpindex; i < RIO_NDROPPREC; i++) {
453 		if (--rp->rio_precstate[i].qlen == 0) {
454 			if (rp->rio_precstate[i].idle == 0) {
455 				rp->rio_precstate[i].idle = 1;
456 				microtime(&rp->rio_precstate[i].last);
457 			}
458 		}
459 	}
460 	return (m);
461 }
462 
463 #ifdef ALTQ3_COMPAT
464 int
465 rioopen(dev_t dev, int flag, int fmt,
466     struct lwp *l)
467 {
468 	/* everything will be done when the queueing scheme is attached. */
469 	return 0;
470 }
471 
472 int
473 rioclose(dev_t dev, int flag, int fmt,
474     struct lwp *l)
475 {
476 	rio_queue_t *rqp;
477 	int err, error = 0;
478 
479 	while ((rqp = rio_list) != NULL) {
480 		/* destroy all */
481 		err = rio_detach(rqp);
482 		if (err != 0 && error == 0)
483 			error = err;
484 	}
485 
486 	return error;
487 }
488 
489 int
490 rioioctl(dev_t dev, ioctlcmd_t cmd, void *addr, int flag,
491     struct lwp *l)
492 {
493 	rio_queue_t *rqp;
494 	struct rio_interface *ifacep;
495 	struct ifnet *ifp;
496 	int	error = 0;
497 
498 	/* check super-user privilege */
499 	switch (cmd) {
500 	case RIO_GETSTATS:
501 		break;
502 	default:
503 #if (__FreeBSD_version > 400000)
504 		if ((error = suser(p)) != 0)
505 			return (error);
506 #else
507 		if ((error = kauth_authorize_network(l->l_cred,
508 		    KAUTH_NETWORK_ALTQ, KAUTH_REQ_NETWORK_ALTQ_RIO, NULL,
509 		    NULL, NULL)) != 0)
510 			return (error);
511 #endif
512 		break;
513 	}
514 
515 	switch (cmd) {
516 
517 	case RIO_ENABLE:
518 		ifacep = (struct rio_interface *)addr;
519 		if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
520 			error = EBADF;
521 			break;
522 		}
523 		error = altq_enable(rqp->rq_ifq);
524 		break;
525 
526 	case RIO_DISABLE:
527 		ifacep = (struct rio_interface *)addr;
528 		if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
529 			error = EBADF;
530 			break;
531 		}
532 		error = altq_disable(rqp->rq_ifq);
533 		break;
534 
535 	case RIO_IF_ATTACH:
536 		ifp = ifunit(((struct rio_interface *)addr)->rio_ifname);
537 		if (ifp == NULL) {
538 			error = ENXIO;
539 			break;
540 		}
541 
542 		/* allocate and initialize rio_queue_t */
543 		rqp = malloc(sizeof(rio_queue_t), M_DEVBUF, M_WAITOK|M_ZERO);
544 		if (rqp == NULL) {
545 			error = ENOMEM;
546 			break;
547 		}
548 
549 		rqp->rq_q = malloc(sizeof(class_queue_t), M_DEVBUF,
550 		    M_WAITOK|M_ZERO);
551 		if (rqp->rq_q == NULL) {
552 			free(rqp, M_DEVBUF);
553 			error = ENOMEM;
554 			break;
555 		}
556 
557 		rqp->rq_rio = rio_alloc(0, NULL, 0, 0);
558 		if (rqp->rq_rio == NULL) {
559 			free(rqp->rq_q, M_DEVBUF);
560 			free(rqp, M_DEVBUF);
561 			error = ENOMEM;
562 			break;
563 		}
564 
565 		rqp->rq_ifq = &ifp->if_snd;
566 		qtail(rqp->rq_q) = NULL;
567 		qlen(rqp->rq_q) = 0;
568 		qlimit(rqp->rq_q) = RIO_LIMIT;
569 		qtype(rqp->rq_q) = Q_RIO;
570 
571 		/*
572 		 * set RIO to this ifnet structure.
573 		 */
574 		error = altq_attach(rqp->rq_ifq, ALTQT_RIO, rqp,
575 				    rio_enqueue, rio_dequeue, rio_request,
576 				    NULL, NULL);
577 		if (error) {
578 			rio_destroy(rqp->rq_rio);
579 			free(rqp->rq_q, M_DEVBUF);
580 			free(rqp, M_DEVBUF);
581 			break;
582 		}
583 
584 		/* add this state to the rio list */
585 		rqp->rq_next = rio_list;
586 		rio_list = rqp;
587 		break;
588 
589 	case RIO_IF_DETACH:
590 		ifacep = (struct rio_interface *)addr;
591 		if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
592 			error = EBADF;
593 			break;
594 		}
595 		error = rio_detach(rqp);
596 		break;
597 
598 	case RIO_GETSTATS:
599 		do {
600 			struct rio_stats *q_stats;
601 			rio_t *rp;
602 			int i;
603 
604 			q_stats = (struct rio_stats *)addr;
605 			if ((rqp = altq_lookup(q_stats->iface.rio_ifname,
606 					       ALTQT_RIO)) == NULL) {
607 				error = EBADF;
608 				break;
609 			}
610 
611 			rp = rqp->rq_rio;
612 
613 			q_stats->q_limit = qlimit(rqp->rq_q);
614 			q_stats->weight	= rp->rio_weight;
615 			q_stats->flags = rp->rio_flags;
616 
617 			for (i = 0; i < RIO_NDROPPREC; i++) {
618 				q_stats->q_len[i] = rp->rio_precstate[i].qlen;
619 				memcpy(&q_stats->q_stats[i], &rp->q_stats[i],
620 				      sizeof(struct redstats));
621 				q_stats->q_stats[i].q_avg =
622 				    rp->rio_precstate[i].avg >> rp->rio_wshift;
623 
624 				q_stats->q_params[i].inv_pmax
625 					= rp->rio_precstate[i].inv_pmax;
626 				q_stats->q_params[i].th_min
627 					= rp->rio_precstate[i].th_min;
628 				q_stats->q_params[i].th_max
629 					= rp->rio_precstate[i].th_max;
630 			}
631 		} while (/*CONSTCOND*/ 0);
632 		break;
633 
634 	case RIO_CONFIG:
635 		do {
636 			struct rio_conf *fc;
637 			rio_t	*new;
638 			int s, limit, i;
639 
640 			fc = (struct rio_conf *)addr;
641 			if ((rqp = altq_lookup(fc->iface.rio_ifname,
642 					       ALTQT_RIO)) == NULL) {
643 				error = EBADF;
644 				break;
645 			}
646 
647 			new = rio_alloc(fc->rio_weight, &fc->q_params[0],
648 					fc->rio_flags, fc->rio_pkttime);
649 			if (new == NULL) {
650 				error = ENOMEM;
651 				break;
652 			}
653 
654 			s = splnet();
655 			_flushq(rqp->rq_q);
656 			limit = fc->rio_limit;
657 			if (limit < fc->q_params[RIO_NDROPPREC-1].th_max)
658 				limit = fc->q_params[RIO_NDROPPREC-1].th_max;
659 			qlimit(rqp->rq_q) = limit;
660 
661 			rio_destroy(rqp->rq_rio);
662 			rqp->rq_rio = new;
663 
664 			splx(s);
665 
666 			/* write back new values */
667 			fc->rio_limit = limit;
668 			for (i = 0; i < RIO_NDROPPREC; i++) {
669 				fc->q_params[i].inv_pmax =
670 					rqp->rq_rio->rio_precstate[i].inv_pmax;
671 				fc->q_params[i].th_min =
672 					rqp->rq_rio->rio_precstate[i].th_min;
673 				fc->q_params[i].th_max =
674 					rqp->rq_rio->rio_precstate[i].th_max;
675 			}
676 		} while (/*CONSTCOND*/ 0);
677 		break;
678 
679 	case RIO_SETDEFAULTS:
680 		do {
681 			struct redparams *rp;
682 			int i;
683 
684 			rp = (struct redparams *)addr;
685 			for (i = 0; i < RIO_NDROPPREC; i++)
686 				default_rio_params[i] = rp[i];
687 		} while (/*CONSTCOND*/ 0);
688 		break;
689 
690 	default:
691 		error = EINVAL;
692 		break;
693 	}
694 
695 	return error;
696 }
697 
698 static int
699 rio_detach(rio_queue_t *rqp)
700 {
701 	rio_queue_t *tmp;
702 	int error = 0;
703 
704 	if (ALTQ_IS_ENABLED(rqp->rq_ifq))
705 		altq_disable(rqp->rq_ifq);
706 
707 	if ((error = altq_detach(rqp->rq_ifq)))
708 		return (error);
709 
710 	if (rio_list == rqp)
711 		rio_list = rqp->rq_next;
712 	else {
713 		for (tmp = rio_list; tmp != NULL; tmp = tmp->rq_next)
714 			if (tmp->rq_next == rqp) {
715 				tmp->rq_next = rqp->rq_next;
716 				break;
717 			}
718 		if (tmp == NULL)
719 			printf("rio_detach: no state found in rio_list!\n");
720 	}
721 
722 	rio_destroy(rqp->rq_rio);
723 	free(rqp->rq_q, M_DEVBUF);
724 	free(rqp, M_DEVBUF);
725 	return (error);
726 }
727 
728 /*
729  * rio support routines
730  */
731 static int
732 rio_request(struct ifaltq *ifq, int req, void *arg)
733 {
734 	rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
735 
736 	switch (req) {
737 	case ALTRQ_PURGE:
738 		_flushq(rqp->rq_q);
739 		if (ALTQ_IS_ENABLED(ifq))
740 			ifq->ifq_len = 0;
741 		break;
742 	}
743 	return (0);
744 }
745 
746 /*
747  * enqueue routine:
748  *
749  *	returns: 0 when successfully queued.
750  *		 ENOBUFS when drop occurs.
751  */
752 static int
753 rio_enqueue(struct ifaltq *ifq, struct mbuf *m)
754 {
755 	struct altq_pktattr pktattr;
756 	rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
757 	int error = 0;
758 
759 	pktattr.pattr_class = m->m_pkthdr.pattr_class;
760 	pktattr.pattr_af = m->m_pkthdr.pattr_af;
761 	pktattr.pattr_hdr = m->m_pkthdr.pattr_hdr;
762 
763 	if (rio_addq(rqp->rq_rio, rqp->rq_q, m, &pktattr) == 0)
764 		ifq->ifq_len++;
765 	else
766 		error = ENOBUFS;
767 	return error;
768 }
769 
770 /*
771  * dequeue routine:
772  *	must be called in splnet.
773  *
774  *	returns: mbuf dequeued.
775  *		 NULL when no packet is available in the queue.
776  */
777 
778 static struct mbuf *
779 rio_dequeue(struct ifaltq *ifq, int op)
780 {
781 	rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
782 	struct mbuf *m = NULL;
783 
784 	if (op == ALTDQ_POLL)
785 		return qhead(rqp->rq_q);
786 
787 	m = rio_getq(rqp->rq_rio, rqp->rq_q);
788 	if (m != NULL)
789 		ifq->ifq_len--;
790 	return m;
791 }
792 
793 #ifdef KLD_MODULE
794 
795 static struct altqsw rio_sw =
796 	{"rio", rioopen, rioclose, rioioctl};
797 
798 ALTQ_MODULE(altq_rio, ALTQT_RIO, &rio_sw);
799 MODULE_VERSION(altq_rio, 1);
800 MODULE_DEPEND(altq_rio, altq_red, 1, 1, 1);
801 
802 #endif /* KLD_MODULE */
803 #endif /* ALTQ3_COMPAT */
804 
805 #endif /* ALTQ_RIO */
806