[net-next-2.6.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 Alexey's sch_red.c
 */

#include <linux/config.h>
#include <linux/module.h>
#include <asm/uaccess.h>
#include <asm/system.h>
#include <linux/bitops.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/string.h>
#include <linux/mm.h>
#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/in.h>
#include <linux/errno.h>
#include <linux/interrupt.h>
#include <linux/if_ether.h>
#include <linux/inet.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/notifier.h>
#include <net/ip.h>
#include <net/route.h>
#include <linux/skbuff.h>
#include <net/sock.h>
#include <net/pkt_sched.h>

#if 1 /* control */
#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
#else
#define DPRINTK(format,args...)
#endif

#if 0 /* data */
#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
#else
#define D2PRINTK(format,args...)
#endif

struct gred_sched_data;
struct gred_sched;

struct gred_sched_data
{
/* Parameters */
	u32		limit;		/* HARD maximal queue length	*/
	u32		qth_min;	/* Min average length threshold: A scaled */
	u32		qth_max;	/* Max average length threshold: A scaled */
	u32      	DP;		/* the drop pramaters */
	char		Wlog;		/* log(W)		*/
	char		Plog;		/* random number bits	*/
	u32		Scell_max;
	u32		Rmask;
	u32		bytesin;	/* bytes seen on virtualQ so far*/
	u32		packetsin;	/* packets seen on virtualQ so far*/
	u32		backlog;	/* bytes on the virtualQ */
	u32		forced;	/* packets dropped for exceeding limits */
	u32		early;	/* packets dropped as a warning */
	u32		other;	/* packets dropped by invoking drop() */
	u32		pdrop;	/* packets dropped because we exceeded physical queue limits */
	char		Scell_log;
	u8		Stab[256];
	u8              prio;        /* the prio of this vq */

/* Variables */
	unsigned long	qave;		/* Average queue length: A scaled */
	int		qcount;		/* Packets since last random number generation */
	u32		qR;		/* Cached random number */

	psched_time_t	qidlestart;	/* Start of idle period	*/
};

enum {
	GRED_WRED_MODE = 1,
	GRED_RIO_MODE,
};

struct gred_sched
{
	struct gred_sched_data *tab[MAX_DPs];
	unsigned long	flags;
	u32 		DPs;   
	u32 		def; 
	u8 		initd; 
};

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 int
gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
{
	psched_time_t now;
	struct gred_sched_data *q=NULL;
	struct gred_sched *t= qdisc_priv(sch);
	unsigned long	qave=0;	
	int i=0;

	if (!t->initd && skb_queue_len(&sch->q) < (sch->dev->tx_queue_len ? : 1)) {
		D2PRINTK("NO GRED Queues setup yet! Enqueued anyway\n");
		goto do_enqueue;
	}


	if ( ((skb->tc_index&0xf) > (t->DPs -1)) || !(q=t->tab[skb->tc_index&0xf])) {
		printk("GRED: setting to default (%d)\n ",t->def);
		if (!(q=t->tab[t->def])) {
			DPRINTK("GRED: setting to default FAILED! dropping!! "
			    "(%d)\n ", t->def);
			goto drop;
		}
		/* fix tc_index? --could be controvesial but needed for
		   requeueing */
		skb->tc_index=(skb->tc_index&0xfffffff0) | t->def;
	}

	D2PRINTK("gred_enqueue virtualQ 0x%x classid %x backlog %d "
	    "general backlog %d\n",skb->tc_index&0xf,sch->handle,q->backlog,
	    sch->qstats.backlog);
	/* sum up all the qaves of prios <= to ours to get the new qave*/
	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
		for (i=0;i<t->DPs;i++) {
			if ((!t->tab[i]) || (i==q->DP))	
				continue; 
				
			if ((t->tab[i]->prio < q->prio) && (PSCHED_IS_PASTPERFECT(t->tab[i]->qidlestart)))
				qave +=t->tab[i]->qave;
		}
			
	}

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

	if (gred_wred_mode(t)) {
		qave=0;
		q->qave=t->tab[t->def]->qave;
		q->qidlestart=t->tab[t->def]->qidlestart;
	}

	if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
		long us_idle;
		PSCHED_GET_TIME(now);
		us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
		PSCHED_SET_PASTPERFECT(q->qidlestart);

		q->qave >>= q->Stab[(us_idle>>q->Scell_log)&0xFF];
	} else {
		if (gred_wred_mode(t)) {
			q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
		} else {
			q->qave += q->backlog - (q->qave >> q->Wlog);
		}

	}
	

	if (gred_wred_mode(t))
		t->tab[t->def]->qave=q->qave;

	if ((q->qave+qave) < q->qth_min) {
		q->qcount = -1;
enqueue:
		if (q->backlog + skb->len <= q->limit) {
			q->backlog += skb->len;
do_enqueue:
			__skb_queue_tail(&sch->q, skb);
			sch->qstats.backlog += skb->len;
			sch->bstats.bytes += skb->len;
			sch->bstats.packets++;
			return 0;
		} else {
			q->pdrop++;
		}

drop:
		kfree_skb(skb);
		sch->qstats.drops++;
		return NET_XMIT_DROP;
	}
	if ((q->qave+qave) >= q->qth_max) {
		q->qcount = -1;
		sch->qstats.overlimits++;
		q->forced++;
		goto drop;
	}
	if (++q->qcount) {
		if ((((qave+q->qave) - q->qth_min)>>q->Wlog)*q->qcount < q->qR)
			goto enqueue;
		q->qcount = 0;
		q->qR = net_random()&q->Rmask;
		sch->qstats.overlimits++;
		q->early++;
		goto drop;
	}
	q->qR = net_random()&q->Rmask;
	goto enqueue;
}

static int
gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
{
	struct gred_sched_data *q;
	struct gred_sched *t= qdisc_priv(sch);
	q= t->tab[(skb->tc_index&0xf)];
/* error checking here -- probably unnecessary */
	PSCHED_SET_PASTPERFECT(q->qidlestart);

	__skb_queue_head(&sch->q, skb);
	sch->qstats.backlog += skb->len;
	sch->qstats.requeues++;
	q->backlog += skb->len;
	return 0;
}

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

	skb = __skb_dequeue(&sch->q);
	if (skb) {
		sch->qstats.backlog -= skb->len;
		q= t->tab[(skb->tc_index&0xf)];
		if (q) {
			q->backlog -= skb->len;
			if (!q->backlog && !gred_wred_mode(t))
				PSCHED_GET_TIME(q->qidlestart);
		} else {
			D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf); 
		}
		return skb;
	}

	if (gred_wred_mode(t)) {
			q= t->tab[t->def];
			if (!q)	
				D2PRINTK("no default VQ set: Results will be "
				       "screwed up\n");
			else
				PSCHED_GET_TIME(q->qidlestart);
	}

	return NULL;
}

static unsigned int gred_drop(struct Qdisc* sch)
{
	struct sk_buff *skb;

	struct gred_sched_data *q;
	struct gred_sched *t= qdisc_priv(sch);

	skb = __skb_dequeue_tail(&sch->q);
	if (skb) {
		unsigned int len = skb->len;
		sch->qstats.backlog -= len;
		sch->qstats.drops++;
		q= t->tab[(skb->tc_index&0xf)];
		if (q) {
			q->backlog -= len;
			q->other++;
			if (!q->backlog && !gred_wred_mode(t))
				PSCHED_GET_TIME(q->qidlestart);
		} else {
			D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf); 
		}

		kfree_skb(skb);
		return len;
	}

	q=t->tab[t->def];
	if (!q) {
		D2PRINTK("no default VQ set: Results might be screwed up\n");
		return 0;
	}

	PSCHED_GET_TIME(q->qidlestart);
	return 0;

}

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

	__skb_queue_purge(&sch->q);

	sch->qstats.backlog = 0;

        for (i=0;i<t->DPs;i++) {
	        q= t->tab[i];
		if (!q)	
			continue; 
		PSCHED_SET_PASTPERFECT(q->qidlestart);
		q->qave = 0;
		q->qcount = -1;
		q->backlog = 0;
		q->other=0;
		q->forced=0;
		q->pdrop=0;
		q->early=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;

	/*
	 * 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;
  		}
	}

	table->initd = 0;

	return 0;
}

static int gred_change(struct Qdisc *sch, struct rtattr *opt)
{
	struct gred_sched *table = qdisc_priv(sch);
	struct gred_sched_data *q;
	struct tc_gred_qopt *ctl;
	struct rtattr *tb[TCA_GRED_STAB];

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

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

	if (!table->DPs || tb[TCA_GRED_PARMS-1] == 0 || tb[TCA_GRED_STAB-1] == 0 ||
		RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
		RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
			return -EINVAL;

	ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
	if (ctl->DP > MAX_DPs-1 ) {
		/* misbehaving is punished! Put in the default drop probability */
		DPRINTK("\nGRED: DP %u not in  the proper range fixed. New DP "
			"set to default at %d\n",ctl->DP,table->def);
		ctl->DP=table->def;
	}
	
	if (table->tab[ctl->DP] == NULL) {
		table->tab[ctl->DP]=kmalloc(sizeof(struct gred_sched_data),
					    GFP_KERNEL);
		if (NULL == table->tab[ctl->DP])
			return -ENOMEM;
		memset(table->tab[ctl->DP], 0, (sizeof(struct gred_sched_data)));
	}
	q= table->tab[ctl->DP]; 

	if (gred_rio_mode(table)) {
		if (ctl->prio <=0) {
			if (table->def && table->tab[table->def]) {
				DPRINTK("\nGRED: DP %u does not have a prio"
					"setting default to %d\n",ctl->DP,
					table->tab[table->def]->prio);
				q->prio=table->tab[table->def]->prio;
			} else { 
				DPRINTK("\nGRED: DP %u does not have a prio"
					" setting default to 8\n",ctl->DP);
				q->prio=8;
			}
		} else {
			q->prio=ctl->prio;
		}
	} else {
		q->prio=8;
	}


	q->DP=ctl->DP;
	q->Wlog = ctl->Wlog;
	q->Plog = ctl->Plog;
	q->limit = ctl->limit;
	q->Scell_log = ctl->Scell_log;
	q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
	q->Scell_max = (255<<q->Scell_log);
	q->qth_min = ctl->qth_min<<ctl->Wlog;
	q->qth_max = ctl->qth_max<<ctl->Wlog;
	q->qave=0;
	q->backlog=0;
	q->qcount = -1;
	q->other=0;
	q->forced=0;
	q->pdrop=0;
	q->early=0;

	PSCHED_SET_PASTPERFECT(q->qidlestart);
	memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);

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

	if (!table->initd) {
		table->initd=1;
		/* 
        	the first entry also goes into the default until
        	over-written 
		*/

		if (table->tab[table->def] == NULL) {
			table->tab[table->def]=
				kmalloc(sizeof(struct gred_sched_data), GFP_KERNEL);
			if (NULL == table->tab[table->def])
				return -ENOMEM;

			memset(table->tab[table->def], 0,
			       (sizeof(struct gred_sched_data)));
		}
		q= table->tab[table->def]; 
		q->DP=table->def;
		q->Wlog = ctl->Wlog;
		q->Plog = ctl->Plog;
		q->limit = ctl->limit;
		q->Scell_log = ctl->Scell_log;
		q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
		q->Scell_max = (255<<q->Scell_log);
		q->qth_min = ctl->qth_min<<ctl->Wlog;
		q->qth_max = ctl->qth_max<<ctl->Wlog;

		if (gred_rio_mode(table))
			q->prio=table->tab[ctl->DP]->prio;
		else
			q->prio=8;

		q->qcount = -1;
		PSCHED_SET_PASTPERFECT(q->qidlestart);
		memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
	}
	return 0;

}

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),
	};

	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->qth_min >> q->Wlog;
		opt.qth_max	= q->qth_max >> q->Wlog;
		opt.Wlog	= q->Wlog;
		opt.Plog	= q->Plog;
		opt.Scell_log	= q->Scell_log;
		opt.other	= q->other;
		opt.early	= q->early;
		opt.forced	= q->forced;
		opt.pdrop	= q->pdrop;
		opt.packets	= q->packetsin;
		opt.bytesin	= q->bytesin;

		if (q->qave) {
			if (gred_wred_mode(table)) {
				q->qidlestart=table->tab[table->def]->qidlestart;
				q->qave=table->tab[table->def]->qave;
			}
			if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
				long idle;
				unsigned long qave;
				psched_time_t now;
				PSCHED_GET_TIME(now);
				idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
				qave  = q->qave >> q->Stab[(idle>>q->Scell_log)&0xFF];
				opt.qave = qave >> q->Wlog;

			} else {
				opt.qave = q->qave >> q->Wlog;
			}
		}

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 = {
	.next		=	NULL,
	.cl_ops		=	NULL,
	.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	*
	18	*
	19	*
	20	* For all the glorious comments look at Alexey's sch_red.c
	21	*/
	22
	23	#include <linux/config.h>
	24	#include <linux/module.h>
	25	#include <asm/uaccess.h>
	26	#include <asm/system.h>
	27	#include <linux/bitops.h>
	28	#include <linux/types.h>
	29	#include <linux/kernel.h>
	30	#include <linux/sched.h>
	31	#include <linux/string.h>
	32	#include <linux/mm.h>
	33	#include <linux/socket.h>
	34	#include <linux/sockios.h>
	35	#include <linux/in.h>
	36	#include <linux/errno.h>
	37	#include <linux/interrupt.h>
	38	#include <linux/if_ether.h>
	39	#include <linux/inet.h>
	40	#include <linux/netdevice.h>
	41	#include <linux/etherdevice.h>
	42	#include <linux/notifier.h>
	43	#include <net/ip.h>
	44	#include <net/route.h>
	45	#include <linux/skbuff.h>
	46	#include <net/sock.h>
	47	#include <net/pkt_sched.h>
	48
	49	#if 1 /* control */
	50	#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
	51	#else
	52	#define DPRINTK(format,args...)
	53	#endif
	54
	55	#if 0 /* data */
	56	#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
	57	#else
	58	#define D2PRINTK(format,args...)
	59	#endif
	60
	61	struct gred_sched_data;
	62	struct gred_sched;
	63
	64	struct gred_sched_data
65	{
66	/* Parameters */
67	u32 limit; /* HARD maximal queue length */
68	u32 qth_min; /* Min average length threshold: A scaled */
69	u32 qth_max; /* Max average length threshold: A scaled */
70	u32 DP; /* the drop pramaters */
71	char Wlog; /* log(W) */
72	char Plog; /* random number bits */
73	u32 Scell_max;
74	u32 Rmask;
75	u32 bytesin; /* bytes seen on virtualQ so far*/
76	u32 packetsin; /* packets seen on virtualQ so far*/
77	u32 backlog; /* bytes on the virtualQ */
78	u32 forced; /* packets dropped for exceeding limits */
79	u32 early; /* packets dropped as a warning */
80	u32 other; /* packets dropped by invoking drop() */
81	u32 pdrop; /* packets dropped because we exceeded physical queue limits */
82	char Scell_log;
83	u8 Stab[256];
84	u8 prio; /* the prio of this vq */
85
86	/* Variables */
87	unsigned long qave; /* Average queue length: A scaled */
88	int qcount; /* Packets since last random number generation */
89	u32 qR; /* Cached random number */
90
91	psched_time_t qidlestart; /* Start of idle period */
92	};
93
dea3f628 TG	94	enum {
dea3f628 TG	95	GRED_WRED_MODE = 1,
d6fd4e96	96	GRED_RIO_MODE,
dea3f628 TG	97	};
dea3f628 TG	98
1da177e4 LT	99	struct gred_sched
	100	{
	101	struct gred_sched_data *tab[MAX_DPs];
dea3f628	102	unsigned long flags;
1da177e4 LT	103	u32 DPs;
	104	u32 def;
	105	u8 initd;
1da177e4 LT	106	};
1da177e4 LT	107
dea3f628 TG	108	static inline int gred_wred_mode(struct gred_sched *table)
	109	{
	110	return test_bit(GRED_WRED_MODE, &table->flags);
	111	}
	112
	113	static inline void gred_enable_wred_mode(struct gred_sched *table)
	114	{
	115	__set_bit(GRED_WRED_MODE, &table->flags);
	116	}
	117
	118	static inline void gred_disable_wred_mode(struct gred_sched *table)
	119	{
	120	__clear_bit(GRED_WRED_MODE, &table->flags);
	121	}
	122
d6fd4e96 TG	123	static inline int gred_rio_mode(struct gred_sched *table)
	124	{
	125	return test_bit(GRED_RIO_MODE, &table->flags);
	126	}
	127
	128	static inline void gred_enable_rio_mode(struct gred_sched *table)
	129	{
	130	__set_bit(GRED_RIO_MODE, &table->flags);
	131	}
	132
	133	static inline void gred_disable_rio_mode(struct gred_sched *table)
	134	{
	135	__clear_bit(GRED_RIO_MODE, &table->flags);
	136	}
	137
dea3f628 TG	138	static inline int gred_wred_mode_check(struct Qdisc *sch)
	139	{
	140	struct gred_sched *table = qdisc_priv(sch);
	141	int i;
	142
	143	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
	144	for (i = 0; i < table->DPs; i++) {
	145	struct gred_sched_data *q = table->tab[i];
	146	int n;
	147
	148	if (q == NULL)
	149	continue;
	150
	151	for (n = 0; n < table->DPs; n++)
	152	if (table->tab[n] && table->tab[n] != q &&
	153	table->tab[n]->prio == q->prio)
	154	return 1;
	155	}
	156
	157	return 0;
	158	}
	159
1da177e4 LT	160	static int
	161	gred_enqueue(struct sk_buff skb, struct Qdisc sch)
	162	{
	163	psched_time_t now;
	164	struct gred_sched_data *q=NULL;
	165	struct gred_sched *t= qdisc_priv(sch);
	166	unsigned long qave=0;
	167	int i=0;
	168
	169	if (!t->initd && skb_queue_len(&sch->q) < (sch->dev->tx_queue_len ? : 1)) {
	170	D2PRINTK("NO GRED Queues setup yet! Enqueued anyway\n");
	171	goto do_enqueue;
	172	}
	173
	174
	175	if ( ((skb->tc_index&0xf) > (t->DPs -1)) \|\| !(q=t->tab[skb->tc_index&0xf])) {
	176	printk("GRED: setting to default (%d)\n ",t->def);
	177	if (!(q=t->tab[t->def])) {
	178	DPRINTK("GRED: setting to default FAILED! dropping!! "
	179	"(%d)\n ", t->def);
	180	goto drop;
	181	}
	182	/* fix tc_index? --could be controvesial but needed for
	183	requeueing */
	184	skb->tc_index=(skb->tc_index&0xfffffff0) \| t->def;
	185	}
	186
	187	D2PRINTK("gred_enqueue virtualQ 0x%x classid %x backlog %d "
	188	"general backlog %d\n",skb->tc_index&0xf,sch->handle,q->backlog,
	189	sch->qstats.backlog);
	190	/* sum up all the qaves of prios <= to ours to get the new qave*/
d6fd4e96	191	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
1da177e4 LT	192	for (i=0;i<t->DPs;i++) {
	193	if ((!t->tab[i]) \|\| (i==q->DP))
	194	continue;
	195
	196	if ((t->tab[i]->prio < q->prio) && (PSCHED_IS_PASTPERFECT(t->tab[i]->qidlestart)))
	197	qave +=t->tab[i]->qave;
	198	}
	199
	200	}
	201
	202	q->packetsin++;
	203	q->bytesin+=skb->len;
	204
dea3f628	205	if (gred_wred_mode(t)) {
1da177e4 LT	206	qave=0;
	207	q->qave=t->tab[t->def]->qave;
	208	q->qidlestart=t->tab[t->def]->qidlestart;
	209	}
	210
	211	if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
	212	long us_idle;
	213	PSCHED_GET_TIME(now);
	214	us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
	215	PSCHED_SET_PASTPERFECT(q->qidlestart);
	216
	217	q->qave >>= q->Stab[(us_idle>>q->Scell_log)&0xFF];
	218	} else {
dea3f628	219	if (gred_wred_mode(t)) {
1da177e4 LT	220	q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
	221	} else {
	222	q->qave += q->backlog - (q->qave >> q->Wlog);
	223	}
	224
	225	}
	226
	227
dea3f628	228	if (gred_wred_mode(t))
1da177e4 LT	229	t->tab[t->def]->qave=q->qave;
	230
	231	if ((q->qave+qave) < q->qth_min) {
	232	q->qcount = -1;
	233	enqueue:
	234	if (q->backlog + skb->len <= q->limit) {
	235	q->backlog += skb->len;
	236	do_enqueue:
	237	__skb_queue_tail(&sch->q, skb);
	238	sch->qstats.backlog += skb->len;
	239	sch->bstats.bytes += skb->len;
	240	sch->bstats.packets++;
	241	return 0;
	242	} else {
	243	q->pdrop++;
	244	}
	245
	246	drop:
	247	kfree_skb(skb);
	248	sch->qstats.drops++;
	249	return NET_XMIT_DROP;
	250	}
	251	if ((q->qave+qave) >= q->qth_max) {
	252	q->qcount = -1;
	253	sch->qstats.overlimits++;
	254	q->forced++;
	255	goto drop;
	256	}
	257	if (++q->qcount) {
	258	if ((((qave+q->qave) - q->qth_min)>>q->Wlog)*q->qcount < q->qR)
	259	goto enqueue;
	260	q->qcount = 0;
	261	q->qR = net_random()&q->Rmask;
	262	sch->qstats.overlimits++;
	263	q->early++;
	264	goto drop;
	265	}
	266	q->qR = net_random()&q->Rmask;
	267	goto enqueue;
	268	}
	269
	270	static int
	271	gred_requeue(struct sk_buff skb, struct Qdisc sch)
	272	{
	273	struct gred_sched_data *q;
	274	struct gred_sched *t= qdisc_priv(sch);
	275	q= t->tab[(skb->tc_index&0xf)];
	276	/* error checking here -- probably unnecessary */
	277	PSCHED_SET_PASTPERFECT(q->qidlestart);
	278
	279	__skb_queue_head(&sch->q, skb);
	280	sch->qstats.backlog += skb->len;
	281	sch->qstats.requeues++;
	282	q->backlog += skb->len;
	283	return 0;
	284	}
	285
	286	static struct sk_buff *
	287	gred_dequeue(struct Qdisc* sch)
	288	{
	289	struct sk_buff *skb;
	290	struct gred_sched_data *q;
	291	struct gred_sched *t= qdisc_priv(sch);
	292
293	skb = __skb_dequeue(&sch->q);
294	if (skb) {
295	sch->qstats.backlog -= skb->len;
296	q= t->tab[(skb->tc_index&0xf)];
297	if (q) {
298	q->backlog -= skb->len;
dea3f628	299	if (!q->backlog && !gred_wred_mode(t))
1da177e4 LT	300	PSCHED_GET_TIME(q->qidlestart);
	301	} else {
	302	D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf);
	303	}
	304	return skb;
	305	}
	306
dea3f628	307	if (gred_wred_mode(t)) {
1da177e4 LT	308	q= t->tab[t->def];
	309	if (!q)
	310	D2PRINTK("no default VQ set: Results will be "
	311	"screwed up\n");
	312	else
	313	PSCHED_GET_TIME(q->qidlestart);
	314	}
	315
	316	return NULL;
	317	}
	318
	319	static unsigned int gred_drop(struct Qdisc* sch)
	320	{
	321	struct sk_buff *skb;
	322
	323	struct gred_sched_data *q;
	324	struct gred_sched *t= qdisc_priv(sch);
	325
	326	skb = __skb_dequeue_tail(&sch->q);
	327	if (skb) {
	328	unsigned int len = skb->len;
	329	sch->qstats.backlog -= len;
	330	sch->qstats.drops++;
	331	q= t->tab[(skb->tc_index&0xf)];
	332	if (q) {
	333	q->backlog -= len;
	334	q->other++;
dea3f628	335	if (!q->backlog && !gred_wred_mode(t))
1da177e4 LT	336	PSCHED_GET_TIME(q->qidlestart);
	337	} else {
	338	D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf);
	339	}
	340
	341	kfree_skb(skb);
	342	return len;
	343	}
	344
	345	q=t->tab[t->def];
	346	if (!q) {
	347	D2PRINTK("no default VQ set: Results might be screwed up\n");
	348	return 0;
	349	}
	350
	351	PSCHED_GET_TIME(q->qidlestart);
	352	return 0;
	353
	354	}
	355
	356	static void gred_reset(struct Qdisc* sch)
	357	{
	358	int i;
	359	struct gred_sched_data *q;
	360	struct gred_sched *t= qdisc_priv(sch);
	361
	362	__skb_queue_purge(&sch->q);
	363
	364	sch->qstats.backlog = 0;
	365
	366	for (i=0;i<t->DPs;i++) {
	367	q= t->tab[i];
	368	if (!q)
	369	continue;
	370	PSCHED_SET_PASTPERFECT(q->qidlestart);
	371	q->qave = 0;
	372	q->qcount = -1;
	373	q->backlog = 0;
	374	q->other=0;
	375	q->forced=0;
	376	q->pdrop=0;
	377	q->early=0;
	378	}
	379	}
	380
6639607e TG	381	static inline void gred_destroy_vq(struct gred_sched_data *q)
	382	{
	383	kfree(q);
	384	}
	385
	386	static inline int gred_change_table_def(struct Qdisc sch, struct rtattr dps)
1da177e4 LT	387	{
1da177e4 LT	388	struct gred_sched *table = qdisc_priv(sch);
1da177e4	389	struct tc_gred_sopt *sopt;
6639607e	390	int i;
1da177e4	391
6639607e	392	if (dps == NULL \|\| RTA_PAYLOAD(dps) < sizeof(*sopt))
1da177e4 LT	393	return -EINVAL;
1da177e4 LT	394
6639607e	395	sopt = RTA_DATA(dps);
1da177e4	396
6639607e TG	397	if (sopt->DPs > MAX_DPs \|\| sopt->DPs == 0 \|\| sopt->def_DP >= sopt->DPs)
6639607e TG	398	return -EINVAL;
1da177e4	399
6639607e TG	400	sch_tree_lock(sch);
	401	table->DPs = sopt->DPs;
	402	table->def = sopt->def_DP;
dea3f628	403
6639607e TG	404	/*
	405	* Every entry point to GRED is synchronized with the above code
	406	* and the DP is checked against DPs, i.e. shadowed VQs can no
	407	* longer be found so we can unlock right here.
	408	*/
	409	sch_tree_unlock(sch);
dea3f628	410
6639607e TG	411	if (sopt->grio) {
	412	gred_enable_rio_mode(table);
	413	gred_disable_wred_mode(table);
	414	if (gred_wred_mode_check(sch))
	415	gred_enable_wred_mode(table);
	416	} else {
	417	gred_disable_rio_mode(table);
	418	gred_disable_wred_mode(table);
	419	}
	420
	421	for (i = table->DPs; i < MAX_DPs; i++) {
	422	if (table->tab[i]) {
	423	printk(KERN_WARNING "GRED: Warning: Destroying "
	424	"shadowed VQ 0x%x\n", i);
	425	gred_destroy_vq(table->tab[i]);
	426	table->tab[i] = NULL;
	427	}
	428	}
1da177e4	429
6639607e TG	430	table->initd = 0;
	431
	432	return 0;
	433	}
	434
	435	static int gred_change(struct Qdisc sch, struct rtattr opt)
	436	{
	437	struct gred_sched *table = qdisc_priv(sch);
	438	struct gred_sched_data *q;
	439	struct tc_gred_qopt *ctl;
	440	struct rtattr *tb[TCA_GRED_STAB];
	441
	442	if (opt == NULL \|\| rtattr_parse_nested(tb, TCA_GRED_STAB, opt))
	443	return -EINVAL;
	444
	445	if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
	446	return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
1da177e4 LT	447
	448	if (!table->DPs \|\| tb[TCA_GRED_PARMS-1] == 0 \|\| tb[TCA_GRED_STAB-1] == 0 \|\|
	449	RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) \|\|
	450	RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
	451	return -EINVAL;
	452
	453	ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
	454	if (ctl->DP > MAX_DPs-1 ) {
	455	/* misbehaving is punished! Put in the default drop probability */
	456	DPRINTK("\nGRED: DP %u not in the proper range fixed. New DP "
	457	"set to default at %d\n",ctl->DP,table->def);
	458	ctl->DP=table->def;
	459	}
	460
	461	if (table->tab[ctl->DP] == NULL) {
	462	table->tab[ctl->DP]=kmalloc(sizeof(struct gred_sched_data),
	463	GFP_KERNEL);
	464	if (NULL == table->tab[ctl->DP])
	465	return -ENOMEM;
	466	memset(table->tab[ctl->DP], 0, (sizeof(struct gred_sched_data)));
	467	}
	468	q= table->tab[ctl->DP];
	469
d6fd4e96	470	if (gred_rio_mode(table)) {
1da177e4 LT	471	if (ctl->prio <=0) {
	472	if (table->def && table->tab[table->def]) {
	473	DPRINTK("\nGRED: DP %u does not have a prio"
	474	"setting default to %d\n",ctl->DP,
	475	table->tab[table->def]->prio);
	476	q->prio=table->tab[table->def]->prio;
	477	} else {
	478	DPRINTK("\nGRED: DP %u does not have a prio"
	479	" setting default to 8\n",ctl->DP);
	480	q->prio=8;
	481	}
	482	} else {
	483	q->prio=ctl->prio;
	484	}
	485	} else {
	486	q->prio=8;
	487	}
	488
	489
	490	q->DP=ctl->DP;
	491	q->Wlog = ctl->Wlog;
	492	q->Plog = ctl->Plog;
	493	q->limit = ctl->limit;
	494	q->Scell_log = ctl->Scell_log;
	495	q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
	496	q->Scell_max = (255<<q->Scell_log);
	497	q->qth_min = ctl->qth_min<<ctl->Wlog;
	498	q->qth_max = ctl->qth_max<<ctl->Wlog;
	499	q->qave=0;
	500	q->backlog=0;
	501	q->qcount = -1;
	502	q->other=0;
	503	q->forced=0;
	504	q->pdrop=0;
	505	q->early=0;
	506
	507	PSCHED_SET_PASTPERFECT(q->qidlestart);
	508	memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
	509
d6fd4e96	510	if (gred_rio_mode(table)) {
dea3f628 TG	511	gred_disable_wred_mode(table);
	512	if (gred_wred_mode_check(sch))
	513	gred_enable_wred_mode(table);
1da177e4 LT	514	}
	515
	516	if (!table->initd) {
	517	table->initd=1;
	518	/*
	519	the first entry also goes into the default until
	520	over-written
	521	*/
	522
	523	if (table->tab[table->def] == NULL) {
	524	table->tab[table->def]=
	525	kmalloc(sizeof(struct gred_sched_data), GFP_KERNEL);
	526	if (NULL == table->tab[table->def])
	527	return -ENOMEM;
	528
	529	memset(table->tab[table->def], 0,
	530	(sizeof(struct gred_sched_data)));
	531	}
	532	q= table->tab[table->def];
	533	q->DP=table->def;
	534	q->Wlog = ctl->Wlog;
	535	q->Plog = ctl->Plog;
	536	q->limit = ctl->limit;
	537	q->Scell_log = ctl->Scell_log;
	538	q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
	539	q->Scell_max = (255<<q->Scell_log);
	540	q->qth_min = ctl->qth_min<<ctl->Wlog;
	541	q->qth_max = ctl->qth_max<<ctl->Wlog;
	542
d6fd4e96	543	if (gred_rio_mode(table))
1da177e4 LT	544	q->prio=table->tab[ctl->DP]->prio;
	545	else
	546	q->prio=8;
	547
	548	q->qcount = -1;
	549	PSCHED_SET_PASTPERFECT(q->qidlestart);
	550	memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
	551	}
	552	return 0;
	553
	554	}
	555
	556	static int gred_init(struct Qdisc sch, struct rtattr opt)
	557	{
6639607e	558	struct rtattr *tb[TCA_GRED_MAX];
1da177e4	559
6639607e	560	if (opt == NULL \|\| rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
1da177e4 LT	561	return -EINVAL;
1da177e4 LT	562
6639607e TG	563	if (tb[TCA_GRED_PARMS-1] \|\| tb[TCA_GRED_STAB-1])
6639607e TG	564	return -EINVAL;
1da177e4	565
6639607e	566	return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
1da177e4 LT	567	}
	568
	569	static int gred_dump(struct Qdisc sch, struct sk_buff skb)
	570	{
1da177e4	571	struct gred_sched *table = qdisc_priv(sch);
05f1cc01	572	struct rtattr parms, opts = NULL;
1da177e4	573	int i;
e0636822 TG	574	struct tc_gred_sopt sopt = {
	575	.DPs = table->DPs,
	576	.def_DP = table->def,
	577	.grio = gred_rio_mode(table),
	578	};
1da177e4	579
05f1cc01	580	opts = RTA_NEST(skb, TCA_OPTIONS);
e0636822	581	RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
05f1cc01	582	parms = RTA_NEST(skb, TCA_GRED_PARMS);
1da177e4	583
05f1cc01 TG	584	for (i = 0; i < MAX_DPs; i++) {
	585	struct gred_sched_data *q = table->tab[i];
	586	struct tc_gred_qopt opt;
1da177e4	587
05f1cc01	588	memset(&opt, 0, sizeof(opt));
1da177e4 LT	589
	590	if (!q) {
	591	/* hack -- fix at some point with proper message
	592	This is how we indicate to tc that there is no VQ
	593	at this DP */
	594
05f1cc01 TG	595	opt.DP = MAX_DPs + i;
05f1cc01 TG	596	goto append_opt;
1da177e4 LT	597	}
1da177e4 LT	598
05f1cc01 TG	599	opt.limit = q->limit;
	600	opt.DP = q->DP;
	601	opt.backlog = q->backlog;
	602	opt.prio = q->prio;
	603	opt.qth_min = q->qth_min >> q->Wlog;
	604	opt.qth_max = q->qth_max >> q->Wlog;
	605	opt.Wlog = q->Wlog;
	606	opt.Plog = q->Plog;
	607	opt.Scell_log = q->Scell_log;
	608	opt.other = q->other;
	609	opt.early = q->early;
	610	opt.forced = q->forced;
	611	opt.pdrop = q->pdrop;
	612	opt.packets = q->packetsin;
	613	opt.bytesin = q->bytesin;
	614
1da177e4	615	if (q->qave) {
dea3f628	616	if (gred_wred_mode(table)) {
1da177e4 LT	617	q->qidlestart=table->tab[table->def]->qidlestart;
	618	q->qave=table->tab[table->def]->qave;
	619	}
	620	if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
	621	long idle;
05f1cc01	622	unsigned long qave;
1da177e4 LT	623	psched_time_t now;
	624	PSCHED_GET_TIME(now);
	625	idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
	626	qave = q->qave >> q->Stab[(idle>>q->Scell_log)&0xFF];
05f1cc01	627	opt.qave = qave >> q->Wlog;
1da177e4 LT	628
1da177e4 LT	629	} else {
05f1cc01	630	opt.qave = q->qave >> q->Wlog;
1da177e4	631	}
1da177e4	632	}
05f1cc01 TG	633
	634	append_opt:
	635	RTA_APPEND(skb, sizeof(opt), &opt);
1da177e4 LT	636	}
1da177e4 LT	637
05f1cc01	638	RTA_NEST_END(skb, parms);
1da177e4	639
05f1cc01	640	return RTA_NEST_END(skb, opts);
1da177e4 LT	641
1da177e4 LT	642	rtattr_failure:
05f1cc01	643	return RTA_NEST_CANCEL(skb, opts);
1da177e4 LT	644	}
	645
	646	static void gred_destroy(struct Qdisc *sch)
	647	{
	648	struct gred_sched *table = qdisc_priv(sch);
	649	int i;
	650
	651	for (i = 0;i < table->DPs; i++) {
	652	if (table->tab[i])
6639607e	653	gred_destroy_vq(table->tab[i]);
1da177e4 LT	654	}
	655	}
	656
	657	static struct Qdisc_ops gred_qdisc_ops = {
	658	.next = NULL,
	659	.cl_ops = NULL,
	660	.id = "gred",
	661	.priv_size = sizeof(struct gred_sched),
	662	.enqueue = gred_enqueue,
	663	.dequeue = gred_dequeue,
	664	.requeue = gred_requeue,
	665	.drop = gred_drop,
	666	.init = gred_init,
	667	.reset = gred_reset,
	668	.destroy = gred_destroy,
	669	.change = gred_change,
	670	.dump = gred_dump,
	671	.owner = THIS_MODULE,
	672	};
	673
	674	static int __init gred_module_init(void)
	675	{
	676	return register_qdisc(&gred_qdisc_ops);
	677	}
	678	static void __exit gred_module_exit(void)
	679	{
	680	unregister_qdisc(&gred_qdisc_ops);
	681	}
	682	module_init(gred_module_init)
	683	module_exit(gred_module_exit)
	684	MODULE_LICENSE("GPL");