[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 sk_buff **to_free)
{
	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, to_free);

congestion_drop:
	qdisc_drop(skb, sch, to_free);
	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 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)
		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 (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog))
		return -EINVAL;

	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 netlink_ext_ack *extack)
{
	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, NULL);
	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 netlink_ext_ack *extack)
{
	struct nlattr *tb[TCA_GRED_MAX + 1];
	int err;

	if (!opt)
		return -EINVAL;

	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy, NULL);
	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,
	.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
520ac30f ED	152	static int gred_enqueue(struct sk_buff skb, struct Qdisc sch,
520ac30f ED	153	struct sk_buff **to_free)
1da177e4	154	{
cc7ec456 ED	155	struct gred_sched_data *q = NULL;
cc7ec456 ED	156	struct gred_sched *t = qdisc_priv(sch);
22b33429	157	unsigned long qavg = 0;
4a591834	158	u16 dp = tc_index_to_dp(skb);
1da177e4	159
cc7ec456	160	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
18e3fb84 TG	161	dp = t->def;
18e3fb84 TG	162
cc7ec456 ED	163	q = t->tab[dp];
cc7ec456 ED	164	if (!q) {
18e3fb84 TG	165	/* Pass through packets not assigned to a DP
	166	* if no default DP has been configured. This
	167	* allows for DP flows to be left untouched.
	168	*/
a3eb95f8 DW	169	if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <=
a3eb95f8 DW	170	sch->limit))
18e3fb84 TG	171	return qdisc_enqueue_tail(skb, sch);
	172	else
	173	goto drop;
1da177e4	174	}
18e3fb84	175
eeca6688	176	/* fix tc_index? --could be controversial but needed for
1da177e4	177	requeueing */
18e3fb84	178	skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) \| dp;
1da177e4 LT	179	}
1da177e4 LT	180
e29fe837	181	/* sum up all the qaves of prios < ours to get the new qave */
d6fd4e96	182	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
1e4dfaf9 TG	183	int i;
	184
	185	for (i = 0; i < t->DPs; i++) {
	186	if (t->tab[i] && t->tab[i]->prio < q->prio &&
eeca6688 ED	187	!red_is_idling(&t->tab[i]->vars))
eeca6688 ED	188	qavg += t->tab[i]->vars.qavg;
1da177e4	189	}
1e4dfaf9	190
1da177e4 LT	191	}
	192
	193	q->packetsin++;
0abf77e5	194	q->bytesin += qdisc_pkt_len(skb);
1da177e4	195
1e4dfaf9	196	if (gred_wred_mode(t))
7051703b	197	gred_load_wred_set(t, q);
1da177e4	198
eeca6688 ED	199	q->vars.qavg = red_calc_qavg(&q->parms,
	200	&q->vars,
	201	gred_backlog(t, q, sch));
1da177e4	202
eeca6688 ED	203	if (red_is_idling(&q->vars))
eeca6688 ED	204	red_end_of_idle_period(&q->vars);
1da177e4	205
dea3f628	206	if (gred_wred_mode(t))
7051703b	207	gred_store_wred_set(t, q);
1da177e4	208
eeca6688	209	switch (red_action(&q->parms, &q->vars, q->vars.qavg + qavg)) {
cc7ec456 ED	210	case RED_DONT_MARK:
	211	break;
	212
	213	case RED_PROB_MARK:
25331d6c	214	qdisc_qstats_overlimit(sch);
cc7ec456 ED	215	if (!gred_use_ecn(t) \|\| !INET_ECN_set_ce(skb)) {
	216	q->stats.prob_drop++;
	217	goto congestion_drop;
	218	}
	219
	220	q->stats.prob_mark++;
	221	break;
	222
	223	case RED_HARD_MARK:
25331d6c	224	qdisc_qstats_overlimit(sch);
cc7ec456 ED	225	if (gred_use_harddrop(t) \|\| !gred_use_ecn(t) \|\|
	226	!INET_ECN_set_ce(skb)) {
	227	q->stats.forced_drop++;
	228	goto congestion_drop;
	229	}
	230	q->stats.forced_mark++;
	231	break;
1da177e4	232	}
22b33429	233
145a42b3	234	if (gred_backlog(t, q, sch) + qdisc_pkt_len(skb) <= q->limit) {
0abf77e5	235	q->backlog += qdisc_pkt_len(skb);
edf7a7b1	236	return qdisc_enqueue_tail(skb, sch);
1da177e4	237	}
22b33429 TG	238
	239	q->stats.pdrop++;
	240	drop:
520ac30f	241	return qdisc_drop(skb, sch, to_free);
c3b553cd TG	242
c3b553cd TG	243	congestion_drop:
520ac30f	244	qdisc_drop(skb, sch, to_free);
c3b553cd	245	return NET_XMIT_CN;
1da177e4 LT	246	}
1da177e4 LT	247
cc7ec456	248	static struct sk_buff gred_dequeue(struct Qdisc sch)
1da177e4 LT	249	{
1da177e4 LT	250	struct sk_buff *skb;
1e4dfaf9	251	struct gred_sched *t = qdisc_priv(sch);
1da177e4	252
edf7a7b1 TG	253	skb = qdisc_dequeue_head(sch);
edf7a7b1 TG	254
1da177e4	255	if (skb) {
1e4dfaf9	256	struct gred_sched_data *q;
18e3fb84 TG	257	u16 dp = tc_index_to_dp(skb);
	258
	259	if (dp >= t->DPs \|\| (q = t->tab[dp]) == NULL) {
e87cc472 JP	260	net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x after dequeue, screwing up backlog\n",
e87cc472 JP	261	tc_index_to_dp(skb));
18e3fb84	262	} else {
0abf77e5	263	q->backlog -= qdisc_pkt_len(skb);
18e3fb84	264
ba1bf474 DW	265	if (gred_wred_mode(t)) {
	266	if (!sch->qstats.backlog)
	267	red_start_of_idle_period(&t->wred_set);
	268	} else {
	269	if (!q->backlog)
	270	red_start_of_idle_period(&q->vars);
	271	}
1da177e4	272	}
18e3fb84	273
1da177e4 LT	274	return skb;
	275	}
	276
1da177e4 LT	277	return NULL;
	278	}
	279
cc7ec456	280	static void gred_reset(struct Qdisc *sch)
1da177e4 LT	281	{
1da177e4 LT	282	int i;
1e4dfaf9	283	struct gred_sched *t = qdisc_priv(sch);
1da177e4	284
edf7a7b1	285	qdisc_reset_queue(sch);
1da177e4	286
10297b99	287	for (i = 0; i < t->DPs; i++) {
1e4dfaf9 TG	288	struct gred_sched_data *q = t->tab[i];
	289
	290	if (!q)
	291	continue;
	292
eeca6688	293	red_restart(&q->vars);
1da177e4	294	q->backlog = 0;
1da177e4 LT	295	}
	296	}
	297
6639607e TG	298	static inline void gred_destroy_vq(struct gred_sched_data *q)
	299	{
	300	kfree(q);
	301	}
	302
1e90474c	303	static inline int gred_change_table_def(struct Qdisc sch, struct nlattr dps)
1da177e4 LT	304	{
1da177e4 LT	305	struct gred_sched *table = qdisc_priv(sch);
1da177e4	306	struct tc_gred_sopt *sopt;
6639607e	307	int i;
1da177e4	308
ac8ef4ab	309	if (!dps)
1da177e4 LT	310	return -EINVAL;
1da177e4 LT	311
1e90474c	312	sopt = nla_data(dps);
1da177e4	313
ac8ef4ab AA	314	if (sopt->DPs > MAX_DPs \|\| sopt->DPs == 0 \|\|
ac8ef4ab AA	315	sopt->def_DP >= sopt->DPs)
6639607e	316	return -EINVAL;
1da177e4	317
6639607e TG	318	sch_tree_lock(sch);
	319	table->DPs = sopt->DPs;
	320	table->def = sopt->def_DP;
b38c7eef	321	table->red_flags = sopt->flags;
dea3f628	322
6639607e TG	323	/*
	324	* Every entry point to GRED is synchronized with the above code
	325	* and the DP is checked against DPs, i.e. shadowed VQs can no
	326	* longer be found so we can unlock right here.
	327	*/
	328	sch_tree_unlock(sch);
dea3f628	329
6639607e TG	330	if (sopt->grio) {
	331	gred_enable_rio_mode(table);
	332	gred_disable_wred_mode(table);
	333	if (gred_wred_mode_check(sch))
	334	gred_enable_wred_mode(table);
	335	} else {
	336	gred_disable_rio_mode(table);
	337	gred_disable_wred_mode(table);
	338	}
	339
	340	for (i = table->DPs; i < MAX_DPs; i++) {
	341	if (table->tab[i]) {
c17988a9 YY	342	pr_warn("GRED: Warning: Destroying shadowed VQ 0x%x\n",
c17988a9 YY	343	i);
6639607e TG	344	gred_destroy_vq(table->tab[i]);
6639607e TG	345	table->tab[i] = NULL;
10297b99	346	}
6639607e	347	}
1da177e4	348
6639607e TG	349	return 0;
	350	}
	351
f62d6b93	352	static inline int gred_change_vq(struct Qdisc *sch, int dp,
a73ed26b	353	struct tc_gred_qopt *ctl, int prio,
869aa410 ED	354	u8 *stab, u32 max_P,
869aa410 ED	355	struct gred_sched_data **prealloc)
6639607e TG	356	{
6639607e TG	357	struct gred_sched *table = qdisc_priv(sch);
869aa410	358	struct gred_sched_data *q = table->tab[dp];
f62d6b93	359
8afa10cb NF	360	if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog))
	361	return -EINVAL;
	362
869aa410 ED	363	if (!q) {
	364	table->tab[dp] = q = *prealloc;
	365	*prealloc = NULL;
	366	if (!q)
f62d6b93	367	return -ENOMEM;
f62d6b93 TG	368	}
f62d6b93 TG	369
f62d6b93 TG	370	q->DP = dp;
f62d6b93 TG	371	q->prio = prio;
a3eb95f8 DW	372	if (ctl->limit > sch->limit)
	373	q->limit = sch->limit;
	374	else
	375	q->limit = ctl->limit;
22b33429 TG	376
22b33429 TG	377	if (q->backlog == 0)
eeca6688	378	red_end_of_idle_period(&q->vars);
22b33429 TG	379
	380	red_set_parms(&q->parms,
	381	ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
a73ed26b	382	ctl->Scell_log, stab, max_P);
eeca6688	383	red_set_vars(&q->vars);
f62d6b93 TG	384	return 0;
	385	}
	386
27a3421e PM	387	static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
	388	[TCA_GRED_PARMS] = { .len = sizeof(struct tc_gred_qopt) },
	389	[TCA_GRED_STAB] = { .len = 256 },
	390	[TCA_GRED_DPS] = { .len = sizeof(struct tc_gred_sopt) },
a73ed26b	391	[TCA_GRED_MAX_P] = { .type = NLA_U32 },
a3eb95f8	392	[TCA_GRED_LIMIT] = { .type = NLA_U32 },
27a3421e PM	393	};
27a3421e PM	394
2030721c AA	395	static int gred_change(struct Qdisc sch, struct nlattr opt,
2030721c AA	396	struct netlink_ext_ack *extack)
f62d6b93 TG	397	{
f62d6b93 TG	398	struct gred_sched *table = qdisc_priv(sch);
6639607e	399	struct tc_gred_qopt *ctl;
1e90474c	400	struct nlattr *tb[TCA_GRED_MAX + 1];
cee63723	401	int err, prio = GRED_DEF_PRIO;
f62d6b93	402	u8 *stab;
a73ed26b	403	u32 max_P;
869aa410	404	struct gred_sched_data *prealloc;
6639607e	405
cee63723	406	if (opt == NULL)
6639607e TG	407	return -EINVAL;
6639607e TG	408
fceb6435	409	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy, NULL);
cee63723 PM	410	if (err < 0)
	411	return err;
	412
a3eb95f8 DW	413	if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL) {
	414	if (tb[TCA_GRED_LIMIT] != NULL)
	415	sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
f62d6b93	416	return gred_change_table_def(sch, opt);
a3eb95f8	417	}
1da177e4	418
1e90474c	419	if (tb[TCA_GRED_PARMS] == NULL \|\|
a3eb95f8 DW	420	tb[TCA_GRED_STAB] == NULL \|\|
a3eb95f8 DW	421	tb[TCA_GRED_LIMIT] != NULL)
f62d6b93	422	return -EINVAL;
1da177e4	423
a73ed26b ED	424	max_P = tb[TCA_GRED_MAX_P] ? nla_get_u32(tb[TCA_GRED_MAX_P]) : 0;
a73ed26b ED	425
cee63723	426	err = -EINVAL;
1e90474c PM	427	ctl = nla_data(tb[TCA_GRED_PARMS]);
1e90474c PM	428	stab = nla_data(tb[TCA_GRED_STAB]);
a8aaa995 TG	429
a8aaa995 TG	430	if (ctl->DP >= table->DPs)
f62d6b93	431	goto errout;
1da177e4	432
d6fd4e96	433	if (gred_rio_mode(table)) {
f62d6b93 TG	434	if (ctl->prio == 0) {
	435	int def_prio = GRED_DEF_PRIO;
	436
	437	if (table->tab[table->def])
	438	def_prio = table->tab[table->def]->prio;
	439
	440	printk(KERN_DEBUG "GRED: DP %u does not have a prio "
	441	"setting default to %d\n", ctl->DP, def_prio);
	442
	443	prio = def_prio;
	444	} else
	445	prio = ctl->prio;
1da177e4 LT	446	}
1da177e4 LT	447
869aa410	448	prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
f62d6b93	449	sch_tree_lock(sch);
1da177e4	450
869aa410	451	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab, max_P, &prealloc);
f62d6b93 TG	452	if (err < 0)
f62d6b93 TG	453	goto errout_locked;
1da177e4	454
d6fd4e96	455	if (gred_rio_mode(table)) {
dea3f628 TG	456	gred_disable_wred_mode(table);
	457	if (gred_wred_mode_check(sch))
	458	gred_enable_wred_mode(table);
1da177e4 LT	459	}
1da177e4 LT	460
f62d6b93	461	err = 0;
1da177e4	462
f62d6b93 TG	463	errout_locked:
f62d6b93 TG	464	sch_tree_unlock(sch);
869aa410	465	kfree(prealloc);
f62d6b93 TG	466	errout:
f62d6b93 TG	467	return err;
1da177e4 LT	468	}
1da177e4 LT	469
e63d7dfd AA	470	static int gred_init(struct Qdisc sch, struct nlattr opt,
e63d7dfd AA	471	struct netlink_ext_ack *extack)
1da177e4	472	{
1e90474c	473	struct nlattr *tb[TCA_GRED_MAX + 1];
cee63723	474	int err;
1da177e4	475
ac8ef4ab	476	if (!opt)
1da177e4 LT	477	return -EINVAL;
1da177e4 LT	478
fceb6435	479	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy, NULL);
cee63723 PM	480	if (err < 0)
	481	return err;
	482
1e90474c	483	if (tb[TCA_GRED_PARMS] \|\| tb[TCA_GRED_STAB])
6639607e	484	return -EINVAL;
1da177e4	485
a3eb95f8 DW	486	if (tb[TCA_GRED_LIMIT])
a3eb95f8 DW	487	sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
348e3435 PS	488	else
	489	sch->limit = qdisc_dev(sch)->tx_queue_len
	490	* psched_mtu(qdisc_dev(sch));
a3eb95f8	491
1e90474c	492	return gred_change_table_def(sch, tb[TCA_GRED_DPS]);
1da177e4 LT	493	}
	494
	495	static int gred_dump(struct Qdisc sch, struct sk_buff skb)
	496	{
1da177e4	497	struct gred_sched *table = qdisc_priv(sch);
1e90474c	498	struct nlattr parms, opts = NULL;
1da177e4	499	int i;
a73ed26b	500	u32 max_p[MAX_DPs];
e0636822 TG	501	struct tc_gred_sopt sopt = {
	502	.DPs = table->DPs,
	503	.def_DP = table->def,
	504	.grio = gred_rio_mode(table),
b38c7eef	505	.flags = table->red_flags,
e0636822	506	};
1da177e4	507
1e90474c PM	508	opts = nla_nest_start(skb, TCA_OPTIONS);
	509	if (opts == NULL)
	510	goto nla_put_failure;
1b34ec43 DM	511	if (nla_put(skb, TCA_GRED_DPS, sizeof(sopt), &sopt))
1b34ec43 DM	512	goto nla_put_failure;
a73ed26b ED	513
	514	for (i = 0; i < MAX_DPs; i++) {
	515	struct gred_sched_data *q = table->tab[i];
	516
	517	max_p[i] = q ? q->parms.max_P : 0;
	518	}
1b34ec43 DM	519	if (nla_put(skb, TCA_GRED_MAX_P, sizeof(max_p), max_p))
1b34ec43 DM	520	goto nla_put_failure;
a73ed26b	521
a3eb95f8 DW	522	if (nla_put_u32(skb, TCA_GRED_LIMIT, sch->limit))
	523	goto nla_put_failure;
	524
1e90474c PM	525	parms = nla_nest_start(skb, TCA_GRED_PARMS);
	526	if (parms == NULL)
	527	goto nla_put_failure;
1da177e4	528
05f1cc01 TG	529	for (i = 0; i < MAX_DPs; i++) {
	530	struct gred_sched_data *q = table->tab[i];
	531	struct tc_gred_qopt opt;
1fe37b10	532	unsigned long qavg;
1da177e4	533
05f1cc01	534	memset(&opt, 0, sizeof(opt));
1da177e4 LT	535
	536	if (!q) {
	537	/* hack -- fix at some point with proper message
	538	This is how we indicate to tc that there is no VQ
	539	at this DP */
	540
05f1cc01 TG	541	opt.DP = MAX_DPs + i;
05f1cc01 TG	542	goto append_opt;
1da177e4 LT	543	}
1da177e4 LT	544
05f1cc01 TG	545	opt.limit = q->limit;
05f1cc01 TG	546	opt.DP = q->DP;
145a42b3	547	opt.backlog = gred_backlog(table, q, sch);
05f1cc01	548	opt.prio = q->prio;
22b33429 TG	549	opt.qth_min = q->parms.qth_min >> q->parms.Wlog;
	550	opt.qth_max = q->parms.qth_max >> q->parms.Wlog;
	551	opt.Wlog = q->parms.Wlog;
	552	opt.Plog = q->parms.Plog;
	553	opt.Scell_log = q->parms.Scell_log;
	554	opt.other = q->stats.other;
	555	opt.early = q->stats.prob_drop;
	556	opt.forced = q->stats.forced_drop;
	557	opt.pdrop = q->stats.pdrop;
05f1cc01 TG	558	opt.packets = q->packetsin;
	559	opt.bytesin = q->bytesin;
	560
244b65db DW	561	if (gred_wred_mode(table))
244b65db DW	562	gred_load_wred_set(table, q);
05f1cc01	563
1fe37b10 DW	564	qavg = red_calc_qavg(&q->parms, &q->vars,
	565	q->vars.qavg >> q->parms.Wlog);
	566	opt.qave = qavg >> q->parms.Wlog;
22b33429	567
05f1cc01	568	append_opt:
1e90474c PM	569	if (nla_append(skb, sizeof(opt), &opt) < 0)
1e90474c PM	570	goto nla_put_failure;
1da177e4 LT	571	}
1da177e4 LT	572
1e90474c	573	nla_nest_end(skb, parms);
1da177e4	574
1e90474c	575	return nla_nest_end(skb, opts);
1da177e4	576
1e90474c	577	nla_put_failure:
bc3ed28c TG	578	nla_nest_cancel(skb, opts);
bc3ed28c TG	579	return -EMSGSIZE;
1da177e4 LT	580	}
	581
	582	static void gred_destroy(struct Qdisc *sch)
	583	{
	584	struct gred_sched *table = qdisc_priv(sch);
	585	int i;
	586
1e4dfaf9	587	for (i = 0; i < table->DPs; i++) {
1da177e4	588	if (table->tab[i])
6639607e	589	gred_destroy_vq(table->tab[i]);
1da177e4 LT	590	}
	591	}
	592
20fea08b	593	static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
1da177e4 LT	594	.id = "gred",
	595	.priv_size = sizeof(struct gred_sched),
	596	.enqueue = gred_enqueue,
	597	.dequeue = gred_dequeue,
8e3af978	598	.peek = qdisc_peek_head,
1da177e4 LT	599	.init = gred_init,
	600	.reset = gred_reset,
	601	.destroy = gred_destroy,
	602	.change = gred_change,
	603	.dump = gred_dump,
	604	.owner = THIS_MODULE,
	605	};
	606
	607	static int __init gred_module_init(void)
	608	{
	609	return register_qdisc(&gred_qdisc_ops);
	610	}
1e4dfaf9 TG	611
1e4dfaf9 TG	612	static void __exit gred_module_exit(void)
1da177e4 LT	613	{
	614	unregister_qdisc(&gred_qdisc_ops);
	615	}
1e4dfaf9	616
1da177e4 LT	617	module_init(gred_module_init)
1da177e4 LT	618	module_exit(gred_module_exit)
1e4dfaf9	619
1da177e4	620	MODULE_LICENSE("GPL");