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

/*
 * net/sched/sch_gred.c	Generic Random Early Detection queue.
 *
 *
 *              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.
 *
 * Authors:    J Hadi Salim (hadi@cyberus.ca) 1998-2002
 *
 *             991129: -  Bug fix with grio mode
 *		       - a better sing. AvgQ mode with Grio(WRED)
 *		       - A finer grained VQ dequeue based on sugestion
 *		         from Ren Liu
 *		       - More error checks
 *
 *  For all the glorious comments look at include/net/red.h
 */

#include <linux/slab.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/skbuff.h>
#include <net/pkt_sched.h>
#include <net/red.h>

#define GRED_DEF_PRIO (MAX_DPs / 2)
#define GRED_VQ_MASK (MAX_DPs - 1)

struct gred_sched_data;
struct gred_sched;

struct gred_sched_data {
	u32		limit;		/* HARD maximal queue length	*/
	u32		DP;		/* the drop parameters */
	u32		bytesin;	/* bytes seen on virtualQ so far*/
	u32		packetsin;	/* packets seen on virtualQ so far*/
	u32		backlog;	/* bytes on the virtualQ */
	u8		prio;		/* the prio of this vq */

	struct red_parms parms;
	struct red_vars  vars;
	struct red_stats stats;
};

enum {
	GRED_WRED_MODE = 1,
	GRED_RIO_MODE,
};

struct gred_sched {
	struct gred_sched_data *tab[MAX_DPs];
	unsigned long	flags;
	u32		red_flags;
	u32 		DPs;
	u32 		def;
	struct red_vars wred_set;
};

static inline int gred_wred_mode(struct gred_sched *table)
{
	return test_bit(GRED_WRED_MODE, &table->flags);
}

static inline void gred_enable_wred_mode(struct gred_sched *table)
{
	__set_bit(GRED_WRED_MODE, &table->flags);
}

static inline void gred_disable_wred_mode(struct gred_sched *table)
{
	__clear_bit(GRED_WRED_MODE, &table->flags);
}

static inline int gred_rio_mode(struct gred_sched *table)
{
	return test_bit(GRED_RIO_MODE, &table->flags);
}

static inline void gred_enable_rio_mode(struct gred_sched *table)
{
	__set_bit(GRED_RIO_MODE, &table->flags);
}

static inline void gred_disable_rio_mode(struct gred_sched *table)
{
	__clear_bit(GRED_RIO_MODE, &table->flags);
}

static inline int gred_wred_mode_check(struct Qdisc *sch)
{
	struct gred_sched *table = qdisc_priv(sch);
	int i;

	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
	for (i = 0; i < table->DPs; i++) {
		struct gred_sched_data *q = table->tab[i];
		int n;

		if (q == NULL)
			continue;

		for (n = i + 1; n < table->DPs; n++)
			if (table->tab[n] && table->tab[n]->prio == q->prio)
				return 1;
	}

	return 0;
}

static inline unsigned int gred_backlog(struct gred_sched *table,
					struct gred_sched_data *q,
					struct Qdisc *sch)
{
	if (gred_wred_mode(table))
		return sch->qstats.backlog;
	else
		return q->backlog;
}

static inline u16 tc_index_to_dp(struct sk_buff *skb)
{
	return skb->tc_index & GRED_VQ_MASK;
}

static inline void gred_load_wred_set(const struct gred_sched *table,
				      struct gred_sched_data *q)
{
	q->vars.qavg = table->wred_set.qavg;
	q->vars.qidlestart = table->wred_set.qidlestart;
}

static inline void gred_store_wred_set(struct gred_sched *table,
				       struct gred_sched_data *q)
{
	table->wred_set.qavg = q->vars.qavg;
	table->wred_set.qidlestart = q->vars.qidlestart;
}

static inline int gred_use_ecn(struct gred_sched *t)
{
	return t->red_flags & TC_RED_ECN;
}

static inline int gred_use_harddrop(struct gred_sched *t)
{
	return t->red_flags & TC_RED_HARDDROP;
}

static int gred_enqueue(struct sk_buff *skb, struct Qdisc *sch)
{
	struct gred_sched_data *q = NULL;
	struct gred_sched *t = qdisc_priv(sch);
	unsigned long qavg = 0;
	u16 dp = tc_index_to_dp(skb);

	if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
		dp = t->def;

		q = t->tab[dp];
		if (!q) {
			/* Pass through packets not assigned to a DP
			 * if no default DP has been configured. This
			 * allows for DP flows to be left untouched.
			 */
			if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <=
					sch->limit))
				return qdisc_enqueue_tail(skb, sch);
			else
				goto drop;
		}

		/* fix tc_index? --could be controversial but needed for
		   requeueing */
		skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
	}

	/* sum up all the qaves of prios < ours to get the new qave */
	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
		int i;

		for (i = 0; i < t->DPs; i++) {
			if (t->tab[i] && t->tab[i]->prio < q->prio &&
			    !red_is_idling(&t->tab[i]->vars))
				qavg += t->tab[i]->vars.qavg;
		}

	}

	q->packetsin++;
	q->bytesin += qdisc_pkt_len(skb);

	if (gred_wred_mode(t))
		gred_load_wred_set(t, q);

	q->vars.qavg = red_calc_qavg(&q->parms,
				     &q->vars,
				     gred_backlog(t, q, sch));

	if (red_is_idling(&q->vars))
		red_end_of_idle_period(&q->vars);

	if (gred_wred_mode(t))
		gred_store_wred_set(t, q);

	switch (red_action(&q->parms, &q->vars, q->vars.qavg + qavg)) {
	case RED_DONT_MARK:
		break;

	case RED_PROB_MARK:
		qdisc_qstats_overlimit(sch);
		if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
			q->stats.prob_drop++;
			goto congestion_drop;
		}

		q->stats.prob_mark++;
		break;

	case RED_HARD_MARK:
		qdisc_qstats_overlimit(sch);
		if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
		    !INET_ECN_set_ce(skb)) {
			q->stats.forced_drop++;
			goto congestion_drop;
		}
		q->stats.forced_mark++;
		break;
	}

	if (gred_backlog(t, q, sch) + qdisc_pkt_len(skb) <= q->limit) {
		q->backlog += qdisc_pkt_len(skb);
		return qdisc_enqueue_tail(skb, sch);
	}

	q->stats.pdrop++;
drop:
	return qdisc_drop(skb, sch);

congestion_drop:
	qdisc_drop(skb, sch);
	return NET_XMIT_CN;
}

static struct sk_buff *gred_dequeue(struct Qdisc *sch)
{
	struct sk_buff *skb;
	struct gred_sched *t = qdisc_priv(sch);

	skb = qdisc_dequeue_head(sch);

	if (skb) {
		struct gred_sched_data *q;
		u16 dp = tc_index_to_dp(skb);

		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
			net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x after dequeue, screwing up backlog\n",
					     tc_index_to_dp(skb));
		} else {
			q->backlog -= qdisc_pkt_len(skb);

			if (gred_wred_mode(t)) {
				if (!sch->qstats.backlog)
					red_start_of_idle_period(&t->wred_set);
			} else {
				if (!q->backlog)
					red_start_of_idle_period(&q->vars);
			}
		}

		return skb;
	}

	return NULL;
}

static unsigned int gred_drop(struct Qdisc *sch)
{
	struct sk_buff *skb;
	struct gred_sched *t = qdisc_priv(sch);

	skb = qdisc_dequeue_tail(sch);
	if (skb) {
		unsigned int len = qdisc_pkt_len(skb);
		struct gred_sched_data *q;
		u16 dp = tc_index_to_dp(skb);

		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
			net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x while dropping, screwing up backlog\n",
					     tc_index_to_dp(skb));
		} else {
			q->backlog -= len;
			q->stats.other++;

			if (gred_wred_mode(t)) {
				if (!sch->qstats.backlog)
					red_start_of_idle_period(&t->wred_set);
			} else {
				if (!q->backlog)
					red_start_of_idle_period(&q->vars);
			}
		}

		qdisc_drop(skb, sch);
		return len;
	}

	return 0;
}

static void gred_reset(struct Qdisc *sch)
{
	int i;
	struct gred_sched *t = qdisc_priv(sch);

	qdisc_reset_queue(sch);

	for (i = 0; i < t->DPs; i++) {
		struct gred_sched_data *q = t->tab[i];

		if (!q)
			continue;

		red_restart(&q->vars);
		q->backlog = 0;
	}
}

static inline void gred_destroy_vq(struct gred_sched_data *q)
{
	kfree(q);
}

static inline int gred_change_table_def(struct Qdisc *sch, struct nlattr *dps)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct tc_gred_sopt *sopt;
	int i;

	if (dps == NULL)
		return -EINVAL;

	sopt = nla_data(dps);

	if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
		return -EINVAL;

	sch_tree_lock(sch);
	table->DPs = sopt->DPs;
	table->def = sopt->def_DP;
	table->red_flags = sopt->flags;

	/*
	 * Every entry point to GRED is synchronized with the above code
	 * and the DP is checked against DPs, i.e. shadowed VQs can no
	 * longer be found so we can unlock right here.
	 */
	sch_tree_unlock(sch);

	if (sopt->grio) {
		gred_enable_rio_mode(table);
		gred_disable_wred_mode(table);
		if (gred_wred_mode_check(sch))
			gred_enable_wred_mode(table);
	} else {
		gred_disable_rio_mode(table);
		gred_disable_wred_mode(table);
	}

	for (i = table->DPs; i < MAX_DPs; i++) {
		if (table->tab[i]) {
			pr_warn("GRED: Warning: Destroying shadowed VQ 0x%x\n",
				i);
			gred_destroy_vq(table->tab[i]);
			table->tab[i] = NULL;
		}
	}

	return 0;
}

static inline int gred_change_vq(struct Qdisc *sch, int dp,
				 struct tc_gred_qopt *ctl, int prio,
				 u8 *stab, u32 max_P,
				 struct gred_sched_data **prealloc)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct gred_sched_data *q = table->tab[dp];

	if (!q) {
		table->tab[dp] = q = *prealloc;
		*prealloc = NULL;
		if (!q)
			return -ENOMEM;
	}

	q->DP = dp;
	q->prio = prio;
	if (ctl->limit > sch->limit)
		q->limit = sch->limit;
	else
		q->limit = ctl->limit;

	if (q->backlog == 0)
		red_end_of_idle_period(&q->vars);

	red_set_parms(&q->parms,
		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
		      ctl->Scell_log, stab, max_P);
	red_set_vars(&q->vars);
	return 0;
}

static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
	[TCA_GRED_PARMS]	= { .len = sizeof(struct tc_gred_qopt) },
	[TCA_GRED_STAB]		= { .len = 256 },
	[TCA_GRED_DPS]		= { .len = sizeof(struct tc_gred_sopt) },
	[TCA_GRED_MAX_P]	= { .type = NLA_U32 },
	[TCA_GRED_LIMIT]	= { .type = NLA_U32 },
};

static int gred_change(struct Qdisc *sch, struct nlattr *opt)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct tc_gred_qopt *ctl;
	struct nlattr *tb[TCA_GRED_MAX + 1];
	int err, prio = GRED_DEF_PRIO;
	u8 *stab;
	u32 max_P;
	struct gred_sched_data *prealloc;

	if (opt == NULL)
		return -EINVAL;

	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
	if (err < 0)
		return err;

	if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL) {
		if (tb[TCA_GRED_LIMIT] != NULL)
			sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
		return gred_change_table_def(sch, opt);
	}

	if (tb[TCA_GRED_PARMS] == NULL ||
	    tb[TCA_GRED_STAB] == NULL ||
	    tb[TCA_GRED_LIMIT] != NULL)
		return -EINVAL;

	max_P = tb[TCA_GRED_MAX_P] ? nla_get_u32(tb[TCA_GRED_MAX_P]) : 0;

	err = -EINVAL;
	ctl = nla_data(tb[TCA_GRED_PARMS]);
	stab = nla_data(tb[TCA_GRED_STAB]);

	if (ctl->DP >= table->DPs)
		goto errout;

	if (gred_rio_mode(table)) {
		if (ctl->prio == 0) {
			int def_prio = GRED_DEF_PRIO;

			if (table->tab[table->def])
				def_prio = table->tab[table->def]->prio;

			printk(KERN_DEBUG "GRED: DP %u does not have a prio "
			       "setting default to %d\n", ctl->DP, def_prio);

			prio = def_prio;
		} else
			prio = ctl->prio;
	}

	prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
	sch_tree_lock(sch);

	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab, max_P, &prealloc);
	if (err < 0)
		goto errout_locked;

	if (gred_rio_mode(table)) {
		gred_disable_wred_mode(table);
		if (gred_wred_mode_check(sch))
			gred_enable_wred_mode(table);
	}

	err = 0;

errout_locked:
	sch_tree_unlock(sch);
	kfree(prealloc);
errout:
	return err;
}

static int gred_init(struct Qdisc *sch, struct nlattr *opt)
{
	struct nlattr *tb[TCA_GRED_MAX + 1];
	int err;

	if (opt == NULL)
		return -EINVAL;

	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
	if (err < 0)
		return err;

	if (tb[TCA_GRED_PARMS] || tb[TCA_GRED_STAB])
		return -EINVAL;

	if (tb[TCA_GRED_LIMIT])
		sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
	else
		sch->limit = qdisc_dev(sch)->tx_queue_len
		             * psched_mtu(qdisc_dev(sch));

	return gred_change_table_def(sch, tb[TCA_GRED_DPS]);
}

static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct nlattr *parms, *opts = NULL;
	int i;
	u32 max_p[MAX_DPs];
	struct tc_gred_sopt sopt = {
		.DPs	= table->DPs,
		.def_DP	= table->def,
		.grio	= gred_rio_mode(table),
		.flags	= table->red_flags,
	};

	opts = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;
	if (nla_put(skb, TCA_GRED_DPS, sizeof(sopt), &sopt))
		goto nla_put_failure;

	for (i = 0; i < MAX_DPs; i++) {
		struct gred_sched_data *q = table->tab[i];

		max_p[i] = q ? q->parms.max_P : 0;
	}
	if (nla_put(skb, TCA_GRED_MAX_P, sizeof(max_p), max_p))
		goto nla_put_failure;

	if (nla_put_u32(skb, TCA_GRED_LIMIT, sch->limit))
		goto nla_put_failure;

	parms = nla_nest_start(skb, TCA_GRED_PARMS);
	if (parms == NULL)
		goto nla_put_failure;

	for (i = 0; i < MAX_DPs; i++) {
		struct gred_sched_data *q = table->tab[i];
		struct tc_gred_qopt opt;
		unsigned long qavg;

		memset(&opt, 0, sizeof(opt));

		if (!q) {
			/* hack -- fix at some point with proper message
			   This is how we indicate to tc that there is no VQ
			   at this DP */

			opt.DP = MAX_DPs + i;
			goto append_opt;
		}

		opt.limit	= q->limit;
		opt.DP		= q->DP;
		opt.backlog	= gred_backlog(table, q, sch);
		opt.prio	= q->prio;
		opt.qth_min	= q->parms.qth_min >> q->parms.Wlog;
		opt.qth_max	= q->parms.qth_max >> q->parms.Wlog;
		opt.Wlog	= q->parms.Wlog;
		opt.Plog	= q->parms.Plog;
		opt.Scell_log	= q->parms.Scell_log;
		opt.other	= q->stats.other;
		opt.early	= q->stats.prob_drop;
		opt.forced	= q->stats.forced_drop;
		opt.pdrop	= q->stats.pdrop;
		opt.packets	= q->packetsin;
		opt.bytesin	= q->bytesin;

		if (gred_wred_mode(table))
			gred_load_wred_set(table, q);

		qavg = red_calc_qavg(&q->parms, &q->vars,
				     q->vars.qavg >> q->parms.Wlog);
		opt.qave = qavg >> q->parms.Wlog;

append_opt:
		if (nla_append(skb, sizeof(opt), &opt) < 0)
			goto nla_put_failure;
	}

	nla_nest_end(skb, parms);

	return nla_nest_end(skb, opts);

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

static void gred_destroy(struct Qdisc *sch)
{
	struct gred_sched *table = qdisc_priv(sch);
	int i;

	for (i = 0; i < table->DPs; i++) {
		if (table->tab[i])
			gred_destroy_vq(table->tab[i]);
	}
}

static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
	.id		=	"gred",
	.priv_size	=	sizeof(struct gred_sched),
	.enqueue	=	gred_enqueue,
	.dequeue	=	gred_dequeue,
	.peek		=	qdisc_peek_head,
	.drop		=	gred_drop,
	.init		=	gred_init,
	.reset		=	gred_reset,
	.destroy	=	gred_destroy,
	.change		=	gred_change,
	.dump		=	gred_dump,
	.owner		=	THIS_MODULE,
};

static int __init gred_module_init(void)
{
	return register_qdisc(&gred_qdisc_ops);
}

static void __exit gred_module_exit(void)
{
	unregister_qdisc(&gred_qdisc_ops);
}

module_init(gred_module_init)
module_exit(gred_module_exit)

MODULE_LICENSE("GPL");
Commit	Line	Data
1da177e4 LT	1	/*
	2	* net/sched/sch_gred.c Generic Random Early Detection queue.
	3	*
	4	*
	5	* This program is free software; you can redistribute it and/or
	6	* modify it under the terms of the GNU General Public License
	7	* as published by the Free Software Foundation; either version
	8	* 2 of the License, or (at your option) any later version.
	9	*
	10	* Authors: J Hadi Salim (hadi@cyberus.ca) 1998-2002
	11	*
	12	* 991129: - Bug fix with grio mode
	13	* - a better sing. AvgQ mode with Grio(WRED)
	14	* - A finer grained VQ dequeue based on sugestion
	15	* from Ren Liu
	16	* - More error checks
	17	*
1e4dfaf9	18	* For all the glorious comments look at include/net/red.h
1da177e4 LT	19	*/
1da177e4 LT	20
5a0e3ad6	21	#include <linux/slab.h>
1da177e4	22	#include <linux/module.h>
1da177e4 LT	23	#include <linux/types.h>
1da177e4 LT	24	#include <linux/kernel.h>
1da177e4	25	#include <linux/skbuff.h>
1da177e4	26	#include <net/pkt_sched.h>
22b33429	27	#include <net/red.h>
1da177e4	28
f62d6b93	29	#define GRED_DEF_PRIO (MAX_DPs / 2)
716a1b40	30	#define GRED_VQ_MASK (MAX_DPs - 1)
f62d6b93	31
1da177e4 LT	32	struct gred_sched_data;
	33	struct gred_sched;
	34
cc7ec456	35	struct gred_sched_data {
1da177e4	36	u32 limit; /* HARD maximal queue length */
a73ed26b	37	u32 DP; /* the drop parameters */
1da177e4 LT	38	u32 bytesin; /* bytes seen on virtualQ so far*/
	39	u32 packetsin; /* packets seen on virtualQ so far*/
	40	u32 backlog; /* bytes on the virtualQ */
1e4dfaf9	41	u8 prio; /* the prio of this vq */
1da177e4	42
22b33429	43	struct red_parms parms;
eeca6688	44	struct red_vars vars;
22b33429	45	struct red_stats stats;
1da177e4 LT	46	};
1da177e4 LT	47
dea3f628 TG	48	enum {
dea3f628 TG	49	GRED_WRED_MODE = 1,
d6fd4e96	50	GRED_RIO_MODE,
dea3f628 TG	51	};
dea3f628 TG	52
cc7ec456	53	struct gred_sched {
1da177e4	54	struct gred_sched_data *tab[MAX_DPs];
dea3f628	55	unsigned long flags;
b38c7eef	56	u32 red_flags;
1e4dfaf9 TG	57	u32 DPs;
1e4dfaf9 TG	58	u32 def;
eeca6688	59	struct red_vars wred_set;
1da177e4 LT	60	};
1da177e4 LT	61
dea3f628 TG	62	static inline int gred_wred_mode(struct gred_sched *table)
	63	{
	64	return test_bit(GRED_WRED_MODE, &table->flags);
	65	}
	66
	67	static inline void gred_enable_wred_mode(struct gred_sched *table)
	68	{
	69	__set_bit(GRED_WRED_MODE, &table->flags);
	70	}
	71
	72	static inline void gred_disable_wred_mode(struct gred_sched *table)
	73	{
	74	__clear_bit(GRED_WRED_MODE, &table->flags);
	75	}
	76
d6fd4e96 TG	77	static inline int gred_rio_mode(struct gred_sched *table)
	78	{
	79	return test_bit(GRED_RIO_MODE, &table->flags);
	80	}
	81
	82	static inline void gred_enable_rio_mode(struct gred_sched *table)
	83	{
	84	__set_bit(GRED_RIO_MODE, &table->flags);
	85	}
	86
	87	static inline void gred_disable_rio_mode(struct gred_sched *table)
	88	{
	89	__clear_bit(GRED_RIO_MODE, &table->flags);
	90	}
	91
dea3f628 TG	92	static inline int gred_wred_mode_check(struct Qdisc *sch)
	93	{
	94	struct gred_sched *table = qdisc_priv(sch);
	95	int i;
	96
	97	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
	98	for (i = 0; i < table->DPs; i++) {
	99	struct gred_sched_data *q = table->tab[i];
	100	int n;
	101
	102	if (q == NULL)
	103	continue;
	104
c22e4640 DW	105	for (n = i + 1; n < table->DPs; n++)
c22e4640 DW	106	if (table->tab[n] && table->tab[n]->prio == q->prio)
dea3f628 TG	107	return 1;
	108	}
	109
	110	return 0;
	111	}
	112
22b33429 TG	113	static inline unsigned int gred_backlog(struct gred_sched *table,
	114	struct gred_sched_data *q,
	115	struct Qdisc *sch)
	116	{
	117	if (gred_wred_mode(table))
	118	return sch->qstats.backlog;
	119	else
	120	return q->backlog;
	121	}
	122
716a1b40 TG	123	static inline u16 tc_index_to_dp(struct sk_buff *skb)
	124	{
	125	return skb->tc_index & GRED_VQ_MASK;
	126	}
	127
eeca6688	128	static inline void gred_load_wred_set(const struct gred_sched *table,
7051703b TG	129	struct gred_sched_data *q)
7051703b TG	130	{
eeca6688 ED	131	q->vars.qavg = table->wred_set.qavg;
eeca6688 ED	132	q->vars.qidlestart = table->wred_set.qidlestart;
7051703b TG	133	}
	134
	135	static inline void gred_store_wred_set(struct gred_sched *table,
	136	struct gred_sched_data *q)
	137	{
eeca6688	138	table->wred_set.qavg = q->vars.qavg;
ba1bf474	139	table->wred_set.qidlestart = q->vars.qidlestart;
7051703b TG	140	}
7051703b TG	141
b38c7eef TG	142	static inline int gred_use_ecn(struct gred_sched *t)
	143	{
	144	return t->red_flags & TC_RED_ECN;
	145	}
	146
bdc450a0 TG	147	static inline int gred_use_harddrop(struct gred_sched *t)
	148	{
	149	return t->red_flags & TC_RED_HARDDROP;
	150	}
	151
cc7ec456	152	static int gred_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	153	{
cc7ec456 ED	154	struct gred_sched_data *q = NULL;
cc7ec456 ED	155	struct gred_sched *t = qdisc_priv(sch);
22b33429	156	unsigned long qavg = 0;
4a591834	157	u16 dp = tc_index_to_dp(skb);
1da177e4	158
cc7ec456	159	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
18e3fb84 TG	160	dp = t->def;
18e3fb84 TG	161
cc7ec456 ED	162	q = t->tab[dp];
cc7ec456 ED	163	if (!q) {
18e3fb84 TG	164	/* Pass through packets not assigned to a DP
	165	* if no default DP has been configured. This
	166	* allows for DP flows to be left untouched.
	167	*/
a3eb95f8 DW	168	if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <=
a3eb95f8 DW	169	sch->limit))
18e3fb84 TG	170	return qdisc_enqueue_tail(skb, sch);
	171	else
	172	goto drop;
1da177e4	173	}
18e3fb84	174
eeca6688	175	/* fix tc_index? --could be controversial but needed for
1da177e4	176	requeueing */
18e3fb84	177	skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) \| dp;
1da177e4 LT	178	}
1da177e4 LT	179
e29fe837	180	/* sum up all the qaves of prios < ours to get the new qave */
d6fd4e96	181	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
1e4dfaf9 TG	182	int i;
	183
	184	for (i = 0; i < t->DPs; i++) {
	185	if (t->tab[i] && t->tab[i]->prio < q->prio &&
eeca6688 ED	186	!red_is_idling(&t->tab[i]->vars))
eeca6688 ED	187	qavg += t->tab[i]->vars.qavg;
1da177e4	188	}
1e4dfaf9	189
1da177e4 LT	190	}
	191
	192	q->packetsin++;
0abf77e5	193	q->bytesin += qdisc_pkt_len(skb);
1da177e4	194
1e4dfaf9	195	if (gred_wred_mode(t))
7051703b	196	gred_load_wred_set(t, q);
1da177e4	197
eeca6688 ED	198	q->vars.qavg = red_calc_qavg(&q->parms,
	199	&q->vars,
	200	gred_backlog(t, q, sch));
1da177e4	201
eeca6688 ED	202	if (red_is_idling(&q->vars))
eeca6688 ED	203	red_end_of_idle_period(&q->vars);
1da177e4	204
dea3f628	205	if (gred_wred_mode(t))
7051703b	206	gred_store_wred_set(t, q);
1da177e4	207
eeca6688	208	switch (red_action(&q->parms, &q->vars, q->vars.qavg + qavg)) {
cc7ec456 ED	209	case RED_DONT_MARK:
	210	break;
	211
	212	case RED_PROB_MARK:
25331d6c	213	qdisc_qstats_overlimit(sch);
cc7ec456 ED	214	if (!gred_use_ecn(t) \|\| !INET_ECN_set_ce(skb)) {
	215	q->stats.prob_drop++;
	216	goto congestion_drop;
	217	}
	218
	219	q->stats.prob_mark++;
	220	break;
	221
	222	case RED_HARD_MARK:
25331d6c	223	qdisc_qstats_overlimit(sch);
cc7ec456 ED	224	if (gred_use_harddrop(t) \|\| !gred_use_ecn(t) \|\|
	225	!INET_ECN_set_ce(skb)) {
	226	q->stats.forced_drop++;
	227	goto congestion_drop;
	228	}
	229	q->stats.forced_mark++;
	230	break;
1da177e4	231	}
22b33429	232
145a42b3	233	if (gred_backlog(t, q, sch) + qdisc_pkt_len(skb) <= q->limit) {
0abf77e5	234	q->backlog += qdisc_pkt_len(skb);
edf7a7b1	235	return qdisc_enqueue_tail(skb, sch);
1da177e4	236	}
22b33429 TG	237
	238	q->stats.pdrop++;
	239	drop:
edf7a7b1	240	return qdisc_drop(skb, sch);
c3b553cd TG	241
c3b553cd TG	242	congestion_drop:
edf7a7b1	243	qdisc_drop(skb, sch);
c3b553cd	244	return NET_XMIT_CN;
1da177e4 LT	245	}
1da177e4 LT	246
cc7ec456	247	static struct sk_buff gred_dequeue(struct Qdisc sch)
1da177e4 LT	248	{
1da177e4 LT	249	struct sk_buff *skb;
1e4dfaf9	250	struct gred_sched *t = qdisc_priv(sch);
1da177e4	251
edf7a7b1 TG	252	skb = qdisc_dequeue_head(sch);
edf7a7b1 TG	253
1da177e4	254	if (skb) {
1e4dfaf9	255	struct gred_sched_data *q;
18e3fb84 TG	256	u16 dp = tc_index_to_dp(skb);
	257
	258	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
e87cc472 JP	259	net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x after dequeue, screwing up backlog\n",
e87cc472 JP	260	tc_index_to_dp(skb));
18e3fb84	261	} else {
0abf77e5	262	q->backlog -= qdisc_pkt_len(skb);
18e3fb84	263
ba1bf474 DW	264	if (gred_wred_mode(t)) {
	265	if (!sch->qstats.backlog)
	266	red_start_of_idle_period(&t->wred_set);
	267	} else {
	268	if (!q->backlog)
	269	red_start_of_idle_period(&q->vars);
	270	}
1da177e4	271	}
18e3fb84	272
1da177e4 LT	273	return skb;
	274	}
	275
1da177e4 LT	276	return NULL;
	277	}
	278
cc7ec456	279	static unsigned int gred_drop(struct Qdisc *sch)
1da177e4 LT	280	{
1da177e4 LT	281	struct sk_buff *skb;
1e4dfaf9	282	struct gred_sched *t = qdisc_priv(sch);
1da177e4	283
edf7a7b1	284	skb = qdisc_dequeue_tail(sch);
1da177e4	285	if (skb) {
0abf77e5	286	unsigned int len = qdisc_pkt_len(skb);
1e4dfaf9	287	struct gred_sched_data *q;
18e3fb84 TG	288	u16 dp = tc_index_to_dp(skb);
	289
	290	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
e87cc472 JP	291	net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x while dropping, screwing up backlog\n",
e87cc472 JP	292	tc_index_to_dp(skb));
18e3fb84	293	} else {
1da177e4	294	q->backlog -= len;
22b33429	295	q->stats.other++;
18e3fb84	296
ba1bf474 DW	297	if (gred_wred_mode(t)) {
	298	if (!sch->qstats.backlog)
	299	red_start_of_idle_period(&t->wred_set);
	300	} else {
	301	if (!q->backlog)
	302	red_start_of_idle_period(&q->vars);
	303	}
1da177e4 LT	304	}
1da177e4 LT	305
edf7a7b1	306	qdisc_drop(skb, sch);
1da177e4 LT	307	return len;
	308	}
	309
1da177e4	310	return 0;
1da177e4 LT	311	}
1da177e4 LT	312
cc7ec456	313	static void gred_reset(struct Qdisc *sch)
1da177e4 LT	314	{
1da177e4 LT	315	int i;
1e4dfaf9	316	struct gred_sched *t = qdisc_priv(sch);
1da177e4	317
edf7a7b1	318	qdisc_reset_queue(sch);
1da177e4	319
10297b99	320	for (i = 0; i < t->DPs; i++) {
1e4dfaf9 TG	321	struct gred_sched_data *q = t->tab[i];
	322
	323	if (!q)
	324	continue;
	325
eeca6688	326	red_restart(&q->vars);
1da177e4	327	q->backlog = 0;
1da177e4 LT	328	}
	329	}
	330
6639607e TG	331	static inline void gred_destroy_vq(struct gred_sched_data *q)
	332	{
	333	kfree(q);
	334	}
	335
1e90474c	336	static inline int gred_change_table_def(struct Qdisc sch, struct nlattr dps)
1da177e4 LT	337	{
1da177e4 LT	338	struct gred_sched *table = qdisc_priv(sch);
1da177e4	339	struct tc_gred_sopt *sopt;
6639607e	340	int i;
1da177e4	341
27a3421e	342	if (dps == NULL)
1da177e4 LT	343	return -EINVAL;
1da177e4 LT	344
1e90474c	345	sopt = nla_data(dps);
1da177e4	346
6639607e TG	347	if (sopt->DPs > MAX_DPs \|\| sopt->DPs == 0 \|\| sopt->def_DP >= sopt->DPs)
6639607e TG	348	return -EINVAL;
1da177e4	349
6639607e TG	350	sch_tree_lock(sch);
	351	table->DPs = sopt->DPs;
	352	table->def = sopt->def_DP;
b38c7eef	353	table->red_flags = sopt->flags;
dea3f628	354
6639607e TG	355	/*
	356	* Every entry point to GRED is synchronized with the above code
	357	* and the DP is checked against DPs, i.e. shadowed VQs can no
	358	* longer be found so we can unlock right here.
	359	*/
	360	sch_tree_unlock(sch);
dea3f628	361
6639607e TG	362	if (sopt->grio) {
	363	gred_enable_rio_mode(table);
	364	gred_disable_wred_mode(table);
	365	if (gred_wred_mode_check(sch))
	366	gred_enable_wred_mode(table);
	367	} else {
	368	gred_disable_rio_mode(table);
	369	gred_disable_wred_mode(table);
	370	}
	371
	372	for (i = table->DPs; i < MAX_DPs; i++) {
	373	if (table->tab[i]) {
c17988a9 YY	374	pr_warn("GRED: Warning: Destroying shadowed VQ 0x%x\n",
c17988a9 YY	375	i);
6639607e TG	376	gred_destroy_vq(table->tab[i]);
6639607e TG	377	table->tab[i] = NULL;
10297b99	378	}
6639607e	379	}
1da177e4	380
6639607e TG	381	return 0;
	382	}
	383
f62d6b93	384	static inline int gred_change_vq(struct Qdisc *sch, int dp,
a73ed26b	385	struct tc_gred_qopt *ctl, int prio,
869aa410 ED	386	u8 *stab, u32 max_P,
869aa410 ED	387	struct gred_sched_data **prealloc)
6639607e TG	388	{
6639607e TG	389	struct gred_sched *table = qdisc_priv(sch);
869aa410	390	struct gred_sched_data *q = table->tab[dp];
f62d6b93	391
869aa410 ED	392	if (!q) {
	393	table->tab[dp] = q = *prealloc;
	394	*prealloc = NULL;
	395	if (!q)
f62d6b93	396	return -ENOMEM;
f62d6b93 TG	397	}
f62d6b93 TG	398
f62d6b93 TG	399	q->DP = dp;
f62d6b93 TG	400	q->prio = prio;
a3eb95f8 DW	401	if (ctl->limit > sch->limit)
	402	q->limit = sch->limit;
	403	else
	404	q->limit = ctl->limit;
22b33429 TG	405
22b33429 TG	406	if (q->backlog == 0)
eeca6688	407	red_end_of_idle_period(&q->vars);
22b33429 TG	408
	409	red_set_parms(&q->parms,
	410	ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
a73ed26b	411	ctl->Scell_log, stab, max_P);
eeca6688	412	red_set_vars(&q->vars);
f62d6b93 TG	413	return 0;
	414	}
	415
27a3421e PM	416	static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
	417	[TCA_GRED_PARMS] = { .len = sizeof(struct tc_gred_qopt) },
	418	[TCA_GRED_STAB] = { .len = 256 },
	419	[TCA_GRED_DPS] = { .len = sizeof(struct tc_gred_sopt) },
a73ed26b	420	[TCA_GRED_MAX_P] = { .type = NLA_U32 },
a3eb95f8	421	[TCA_GRED_LIMIT] = { .type = NLA_U32 },
27a3421e PM	422	};
27a3421e PM	423
1e90474c	424	static int gred_change(struct Qdisc sch, struct nlattr opt)
f62d6b93 TG	425	{
f62d6b93 TG	426	struct gred_sched *table = qdisc_priv(sch);
6639607e	427	struct tc_gred_qopt *ctl;
1e90474c	428	struct nlattr *tb[TCA_GRED_MAX + 1];
cee63723	429	int err, prio = GRED_DEF_PRIO;
f62d6b93	430	u8 *stab;
a73ed26b	431	u32 max_P;
869aa410	432	struct gred_sched_data *prealloc;
6639607e	433
cee63723	434	if (opt == NULL)
6639607e TG	435	return -EINVAL;
6639607e TG	436
27a3421e	437	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
cee63723 PM	438	if (err < 0)
	439	return err;
	440
a3eb95f8 DW	441	if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL) {
	442	if (tb[TCA_GRED_LIMIT] != NULL)
	443	sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
f62d6b93	444	return gred_change_table_def(sch, opt);
a3eb95f8	445	}
1da177e4	446
1e90474c	447	if (tb[TCA_GRED_PARMS] == NULL \|\|
a3eb95f8 DW	448	tb[TCA_GRED_STAB] == NULL \|\|
a3eb95f8 DW	449	tb[TCA_GRED_LIMIT] != NULL)
f62d6b93	450	return -EINVAL;
1da177e4	451
a73ed26b ED	452	max_P = tb[TCA_GRED_MAX_P] ? nla_get_u32(tb[TCA_GRED_MAX_P]) : 0;
a73ed26b ED	453
cee63723	454	err = -EINVAL;
1e90474c PM	455	ctl = nla_data(tb[TCA_GRED_PARMS]);
1e90474c PM	456	stab = nla_data(tb[TCA_GRED_STAB]);
a8aaa995 TG	457
a8aaa995 TG	458	if (ctl->DP >= table->DPs)
f62d6b93	459	goto errout;
1da177e4	460
d6fd4e96	461	if (gred_rio_mode(table)) {
f62d6b93 TG	462	if (ctl->prio == 0) {
	463	int def_prio = GRED_DEF_PRIO;
	464
	465	if (table->tab[table->def])
	466	def_prio = table->tab[table->def]->prio;
	467
	468	printk(KERN_DEBUG "GRED: DP %u does not have a prio "
	469	"setting default to %d\n", ctl->DP, def_prio);
	470
	471	prio = def_prio;
	472	} else
	473	prio = ctl->prio;
1da177e4 LT	474	}
1da177e4 LT	475
869aa410	476	prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
f62d6b93	477	sch_tree_lock(sch);
1da177e4	478
869aa410	479	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab, max_P, &prealloc);
f62d6b93 TG	480	if (err < 0)
f62d6b93 TG	481	goto errout_locked;
1da177e4	482
d6fd4e96	483	if (gred_rio_mode(table)) {
dea3f628 TG	484	gred_disable_wred_mode(table);
	485	if (gred_wred_mode_check(sch))
	486	gred_enable_wred_mode(table);
1da177e4 LT	487	}
1da177e4 LT	488
f62d6b93	489	err = 0;
1da177e4	490
f62d6b93 TG	491	errout_locked:
f62d6b93 TG	492	sch_tree_unlock(sch);
869aa410	493	kfree(prealloc);
f62d6b93 TG	494	errout:
f62d6b93 TG	495	return err;
1da177e4 LT	496	}
1da177e4 LT	497
1e90474c	498	static int gred_init(struct Qdisc sch, struct nlattr opt)
1da177e4	499	{
1e90474c	500	struct nlattr *tb[TCA_GRED_MAX + 1];
cee63723	501	int err;
1da177e4	502
cee63723	503	if (opt == NULL)
1da177e4 LT	504	return -EINVAL;
1da177e4 LT	505
27a3421e	506	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
cee63723 PM	507	if (err < 0)
	508	return err;
	509
1e90474c	510	if (tb[TCA_GRED_PARMS] \|\| tb[TCA_GRED_STAB])
6639607e	511	return -EINVAL;
1da177e4	512
a3eb95f8 DW	513	if (tb[TCA_GRED_LIMIT])
a3eb95f8 DW	514	sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
348e3435 PS	515	else
	516	sch->limit = qdisc_dev(sch)->tx_queue_len
	517	* psched_mtu(qdisc_dev(sch));
a3eb95f8	518
1e90474c	519	return gred_change_table_def(sch, tb[TCA_GRED_DPS]);
1da177e4 LT	520	}
	521
	522	static int gred_dump(struct Qdisc sch, struct sk_buff skb)
	523	{
1da177e4	524	struct gred_sched *table = qdisc_priv(sch);
1e90474c	525	struct nlattr parms, opts = NULL;
1da177e4	526	int i;
a73ed26b	527	u32 max_p[MAX_DPs];
e0636822 TG	528	struct tc_gred_sopt sopt = {
	529	.DPs = table->DPs,
	530	.def_DP = table->def,
	531	.grio = gred_rio_mode(table),
b38c7eef	532	.flags = table->red_flags,
e0636822	533	};
1da177e4	534
1e90474c PM	535	opts = nla_nest_start(skb, TCA_OPTIONS);
	536	if (opts == NULL)
	537	goto nla_put_failure;
1b34ec43 DM	538	if (nla_put(skb, TCA_GRED_DPS, sizeof(sopt), &sopt))
1b34ec43 DM	539	goto nla_put_failure;
a73ed26b ED	540
	541	for (i = 0; i < MAX_DPs; i++) {
	542	struct gred_sched_data *q = table->tab[i];
	543
	544	max_p[i] = q ? q->parms.max_P : 0;
	545	}
1b34ec43 DM	546	if (nla_put(skb, TCA_GRED_MAX_P, sizeof(max_p), max_p))
1b34ec43 DM	547	goto nla_put_failure;
a73ed26b	548
a3eb95f8 DW	549	if (nla_put_u32(skb, TCA_GRED_LIMIT, sch->limit))
	550	goto nla_put_failure;
	551
1e90474c PM	552	parms = nla_nest_start(skb, TCA_GRED_PARMS);
	553	if (parms == NULL)
	554	goto nla_put_failure;
1da177e4	555
05f1cc01 TG	556	for (i = 0; i < MAX_DPs; i++) {
	557	struct gred_sched_data *q = table->tab[i];
	558	struct tc_gred_qopt opt;
1fe37b10	559	unsigned long qavg;
1da177e4	560
05f1cc01	561	memset(&opt, 0, sizeof(opt));
1da177e4 LT	562
	563	if (!q) {
	564	/* hack -- fix at some point with proper message
	565	This is how we indicate to tc that there is no VQ
	566	at this DP */
	567
05f1cc01 TG	568	opt.DP = MAX_DPs + i;
05f1cc01 TG	569	goto append_opt;
1da177e4 LT	570	}
1da177e4 LT	571
05f1cc01 TG	572	opt.limit = q->limit;
05f1cc01 TG	573	opt.DP = q->DP;
145a42b3	574	opt.backlog = gred_backlog(table, q, sch);
05f1cc01	575	opt.prio = q->prio;
22b33429 TG	576	opt.qth_min = q->parms.qth_min >> q->parms.Wlog;
	577	opt.qth_max = q->parms.qth_max >> q->parms.Wlog;
	578	opt.Wlog = q->parms.Wlog;
	579	opt.Plog = q->parms.Plog;
	580	opt.Scell_log = q->parms.Scell_log;
	581	opt.other = q->stats.other;
	582	opt.early = q->stats.prob_drop;
	583	opt.forced = q->stats.forced_drop;
	584	opt.pdrop = q->stats.pdrop;
05f1cc01 TG	585	opt.packets = q->packetsin;
	586	opt.bytesin = q->bytesin;
	587
244b65db DW	588	if (gred_wred_mode(table))
244b65db DW	589	gred_load_wred_set(table, q);
05f1cc01	590
1fe37b10 DW	591	qavg = red_calc_qavg(&q->parms, &q->vars,
	592	q->vars.qavg >> q->parms.Wlog);
	593	opt.qave = qavg >> q->parms.Wlog;
22b33429	594
05f1cc01	595	append_opt:
1e90474c PM	596	if (nla_append(skb, sizeof(opt), &opt) < 0)
1e90474c PM	597	goto nla_put_failure;
1da177e4 LT	598	}
1da177e4 LT	599
1e90474c	600	nla_nest_end(skb, parms);
1da177e4	601
1e90474c	602	return nla_nest_end(skb, opts);
1da177e4	603
1e90474c	604	nla_put_failure:
bc3ed28c TG	605	nla_nest_cancel(skb, opts);
bc3ed28c TG	606	return -EMSGSIZE;
1da177e4 LT	607	}
	608
	609	static void gred_destroy(struct Qdisc *sch)
	610	{
	611	struct gred_sched *table = qdisc_priv(sch);
	612	int i;
	613
1e4dfaf9	614	for (i = 0; i < table->DPs; i++) {
1da177e4	615	if (table->tab[i])
6639607e	616	gred_destroy_vq(table->tab[i]);
1da177e4 LT	617	}
	618	}
	619
20fea08b	620	static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
1da177e4 LT	621	.id = "gred",
	622	.priv_size = sizeof(struct gred_sched),
	623	.enqueue = gred_enqueue,
	624	.dequeue = gred_dequeue,
8e3af978	625	.peek = qdisc_peek_head,
1da177e4 LT	626	.drop = gred_drop,
	627	.init = gred_init,
	628	.reset = gred_reset,
	629	.destroy = gred_destroy,
	630	.change = gred_change,
	631	.dump = gred_dump,
	632	.owner = THIS_MODULE,
	633	};
	634
	635	static int __init gred_module_init(void)
	636	{
	637	return register_qdisc(&gred_qdisc_ops);
	638	}
1e4dfaf9 TG	639
1e4dfaf9 TG	640	static void __exit gred_module_exit(void)
1da177e4 LT	641	{
	642	unregister_qdisc(&gred_qdisc_ops);
	643	}
1e4dfaf9	644
1da177e4 LT	645	module_init(gred_module_init)
1da177e4 LT	646	module_exit(gred_module_exit)
1e4dfaf9	647
1da177e4	648	MODULE_LICENSE("GPL");