[mirror_ubuntu-artful-kernel.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/config.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/netdevice.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 pramaters */
	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_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_parms 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 = 0; n < table->DPs; n++)
			if (table->tab[n] && table->tab[n] != q &&
			    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(struct gred_sched *table,
				      struct gred_sched_data *q)
{
	q->parms.qavg = table->wred_set.qavg;
	q->parms.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->parms.qavg;
}

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

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;

		if ((q = t->tab[dp]) == NULL) {
			/* 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 (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
				return qdisc_enqueue_tail(skb, sch);
			else
				goto drop;
		}

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

	/* sum up all the qaves of prios <= to 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]->parms))
				qavg +=t->tab[i]->parms.qavg;
		}

	}

	q->packetsin++;
	q->bytesin += skb->len;

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

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

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

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

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

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

	if (q->backlog + skb->len <= q->limit) {
		q->backlog += skb->len;
		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 int gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
{
	struct gred_sched *t = qdisc_priv(sch);
	struct gred_sched_data *q;
	u16 dp = tc_index_to_dp(skb);

	if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
		if (net_ratelimit())
			printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
			       "for requeue, screwing up backlog.\n",
			       tc_index_to_dp(skb));
	} else {
		if (red_is_idling(&q->parms))
			red_end_of_idle_period(&q->parms);
		q->backlog += skb->len;
	}

	return qdisc_requeue(skb, sch);
}

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) {
			if (net_ratelimit())
				printk(KERN_WARNING "GRED: Unable to relocate "
				       "VQ 0x%x after dequeue, screwing up "
				       "backlog.\n", tc_index_to_dp(skb));
		} else {
			q->backlog -= skb->len;

			if (!q->backlog && !gred_wred_mode(t))
				red_start_of_idle_period(&q->parms);
		}

		return skb;
	}

	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
		red_start_of_idle_period(&t->wred_set);

	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 = skb->len;
		struct gred_sched_data *q;
		u16 dp = tc_index_to_dp(skb);

		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
			if (net_ratelimit())
				printk(KERN_WARNING "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 (!q->backlog && !gred_wred_mode(t))
				red_start_of_idle_period(&q->parms);
		}

		qdisc_drop(skb, sch);
		return len;
	}

	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
		red_start_of_idle_period(&t->wred_set);

	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->parms);
		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 rtattr *dps)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct tc_gred_sopt *sopt;
	int i;

	if (dps == NULL || RTA_PAYLOAD(dps) < sizeof(*sopt))
		return -EINVAL;

	sopt = RTA_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]) {
			printk(KERN_WARNING "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)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct gred_sched_data *q;

	if (table->tab[dp] == NULL) {
		table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
		if (table->tab[dp] == NULL)
			return -ENOMEM;
		memset(table->tab[dp], 0, sizeof(*q));
	}

	q = table->tab[dp];
	q->DP = dp;
	q->prio = prio;
	q->limit = ctl->limit;

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

	red_set_parms(&q->parms,
		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
		      ctl->Scell_log, stab);

	return 0;
}

static int gred_change(struct Qdisc *sch, struct rtattr *opt)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct tc_gred_qopt *ctl;
	struct rtattr *tb[TCA_GRED_MAX];
	int err = -EINVAL, prio = GRED_DEF_PRIO;
	u8 *stab;

	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
		return -EINVAL;

	if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
		return gred_change_table_def(sch, opt);

	if (tb[TCA_GRED_PARMS-1] == NULL ||
	    RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
	    tb[TCA_GRED_STAB-1] == NULL ||
	    RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
		return -EINVAL;

	ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
	stab = RTA_DATA(tb[TCA_GRED_STAB-1]);

	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;
	}

	sch_tree_lock(sch);

	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
	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);
errout:
	return err;
}

static int gred_init(struct Qdisc *sch, struct rtattr *opt)
{
	struct rtattr *tb[TCA_GRED_MAX];

	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
		return -EINVAL;

	if (tb[TCA_GRED_PARMS-1] || tb[TCA_GRED_STAB-1])
		return -EINVAL;

	return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
}

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

	opts = RTA_NEST(skb, TCA_OPTIONS);
	RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
	parms = RTA_NEST(skb, TCA_GRED_PARMS);

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

		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	= q->backlog;
		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)) {
			q->parms.qidlestart =
				table->tab[table->def]->parms.qidlestart;
			q->parms.qavg = table->tab[table->def]->parms.qavg;
		}

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

append_opt:
		RTA_APPEND(skb, sizeof(opt), &opt);
	}

	RTA_NEST_END(skb, parms);

	return RTA_NEST_END(skb, opts);

rtattr_failure:
	return RTA_NEST_CANCEL(skb, opts);
}

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 = {
	.id		=	"gred",
	.priv_size	=	sizeof(struct gred_sched),
	.enqueue	=	gred_enqueue,
	.dequeue	=	gred_dequeue,
	.requeue	=	gred_requeue,
	.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	*/
	20
	21	#include <linux/config.h>
	22	#include <linux/module.h>
1da177e4 LT	23	#include <linux/types.h>
1da177e4 LT	24	#include <linux/kernel.h>
1da177e4	25	#include <linux/netdevice.h>
1da177e4	26	#include <linux/skbuff.h>
1da177e4	27	#include <net/pkt_sched.h>
22b33429	28	#include <net/red.h>
1da177e4	29
f62d6b93	30	#define GRED_DEF_PRIO (MAX_DPs / 2)
716a1b40	31	#define GRED_VQ_MASK (MAX_DPs - 1)
f62d6b93	32
1da177e4 LT	33	struct gred_sched_data;
	34	struct gred_sched;
	35
	36	struct gred_sched_data
	37	{
1da177e4	38	u32 limit; /* HARD maximal queue length */
1da177e4	39	u32 DP; /* the drop pramaters */
1da177e4 LT	40	u32 bytesin; /* bytes seen on virtualQ so far*/
	41	u32 packetsin; /* packets seen on virtualQ so far*/
	42	u32 backlog; /* bytes on the virtualQ */
1e4dfaf9	43	u8 prio; /* the prio of this vq */
1da177e4	44
22b33429 TG	45	struct red_parms parms;
22b33429 TG	46	struct red_stats stats;
1da177e4 LT	47	};
1da177e4 LT	48
dea3f628 TG	49	enum {
dea3f628 TG	50	GRED_WRED_MODE = 1,
d6fd4e96	51	GRED_RIO_MODE,
dea3f628 TG	52	};
dea3f628 TG	53
1da177e4 LT	54	struct gred_sched
	55	{
	56	struct gred_sched_data *tab[MAX_DPs];
dea3f628	57	unsigned long flags;
b38c7eef	58	u32 red_flags;
1e4dfaf9 TG	59	u32 DPs;
1e4dfaf9 TG	60	u32 def;
7051703b	61	struct red_parms wred_set;
1da177e4 LT	62	};
1da177e4 LT	63
dea3f628 TG	64	static inline int gred_wred_mode(struct gred_sched *table)
	65	{
	66	return test_bit(GRED_WRED_MODE, &table->flags);
	67	}
	68
	69	static inline void gred_enable_wred_mode(struct gred_sched *table)
	70	{
	71	__set_bit(GRED_WRED_MODE, &table->flags);
	72	}
	73
	74	static inline void gred_disable_wred_mode(struct gred_sched *table)
	75	{
	76	__clear_bit(GRED_WRED_MODE, &table->flags);
	77	}
	78
d6fd4e96 TG	79	static inline int gred_rio_mode(struct gred_sched *table)
	80	{
	81	return test_bit(GRED_RIO_MODE, &table->flags);
	82	}
	83
	84	static inline void gred_enable_rio_mode(struct gred_sched *table)
	85	{
	86	__set_bit(GRED_RIO_MODE, &table->flags);
	87	}
	88
	89	static inline void gred_disable_rio_mode(struct gred_sched *table)
	90	{
	91	__clear_bit(GRED_RIO_MODE, &table->flags);
	92	}
	93
dea3f628 TG	94	static inline int gred_wred_mode_check(struct Qdisc *sch)
	95	{
	96	struct gred_sched *table = qdisc_priv(sch);
	97	int i;
	98
	99	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
	100	for (i = 0; i < table->DPs; i++) {
	101	struct gred_sched_data *q = table->tab[i];
	102	int n;
	103
	104	if (q == NULL)
	105	continue;
	106
	107	for (n = 0; n < table->DPs; n++)
	108	if (table->tab[n] && table->tab[n] != q &&
	109	table->tab[n]->prio == q->prio)
	110	return 1;
	111	}
	112
	113	return 0;
	114	}
	115
22b33429 TG	116	static inline unsigned int gred_backlog(struct gred_sched *table,
	117	struct gred_sched_data *q,
	118	struct Qdisc *sch)
	119	{
	120	if (gred_wred_mode(table))
	121	return sch->qstats.backlog;
	122	else
	123	return q->backlog;
	124	}
	125
716a1b40 TG	126	static inline u16 tc_index_to_dp(struct sk_buff *skb)
	127	{
	128	return skb->tc_index & GRED_VQ_MASK;
	129	}
	130
7051703b TG	131	static inline void gred_load_wred_set(struct gred_sched *table,
	132	struct gred_sched_data *q)
	133	{
	134	q->parms.qavg = table->wred_set.qavg;
	135	q->parms.qidlestart = table->wred_set.qidlestart;
	136	}
	137
	138	static inline void gred_store_wred_set(struct gred_sched *table,
	139	struct gred_sched_data *q)
	140	{
	141	table->wred_set.qavg = q->parms.qavg;
	142	}
	143
b38c7eef TG	144	static inline int gred_use_ecn(struct gred_sched *t)
	145	{
	146	return t->red_flags & TC_RED_ECN;
	147	}
	148
1e4dfaf9	149	static int gred_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	150	{
1da177e4 LT	151	struct gred_sched_data *q=NULL;
1da177e4 LT	152	struct gred_sched *t= qdisc_priv(sch);
22b33429	153	unsigned long qavg = 0;
4a591834	154	u16 dp = tc_index_to_dp(skb);
1da177e4	155
716a1b40	156	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
18e3fb84 TG	157	dp = t->def;
	158
	159	if ((q = t->tab[dp]) == NULL) {
	160	/* Pass through packets not assigned to a DP
	161	* if no default DP has been configured. This
	162	* allows for DP flows to be left untouched.
	163	*/
	164	if (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
	165	return qdisc_enqueue_tail(skb, sch);
	166	else
	167	goto drop;
1da177e4	168	}
18e3fb84	169
1da177e4 LT	170	/* fix tc_index? --could be controvesial but needed for
1da177e4 LT	171	requeueing */
18e3fb84	172	skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) \| dp;
1da177e4 LT	173	}
1da177e4 LT	174
1e4dfaf9	175	/* sum up all the qaves of prios <= to ours to get the new qave */
d6fd4e96	176	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
1e4dfaf9 TG	177	int i;
	178
	179	for (i = 0; i < t->DPs; i++) {
	180	if (t->tab[i] && t->tab[i]->prio < q->prio &&
22b33429 TG	181	!red_is_idling(&t->tab[i]->parms))
22b33429 TG	182	qavg +=t->tab[i]->parms.qavg;
1da177e4	183	}
1e4dfaf9	184
1da177e4 LT	185	}
	186
	187	q->packetsin++;
1e4dfaf9	188	q->bytesin += skb->len;
1da177e4	189
1e4dfaf9	190	if (gred_wred_mode(t))
7051703b	191	gred_load_wred_set(t, q);
1da177e4	192
22b33429	193	q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
1da177e4	194
22b33429 TG	195	if (red_is_idling(&q->parms))
22b33429 TG	196	red_end_of_idle_period(&q->parms);
1da177e4	197
dea3f628	198	if (gred_wred_mode(t))
7051703b	199	gred_store_wred_set(t, q);
1da177e4	200
22b33429 TG	201	switch (red_action(&q->parms, q->parms.qavg + qavg)) {
	202	case RED_DONT_MARK:
	203	break;
1da177e4	204
22b33429 TG	205	case RED_PROB_MARK:
22b33429 TG	206	sch->qstats.overlimits++;
b38c7eef TG	207	if (!gred_use_ecn(t) \|\| !INET_ECN_set_ce(skb)) {
	208	q->stats.prob_drop++;
	209	goto congestion_drop;
	210	}
	211
	212	q->stats.prob_mark++;
	213	break;
22b33429 TG	214
	215	case RED_HARD_MARK:
	216	sch->qstats.overlimits++;
b38c7eef TG	217	if (!gred_use_ecn(t) \|\| !INET_ECN_set_ce(skb)) {
	218	q->stats.forced_drop++;
	219	goto congestion_drop;
	220	}
	221	q->stats.forced_mark++;
	222	break;
1da177e4	223	}
22b33429 TG	224
	225	if (q->backlog + skb->len <= q->limit) {
	226	q->backlog += skb->len;
edf7a7b1	227	return qdisc_enqueue_tail(skb, sch);
1da177e4	228	}
22b33429 TG	229
	230	q->stats.pdrop++;
	231	drop:
edf7a7b1	232	return qdisc_drop(skb, sch);
c3b553cd TG	233
c3b553cd TG	234	congestion_drop:
edf7a7b1	235	qdisc_drop(skb, sch);
c3b553cd	236	return NET_XMIT_CN;
1da177e4 LT	237	}
1da177e4 LT	238
1e4dfaf9	239	static int gred_requeue(struct sk_buff skb, struct Qdisc sch)
1da177e4	240	{
716a1b40	241	struct gred_sched *t = qdisc_priv(sch);
18e3fb84 TG	242	struct gred_sched_data *q;
	243	u16 dp = tc_index_to_dp(skb);
	244
	245	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
	246	if (net_ratelimit())
	247	printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
	248	"for requeue, screwing up backlog.\n",
	249	tc_index_to_dp(skb));
	250	} else {
	251	if (red_is_idling(&q->parms))
	252	red_end_of_idle_period(&q->parms);
	253	q->backlog += skb->len;
	254	}
1da177e4	255
edf7a7b1	256	return qdisc_requeue(skb, sch);
1da177e4 LT	257	}
1da177e4 LT	258
1e4dfaf9	259	static struct sk_buff gred_dequeue(struct Qdisc sch)
1da177e4 LT	260	{
1da177e4 LT	261	struct sk_buff *skb;
1e4dfaf9	262	struct gred_sched *t = qdisc_priv(sch);
1da177e4	263
edf7a7b1 TG	264	skb = qdisc_dequeue_head(sch);
edf7a7b1 TG	265
1da177e4	266	if (skb) {
1e4dfaf9	267	struct gred_sched_data *q;
18e3fb84 TG	268	u16 dp = tc_index_to_dp(skb);
	269
	270	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
	271	if (net_ratelimit())
	272	printk(KERN_WARNING "GRED: Unable to relocate "
	273	"VQ 0x%x after dequeue, screwing up "
	274	"backlog.\n", tc_index_to_dp(skb));
	275	} else {
1da177e4	276	q->backlog -= skb->len;
18e3fb84	277
dea3f628	278	if (!q->backlog && !gred_wred_mode(t))
22b33429	279	red_start_of_idle_period(&q->parms);
1da177e4	280	}
18e3fb84	281
1da177e4 LT	282	return skb;
	283	}
	284
d8f64e19	285	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
7051703b	286	red_start_of_idle_period(&t->wred_set);
1da177e4 LT	287
	288	return NULL;
	289	}
	290
	291	static unsigned int gred_drop(struct Qdisc* sch)
	292	{
	293	struct sk_buff *skb;
1e4dfaf9	294	struct gred_sched *t = qdisc_priv(sch);
1da177e4	295
edf7a7b1	296	skb = qdisc_dequeue_tail(sch);
1da177e4 LT	297	if (skb) {
1da177e4 LT	298	unsigned int len = skb->len;
1e4dfaf9	299	struct gred_sched_data *q;
18e3fb84 TG	300	u16 dp = tc_index_to_dp(skb);
	301
	302	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
	303	if (net_ratelimit())
	304	printk(KERN_WARNING "GRED: Unable to relocate "
	305	"VQ 0x%x while dropping, screwing up "
	306	"backlog.\n", tc_index_to_dp(skb));
	307	} else {
1da177e4	308	q->backlog -= len;
22b33429	309	q->stats.other++;
18e3fb84	310
dea3f628	311	if (!q->backlog && !gred_wred_mode(t))
22b33429	312	red_start_of_idle_period(&q->parms);
1da177e4 LT	313	}
1da177e4 LT	314
edf7a7b1	315	qdisc_drop(skb, sch);
1da177e4 LT	316	return len;
	317	}
	318
d8f64e19	319	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
7051703b	320	red_start_of_idle_period(&t->wred_set);
1da177e4	321
1da177e4 LT	322	return 0;
	323
	324	}
	325
	326	static void gred_reset(struct Qdisc* sch)
	327	{
	328	int i;
1e4dfaf9	329	struct gred_sched *t = qdisc_priv(sch);
1da177e4	330
edf7a7b1	331	qdisc_reset_queue(sch);
1da177e4	332
1e4dfaf9 TG	333	for (i = 0; i < t->DPs; i++) {
	334	struct gred_sched_data *q = t->tab[i];
	335
	336	if (!q)
	337	continue;
	338
22b33429	339	red_restart(&q->parms);
1da177e4	340	q->backlog = 0;
1da177e4 LT	341	}
	342	}
	343
6639607e TG	344	static inline void gred_destroy_vq(struct gred_sched_data *q)
	345	{
	346	kfree(q);
	347	}
	348
	349	static inline int gred_change_table_def(struct Qdisc sch, struct rtattr dps)
1da177e4 LT	350	{
1da177e4 LT	351	struct gred_sched *table = qdisc_priv(sch);
1da177e4	352	struct tc_gred_sopt *sopt;
6639607e	353	int i;
1da177e4	354
6639607e	355	if (dps == NULL \|\| RTA_PAYLOAD(dps) < sizeof(*sopt))
1da177e4 LT	356	return -EINVAL;
1da177e4 LT	357
6639607e	358	sopt = RTA_DATA(dps);
1da177e4	359
6639607e TG	360	if (sopt->DPs > MAX_DPs \|\| sopt->DPs == 0 \|\| sopt->def_DP >= sopt->DPs)
6639607e TG	361	return -EINVAL;
1da177e4	362
6639607e TG	363	sch_tree_lock(sch);
	364	table->DPs = sopt->DPs;
	365	table->def = sopt->def_DP;
b38c7eef	366	table->red_flags = sopt->flags;
dea3f628	367
6639607e TG	368	/*
	369	* Every entry point to GRED is synchronized with the above code
	370	* and the DP is checked against DPs, i.e. shadowed VQs can no
	371	* longer be found so we can unlock right here.
	372	*/
	373	sch_tree_unlock(sch);
dea3f628	374
6639607e TG	375	if (sopt->grio) {
	376	gred_enable_rio_mode(table);
	377	gred_disable_wred_mode(table);
	378	if (gred_wred_mode_check(sch))
	379	gred_enable_wred_mode(table);
	380	} else {
	381	gred_disable_rio_mode(table);
	382	gred_disable_wred_mode(table);
	383	}
	384
	385	for (i = table->DPs; i < MAX_DPs; i++) {
	386	if (table->tab[i]) {
	387	printk(KERN_WARNING "GRED: Warning: Destroying "
	388	"shadowed VQ 0x%x\n", i);
	389	gred_destroy_vq(table->tab[i]);
	390	table->tab[i] = NULL;
	391	}
	392	}
1da177e4	393
6639607e TG	394	return 0;
	395	}
	396
f62d6b93 TG	397	static inline int gred_change_vq(struct Qdisc *sch, int dp,
f62d6b93 TG	398	struct tc_gred_qopt ctl, int prio, u8 stab)
6639607e TG	399	{
	400	struct gred_sched *table = qdisc_priv(sch);
	401	struct gred_sched_data *q;
f62d6b93 TG	402
	403	if (table->tab[dp] == NULL) {
	404	table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
	405	if (table->tab[dp] == NULL)
	406	return -ENOMEM;
	407	memset(table->tab[dp], 0, sizeof(*q));
	408	}
	409
	410	q = table->tab[dp];
	411	q->DP = dp;
	412	q->prio = prio;
f62d6b93	413	q->limit = ctl->limit;
22b33429 TG	414
	415	if (q->backlog == 0)
	416	red_end_of_idle_period(&q->parms);
	417
	418	red_set_parms(&q->parms,
	419	ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
	420	ctl->Scell_log, stab);
	421
f62d6b93 TG	422	return 0;
	423	}
	424
	425	static int gred_change(struct Qdisc sch, struct rtattr opt)
	426	{
	427	struct gred_sched *table = qdisc_priv(sch);
6639607e	428	struct tc_gred_qopt *ctl;
f62d6b93 TG	429	struct rtattr *tb[TCA_GRED_MAX];
	430	int err = -EINVAL, prio = GRED_DEF_PRIO;
	431	u8 *stab;
6639607e	432
f62d6b93	433	if (opt == NULL \|\| rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
6639607e TG	434	return -EINVAL;
	435
	436	if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
f62d6b93	437	return gred_change_table_def(sch, opt);
1da177e4	438
f62d6b93 TG	439	if (tb[TCA_GRED_PARMS-1] == NULL \|\|
	440	RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) \|\|
	441	tb[TCA_GRED_STAB-1] == NULL \|\|
	442	RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
	443	return -EINVAL;
1da177e4 LT	444
1da177e4 LT	445	ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
f62d6b93	446	stab = RTA_DATA(tb[TCA_GRED_STAB-1]);
a8aaa995 TG	447
a8aaa995 TG	448	if (ctl->DP >= table->DPs)
f62d6b93	449	goto errout;
1da177e4	450
d6fd4e96	451	if (gred_rio_mode(table)) {
f62d6b93 TG	452	if (ctl->prio == 0) {
	453	int def_prio = GRED_DEF_PRIO;
	454
	455	if (table->tab[table->def])
	456	def_prio = table->tab[table->def]->prio;
	457
	458	printk(KERN_DEBUG "GRED: DP %u does not have a prio "
	459	"setting default to %d\n", ctl->DP, def_prio);
	460
	461	prio = def_prio;
	462	} else
	463	prio = ctl->prio;
1da177e4 LT	464	}
1da177e4 LT	465
f62d6b93	466	sch_tree_lock(sch);
1da177e4	467
f62d6b93 TG	468	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
	469	if (err < 0)
	470	goto errout_locked;
1da177e4	471
d6fd4e96	472	if (gred_rio_mode(table)) {
dea3f628 TG	473	gred_disable_wred_mode(table);
	474	if (gred_wred_mode_check(sch))
	475	gred_enable_wred_mode(table);
1da177e4 LT	476	}
1da177e4 LT	477
f62d6b93	478	err = 0;
1da177e4	479
f62d6b93 TG	480	errout_locked:
	481	sch_tree_unlock(sch);
	482	errout:
	483	return err;
1da177e4 LT	484	}
	485
	486	static int gred_init(struct Qdisc sch, struct rtattr opt)
	487	{
6639607e	488	struct rtattr *tb[TCA_GRED_MAX];
1da177e4	489
6639607e	490	if (opt == NULL \|\| rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
1da177e4 LT	491	return -EINVAL;
1da177e4 LT	492
6639607e TG	493	if (tb[TCA_GRED_PARMS-1] \|\| tb[TCA_GRED_STAB-1])
6639607e TG	494	return -EINVAL;
1da177e4	495
6639607e	496	return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
1da177e4 LT	497	}
	498
	499	static int gred_dump(struct Qdisc sch, struct sk_buff skb)
	500	{
1da177e4	501	struct gred_sched *table = qdisc_priv(sch);
05f1cc01	502	struct rtattr parms, opts = NULL;
1da177e4	503	int i;
e0636822 TG	504	struct tc_gred_sopt sopt = {
	505	.DPs = table->DPs,
	506	.def_DP = table->def,
	507	.grio = gred_rio_mode(table),
b38c7eef	508	.flags = table->red_flags,
e0636822	509	};
1da177e4	510
05f1cc01	511	opts = RTA_NEST(skb, TCA_OPTIONS);
e0636822	512	RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
05f1cc01	513	parms = RTA_NEST(skb, TCA_GRED_PARMS);
1da177e4	514
05f1cc01 TG	515	for (i = 0; i < MAX_DPs; i++) {
	516	struct gred_sched_data *q = table->tab[i];
	517	struct tc_gred_qopt opt;
1da177e4	518
05f1cc01	519	memset(&opt, 0, sizeof(opt));
1da177e4 LT	520
	521	if (!q) {
	522	/* hack -- fix at some point with proper message
	523	This is how we indicate to tc that there is no VQ
	524	at this DP */
	525
05f1cc01 TG	526	opt.DP = MAX_DPs + i;
05f1cc01 TG	527	goto append_opt;
1da177e4 LT	528	}
1da177e4 LT	529
05f1cc01 TG	530	opt.limit = q->limit;
	531	opt.DP = q->DP;
	532	opt.backlog = q->backlog;
	533	opt.prio = q->prio;
22b33429 TG	534	opt.qth_min = q->parms.qth_min >> q->parms.Wlog;
	535	opt.qth_max = q->parms.qth_max >> q->parms.Wlog;
	536	opt.Wlog = q->parms.Wlog;
	537	opt.Plog = q->parms.Plog;
	538	opt.Scell_log = q->parms.Scell_log;
	539	opt.other = q->stats.other;
	540	opt.early = q->stats.prob_drop;
	541	opt.forced = q->stats.forced_drop;
	542	opt.pdrop = q->stats.pdrop;
05f1cc01 TG	543	opt.packets = q->packetsin;
	544	opt.bytesin = q->bytesin;
	545
22b33429 TG	546	if (gred_wred_mode(table)) {
	547	q->parms.qidlestart =
	548	table->tab[table->def]->parms.qidlestart;
	549	q->parms.qavg = table->tab[table->def]->parms.qavg;
1da177e4	550	}
05f1cc01	551
22b33429 TG	552	opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
22b33429 TG	553
05f1cc01 TG	554	append_opt:
05f1cc01 TG	555	RTA_APPEND(skb, sizeof(opt), &opt);
1da177e4 LT	556	}
1da177e4 LT	557
05f1cc01	558	RTA_NEST_END(skb, parms);
1da177e4	559
05f1cc01	560	return RTA_NEST_END(skb, opts);
1da177e4 LT	561
1da177e4 LT	562	rtattr_failure:
05f1cc01	563	return RTA_NEST_CANCEL(skb, opts);
1da177e4 LT	564	}
	565
	566	static void gred_destroy(struct Qdisc *sch)
	567	{
	568	struct gred_sched *table = qdisc_priv(sch);
	569	int i;
	570
1e4dfaf9	571	for (i = 0; i < table->DPs; i++) {
1da177e4	572	if (table->tab[i])
6639607e	573	gred_destroy_vq(table->tab[i]);
1da177e4 LT	574	}
	575	}
	576
	577	static struct Qdisc_ops gred_qdisc_ops = {
1da177e4 LT	578	.id = "gred",
	579	.priv_size = sizeof(struct gred_sched),
	580	.enqueue = gred_enqueue,
	581	.dequeue = gred_dequeue,
	582	.requeue = gred_requeue,
	583	.drop = gred_drop,
	584	.init = gred_init,
	585	.reset = gred_reset,
	586	.destroy = gred_destroy,
	587	.change = gred_change,
	588	.dump = gred_dump,
	589	.owner = THIS_MODULE,
	590	};
	591
	592	static int __init gred_module_init(void)
	593	{
	594	return register_qdisc(&gred_qdisc_ops);
	595	}
1e4dfaf9 TG	596
1e4dfaf9 TG	597	static void __exit gred_module_exit(void)
1da177e4 LT	598	{
	599	unregister_qdisc(&gred_qdisc_ops);
	600	}
1e4dfaf9	601
1da177e4 LT	602	module_init(gred_module_init)
1da177e4 LT	603	module_exit(gred_module_exit)
1e4dfaf9	604
1da177e4	605	MODULE_LICENSE("GPL");