xref: /netbsd-src/sys/altq/altq_red.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: altq_red.c,v 1.26 2007/03/26 22:43:19 hubertf Exp $	*/
2 /*	$KAME: altq_red.c,v 1.20 2005/04/13 03:44:25 suz Exp $	*/
3 
4 /*
5  * Copyright (C) 1997-2003
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 /*
31  * Copyright (c) 1990-1994 Regents of the University of California.
32  * All rights reserved.
33  *
34  * Redistribution and use in source and binary forms, with or without
35  * modification, are permitted provided that the following conditions
36  * are met:
37  * 1. Redistributions of source code must retain the above copyright
38  *    notice, this list of conditions and the following disclaimer.
39  * 2. Redistributions in binary form must reproduce the above copyright
40  *    notice, this list of conditions and the following disclaimer in the
41  *    documentation and/or other materials provided with the distribution.
42  * 3. All advertising materials mentioning features or use of this software
43  *    must display the following acknowledgement:
44  *	This product includes software developed by the Computer Systems
45  *	Engineering Group at Lawrence Berkeley Laboratory.
46  * 4. Neither the name of the University nor of the Laboratory may be used
47  *    to endorse or promote products derived from this software without
48  *    specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  */
62 
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: altq_red.c,v 1.26 2007/03/26 22:43:19 hubertf Exp $");
65 
66 #ifdef _KERNEL_OPT
67 #include "opt_altq.h"
68 #include "opt_inet.h"
69 #include "pf.h"
70 #endif
71 
72 #ifdef ALTQ_RED	/* red is enabled by ALTQ_RED option in opt_altq.h */
73 
74 #include <sys/param.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/systm.h>
79 #include <sys/errno.h>
80 #include <sys/kauth.h>
81 #if 1 /* ALTQ3_COMPAT */
82 #include <sys/sockio.h>
83 #include <sys/proc.h>
84 #include <sys/kernel.h>
85 #ifdef ALTQ_FLOWVALVE
86 #include <sys/queue.h>
87 #include <sys/time.h>
88 #endif
89 #endif /* ALTQ3_COMPAT */
90 
91 #include <net/if.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #ifdef INET6
97 #include <netinet/ip6.h>
98 #endif
99 
100 #if NPF > 0
101 #include <net/pfvar.h>
102 #endif
103 #include <altq/altq.h>
104 #include <altq/altq_red.h>
105 #ifdef ALTQ3_COMPAT
106 #include <altq/altq_conf.h>
107 #ifdef ALTQ_FLOWVALVE
108 #include <altq/altq_flowvalve.h>
109 #endif
110 #endif
111 
112 /*
113  * ALTQ/RED (Random Early Detection) implementation using 32-bit
114  * fixed-point calculation.
115  *
116  * written by kjc using the ns code as a reference.
117  * you can learn more about red and ns from Sally's home page at
118  * http://www-nrg.ee.lbl.gov/floyd/
119  *
120  * most of the red parameter values are fixed in this implementation
121  * to prevent fixed-point overflow/underflow.
122  * if you change the parameters, watch out for overflow/underflow!
123  *
124  * the parameters used are recommended values by Sally.
125  * the corresponding ns config looks:
126  *	q_weight=0.00195
127  *	minthresh=5 maxthresh=15 queue-size=60
128  *	linterm=30
129  *	dropmech=drop-tail
130  *	bytes=false (can't be handled by 32-bit fixed-point)
131  *	doubleq=false dqthresh=false
132  *	wait=true
133  */
134 /*
135  * alternative red parameters for a slow link.
136  *
137  * assume the queue length becomes from zero to L and keeps L, it takes
138  * N packets for q_avg to reach 63% of L.
139  * when q_weight is 0.002, N is about 500 packets.
140  * for a slow link like dial-up, 500 packets takes more than 1 minute!
141  * when q_weight is 0.008, N is about 127 packets.
142  * when q_weight is 0.016, N is about 63 packets.
143  * bursts of 50 packets are allowed for 0.002, bursts of 25 packets
144  * are allowed for 0.016.
145  * see Sally's paper for more details.
146  */
147 /* normal red parameters */
148 #define	W_WEIGHT	512	/* inverse of weight of EWMA (511/512) */
149 				/* q_weight = 0.00195 */
150 
151 /* red parameters for a slow link */
152 #define	W_WEIGHT_1	128	/* inverse of weight of EWMA (127/128) */
153 				/* q_weight = 0.0078125 */
154 
155 /* red parameters for a very slow link (e.g., dialup) */
156 #define	W_WEIGHT_2	64	/* inverse of weight of EWMA (63/64) */
157 				/* q_weight = 0.015625 */
158 
159 /* fixed-point uses 12-bit decimal places */
160 #define	FP_SHIFT	12	/* fixed-point shift */
161 
162 /* red parameters for drop probability */
163 #define	INV_P_MAX	10	/* inverse of max drop probability */
164 #define	TH_MIN		5	/* min threshold */
165 #define	TH_MAX		15	/* max threshold */
166 
167 #define	RED_LIMIT	60	/* default max queue lenght */
168 #define	RED_STATS		/* collect statistics */
169 
170 /*
171  * our default policy for forced-drop is drop-tail.
172  * (in altq-1.1.2 or earlier, the default was random-drop.
173  * but it makes more sense to punish the cause of the surge.)
174  * to switch to the random-drop policy, define "RED_RANDOM_DROP".
175  */
176 
177 #ifdef ALTQ3_COMPAT
178 #ifdef ALTQ_FLOWVALVE
179 /*
180  * flow-valve is an extention to protect red from unresponsive flows
181  * and to promote end-to-end congestion control.
182  * flow-valve observes the average drop rates of the flows that have
183  * experienced packet drops in the recent past.
184  * when the average drop rate exceeds the threshold, the flow is
185  * blocked by the flow-valve.  the trapped flow should back off
186  * exponentially to escape from the flow-valve.
187  */
188 #ifdef RED_RANDOM_DROP
189 #error "random-drop can't be used with flow-valve!"
190 #endif
191 #endif /* ALTQ_FLOWVALVE */
192 
193 /* red_list keeps all red_queue_t's allocated. */
194 static red_queue_t *red_list = NULL;
195 
196 #endif /* ALTQ3_COMPAT */
197 
198 /* default red parameter values */
199 static int default_th_min = TH_MIN;
200 static int default_th_max = TH_MAX;
201 static int default_inv_pmax = INV_P_MAX;
202 
203 #ifdef ALTQ3_COMPAT
204 /* internal function prototypes */
205 static int red_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
206 static struct mbuf *red_dequeue(struct ifaltq *, int);
207 static int red_request(struct ifaltq *, int, void *);
208 static void red_purgeq(red_queue_t *);
209 static int red_detach(red_queue_t *);
210 #ifdef ALTQ_FLOWVALVE
211 static inline struct fve *flowlist_lookup(struct flowvalve *,
212 			 struct altq_pktattr *, struct timeval *);
213 static inline struct fve *flowlist_reclaim(struct flowvalve *,
214 					     struct altq_pktattr *);
215 static inline void flowlist_move_to_head(struct flowvalve *, struct fve *);
216 static inline int fv_p2f(struct flowvalve *, int);
217 static struct flowvalve *fv_alloc(struct red *);
218 static void fv_destroy(struct flowvalve *);
219 static int fv_checkflow(struct flowvalve *, struct altq_pktattr *,
220 			struct fve **);
221 static void fv_dropbyred(struct flowvalve *fv, struct altq_pktattr *,
222 			 struct fve *);
223 #endif
224 #endif /* ALTQ3_COMPAT */
225 
226 /*
227  * red support routines
228  */
229 red_t *
230 red_alloc(int weight, int inv_pmax, int th_min, int th_max, int flags,
231    int pkttime)
232 {
233 	red_t	*rp;
234 	int	 w, i;
235 	int	 npkts_per_sec;
236 
237 	rp = malloc(sizeof(red_t), M_DEVBUF, M_WAITOK|M_ZERO);
238 	if (rp == NULL)
239 		return (NULL);
240 
241 	rp->red_avg = 0;
242 	rp->red_idle = 1;
243 
244 	if (weight == 0)
245 		rp->red_weight = W_WEIGHT;
246 	else
247 		rp->red_weight = weight;
248 	if (inv_pmax == 0)
249 		rp->red_inv_pmax = default_inv_pmax;
250 	else
251 		rp->red_inv_pmax = inv_pmax;
252 	if (th_min == 0)
253 		rp->red_thmin = default_th_min;
254 	else
255 		rp->red_thmin = th_min;
256 	if (th_max == 0)
257 		rp->red_thmax = default_th_max;
258 	else
259 		rp->red_thmax = th_max;
260 
261 	rp->red_flags = flags;
262 
263 	if (pkttime == 0)
264 		/* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */
265 		rp->red_pkttime = 800;
266 	else
267 		rp->red_pkttime = pkttime;
268 
269 	if (weight == 0) {
270 		/* when the link is very slow, adjust red parameters */
271 		npkts_per_sec = 1000000 / rp->red_pkttime;
272 		if (npkts_per_sec < 50) {
273 			/* up to about 400Kbps */
274 			rp->red_weight = W_WEIGHT_2;
275 		} else if (npkts_per_sec < 300) {
276 			/* up to about 2.4Mbps */
277 			rp->red_weight = W_WEIGHT_1;
278 		}
279 	}
280 
281 	/* calculate wshift.  weight must be power of 2 */
282 	w = rp->red_weight;
283 	for (i = 0; w > 1; i++)
284 		w = w >> 1;
285 	rp->red_wshift = i;
286 	w = 1 << rp->red_wshift;
287 	if (w != rp->red_weight) {
288 		printf("invalid weight value %d for red! use %d\n",
289 		       rp->red_weight, w);
290 		rp->red_weight = w;
291 	}
292 
293 	/*
294 	 * thmin_s and thmax_s are scaled versions of th_min and th_max
295 	 * to be compared with avg.
296 	 */
297 	rp->red_thmin_s = rp->red_thmin << (rp->red_wshift + FP_SHIFT);
298 	rp->red_thmax_s = rp->red_thmax << (rp->red_wshift + FP_SHIFT);
299 
300 	/*
301 	 * precompute probability denominator
302 	 *  probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point
303 	 */
304 	rp->red_probd = (2 * (rp->red_thmax - rp->red_thmin)
305 			 * rp->red_inv_pmax) << FP_SHIFT;
306 
307 	/* allocate weight table */
308 	rp->red_wtab = wtab_alloc(rp->red_weight);
309 
310 	microtime(&rp->red_last);
311 #ifdef ALTQ3_COMPAT
312 #ifdef ALTQ_FLOWVALVE
313 	if (flags & REDF_FLOWVALVE)
314 		rp->red_flowvalve = fv_alloc(rp);
315 	/* if fv_alloc failes, flowvalve is just disabled */
316 #endif
317 #endif /* ALTQ3_COMPAT */
318 	return (rp);
319 }
320 
321 void
322 red_destroy(red_t *rp)
323 {
324 #ifdef ALTQ3_COMPAT
325 #ifdef ALTQ_FLOWVALVE
326 	if (rp->red_flowvalve != NULL)
327 		fv_destroy(rp->red_flowvalve);
328 #endif
329 #endif /* ALTQ3_COMPAT */
330 	wtab_destroy(rp->red_wtab);
331 	free(rp, M_DEVBUF);
332 }
333 
334 void
335 red_getstats(red_t *rp, struct redstats *sp)
336 {
337 	sp->q_avg		= rp->red_avg >> rp->red_wshift;
338 	sp->xmit_cnt		= rp->red_stats.xmit_cnt;
339 	sp->drop_cnt		= rp->red_stats.drop_cnt;
340 	sp->drop_forced		= rp->red_stats.drop_forced;
341 	sp->drop_unforced	= rp->red_stats.drop_unforced;
342 	sp->marked_packets	= rp->red_stats.marked_packets;
343 }
344 
345 int
346 red_addq(red_t *rp, class_queue_t *q, struct mbuf *m,
347     struct altq_pktattr *pktattr)
348 {
349 	int avg, droptype;
350 	int n;
351 #ifdef ALTQ3_COMPAT
352 #ifdef ALTQ_FLOWVALVE
353 	struct fve *fve = NULL;
354 
355 	if (rp->red_flowvalve != NULL && rp->red_flowvalve->fv_flows > 0)
356 		if (fv_checkflow(rp->red_flowvalve, pktattr, &fve)) {
357 			m_freem(m);
358 			return (-1);
359 		}
360 #endif
361 #endif /* ALTQ3_COMPAT */
362 
363 	avg = rp->red_avg;
364 
365 	/*
366 	 * if we were idle, we pretend that n packets arrived during
367 	 * the idle period.
368 	 */
369 	if (rp->red_idle) {
370 		struct timeval now;
371 		int t;
372 
373 		rp->red_idle = 0;
374 		microtime(&now);
375 		t = (now.tv_sec - rp->red_last.tv_sec);
376 		if (t > 60) {
377 			/*
378 			 * being idle for more than 1 minute, set avg to zero.
379 			 * this prevents t from overflow.
380 			 */
381 			avg = 0;
382 		} else {
383 			t = t * 1000000 + (now.tv_usec - rp->red_last.tv_usec);
384 			n = t / rp->red_pkttime - 1;
385 
386 			/* the following line does (avg = (1 - Wq)^n * avg) */
387 			if (n > 0)
388 				avg = (avg >> FP_SHIFT) *
389 				    pow_w(rp->red_wtab, n);
390 		}
391 	}
392 
393 	/* run estimator. (note: avg is scaled by WEIGHT in fixed-point) */
394 	avg += (qlen(q) << FP_SHIFT) - (avg >> rp->red_wshift);
395 	rp->red_avg = avg;		/* save the new value */
396 
397 	/*
398 	 * red_count keeps a tally of arriving traffic that has not
399 	 * been dropped.
400 	 */
401 	rp->red_count++;
402 
403 	/* see if we drop early */
404 	droptype = DTYPE_NODROP;
405 	if (avg >= rp->red_thmin_s && qlen(q) > 1) {
406 		if (avg >= rp->red_thmax_s) {
407 			/* avg >= th_max: forced drop */
408 			droptype = DTYPE_FORCED;
409 		} else if (rp->red_old == 0) {
410 			/* first exceeds th_min */
411 			rp->red_count = 1;
412 			rp->red_old = 1;
413 		} else if (drop_early((avg - rp->red_thmin_s) >> rp->red_wshift,
414 				      rp->red_probd, rp->red_count)) {
415 			/* mark or drop by red */
416 			if ((rp->red_flags & REDF_ECN) &&
417 			    mark_ecn(m, pktattr, rp->red_flags)) {
418 				/* successfully marked.  do not drop. */
419 				rp->red_count = 0;
420 #ifdef RED_STATS
421 				rp->red_stats.marked_packets++;
422 #endif
423 			} else {
424 				/* unforced drop by red */
425 				droptype = DTYPE_EARLY;
426 			}
427 		}
428 	} else {
429 		/* avg < th_min */
430 		rp->red_old = 0;
431 	}
432 
433 	/*
434 	 * if the queue length hits the hard limit, it's a forced drop.
435 	 */
436 	if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q))
437 		droptype = DTYPE_FORCED;
438 
439 #ifdef RED_RANDOM_DROP
440 	/* if successful or forced drop, enqueue this packet. */
441 	if (droptype != DTYPE_EARLY)
442 		_addq(q, m);
443 #else
444 	/* if successful, enqueue this packet. */
445 	if (droptype == DTYPE_NODROP)
446 		_addq(q, m);
447 #endif
448 	if (droptype != DTYPE_NODROP) {
449 		if (droptype == DTYPE_EARLY) {
450 			/* drop the incoming packet */
451 #ifdef RED_STATS
452 			rp->red_stats.drop_unforced++;
453 #endif
454 		} else {
455 			/* forced drop, select a victim packet in the queue. */
456 #ifdef RED_RANDOM_DROP
457 			m = _getq_random(q);
458 #endif
459 #ifdef RED_STATS
460 			rp->red_stats.drop_forced++;
461 #endif
462 		}
463 #ifdef RED_STATS
464 		PKTCNTR_ADD(&rp->red_stats.drop_cnt, m_pktlen(m));
465 #endif
466 		rp->red_count = 0;
467 #ifdef ALTQ3_COMPAT
468 #ifdef ALTQ_FLOWVALVE
469 		if (rp->red_flowvalve != NULL)
470 			fv_dropbyred(rp->red_flowvalve, pktattr, fve);
471 #endif
472 #endif /* ALTQ3_COMPAT */
473 		m_freem(m);
474 		return (-1);
475 	}
476 	/* successfully queued */
477 #ifdef RED_STATS
478 	PKTCNTR_ADD(&rp->red_stats.xmit_cnt, m_pktlen(m));
479 #endif
480 	return (0);
481 }
482 
483 /*
484  * early-drop probability is calculated as follows:
485  *   prob = p_max * (avg - th_min) / (th_max - th_min)
486  *   prob_a = prob / (2 - count*prob)
487  *	    = (avg-th_min) / (2*(th_max-th_min)*inv_p_max - count*(avg-th_min))
488  * here prob_a increases as successive undrop count increases.
489  * (prob_a starts from prob/2, becomes prob when (count == (1 / prob)),
490  * becomes 1 when (count >= (2 / prob))).
491  */
492 int
493 drop_early(int fp_len, int fp_probd, int count)
494 {
495 	int	d;		/* denominator of drop-probability */
496 
497 	d = fp_probd - count * fp_len;
498 	if (d <= 0)
499 		/* count exceeds the hard limit: drop or mark */
500 		return (1);
501 
502 	/*
503 	 * now the range of d is [1..600] in fixed-point. (when
504 	 * th_max-th_min=10 and p_max=1/30)
505 	 * drop probability = (avg - TH_MIN) / d
506 	 */
507 
508 	if ((arc4random() % d) < fp_len) {
509 		/* drop or mark */
510 		return (1);
511 	}
512 	/* no drop/mark */
513 	return (0);
514 }
515 
516 /*
517  * try to mark CE bit to the packet.
518  *    returns 1 if successfully marked, 0 otherwise.
519  */
520 int
521 mark_ecn(struct mbuf *m, struct altq_pktattr *pktattr, int flags)
522 {
523 	struct mbuf	*m0;
524 	struct m_tag	*t;
525 	struct altq_tag	*at;
526 	void		*hdr;
527 	int		 af;
528 
529 	t = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
530 	if (t != NULL) {
531 		at = (struct altq_tag *)(t + 1);
532 		if (at == NULL)
533 			return (0);
534 		af = at->af;
535 		hdr = at->hdr;
536 #ifdef ALTQ3_COMPAT
537 	} else if (pktattr != NULL) {
538 		af = pktattr->pattr_af;
539 		hdr = pktattr->pattr_hdr;
540 #endif /* ALTQ3_COMPAT */
541 	} else
542 		return (0);
543 
544 	if (af != AF_INET && af != AF_INET6)
545 		return (0);
546 
547 	/* verify that pattr_hdr is within the mbuf data */
548 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
549 		if (((char *)hdr >= m0->m_data) &&
550 		    ((char *)hdr < m0->m_data + m0->m_len))
551 			break;
552 	if (m0 == NULL) {
553 		/* ick, tag info is stale */
554 		return (0);
555 	}
556 
557 	switch (af) {
558 	case AF_INET:
559 		if (flags & REDF_ECN4) {
560 			struct ip *ip = hdr;
561 			u_int8_t otos;
562 			int sum;
563 
564 			if (ip->ip_v != 4)
565 				return (0);	/* version mismatch! */
566 
567 			if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_NOTECT)
568 				return (0);	/* not-ECT */
569 			if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_CE)
570 				return (1);	/* already marked */
571 
572 			/*
573 			 * ecn-capable but not marked,
574 			 * mark CE and update checksum
575 			 */
576 			otos = ip->ip_tos;
577 			ip->ip_tos |= IPTOS_ECN_CE;
578 			/*
579 			 * update checksum (from RFC1624)
580 			 *	   HC' = ~(~HC + ~m + m')
581 			 */
582 			sum = ~ntohs(ip->ip_sum) & 0xffff;
583 			sum += (~otos & 0xffff) + ip->ip_tos;
584 			sum = (sum >> 16) + (sum & 0xffff);
585 			sum += (sum >> 16);  /* add carry */
586 			ip->ip_sum = htons(~sum & 0xffff);
587 			return (1);
588 		}
589 		break;
590 #ifdef INET6
591 	case AF_INET6:
592 		if (flags & REDF_ECN6) {
593 			struct ip6_hdr *ip6 = hdr;
594 			u_int32_t flowlabel;
595 
596 			flowlabel = ntohl(ip6->ip6_flow);
597 			if ((flowlabel >> 28) != 6)
598 				return (0);	/* version mismatch! */
599 			if ((flowlabel & (IPTOS_ECN_MASK << 20)) ==
600 			    (IPTOS_ECN_NOTECT << 20))
601 				return (0);	/* not-ECT */
602 			if ((flowlabel & (IPTOS_ECN_MASK << 20)) ==
603 			    (IPTOS_ECN_CE << 20))
604 				return (1);	/* already marked */
605 			/*
606 			 * ecn-capable but not marked,  mark CE
607 			 */
608 			flowlabel |= (IPTOS_ECN_CE << 20);
609 			ip6->ip6_flow = htonl(flowlabel);
610 			return (1);
611 		}
612 		break;
613 #endif  /* INET6 */
614 	}
615 
616 	/* not marked */
617 	return (0);
618 }
619 
620 struct mbuf *
621 red_getq(red_t *rp, class_queue_t *q)
622 {
623 	struct mbuf *m;
624 
625 	if ((m = _getq(q)) == NULL) {
626 		if (rp->red_idle == 0) {
627 			rp->red_idle = 1;
628 			microtime(&rp->red_last);
629 		}
630 		return NULL;
631 	}
632 
633 	rp->red_idle = 0;
634 	return (m);
635 }
636 
637 /*
638  * helper routine to calibrate avg during idle.
639  * pow_w(wtab, n) returns (1 - Wq)^n in fixed-point
640  * here Wq = 1/weight and the code assumes Wq is close to zero.
641  *
642  * w_tab[n] holds ((1 - Wq)^(2^n)) in fixed-point.
643  */
644 static struct wtab *wtab_list = NULL;	/* pointer to wtab list */
645 
646 struct wtab *
647 wtab_alloc(int weight)
648 {
649 	struct wtab	*w;
650 	int		 i;
651 
652 	for (w = wtab_list; w != NULL; w = w->w_next)
653 		if (w->w_weight == weight) {
654 			w->w_refcount++;
655 			return (w);
656 		}
657 
658 	w = malloc(sizeof(struct wtab), M_DEVBUF, M_WAITOK|M_ZERO);
659 	if (w == NULL)
660 		panic("wtab_alloc: malloc failed!");
661 	w->w_weight = weight;
662 	w->w_refcount = 1;
663 	w->w_next = wtab_list;
664 	wtab_list = w;
665 
666 	/* initialize the weight table */
667 	w->w_tab[0] = ((weight - 1) << FP_SHIFT) / weight;
668 	for (i = 1; i < 32; i++) {
669 		w->w_tab[i] = (w->w_tab[i-1] * w->w_tab[i-1]) >> FP_SHIFT;
670 		if (w->w_tab[i] == 0 && w->w_param_max == 0)
671 			w->w_param_max = 1 << i;
672 	}
673 
674 	return (w);
675 }
676 
677 int
678 wtab_destroy(struct wtab *w)
679 {
680 	struct wtab	*prev;
681 
682 	if (--w->w_refcount > 0)
683 		return (0);
684 
685 	if (wtab_list == w)
686 		wtab_list = w->w_next;
687 	else for (prev = wtab_list; prev->w_next != NULL; prev = prev->w_next)
688 		if (prev->w_next == w) {
689 			prev->w_next = w->w_next;
690 			break;
691 		}
692 
693 	free(w, M_DEVBUF);
694 	return (0);
695 }
696 
697 int32_t
698 pow_w(struct wtab *w, int n)
699 {
700 	int	i, bit;
701 	int32_t	val;
702 
703 	if (n >= w->w_param_max)
704 		return (0);
705 
706 	val = 1 << FP_SHIFT;
707 	if (n <= 0)
708 		return (val);
709 
710 	bit = 1;
711 	i = 0;
712 	while (n) {
713 		if (n & bit) {
714 			val = (val * w->w_tab[i]) >> FP_SHIFT;
715 			n &= ~bit;
716 		}
717 		i++;
718 		bit <<=  1;
719 	}
720 	return (val);
721 }
722 
723 #ifdef ALTQ3_COMPAT
724 /*
725  * red device interface
726  */
727 altqdev_decl(red);
728 
729 int
730 redopen(dev_t dev, int flag, int fmt,
731     struct lwp *l)
732 {
733 	/* everything will be done when the queueing scheme is attached. */
734 	return 0;
735 }
736 
737 int
738 redclose(dev_t dev, int flag, int fmt,
739     struct lwp *l)
740 {
741 	red_queue_t *rqp;
742 	int err, error = 0;
743 
744 	while ((rqp = red_list) != NULL) {
745 		/* destroy all */
746 		err = red_detach(rqp);
747 		if (err != 0 && error == 0)
748 			error = err;
749 	}
750 
751 	return error;
752 }
753 
754 int
755 redioctl(dev_t dev, ioctlcmd_t cmd, void *addr, int flag,
756     struct lwp *l)
757 {
758 	red_queue_t *rqp;
759 	struct red_interface *ifacep;
760 	struct ifnet *ifp;
761 	struct proc *p = l->l_proc;
762 	int	error = 0;
763 
764 	/* check super-user privilege */
765 	switch (cmd) {
766 	case RED_GETSTATS:
767 		break;
768 	default:
769 #if (__FreeBSD_version > 400000)
770 		if ((error = suser(p)) != 0)
771 #else
772 		if ((error = kauth_authorize_network(p->p_cred,
773 		    KAUTH_NETWORK_ALTQ, KAUTH_REQ_NETWORK_ALTQ_RED, NULL,
774 		    NULL, NULL)) != 0)
775 #endif
776 			return (error);
777 		break;
778 	}
779 
780 	switch (cmd) {
781 
782 	case RED_ENABLE:
783 		ifacep = (struct red_interface *)addr;
784 		if ((rqp = altq_lookup(ifacep->red_ifname, ALTQT_RED)) == NULL) {
785 			error = EBADF;
786 			break;
787 		}
788 		error = altq_enable(rqp->rq_ifq);
789 		break;
790 
791 	case RED_DISABLE:
792 		ifacep = (struct red_interface *)addr;
793 		if ((rqp = altq_lookup(ifacep->red_ifname, ALTQT_RED)) == NULL) {
794 			error = EBADF;
795 			break;
796 		}
797 		error = altq_disable(rqp->rq_ifq);
798 		break;
799 
800 	case RED_IF_ATTACH:
801 		ifp = ifunit(((struct red_interface *)addr)->red_ifname);
802 		if (ifp == NULL) {
803 			error = ENXIO;
804 			break;
805 		}
806 
807 		/* allocate and initialize red_queue_t */
808 		rqp = malloc(sizeof(red_queue_t), M_DEVBUF, M_WAITOK|M_ZERO);
809 		if (rqp == NULL) {
810 			error = ENOMEM;
811 			break;
812 		}
813 
814 		rqp->rq_q = malloc(sizeof(class_queue_t), M_DEVBUF,
815 		    M_WAITOK|M_ZERO);
816 		if (rqp->rq_q == NULL) {
817 			free(rqp, M_DEVBUF);
818 			error = ENOMEM;
819 			break;
820 		}
821 
822 		rqp->rq_red = red_alloc(0, 0, 0, 0, 0, 0);
823 		if (rqp->rq_red == NULL) {
824 			free(rqp->rq_q, M_DEVBUF);
825 			free(rqp, M_DEVBUF);
826 			error = ENOMEM;
827 			break;
828 		}
829 
830 		rqp->rq_ifq = &ifp->if_snd;
831 		qtail(rqp->rq_q) = NULL;
832 		qlen(rqp->rq_q) = 0;
833 		qlimit(rqp->rq_q) = RED_LIMIT;
834 		qtype(rqp->rq_q) = Q_RED;
835 
836 		/*
837 		 * set RED to this ifnet structure.
838 		 */
839 		error = altq_attach(rqp->rq_ifq, ALTQT_RED, rqp,
840 				    red_enqueue, red_dequeue, red_request,
841 				    NULL, NULL);
842 		if (error) {
843 			red_destroy(rqp->rq_red);
844 			free(rqp->rq_q, M_DEVBUF);
845 			free(rqp, M_DEVBUF);
846 			break;
847 		}
848 
849 		/* add this state to the red list */
850 		rqp->rq_next = red_list;
851 		red_list = rqp;
852 		break;
853 
854 	case RED_IF_DETACH:
855 		ifacep = (struct red_interface *)addr;
856 		if ((rqp = altq_lookup(ifacep->red_ifname, ALTQT_RED)) == NULL) {
857 			error = EBADF;
858 			break;
859 		}
860 		error = red_detach(rqp);
861 		break;
862 
863 	case RED_GETSTATS:
864 		do {
865 			struct red_stats *q_stats;
866 			red_t *rp;
867 
868 			q_stats = (struct red_stats *)addr;
869 			if ((rqp = altq_lookup(q_stats->iface.red_ifname,
870 					     ALTQT_RED)) == NULL) {
871 				error = EBADF;
872 				break;
873 			}
874 
875 			q_stats->q_len 	   = qlen(rqp->rq_q);
876 			q_stats->q_limit   = qlimit(rqp->rq_q);
877 
878 			rp = rqp->rq_red;
879 			q_stats->q_avg 	   = rp->red_avg >> rp->red_wshift;
880 			q_stats->xmit_cnt  = rp->red_stats.xmit_cnt;
881 			q_stats->drop_cnt  = rp->red_stats.drop_cnt;
882 			q_stats->drop_forced   = rp->red_stats.drop_forced;
883 			q_stats->drop_unforced = rp->red_stats.drop_unforced;
884 			q_stats->marked_packets = rp->red_stats.marked_packets;
885 
886 			q_stats->weight		= rp->red_weight;
887 			q_stats->inv_pmax	= rp->red_inv_pmax;
888 			q_stats->th_min		= rp->red_thmin;
889 			q_stats->th_max		= rp->red_thmax;
890 
891 #ifdef ALTQ_FLOWVALVE
892 			if (rp->red_flowvalve != NULL) {
893 				struct flowvalve *fv = rp->red_flowvalve;
894 				q_stats->fv_flows    = fv->fv_flows;
895 				q_stats->fv_pass     = fv->fv_stats.pass;
896 				q_stats->fv_predrop  = fv->fv_stats.predrop;
897 				q_stats->fv_alloc    = fv->fv_stats.alloc;
898 				q_stats->fv_escape   = fv->fv_stats.escape;
899 			} else {
900 #endif /* ALTQ_FLOWVALVE */
901 				q_stats->fv_flows    = 0;
902 				q_stats->fv_pass     = 0;
903 				q_stats->fv_predrop  = 0;
904 				q_stats->fv_alloc    = 0;
905 				q_stats->fv_escape   = 0;
906 #ifdef ALTQ_FLOWVALVE
907 			}
908 #endif /* ALTQ_FLOWVALVE */
909 		} while (/*CONSTCOND*/ 0);
910 		break;
911 
912 	case RED_CONFIG:
913 		do {
914 			struct red_conf *fc;
915 			red_t *new;
916 			int s, limit;
917 
918 			fc = (struct red_conf *)addr;
919 			if ((rqp = altq_lookup(fc->iface.red_ifname,
920 					       ALTQT_RED)) == NULL) {
921 				error = EBADF;
922 				break;
923 			}
924 			new = red_alloc(fc->red_weight,
925 					fc->red_inv_pmax,
926 					fc->red_thmin,
927 					fc->red_thmax,
928 					fc->red_flags,
929 					fc->red_pkttime);
930 			if (new == NULL) {
931 				error = ENOMEM;
932 				break;
933 			}
934 
935 			s = splnet();
936 			red_purgeq(rqp);
937 			limit = fc->red_limit;
938 			if (limit < fc->red_thmax)
939 				limit = fc->red_thmax;
940 			qlimit(rqp->rq_q) = limit;
941 			fc->red_limit = limit;	/* write back the new value */
942 
943 			red_destroy(rqp->rq_red);
944 			rqp->rq_red = new;
945 
946 			splx(s);
947 
948 			/* write back new values */
949 			fc->red_limit = limit;
950 			fc->red_inv_pmax = rqp->rq_red->red_inv_pmax;
951 			fc->red_thmin = rqp->rq_red->red_thmin;
952 			fc->red_thmax = rqp->rq_red->red_thmax;
953 
954 		} while (/*CONSTCOND*/ 0);
955 		break;
956 
957 	case RED_SETDEFAULTS:
958 		do {
959 			struct redparams *rp;
960 
961 			rp = (struct redparams *)addr;
962 
963 			default_th_min = rp->th_min;
964 			default_th_max = rp->th_max;
965 			default_inv_pmax = rp->inv_pmax;
966 		} while (/*CONSTCOND*/ 0);
967 		break;
968 
969 	default:
970 		error = EINVAL;
971 		break;
972 	}
973 	return error;
974 }
975 
976 static int
977 red_detach(red_queue_t *rqp)
978 {
979 	red_queue_t *tmp;
980 	int error = 0;
981 
982 	if (ALTQ_IS_ENABLED(rqp->rq_ifq))
983 		altq_disable(rqp->rq_ifq);
984 
985 	if ((error = altq_detach(rqp->rq_ifq)))
986 		return (error);
987 
988 	if (red_list == rqp)
989 		red_list = rqp->rq_next;
990 	else {
991 		for (tmp = red_list; tmp != NULL; tmp = tmp->rq_next)
992 			if (tmp->rq_next == rqp) {
993 				tmp->rq_next = rqp->rq_next;
994 				break;
995 			}
996 		if (tmp == NULL)
997 			printf("red_detach: no state found in red_list!\n");
998 	}
999 
1000 	red_destroy(rqp->rq_red);
1001 	free(rqp->rq_q, M_DEVBUF);
1002 	free(rqp, M_DEVBUF);
1003 	return (error);
1004 }
1005 
1006 /*
1007  * enqueue routine:
1008  *
1009  *	returns: 0 when successfully queued.
1010  *		 ENOBUFS when drop occurs.
1011  */
1012 static int
1013 red_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr)
1014 {
1015 	red_queue_t *rqp = (red_queue_t *)ifq->altq_disc;
1016 
1017 	if (red_addq(rqp->rq_red, rqp->rq_q, m, pktattr) < 0)
1018 		return ENOBUFS;
1019 	ifq->ifq_len++;
1020 	return 0;
1021 }
1022 
1023 /*
1024  * dequeue routine:
1025  *	must be called in splnet.
1026  *
1027  *	returns: mbuf dequeued.
1028  *		 NULL when no packet is available in the queue.
1029  */
1030 
1031 static struct mbuf *
1032 red_dequeue(struct ifaltq *ifq, int op)
1033 {
1034 	red_queue_t *rqp = (red_queue_t *)ifq->altq_disc;
1035 	struct mbuf *m;
1036 
1037 	if (op == ALTDQ_POLL)
1038 		return qhead(rqp->rq_q);
1039 
1040 	/* op == ALTDQ_REMOVE */
1041 	m =  red_getq(rqp->rq_red, rqp->rq_q);
1042 	if (m != NULL)
1043 		ifq->ifq_len--;
1044 	return (m);
1045 }
1046 
1047 static int
1048 red_request(struct ifaltq *ifq, int req, void *arg)
1049 {
1050 	red_queue_t *rqp = (red_queue_t *)ifq->altq_disc;
1051 
1052 	switch (req) {
1053 	case ALTRQ_PURGE:
1054 		red_purgeq(rqp);
1055 		break;
1056 	}
1057 	return (0);
1058 }
1059 
1060 static void
1061 red_purgeq(red_queue_t *rqp)
1062 {
1063 	_flushq(rqp->rq_q);
1064 	if (ALTQ_IS_ENABLED(rqp->rq_ifq))
1065 		rqp->rq_ifq->ifq_len = 0;
1066 }
1067 
1068 #ifdef ALTQ_FLOWVALVE
1069 
1070 #define	FV_PSHIFT	7	/* weight of average drop rate -- 1/128 */
1071 #define	FV_PSCALE(x)	((x) << FV_PSHIFT)
1072 #define	FV_PUNSCALE(x)	((x) >> FV_PSHIFT)
1073 #define	FV_FSHIFT	5	/* weight of average fraction -- 1/32 */
1074 #define	FV_FSCALE(x)	((x) << FV_FSHIFT)
1075 #define	FV_FUNSCALE(x)	((x) >> FV_FSHIFT)
1076 
1077 #define	FV_TIMER	(3 * hz)	/* timer value for garbage collector */
1078 #define	FV_FLOWLISTSIZE		64	/* how many flows in flowlist */
1079 
1080 #define	FV_N			10	/* update fve_f every FV_N packets */
1081 
1082 #define	FV_BACKOFFTHRESH	1  /* backoff threshold interval in second */
1083 #define	FV_TTHRESH		3  /* time threshold to delete fve */
1084 #define	FV_ALPHA		5  /* extra packet count */
1085 
1086 #define	FV_STATS
1087 
1088 #if (__FreeBSD_version > 300000) || defined(__HAVE_TIMECOUNTER)
1089 #define	FV_TIMESTAMP(tp)	getmicrotime(tp)
1090 #else
1091 #define	FV_TIMESTAMP(tp)	{ (*(tp)) = time; }
1092 #endif
1093 
1094 /*
1095  * Brtt table: 127 entry table to convert drop rate (p) to
1096  * the corresponding bandwidth fraction (f)
1097  * the following equation is implemented to use scaled values,
1098  * fve_p and fve_f, in the fixed point format.
1099  *
1100  *   Brtt(p) = 1 /(sqrt(4*p/3) + min(1,3*sqrt(p*6/8)) * p * (1+32 * p*p))
1101  *   f = Brtt(p) / (max_th + alpha)
1102  */
1103 #define	BRTT_SIZE	128
1104 #define	BRTT_SHIFT	12
1105 #define	BRTT_MASK	0x0007f000
1106 #define	BRTT_PMAX	(1 << (FV_PSHIFT + FP_SHIFT))
1107 
1108 const int brtt_tab[BRTT_SIZE] = {
1109 	0, 1262010, 877019, 703694, 598706, 525854, 471107, 427728,
1110 	392026, 361788, 335598, 312506, 291850, 273158, 256081, 240361,
1111 	225800, 212247, 199585, 187788, 178388, 169544, 161207, 153333,
1112 	145888, 138841, 132165, 125836, 119834, 114141, 108739, 103612,
1113 	98747, 94129, 89746, 85585, 81637, 77889, 74333, 70957,
1114 	67752, 64711, 61824, 59084, 56482, 54013, 51667, 49440,
1115 	47325, 45315, 43406, 41591, 39866, 38227, 36667, 35184,
1116 	33773, 32430, 31151, 29933, 28774, 27668, 26615, 25611,
1117 	24653, 23740, 22868, 22035, 21240, 20481, 19755, 19062,
1118 	18399, 17764, 17157, 16576, 16020, 15487, 14976, 14487,
1119 	14017, 13567, 13136, 12721, 12323, 11941, 11574, 11222,
1120 	10883, 10557, 10243, 9942, 9652, 9372, 9103, 8844,
1121 	8594, 8354, 8122, 7898, 7682, 7474, 7273, 7079,
1122 	6892, 6711, 6536, 6367, 6204, 6046, 5893, 5746,
1123 	5603, 5464, 5330, 5201, 5075, 4954, 4836, 4722,
1124 	4611, 4504, 4400, 4299, 4201, 4106, 4014, 3924
1125 };
1126 
1127 static inline struct fve *
1128 flowlist_lookup(struct flowvalve *fv, struct altq_pktattr *pktattr,
1129     struct timeval *now)
1130 {
1131 	struct fve *fve;
1132 	int flows;
1133 	struct ip *ip;
1134 #ifdef INET6
1135 	struct ip6_hdr *ip6;
1136 #endif
1137 	struct timeval tthresh;
1138 
1139 	if (pktattr == NULL)
1140 		return (NULL);
1141 
1142 	tthresh.tv_sec = now->tv_sec - FV_TTHRESH;
1143 	flows = 0;
1144 	/*
1145 	 * search the flow list
1146 	 */
1147 	switch (pktattr->pattr_af) {
1148 	case AF_INET:
1149 		ip = (struct ip *)pktattr->pattr_hdr;
1150 		TAILQ_FOREACH(fve, &fv->fv_flowlist, fve_lru){
1151 			if (fve->fve_lastdrop.tv_sec == 0)
1152 				break;
1153 			if (fve->fve_lastdrop.tv_sec < tthresh.tv_sec) {
1154 				fve->fve_lastdrop.tv_sec = 0;
1155 				break;
1156 			}
1157 			if (fve->fve_flow.flow_af == AF_INET &&
1158 			    fve->fve_flow.flow_ip.ip_src.s_addr ==
1159 			    ip->ip_src.s_addr &&
1160 			    fve->fve_flow.flow_ip.ip_dst.s_addr ==
1161 			    ip->ip_dst.s_addr)
1162 				return (fve);
1163 			flows++;
1164 		}
1165 		break;
1166 #ifdef INET6
1167 	case AF_INET6:
1168 		ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
1169 		TAILQ_FOREACH(fve, &fv->fv_flowlist, fve_lru){
1170 			if (fve->fve_lastdrop.tv_sec == 0)
1171 				break;
1172 			if (fve->fve_lastdrop.tv_sec < tthresh.tv_sec) {
1173 				fve->fve_lastdrop.tv_sec = 0;
1174 				break;
1175 			}
1176 			if (fve->fve_flow.flow_af == AF_INET6 &&
1177 			    IN6_ARE_ADDR_EQUAL(&fve->fve_flow.flow_ip6.ip6_src,
1178 					       &ip6->ip6_src) &&
1179 			    IN6_ARE_ADDR_EQUAL(&fve->fve_flow.flow_ip6.ip6_dst,
1180 					       &ip6->ip6_dst))
1181 				return (fve);
1182 			flows++;
1183 		}
1184 		break;
1185 #endif /* INET6 */
1186 
1187 	default:
1188 		/* unknown protocol.  no drop. */
1189 		return (NULL);
1190 	}
1191 	fv->fv_flows = flows;	/* save the number of active fve's */
1192 	return (NULL);
1193 }
1194 
1195 static inline struct fve *
1196 flowlist_reclaim(struct flowvalve *fv, struct altq_pktattr *pktattr)
1197 {
1198 	struct fve *fve;
1199 	struct ip *ip;
1200 #ifdef INET6
1201 	struct ip6_hdr *ip6;
1202 #endif
1203 
1204 	/*
1205 	 * get an entry from the tail of the LRU list.
1206 	 */
1207 	fve = TAILQ_LAST(&fv->fv_flowlist, fv_flowhead);
1208 
1209 	switch (pktattr->pattr_af) {
1210 	case AF_INET:
1211 		ip = (struct ip *)pktattr->pattr_hdr;
1212 		fve->fve_flow.flow_af = AF_INET;
1213 		fve->fve_flow.flow_ip.ip_src = ip->ip_src;
1214 		fve->fve_flow.flow_ip.ip_dst = ip->ip_dst;
1215 		break;
1216 #ifdef INET6
1217 	case AF_INET6:
1218 		ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
1219 		fve->fve_flow.flow_af = AF_INET6;
1220 		fve->fve_flow.flow_ip6.ip6_src = ip6->ip6_src;
1221 		fve->fve_flow.flow_ip6.ip6_dst = ip6->ip6_dst;
1222 		break;
1223 #endif
1224 	}
1225 
1226 	fve->fve_state = Green;
1227 	fve->fve_p = 0.0;
1228 	fve->fve_f = 0.0;
1229 	fve->fve_ifseq = fv->fv_ifseq - 1;
1230 	fve->fve_count = 0;
1231 
1232 	fv->fv_flows++;
1233 #ifdef FV_STATS
1234 	fv->fv_stats.alloc++;
1235 #endif
1236 	return (fve);
1237 }
1238 
1239 static inline void
1240 flowlist_move_to_head(struct flowvalve *fv, struct fve *fve)
1241 {
1242 	if (TAILQ_FIRST(&fv->fv_flowlist) != fve) {
1243 		TAILQ_REMOVE(&fv->fv_flowlist, fve, fve_lru);
1244 		TAILQ_INSERT_HEAD(&fv->fv_flowlist, fve, fve_lru);
1245 	}
1246 }
1247 
1248 /*
1249  * allocate flowvalve structure
1250  */
1251 static struct flowvalve *
1252 fv_alloc(struct red *rp)
1253 {
1254 	struct flowvalve *fv;
1255 	struct fve *fve;
1256 	int i, num;
1257 
1258 	num = FV_FLOWLISTSIZE;
1259 	fv = malloc(sizeof(struct flowvalve), M_DEVBUF, M_WAITOK|M_ZERO);
1260 	if (fv == NULL)
1261 		return (NULL);
1262 
1263 	fv->fv_fves = malloc(sizeof(struct fve) * num, M_DEVBUF,
1264 	    M_WAITOK|M_ZERO);
1265 	if (fv->fv_fves == NULL) {
1266 		free(fv, M_DEVBUF);
1267 		return (NULL);
1268 	}
1269 
1270 	fv->fv_flows = 0;
1271 	TAILQ_INIT(&fv->fv_flowlist);
1272 	for (i = 0; i < num; i++) {
1273 		fve = &fv->fv_fves[i];
1274 		fve->fve_lastdrop.tv_sec = 0;
1275 		TAILQ_INSERT_TAIL(&fv->fv_flowlist, fve, fve_lru);
1276 	}
1277 
1278 	/* initialize drop rate threshold in scaled fixed-point */
1279 	fv->fv_pthresh = (FV_PSCALE(1) << FP_SHIFT) / rp->red_inv_pmax;
1280 
1281 	/* initialize drop rate to fraction table */
1282 	fv->fv_p2ftab = malloc(sizeof(int) * BRTT_SIZE, M_DEVBUF, M_WAITOK);
1283 	if (fv->fv_p2ftab == NULL) {
1284 		free(fv->fv_fves, M_DEVBUF);
1285 		free(fv, M_DEVBUF);
1286 		return (NULL);
1287 	}
1288 	/*
1289 	 * create the p2f table.
1290 	 * (shift is used to keep the precision)
1291 	 */
1292 	for (i = 1; i < BRTT_SIZE; i++) {
1293 		int f;
1294 
1295 		f = brtt_tab[i] << 8;
1296 		fv->fv_p2ftab[i] = (f / (rp->red_thmax + FV_ALPHA)) >> 8;
1297 	}
1298 
1299 	return (fv);
1300 }
1301 
1302 static void
1303 fv_destroy(struct flowvalve *fv)
1304 {
1305 	free(fv->fv_p2ftab, M_DEVBUF);
1306 	free(fv->fv_fves, M_DEVBUF);
1307 	free(fv, M_DEVBUF);
1308 }
1309 
1310 static inline int
1311 fv_p2f(struct flowvalve *fv, int p)
1312 {
1313 	int val, f;
1314 
1315 	if (p >= BRTT_PMAX)
1316 		f = fv->fv_p2ftab[BRTT_SIZE-1];
1317 	else if ((val = (p & BRTT_MASK)))
1318 		f = fv->fv_p2ftab[(val >> BRTT_SHIFT)];
1319 	else
1320 		f = fv->fv_p2ftab[1];
1321 	return (f);
1322 }
1323 
1324 /*
1325  * check if an arriving packet should be pre-dropped.
1326  * called from red_addq() when a packet arrives.
1327  * returns 1 when the packet should be pre-dropped.
1328  * should be called in splnet.
1329  */
1330 static int
1331 fv_checkflow(struct flowvalve *fv, struct altq_pktattr *pktattr,
1332     struct fve **fcache)
1333 {
1334 	struct fve *fve;
1335 	struct timeval now;
1336 
1337 	fv->fv_ifseq++;
1338 	FV_TIMESTAMP(&now);
1339 
1340 	if ((fve = flowlist_lookup(fv, pktattr, &now)) == NULL)
1341 		/* no matching entry in the flowlist */
1342 		return (0);
1343 
1344 	*fcache = fve;
1345 
1346 	/* update fraction f for every FV_N packets */
1347 	if (++fve->fve_count == FV_N) {
1348 		/*
1349 		 * f = Wf * N / (fv_ifseq - fve_ifseq) + (1 - Wf) * f
1350 		 */
1351 		fve->fve_f =
1352 			(FV_N << FP_SHIFT) / (fv->fv_ifseq - fve->fve_ifseq)
1353 			+ fve->fve_f - FV_FUNSCALE(fve->fve_f);
1354 		fve->fve_ifseq = fv->fv_ifseq;
1355 		fve->fve_count = 0;
1356 	}
1357 
1358 	/*
1359 	 * overpumping test
1360 	 */
1361 	if (fve->fve_state == Green && fve->fve_p > fv->fv_pthresh) {
1362 		int fthresh;
1363 
1364 		/* calculate a threshold */
1365 		fthresh = fv_p2f(fv, fve->fve_p);
1366 		if (fve->fve_f > fthresh)
1367 			fve->fve_state = Red;
1368 	}
1369 
1370 	if (fve->fve_state == Red) {
1371 		/*
1372 		 * backoff test
1373 		 */
1374 		if (now.tv_sec - fve->fve_lastdrop.tv_sec > FV_BACKOFFTHRESH) {
1375 			/* no drop for at least FV_BACKOFFTHRESH sec */
1376 			fve->fve_p = 0;
1377 			fve->fve_state = Green;
1378 #ifdef FV_STATS
1379 			fv->fv_stats.escape++;
1380 #endif
1381 		} else {
1382 			/* block this flow */
1383 			flowlist_move_to_head(fv, fve);
1384 			fve->fve_lastdrop = now;
1385 #ifdef FV_STATS
1386 			fv->fv_stats.predrop++;
1387 #endif
1388 			return (1);
1389 		}
1390 	}
1391 
1392 	/*
1393 	 * p = (1 - Wp) * p
1394 	 */
1395 	fve->fve_p -= FV_PUNSCALE(fve->fve_p);
1396 	if (fve->fve_p < 0)
1397 		fve->fve_p = 0;
1398 #ifdef FV_STATS
1399 	fv->fv_stats.pass++;
1400 #endif
1401 	return (0);
1402 }
1403 
1404 /*
1405  * called from red_addq when a packet is dropped by red.
1406  * should be called in splnet.
1407  */
1408 static void
1409 fv_dropbyred(struct flowvalve *fv, struct altq_pktattr *pktattr,
1410     struct fve *fcache)
1411 {
1412 	struct fve *fve;
1413 	struct timeval now;
1414 
1415 	if (pktattr == NULL)
1416 		return;
1417 	FV_TIMESTAMP(&now);
1418 
1419 	if (fcache != NULL)
1420 		/* the fve of this packet is already cached */
1421 		fve = fcache;
1422 	else if ((fve = flowlist_lookup(fv, pktattr, &now)) == NULL)
1423 		fve = flowlist_reclaim(fv, pktattr);
1424 
1425 	flowlist_move_to_head(fv, fve);
1426 
1427 	/*
1428 	 * update p:  the following line cancels the update
1429 	 *	      in fv_checkflow() and calculate
1430 	 *	p = Wp + (1 - Wp) * p
1431 	 */
1432 	fve->fve_p = (1 << FP_SHIFT) + fve->fve_p;
1433 
1434 	fve->fve_lastdrop = now;
1435 }
1436 
1437 #endif /* ALTQ_FLOWVALVE */
1438 
1439 #ifdef KLD_MODULE
1440 
1441 static struct altqsw red_sw =
1442 	{"red", redopen, redclose, redioctl};
1443 
1444 ALTQ_MODULE(altq_red, ALTQT_RED, &red_sw);
1445 MODULE_VERSION(altq_red, 1);
1446 
1447 #endif /* KLD_MODULE */
1448 #endif /* ALTQ3_COMPAT */
1449 
1450 #endif /* ALTQ_RED */
1451