xref: /linux/net/sched/sch_sfb.c (revision cff4fa8415a3224a5abdd2b1dd7f431e4ea49366) 
1 /*
2  * net/sched/sch_sfb.c	  Stochastic Fair Blue
3  *
4  * Copyright (c) 2008-2011 Juliusz Chroboczek <jch@pps.jussieu.fr>
5  * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com>
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * version 2 as published by the Free Software Foundation.
10  *
11  * W. Feng, D. Kandlur, D. Saha, K. Shin. Blue:
12  * A New Class of Active Queue Management Algorithms.
13  * U. Michigan CSE-TR-387-99, April 1999.
14  *
15  * http://www.thefengs.com/wuchang/blue/CSE-TR-387-99.pdf
16  *
17  */
18 
19 #include <linux/module.h>
20 #include <linux/types.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/skbuff.h>
24 #include <linux/random.h>
25 #include <linux/jhash.h>
26 #include <net/ip.h>
27 #include <net/pkt_sched.h>
28 #include <net/inet_ecn.h>
29 
30 /*
31  * SFB uses two B[l][n] : L x N arrays of bins (L levels, N bins per level)
32  * This implementation uses L = 8 and N = 16
33  * This permits us to split one 32bit hash (provided per packet by rxhash or
34  * external classifier) into 8 subhashes of 4 bits.
35  */
36 #define SFB_BUCKET_SHIFT 4
37 #define SFB_NUMBUCKETS	(1 << SFB_BUCKET_SHIFT) /* N bins per Level */
38 #define SFB_BUCKET_MASK (SFB_NUMBUCKETS - 1)
39 #define SFB_LEVELS	(32 / SFB_BUCKET_SHIFT) /* L */
40 
41 /* SFB algo uses a virtual queue, named "bin" */
42 struct sfb_bucket {
43 	u16		qlen; /* length of virtual queue */
44 	u16		p_mark; /* marking probability */
45 };
46 
47 /* We use a double buffering right before hash change
48  * (Section 4.4 of SFB reference : moving hash functions)
49  */
50 struct sfb_bins {
51 	u32		  perturbation; /* jhash perturbation */
52 	struct sfb_bucket bins[SFB_LEVELS][SFB_NUMBUCKETS];
53 };
54 
55 struct sfb_sched_data {
56 	struct Qdisc	*qdisc;
57 	struct tcf_proto *filter_list;
58 	unsigned long	rehash_interval;
59 	unsigned long	warmup_time;	/* double buffering warmup time in jiffies */
60 	u32		max;
61 	u32		bin_size;	/* maximum queue length per bin */
62 	u32		increment;	/* d1 */
63 	u32		decrement;	/* d2 */
64 	u32		limit;		/* HARD maximal queue length */
65 	u32		penalty_rate;
66 	u32		penalty_burst;
67 	u32		tokens_avail;
68 	unsigned long	rehash_time;
69 	unsigned long	token_time;
70 
71 	u8		slot;		/* current active bins (0 or 1) */
72 	bool		double_buffering;
73 	struct sfb_bins bins[2];
74 
75 	struct {
76 		u32	earlydrop;
77 		u32	penaltydrop;
78 		u32	bucketdrop;
79 		u32	queuedrop;
80 		u32	childdrop;	/* drops in child qdisc */
81 		u32	marked;		/* ECN mark */
82 	} stats;
83 };
84 
85 /*
86  * Each queued skb might be hashed on one or two bins
87  * We store in skb_cb the two hash values.
88  * (A zero value means double buffering was not used)
89  */
90 struct sfb_skb_cb {
91 	u32 hashes[2];
92 };
93 
94 static inline struct sfb_skb_cb *sfb_skb_cb(const struct sk_buff *skb)
95 {
96 	BUILD_BUG_ON(sizeof(skb->cb) <
97 		sizeof(struct qdisc_skb_cb) + sizeof(struct sfb_skb_cb));
98 	return (struct sfb_skb_cb *)qdisc_skb_cb(skb)->data;
99 }
100 
101 /*
102  * If using 'internal' SFB flow classifier, hash comes from skb rxhash
103  * If using external classifier, hash comes from the classid.
104  */
105 static u32 sfb_hash(const struct sk_buff *skb, u32 slot)
106 {
107 	return sfb_skb_cb(skb)->hashes[slot];
108 }
109 
110 /* Probabilities are coded as Q0.16 fixed-point values,
111  * with 0xFFFF representing 65535/65536 (almost 1.0)
112  * Addition and subtraction are saturating in [0, 65535]
113  */
114 static u32 prob_plus(u32 p1, u32 p2)
115 {
116 	u32 res = p1 + p2;
117 
118 	return min_t(u32, res, SFB_MAX_PROB);
119 }
120 
121 static u32 prob_minus(u32 p1, u32 p2)
122 {
123 	return p1 > p2 ? p1 - p2 : 0;
124 }
125 
126 static void increment_one_qlen(u32 sfbhash, u32 slot, struct sfb_sched_data *q)
127 {
128 	int i;
129 	struct sfb_bucket *b = &q->bins[slot].bins[0][0];
130 
131 	for (i = 0; i < SFB_LEVELS; i++) {
132 		u32 hash = sfbhash & SFB_BUCKET_MASK;
133 
134 		sfbhash >>= SFB_BUCKET_SHIFT;
135 		if (b[hash].qlen < 0xFFFF)
136 			b[hash].qlen++;
137 		b += SFB_NUMBUCKETS; /* next level */
138 	}
139 }
140 
141 static void increment_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
142 {
143 	u32 sfbhash;
144 
145 	sfbhash = sfb_hash(skb, 0);
146 	if (sfbhash)
147 		increment_one_qlen(sfbhash, 0, q);
148 
149 	sfbhash = sfb_hash(skb, 1);
150 	if (sfbhash)
151 		increment_one_qlen(sfbhash, 1, q);
152 }
153 
154 static void decrement_one_qlen(u32 sfbhash, u32 slot,
155 			       struct sfb_sched_data *q)
156 {
157 	int i;
158 	struct sfb_bucket *b = &q->bins[slot].bins[0][0];
159 
160 	for (i = 0; i < SFB_LEVELS; i++) {
161 		u32 hash = sfbhash & SFB_BUCKET_MASK;
162 
163 		sfbhash >>= SFB_BUCKET_SHIFT;
164 		if (b[hash].qlen > 0)
165 			b[hash].qlen--;
166 		b += SFB_NUMBUCKETS; /* next level */
167 	}
168 }
169 
170 static void decrement_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
171 {
172 	u32 sfbhash;
173 
174 	sfbhash = sfb_hash(skb, 0);
175 	if (sfbhash)
176 		decrement_one_qlen(sfbhash, 0, q);
177 
178 	sfbhash = sfb_hash(skb, 1);
179 	if (sfbhash)
180 		decrement_one_qlen(sfbhash, 1, q);
181 }
182 
183 static void decrement_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
184 {
185 	b->p_mark = prob_minus(b->p_mark, q->decrement);
186 }
187 
188 static void increment_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
189 {
190 	b->p_mark = prob_plus(b->p_mark, q->increment);
191 }
192 
193 static void sfb_zero_all_buckets(struct sfb_sched_data *q)
194 {
195 	memset(&q->bins, 0, sizeof(q->bins));
196 }
197 
198 /*
199  * compute max qlen, max p_mark, and avg p_mark
200  */
201 static u32 sfb_compute_qlen(u32 *prob_r, u32 *avgpm_r, const struct sfb_sched_data *q)
202 {
203 	int i;
204 	u32 qlen = 0, prob = 0, totalpm = 0;
205 	const struct sfb_bucket *b = &q->bins[q->slot].bins[0][0];
206 
207 	for (i = 0; i < SFB_LEVELS * SFB_NUMBUCKETS; i++) {
208 		if (qlen < b->qlen)
209 			qlen = b->qlen;
210 		totalpm += b->p_mark;
211 		if (prob < b->p_mark)
212 			prob = b->p_mark;
213 		b++;
214 	}
215 	*prob_r = prob;
216 	*avgpm_r = totalpm / (SFB_LEVELS * SFB_NUMBUCKETS);
217 	return qlen;
218 }
219 
220 
221 static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q)
222 {
223 	q->bins[slot].perturbation = net_random();
224 }
225 
226 static void sfb_swap_slot(struct sfb_sched_data *q)
227 {
228 	sfb_init_perturbation(q->slot, q);
229 	q->slot ^= 1;
230 	q->double_buffering = false;
231 }
232 
233 /* Non elastic flows are allowed to use part of the bandwidth, expressed
234  * in "penalty_rate" packets per second, with "penalty_burst" burst
235  */
236 static bool sfb_rate_limit(struct sk_buff *skb, struct sfb_sched_data *q)
237 {
238 	if (q->penalty_rate == 0 || q->penalty_burst == 0)
239 		return true;
240 
241 	if (q->tokens_avail < 1) {
242 		unsigned long age = min(10UL * HZ, jiffies - q->token_time);
243 
244 		q->tokens_avail = (age * q->penalty_rate) / HZ;
245 		if (q->tokens_avail > q->penalty_burst)
246 			q->tokens_avail = q->penalty_burst;
247 		q->token_time = jiffies;
248 		if (q->tokens_avail < 1)
249 			return true;
250 	}
251 
252 	q->tokens_avail--;
253 	return false;
254 }
255 
256 static bool sfb_classify(struct sk_buff *skb, struct sfb_sched_data *q,
257 			 int *qerr, u32 *salt)
258 {
259 	struct tcf_result res;
260 	int result;
261 
262 	result = tc_classify(skb, q->filter_list, &res);
263 	if (result >= 0) {
264 #ifdef CONFIG_NET_CLS_ACT
265 		switch (result) {
266 		case TC_ACT_STOLEN:
267 		case TC_ACT_QUEUED:
268 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
269 		case TC_ACT_SHOT:
270 			return false;
271 		}
272 #endif
273 		*salt = TC_H_MIN(res.classid);
274 		return true;
275 	}
276 	return false;
277 }
278 
279 static int sfb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
280 {
281 
282 	struct sfb_sched_data *q = qdisc_priv(sch);
283 	struct Qdisc *child = q->qdisc;
284 	int i;
285 	u32 p_min = ~0;
286 	u32 minqlen = ~0;
287 	u32 r, slot, salt, sfbhash;
288 	int ret = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
289 
290 	if (q->rehash_interval > 0) {
291 		unsigned long limit = q->rehash_time + q->rehash_interval;
292 
293 		if (unlikely(time_after(jiffies, limit))) {
294 			sfb_swap_slot(q);
295 			q->rehash_time = jiffies;
296 		} else if (unlikely(!q->double_buffering && q->warmup_time > 0 &&
297 				    time_after(jiffies, limit - q->warmup_time))) {
298 			q->double_buffering = true;
299 		}
300 	}
301 
302 	if (q->filter_list) {
303 		/* If using external classifiers, get result and record it. */
304 		if (!sfb_classify(skb, q, &ret, &salt))
305 			goto other_drop;
306 	} else {
307 		salt = skb_get_rxhash(skb);
308 	}
309 
310 	slot = q->slot;
311 
312 	sfbhash = jhash_1word(salt, q->bins[slot].perturbation);
313 	if (!sfbhash)
314 		sfbhash = 1;
315 	sfb_skb_cb(skb)->hashes[slot] = sfbhash;
316 
317 	for (i = 0; i < SFB_LEVELS; i++) {
318 		u32 hash = sfbhash & SFB_BUCKET_MASK;
319 		struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
320 
321 		sfbhash >>= SFB_BUCKET_SHIFT;
322 		if (b->qlen == 0)
323 			decrement_prob(b, q);
324 		else if (b->qlen >= q->bin_size)
325 			increment_prob(b, q);
326 		if (minqlen > b->qlen)
327 			minqlen = b->qlen;
328 		if (p_min > b->p_mark)
329 			p_min = b->p_mark;
330 	}
331 
332 	slot ^= 1;
333 	sfb_skb_cb(skb)->hashes[slot] = 0;
334 
335 	if (unlikely(minqlen >= q->max || sch->q.qlen >= q->limit)) {
336 		sch->qstats.overlimits++;
337 		if (minqlen >= q->max)
338 			q->stats.bucketdrop++;
339 		else
340 			q->stats.queuedrop++;
341 		goto drop;
342 	}
343 
344 	if (unlikely(p_min >= SFB_MAX_PROB)) {
345 		/* Inelastic flow */
346 		if (q->double_buffering) {
347 			sfbhash = jhash_1word(salt, q->bins[slot].perturbation);
348 			if (!sfbhash)
349 				sfbhash = 1;
350 			sfb_skb_cb(skb)->hashes[slot] = sfbhash;
351 
352 			for (i = 0; i < SFB_LEVELS; i++) {
353 				u32 hash = sfbhash & SFB_BUCKET_MASK;
354 				struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
355 
356 				sfbhash >>= SFB_BUCKET_SHIFT;
357 				if (b->qlen == 0)
358 					decrement_prob(b, q);
359 				else if (b->qlen >= q->bin_size)
360 					increment_prob(b, q);
361 			}
362 		}
363 		if (sfb_rate_limit(skb, q)) {
364 			sch->qstats.overlimits++;
365 			q->stats.penaltydrop++;
366 			goto drop;
367 		}
368 		goto enqueue;
369 	}
370 
371 	r = net_random() & SFB_MAX_PROB;
372 
373 	if (unlikely(r < p_min)) {
374 		if (unlikely(p_min > SFB_MAX_PROB / 2)) {
375 			/* If we're marking that many packets, then either
376 			 * this flow is unresponsive, or we're badly congested.
377 			 * In either case, we want to start dropping packets.
378 			 */
379 			if (r < (p_min - SFB_MAX_PROB / 2) * 2) {
380 				q->stats.earlydrop++;
381 				goto drop;
382 			}
383 		}
384 		if (INET_ECN_set_ce(skb)) {
385 			q->stats.marked++;
386 		} else {
387 			q->stats.earlydrop++;
388 			goto drop;
389 		}
390 	}
391 
392 enqueue:
393 	ret = qdisc_enqueue(skb, child);
394 	if (likely(ret == NET_XMIT_SUCCESS)) {
395 		sch->q.qlen++;
396 		increment_qlen(skb, q);
397 	} else if (net_xmit_drop_count(ret)) {
398 		q->stats.childdrop++;
399 		sch->qstats.drops++;
400 	}
401 	return ret;
402 
403 drop:
404 	qdisc_drop(skb, sch);
405 	return NET_XMIT_CN;
406 other_drop:
407 	if (ret & __NET_XMIT_BYPASS)
408 		sch->qstats.drops++;
409 	kfree_skb(skb);
410 	return ret;
411 }
412 
413 static struct sk_buff *sfb_dequeue(struct Qdisc *sch)
414 {
415 	struct sfb_sched_data *q = qdisc_priv(sch);
416 	struct Qdisc *child = q->qdisc;
417 	struct sk_buff *skb;
418 
419 	skb = child->dequeue(q->qdisc);
420 
421 	if (skb) {
422 		qdisc_bstats_update(sch, skb);
423 		sch->q.qlen--;
424 		decrement_qlen(skb, q);
425 	}
426 
427 	return skb;
428 }
429 
430 static struct sk_buff *sfb_peek(struct Qdisc *sch)
431 {
432 	struct sfb_sched_data *q = qdisc_priv(sch);
433 	struct Qdisc *child = q->qdisc;
434 
435 	return child->ops->peek(child);
436 }
437 
438 /* No sfb_drop -- impossible since the child doesn't return the dropped skb. */
439 
440 static void sfb_reset(struct Qdisc *sch)
441 {
442 	struct sfb_sched_data *q = qdisc_priv(sch);
443 
444 	qdisc_reset(q->qdisc);
445 	sch->q.qlen = 0;
446 	q->slot = 0;
447 	q->double_buffering = false;
448 	sfb_zero_all_buckets(q);
449 	sfb_init_perturbation(0, q);
450 }
451 
452 static void sfb_destroy(struct Qdisc *sch)
453 {
454 	struct sfb_sched_data *q = qdisc_priv(sch);
455 
456 	tcf_destroy_chain(&q->filter_list);
457 	qdisc_destroy(q->qdisc);
458 }
459 
460 static const struct nla_policy sfb_policy[TCA_SFB_MAX + 1] = {
461 	[TCA_SFB_PARMS]	= { .len = sizeof(struct tc_sfb_qopt) },
462 };
463 
464 static const struct tc_sfb_qopt sfb_default_ops = {
465 	.rehash_interval = 600 * MSEC_PER_SEC,
466 	.warmup_time = 60 * MSEC_PER_SEC,
467 	.limit = 0,
468 	.max = 25,
469 	.bin_size = 20,
470 	.increment = (SFB_MAX_PROB + 500) / 1000, /* 0.1 % */
471 	.decrement = (SFB_MAX_PROB + 3000) / 6000,
472 	.penalty_rate = 10,
473 	.penalty_burst = 20,
474 };
475 
476 static int sfb_change(struct Qdisc *sch, struct nlattr *opt)
477 {
478 	struct sfb_sched_data *q = qdisc_priv(sch);
479 	struct Qdisc *child;
480 	struct nlattr *tb[TCA_SFB_MAX + 1];
481 	const struct tc_sfb_qopt *ctl = &sfb_default_ops;
482 	u32 limit;
483 	int err;
484 
485 	if (opt) {
486 		err = nla_parse_nested(tb, TCA_SFB_MAX, opt, sfb_policy);
487 		if (err < 0)
488 			return -EINVAL;
489 
490 		if (tb[TCA_SFB_PARMS] == NULL)
491 			return -EINVAL;
492 
493 		ctl = nla_data(tb[TCA_SFB_PARMS]);
494 	}
495 
496 	limit = ctl->limit;
497 	if (limit == 0)
498 		limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1);
499 
500 	child = fifo_create_dflt(sch, &pfifo_qdisc_ops, limit);
501 	if (IS_ERR(child))
502 		return PTR_ERR(child);
503 
504 	sch_tree_lock(sch);
505 
506 	qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
507 	qdisc_destroy(q->qdisc);
508 	q->qdisc = child;
509 
510 	q->rehash_interval = msecs_to_jiffies(ctl->rehash_interval);
511 	q->warmup_time = msecs_to_jiffies(ctl->warmup_time);
512 	q->rehash_time = jiffies;
513 	q->limit = limit;
514 	q->increment = ctl->increment;
515 	q->decrement = ctl->decrement;
516 	q->max = ctl->max;
517 	q->bin_size = ctl->bin_size;
518 	q->penalty_rate = ctl->penalty_rate;
519 	q->penalty_burst = ctl->penalty_burst;
520 	q->tokens_avail = ctl->penalty_burst;
521 	q->token_time = jiffies;
522 
523 	q->slot = 0;
524 	q->double_buffering = false;
525 	sfb_zero_all_buckets(q);
526 	sfb_init_perturbation(0, q);
527 	sfb_init_perturbation(1, q);
528 
529 	sch_tree_unlock(sch);
530 
531 	return 0;
532 }
533 
534 static int sfb_init(struct Qdisc *sch, struct nlattr *opt)
535 {
536 	struct sfb_sched_data *q = qdisc_priv(sch);
537 
538 	q->qdisc = &noop_qdisc;
539 	return sfb_change(sch, opt);
540 }
541 
542 static int sfb_dump(struct Qdisc *sch, struct sk_buff *skb)
543 {
544 	struct sfb_sched_data *q = qdisc_priv(sch);
545 	struct nlattr *opts;
546 	struct tc_sfb_qopt opt = {
547 		.rehash_interval = jiffies_to_msecs(q->rehash_interval),
548 		.warmup_time = jiffies_to_msecs(q->warmup_time),
549 		.limit = q->limit,
550 		.max = q->max,
551 		.bin_size = q->bin_size,
552 		.increment = q->increment,
553 		.decrement = q->decrement,
554 		.penalty_rate = q->penalty_rate,
555 		.penalty_burst = q->penalty_burst,
556 	};
557 
558 	sch->qstats.backlog = q->qdisc->qstats.backlog;
559 	opts = nla_nest_start(skb, TCA_OPTIONS);
560 	NLA_PUT(skb, TCA_SFB_PARMS, sizeof(opt), &opt);
561 	return nla_nest_end(skb, opts);
562 
563 nla_put_failure:
564 	nla_nest_cancel(skb, opts);
565 	return -EMSGSIZE;
566 }
567 
568 static int sfb_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
569 {
570 	struct sfb_sched_data *q = qdisc_priv(sch);
571 	struct tc_sfb_xstats st = {
572 		.earlydrop = q->stats.earlydrop,
573 		.penaltydrop = q->stats.penaltydrop,
574 		.bucketdrop = q->stats.bucketdrop,
575 		.queuedrop = q->stats.queuedrop,
576 		.childdrop = q->stats.childdrop,
577 		.marked = q->stats.marked,
578 	};
579 
580 	st.maxqlen = sfb_compute_qlen(&st.maxprob, &st.avgprob, q);
581 
582 	return gnet_stats_copy_app(d, &st, sizeof(st));
583 }
584 
585 static int sfb_dump_class(struct Qdisc *sch, unsigned long cl,
586 			  struct sk_buff *skb, struct tcmsg *tcm)
587 {
588 	return -ENOSYS;
589 }
590 
591 static int sfb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
592 		     struct Qdisc **old)
593 {
594 	struct sfb_sched_data *q = qdisc_priv(sch);
595 
596 	if (new == NULL)
597 		new = &noop_qdisc;
598 
599 	sch_tree_lock(sch);
600 	*old = q->qdisc;
601 	q->qdisc = new;
602 	qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
603 	qdisc_reset(*old);
604 	sch_tree_unlock(sch);
605 	return 0;
606 }
607 
608 static struct Qdisc *sfb_leaf(struct Qdisc *sch, unsigned long arg)
609 {
610 	struct sfb_sched_data *q = qdisc_priv(sch);
611 
612 	return q->qdisc;
613 }
614 
615 static unsigned long sfb_get(struct Qdisc *sch, u32 classid)
616 {
617 	return 1;
618 }
619 
620 static void sfb_put(struct Qdisc *sch, unsigned long arg)
621 {
622 }
623 
624 static int sfb_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
625 			    struct nlattr **tca, unsigned long *arg)
626 {
627 	return -ENOSYS;
628 }
629 
630 static int sfb_delete(struct Qdisc *sch, unsigned long cl)
631 {
632 	return -ENOSYS;
633 }
634 
635 static void sfb_walk(struct Qdisc *sch, struct qdisc_walker *walker)
636 {
637 	if (!walker->stop) {
638 		if (walker->count >= walker->skip)
639 			if (walker->fn(sch, 1, walker) < 0) {
640 				walker->stop = 1;
641 				return;
642 			}
643 		walker->count++;
644 	}
645 }
646 
647 static struct tcf_proto **sfb_find_tcf(struct Qdisc *sch, unsigned long cl)
648 {
649 	struct sfb_sched_data *q = qdisc_priv(sch);
650 
651 	if (cl)
652 		return NULL;
653 	return &q->filter_list;
654 }
655 
656 static unsigned long sfb_bind(struct Qdisc *sch, unsigned long parent,
657 			      u32 classid)
658 {
659 	return 0;
660 }
661 
662 
663 static const struct Qdisc_class_ops sfb_class_ops = {
664 	.graft		=	sfb_graft,
665 	.leaf		=	sfb_leaf,
666 	.get		=	sfb_get,
667 	.put		=	sfb_put,
668 	.change		=	sfb_change_class,
669 	.delete		=	sfb_delete,
670 	.walk		=	sfb_walk,
671 	.tcf_chain	=	sfb_find_tcf,
672 	.bind_tcf	=	sfb_bind,
673 	.unbind_tcf	=	sfb_put,
674 	.dump		=	sfb_dump_class,
675 };
676 
677 static struct Qdisc_ops sfb_qdisc_ops __read_mostly = {
678 	.id		=	"sfb",
679 	.priv_size	=	sizeof(struct sfb_sched_data),
680 	.cl_ops		=	&sfb_class_ops,
681 	.enqueue	=	sfb_enqueue,
682 	.dequeue	=	sfb_dequeue,
683 	.peek		=	sfb_peek,
684 	.init		=	sfb_init,
685 	.reset		=	sfb_reset,
686 	.destroy	=	sfb_destroy,
687 	.change		=	sfb_change,
688 	.dump		=	sfb_dump,
689 	.dump_stats	=	sfb_dump_stats,
690 	.owner		=	THIS_MODULE,
691 };
692 
693 static int __init sfb_module_init(void)
694 {
695 	return register_qdisc(&sfb_qdisc_ops);
696 }
697 
698 static void __exit sfb_module_exit(void)
699 {
700 	unregister_qdisc(&sfb_qdisc_ops);
701 }
702 
703 module_init(sfb_module_init)
704 module_exit(sfb_module_exit)
705 
706 MODULE_DESCRIPTION("Stochastic Fair Blue queue discipline");
707 MODULE_AUTHOR("Juliusz Chroboczek");
708 MODULE_AUTHOR("Eric Dumazet");
709 MODULE_LICENSE("GPL");
710