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

/* net/sched/sch_teql.c	"True" (or "trivial") link equalizer.
 *
 *		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:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/if_arp.h>
#include <linux/netdevice.h>
#include <linux/init.h>
#include <linux/skbuff.h>
#include <linux/moduleparam.h>
#include <net/dst.h>
#include <net/neighbour.h>
#include <net/pkt_sched.h>

/*
   How to setup it.
   ----------------

   After loading this module you will find a new device teqlN
   and new qdisc with the same name. To join a slave to the equalizer
   you should just set this qdisc on a device f.e.

   # tc qdisc add dev eth0 root teql0
   # tc qdisc add dev eth1 root teql0

   That's all. Full PnP 8)

   Applicability.
   --------------

   1. Slave devices MUST be active devices, i.e., they must raise the tbusy
      signal and generate EOI events. If you want to equalize virtual devices
      like tunnels, use a normal eql device.
   2. This device puts no limitations on physical slave characteristics
      f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
      Certainly, large difference in link speeds will make the resulting
      eqalized link unusable, because of huge packet reordering.
      I estimate an upper useful difference as ~10 times.
   3. If the slave requires address resolution, only protocols using
      neighbour cache (IPv4/IPv6) will work over the equalized link.
      Other protocols are still allowed to use the slave device directly,
      which will not break load balancing, though native slave
      traffic will have the highest priority.  */

struct teql_master
{
	struct Qdisc_ops qops;
	struct net_device *dev;
	struct Qdisc *slaves;
	struct list_head master_list;
	struct net_device_stats stats;
};

struct teql_sched_data
{
	struct Qdisc *next;
	struct teql_master *m;
	struct neighbour *ncache;
	struct sk_buff_head q;
};

#define NEXT_SLAVE(q) (((struct teql_sched_data*)qdisc_priv(q))->next)

#define FMASK (IFF_BROADCAST|IFF_POINTOPOINT)

/* "teql*" qdisc routines */

static int
teql_enqueue(struct sk_buff *skb, struct Qdisc* sch)
{
	struct net_device *dev = qdisc_dev(sch);
	struct teql_sched_data *q = qdisc_priv(sch);

	if (q->q.qlen < dev->tx_queue_len) {
		__skb_queue_tail(&q->q, skb);
		sch->bstats.bytes += qdisc_pkt_len(skb);
		sch->bstats.packets++;
		return 0;
	}

	kfree_skb(skb);
	sch->qstats.drops++;
	return NET_XMIT_DROP;
}

static struct sk_buff *
teql_dequeue(struct Qdisc* sch)
{
	struct teql_sched_data *dat = qdisc_priv(sch);
	struct netdev_queue *dat_queue;
	struct sk_buff *skb;

	skb = __skb_dequeue(&dat->q);
	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
	if (skb == NULL) {
		struct net_device *m = qdisc_dev(dat_queue->qdisc);
		if (m) {
			dat->m->slaves = sch;
			netif_wake_queue(m);
		}
	}
	sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
	return skb;
}

static struct sk_buff *
teql_peek(struct Qdisc* sch)
{
	/* teql is meant to be used as root qdisc */
	return NULL;
}

static __inline__ void
teql_neigh_release(struct neighbour *n)
{
	if (n)
		neigh_release(n);
}

static void
teql_reset(struct Qdisc* sch)
{
	struct teql_sched_data *dat = qdisc_priv(sch);

	skb_queue_purge(&dat->q);
	sch->q.qlen = 0;
	teql_neigh_release(xchg(&dat->ncache, NULL));
}

static void
teql_destroy(struct Qdisc* sch)
{
	struct Qdisc *q, *prev;
	struct teql_sched_data *dat = qdisc_priv(sch);
	struct teql_master *master = dat->m;

	if ((prev = master->slaves) != NULL) {
		do {
			q = NEXT_SLAVE(prev);
			if (q == sch) {
				NEXT_SLAVE(prev) = NEXT_SLAVE(q);
				if (q == master->slaves) {
					master->slaves = NEXT_SLAVE(q);
					if (q == master->slaves) {
						struct netdev_queue *txq;
						spinlock_t *root_lock;

						txq = netdev_get_tx_queue(master->dev, 0);
						master->slaves = NULL;

						root_lock = qdisc_root_sleeping_lock(txq->qdisc);
						spin_lock_bh(root_lock);
						qdisc_reset(txq->qdisc);
						spin_unlock_bh(root_lock);
					}
				}
				skb_queue_purge(&dat->q);
				teql_neigh_release(xchg(&dat->ncache, NULL));
				break;
			}

		} while ((prev = q) != master->slaves);
	}
}

static int teql_qdisc_init(struct Qdisc *sch, struct nlattr *opt)
{
	struct net_device *dev = qdisc_dev(sch);
	struct teql_master *m = (struct teql_master*)sch->ops;
	struct teql_sched_data *q = qdisc_priv(sch);

	if (dev->hard_header_len > m->dev->hard_header_len)
		return -EINVAL;

	if (m->dev == dev)
		return -ELOOP;

	q->m = m;

	skb_queue_head_init(&q->q);

	if (m->slaves) {
		if (m->dev->flags & IFF_UP) {
			if ((m->dev->flags&IFF_POINTOPOINT && !(dev->flags&IFF_POINTOPOINT))
			    || (m->dev->flags&IFF_BROADCAST && !(dev->flags&IFF_BROADCAST))
			    || (m->dev->flags&IFF_MULTICAST && !(dev->flags&IFF_MULTICAST))
			    || dev->mtu < m->dev->mtu)
				return -EINVAL;
		} else {
			if (!(dev->flags&IFF_POINTOPOINT))
				m->dev->flags &= ~IFF_POINTOPOINT;
			if (!(dev->flags&IFF_BROADCAST))
				m->dev->flags &= ~IFF_BROADCAST;
			if (!(dev->flags&IFF_MULTICAST))
				m->dev->flags &= ~IFF_MULTICAST;
			if (dev->mtu < m->dev->mtu)
				m->dev->mtu = dev->mtu;
		}
		q->next = NEXT_SLAVE(m->slaves);
		NEXT_SLAVE(m->slaves) = sch;
	} else {
		q->next = sch;
		m->slaves = sch;
		m->dev->mtu = dev->mtu;
		m->dev->flags = (m->dev->flags&~FMASK)|(dev->flags&FMASK);
	}
	return 0;
}


static int
__teql_resolve(struct sk_buff *skb, struct sk_buff *skb_res, struct net_device *dev)
{
	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
	struct teql_sched_data *q = qdisc_priv(dev_queue->qdisc);
	struct neighbour *mn = skb->dst->neighbour;
	struct neighbour *n = q->ncache;

	if (mn->tbl == NULL)
		return -EINVAL;
	if (n && n->tbl == mn->tbl &&
	    memcmp(n->primary_key, mn->primary_key, mn->tbl->key_len) == 0) {
		atomic_inc(&n->refcnt);
	} else {
		n = __neigh_lookup_errno(mn->tbl, mn->primary_key, dev);
		if (IS_ERR(n))
			return PTR_ERR(n);
	}
	if (neigh_event_send(n, skb_res) == 0) {
		int err;

		read_lock(&n->lock);
		err = dev_hard_header(skb, dev, ntohs(skb->protocol),
				      n->ha, NULL, skb->len);
		read_unlock(&n->lock);

		if (err < 0) {
			neigh_release(n);
			return -EINVAL;
		}
		teql_neigh_release(xchg(&q->ncache, n));
		return 0;
	}
	neigh_release(n);
	return (skb_res == NULL) ? -EAGAIN : 1;
}

static inline int teql_resolve(struct sk_buff *skb,
			       struct sk_buff *skb_res, struct net_device *dev)
{
	struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
	if (txq->qdisc == &noop_qdisc)
		return -ENODEV;

	if (dev->header_ops == NULL ||
	    skb->dst == NULL ||
	    skb->dst->neighbour == NULL)
		return 0;
	return __teql_resolve(skb, skb_res, dev);
}

static int teql_master_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct teql_master *master = netdev_priv(dev);
	struct Qdisc *start, *q;
	int busy;
	int nores;
	int subq = skb_get_queue_mapping(skb);
	struct sk_buff *skb_res = NULL;

	start = master->slaves;

restart:
	nores = 0;
	busy = 0;

	if ((q = start) == NULL)
		goto drop;

	do {
		struct net_device *slave = qdisc_dev(q);
		struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
		const struct net_device_ops *slave_ops = slave->netdev_ops;

		if (slave_txq->qdisc_sleeping != q)
			continue;
		if (__netif_subqueue_stopped(slave, subq) ||
		    !netif_running(slave)) {
			busy = 1;
			continue;
		}

		switch (teql_resolve(skb, skb_res, slave)) {
		case 0:
			if (__netif_tx_trylock(slave_txq)) {
				unsigned int length = qdisc_pkt_len(skb);

				if (!netif_tx_queue_stopped(slave_txq) &&
				    !netif_tx_queue_frozen(slave_txq) &&
				    slave_ops->ndo_start_xmit(skb, slave) == 0) {
					__netif_tx_unlock(slave_txq);
					master->slaves = NEXT_SLAVE(q);
					netif_wake_queue(dev);
					master->stats.tx_packets++;
					master->stats.tx_bytes += length;
					return 0;
				}
				__netif_tx_unlock(slave_txq);
			}
			if (netif_queue_stopped(dev))
				busy = 1;
			break;
		case 1:
			master->slaves = NEXT_SLAVE(q);
			return 0;
		default:
			nores = 1;
			break;
		}
		__skb_pull(skb, skb_network_offset(skb));
	} while ((q = NEXT_SLAVE(q)) != start);

	if (nores && skb_res == NULL) {
		skb_res = skb;
		goto restart;
	}

	if (busy) {
		netif_stop_queue(dev);
		return 1;
	}
	master->stats.tx_errors++;

drop:
	master->stats.tx_dropped++;
	dev_kfree_skb(skb);
	return 0;
}

static int teql_master_open(struct net_device *dev)
{
	struct Qdisc * q;
	struct teql_master *m = netdev_priv(dev);
	int mtu = 0xFFFE;
	unsigned flags = IFF_NOARP|IFF_MULTICAST;

	if (m->slaves == NULL)
		return -EUNATCH;

	flags = FMASK;

	q = m->slaves;
	do {
		struct net_device *slave = qdisc_dev(q);

		if (slave == NULL)
			return -EUNATCH;

		if (slave->mtu < mtu)
			mtu = slave->mtu;
		if (slave->hard_header_len > LL_MAX_HEADER)
			return -EINVAL;

		/* If all the slaves are BROADCAST, master is BROADCAST
		   If all the slaves are PtP, master is PtP
		   Otherwise, master is NBMA.
		 */
		if (!(slave->flags&IFF_POINTOPOINT))
			flags &= ~IFF_POINTOPOINT;
		if (!(slave->flags&IFF_BROADCAST))
			flags &= ~IFF_BROADCAST;
		if (!(slave->flags&IFF_MULTICAST))
			flags &= ~IFF_MULTICAST;
	} while ((q = NEXT_SLAVE(q)) != m->slaves);

	m->dev->mtu = mtu;
	m->dev->flags = (m->dev->flags&~FMASK) | flags;
	netif_start_queue(m->dev);
	return 0;
}

static int teql_master_close(struct net_device *dev)
{
	netif_stop_queue(dev);
	return 0;
}

static struct net_device_stats *teql_master_stats(struct net_device *dev)
{
	struct teql_master *m = netdev_priv(dev);
	return &m->stats;
}

static int teql_master_mtu(struct net_device *dev, int new_mtu)
{
	struct teql_master *m = netdev_priv(dev);
	struct Qdisc *q;

	if (new_mtu < 68)
		return -EINVAL;

	q = m->slaves;
	if (q) {
		do {
			if (new_mtu > qdisc_dev(q)->mtu)
				return -EINVAL;
		} while ((q=NEXT_SLAVE(q)) != m->slaves);
	}

	dev->mtu = new_mtu;
	return 0;
}

static const struct net_device_ops teql_netdev_ops = {
	.ndo_open	= teql_master_open,
	.ndo_stop	= teql_master_close,
	.ndo_start_xmit	= teql_master_xmit,
	.ndo_get_stats	= teql_master_stats,
	.ndo_change_mtu	= teql_master_mtu,
};

static __init void teql_master_setup(struct net_device *dev)
{
	struct teql_master *master = netdev_priv(dev);
	struct Qdisc_ops *ops = &master->qops;

	master->dev	= dev;
	ops->priv_size  = sizeof(struct teql_sched_data);

	ops->enqueue	=	teql_enqueue;
	ops->dequeue	=	teql_dequeue;
	ops->peek	=	teql_peek;
	ops->init	=	teql_qdisc_init;
	ops->reset	=	teql_reset;
	ops->destroy	=	teql_destroy;
	ops->owner	=	THIS_MODULE;

	dev->netdev_ops =       &teql_netdev_ops;
	dev->type		= ARPHRD_VOID;
	dev->mtu		= 1500;
	dev->tx_queue_len	= 100;
	dev->flags		= IFF_NOARP;
	dev->hard_header_len	= LL_MAX_HEADER;
}

static LIST_HEAD(master_dev_list);
static int max_equalizers = 1;
module_param(max_equalizers, int, 0);
MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");

static int __init teql_init(void)
{
	int i;
	int err = -ENODEV;

	for (i = 0; i < max_equalizers; i++) {
		struct net_device *dev;
		struct teql_master *master;

		dev = alloc_netdev(sizeof(struct teql_master),
				  "teql%d", teql_master_setup);
		if (!dev) {
			err = -ENOMEM;
			break;
		}

		if ((err = register_netdev(dev))) {
			free_netdev(dev);
			break;
		}

		master = netdev_priv(dev);

		strlcpy(master->qops.id, dev->name, IFNAMSIZ);
		err = register_qdisc(&master->qops);

		if (err) {
			unregister_netdev(dev);
			free_netdev(dev);
			break;
		}

		list_add_tail(&master->master_list, &master_dev_list);
	}
	return i ? 0 : err;
}

static void __exit teql_exit(void)
{
	struct teql_master *master, *nxt;

	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {

		list_del(&master->master_list);

		unregister_qdisc(&master->qops);
		unregister_netdev(master->dev);
		free_netdev(master->dev);
	}
}

module_init(teql_init);
module_exit(teql_exit);

MODULE_LICENSE("GPL");
Commit	Line	Data
1da177e4 LT	1	/* net/sched/sch_teql.c "True" (or "trivial") link equalizer.
	2	*
	3	* This program is free software; you can redistribute it and/or
	4	* modify it under the terms of the GNU General Public License
	5	* as published by the Free Software Foundation; either version
	6	* 2 of the License, or (at your option) any later version.
	7	*
	8	* Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
	9	*/
	10
	11	#include <linux/module.h>
1da177e4 LT	12	#include <linux/types.h>
1da177e4 LT	13	#include <linux/kernel.h>
1da177e4	14	#include <linux/string.h>
1da177e4	15	#include <linux/errno.h>
14c85021	16	#include <linux/if_arp.h>
1da177e4	17	#include <linux/netdevice.h>
1da177e4	18	#include <linux/init.h>
1da177e4 LT	19	#include <linux/skbuff.h>
1da177e4 LT	20	#include <linux/moduleparam.h>
0ba48053 PM	21	#include <net/dst.h>
0ba48053 PM	22	#include <net/neighbour.h>
1da177e4 LT	23	#include <net/pkt_sched.h>
	24
	25	/*
	26	How to setup it.
	27	----------------
	28
	29	After loading this module you will find a new device teqlN
	30	and new qdisc with the same name. To join a slave to the equalizer
	31	you should just set this qdisc on a device f.e.
	32
	33	# tc qdisc add dev eth0 root teql0
	34	# tc qdisc add dev eth1 root teql0
	35
	36	That's all. Full PnP 8)
	37
	38	Applicability.
	39	--------------
	40
	41	1. Slave devices MUST be active devices, i.e., they must raise the tbusy
	42	signal and generate EOI events. If you want to equalize virtual devices
	43	like tunnels, use a normal eql device.
	44	2. This device puts no limitations on physical slave characteristics
	45	f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
	46	Certainly, large difference in link speeds will make the resulting
	47	eqalized link unusable, because of huge packet reordering.
	48	I estimate an upper useful difference as ~10 times.
	49	3. If the slave requires address resolution, only protocols using
	50	neighbour cache (IPv4/IPv6) will work over the equalized link.
	51	Other protocols are still allowed to use the slave device directly,
	52	which will not break load balancing, though native slave
	53	traffic will have the highest priority. */
	54
	55	struct teql_master
	56	{
	57	struct Qdisc_ops qops;
	58	struct net_device *dev;
	59	struct Qdisc *slaves;
	60	struct list_head master_list;
	61	struct net_device_stats stats;
	62	};
	63
	64	struct teql_sched_data
	65	{
	66	struct Qdisc *next;
	67	struct teql_master *m;
	68	struct neighbour *ncache;
	69	struct sk_buff_head q;
	70	};
	71
	72	#define NEXT_SLAVE(q) (((struct teql_sched_data*)qdisc_priv(q))->next)
	73
cc8fd14d	74	#define FMASK (IFF_BROADCAST\|IFF_POINTOPOINT)
1da177e4 LT	75
	76	/* "teql" qdisc routines /
	77
	78	static int
	79	teql_enqueue(struct sk_buff skb, struct Qdisc sch)
	80	{
5ce2d488	81	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	82	struct teql_sched_data *q = qdisc_priv(sch);
1da177e4 LT	83
4cd8c9e8 KK	84	if (q->q.qlen < dev->tx_queue_len) {
4cd8c9e8 KK	85	__skb_queue_tail(&q->q, skb);
0abf77e5	86	sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4 LT	87	sch->bstats.packets++;
	88	return 0;
	89	}
	90
1da177e4 LT	91	kfree_skb(skb);
	92	sch->qstats.drops++;
	93	return NET_XMIT_DROP;
	94	}
	95
1da177e4 LT	96	static struct sk_buff *
	97	teql_dequeue(struct Qdisc* sch)
	98	{
	99	struct teql_sched_data *dat = qdisc_priv(sch);
b0e1e646	100	struct netdev_queue *dat_queue;
1da177e4 LT	101	struct sk_buff *skb;
	102
	103	skb = __skb_dequeue(&dat->q);
e8a0464c	104	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
1da177e4	105	if (skb == NULL) {
b0e1e646	106	struct net_device *m = qdisc_dev(dat_queue->qdisc);
1da177e4 LT	107	if (m) {
	108	dat->m->slaves = sch;
	109	netif_wake_queue(m);
	110	}
	111	}
b0e1e646	112	sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
1da177e4 LT	113	return skb;
	114	}
	115
8e3af978 JP	116	static struct sk_buff *
	117	teql_peek(struct Qdisc* sch)
	118	{
	119	/* teql is meant to be used as root qdisc */
	120	return NULL;
	121	}
	122
1da177e4 LT	123	static __inline__ void
	124	teql_neigh_release(struct neighbour *n)
	125	{
	126	if (n)
	127	neigh_release(n);
	128	}
	129
	130	static void
	131	teql_reset(struct Qdisc* sch)
	132	{
	133	struct teql_sched_data *dat = qdisc_priv(sch);
	134
	135	skb_queue_purge(&dat->q);
	136	sch->q.qlen = 0;
	137	teql_neigh_release(xchg(&dat->ncache, NULL));
	138	}
	139
	140	static void
	141	teql_destroy(struct Qdisc* sch)
	142	{
	143	struct Qdisc q, prev;
	144	struct teql_sched_data *dat = qdisc_priv(sch);
	145	struct teql_master *master = dat->m;
	146
	147	if ((prev = master->slaves) != NULL) {
	148	do {
	149	q = NEXT_SLAVE(prev);
	150	if (q == sch) {
	151	NEXT_SLAVE(prev) = NEXT_SLAVE(q);
	152	if (q == master->slaves) {
	153	master->slaves = NEXT_SLAVE(q);
	154	if (q == master->slaves) {
e8a0464c	155	struct netdev_queue *txq;
83874000	156	spinlock_t *root_lock;
e8a0464c DM	157
e8a0464c DM	158	txq = netdev_get_tx_queue(master->dev, 0);
1da177e4	159	master->slaves = NULL;
83874000	160
102396ae	161	root_lock = qdisc_root_sleeping_lock(txq->qdisc);
83874000	162	spin_lock_bh(root_lock);
e8a0464c	163	qdisc_reset(txq->qdisc);
83874000	164	spin_unlock_bh(root_lock);
1da177e4 LT	165	}
	166	}
	167	skb_queue_purge(&dat->q);
	168	teql_neigh_release(xchg(&dat->ncache, NULL));
	169	break;
	170	}
10297b99	171
1da177e4 LT	172	} while ((prev = q) != master->slaves);
	173	}
	174	}
	175
1e90474c	176	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt)
1da177e4	177	{
5ce2d488	178	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	179	struct teql_master m = (struct teql_master)sch->ops;
	180	struct teql_sched_data *q = qdisc_priv(sch);
	181
	182	if (dev->hard_header_len > m->dev->hard_header_len)
	183	return -EINVAL;
	184
	185	if (m->dev == dev)
	186	return -ELOOP;
	187
	188	q->m = m;
	189
	190	skb_queue_head_init(&q->q);
	191
	192	if (m->slaves) {
	193	if (m->dev->flags & IFF_UP) {
	194	if ((m->dev->flags&IFF_POINTOPOINT && !(dev->flags&IFF_POINTOPOINT))
	195	\|\| (m->dev->flags&IFF_BROADCAST && !(dev->flags&IFF_BROADCAST))
	196	\|\| (m->dev->flags&IFF_MULTICAST && !(dev->flags&IFF_MULTICAST))
	197	\|\| dev->mtu < m->dev->mtu)
	198	return -EINVAL;
	199	} else {
	200	if (!(dev->flags&IFF_POINTOPOINT))
	201	m->dev->flags &= ~IFF_POINTOPOINT;
	202	if (!(dev->flags&IFF_BROADCAST))
	203	m->dev->flags &= ~IFF_BROADCAST;
	204	if (!(dev->flags&IFF_MULTICAST))
	205	m->dev->flags &= ~IFF_MULTICAST;
	206	if (dev->mtu < m->dev->mtu)
	207	m->dev->mtu = dev->mtu;
	208	}
	209	q->next = NEXT_SLAVE(m->slaves);
	210	NEXT_SLAVE(m->slaves) = sch;
	211	} else {
	212	q->next = sch;
	213	m->slaves = sch;
	214	m->dev->mtu = dev->mtu;
	215	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	216	}
	217	return 0;
	218	}
	219
1da177e4 LT	220
	221	static int
	222	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res, struct net_device *dev)
	223	{
e8a0464c DM	224	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
e8a0464c DM	225	struct teql_sched_data *q = qdisc_priv(dev_queue->qdisc);
1da177e4 LT	226	struct neighbour *mn = skb->dst->neighbour;
	227	struct neighbour *n = q->ncache;
	228
	229	if (mn->tbl == NULL)
	230	return -EINVAL;
	231	if (n && n->tbl == mn->tbl &&
	232	memcmp(n->primary_key, mn->primary_key, mn->tbl->key_len) == 0) {
	233	atomic_inc(&n->refcnt);
	234	} else {
	235	n = __neigh_lookup_errno(mn->tbl, mn->primary_key, dev);
	236	if (IS_ERR(n))
	237	return PTR_ERR(n);
	238	}
	239	if (neigh_event_send(n, skb_res) == 0) {
	240	int err;
0c4e8581	241
1da177e4	242	read_lock(&n->lock);
0c4e8581 SH	243	err = dev_hard_header(skb, dev, ntohs(skb->protocol),
0c4e8581 SH	244	n->ha, NULL, skb->len);
1da177e4	245	read_unlock(&n->lock);
0c4e8581	246
1da177e4 LT	247	if (err < 0) {
	248	neigh_release(n);
	249	return -EINVAL;
	250	}
	251	teql_neigh_release(xchg(&q->ncache, n));
	252	return 0;
	253	}
	254	neigh_release(n);
	255	return (skb_res == NULL) ? -EAGAIN : 1;
	256	}
	257
3b04ddde SH	258	static inline int teql_resolve(struct sk_buff *skb,
3b04ddde SH	259	struct sk_buff skb_res, struct net_device dev)
1da177e4	260	{
e8a0464c DM	261	struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
e8a0464c DM	262	if (txq->qdisc == &noop_qdisc)
4f9f8311 EP	263	return -ENODEV;
4f9f8311 EP	264
3b04ddde	265	if (dev->header_ops == NULL \|\|
1da177e4 LT	266	skb->dst == NULL \|\|
	267	skb->dst->neighbour == NULL)
	268	return 0;
	269	return __teql_resolve(skb, skb_res, dev);
	270	}
	271
	272	static int teql_master_xmit(struct sk_buff skb, struct net_device dev)
	273	{
2941a486	274	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	275	struct Qdisc start, q;
	276	int busy;
	277	int nores;
4e3ab47a	278	int subq = skb_get_queue_mapping(skb);
1da177e4 LT	279	struct sk_buff *skb_res = NULL;
	280
	281	start = master->slaves;
	282
	283	restart:
	284	nores = 0;
	285	busy = 0;
	286
	287	if ((q = start) == NULL)
	288	goto drop;
	289
	290	do {
5ce2d488	291	struct net_device *slave = qdisc_dev(q);
61294e2e SH	292	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
61294e2e SH	293	const struct net_device_ops *slave_ops = slave->netdev_ops;
10297b99	294
e8a0464c	295	if (slave_txq->qdisc_sleeping != q)
1da177e4	296	continue;
fd2ea0a7	297	if (__netif_subqueue_stopped(slave, subq) \|\|
f25f4e44	298	!netif_running(slave)) {
1da177e4 LT	299	busy = 1;
	300	continue;
	301	}
	302
	303	switch (teql_resolve(skb, skb_res, slave)) {
	304	case 0:
c3f26a26	305	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	306	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	307
c3f26a26 DM	308	if (!netif_tx_queue_stopped(slave_txq) &&
c3f26a26 DM	309	!netif_tx_queue_frozen(slave_txq) &&
61294e2e	310	slave_ops->ndo_start_xmit(skb, slave) == 0) {
c3f26a26	311	__netif_tx_unlock(slave_txq);
1da177e4 LT	312	master->slaves = NEXT_SLAVE(q);
	313	netif_wake_queue(dev);
	314	master->stats.tx_packets++;
c0f84d0d	315	master->stats.tx_bytes += length;
1da177e4 LT	316	return 0;
1da177e4 LT	317	}
c3f26a26	318	__netif_tx_unlock(slave_txq);
1da177e4 LT	319	}
	320	if (netif_queue_stopped(dev))
	321	busy = 1;
	322	break;
	323	case 1:
	324	master->slaves = NEXT_SLAVE(q);
	325	return 0;
	326	default:
	327	nores = 1;
	328	break;
	329	}
bbe735e4	330	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	331	} while ((q = NEXT_SLAVE(q)) != start);
	332
	333	if (nores && skb_res == NULL) {
	334	skb_res = skb;
	335	goto restart;
	336	}
	337
	338	if (busy) {
	339	netif_stop_queue(dev);
	340	return 1;
	341	}
	342	master->stats.tx_errors++;
	343
	344	drop:
	345	master->stats.tx_dropped++;
	346	dev_kfree_skb(skb);
	347	return 0;
	348	}
	349
	350	static int teql_master_open(struct net_device *dev)
	351	{
	352	struct Qdisc * q;
2941a486	353	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	354	int mtu = 0xFFFE;
	355	unsigned flags = IFF_NOARP\|IFF_MULTICAST;
	356
	357	if (m->slaves == NULL)
	358	return -EUNATCH;
	359
	360	flags = FMASK;
	361
	362	q = m->slaves;
	363	do {
5ce2d488	364	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	365
	366	if (slave == NULL)
	367	return -EUNATCH;
	368
	369	if (slave->mtu < mtu)
	370	mtu = slave->mtu;
	371	if (slave->hard_header_len > LL_MAX_HEADER)
	372	return -EINVAL;
	373
	374	/* If all the slaves are BROADCAST, master is BROADCAST
	375	If all the slaves are PtP, master is PtP
	376	Otherwise, master is NBMA.
	377	*/
	378	if (!(slave->flags&IFF_POINTOPOINT))
	379	flags &= ~IFF_POINTOPOINT;
	380	if (!(slave->flags&IFF_BROADCAST))
	381	flags &= ~IFF_BROADCAST;
	382	if (!(slave->flags&IFF_MULTICAST))
	383	flags &= ~IFF_MULTICAST;
	384	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	385
	386	m->dev->mtu = mtu;
	387	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	388	netif_start_queue(m->dev);
	389	return 0;
	390	}
	391
	392	static int teql_master_close(struct net_device *dev)
	393	{
	394	netif_stop_queue(dev);
	395	return 0;
	396	}
	397
	398	static struct net_device_stats teql_master_stats(struct net_device dev)
	399	{
2941a486	400	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	401	return &m->stats;
	402	}
	403
	404	static int teql_master_mtu(struct net_device *dev, int new_mtu)
	405	{
2941a486	406	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	407	struct Qdisc *q;
	408
	409	if (new_mtu < 68)
	410	return -EINVAL;
	411
	412	q = m->slaves;
	413	if (q) {
	414	do {
5ce2d488	415	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4 LT	416	return -EINVAL;
	417	} while ((q=NEXT_SLAVE(q)) != m->slaves);
	418	}
	419
	420	dev->mtu = new_mtu;
	421	return 0;
	422	}
	423
61294e2e SH	424	static const struct net_device_ops teql_netdev_ops = {
	425	.ndo_open = teql_master_open,
	426	.ndo_stop = teql_master_close,
	427	.ndo_start_xmit = teql_master_xmit,
	428	.ndo_get_stats = teql_master_stats,
	429	.ndo_change_mtu = teql_master_mtu,
	430	};
	431
1da177e4 LT	432	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	433	{
2941a486	434	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	435	struct Qdisc_ops *ops = &master->qops;
	436
	437	master->dev = dev;
	438	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	439
1da177e4 LT	440	ops->enqueue = teql_enqueue;
1da177e4 LT	441	ops->dequeue = teql_dequeue;
8e3af978	442	ops->peek = teql_peek;
1da177e4 LT	443	ops->init = teql_qdisc_init;
	444	ops->reset = teql_reset;
	445	ops->destroy = teql_destroy;
	446	ops->owner = THIS_MODULE;
	447
61294e2e	448	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	449	dev->type = ARPHRD_VOID;
	450	dev->mtu = 1500;
	451	dev->tx_queue_len = 100;
	452	dev->flags = IFF_NOARP;
	453	dev->hard_header_len = LL_MAX_HEADER;
1da177e4 LT	454	}
	455
	456	static LIST_HEAD(master_dev_list);
	457	static int max_equalizers = 1;
	458	module_param(max_equalizers, int, 0);
	459	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	460
	461	static int __init teql_init(void)
	462	{
	463	int i;
	464	int err = -ENODEV;
	465
	466	for (i = 0; i < max_equalizers; i++) {
	467	struct net_device *dev;
	468	struct teql_master *master;
	469
	470	dev = alloc_netdev(sizeof(struct teql_master),
	471	"teql%d", teql_master_setup);
	472	if (!dev) {
	473	err = -ENOMEM;
	474	break;
	475	}
	476
	477	if ((err = register_netdev(dev))) {
	478	free_netdev(dev);
	479	break;
	480	}
	481
2941a486	482	master = netdev_priv(dev);
1da177e4 LT	483
	484	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	485	err = register_qdisc(&master->qops);
	486
	487	if (err) {
	488	unregister_netdev(dev);
	489	free_netdev(dev);
	490	break;
	491	}
	492
	493	list_add_tail(&master->master_list, &master_dev_list);
	494	}
	495	return i ? 0 : err;
	496	}
	497
10297b99	498	static void __exit teql_exit(void)
1da177e4 LT	499	{
	500	struct teql_master master, nxt;
	501
	502	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	503
	504	list_del(&master->master_list);
	505
	506	unregister_qdisc(&master->qops);
	507	unregister_netdev(master->dev);
	508	free_netdev(master->dev);
	509	}
	510	}
	511
	512	module_init(teql_init);
	513	module_exit(teql_exit);
	514
	515	MODULE_LICENSE("GPL");