[deliverable/linux.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/slab.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;
	unsigned long	tx_bytes;
	unsigned long	tx_packets;
	unsigned long	tx_errors;
	unsigned long	tx_dropped;
};

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);
		return NET_XMIT_SUCCESS;
	}

	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);
		}
	} else {
		qdisc_bstats_update(sch, skb);
	}
	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;

	prev = master->slaves;
	if (prev) {
		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 = dst_get_neighbour(skb_dst(skb));
	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;
		char haddr[MAX_ADDR_LEN];

		neigh_ha_snapshot(haddr, n, dev);
		err = dev_hard_header(skb, dev, ntohs(skb->protocol), haddr,
				      NULL, skb->len);

		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(skb) == NULL ||
	    dst_get_neighbour(skb_dst(skb)) == NULL)
		return 0;
	return __teql_resolve(skb, skb_res, dev);
}

static netdev_tx_t 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;

	q = start;
	if (!q)
		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_frozen_or_stopped(slave_txq) &&
				    slave_ops->ndo_start_xmit(skb, slave) == NETDEV_TX_OK) {
					txq_trans_update(slave_txq);
					__netif_tx_unlock(slave_txq);
					master->slaves = NEXT_SLAVE(q);
					netif_wake_queue(dev);
					master->tx_packets++;
					master->tx_bytes += length;
					return NETDEV_TX_OK;
				}
				__netif_tx_unlock(slave_txq);
			}
			if (netif_queue_stopped(dev))
				busy = 1;
			break;
		case 1:
			master->slaves = NEXT_SLAVE(q);
			return NETDEV_TX_OK;
		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 NETDEV_TX_BUSY;
	}
	master->tx_errors++;

drop:
	master->tx_dropped++;
	dev_kfree_skb(skb);
	return NETDEV_TX_OK;
}

static int teql_master_open(struct net_device *dev)
{
	struct Qdisc *q;
	struct teql_master *m = netdev_priv(dev);
	int mtu = 0xFFFE;
	unsigned int 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 rtnl_link_stats64 *teql_master_stats64(struct net_device *dev,
						     struct rtnl_link_stats64 *stats)
{
	struct teql_master *m = netdev_priv(dev);

	stats->tx_packets	= m->tx_packets;
	stats->tx_bytes		= m->tx_bytes;
	stats->tx_errors	= m->tx_errors;
	stats->tx_dropped	= m->tx_dropped;
	return 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_stats64 = teql_master_stats64,
	.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;
	dev->priv_flags		&= ~IFF_XMIT_DST_RELEASE;
}

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>
5a0e3ad6	14	#include <linux/slab.h>
1da177e4	15	#include <linux/string.h>
1da177e4	16	#include <linux/errno.h>
14c85021	17	#include <linux/if_arp.h>
1da177e4	18	#include <linux/netdevice.h>
1da177e4	19	#include <linux/init.h>
1da177e4 LT	20	#include <linux/skbuff.h>
1da177e4 LT	21	#include <linux/moduleparam.h>
0ba48053 PM	22	#include <net/dst.h>
0ba48053 PM	23	#include <net/neighbour.h>
1da177e4 LT	24	#include <net/pkt_sched.h>
	25
	26	/*
	27	How to setup it.
	28	----------------
	29
	30	After loading this module you will find a new device teqlN
	31	and new qdisc with the same name. To join a slave to the equalizer
	32	you should just set this qdisc on a device f.e.
	33
	34	# tc qdisc add dev eth0 root teql0
	35	# tc qdisc add dev eth1 root teql0
	36
	37	That's all. Full PnP 8)
	38
	39	Applicability.
	40	--------------
	41
	42	1. Slave devices MUST be active devices, i.e., they must raise the tbusy
	43	signal and generate EOI events. If you want to equalize virtual devices
	44	like tunnels, use a normal eql device.
	45	2. This device puts no limitations on physical slave characteristics
	46	f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
	47	Certainly, large difference in link speeds will make the resulting
	48	eqalized link unusable, because of huge packet reordering.
	49	I estimate an upper useful difference as ~10 times.
	50	3. If the slave requires address resolution, only protocols using
	51	neighbour cache (IPv4/IPv6) will work over the equalized link.
	52	Other protocols are still allowed to use the slave device directly,
	53	which will not break load balancing, though native slave
	54	traffic will have the highest priority. */
	55
cc7ec456	56	struct teql_master {
1da177e4 LT	57	struct Qdisc_ops qops;
	58	struct net_device *dev;
	59	struct Qdisc *slaves;
	60	struct list_head master_list;
1ac9ad13 ED	61	unsigned long tx_bytes;
	62	unsigned long tx_packets;
	63	unsigned long tx_errors;
	64	unsigned long tx_dropped;
1da177e4 LT	65	};
1da177e4 LT	66
cc7ec456	67	struct teql_sched_data {
1da177e4 LT	68	struct Qdisc *next;
	69	struct teql_master *m;
	70	struct neighbour *ncache;
	71	struct sk_buff_head q;
	72	};
	73
cc7ec456	74	#define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
1da177e4	75
cc7ec456	76	#define FMASK (IFF_BROADCAST \| IFF_POINTOPOINT)
1da177e4 LT	77
	78	/* "teql" qdisc routines /
	79
	80	static int
cc7ec456	81	teql_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	82	{
5ce2d488	83	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	84	struct teql_sched_data *q = qdisc_priv(sch);
1da177e4 LT	85
4cd8c9e8 KK	86	if (q->q.qlen < dev->tx_queue_len) {
4cd8c9e8 KK	87	__skb_queue_tail(&q->q, skb);
9871e50e	88	return NET_XMIT_SUCCESS;
1da177e4 LT	89	}
1da177e4 LT	90
1da177e4 LT	91	kfree_skb(skb);
	92	sch->qstats.drops++;
	93	return NET_XMIT_DROP;
	94	}
	95
1da177e4	96	static struct sk_buff *
cc7ec456	97	teql_dequeue(struct Qdisc *sch)
1da177e4 LT	98	{
1da177e4 LT	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	}
9190b3b3 ED	111	} else {
9190b3b3 ED	112	qdisc_bstats_update(sch, skb);
1da177e4	113	}
b0e1e646	114	sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
1da177e4 LT	115	return skb;
	116	}
	117
8e3af978	118	static struct sk_buff *
cc7ec456	119	teql_peek(struct Qdisc *sch)
8e3af978 JP	120	{
	121	/* teql is meant to be used as root qdisc */
	122	return NULL;
	123	}
	124
cc7ec456	125	static inline void
1da177e4 LT	126	teql_neigh_release(struct neighbour *n)
	127	{
	128	if (n)
	129	neigh_release(n);
	130	}
	131
	132	static void
cc7ec456	133	teql_reset(struct Qdisc *sch)
1da177e4 LT	134	{
	135	struct teql_sched_data *dat = qdisc_priv(sch);
	136
	137	skb_queue_purge(&dat->q);
	138	sch->q.qlen = 0;
	139	teql_neigh_release(xchg(&dat->ncache, NULL));
	140	}
	141
	142	static void
cc7ec456	143	teql_destroy(struct Qdisc *sch)
1da177e4 LT	144	{
	145	struct Qdisc q, prev;
	146	struct teql_sched_data *dat = qdisc_priv(sch);
	147	struct teql_master *master = dat->m;
	148
cc7ec456 ED	149	prev = master->slaves;
cc7ec456 ED	150	if (prev) {
1da177e4 LT	151	do {
	152	q = NEXT_SLAVE(prev);
	153	if (q == sch) {
	154	NEXT_SLAVE(prev) = NEXT_SLAVE(q);
	155	if (q == master->slaves) {
	156	master->slaves = NEXT_SLAVE(q);
	157	if (q == master->slaves) {
e8a0464c	158	struct netdev_queue *txq;
83874000	159	spinlock_t *root_lock;
e8a0464c DM	160
e8a0464c DM	161	txq = netdev_get_tx_queue(master->dev, 0);
1da177e4	162	master->slaves = NULL;
83874000	163
102396ae	164	root_lock = qdisc_root_sleeping_lock(txq->qdisc);
83874000	165	spin_lock_bh(root_lock);
e8a0464c	166	qdisc_reset(txq->qdisc);
83874000	167	spin_unlock_bh(root_lock);
1da177e4 LT	168	}
	169	}
	170	skb_queue_purge(&dat->q);
	171	teql_neigh_release(xchg(&dat->ncache, NULL));
	172	break;
	173	}
10297b99	174
1da177e4 LT	175	} while ((prev = q) != master->slaves);
	176	}
	177	}
	178
1e90474c	179	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt)
1da177e4	180	{
5ce2d488	181	struct net_device *dev = qdisc_dev(sch);
cc7ec456	182	struct teql_master m = (struct teql_master )sch->ops;
1da177e4 LT	183	struct teql_sched_data *q = qdisc_priv(sch);
	184
	185	if (dev->hard_header_len > m->dev->hard_header_len)
	186	return -EINVAL;
	187
	188	if (m->dev == dev)
	189	return -ELOOP;
	190
	191	q->m = m;
	192
	193	skb_queue_head_init(&q->q);
	194
	195	if (m->slaves) {
	196	if (m->dev->flags & IFF_UP) {
f64f9e71 JP	197	if ((m->dev->flags & IFF_POINTOPOINT &&
	198	!(dev->flags & IFF_POINTOPOINT)) \|\|
	199	(m->dev->flags & IFF_BROADCAST &&
	200	!(dev->flags & IFF_BROADCAST)) \|\|
	201	(m->dev->flags & IFF_MULTICAST &&
	202	!(dev->flags & IFF_MULTICAST)) \|\|
	203	dev->mtu < m->dev->mtu)
1da177e4 LT	204	return -EINVAL;
	205	} else {
	206	if (!(dev->flags&IFF_POINTOPOINT))
	207	m->dev->flags &= ~IFF_POINTOPOINT;
	208	if (!(dev->flags&IFF_BROADCAST))
	209	m->dev->flags &= ~IFF_BROADCAST;
	210	if (!(dev->flags&IFF_MULTICAST))
	211	m->dev->flags &= ~IFF_MULTICAST;
	212	if (dev->mtu < m->dev->mtu)
	213	m->dev->mtu = dev->mtu;
	214	}
	215	q->next = NEXT_SLAVE(m->slaves);
	216	NEXT_SLAVE(m->slaves) = sch;
	217	} else {
	218	q->next = sch;
	219	m->slaves = sch;
	220	m->dev->mtu = dev->mtu;
	221	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	222	}
	223	return 0;
	224	}
	225
1da177e4 LT	226
	227	static int
	228	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res, struct net_device *dev)
	229	{
e8a0464c DM	230	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
e8a0464c DM	231	struct teql_sched_data *q = qdisc_priv(dev_queue->qdisc);
69cce1d1	232	struct neighbour *mn = dst_get_neighbour(skb_dst(skb));
1da177e4 LT	233	struct neighbour *n = q->ncache;
	234
	235	if (mn->tbl == NULL)
	236	return -EINVAL;
	237	if (n && n->tbl == mn->tbl &&
	238	memcmp(n->primary_key, mn->primary_key, mn->tbl->key_len) == 0) {
	239	atomic_inc(&n->refcnt);
	240	} else {
	241	n = __neigh_lookup_errno(mn->tbl, mn->primary_key, dev);
	242	if (IS_ERR(n))
	243	return PTR_ERR(n);
	244	}
	245	if (neigh_event_send(n, skb_res) == 0) {
	246	int err;
0ed8ddf4	247	char haddr[MAX_ADDR_LEN];
0c4e8581	248
0ed8ddf4 ED	249	neigh_ha_snapshot(haddr, n, dev);
	250	err = dev_hard_header(skb, dev, ntohs(skb->protocol), haddr,
	251	NULL, skb->len);
0c4e8581	252
1da177e4 LT	253	if (err < 0) {
	254	neigh_release(n);
	255	return -EINVAL;
	256	}
	257	teql_neigh_release(xchg(&q->ncache, n));
	258	return 0;
	259	}
	260	neigh_release(n);
	261	return (skb_res == NULL) ? -EAGAIN : 1;
	262	}
	263
3b04ddde SH	264	static inline int teql_resolve(struct sk_buff *skb,
3b04ddde SH	265	struct sk_buff skb_res, struct net_device dev)
1da177e4	266	{
e8a0464c DM	267	struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
e8a0464c DM	268	if (txq->qdisc == &noop_qdisc)
4f9f8311 EP	269	return -ENODEV;
4f9f8311 EP	270
3b04ddde	271	if (dev->header_ops == NULL \|\|
adf30907	272	skb_dst(skb) == NULL \|\|
69cce1d1	273	dst_get_neighbour(skb_dst(skb)) == NULL)
1da177e4 LT	274	return 0;
	275	return __teql_resolve(skb, skb_res, dev);
	276	}
	277
6fef4c0c	278	static netdev_tx_t teql_master_xmit(struct sk_buff skb, struct net_device dev)
1da177e4	279	{
2941a486	280	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	281	struct Qdisc start, q;
	282	int busy;
	283	int nores;
4e3ab47a	284	int subq = skb_get_queue_mapping(skb);
1da177e4 LT	285	struct sk_buff *skb_res = NULL;
	286
	287	start = master->slaves;
	288
	289	restart:
	290	nores = 0;
	291	busy = 0;
	292
cc7ec456 ED	293	q = start;
cc7ec456 ED	294	if (!q)
1da177e4 LT	295	goto drop;
	296
	297	do {
5ce2d488	298	struct net_device *slave = qdisc_dev(q);
61294e2e SH	299	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
61294e2e SH	300	const struct net_device_ops *slave_ops = slave->netdev_ops;
10297b99	301
e8a0464c	302	if (slave_txq->qdisc_sleeping != q)
1da177e4	303	continue;
fd2ea0a7	304	if (__netif_subqueue_stopped(slave, subq) \|\|
f25f4e44	305	!netif_running(slave)) {
1da177e4 LT	306	busy = 1;
	307	continue;
	308	}
	309
	310	switch (teql_resolve(skb, skb_res, slave)) {
	311	case 0:
c3f26a26	312	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	313	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	314
5a0d2268	315	if (!netif_tx_queue_frozen_or_stopped(slave_txq) &&
6fef4c0c	316	slave_ops->ndo_start_xmit(skb, slave) == NETDEV_TX_OK) {
08baf561	317	txq_trans_update(slave_txq);
c3f26a26	318	__netif_tx_unlock(slave_txq);
1da177e4 LT	319	master->slaves = NEXT_SLAVE(q);
1da177e4 LT	320	netif_wake_queue(dev);
1ac9ad13 ED	321	master->tx_packets++;
1ac9ad13 ED	322	master->tx_bytes += length;
6ed10654	323	return NETDEV_TX_OK;
1da177e4	324	}
c3f26a26	325	__netif_tx_unlock(slave_txq);
1da177e4 LT	326	}
	327	if (netif_queue_stopped(dev))
	328	busy = 1;
	329	break;
	330	case 1:
	331	master->slaves = NEXT_SLAVE(q);
6ed10654	332	return NETDEV_TX_OK;
1da177e4 LT	333	default:
	334	nores = 1;
	335	break;
	336	}
bbe735e4	337	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	338	} while ((q = NEXT_SLAVE(q)) != start);
	339
	340	if (nores && skb_res == NULL) {
	341	skb_res = skb;
	342	goto restart;
	343	}
	344
	345	if (busy) {
	346	netif_stop_queue(dev);
5b548140	347	return NETDEV_TX_BUSY;
1da177e4	348	}
1ac9ad13	349	master->tx_errors++;
1da177e4 LT	350
1da177e4 LT	351	drop:
1ac9ad13	352	master->tx_dropped++;
1da177e4	353	dev_kfree_skb(skb);
6ed10654	354	return NETDEV_TX_OK;
1da177e4 LT	355	}
	356
	357	static int teql_master_open(struct net_device *dev)
	358	{
cc7ec456	359	struct Qdisc *q;
2941a486	360	struct teql_master *m = netdev_priv(dev);
1da177e4	361	int mtu = 0xFFFE;
cc7ec456	362	unsigned int flags = IFF_NOARP \| IFF_MULTICAST;
1da177e4 LT	363
	364	if (m->slaves == NULL)
	365	return -EUNATCH;
	366
	367	flags = FMASK;
	368
	369	q = m->slaves;
	370	do {
5ce2d488	371	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	372
	373	if (slave == NULL)
	374	return -EUNATCH;
	375
	376	if (slave->mtu < mtu)
	377	mtu = slave->mtu;
	378	if (slave->hard_header_len > LL_MAX_HEADER)
	379	return -EINVAL;
	380
	381	/* If all the slaves are BROADCAST, master is BROADCAST
	382	If all the slaves are PtP, master is PtP
	383	Otherwise, master is NBMA.
	384	*/
	385	if (!(slave->flags&IFF_POINTOPOINT))
	386	flags &= ~IFF_POINTOPOINT;
	387	if (!(slave->flags&IFF_BROADCAST))
	388	flags &= ~IFF_BROADCAST;
	389	if (!(slave->flags&IFF_MULTICAST))
	390	flags &= ~IFF_MULTICAST;
	391	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	392
	393	m->dev->mtu = mtu;
	394	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	395	netif_start_queue(m->dev);
	396	return 0;
	397	}
	398
	399	static int teql_master_close(struct net_device *dev)
	400	{
	401	netif_stop_queue(dev);
	402	return 0;
	403	}
	404
1ac9ad13 ED	405	static struct rtnl_link_stats64 teql_master_stats64(struct net_device dev,
	406	struct rtnl_link_stats64 *stats)
	407	{
	408	struct teql_master *m = netdev_priv(dev);
	409
	410	stats->tx_packets = m->tx_packets;
	411	stats->tx_bytes = m->tx_bytes;
	412	stats->tx_errors = m->tx_errors;
	413	stats->tx_dropped = m->tx_dropped;
	414	return stats;
	415	}
	416
1da177e4 LT	417	static int teql_master_mtu(struct net_device *dev, int new_mtu)
1da177e4 LT	418	{
2941a486	419	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	420	struct Qdisc *q;
	421
	422	if (new_mtu < 68)
	423	return -EINVAL;
	424
	425	q = m->slaves;
	426	if (q) {
	427	do {
5ce2d488	428	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4	429	return -EINVAL;
cc7ec456	430	} while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4 LT	431	}
	432
	433	dev->mtu = new_mtu;
	434	return 0;
	435	}
	436
61294e2e SH	437	static const struct net_device_ops teql_netdev_ops = {
	438	.ndo_open = teql_master_open,
	439	.ndo_stop = teql_master_close,
	440	.ndo_start_xmit = teql_master_xmit,
1ac9ad13	441	.ndo_get_stats64 = teql_master_stats64,
61294e2e SH	442	.ndo_change_mtu = teql_master_mtu,
	443	};
	444
1da177e4 LT	445	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	446	{
2941a486	447	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	448	struct Qdisc_ops *ops = &master->qops;
	449
	450	master->dev = dev;
	451	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	452
1da177e4 LT	453	ops->enqueue = teql_enqueue;
1da177e4 LT	454	ops->dequeue = teql_dequeue;
8e3af978	455	ops->peek = teql_peek;
1da177e4 LT	456	ops->init = teql_qdisc_init;
	457	ops->reset = teql_reset;
	458	ops->destroy = teql_destroy;
	459	ops->owner = THIS_MODULE;
	460
61294e2e	461	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	462	dev->type = ARPHRD_VOID;
	463	dev->mtu = 1500;
	464	dev->tx_queue_len = 100;
	465	dev->flags = IFF_NOARP;
	466	dev->hard_header_len = LL_MAX_HEADER;
fa68a782	467	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1da177e4 LT	468	}
	469
	470	static LIST_HEAD(master_dev_list);
	471	static int max_equalizers = 1;
	472	module_param(max_equalizers, int, 0);
	473	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	474
	475	static int __init teql_init(void)
	476	{
	477	int i;
	478	int err = -ENODEV;
	479
	480	for (i = 0; i < max_equalizers; i++) {
	481	struct net_device *dev;
	482	struct teql_master *master;
	483
	484	dev = alloc_netdev(sizeof(struct teql_master),
	485	"teql%d", teql_master_setup);
	486	if (!dev) {
	487	err = -ENOMEM;
	488	break;
	489	}
	490
	491	if ((err = register_netdev(dev))) {
	492	free_netdev(dev);
	493	break;
	494	}
	495
2941a486	496	master = netdev_priv(dev);
1da177e4 LT	497
	498	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	499	err = register_qdisc(&master->qops);
	500
	501	if (err) {
	502	unregister_netdev(dev);
	503	free_netdev(dev);
	504	break;
	505	}
	506
	507	list_add_tail(&master->master_list, &master_dev_list);
	508	}
	509	return i ? 0 : err;
	510	}
	511
10297b99	512	static void __exit teql_exit(void)
1da177e4 LT	513	{
	514	struct teql_master master, nxt;
	515
	516	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	517
	518	list_del(&master->master_list);
	519
	520	unregister_qdisc(&master->qops);
	521	unregister_netdev(master->dev);
	522	free_netdev(master->dev);
	523	}
	524	}
	525
	526	module_init(teql_init);
	527	module_exit(teql_exit);
	528
	529	MODULE_LICENSE("GPL");