[linux-2.6-block.git] / net / sched / sch_fq_codel.c

/*
 * Fair Queue CoDel discipline
 *
 *	This program is free software; you can redistribute it and/or
 *	modify it under the terms of the GNU General Public License
 *	as published by the Free Software Foundation; either version
 *	2 of the License, or (at your option) any later version.
 *
 *  Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/jiffies.h>
#include <linux/string.h>
#include <linux/in.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/skbuff.h>
#include <linux/jhash.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <net/netlink.h>
#include <net/pkt_sched.h>
#include <net/codel.h>

/*	Fair Queue CoDel.
 *
 * Principles :
 * Packets are classified (internal classifier or external) on flows.
 * This is a Stochastic model (as we use a hash, several flows
 *			       might be hashed on same slot)
 * Each flow has a CoDel managed queue.
 * Flows are linked onto two (Round Robin) lists,
 * so that new flows have priority on old ones.
 *
 * For a given flow, packets are not reordered (CoDel uses a FIFO)
 * head drops only.
 * ECN capability is on by default.
 * Low memory footprint (64 bytes per flow)
 */

struct fq_codel_flow {
	struct sk_buff	  *head;
	struct sk_buff	  *tail;
	struct list_head  flowchain;
	int		  deficit;
	u32		  dropped; /* number of drops (or ECN marks) on this flow */
	struct codel_vars cvars;
}; /* please try to keep this structure <= 64 bytes */

struct fq_codel_sched_data {
	struct tcf_proto __rcu *filter_list; /* optional external classifier */
	struct fq_codel_flow *flows;	/* Flows table [flows_cnt] */
	u32		*backlogs;	/* backlog table [flows_cnt] */
	u32		flows_cnt;	/* number of flows */
	u32		perturbation;	/* hash perturbation */
	u32		quantum;	/* psched_mtu(qdisc_dev(sch)); */
	struct codel_params cparams;
	struct codel_stats cstats;
	u32		drop_overlimit;
	u32		new_flow_count;

	struct list_head new_flows;	/* list of new flows */
	struct list_head old_flows;	/* list of old flows */
};

static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q,
				  struct sk_buff *skb)
{
	u32 hash = skb_get_hash_perturb(skb, q->perturbation);

	return reciprocal_scale(hash, q->flows_cnt);
}

static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch,
				      int *qerr)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct tcf_proto *filter;
	struct tcf_result res;
	int result;

	if (TC_H_MAJ(skb->priority) == sch->handle &&
	    TC_H_MIN(skb->priority) > 0 &&
	    TC_H_MIN(skb->priority) <= q->flows_cnt)
		return TC_H_MIN(skb->priority);

	filter = rcu_dereference_bh(q->filter_list);
	if (!filter)
		return fq_codel_hash(q, skb) + 1;

	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
	result = tc_classify(skb, filter, &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 0;
		}
#endif
		if (TC_H_MIN(res.classid) <= q->flows_cnt)
			return TC_H_MIN(res.classid);
	}
	return 0;
}

/* helper functions : might be changed when/if skb use a standard list_head */

/* remove one skb from head of slot queue */
static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow)
{
	struct sk_buff *skb = flow->head;

	flow->head = skb->next;
	skb->next = NULL;
	return skb;
}

/* add skb to flow queue (tail add) */
static inline void flow_queue_add(struct fq_codel_flow *flow,
				  struct sk_buff *skb)
{
	if (flow->head == NULL)
		flow->head = skb;
	else
		flow->tail->next = skb;
	flow->tail = skb;
	skb->next = NULL;
}

static unsigned int fq_codel_drop(struct Qdisc *sch)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct sk_buff *skb;
	unsigned int maxbacklog = 0, idx = 0, i, len;
	struct fq_codel_flow *flow;

	/* Queue is full! Find the fat flow and drop packet from it.
	 * This might sound expensive, but with 1024 flows, we scan
	 * 4KB of memory, and we dont need to handle a complex tree
	 * in fast path (packet queue/enqueue) with many cache misses.
	 */
	for (i = 0; i < q->flows_cnt; i++) {
		if (q->backlogs[i] > maxbacklog) {
			maxbacklog = q->backlogs[i];
			idx = i;
		}
	}
	flow = &q->flows[idx];
	skb = dequeue_head(flow);
	len = qdisc_pkt_len(skb);
	q->backlogs[idx] -= len;
	sch->q.qlen--;
	qdisc_qstats_drop(sch);
	qdisc_qstats_backlog_dec(sch, skb);
	kfree_skb(skb);
	flow->dropped++;
	return idx;
}

static unsigned int fq_codel_qdisc_drop(struct Qdisc *sch)
{
	unsigned int prev_backlog;

	prev_backlog = sch->qstats.backlog;
	fq_codel_drop(sch);
	return prev_backlog - sch->qstats.backlog;
}

static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	unsigned int idx, prev_backlog;
	struct fq_codel_flow *flow;
	int uninitialized_var(ret);

	idx = fq_codel_classify(skb, sch, &ret);
	if (idx == 0) {
		if (ret & __NET_XMIT_BYPASS)
			qdisc_qstats_drop(sch);
		kfree_skb(skb);
		return ret;
	}
	idx--;

	codel_set_enqueue_time(skb);
	flow = &q->flows[idx];
	flow_queue_add(flow, skb);
	q->backlogs[idx] += qdisc_pkt_len(skb);
	qdisc_qstats_backlog_inc(sch, skb);

	if (list_empty(&flow->flowchain)) {
		list_add_tail(&flow->flowchain, &q->new_flows);
		q->new_flow_count++;
		flow->deficit = q->quantum;
		flow->dropped = 0;
	}
	if (++sch->q.qlen <= sch->limit)
		return NET_XMIT_SUCCESS;

	prev_backlog = sch->qstats.backlog;
	q->drop_overlimit++;
	/* Return Congestion Notification only if we dropped a packet
	 * from this flow.
	 */
	if (fq_codel_drop(sch) == idx)
		return NET_XMIT_CN;

	/* As we dropped a packet, better let upper stack know this */
	qdisc_tree_reduce_backlog(sch, 1, prev_backlog - sch->qstats.backlog);
	return NET_XMIT_SUCCESS;
}

/* This is the specific function called from codel_dequeue()
 * to dequeue a packet from queue. Note: backlog is handled in
 * codel, we dont need to reduce it here.
 */
static struct sk_buff *dequeue(struct codel_vars *vars, struct Qdisc *sch)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct fq_codel_flow *flow;
	struct sk_buff *skb = NULL;

	flow = container_of(vars, struct fq_codel_flow, cvars);
	if (flow->head) {
		skb = dequeue_head(flow);
		q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb);
		sch->q.qlen--;
	}
	return skb;
}

static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct sk_buff *skb;
	struct fq_codel_flow *flow;
	struct list_head *head;
	u32 prev_drop_count, prev_ecn_mark;
	unsigned int prev_backlog;

begin:
	head = &q->new_flows;
	if (list_empty(head)) {
		head = &q->old_flows;
		if (list_empty(head))
			return NULL;
	}
	flow = list_first_entry(head, struct fq_codel_flow, flowchain);

	if (flow->deficit <= 0) {
		flow->deficit += q->quantum;
		list_move_tail(&flow->flowchain, &q->old_flows);
		goto begin;
	}

	prev_drop_count = q->cstats.drop_count;
	prev_ecn_mark = q->cstats.ecn_mark;
	prev_backlog = sch->qstats.backlog;

	skb = codel_dequeue(sch, &q->cparams, &flow->cvars, &q->cstats,
			    dequeue);

	flow->dropped += q->cstats.drop_count - prev_drop_count;
	flow->dropped += q->cstats.ecn_mark - prev_ecn_mark;

	if (!skb) {
		/* force a pass through old_flows to prevent starvation */
		if ((head == &q->new_flows) && !list_empty(&q->old_flows))
			list_move_tail(&flow->flowchain, &q->old_flows);
		else
			list_del_init(&flow->flowchain);
		goto begin;
	}
	qdisc_bstats_update(sch, skb);
	flow->deficit -= qdisc_pkt_len(skb);
	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
	 * or HTB crashes. Defer it for next round.
	 */
	if (q->cstats.drop_count && sch->q.qlen) {
		qdisc_tree_reduce_backlog(sch, q->cstats.drop_count,
					  q->cstats.drop_len);
		q->cstats.drop_count = 0;
		q->cstats.drop_len = 0;
	}
	return skb;
}

static void fq_codel_reset(struct Qdisc *sch)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	int i;

	INIT_LIST_HEAD(&q->new_flows);
	INIT_LIST_HEAD(&q->old_flows);
	for (i = 0; i < q->flows_cnt; i++) {
		struct fq_codel_flow *flow = q->flows + i;

		while (flow->head) {
			struct sk_buff *skb = dequeue_head(flow);

			qdisc_qstats_backlog_dec(sch, skb);
			kfree_skb(skb);
		}

		INIT_LIST_HEAD(&flow->flowchain);
		codel_vars_init(&flow->cvars);
	}
	memset(q->backlogs, 0, q->flows_cnt * sizeof(u32));
	sch->q.qlen = 0;
}

static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = {
	[TCA_FQ_CODEL_TARGET]	= { .type = NLA_U32 },
	[TCA_FQ_CODEL_LIMIT]	= { .type = NLA_U32 },
	[TCA_FQ_CODEL_INTERVAL]	= { .type = NLA_U32 },
	[TCA_FQ_CODEL_ECN]	= { .type = NLA_U32 },
	[TCA_FQ_CODEL_FLOWS]	= { .type = NLA_U32 },
	[TCA_FQ_CODEL_QUANTUM]	= { .type = NLA_U32 },
	[TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 },
};

static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct nlattr *tb[TCA_FQ_CODEL_MAX + 1];
	int err;

	if (!opt)
		return -EINVAL;

	err = nla_parse_nested(tb, TCA_FQ_CODEL_MAX, opt, fq_codel_policy);
	if (err < 0)
		return err;
	if (tb[TCA_FQ_CODEL_FLOWS]) {
		if (q->flows)
			return -EINVAL;
		q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]);
		if (!q->flows_cnt ||
		    q->flows_cnt > 65536)
			return -EINVAL;
	}
	sch_tree_lock(sch);

	if (tb[TCA_FQ_CODEL_TARGET]) {
		u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]);

		q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT;
	}

	if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) {
		u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]);

		q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
	}

	if (tb[TCA_FQ_CODEL_INTERVAL]) {
		u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]);

		q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT;
	}

	if (tb[TCA_FQ_CODEL_LIMIT])
		sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]);

	if (tb[TCA_FQ_CODEL_ECN])
		q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]);

	if (tb[TCA_FQ_CODEL_QUANTUM])
		q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM]));

	while (sch->q.qlen > sch->limit) {
		struct sk_buff *skb = fq_codel_dequeue(sch);

		q->cstats.drop_len += qdisc_pkt_len(skb);
		kfree_skb(skb);
		q->cstats.drop_count++;
	}
	qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len);
	q->cstats.drop_count = 0;
	q->cstats.drop_len = 0;

	sch_tree_unlock(sch);
	return 0;
}

static void *fq_codel_zalloc(size_t sz)
{
	void *ptr = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN);

	if (!ptr)
		ptr = vzalloc(sz);
	return ptr;
}

static void fq_codel_free(void *addr)
{
	kvfree(addr);
}

static void fq_codel_destroy(struct Qdisc *sch)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);

	tcf_destroy_chain(&q->filter_list);
	fq_codel_free(q->backlogs);
	fq_codel_free(q->flows);
}

static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	int i;

	sch->limit = 10*1024;
	q->flows_cnt = 1024;
	q->quantum = psched_mtu(qdisc_dev(sch));
	q->perturbation = prandom_u32();
	INIT_LIST_HEAD(&q->new_flows);
	INIT_LIST_HEAD(&q->old_flows);
	codel_params_init(&q->cparams, sch);
	codel_stats_init(&q->cstats);
	q->cparams.ecn = true;

	if (opt) {
		int err = fq_codel_change(sch, opt);
		if (err)
			return err;
	}

	if (!q->flows) {
		q->flows = fq_codel_zalloc(q->flows_cnt *
					   sizeof(struct fq_codel_flow));
		if (!q->flows)
			return -ENOMEM;
		q->backlogs = fq_codel_zalloc(q->flows_cnt * sizeof(u32));
		if (!q->backlogs) {
			fq_codel_free(q->flows);
			return -ENOMEM;
		}
		for (i = 0; i < q->flows_cnt; i++) {
			struct fq_codel_flow *flow = q->flows + i;

			INIT_LIST_HEAD(&flow->flowchain);
			codel_vars_init(&flow->cvars);
		}
	}
	if (sch->limit >= 1)
		sch->flags |= TCQ_F_CAN_BYPASS;
	else
		sch->flags &= ~TCQ_F_CAN_BYPASS;
	return 0;
}

static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct nlattr *opts;

	opts = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;

	if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET,
			codel_time_to_us(q->cparams.target)) ||
	    nla_put_u32(skb, TCA_FQ_CODEL_LIMIT,
			sch->limit) ||
	    nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL,
			codel_time_to_us(q->cparams.interval)) ||
	    nla_put_u32(skb, TCA_FQ_CODEL_ECN,
			q->cparams.ecn) ||
	    nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM,
			q->quantum) ||
	    nla_put_u32(skb, TCA_FQ_CODEL_FLOWS,
			q->flows_cnt))
		goto nla_put_failure;

	if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD &&
	    nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
			codel_time_to_us(q->cparams.ce_threshold)))
		goto nla_put_failure;

	return nla_nest_end(skb, opts);

nla_put_failure:
	return -1;
}

static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	struct tc_fq_codel_xstats st = {
		.type				= TCA_FQ_CODEL_XSTATS_QDISC,
	};
	struct list_head *pos;

	st.qdisc_stats.maxpacket = q->cstats.maxpacket;
	st.qdisc_stats.drop_overlimit = q->drop_overlimit;
	st.qdisc_stats.ecn_mark = q->cstats.ecn_mark;
	st.qdisc_stats.new_flow_count = q->new_flow_count;
	st.qdisc_stats.ce_mark = q->cstats.ce_mark;

	list_for_each(pos, &q->new_flows)
		st.qdisc_stats.new_flows_len++;

	list_for_each(pos, &q->old_flows)
		st.qdisc_stats.old_flows_len++;

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

static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg)
{
	return NULL;
}

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

static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent,
			      u32 classid)
{
	/* we cannot bypass queue discipline anymore */
	sch->flags &= ~TCQ_F_CAN_BYPASS;
	return 0;
}

static void fq_codel_put(struct Qdisc *q, unsigned long cl)
{
}

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

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

static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl,
			  struct sk_buff *skb, struct tcmsg *tcm)
{
	tcm->tcm_handle |= TC_H_MIN(cl);
	return 0;
}

static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl,
				     struct gnet_dump *d)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	u32 idx = cl - 1;
	struct gnet_stats_queue qs = { 0 };
	struct tc_fq_codel_xstats xstats;

	if (idx < q->flows_cnt) {
		const struct fq_codel_flow *flow = &q->flows[idx];
		const struct sk_buff *skb = flow->head;

		memset(&xstats, 0, sizeof(xstats));
		xstats.type = TCA_FQ_CODEL_XSTATS_CLASS;
		xstats.class_stats.deficit = flow->deficit;
		xstats.class_stats.ldelay =
			codel_time_to_us(flow->cvars.ldelay);
		xstats.class_stats.count = flow->cvars.count;
		xstats.class_stats.lastcount = flow->cvars.lastcount;
		xstats.class_stats.dropping = flow->cvars.dropping;
		if (flow->cvars.dropping) {
			codel_tdiff_t delta = flow->cvars.drop_next -
					      codel_get_time();

			xstats.class_stats.drop_next = (delta >= 0) ?
				codel_time_to_us(delta) :
				-codel_time_to_us(-delta);
		}
		while (skb) {
			qs.qlen++;
			skb = skb->next;
		}
		qs.backlog = q->backlogs[idx];
		qs.drops = flow->dropped;
	}
	if (gnet_stats_copy_queue(d, NULL, &qs, 0) < 0)
		return -1;
	if (idx < q->flows_cnt)
		return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
	return 0;
}

static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg)
{
	struct fq_codel_sched_data *q = qdisc_priv(sch);
	unsigned int i;

	if (arg->stop)
		return;

	for (i = 0; i < q->flows_cnt; i++) {
		if (list_empty(&q->flows[i].flowchain) ||
		    arg->count < arg->skip) {
			arg->count++;
			continue;
		}
		if (arg->fn(sch, i + 1, arg) < 0) {
			arg->stop = 1;
			break;
		}
		arg->count++;
	}
}

static const struct Qdisc_class_ops fq_codel_class_ops = {
	.leaf		=	fq_codel_leaf,
	.get		=	fq_codel_get,
	.put		=	fq_codel_put,
	.tcf_chain	=	fq_codel_find_tcf,
	.bind_tcf	=	fq_codel_bind,
	.unbind_tcf	=	fq_codel_put,
	.dump		=	fq_codel_dump_class,
	.dump_stats	=	fq_codel_dump_class_stats,
	.walk		=	fq_codel_walk,
};

static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = {
	.cl_ops		=	&fq_codel_class_ops,
	.id		=	"fq_codel",
	.priv_size	=	sizeof(struct fq_codel_sched_data),
	.enqueue	=	fq_codel_enqueue,
	.dequeue	=	fq_codel_dequeue,
	.peek		=	qdisc_peek_dequeued,
	.drop		=	fq_codel_qdisc_drop,
	.init		=	fq_codel_init,
	.reset		=	fq_codel_reset,
	.destroy	=	fq_codel_destroy,
	.change		=	fq_codel_change,
	.dump		=	fq_codel_dump,
	.dump_stats =	fq_codel_dump_stats,
	.owner		=	THIS_MODULE,
};

static int __init fq_codel_module_init(void)
{
	return register_qdisc(&fq_codel_qdisc_ops);
}

static void __exit fq_codel_module_exit(void)
{
	unregister_qdisc(&fq_codel_qdisc_ops);
}

module_init(fq_codel_module_init)
module_exit(fq_codel_module_exit)
MODULE_AUTHOR("Eric Dumazet");
MODULE_LICENSE("GPL");
Commit	Line	Data
4b549a2e ED	1	/*
	2	* Fair Queue CoDel discipline
	3	*
	4	* This program is free software; you can redistribute it and/or
	5	* modify it under the terms of the GNU General Public License
	6	* as published by the Free Software Foundation; either version
	7	* 2 of the License, or (at your option) any later version.
	8	*
80ba92fa	9	* Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
4b549a2e ED	10	*/
	11
	12	#include <linux/module.h>
	13	#include <linux/types.h>
	14	#include <linux/kernel.h>
	15	#include <linux/jiffies.h>
	16	#include <linux/string.h>
	17	#include <linux/in.h>
	18	#include <linux/errno.h>
	19	#include <linux/init.h>
	20	#include <linux/skbuff.h>
	21	#include <linux/jhash.h>
	22	#include <linux/slab.h>
	23	#include <linux/vmalloc.h>
	24	#include <net/netlink.h>
	25	#include <net/pkt_sched.h>
4b549a2e ED	26	#include <net/codel.h>
	27
	28	/* Fair Queue CoDel.
	29	*
	30	* Principles :
	31	* Packets are classified (internal classifier or external) on flows.
	32	* This is a Stochastic model (as we use a hash, several flows
	33	* might be hashed on same slot)
	34	* Each flow has a CoDel managed queue.
	35	* Flows are linked onto two (Round Robin) lists,
	36	* so that new flows have priority on old ones.
	37	*
	38	* For a given flow, packets are not reordered (CoDel uses a FIFO)
	39	* head drops only.
	40	* ECN capability is on by default.
	41	* Low memory footprint (64 bytes per flow)
	42	*/
	43
	44	struct fq_codel_flow {
	45	struct sk_buff *head;
	46	struct sk_buff *tail;
	47	struct list_head flowchain;
	48	int deficit;
	49	u32 dropped; /* number of drops (or ECN marks) on this flow */
	50	struct codel_vars cvars;
	51	}; /* please try to keep this structure <= 64 bytes */
	52
	53	struct fq_codel_sched_data {
25d8c0d5	54	struct tcf_proto __rcu filter_list; / optional external classifier */
4b549a2e ED	55	struct fq_codel_flow flows; / Flows table [flows_cnt] */
	56	u32 backlogs; / backlog table [flows_cnt] */
	57	u32 flows_cnt; /* number of flows */
	58	u32 perturbation; /* hash perturbation */
	59	u32 quantum; /* psched_mtu(qdisc_dev(sch)); */
	60	struct codel_params cparams;
	61	struct codel_stats cstats;
	62	u32 drop_overlimit;
	63	u32 new_flow_count;
	64
	65	struct list_head new_flows; /* list of new flows */
	66	struct list_head old_flows; /* list of old flows */
	67	};
	68
	69	static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q,
342db221	70	struct sk_buff *skb)
4b549a2e	71	{
342db221	72	u32 hash = skb_get_hash_perturb(skb, q->perturbation);
8fc54f68 DB	73
8fc54f68 DB	74	return reciprocal_scale(hash, q->flows_cnt);
4b549a2e ED	75	}
	76
	77	static unsigned int fq_codel_classify(struct sk_buff skb, struct Qdisc sch,
	78	int *qerr)
	79	{
	80	struct fq_codel_sched_data *q = qdisc_priv(sch);
25d8c0d5	81	struct tcf_proto *filter;
4b549a2e ED	82	struct tcf_result res;
	83	int result;
	84
	85	if (TC_H_MAJ(skb->priority) == sch->handle &&
	86	TC_H_MIN(skb->priority) > 0 &&
	87	TC_H_MIN(skb->priority) <= q->flows_cnt)
	88	return TC_H_MIN(skb->priority);
	89
69204cf7	90	filter = rcu_dereference_bh(q->filter_list);
25d8c0d5	91	if (!filter)
4b549a2e ED	92	return fq_codel_hash(q, skb) + 1;
	93
	94	*qerr = NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
3b3ae880	95	result = tc_classify(skb, filter, &res, false);
4b549a2e ED	96	if (result >= 0) {
	97	#ifdef CONFIG_NET_CLS_ACT
	98	switch (result) {
	99	case TC_ACT_STOLEN:
	100	case TC_ACT_QUEUED:
	101	*qerr = NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
	102	case TC_ACT_SHOT:
	103	return 0;
	104	}
	105	#endif
	106	if (TC_H_MIN(res.classid) <= q->flows_cnt)
	107	return TC_H_MIN(res.classid);
	108	}
	109	return 0;
	110	}
	111
	112	/* helper functions : might be changed when/if skb use a standard list_head */
	113
	114	/* remove one skb from head of slot queue */
	115	static inline struct sk_buff dequeue_head(struct fq_codel_flow flow)
	116	{
	117	struct sk_buff *skb = flow->head;
	118
	119	flow->head = skb->next;
	120	skb->next = NULL;
	121	return skb;
	122	}
	123
	124	/* add skb to flow queue (tail add) */
	125	static inline void flow_queue_add(struct fq_codel_flow *flow,
	126	struct sk_buff *skb)
	127	{
	128	if (flow->head == NULL)
	129	flow->head = skb;
	130	else
	131	flow->tail->next = skb;
	132	flow->tail = skb;
	133	skb->next = NULL;
	134	}
	135
	136	static unsigned int fq_codel_drop(struct Qdisc *sch)
	137	{
	138	struct fq_codel_sched_data *q = qdisc_priv(sch);
	139	struct sk_buff *skb;
	140	unsigned int maxbacklog = 0, idx = 0, i, len;
	141	struct fq_codel_flow *flow;
	142
	143	/* Queue is full! Find the fat flow and drop packet from it.
	144	* This might sound expensive, but with 1024 flows, we scan
	145	* 4KB of memory, and we dont need to handle a complex tree
	146	* in fast path (packet queue/enqueue) with many cache misses.
	147	*/
	148	for (i = 0; i < q->flows_cnt; i++) {
	149	if (q->backlogs[i] > maxbacklog) {
	150	maxbacklog = q->backlogs[i];
	151	idx = i;
	152	}
	153	}
	154	flow = &q->flows[idx];
	155	skb = dequeue_head(flow);
	156	len = qdisc_pkt_len(skb);
	157	q->backlogs[idx] -= len;
4b549a2e	158	sch->q.qlen--;
25331d6c JF	159	qdisc_qstats_drop(sch);
25331d6c JF	160	qdisc_qstats_backlog_dec(sch, skb);
052cbda4	161	kfree_skb(skb);
4b549a2e ED	162	flow->dropped++;
	163	return idx;
	164	}
	165
c0afd9ce WC	166	static unsigned int fq_codel_qdisc_drop(struct Qdisc *sch)
	167	{
	168	unsigned int prev_backlog;
	169
	170	prev_backlog = sch->qstats.backlog;
	171	fq_codel_drop(sch);
	172	return prev_backlog - sch->qstats.backlog;
	173	}
	174
4b549a2e ED	175	static int fq_codel_enqueue(struct sk_buff skb, struct Qdisc sch)
	176	{
	177	struct fq_codel_sched_data *q = qdisc_priv(sch);
2ccccf5f	178	unsigned int idx, prev_backlog;
4b549a2e ED	179	struct fq_codel_flow *flow;
	180	int uninitialized_var(ret);
	181
	182	idx = fq_codel_classify(skb, sch, &ret);
	183	if (idx == 0) {
	184	if (ret & __NET_XMIT_BYPASS)
25331d6c	185	qdisc_qstats_drop(sch);
4b549a2e ED	186	kfree_skb(skb);
	187	return ret;
	188	}
	189	idx--;
	190
	191	codel_set_enqueue_time(skb);
	192	flow = &q->flows[idx];
	193	flow_queue_add(flow, skb);
	194	q->backlogs[idx] += qdisc_pkt_len(skb);
25331d6c	195	qdisc_qstats_backlog_inc(sch, skb);
4b549a2e ED	196
	197	if (list_empty(&flow->flowchain)) {
	198	list_add_tail(&flow->flowchain, &q->new_flows);
4b549a2e ED	199	q->new_flow_count++;
	200	flow->deficit = q->quantum;
	201	flow->dropped = 0;
	202	}
cd68ddd4	203	if (++sch->q.qlen <= sch->limit)
4b549a2e ED	204	return NET_XMIT_SUCCESS;
4b549a2e ED	205
2ccccf5f	206	prev_backlog = sch->qstats.backlog;
4b549a2e ED	207	q->drop_overlimit++;
	208	/* Return Congestion Notification only if we dropped a packet
	209	* from this flow.
	210	*/
	211	if (fq_codel_drop(sch) == idx)
	212	return NET_XMIT_CN;
	213
	214	/* As we dropped a packet, better let upper stack know this */
2ccccf5f	215	qdisc_tree_reduce_backlog(sch, 1, prev_backlog - sch->qstats.backlog);
4b549a2e ED	216	return NET_XMIT_SUCCESS;
	217	}
	218
	219	/* This is the specific function called from codel_dequeue()
	220	* to dequeue a packet from queue. Note: backlog is handled in
	221	* codel, we dont need to reduce it here.
	222	*/
	223	static struct sk_buff dequeue(struct codel_vars vars, struct Qdisc *sch)
	224	{
865ec552	225	struct fq_codel_sched_data *q = qdisc_priv(sch);
4b549a2e ED	226	struct fq_codel_flow *flow;
	227	struct sk_buff *skb = NULL;
	228
	229	flow = container_of(vars, struct fq_codel_flow, cvars);
	230	if (flow->head) {
	231	skb = dequeue_head(flow);
865ec552	232	q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb);
4b549a2e ED	233	sch->q.qlen--;
	234	}
	235	return skb;
	236	}
	237
	238	static struct sk_buff fq_codel_dequeue(struct Qdisc sch)
	239	{
	240	struct fq_codel_sched_data *q = qdisc_priv(sch);
	241	struct sk_buff *skb;
	242	struct fq_codel_flow *flow;
	243	struct list_head *head;
	244	u32 prev_drop_count, prev_ecn_mark;
2ccccf5f	245	unsigned int prev_backlog;
4b549a2e ED	246
	247	begin:
	248	head = &q->new_flows;
	249	if (list_empty(head)) {
	250	head = &q->old_flows;
	251	if (list_empty(head))
	252	return NULL;
	253	}
	254	flow = list_first_entry(head, struct fq_codel_flow, flowchain);
	255
	256	if (flow->deficit <= 0) {
	257	flow->deficit += q->quantum;
	258	list_move_tail(&flow->flowchain, &q->old_flows);
	259	goto begin;
	260	}
	261
	262	prev_drop_count = q->cstats.drop_count;
	263	prev_ecn_mark = q->cstats.ecn_mark;
2ccccf5f	264	prev_backlog = sch->qstats.backlog;
4b549a2e ED	265
4b549a2e ED	266	skb = codel_dequeue(sch, &q->cparams, &flow->cvars, &q->cstats,
865ec552	267	dequeue);
4b549a2e ED	268
	269	flow->dropped += q->cstats.drop_count - prev_drop_count;
	270	flow->dropped += q->cstats.ecn_mark - prev_ecn_mark;
	271
	272	if (!skb) {
	273	/* force a pass through old_flows to prevent starvation */
	274	if ((head == &q->new_flows) && !list_empty(&q->old_flows))
	275	list_move_tail(&flow->flowchain, &q->old_flows);
	276	else
	277	list_del_init(&flow->flowchain);
	278	goto begin;
	279	}
	280	qdisc_bstats_update(sch, skb);
	281	flow->deficit -= qdisc_pkt_len(skb);
2ccccf5f	282	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
4b549a2e ED	283	* or HTB crashes. Defer it for next round.
	284	*/
	285	if (q->cstats.drop_count && sch->q.qlen) {
2ccccf5f WC	286	qdisc_tree_reduce_backlog(sch, q->cstats.drop_count,
2ccccf5f WC	287	q->cstats.drop_len);
4b549a2e	288	q->cstats.drop_count = 0;
2ccccf5f	289	q->cstats.drop_len = 0;
4b549a2e ED	290	}
	291	return skb;
	292	}
	293
	294	static void fq_codel_reset(struct Qdisc *sch)
	295	{
3d0e0af4 ED	296	struct fq_codel_sched_data *q = qdisc_priv(sch);
3d0e0af4 ED	297	int i;
4b549a2e	298
3d0e0af4 ED	299	INIT_LIST_HEAD(&q->new_flows);
	300	INIT_LIST_HEAD(&q->old_flows);
	301	for (i = 0; i < q->flows_cnt; i++) {
	302	struct fq_codel_flow *flow = q->flows + i;
	303
	304	while (flow->head) {
	305	struct sk_buff *skb = dequeue_head(flow);
	306
	307	qdisc_qstats_backlog_dec(sch, skb);
	308	kfree_skb(skb);
	309	}
	310
	311	INIT_LIST_HEAD(&flow->flowchain);
	312	codel_vars_init(&flow->cvars);
	313	}
	314	memset(q->backlogs, 0, q->flows_cnt * sizeof(u32));
	315	sch->q.qlen = 0;
4b549a2e ED	316	}
	317
	318	static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = {
	319	[TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 },
	320	[TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 },
	321	[TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 },
	322	[TCA_FQ_CODEL_ECN] = { .type = NLA_U32 },
	323	[TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 },
	324	[TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 },
80ba92fa	325	[TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 },
4b549a2e ED	326	};
	327
	328	static int fq_codel_change(struct Qdisc sch, struct nlattr opt)
	329	{
	330	struct fq_codel_sched_data *q = qdisc_priv(sch);
	331	struct nlattr *tb[TCA_FQ_CODEL_MAX + 1];
	332	int err;
	333
	334	if (!opt)
	335	return -EINVAL;
	336
	337	err = nla_parse_nested(tb, TCA_FQ_CODEL_MAX, opt, fq_codel_policy);
	338	if (err < 0)
	339	return err;
	340	if (tb[TCA_FQ_CODEL_FLOWS]) {
	341	if (q->flows)
	342	return -EINVAL;
	343	q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]);
	344	if (!q->flows_cnt \|\|
	345	q->flows_cnt > 65536)
	346	return -EINVAL;
	347	}
	348	sch_tree_lock(sch);
	349
	350	if (tb[TCA_FQ_CODEL_TARGET]) {
	351	u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]);
	352
	353	q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT;
	354	}
	355
80ba92fa ED	356	if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) {
	357	u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]);
	358
	359	q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
	360	}
	361
4b549a2e ED	362	if (tb[TCA_FQ_CODEL_INTERVAL]) {
	363	u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]);
	364
	365	q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT;
	366	}
	367
	368	if (tb[TCA_FQ_CODEL_LIMIT])
	369	sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]);
	370
	371	if (tb[TCA_FQ_CODEL_ECN])
	372	q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]);
	373
	374	if (tb[TCA_FQ_CODEL_QUANTUM])
	375	q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM]));
	376
	377	while (sch->q.qlen > sch->limit) {
	378	struct sk_buff *skb = fq_codel_dequeue(sch);
	379
2ccccf5f	380	q->cstats.drop_len += qdisc_pkt_len(skb);
4b549a2e ED	381	kfree_skb(skb);
	382	q->cstats.drop_count++;
	383	}
2ccccf5f	384	qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len);
4b549a2e	385	q->cstats.drop_count = 0;
2ccccf5f	386	q->cstats.drop_len = 0;
4b549a2e ED	387
	388	sch_tree_unlock(sch);
	389	return 0;
	390	}
	391
	392	static void *fq_codel_zalloc(size_t sz)
	393	{
	394	void *ptr = kzalloc(sz, GFP_KERNEL \| __GFP_NOWARN);
	395
	396	if (!ptr)
	397	ptr = vzalloc(sz);
	398	return ptr;
	399	}
	400
	401	static void fq_codel_free(void *addr)
	402	{
4cb28970	403	kvfree(addr);
4b549a2e ED	404	}
	405
	406	static void fq_codel_destroy(struct Qdisc *sch)
	407	{
	408	struct fq_codel_sched_data *q = qdisc_priv(sch);
	409
	410	tcf_destroy_chain(&q->filter_list);
	411	fq_codel_free(q->backlogs);
	412	fq_codel_free(q->flows);
	413	}
	414
	415	static int fq_codel_init(struct Qdisc sch, struct nlattr opt)
	416	{
	417	struct fq_codel_sched_data *q = qdisc_priv(sch);
	418	int i;
	419
	420	sch->limit = 10*1024;
	421	q->flows_cnt = 1024;
	422	q->quantum = psched_mtu(qdisc_dev(sch));
63862b5b	423	q->perturbation = prandom_u32();
4b549a2e ED	424	INIT_LIST_HEAD(&q->new_flows);
4b549a2e ED	425	INIT_LIST_HEAD(&q->old_flows);
a5d28090	426	codel_params_init(&q->cparams, sch);
4b549a2e ED	427	codel_stats_init(&q->cstats);
	428	q->cparams.ecn = true;
	429
	430	if (opt) {
	431	int err = fq_codel_change(sch, opt);
	432	if (err)
	433	return err;
	434	}
	435
	436	if (!q->flows) {
	437	q->flows = fq_codel_zalloc(q->flows_cnt *
	438	sizeof(struct fq_codel_flow));
	439	if (!q->flows)
	440	return -ENOMEM;
	441	q->backlogs = fq_codel_zalloc(q->flows_cnt * sizeof(u32));
	442	if (!q->backlogs) {
	443	fq_codel_free(q->flows);
	444	return -ENOMEM;
	445	}
	446	for (i = 0; i < q->flows_cnt; i++) {
	447	struct fq_codel_flow *flow = q->flows + i;
	448
	449	INIT_LIST_HEAD(&flow->flowchain);
b379135c	450	codel_vars_init(&flow->cvars);
4b549a2e ED	451	}
	452	}
	453	if (sch->limit >= 1)
	454	sch->flags \|= TCQ_F_CAN_BYPASS;
	455	else
	456	sch->flags &= ~TCQ_F_CAN_BYPASS;
	457	return 0;
	458	}
	459
	460	static int fq_codel_dump(struct Qdisc sch, struct sk_buff skb)
	461	{
	462	struct fq_codel_sched_data *q = qdisc_priv(sch);
	463	struct nlattr *opts;
	464
	465	opts = nla_nest_start(skb, TCA_OPTIONS);
	466	if (opts == NULL)
	467	goto nla_put_failure;
	468
	469	if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET,
	470	codel_time_to_us(q->cparams.target)) \|\|
	471	nla_put_u32(skb, TCA_FQ_CODEL_LIMIT,
	472	sch->limit) \|\|
	473	nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL,
	474	codel_time_to_us(q->cparams.interval)) \|\|
	475	nla_put_u32(skb, TCA_FQ_CODEL_ECN,
	476	q->cparams.ecn) \|\|
	477	nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM,
	478	q->quantum) \|\|
	479	nla_put_u32(skb, TCA_FQ_CODEL_FLOWS,
	480	q->flows_cnt))
	481	goto nla_put_failure;
	482
80ba92fa ED	483	if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD &&
	484	nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
	485	codel_time_to_us(q->cparams.ce_threshold)))
	486	goto nla_put_failure;
	487
d59b7d80	488	return nla_nest_end(skb, opts);
4b549a2e ED	489
	490	nla_put_failure:
	491	return -1;
	492	}
	493
	494	static int fq_codel_dump_stats(struct Qdisc sch, struct gnet_dump d)
	495	{
	496	struct fq_codel_sched_data *q = qdisc_priv(sch);
	497	struct tc_fq_codel_xstats st = {
	498	.type = TCA_FQ_CODEL_XSTATS_QDISC,
4b549a2e ED	499	};
	500	struct list_head *pos;
	501
669d67bf SL	502	st.qdisc_stats.maxpacket = q->cstats.maxpacket;
	503	st.qdisc_stats.drop_overlimit = q->drop_overlimit;
	504	st.qdisc_stats.ecn_mark = q->cstats.ecn_mark;
	505	st.qdisc_stats.new_flow_count = q->new_flow_count;
80ba92fa	506	st.qdisc_stats.ce_mark = q->cstats.ce_mark;
669d67bf	507
4b549a2e ED	508	list_for_each(pos, &q->new_flows)
	509	st.qdisc_stats.new_flows_len++;
	510
	511	list_for_each(pos, &q->old_flows)
	512	st.qdisc_stats.old_flows_len++;
	513
	514	return gnet_stats_copy_app(d, &st, sizeof(st));
	515	}
	516
	517	static struct Qdisc fq_codel_leaf(struct Qdisc sch, unsigned long arg)
	518	{
	519	return NULL;
	520	}
	521
	522	static unsigned long fq_codel_get(struct Qdisc *sch, u32 classid)
	523	{
	524	return 0;
	525	}
	526
	527	static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent,
	528	u32 classid)
	529	{
	530	/* we cannot bypass queue discipline anymore */
	531	sch->flags &= ~TCQ_F_CAN_BYPASS;
	532	return 0;
	533	}
	534
	535	static void fq_codel_put(struct Qdisc *q, unsigned long cl)
	536	{
	537	}
	538
25d8c0d5 JF	539	static struct tcf_proto __rcu *fq_codel_find_tcf(struct Qdisc sch,
25d8c0d5 JF	540	unsigned long cl)
4b549a2e ED	541	{
	542	struct fq_codel_sched_data *q = qdisc_priv(sch);
	543
	544	if (cl)
	545	return NULL;
	546	return &q->filter_list;
	547	}
	548
	549	static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl,
	550	struct sk_buff skb, struct tcmsg tcm)
	551	{
	552	tcm->tcm_handle \|= TC_H_MIN(cl);
	553	return 0;
	554	}
	555
	556	static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl,
	557	struct gnet_dump *d)
	558	{
	559	struct fq_codel_sched_data *q = qdisc_priv(sch);
	560	u32 idx = cl - 1;
	561	struct gnet_stats_queue qs = { 0 };
	562	struct tc_fq_codel_xstats xstats;
	563
	564	if (idx < q->flows_cnt) {
	565	const struct fq_codel_flow *flow = &q->flows[idx];
	566	const struct sk_buff *skb = flow->head;
	567
	568	memset(&xstats, 0, sizeof(xstats));
	569	xstats.type = TCA_FQ_CODEL_XSTATS_CLASS;
	570	xstats.class_stats.deficit = flow->deficit;
	571	xstats.class_stats.ldelay =
	572	codel_time_to_us(flow->cvars.ldelay);
	573	xstats.class_stats.count = flow->cvars.count;
	574	xstats.class_stats.lastcount = flow->cvars.lastcount;
	575	xstats.class_stats.dropping = flow->cvars.dropping;
	576	if (flow->cvars.dropping) {
	577	codel_tdiff_t delta = flow->cvars.drop_next -
	578	codel_get_time();
	579
	580	xstats.class_stats.drop_next = (delta >= 0) ?
	581	codel_time_to_us(delta) :
	582	-codel_time_to_us(-delta);
	583	}
	584	while (skb) {
	585	qs.qlen++;
	586	skb = skb->next;
	587	}
	588	qs.backlog = q->backlogs[idx];
	589	qs.drops = flow->dropped;
	590	}
b0ab6f92	591	if (gnet_stats_copy_queue(d, NULL, &qs, 0) < 0)
4b549a2e ED	592	return -1;
	593	if (idx < q->flows_cnt)
	594	return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
	595	return 0;
	596	}
	597
	598	static void fq_codel_walk(struct Qdisc sch, struct qdisc_walker arg)
	599	{
	600	struct fq_codel_sched_data *q = qdisc_priv(sch);
	601	unsigned int i;
	602
	603	if (arg->stop)
	604	return;
	605
	606	for (i = 0; i < q->flows_cnt; i++) {
	607	if (list_empty(&q->flows[i].flowchain) \|\|
	608	arg->count < arg->skip) {
	609	arg->count++;
	610	continue;
	611	}
	612	if (arg->fn(sch, i + 1, arg) < 0) {
	613	arg->stop = 1;
	614	break;
	615	}
	616	arg->count++;
	617	}
	618	}
	619
	620	static const struct Qdisc_class_ops fq_codel_class_ops = {
	621	.leaf = fq_codel_leaf,
	622	.get = fq_codel_get,
	623	.put = fq_codel_put,
	624	.tcf_chain = fq_codel_find_tcf,
	625	.bind_tcf = fq_codel_bind,
	626	.unbind_tcf = fq_codel_put,
	627	.dump = fq_codel_dump_class,
	628	.dump_stats = fq_codel_dump_class_stats,
	629	.walk = fq_codel_walk,
	630	};
	631
	632	static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = {
	633	.cl_ops = &fq_codel_class_ops,
	634	.id = "fq_codel",
	635	.priv_size = sizeof(struct fq_codel_sched_data),
	636	.enqueue = fq_codel_enqueue,
	637	.dequeue = fq_codel_dequeue,
	638	.peek = qdisc_peek_dequeued,
c0afd9ce	639	.drop = fq_codel_qdisc_drop,
4b549a2e ED	640	.init = fq_codel_init,
	641	.reset = fq_codel_reset,
	642	.destroy = fq_codel_destroy,
	643	.change = fq_codel_change,
	644	.dump = fq_codel_dump,
	645	.dump_stats = fq_codel_dump_stats,
	646	.owner = THIS_MODULE,
	647	};
	648
	649	static int __init fq_codel_module_init(void)
	650	{
	651	return register_qdisc(&fq_codel_qdisc_ops);
	652	}
	653
	654	static void __exit fq_codel_module_exit(void)
	655	{
	656	unregister_qdisc(&fq_codel_qdisc_ops);
	657	}
	658
	659	module_init(fq_codel_module_init)
	660	module_exit(fq_codel_module_exit)
	661	MODULE_AUTHOR("Eric Dumazet");
	662	MODULE_LICENSE("GPL");