2 /* $KAME: altq_rio.c,v 1.17 2003/07/10 12:07:49 kjc Exp $ */
5 * Copyright (C) 1998-2003
6 * Sony Computer Science Laboratories Inc. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
30 * Copyright (c) 1990-1994 Regents of the University of California.
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
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.
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
62 #if defined(__FreeBSD__) || defined(__NetBSD__)
66 #include "opt_inet6.h"
68 #endif /* __FreeBSD__ || __NetBSD__ */
69 #ifdef ALTQ_RIO /* rio is enabled by ALTQ_RIO option in opt_altq.h */
71 #include <sys/param.h>
72 #include <sys/malloc.h>
74 #include <sys/socket.h>
75 #include <sys/systm.h>
76 #include <sys/errno.h>
77 #if 1 /* ALTQ3_COMPAT */
79 #include <sys/sockio.h>
80 #include <sys/kernel.h>
85 #include <netinet/in.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/ip.h>
89 #include <netinet/ip6.h>
92 #include <net/pfvar.h>
93 #include <altq/altq.h>
94 #include <altq/altq_cdnr.h>
95 #include <altq/altq_red.h>
96 #include <altq/altq_rio.h>
98 #include <altq/altq_conf.h>
102 * RIO: RED with IN/OUT bit
104 * "Explicit Allocation of Best Effort Packet Delivery Service"
105 * David D. Clark and Wenjia Fang, MIT Lab for Computer Science
106 * http://diffserv.lcs.mit.edu/Papers/exp-alloc-ddc-wf.{ps,pdf}
108 * this implementation is extended to support more than 2 drop precedence
109 * values as described in RFC2597 (Assured Forwarding PHB Group).
113 * AF DS (differentiated service) codepoints.
114 * (classes can be mapped to CBQ or H-FSC classes.)
117 * +---+---+---+---+---+---+---+---+
118 * | CLASS |DropPre| 0 | CU |
119 * +---+---+---+---+---+---+---+---+
127 * medium drop prec: 10
131 /* normal red parameters */
132 #define W_WEIGHT 512 /* inverse of weight of EWMA (511/512) */
133 /* q_weight = 0.00195 */
135 /* red parameters for a slow link */
136 #define W_WEIGHT_1 128 /* inverse of weight of EWMA (127/128) */
137 /* q_weight = 0.0078125 */
139 /* red parameters for a very slow link (e.g., dialup) */
140 #define W_WEIGHT_2 64 /* inverse of weight of EWMA (63/64) */
141 /* q_weight = 0.015625 */
143 /* fixed-point uses 12-bit decimal places */
144 #define FP_SHIFT 12 /* fixed-point shift */
146 /* red parameters for drop probability */
147 #define INV_P_MAX 10 /* inverse of max drop probability */
148 #define TH_MIN 5 /* min threshold */
149 #define TH_MAX 15 /* max threshold */
151 #define RIO_LIMIT 60 /* default max queue lenght */
152 #define RIO_STATS /* collect statistics */
154 #define TV_DELTA(a, b, delta) { \
157 delta = (a)->tv_usec - (b)->tv_usec; \
158 if ((xxs = (a)->tv_sec - (b)->tv_sec) != 0) { \
161 } else if (xxs > 4) { \
165 delta += xxs * 1000000; \
166 } else while (xxs > 0) { \
174 /* rio_list keeps all rio_queue_t's allocated. */
175 static rio_queue_t *rio_list = NULL;
177 /* default rio parameter values */
178 static struct redparams default_rio_params[RIO_NDROPPREC] = {
179 /* th_min, th_max, inv_pmax */
180 { TH_MAX * 2 + TH_MIN, TH_MAX * 3, INV_P_MAX }, /* low drop precedence */
181 { TH_MAX + TH_MIN, TH_MAX * 2, INV_P_MAX }, /* medium drop precedence */
182 { TH_MIN, TH_MAX, INV_P_MAX } /* high drop precedence */
185 /* internal function prototypes */
186 static int dscp2index(u_int8_t);
188 static int rio_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
189 static struct mbuf *rio_dequeue(struct ifaltq *, int);
190 static int rio_request(struct ifaltq *, int, void *);
191 static int rio_detach(rio_queue_t *);
194 * rio device interface
198 #endif /* ALTQ3_COMPAT */
201 rio_alloc(int weight, struct redparams *params, int flags, int pkttime)
207 rp = malloc(sizeof(rio_t), M_DEVBUF, M_WAITOK);
210 bzero(rp, sizeof(rio_t));
212 rp->rio_flags = flags;
214 /* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */
215 rp->rio_pkttime = 800;
217 rp->rio_pkttime = pkttime;
220 rp->rio_weight = weight;
223 rp->rio_weight = W_WEIGHT;
225 /* when the link is very slow, adjust red parameters */
226 npkts_per_sec = 1000000 / rp->rio_pkttime;
227 if (npkts_per_sec < 50) {
228 /* up to about 400Kbps */
229 rp->rio_weight = W_WEIGHT_2;
230 } else if (npkts_per_sec < 300) {
231 /* up to about 2.4Mbps */
232 rp->rio_weight = W_WEIGHT_1;
236 /* calculate wshift. weight must be power of 2 */
238 for (i = 0; w > 1; i++)
241 w = 1 << rp->rio_wshift;
242 if (w != rp->rio_weight) {
243 printf("invalid weight value %d for red! use %d\n",
248 /* allocate weight table */
249 rp->rio_wtab = wtab_alloc(rp->rio_weight);
251 for (i = 0; i < RIO_NDROPPREC; i++) {
252 struct dropprec_state *prec = &rp->rio_precstate[i];
257 if (params == NULL || params[i].inv_pmax == 0)
258 prec->inv_pmax = default_rio_params[i].inv_pmax;
260 prec->inv_pmax = params[i].inv_pmax;
261 if (params == NULL || params[i].th_min == 0)
262 prec->th_min = default_rio_params[i].th_min;
264 prec->th_min = params[i].th_min;
265 if (params == NULL || params[i].th_max == 0)
266 prec->th_max = default_rio_params[i].th_max;
268 prec->th_max = params[i].th_max;
271 * th_min_s and th_max_s are scaled versions of th_min
272 * and th_max to be compared with avg.
274 prec->th_min_s = prec->th_min << (rp->rio_wshift + FP_SHIFT);
275 prec->th_max_s = prec->th_max << (rp->rio_wshift + FP_SHIFT);
278 * precompute probability denominator
279 * probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point
281 prec->probd = (2 * (prec->th_max - prec->th_min)
282 * prec->inv_pmax) << FP_SHIFT;
284 microtime(&prec->last);
291 rio_destroy(rio_t *rp)
293 wtab_destroy(rp->rio_wtab);
298 rio_getstats(rio_t *rp, struct redstats *sp)
302 for (i = 0; i < RIO_NDROPPREC; i++) {
303 bcopy(&rp->q_stats[i], sp, sizeof(struct redstats));
304 sp->q_avg = rp->rio_precstate[i].avg >> rp->rio_wshift;
309 #if (RIO_NDROPPREC == 3)
311 * internally, a drop precedence value is converted to an index
315 dscp2index(u_int8_t dscp)
317 int dpindex = dscp & AF_DROPPRECMASK;
321 return ((dpindex >> 3) - 1);
327 * kludge: when a packet is dequeued, we need to know its drop precedence
328 * in order to keep the queue length of each drop precedence.
329 * use m_pkthdr.rcvif to pass this info.
331 #define RIOM_SET_PRECINDEX(m, idx) \
332 do { (m)->m_pkthdr.rcvif = (void *)((long)(idx)); } while (0)
333 #define RIOM_GET_PRECINDEX(m) \
334 ({ long idx; idx = (long)((m)->m_pkthdr.rcvif); \
335 (m)->m_pkthdr.rcvif = NULL; idx; })
339 rio_addq(rio_t *rp, class_queue_t *q, struct mbuf *m,
340 struct altq_pktattr *pktattr)
343 u_int8_t dsfield, odsfield;
344 int dpindex, i, n, t;
346 struct dropprec_state *prec;
348 dsfield = odsfield = read_dsfield(m, pktattr);
349 dpindex = dscp2index(dsfield);
352 * update avg of the precedence states whose drop precedence
353 * is larger than or equal to the drop precedence of the packet
356 for (i = dpindex; i < RIO_NDROPPREC; i++) {
357 prec = &rp->rio_precstate[i];
363 t = (now.tv_sec - prec->last.tv_sec);
368 (now.tv_usec - prec->last.tv_usec);
369 n = t / rp->rio_pkttime;
370 /* calculate (avg = (1 - Wq)^n * avg) */
372 avg = (avg >> FP_SHIFT) *
373 pow_w(rp->rio_wtab, n);
377 /* run estimator. (avg is scaled by WEIGHT in fixed-point) */
378 avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift);
379 prec->avg = avg; /* save the new value */
381 * count keeps a tally of arriving traffic that has not
387 prec = &rp->rio_precstate[dpindex];
390 /* see if we drop early */
391 droptype = DTYPE_NODROP;
392 if (avg >= prec->th_min_s && prec->qlen > 1) {
393 if (avg >= prec->th_max_s) {
394 /* avg >= th_max: forced drop */
395 droptype = DTYPE_FORCED;
396 } else if (prec->old == 0) {
397 /* first exceeds th_min */
400 } else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift,
401 prec->probd, prec->count)) {
402 /* unforced drop by red */
403 droptype = DTYPE_EARLY;
411 * if the queue length hits the hard limit, it's a forced drop.
413 if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q))
414 droptype = DTYPE_FORCED;
416 if (droptype != DTYPE_NODROP) {
417 /* always drop incoming packet (as opposed to randomdrop) */
418 for (i = dpindex; i < RIO_NDROPPREC; i++)
419 rp->rio_precstate[i].count = 0;
421 if (droptype == DTYPE_EARLY)
422 rp->q_stats[dpindex].drop_unforced++;
424 rp->q_stats[dpindex].drop_forced++;
425 PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m));
431 for (i = dpindex; i < RIO_NDROPPREC; i++)
432 rp->rio_precstate[i].qlen++;
434 /* save drop precedence index in mbuf hdr */
435 RIOM_SET_PRECINDEX(m, dpindex);
437 if (rp->rio_flags & RIOF_CLEARDSCP)
438 dsfield &= ~DSCP_MASK;
440 if (dsfield != odsfield)
441 write_dsfield(m, pktattr, dsfield);
446 PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m));
452 rio_getq(rio_t *rp, class_queue_t *q)
457 if ((m = _getq(q)) == NULL)
460 dpindex = RIOM_GET_PRECINDEX(m);
461 for (i = dpindex; i < RIO_NDROPPREC; i++) {
462 if (--rp->rio_precstate[i].qlen == 0) {
463 if (rp->rio_precstate[i].idle == 0) {
464 rp->rio_precstate[i].idle = 1;
465 microtime(&rp->rio_precstate[i].last);
474 rioopen(dev, flag, fmt, p)
477 #if (__FreeBSD_version > 500000)
483 /* everything will be done when the queueing scheme is attached. */
488 rioclose(dev, flag, fmt, p)
491 #if (__FreeBSD_version > 500000)
500 while ((rqp = rio_list) != NULL) {
502 err = rio_detach(rqp);
503 if (err != 0 && error == 0)
511 rioioctl(dev, cmd, addr, flag, p)
516 #if (__FreeBSD_version > 500000)
523 struct rio_interface *ifacep;
527 /* check super-user privilege */
532 #if (__FreeBSD_version > 700000)
533 if ((error = priv_check(p, PRIV_ALTQ_MANAGE)) != 0)
535 #elsif (__FreeBSD_version > 400000)
536 if ((error = suser(p)) != 0)
539 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
548 ifacep = (struct rio_interface *)addr;
549 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
553 error = altq_enable(rqp->rq_ifq);
557 ifacep = (struct rio_interface *)addr;
558 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
562 error = altq_disable(rqp->rq_ifq);
566 ifp = ifunit(((struct rio_interface *)addr)->rio_ifname);
572 /* allocate and initialize rio_queue_t */
573 rqp = malloc(sizeof(rio_queue_t), M_DEVBUF, M_WAITOK);
578 bzero(rqp, sizeof(rio_queue_t));
580 rqp->rq_q = malloc(sizeof(class_queue_t),
582 if (rqp->rq_q == NULL) {
587 bzero(rqp->rq_q, sizeof(class_queue_t));
589 rqp->rq_rio = rio_alloc(0, NULL, 0, 0);
590 if (rqp->rq_rio == NULL) {
591 free(rqp->rq_q, M_DEVBUF);
597 rqp->rq_ifq = &ifp->if_snd;
598 qtail(rqp->rq_q) = NULL;
600 qlimit(rqp->rq_q) = RIO_LIMIT;
601 qtype(rqp->rq_q) = Q_RIO;
604 * set RIO to this ifnet structure.
606 error = altq_attach(rqp->rq_ifq, ALTQT_RIO, rqp,
607 rio_enqueue, rio_dequeue, rio_request,
610 rio_destroy(rqp->rq_rio);
611 free(rqp->rq_q, M_DEVBUF);
616 /* add this state to the rio list */
617 rqp->rq_next = rio_list;
622 ifacep = (struct rio_interface *)addr;
623 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) {
627 error = rio_detach(rqp);
632 struct rio_stats *q_stats;
636 q_stats = (struct rio_stats *)addr;
637 if ((rqp = altq_lookup(q_stats->iface.rio_ifname,
638 ALTQT_RIO)) == NULL) {
645 q_stats->q_limit = qlimit(rqp->rq_q);
646 q_stats->weight = rp->rio_weight;
647 q_stats->flags = rp->rio_flags;
649 for (i = 0; i < RIO_NDROPPREC; i++) {
650 q_stats->q_len[i] = rp->rio_precstate[i].qlen;
651 bcopy(&rp->q_stats[i], &q_stats->q_stats[i],
652 sizeof(struct redstats));
653 q_stats->q_stats[i].q_avg =
654 rp->rio_precstate[i].avg >> rp->rio_wshift;
656 q_stats->q_params[i].inv_pmax
657 = rp->rio_precstate[i].inv_pmax;
658 q_stats->q_params[i].th_min
659 = rp->rio_precstate[i].th_min;
660 q_stats->q_params[i].th_max
661 = rp->rio_precstate[i].th_max;
663 } while (/*CONSTCOND*/ 0);
672 fc = (struct rio_conf *)addr;
673 if ((rqp = altq_lookup(fc->iface.rio_ifname,
674 ALTQT_RIO)) == NULL) {
679 new = rio_alloc(fc->rio_weight, &fc->q_params[0],
680 fc->rio_flags, fc->rio_pkttime);
692 limit = fc->rio_limit;
693 if (limit < fc->q_params[RIO_NDROPPREC-1].th_max)
694 limit = fc->q_params[RIO_NDROPPREC-1].th_max;
695 qlimit(rqp->rq_q) = limit;
697 rio_destroy(rqp->rq_rio);
702 /* write back new values */
703 fc->rio_limit = limit;
704 for (i = 0; i < RIO_NDROPPREC; i++) {
705 fc->q_params[i].inv_pmax =
706 rqp->rq_rio->rio_precstate[i].inv_pmax;
707 fc->q_params[i].th_min =
708 rqp->rq_rio->rio_precstate[i].th_min;
709 fc->q_params[i].th_max =
710 rqp->rq_rio->rio_precstate[i].th_max;
712 } while (/*CONSTCOND*/ 0);
715 case RIO_SETDEFAULTS:
717 struct redparams *rp;
720 rp = (struct redparams *)addr;
721 for (i = 0; i < RIO_NDROPPREC; i++)
722 default_rio_params[i] = rp[i];
723 } while (/*CONSTCOND*/ 0);
741 if (ALTQ_IS_ENABLED(rqp->rq_ifq))
742 altq_disable(rqp->rq_ifq);
744 if ((error = altq_detach(rqp->rq_ifq)))
748 rio_list = rqp->rq_next;
750 for (tmp = rio_list; tmp != NULL; tmp = tmp->rq_next)
751 if (tmp->rq_next == rqp) {
752 tmp->rq_next = rqp->rq_next;
756 printf("rio_detach: no state found in rio_list!\n");
759 rio_destroy(rqp->rq_rio);
760 free(rqp->rq_q, M_DEVBUF);
766 * rio support routines
769 rio_request(ifq, req, arg)
774 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
776 IFQ_LOCK_ASSERT(ifq);
781 if (ALTQ_IS_ENABLED(ifq))
791 * returns: 0 when successfully queued.
792 * ENOBUFS when drop occurs.
795 rio_enqueue(ifq, m, pktattr)
798 struct altq_pktattr *pktattr;
800 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
803 IFQ_LOCK_ASSERT(ifq);
805 if (rio_addq(rqp->rq_rio, rqp->rq_q, m, pktattr) == 0)
814 * must be called in splimp.
816 * returns: mbuf dequeued.
817 * NULL when no packet is available in the queue.
825 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc;
826 struct mbuf *m = NULL;
828 IFQ_LOCK_ASSERT(ifq);
830 if (op == ALTDQ_POLL)
831 return qhead(rqp->rq_q);
833 m = rio_getq(rqp->rq_rio, rqp->rq_q);
841 static struct altqsw rio_sw =
842 {"rio", rioopen, rioclose, rioioctl};
844 ALTQ_MODULE(altq_rio, ALTQT_RIO, &rio_sw);
845 MODULE_VERSION(altq_rio, 1);
846 MODULE_DEPEND(altq_rio, altq_red, 1, 1, 1);
848 #endif /* KLD_MODULE */
849 #endif /* ALTQ3_COMPAT */
851 #endif /* ALTQ_RIO */