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

#define GRED_DEF_PRIO (MAX_DPs / 2)

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 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->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, stab, 256);

	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 (table->tab[table->def] == NULL) {
		if (gred_rio_mode(table))
			prio = table->tab[ctl->DP]->prio;

		err = gred_change_vq(sch, table->def, ctl, prio, stab);
		if (err < 0)
			goto errout_locked;
	}

	table->initd = 1;

	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),
	};

	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");