xref: /netbsd-src/sys/altq/altq_rio.c (revision 413d532bcc3f62d122e56d92e13ac64825a40baf)
1 /*	$NetBSD: altq_rio.c,v 1.21 2009/04/18 14:58:02 tsutsui 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 1.21 2009/04/18 14:58:02 tsutsui 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 *, struct altq_pktattr *);
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 #if 1
330 /*
331  * kludge: when a packet is dequeued, we need to know its drop precedence
332  * in order to keep the queue length of each drop precedence.
333  * use m_pkthdr.rcvif to pass this info.
334  */
335 #define	RIOM_SET_PRECINDEX(m, idx)	\
336 	do { (m)->m_pkthdr.rcvif = (struct ifnet *)((long)(idx)); } while (0)
337 #define	RIOM_GET_PRECINDEX(m)	\
338 	({ long idx; idx = (long)((m)->m_pkthdr.rcvif); \
339 	(m)->m_pkthdr.rcvif = NULL; idx; })
340 #endif
341 
342 int
343 rio_addq(rio_t *rp, class_queue_t *q, struct mbuf *m,
344     struct altq_pktattr *pktattr)
345 {
346 	int			 avg, droptype;
347 	u_int8_t		 dsfield, odsfield;
348 	int			 dpindex, i, n, t;
349 	struct timeval		 now;
350 	struct dropprec_state	*prec;
351 
352 	dsfield = odsfield = read_dsfield(m, pktattr);
353 	dpindex = dscp2index(dsfield);
354 
355 	/*
356 	 * update avg of the precedence states whose drop precedence
357 	 * is larger than or equal to the drop precedence of the packet
358 	 */
359 	now.tv_sec = 0;
360 	for (i = dpindex; i < RIO_NDROPPREC; i++) {
361 		prec = &rp->rio_precstate[i];
362 		avg = prec->avg;
363 		if (prec->idle) {
364 			prec->idle = 0;
365 			if (now.tv_sec == 0)
366 				microtime(&now);
367 			t = (now.tv_sec - prec->last.tv_sec);
368 			if (t > 60)
369 				avg = 0;
370 			else {
371 				t = t * 1000000 +
372 					(now.tv_usec - prec->last.tv_usec);
373 				n = t / rp->rio_pkttime;
374 				/* calculate (avg = (1 - Wq)^n * avg) */
375 				if (n > 0)
376 					avg = (avg >> FP_SHIFT) *
377 						pow_w(rp->rio_wtab, n);
378 			}
379 		}
380 
381 		/* run estimator. (avg is scaled by WEIGHT in fixed-point) */
382 		avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift);
383 		prec->avg = avg;		/* save the new value */
384 		/*
385 		 * count keeps a tally of arriving traffic that has not
386 		 * been dropped.
387 		 */
388 		prec->count++;
389 	}
390 
391 	prec = &rp->rio_precstate[dpindex];
392 	avg = prec->avg;
393 
394 	/* see if we drop early */
395 	droptype = DTYPE_NODROP;
396 	if (avg >= prec->th_min_s && prec->qlen > 1) {
397 		if (avg >= prec->th_max_s) {
398 			/* avg >= th_max: forced drop */
399 			droptype = DTYPE_FORCED;
400 		} else if (prec->old == 0) {
401 			/* first exceeds th_min */
402 			prec->count = 1;
403 			prec->old = 1;
404 		} else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift,
405 				      prec->probd, prec->count)) {
406 			/* unforced drop by red */
407 			droptype = DTYPE_EARLY;
408 		}
409 	} else {
410 		/* avg < th_min */
411 		prec->old = 0;
412 	}
413 
414 	/*
415 	 * if the queue length hits the hard limit, it's a forced drop.
416 	 */
417 	if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q))
418 		droptype = DTYPE_FORCED;
419 
420 	if (droptype != DTYPE_NODROP) {
421 		/* always drop incoming packet (as opposed to randomdrop) */
422 		for (i = dpindex; i < RIO_NDROPPREC; i++)
423 			rp->rio_precstate[i].count = 0;
424 #ifdef RIO_STATS
425 		if (droptype == DTYPE_EARLY)
426 			rp->q_stats[dpindex].drop_unforced++;
427 		else
428 			rp->q_stats[dpindex].drop_forced++;
429 		PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m));
430 #endif
431 		m_freem(m);
432 		return (-1);
433 	}
434 
435 	for (i = dpindex; i < RIO_NDROPPREC; i++)
436 		rp->rio_precstate[i].qlen++;
437 
438 	/* save drop precedence index in mbuf hdr */
439 	RIOM_SET_PRECINDEX(m, dpindex);
440 
441 	if (rp->rio_flags & RIOF_CLEARDSCP)
442 		dsfield &= ~DSCP_MASK;
443 
444 	if (dsfield != odsfield)
445 		write_dsfield(m, pktattr, dsfield);
446 
447 	_addq(q, m);
448 
449 #ifdef RIO_STATS
450 	PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m));
451 #endif
452 	return (0);
453 }
454 
455 struct mbuf *
456 rio_getq(rio_t *rp, class_queue_t *q)
457 {
458 	struct mbuf	*m;
459 	int		 dpindex, i;
460 
461 	if ((m = _getq(q)) == NULL)
462 		return NULL;
463 
464 	dpindex = RIOM_GET_PRECINDEX(m);
465 	for (i = dpindex; i < RIO_NDROPPREC; i++) {
466 		if (--rp->rio_precstate[i].qlen == 0) {
467 			if (rp->rio_precstate[i].idle == 0) {
468 				rp->rio_precstate[i].idle = 1;
469 				microtime(&rp->rio_precstate[i].last);
470 			}
471 		}
472 	}
473 	return (m);
474 }
475 
476 #ifdef ALTQ3_COMPAT
477 int
478 rioopen(dev_t dev, int flag, int fmt,
479     struct lwp *l)
480 {
481 	/* everything will be done when the queueing scheme is attached. */
482 	return 0;
483 }
484 
485 int
486 rioclose(dev_t dev, int flag, int fmt,
487     struct lwp *l)
488 {
489 	rio_queue_t *rqp;
490 	int err, error = 0;
491 
492 	while ((rqp = rio_list) != NULL) {
493 		/* destroy all */
494 		err = rio_detach(rqp);
495 		if (err != 0 && error == 0)
496 			error = err;
497 	}
498 
499 	return error;
500 }
501 
502 int
503 rioioctl(dev_t dev, ioctlcmd_t cmd, void *addr, int flag,
504     struct lwp *l)
505 {
506 	rio_queue_t *rqp;
507 	struct rio_interface *ifacep;
508 	struct ifnet *ifp;
509 	int	error = 0;
510 
511 	/* check super-user privilege */
512 	switch (cmd) {
513 	case RIO_GETSTATS:
514 		break;
515 	default:
516 #if (__FreeBSD_version > 400000)
517 		if ((error = suser(p)) != 0)
518 			return (error);
519 #else
520 		if ((error = kauth_authorize_network(l->l_cred,
521 		    KAUTH_NETWORK_ALTQ, KAUTH_REQ_NETWORK_ALTQ_RIO, NULL,
522 		    NULL, NULL)) != 0)
523 			return (error);
524 #endif
525 		break;
526 	}
527 
528 	switch (cmd) {
529 
530 	case RIO_ENABLE:
531 		ifacep = (struct rio_interface *)addr;
532 		if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
533 			error = EBADF;
534 			break;
535 		}
536 		error = altq_enable(rqp->rq_ifq);
537 		break;
538 
539 	case RIO_DISABLE:
540 		ifacep = (struct rio_interface *)addr;
541 		if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
542 			error = EBADF;
543 			break;
544 		}
545 		error = altq_disable(rqp->rq_ifq);
546 		break;
547 
548 	case RIO_IF_ATTACH:
549 		ifp = ifunit(((struct rio_interface *)addr)->rio_ifname);
550 		if (ifp == NULL) {
551 			error = ENXIO;
552 			break;
553 		}
554 
555 		/* allocate and initialize rio_queue_t */
556 		rqp = malloc(sizeof(rio_queue_t), M_DEVBUF, M_WAITOK|M_ZERO);
557 		if (rqp == NULL) {
558 			error = ENOMEM;
559 			break;
560 		}
561 
562 		rqp->rq_q = malloc(sizeof(class_queue_t), M_DEVBUF,
563 		    M_WAITOK|M_ZERO);
564 		if (rqp->rq_q == NULL) {
565 			free(rqp, M_DEVBUF);
566 			error = ENOMEM;
567 			break;
568 		}
569 
570 		rqp->rq_rio = rio_alloc(0, NULL, 0, 0);
571 		if (rqp->rq_rio == NULL) {
572 			free(rqp->rq_q, M_DEVBUF);
573 			free(rqp, M_DEVBUF);
574 			error = ENOMEM;
575 			break;
576 		}
577 
578 		rqp->rq_ifq = &ifp->if_snd;
579 		qtail(rqp->rq_q) = NULL;
580 		qlen(rqp->rq_q) = 0;
581 		qlimit(rqp->rq_q) = RIO_LIMIT;
582 		qtype(rqp->rq_q) = Q_RIO;
583 
584 		/*
585 		 * set RIO to this ifnet structure.
586 		 */
587 		error = altq_attach(rqp->rq_ifq, ALTQT_RIO, rqp,
588 				    rio_enqueue, rio_dequeue, rio_request,
589 				    NULL, NULL);
590 		if (error) {
591 			rio_destroy(rqp->rq_rio);
592 			free(rqp->rq_q, M_DEVBUF);
593 			free(rqp, M_DEVBUF);
594 			break;
595 		}
596 
597 		/* add this state to the rio list */
598 		rqp->rq_next = rio_list;
599 		rio_list = rqp;
600 		break;
601 
602 	case RIO_IF_DETACH:
603 		ifacep = (struct rio_interface *)addr;
604 		if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
605 			error = EBADF;
606 			break;
607 		}
608 		error = rio_detach(rqp);
609 		break;
610 
611 	case RIO_GETSTATS:
612 		do {
613 			struct rio_stats *q_stats;
614 			rio_t *rp;
615 			int i;
616 
617 			q_stats = (struct rio_stats *)addr;
618 			if ((rqp = altq_lookup(q_stats->iface.rio_ifname,
619 					       ALTQT_RIO)) == NULL) {
620 				error = EBADF;
621 				break;
622 			}
623 
624 			rp = rqp->rq_rio;
625 
626 			q_stats->q_limit = qlimit(rqp->rq_q);
627 			q_stats->weight	= rp->rio_weight;
628 			q_stats->flags = rp->rio_flags;
629 
630 			for (i = 0; i < RIO_NDROPPREC; i++) {
631 				q_stats->q_len[i] = rp->rio_precstate[i].qlen;
632 				memcpy(&q_stats->q_stats[i], &rp->q_stats[i],
633 				      sizeof(struct redstats));
634 				q_stats->q_stats[i].q_avg =
635 				    rp->rio_precstate[i].avg >> rp->rio_wshift;
636 
637 				q_stats->q_params[i].inv_pmax
638 					= rp->rio_precstate[i].inv_pmax;
639 				q_stats->q_params[i].th_min
640 					= rp->rio_precstate[i].th_min;
641 				q_stats->q_params[i].th_max
642 					= rp->rio_precstate[i].th_max;
643 			}
644 		} while (/*CONSTCOND*/ 0);
645 		break;
646 
647 	case RIO_CONFIG:
648 		do {
649 			struct rio_conf *fc;
650 			rio_t	*new;
651 			int s, limit, i;
652 
653 			fc = (struct rio_conf *)addr;
654 			if ((rqp = altq_lookup(fc->iface.rio_ifname,
655 					       ALTQT_RIO)) == NULL) {
656 				error = EBADF;
657 				break;
658 			}
659 
660 			new = rio_alloc(fc->rio_weight, &fc->q_params[0],
661 					fc->rio_flags, fc->rio_pkttime);
662 			if (new == NULL) {
663 				error = ENOMEM;
664 				break;
665 			}
666 
667 			s = splnet();
668 			_flushq(rqp->rq_q);
669 			limit = fc->rio_limit;
670 			if (limit < fc->q_params[RIO_NDROPPREC-1].th_max)
671 				limit = fc->q_params[RIO_NDROPPREC-1].th_max;
672 			qlimit(rqp->rq_q) = limit;
673 
674 			rio_destroy(rqp->rq_rio);
675 			rqp->rq_rio = new;
676 
677 			splx(s);
678 
679 			/* write back new values */
680 			fc->rio_limit = limit;
681 			for (i = 0; i < RIO_NDROPPREC; i++) {
682 				fc->q_params[i].inv_pmax =
683 					rqp->rq_rio->rio_precstate[i].inv_pmax;
684 				fc->q_params[i].th_min =
685 					rqp->rq_rio->rio_precstate[i].th_min;
686 				fc->q_params[i].th_max =
687 					rqp->rq_rio->rio_precstate[i].th_max;
688 			}
689 		} while (/*CONSTCOND*/ 0);
690 		break;
691 
692 	case RIO_SETDEFAULTS:
693 		do {
694 			struct redparams *rp;
695 			int i;
696 
697 			rp = (struct redparams *)addr;
698 			for (i = 0; i < RIO_NDROPPREC; i++)
699 				default_rio_params[i] = rp[i];
700 		} while (/*CONSTCOND*/ 0);
701 		break;
702 
703 	default:
704 		error = EINVAL;
705 		break;
706 	}
707 
708 	return error;
709 }
710 
711 static int
712 rio_detach(rio_queue_t *rqp)
713 {
714 	rio_queue_t *tmp;
715 	int error = 0;
716 
717 	if (ALTQ_IS_ENABLED(rqp->rq_ifq))
718 		altq_disable(rqp->rq_ifq);
719 
720 	if ((error = altq_detach(rqp->rq_ifq)))
721 		return (error);
722 
723 	if (rio_list == rqp)
724 		rio_list = rqp->rq_next;
725 	else {
726 		for (tmp = rio_list; tmp != NULL; tmp = tmp->rq_next)
727 			if (tmp->rq_next == rqp) {
728 				tmp->rq_next = rqp->rq_next;
729 				break;
730 			}
731 		if (tmp == NULL)
732 			printf("rio_detach: no state found in rio_list!\n");
733 	}
734 
735 	rio_destroy(rqp->rq_rio);
736 	free(rqp->rq_q, M_DEVBUF);
737 	free(rqp, M_DEVBUF);
738 	return (error);
739 }
740 
741 /*
742  * rio support routines
743  */
744 static int
745 rio_request(struct ifaltq *ifq, int req, void *arg)
746 {
747 	rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
748 
749 	switch (req) {
750 	case ALTRQ_PURGE:
751 		_flushq(rqp->rq_q);
752 		if (ALTQ_IS_ENABLED(ifq))
753 			ifq->ifq_len = 0;
754 		break;
755 	}
756 	return (0);
757 }
758 
759 /*
760  * enqueue routine:
761  *
762  *	returns: 0 when successfully queued.
763  *		 ENOBUFS when drop occurs.
764  */
765 static int
766 rio_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr)
767 {
768 	rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
769 	int error = 0;
770 
771 	if (rio_addq(rqp->rq_rio, rqp->rq_q, m, pktattr) == 0)
772 		ifq->ifq_len++;
773 	else
774 		error = ENOBUFS;
775 	return error;
776 }
777 
778 /*
779  * dequeue routine:
780  *	must be called in splnet.
781  *
782  *	returns: mbuf dequeued.
783  *		 NULL when no packet is available in the queue.
784  */
785 
786 static struct mbuf *
787 rio_dequeue(struct ifaltq *ifq, int op)
788 {
789 	rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
790 	struct mbuf *m = NULL;
791 
792 	if (op == ALTDQ_POLL)
793 		return qhead(rqp->rq_q);
794 
795 	m = rio_getq(rqp->rq_rio, rqp->rq_q);
796 	if (m != NULL)
797 		ifq->ifq_len--;
798 	return m;
799 }
800 
801 #ifdef KLD_MODULE
802 
803 static struct altqsw rio_sw =
804 	{"rio", rioopen, rioclose, rioioctl};
805 
806 ALTQ_MODULE(altq_rio, ALTQT_RIO, &rio_sw);
807 MODULE_VERSION(altq_rio, 1);
808 MODULE_DEPEND(altq_rio, altq_red, 1, 1, 1);
809 
810 #endif /* KLD_MODULE */
811 #endif /* ALTQ3_COMPAT */
812 
813 #endif /* ALTQ_RIO */
814