[deliverable/linux.git] / net / sched / sch_sfb.c

/*
 * net/sched/sch_sfb.c	  Stochastic Fair Blue
 *
 * Copyright (c) 2008-2011 Juliusz Chroboczek <jch@pps.jussieu.fr>
 * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * version 2 as published by the Free Software Foundation.
 *
 * W. Feng, D. Kandlur, D. Saha, K. Shin. Blue:
 * A New Class of Active Queue Management Algorithms.
 * U. Michigan CSE-TR-387-99, April 1999.
 *
 * http://www.thefengs.com/wuchang/blue/CSE-TR-387-99.pdf
 *
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <linux/random.h>
#include <linux/jhash.h>
#include <net/ip.h>
#include <net/pkt_sched.h>
#include <net/inet_ecn.h>

/*
 * SFB uses two B[l][n] : L x N arrays of bins (L levels, N bins per level)
 * This implementation uses L = 8 and N = 16
 * This permits us to split one 32bit hash (provided per packet by rxhash or
 * external classifier) into 8 subhashes of 4 bits.
 */
#define SFB_BUCKET_SHIFT 4
#define SFB_NUMBUCKETS	(1 << SFB_BUCKET_SHIFT) /* N bins per Level */
#define SFB_BUCKET_MASK (SFB_NUMBUCKETS - 1)
#define SFB_LEVELS	(32 / SFB_BUCKET_SHIFT) /* L */

/* SFB algo uses a virtual queue, named "bin" */
struct sfb_bucket {
	u16		qlen; /* length of virtual queue */
	u16		p_mark; /* marking probability */
};

/* We use a double buffering right before hash change
 * (Section 4.4 of SFB reference : moving hash functions)
 */
struct sfb_bins {
	u32		  perturbation; /* jhash perturbation */
	struct sfb_bucket bins[SFB_LEVELS][SFB_NUMBUCKETS];
};

struct sfb_sched_data {
	struct Qdisc	*qdisc;
	struct tcf_proto __rcu *filter_list;
	unsigned long	rehash_interval;
	unsigned long	warmup_time;	/* double buffering warmup time in jiffies */
	u32		max;
	u32		bin_size;	/* maximum queue length per bin */
	u32		increment;	/* d1 */
	u32		decrement;	/* d2 */
	u32		limit;		/* HARD maximal queue length */
	u32		penalty_rate;
	u32		penalty_burst;
	u32		tokens_avail;
	unsigned long	rehash_time;
	unsigned long	token_time;

	u8		slot;		/* current active bins (0 or 1) */
	bool		double_buffering;
	struct sfb_bins bins[2];

	struct {
		u32	earlydrop;
		u32	penaltydrop;
		u32	bucketdrop;
		u32	queuedrop;
		u32	childdrop;	/* drops in child qdisc */
		u32	marked;		/* ECN mark */
	} stats;
};

/*
 * Each queued skb might be hashed on one or two bins
 * We store in skb_cb the two hash values.
 * (A zero value means double buffering was not used)
 */
struct sfb_skb_cb {
	u32 hashes[2];
};

static inline struct sfb_skb_cb *sfb_skb_cb(const struct sk_buff *skb)
{
	qdisc_cb_private_validate(skb, sizeof(struct sfb_skb_cb));
	return (struct sfb_skb_cb *)qdisc_skb_cb(skb)->data;
}

/*
 * If using 'internal' SFB flow classifier, hash comes from skb rxhash
 * If using external classifier, hash comes from the classid.
 */
static u32 sfb_hash(const struct sk_buff *skb, u32 slot)
{
	return sfb_skb_cb(skb)->hashes[slot];
}

/* Probabilities are coded as Q0.16 fixed-point values,
 * with 0xFFFF representing 65535/65536 (almost 1.0)
 * Addition and subtraction are saturating in [0, 65535]
 */
static u32 prob_plus(u32 p1, u32 p2)
{
	u32 res = p1 + p2;

	return min_t(u32, res, SFB_MAX_PROB);
}

static u32 prob_minus(u32 p1, u32 p2)
{
	return p1 > p2 ? p1 - p2 : 0;
}

static void increment_one_qlen(u32 sfbhash, u32 slot, struct sfb_sched_data *q)
{
	int i;
	struct sfb_bucket *b = &q->bins[slot].bins[0][0];

	for (i = 0; i < SFB_LEVELS; i++) {
		u32 hash = sfbhash & SFB_BUCKET_MASK;

		sfbhash >>= SFB_BUCKET_SHIFT;
		if (b[hash].qlen < 0xFFFF)
			b[hash].qlen++;
		b += SFB_NUMBUCKETS; /* next level */
	}
}

static void increment_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
{
	u32 sfbhash;

	sfbhash = sfb_hash(skb, 0);
	if (sfbhash)
		increment_one_qlen(sfbhash, 0, q);

	sfbhash = sfb_hash(skb, 1);
	if (sfbhash)
		increment_one_qlen(sfbhash, 1, q);
}

static void decrement_one_qlen(u32 sfbhash, u32 slot,
			       struct sfb_sched_data *q)
{
	int i;
	struct sfb_bucket *b = &q->bins[slot].bins[0][0];

	for (i = 0; i < SFB_LEVELS; i++) {
		u32 hash = sfbhash & SFB_BUCKET_MASK;

		sfbhash >>= SFB_BUCKET_SHIFT;
		if (b[hash].qlen > 0)
			b[hash].qlen--;
		b += SFB_NUMBUCKETS; /* next level */
	}
}

static void decrement_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
{
	u32 sfbhash;

	sfbhash = sfb_hash(skb, 0);
	if (sfbhash)
		decrement_one_qlen(sfbhash, 0, q);

	sfbhash = sfb_hash(skb, 1);
	if (sfbhash)
		decrement_one_qlen(sfbhash, 1, q);
}

static void decrement_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
{
	b->p_mark = prob_minus(b->p_mark, q->decrement);
}

static void increment_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
{
	b->p_mark = prob_plus(b->p_mark, q->increment);
}

static void sfb_zero_all_buckets(struct sfb_sched_data *q)
{
	memset(&q->bins, 0, sizeof(q->bins));
}

/*
 * compute max qlen, max p_mark, and avg p_mark
 */
static u32 sfb_compute_qlen(u32 *prob_r, u32 *avgpm_r, const struct sfb_sched_data *q)
{
	int i;
	u32 qlen = 0, prob = 0, totalpm = 0;
	const struct sfb_bucket *b = &q->bins[q->slot].bins[0][0];

	for (i = 0; i < SFB_LEVELS * SFB_NUMBUCKETS; i++) {
		if (qlen < b->qlen)
			qlen = b->qlen;
		totalpm += b->p_mark;
		if (prob < b->p_mark)
			prob = b->p_mark;
		b++;
	}
	*prob_r = prob;
	*avgpm_r = totalpm / (SFB_LEVELS * SFB_NUMBUCKETS);
	return qlen;
}


static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q)
{
	q->bins[slot].perturbation = prandom_u32();
}

static void sfb_swap_slot(struct sfb_sched_data *q)
{
	sfb_init_perturbation(q->slot, q);
	q->slot ^= 1;
	q->double_buffering = false;
}

/* Non elastic flows are allowed to use part of the bandwidth, expressed
 * in "penalty_rate" packets per second, with "penalty_burst" burst
 */
static bool sfb_rate_limit(struct sk_buff *skb, struct sfb_sched_data *q)
{
	if (q->penalty_rate == 0 || q->penalty_burst == 0)
		return true;

	if (q->tokens_avail < 1) {
		unsigned long age = min(10UL * HZ, jiffies - q->token_time);

		q->tokens_avail = (age * q->penalty_rate) / HZ;
		if (q->tokens_avail > q->penalty_burst)
			q->tokens_avail = q->penalty_burst;
		q->token_time = jiffies;
		if (q->tokens_avail < 1)
			return true;
	}

	q->tokens_avail--;
	return false;
}

static bool sfb_classify(struct sk_buff *skb, struct tcf_proto *fl,
			 int *qerr, u32 *salt)
{
	struct tcf_result res;
	int result;

	result = tc_classify(skb, fl, &res, false);
	if (result >= 0) {
#ifdef CONFIG_NET_CLS_ACT
		switch (result) {
		case TC_ACT_STOLEN:
		case TC_ACT_QUEUED:
			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
		case TC_ACT_SHOT:
			return false;
		}
#endif
		*salt = TC_H_MIN(res.classid);
		return true;
	}
	return false;
}

static int sfb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
{

	struct sfb_sched_data *q = qdisc_priv(sch);
	struct Qdisc *child = q->qdisc;
	struct tcf_proto *fl;
	int i;
	u32 p_min = ~0;
	u32 minqlen = ~0;
	u32 r, sfbhash;
	u32 slot = q->slot;
	int ret = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;

	if (unlikely(sch->q.qlen >= q->limit)) {
		qdisc_qstats_overlimit(sch);
		q->stats.queuedrop++;
		goto drop;
	}

	if (q->rehash_interval > 0) {
		unsigned long limit = q->rehash_time + q->rehash_interval;

		if (unlikely(time_after(jiffies, limit))) {
			sfb_swap_slot(q);
			q->rehash_time = jiffies;
		} else if (unlikely(!q->double_buffering && q->warmup_time > 0 &&
				    time_after(jiffies, limit - q->warmup_time))) {
			q->double_buffering = true;
		}
	}

	fl = rcu_dereference_bh(q->filter_list);
	if (fl) {
		u32 salt;

		/* If using external classifiers, get result and record it. */
		if (!sfb_classify(skb, fl, &ret, &salt))
			goto other_drop;
		sfbhash = jhash_1word(salt, q->bins[slot].perturbation);
	} else {
		sfbhash = skb_get_hash_perturb(skb, q->bins[slot].perturbation);
	}


	if (!sfbhash)
		sfbhash = 1;
	sfb_skb_cb(skb)->hashes[slot] = sfbhash;

	for (i = 0; i < SFB_LEVELS; i++) {
		u32 hash = sfbhash & SFB_BUCKET_MASK;
		struct sfb_bucket *b = &q->bins[slot].bins[i][hash];

		sfbhash >>= SFB_BUCKET_SHIFT;
		if (b->qlen == 0)
			decrement_prob(b, q);
		else if (b->qlen >= q->bin_size)
			increment_prob(b, q);
		if (minqlen > b->qlen)
			minqlen = b->qlen;
		if (p_min > b->p_mark)
			p_min = b->p_mark;
	}

	slot ^= 1;
	sfb_skb_cb(skb)->hashes[slot] = 0;

	if (unlikely(minqlen >= q->max)) {
		qdisc_qstats_overlimit(sch);
		q->stats.bucketdrop++;
		goto drop;
	}

	if (unlikely(p_min >= SFB_MAX_PROB)) {
		/* Inelastic flow */
		if (q->double_buffering) {
			sfbhash = skb_get_hash_perturb(skb,
			    q->bins[slot].perturbation);
			if (!sfbhash)
				sfbhash = 1;
			sfb_skb_cb(skb)->hashes[slot] = sfbhash;

			for (i = 0; i < SFB_LEVELS; i++) {
				u32 hash = sfbhash & SFB_BUCKET_MASK;
				struct sfb_bucket *b = &q->bins[slot].bins[i][hash];

				sfbhash >>= SFB_BUCKET_SHIFT;
				if (b->qlen == 0)
					decrement_prob(b, q);
				else if (b->qlen >= q->bin_size)
					increment_prob(b, q);
			}
		}
		if (sfb_rate_limit(skb, q)) {
			qdisc_qstats_overlimit(sch);
			q->stats.penaltydrop++;
			goto drop;
		}
		goto enqueue;
	}

	r = prandom_u32() & SFB_MAX_PROB;

	if (unlikely(r < p_min)) {
		if (unlikely(p_min > SFB_MAX_PROB / 2)) {
			/* If we're marking that many packets, then either
			 * this flow is unresponsive, or we're badly congested.
			 * In either case, we want to start dropping packets.
			 */
			if (r < (p_min - SFB_MAX_PROB / 2) * 2) {
				q->stats.earlydrop++;
				goto drop;
			}
		}
		if (INET_ECN_set_ce(skb)) {
			q->stats.marked++;
		} else {
			q->stats.earlydrop++;
			goto drop;
		}
	}

enqueue:
	ret = qdisc_enqueue(skb, child);
	if (likely(ret == NET_XMIT_SUCCESS)) {
		sch->q.qlen++;
		increment_qlen(skb, q);
	} else if (net_xmit_drop_count(ret)) {
		q->stats.childdrop++;
		qdisc_qstats_drop(sch);
	}
	return ret;

drop:
	qdisc_drop(skb, sch);
	return NET_XMIT_CN;
other_drop:
	if (ret & __NET_XMIT_BYPASS)
		qdisc_qstats_drop(sch);
	kfree_skb(skb);
	return ret;
}

static struct sk_buff *sfb_dequeue(struct Qdisc *sch)
{
	struct sfb_sched_data *q = qdisc_priv(sch);
	struct Qdisc *child = q->qdisc;
	struct sk_buff *skb;

	skb = child->dequeue(q->qdisc);

	if (skb) {
		qdisc_bstats_update(sch, skb);
		sch->q.qlen--;
		decrement_qlen(skb, q);
	}

	return skb;
}

static struct sk_buff *sfb_peek(struct Qdisc *sch)
{
	struct sfb_sched_data *q = qdisc_priv(sch);
	struct Qdisc *child = q->qdisc;

	return child->ops->peek(child);
}

/* No sfb_drop -- impossible since the child doesn't return the dropped skb. */

static void sfb_reset(struct Qdisc *sch)
{
	struct sfb_sched_data *q = qdisc_priv(sch);

	qdisc_reset(q->qdisc);
	sch->q.qlen = 0;
	q->slot = 0;
	q->double_buffering = false;
	sfb_zero_all_buckets(q);
	sfb_init_perturbation(0, q);
}

static void sfb_destroy(struct Qdisc *sch)
{
	struct sfb_sched_data *q = qdisc_priv(sch);

	tcf_destroy_chain(&q->filter_list);
	qdisc_destroy(q->qdisc);
}

static const struct nla_policy sfb_policy[TCA_SFB_MAX + 1] = {
	[TCA_SFB_PARMS]	= { .len = sizeof(struct tc_sfb_qopt) },
};

static const struct tc_sfb_qopt sfb_default_ops = {
	.rehash_interval = 600 * MSEC_PER_SEC,
	.warmup_time = 60 * MSEC_PER_SEC,
	.limit = 0,
	.max = 25,
	.bin_size = 20,
	.increment = (SFB_MAX_PROB + 500) / 1000, /* 0.1 % */
	.decrement = (SFB_MAX_PROB + 3000) / 6000,
	.penalty_rate = 10,
	.penalty_burst = 20,
};

static int sfb_change(struct Qdisc *sch, struct nlattr *opt)
{
	struct sfb_sched_data *q = qdisc_priv(sch);
	struct Qdisc *child;
	struct nlattr *tb[TCA_SFB_MAX + 1];
	const struct tc_sfb_qopt *ctl = &sfb_default_ops;
	u32 limit;
	int err;

	if (opt) {
		err = nla_parse_nested(tb, TCA_SFB_MAX, opt, sfb_policy);
		if (err < 0)
			return -EINVAL;

		if (tb[TCA_SFB_PARMS] == NULL)
			return -EINVAL;

		ctl = nla_data(tb[TCA_SFB_PARMS]);
	}

	limit = ctl->limit;
	if (limit == 0)
		limit = qdisc_dev(sch)->tx_queue_len;

	child = fifo_create_dflt(sch, &pfifo_qdisc_ops, limit);
	if (IS_ERR(child))
		return PTR_ERR(child);

	sch_tree_lock(sch);

	qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen,
				  q->qdisc->qstats.backlog);
	qdisc_destroy(q->qdisc);
	q->qdisc = child;

	q->rehash_interval = msecs_to_jiffies(ctl->rehash_interval);
	q->warmup_time = msecs_to_jiffies(ctl->warmup_time);
	q->rehash_time = jiffies;
	q->limit = limit;
	q->increment = ctl->increment;
	q->decrement = ctl->decrement;
	q->max = ctl->max;
	q->bin_size = ctl->bin_size;
	q->penalty_rate = ctl->penalty_rate;
	q->penalty_burst = ctl->penalty_burst;
	q->tokens_avail = ctl->penalty_burst;
	q->token_time = jiffies;

	q->slot = 0;
	q->double_buffering = false;
	sfb_zero_all_buckets(q);
	sfb_init_perturbation(0, q);
	sfb_init_perturbation(1, q);

	sch_tree_unlock(sch);

	return 0;
}

static int sfb_init(struct Qdisc *sch, struct nlattr *opt)
{
	struct sfb_sched_data *q = qdisc_priv(sch);

	q->qdisc = &noop_qdisc;
	return sfb_change(sch, opt);
}

static int sfb_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct sfb_sched_data *q = qdisc_priv(sch);
	struct nlattr *opts;
	struct tc_sfb_qopt opt = {
		.rehash_interval = jiffies_to_msecs(q->rehash_interval),
		.warmup_time = jiffies_to_msecs(q->warmup_time),
		.limit = q->limit,
		.max = q->max,
		.bin_size = q->bin_size,
		.increment = q->increment,
		.decrement = q->decrement,
		.penalty_rate = q->penalty_rate,
		.penalty_burst = q->penalty_burst,
	};

	sch->qstats.backlog = q->qdisc->qstats.backlog;
	opts = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;
	if (nla_put(skb, TCA_SFB_PARMS, sizeof(opt), &opt))
		goto nla_put_failure;
	return nla_nest_end(skb, opts);

nla_put_failure:
	nla_nest_cancel(skb, opts);
	return -EMSGSIZE;
}

static int sfb_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
{
	struct sfb_sched_data *q = qdisc_priv(sch);
	struct tc_sfb_xstats st = {
		.earlydrop = q->stats.earlydrop,
		.penaltydrop = q->stats.penaltydrop,
		.bucketdrop = q->stats.bucketdrop,
		.queuedrop = q->stats.queuedrop,
		.childdrop = q->stats.childdrop,
		.marked = q->stats.marked,
	};

	st.maxqlen = sfb_compute_qlen(&st.maxprob, &st.avgprob, q);

	return gnet_stats_copy_app(d, &st, sizeof(st));
}

static int sfb_dump_class(struct Qdisc *sch, unsigned long cl,
			  struct sk_buff *skb, struct tcmsg *tcm)
{
	return -ENOSYS;
}

static int sfb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
		     struct Qdisc **old)
{
	struct sfb_sched_data *q = qdisc_priv(sch);

	if (new == NULL)
		new = &noop_qdisc;

	*old = qdisc_replace(sch, new, &q->qdisc);
	return 0;
}

static struct Qdisc *sfb_leaf(struct Qdisc *sch, unsigned long arg)
{
	struct sfb_sched_data *q = qdisc_priv(sch);

	return q->qdisc;
}

static unsigned long sfb_get(struct Qdisc *sch, u32 classid)
{
	return 1;
}

static void sfb_put(struct Qdisc *sch, unsigned long arg)
{
}

static int sfb_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
			    struct nlattr **tca, unsigned long *arg)
{
	return -ENOSYS;
}

static int sfb_delete(struct Qdisc *sch, unsigned long cl)
{
	return -ENOSYS;
}

static void sfb_walk(struct Qdisc *sch, struct qdisc_walker *walker)
{
	if (!walker->stop) {
		if (walker->count >= walker->skip)
			if (walker->fn(sch, 1, walker) < 0) {
				walker->stop = 1;
				return;
			}
		walker->count++;
	}
}

static struct tcf_proto __rcu **sfb_find_tcf(struct Qdisc *sch,
					     unsigned long cl)
{
	struct sfb_sched_data *q = qdisc_priv(sch);

	if (cl)
		return NULL;
	return &q->filter_list;
}

static unsigned long sfb_bind(struct Qdisc *sch, unsigned long parent,
			      u32 classid)
{
	return 0;
}


static const struct Qdisc_class_ops sfb_class_ops = {
	.graft		=	sfb_graft,
	.leaf		=	sfb_leaf,
	.get		=	sfb_get,
	.put		=	sfb_put,
	.change		=	sfb_change_class,
	.delete		=	sfb_delete,
	.walk		=	sfb_walk,
	.tcf_chain	=	sfb_find_tcf,
	.bind_tcf	=	sfb_bind,
	.unbind_tcf	=	sfb_put,
	.dump		=	sfb_dump_class,
};

static struct Qdisc_ops sfb_qdisc_ops __read_mostly = {
	.id		=	"sfb",
	.priv_size	=	sizeof(struct sfb_sched_data),
	.cl_ops		=	&sfb_class_ops,
	.enqueue	=	sfb_enqueue,
	.dequeue	=	sfb_dequeue,
	.peek		=	sfb_peek,
	.init		=	sfb_init,
	.reset		=	sfb_reset,
	.destroy	=	sfb_destroy,
	.change		=	sfb_change,
	.dump		=	sfb_dump,
	.dump_stats	=	sfb_dump_stats,
	.owner		=	THIS_MODULE,
};

static int __init sfb_module_init(void)
{
	return register_qdisc(&sfb_qdisc_ops);
}

static void __exit sfb_module_exit(void)
{
	unregister_qdisc(&sfb_qdisc_ops);
}

module_init(sfb_module_init)
module_exit(sfb_module_exit)

MODULE_DESCRIPTION("Stochastic Fair Blue queue discipline");
MODULE_AUTHOR("Juliusz Chroboczek");
MODULE_AUTHOR("Eric Dumazet");
MODULE_LICENSE("GPL");
Commit	Line	Data
e13e02a3 ED	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;
25d8c0d5	57	struct tcf_proto __rcu *filter_list;
e13e02a3 ED	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	{
16bda13d	96	qdisc_cb_private_validate(skb, sizeof(struct sfb_skb_cb));
e13e02a3 ED	97	return (struct sfb_skb_cb *)qdisc_skb_cb(skb)->data;
	98	}
	99
	100	/*
	101	* If using 'internal' SFB flow classifier, hash comes from skb rxhash
	102	* If using external classifier, hash comes from the classid.
	103	*/
	104	static u32 sfb_hash(const struct sk_buff *skb, u32 slot)
	105	{
	106	return sfb_skb_cb(skb)->hashes[slot];
	107	}
	108
	109	/* Probabilities are coded as Q0.16 fixed-point values,
	110	* with 0xFFFF representing 65535/65536 (almost 1.0)
	111	* Addition and subtraction are saturating in [0, 65535]
	112	*/
	113	static u32 prob_plus(u32 p1, u32 p2)
	114	{
	115	u32 res = p1 + p2;
	116
	117	return min_t(u32, res, SFB_MAX_PROB);
	118	}
	119
	120	static u32 prob_minus(u32 p1, u32 p2)
	121	{
	122	return p1 > p2 ? p1 - p2 : 0;
	123	}
	124
	125	static void increment_one_qlen(u32 sfbhash, u32 slot, struct sfb_sched_data *q)
	126	{
	127	int i;
	128	struct sfb_bucket *b = &q->bins[slot].bins[0][0];
	129
	130	for (i = 0; i < SFB_LEVELS; i++) {
	131	u32 hash = sfbhash & SFB_BUCKET_MASK;
	132
	133	sfbhash >>= SFB_BUCKET_SHIFT;
	134	if (b[hash].qlen < 0xFFFF)
	135	b[hash].qlen++;
	136	b += SFB_NUMBUCKETS; /* next level */
	137	}
	138	}
	139
	140	static void increment_qlen(const struct sk_buff skb, struct sfb_sched_data q)
	141	{
	142	u32 sfbhash;
	143
	144	sfbhash = sfb_hash(skb, 0);
	145	if (sfbhash)
	146	increment_one_qlen(sfbhash, 0, q);
	147
	148	sfbhash = sfb_hash(skb, 1);
	149	if (sfbhash)
	150	increment_one_qlen(sfbhash, 1, q);
	151	}
	152
	153	static void decrement_one_qlen(u32 sfbhash, u32 slot,
	154	struct sfb_sched_data *q)
	155	{
	156	int i;
	157	struct sfb_bucket *b = &q->bins[slot].bins[0][0];
	158
	159	for (i = 0; i < SFB_LEVELS; i++) {
	160	u32 hash = sfbhash & SFB_BUCKET_MASK;
161
162	sfbhash >>= SFB_BUCKET_SHIFT;
163	if (b[hash].qlen > 0)
164	b[hash].qlen--;
165	b += SFB_NUMBUCKETS; /* next level */
166	}
167	}
168
169	static void decrement_qlen(const struct sk_buff skb, struct sfb_sched_data q)
170	{
171	u32 sfbhash;
172
173	sfbhash = sfb_hash(skb, 0);
174	if (sfbhash)
175	decrement_one_qlen(sfbhash, 0, q);
176
177	sfbhash = sfb_hash(skb, 1);
178	if (sfbhash)
179	decrement_one_qlen(sfbhash, 1, q);
180	}
181
182	static void decrement_prob(struct sfb_bucket b, struct sfb_sched_data q)
183	{
184	b->p_mark = prob_minus(b->p_mark, q->decrement);
185	}
186
187	static void increment_prob(struct sfb_bucket b, struct sfb_sched_data q)
188	{
189	b->p_mark = prob_plus(b->p_mark, q->increment);
190	}
191
192	static void sfb_zero_all_buckets(struct sfb_sched_data *q)
193	{
194	memset(&q->bins, 0, sizeof(q->bins));
195	}
196
197	/*
198	* compute max qlen, max p_mark, and avg p_mark
199	*/
200	static u32 sfb_compute_qlen(u32 prob_r, u32 avgpm_r, const struct sfb_sched_data *q)
201	{
202	int i;
203	u32 qlen = 0, prob = 0, totalpm = 0;
204	const struct sfb_bucket *b = &q->bins[q->slot].bins[0][0];
205
206	for (i = 0; i < SFB_LEVELS * SFB_NUMBUCKETS; i++) {
207	if (qlen < b->qlen)
208	qlen = b->qlen;
209	totalpm += b->p_mark;
210	if (prob < b->p_mark)
211	prob = b->p_mark;
212	b++;
213	}
214	*prob_r = prob;
215	avgpm_r = totalpm / (SFB_LEVELS SFB_NUMBUCKETS);
216	return qlen;
217	}
218
219
220	static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q)
221	{
63862b5b	222	q->bins[slot].perturbation = prandom_u32();
e13e02a3 ED	223	}
	224
	225	static void sfb_swap_slot(struct sfb_sched_data *q)
	226	{
	227	sfb_init_perturbation(q->slot, q);
	228	q->slot ^= 1;
	229	q->double_buffering = false;
	230	}
	231
	232	/* Non elastic flows are allowed to use part of the bandwidth, expressed
	233	* in "penalty_rate" packets per second, with "penalty_burst" burst
	234	*/
	235	static bool sfb_rate_limit(struct sk_buff skb, struct sfb_sched_data q)
	236	{
	237	if (q->penalty_rate == 0 \|\| q->penalty_burst == 0)
	238	return true;
	239
	240	if (q->tokens_avail < 1) {
	241	unsigned long age = min(10UL * HZ, jiffies - q->token_time);
	242
	243	q->tokens_avail = (age * q->penalty_rate) / HZ;
	244	if (q->tokens_avail > q->penalty_burst)
	245	q->tokens_avail = q->penalty_burst;
	246	q->token_time = jiffies;
	247	if (q->tokens_avail < 1)
	248	return true;
	249	}
	250
	251	q->tokens_avail--;
	252	return false;
	253	}
	254
25d8c0d5	255	static bool sfb_classify(struct sk_buff skb, struct tcf_proto fl,
e13e02a3 ED	256	int qerr, u32 salt)
	257	{
	258	struct tcf_result res;
	259	int result;
	260
3b3ae880	261	result = tc_classify(skb, fl, &res, false);
e13e02a3 ED	262	if (result >= 0) {
	263	#ifdef CONFIG_NET_CLS_ACT
	264	switch (result) {
	265	case TC_ACT_STOLEN:
	266	case TC_ACT_QUEUED:
	267	*qerr = NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
	268	case TC_ACT_SHOT:
	269	return false;
	270	}
	271	#endif
	272	*salt = TC_H_MIN(res.classid);
	273	return true;
	274	}
	275	return false;
	276	}
	277
	278	static int sfb_enqueue(struct sk_buff skb, struct Qdisc sch)
	279	{
	280
	281	struct sfb_sched_data *q = qdisc_priv(sch);
	282	struct Qdisc *child = q->qdisc;
25d8c0d5	283	struct tcf_proto *fl;
e13e02a3 ED	284	int i;
	285	u32 p_min = ~0;
	286	u32 minqlen = ~0;
63c0ad4d TH	287	u32 r, sfbhash;
63c0ad4d TH	288	u32 slot = q->slot;
e13e02a3 ED	289	int ret = NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
e13e02a3 ED	290
363437f4	291	if (unlikely(sch->q.qlen >= q->limit)) {
25331d6c	292	qdisc_qstats_overlimit(sch);
363437f4 ED	293	q->stats.queuedrop++;
	294	goto drop;
	295	}
	296
e13e02a3 ED	297	if (q->rehash_interval > 0) {
	298	unsigned long limit = q->rehash_time + q->rehash_interval;
	299
	300	if (unlikely(time_after(jiffies, limit))) {
	301	sfb_swap_slot(q);
	302	q->rehash_time = jiffies;
	303	} else if (unlikely(!q->double_buffering && q->warmup_time > 0 &&
	304	time_after(jiffies, limit - q->warmup_time))) {
	305	q->double_buffering = true;
	306	}
	307	}
	308
25d8c0d5 JF	309	fl = rcu_dereference_bh(q->filter_list);
25d8c0d5 JF	310	if (fl) {
63c0ad4d TH	311	u32 salt;
63c0ad4d TH	312
e13e02a3	313	/* If using external classifiers, get result and record it. */
25d8c0d5	314	if (!sfb_classify(skb, fl, &ret, &salt))
e13e02a3	315	goto other_drop;
63c0ad4d	316	sfbhash = jhash_1word(salt, q->bins[slot].perturbation);
e13e02a3	317	} else {
63c0ad4d	318	sfbhash = skb_get_hash_perturb(skb, q->bins[slot].perturbation);
e13e02a3 ED	319	}
e13e02a3 ED	320
e13e02a3	321
e13e02a3 ED	322	if (!sfbhash)
	323	sfbhash = 1;
	324	sfb_skb_cb(skb)->hashes[slot] = sfbhash;
	325
	326	for (i = 0; i < SFB_LEVELS; i++) {
	327	u32 hash = sfbhash & SFB_BUCKET_MASK;
	328	struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
	329
	330	sfbhash >>= SFB_BUCKET_SHIFT;
	331	if (b->qlen == 0)
	332	decrement_prob(b, q);
	333	else if (b->qlen >= q->bin_size)
	334	increment_prob(b, q);
	335	if (minqlen > b->qlen)
	336	minqlen = b->qlen;
	337	if (p_min > b->p_mark)
	338	p_min = b->p_mark;
	339	}
	340
	341	slot ^= 1;
	342	sfb_skb_cb(skb)->hashes[slot] = 0;
	343
363437f4	344	if (unlikely(minqlen >= q->max)) {
25331d6c	345	qdisc_qstats_overlimit(sch);
363437f4	346	q->stats.bucketdrop++;
e13e02a3 ED	347	goto drop;
	348	}
	349
	350	if (unlikely(p_min >= SFB_MAX_PROB)) {
	351	/* Inelastic flow */
	352	if (q->double_buffering) {
63c0ad4d TH	353	sfbhash = skb_get_hash_perturb(skb,
63c0ad4d TH	354	q->bins[slot].perturbation);
e13e02a3 ED	355	if (!sfbhash)
	356	sfbhash = 1;
	357	sfb_skb_cb(skb)->hashes[slot] = sfbhash;
	358
	359	for (i = 0; i < SFB_LEVELS; i++) {
	360	u32 hash = sfbhash & SFB_BUCKET_MASK;
	361	struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
	362
	363	sfbhash >>= SFB_BUCKET_SHIFT;
	364	if (b->qlen == 0)
	365	decrement_prob(b, q);
	366	else if (b->qlen >= q->bin_size)
	367	increment_prob(b, q);
	368	}
	369	}
	370	if (sfb_rate_limit(skb, q)) {
25331d6c	371	qdisc_qstats_overlimit(sch);
e13e02a3 ED	372	q->stats.penaltydrop++;
	373	goto drop;
	374	}
	375	goto enqueue;
	376	}
	377
63862b5b	378	r = prandom_u32() & SFB_MAX_PROB;
e13e02a3 ED	379
	380	if (unlikely(r < p_min)) {
	381	if (unlikely(p_min > SFB_MAX_PROB / 2)) {
	382	/* If we're marking that many packets, then either
	383	* this flow is unresponsive, or we're badly congested.
	384	* In either case, we want to start dropping packets.
	385	*/
	386	if (r < (p_min - SFB_MAX_PROB / 2) * 2) {
	387	q->stats.earlydrop++;
	388	goto drop;
	389	}
	390	}
	391	if (INET_ECN_set_ce(skb)) {
	392	q->stats.marked++;
	393	} else {
	394	q->stats.earlydrop++;
	395	goto drop;
	396	}
	397	}
	398
	399	enqueue:
	400	ret = qdisc_enqueue(skb, child);
	401	if (likely(ret == NET_XMIT_SUCCESS)) {
	402	sch->q.qlen++;
	403	increment_qlen(skb, q);
	404	} else if (net_xmit_drop_count(ret)) {
	405	q->stats.childdrop++;
25331d6c	406	qdisc_qstats_drop(sch);
e13e02a3 ED	407	}
	408	return ret;
	409
	410	drop:
	411	qdisc_drop(skb, sch);
	412	return NET_XMIT_CN;
	413	other_drop:
	414	if (ret & __NET_XMIT_BYPASS)
25331d6c	415	qdisc_qstats_drop(sch);
e13e02a3 ED	416	kfree_skb(skb);
	417	return ret;
	418	}
	419
	420	static struct sk_buff sfb_dequeue(struct Qdisc sch)
	421	{
	422	struct sfb_sched_data *q = qdisc_priv(sch);
	423	struct Qdisc *child = q->qdisc;
	424	struct sk_buff *skb;
	425
	426	skb = child->dequeue(q->qdisc);
	427
	428	if (skb) {
	429	qdisc_bstats_update(sch, skb);
	430	sch->q.qlen--;
	431	decrement_qlen(skb, q);
	432	}
	433
	434	return skb;
	435	}
	436
	437	static struct sk_buff sfb_peek(struct Qdisc sch)
	438	{
	439	struct sfb_sched_data *q = qdisc_priv(sch);
	440	struct Qdisc *child = q->qdisc;
	441
	442	return child->ops->peek(child);
	443	}
	444
	445	/* No sfb_drop -- impossible since the child doesn't return the dropped skb. */
	446
	447	static void sfb_reset(struct Qdisc *sch)
	448	{
	449	struct sfb_sched_data *q = qdisc_priv(sch);
	450
	451	qdisc_reset(q->qdisc);
	452	sch->q.qlen = 0;
	453	q->slot = 0;
	454	q->double_buffering = false;
	455	sfb_zero_all_buckets(q);
	456	sfb_init_perturbation(0, q);
	457	}
	458
	459	static void sfb_destroy(struct Qdisc *sch)
	460	{
	461	struct sfb_sched_data *q = qdisc_priv(sch);
	462
	463	tcf_destroy_chain(&q->filter_list);
	464	qdisc_destroy(q->qdisc);
	465	}
	466
	467	static const struct nla_policy sfb_policy[TCA_SFB_MAX + 1] = {
	468	[TCA_SFB_PARMS] = { .len = sizeof(struct tc_sfb_qopt) },
	469	};
	470
	471	static const struct tc_sfb_qopt sfb_default_ops = {
	472	.rehash_interval = 600 * MSEC_PER_SEC,
	473	.warmup_time = 60 * MSEC_PER_SEC,
	474	.limit = 0,
	475	.max = 25,
	476	.bin_size = 20,
	477	.increment = (SFB_MAX_PROB + 500) / 1000, /* 0.1 % */
	478	.decrement = (SFB_MAX_PROB + 3000) / 6000,
	479	.penalty_rate = 10,
480	.penalty_burst = 20,
481	};
482
483	static int sfb_change(struct Qdisc sch, struct nlattr opt)
484	{
485	struct sfb_sched_data *q = qdisc_priv(sch);
486	struct Qdisc *child;
487	struct nlattr *tb[TCA_SFB_MAX + 1];
488	const struct tc_sfb_qopt *ctl = &sfb_default_ops;
489	u32 limit;
490	int err;
491
492	if (opt) {
493	err = nla_parse_nested(tb, TCA_SFB_MAX, opt, sfb_policy);
494	if (err < 0)
495	return -EINVAL;
496
497	if (tb[TCA_SFB_PARMS] == NULL)
498	return -EINVAL;
499
500	ctl = nla_data(tb[TCA_SFB_PARMS]);
501	}
502
503	limit = ctl->limit;
504	if (limit == 0)
348e3435	505	limit = qdisc_dev(sch)->tx_queue_len;
e13e02a3 ED	506
	507	child = fifo_create_dflt(sch, &pfifo_qdisc_ops, limit);
	508	if (IS_ERR(child))
	509	return PTR_ERR(child);
	510
	511	sch_tree_lock(sch);
	512
2ccccf5f WC	513	qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen,
2ccccf5f WC	514	q->qdisc->qstats.backlog);
e13e02a3 ED	515	qdisc_destroy(q->qdisc);
	516	q->qdisc = child;
	517
	518	q->rehash_interval = msecs_to_jiffies(ctl->rehash_interval);
	519	q->warmup_time = msecs_to_jiffies(ctl->warmup_time);
	520	q->rehash_time = jiffies;
	521	q->limit = limit;
	522	q->increment = ctl->increment;
	523	q->decrement = ctl->decrement;
	524	q->max = ctl->max;
	525	q->bin_size = ctl->bin_size;
	526	q->penalty_rate = ctl->penalty_rate;
	527	q->penalty_burst = ctl->penalty_burst;
	528	q->tokens_avail = ctl->penalty_burst;
	529	q->token_time = jiffies;
	530
	531	q->slot = 0;
	532	q->double_buffering = false;
	533	sfb_zero_all_buckets(q);
	534	sfb_init_perturbation(0, q);
	535	sfb_init_perturbation(1, q);
	536
	537	sch_tree_unlock(sch);
	538
	539	return 0;
	540	}
	541
	542	static int sfb_init(struct Qdisc sch, struct nlattr opt)
	543	{
	544	struct sfb_sched_data *q = qdisc_priv(sch);
	545
	546	q->qdisc = &noop_qdisc;
	547	return sfb_change(sch, opt);
	548	}
	549
	550	static int sfb_dump(struct Qdisc sch, struct sk_buff skb)
	551	{
	552	struct sfb_sched_data *q = qdisc_priv(sch);
	553	struct nlattr *opts;
	554	struct tc_sfb_qopt opt = {
	555	.rehash_interval = jiffies_to_msecs(q->rehash_interval),
	556	.warmup_time = jiffies_to_msecs(q->warmup_time),
	557	.limit = q->limit,
	558	.max = q->max,
	559	.bin_size = q->bin_size,
	560	.increment = q->increment,
	561	.decrement = q->decrement,
	562	.penalty_rate = q->penalty_rate,
	563	.penalty_burst = q->penalty_burst,
	564	};
	565
	566	sch->qstats.backlog = q->qdisc->qstats.backlog;
	567	opts = nla_nest_start(skb, TCA_OPTIONS);
7ac2908e AC	568	if (opts == NULL)
7ac2908e AC	569	goto nla_put_failure;
1b34ec43 DM	570	if (nla_put(skb, TCA_SFB_PARMS, sizeof(opt), &opt))
1b34ec43 DM	571	goto nla_put_failure;
e13e02a3 ED	572	return nla_nest_end(skb, opts);
	573
	574	nla_put_failure:
	575	nla_nest_cancel(skb, opts);
	576	return -EMSGSIZE;
	577	}
	578
	579	static int sfb_dump_stats(struct Qdisc sch, struct gnet_dump d)
	580	{
	581	struct sfb_sched_data *q = qdisc_priv(sch);
	582	struct tc_sfb_xstats st = {
	583	.earlydrop = q->stats.earlydrop,
	584	.penaltydrop = q->stats.penaltydrop,
	585	.bucketdrop = q->stats.bucketdrop,
	586	.queuedrop = q->stats.queuedrop,
	587	.childdrop = q->stats.childdrop,
	588	.marked = q->stats.marked,
	589	};
	590
	591	st.maxqlen = sfb_compute_qlen(&st.maxprob, &st.avgprob, q);
	592
	593	return gnet_stats_copy_app(d, &st, sizeof(st));
	594	}
	595
	596	static int sfb_dump_class(struct Qdisc *sch, unsigned long cl,
	597	struct sk_buff skb, struct tcmsg tcm)
	598	{
	599	return -ENOSYS;
	600	}
	601
	602	static int sfb_graft(struct Qdisc sch, unsigned long arg, struct Qdisc new,
	603	struct Qdisc **old)
	604	{
	605	struct sfb_sched_data *q = qdisc_priv(sch);
	606
	607	if (new == NULL)
	608	new = &noop_qdisc;
	609
86a7996c	610	*old = qdisc_replace(sch, new, &q->qdisc);
e13e02a3 ED	611	return 0;
	612	}
	613
	614	static struct Qdisc sfb_leaf(struct Qdisc sch, unsigned long arg)
	615	{
	616	struct sfb_sched_data *q = qdisc_priv(sch);
	617
	618	return q->qdisc;
	619	}
	620
	621	static unsigned long sfb_get(struct Qdisc *sch, u32 classid)
	622	{
	623	return 1;
	624	}
	625
	626	static void sfb_put(struct Qdisc *sch, unsigned long arg)
	627	{
	628	}
	629
	630	static int sfb_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
	631	struct nlattr *tca, unsigned long arg)
	632	{
	633	return -ENOSYS;
	634	}
	635
	636	static int sfb_delete(struct Qdisc *sch, unsigned long cl)
	637	{
	638	return -ENOSYS;
	639	}
	640
	641	static void sfb_walk(struct Qdisc sch, struct qdisc_walker walker)
	642	{
	643	if (!walker->stop) {
	644	if (walker->count >= walker->skip)
	645	if (walker->fn(sch, 1, walker) < 0) {
	646	walker->stop = 1;
	647	return;
	648	}
	649	walker->count++;
	650	}
	651	}
	652
25d8c0d5 JF	653	static struct tcf_proto __rcu *sfb_find_tcf(struct Qdisc sch,
25d8c0d5 JF	654	unsigned long cl)
e13e02a3 ED	655	{
	656	struct sfb_sched_data *q = qdisc_priv(sch);
	657
	658	if (cl)
	659	return NULL;
	660	return &q->filter_list;
	661	}
	662
	663	static unsigned long sfb_bind(struct Qdisc *sch, unsigned long parent,
	664	u32 classid)
	665	{
	666	return 0;
	667	}
	668
	669
	670	static const struct Qdisc_class_ops sfb_class_ops = {
	671	.graft = sfb_graft,
	672	.leaf = sfb_leaf,
	673	.get = sfb_get,
	674	.put = sfb_put,
	675	.change = sfb_change_class,
	676	.delete = sfb_delete,
	677	.walk = sfb_walk,
	678	.tcf_chain = sfb_find_tcf,
	679	.bind_tcf = sfb_bind,
	680	.unbind_tcf = sfb_put,
	681	.dump = sfb_dump_class,
	682	};
	683
	684	static struct Qdisc_ops sfb_qdisc_ops __read_mostly = {
	685	.id = "sfb",
	686	.priv_size = sizeof(struct sfb_sched_data),
	687	.cl_ops = &sfb_class_ops,
	688	.enqueue = sfb_enqueue,
	689	.dequeue = sfb_dequeue,
	690	.peek = sfb_peek,
	691	.init = sfb_init,
	692	.reset = sfb_reset,
	693	.destroy = sfb_destroy,
	694	.change = sfb_change,
	695	.dump = sfb_dump,
	696	.dump_stats = sfb_dump_stats,
	697	.owner = THIS_MODULE,
	698	};
	699
	700	static int __init sfb_module_init(void)
	701	{
	702	return register_qdisc(&sfb_qdisc_ops);
	703	}
	704
	705	static void __exit sfb_module_exit(void)
	706	{
	707	unregister_qdisc(&sfb_qdisc_ops);
	708	}
	709
	710	module_init(sfb_module_init)
	711	module_exit(sfb_module_exit)
	712
	713	MODULE_DESCRIPTION("Stochastic Fair Blue queue discipline");
	714	MODULE_AUTHOR("Juliusz Chroboczek");
	715	MODULE_AUTHOR("Eric Dumazet");
	716	MODULE_LICENSE("GPL");