[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 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;
	}

	return qdisc_drop(skb, sch);
}

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;
}

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);
				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 *txq,
	       struct dst_entry *dst)
{
	struct neighbour *n;
	int err = 0;

	n = dst_neigh_lookup_skb(dst, skb);
	if (!n)
		return -ENOENT;

	if (dst->dev != dev) {
		struct neighbour *mn;

		mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
		neigh_release(n);
		if (IS_ERR(mn))
			return PTR_ERR(mn);
		n = mn;
	}

	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)
			err = -EINVAL;
	} else {
		err = (skb_res == NULL) ? -EAGAIN : 1;
	}
	neigh_release(n);
	return err;
}

static inline int teql_resolve(struct sk_buff *skb,
			       struct sk_buff *skb_res,
			       struct net_device *dev,
			       struct netdev_queue *txq)
{
	struct dst_entry *dst = skb_dst(skb);
	int res;

	if (txq->qdisc == &noop_qdisc)
		return -ENODEV;

	if (!dev->header_ops || !dst)
		return 0;

	rcu_read_lock();
	res = __teql_resolve(skb, skb_res, dev, txq, dst);
	rcu_read_unlock();

	return res;
}

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);

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

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

				if (!netif_xmit_frozen_or_stopped(slave_txq) &&
				    netdev_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_xmit_stopped(netdev_get_tx_queue(dev, 0)))
				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",
				   NET_NAME_UNKNOWN, 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;
1da177e4 LT	69	struct teql_master *m;
1da177e4 LT	70	struct sk_buff_head q;
	71	};
	72
cc7ec456	73	#define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
1da177e4	74
cc7ec456	75	#define FMASK (IFF_BROADCAST \| IFF_POINTOPOINT)
1da177e4 LT	76
	77	/* "teql" qdisc routines /
	78
	79	static int
cc7ec456	80	teql_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	81	{
5ce2d488	82	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	83	struct teql_sched_data *q = qdisc_priv(sch);
1da177e4 LT	84
4cd8c9e8 KK	85	if (q->q.qlen < dev->tx_queue_len) {
4cd8c9e8 KK	86	__skb_queue_tail(&q->q, skb);
9871e50e	87	return NET_XMIT_SUCCESS;
1da177e4 LT	88	}
1da177e4 LT	89
17045755	90	return qdisc_drop(skb, sch);
1da177e4 LT	91	}
1da177e4 LT	92
1da177e4	93	static struct sk_buff *
cc7ec456	94	teql_dequeue(struct Qdisc *sch)
1da177e4 LT	95	{
1da177e4 LT	96	struct teql_sched_data *dat = qdisc_priv(sch);
b0e1e646	97	struct netdev_queue *dat_queue;
1da177e4 LT	98	struct sk_buff *skb;
	99
	100	skb = __skb_dequeue(&dat->q);
e8a0464c	101	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
1da177e4	102	if (skb == NULL) {
b0e1e646	103	struct net_device *m = qdisc_dev(dat_queue->qdisc);
1da177e4 LT	104	if (m) {
	105	dat->m->slaves = sch;
	106	netif_wake_queue(m);
	107	}
9190b3b3 ED	108	} else {
9190b3b3 ED	109	qdisc_bstats_update(sch, skb);
1da177e4	110	}
b0e1e646	111	sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
1da177e4 LT	112	return skb;
	113	}
	114
8e3af978	115	static struct sk_buff *
cc7ec456	116	teql_peek(struct Qdisc *sch)
8e3af978 JP	117	{
	118	/* teql is meant to be used as root qdisc */
	119	return NULL;
	120	}
	121
cc7ec456	122	static inline void
1da177e4 LT	123	teql_neigh_release(struct neighbour *n)
	124	{
	125	if (n)
	126	neigh_release(n);
	127	}
	128
	129	static void
cc7ec456	130	teql_reset(struct Qdisc *sch)
1da177e4 LT	131	{
	132	struct teql_sched_data *dat = qdisc_priv(sch);
	133
	134	skb_queue_purge(&dat->q);
	135	sch->q.qlen = 0;
1da177e4 LT	136	}
	137
	138	static void
cc7ec456	139	teql_destroy(struct Qdisc *sch)
1da177e4 LT	140	{
	141	struct Qdisc q, prev;
	142	struct teql_sched_data *dat = qdisc_priv(sch);
	143	struct teql_master *master = dat->m;
	144
cc7ec456 ED	145	prev = master->slaves;
cc7ec456 ED	146	if (prev) {
1da177e4 LT	147	do {
	148	q = NEXT_SLAVE(prev);
	149	if (q == sch) {
	150	NEXT_SLAVE(prev) = NEXT_SLAVE(q);
	151	if (q == master->slaves) {
	152	master->slaves = NEXT_SLAVE(q);
	153	if (q == master->slaves) {
e8a0464c	154	struct netdev_queue *txq;
83874000	155	spinlock_t *root_lock;
e8a0464c DM	156
e8a0464c DM	157	txq = netdev_get_tx_queue(master->dev, 0);
1da177e4	158	master->slaves = NULL;
83874000	159
102396ae	160	root_lock = qdisc_root_sleeping_lock(txq->qdisc);
83874000	161	spin_lock_bh(root_lock);
e8a0464c	162	qdisc_reset(txq->qdisc);
83874000	163	spin_unlock_bh(root_lock);
1da177e4 LT	164	}
	165	}
	166	skb_queue_purge(&dat->q);
1da177e4 LT	167	break;
1da177e4 LT	168	}
10297b99	169
1da177e4 LT	170	} while ((prev = q) != master->slaves);
	171	}
	172	}
	173
1e90474c	174	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt)
1da177e4	175	{
5ce2d488	176	struct net_device *dev = qdisc_dev(sch);
cc7ec456	177	struct teql_master m = (struct teql_master )sch->ops;
1da177e4 LT	178	struct teql_sched_data *q = qdisc_priv(sch);
	179
	180	if (dev->hard_header_len > m->dev->hard_header_len)
	181	return -EINVAL;
	182
	183	if (m->dev == dev)
	184	return -ELOOP;
	185
	186	q->m = m;
	187
	188	skb_queue_head_init(&q->q);
	189
	190	if (m->slaves) {
	191	if (m->dev->flags & IFF_UP) {
f64f9e71 JP	192	if ((m->dev->flags & IFF_POINTOPOINT &&
	193	!(dev->flags & IFF_POINTOPOINT)) \|\|
	194	(m->dev->flags & IFF_BROADCAST &&
	195	!(dev->flags & IFF_BROADCAST)) \|\|
	196	(m->dev->flags & IFF_MULTICAST &&
	197	!(dev->flags & IFF_MULTICAST)) \|\|
	198	dev->mtu < m->dev->mtu)
1da177e4 LT	199	return -EINVAL;
	200	} else {
	201	if (!(dev->flags&IFF_POINTOPOINT))
	202	m->dev->flags &= ~IFF_POINTOPOINT;
	203	if (!(dev->flags&IFF_BROADCAST))
	204	m->dev->flags &= ~IFF_BROADCAST;
	205	if (!(dev->flags&IFF_MULTICAST))
	206	m->dev->flags &= ~IFF_MULTICAST;
	207	if (dev->mtu < m->dev->mtu)
	208	m->dev->mtu = dev->mtu;
	209	}
	210	q->next = NEXT_SLAVE(m->slaves);
	211	NEXT_SLAVE(m->slaves) = sch;
	212	} else {
	213	q->next = sch;
	214	m->slaves = sch;
	215	m->dev->mtu = dev->mtu;
	216	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	217	}
	218	return 0;
	219	}
	220
1da177e4 LT	221
1da177e4 LT	222	static int
f7e57044 ED	223	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res,
f7e57044 ED	224	struct net_device dev, struct netdev_queue txq,
dbedbe6d	225	struct dst_entry *dst)
1da177e4	226	{
dbedbe6d DM	227	struct neighbour *n;
dbedbe6d DM	228	int err = 0;
1da177e4	229
dbedbe6d DM	230	n = dst_neigh_lookup_skb(dst, skb);
	231	if (!n)
	232	return -ENOENT;
	233
	234	if (dst->dev != dev) {
	235	struct neighbour *mn;
	236
	237	mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
	238	neigh_release(n);
	239	if (IS_ERR(mn))
	240	return PTR_ERR(mn);
	241	n = mn;
1da177e4	242	}
dbedbe6d	243
1da177e4 LT	244	if (neigh_event_send(n, skb_res) == 0) {
1da177e4 LT	245	int err;
0ed8ddf4	246	char haddr[MAX_ADDR_LEN];
0c4e8581	247
0ed8ddf4 ED	248	neigh_ha_snapshot(haddr, n, dev);
	249	err = dev_hard_header(skb, dev, ntohs(skb->protocol), haddr,
	250	NULL, skb->len);
0c4e8581	251
dbedbe6d DM	252	if (err < 0)
	253	err = -EINVAL;
	254	} else {
	255	err = (skb_res == NULL) ? -EAGAIN : 1;
1da177e4 LT	256	}
1da177e4 LT	257	neigh_release(n);
dbedbe6d	258	return err;
1da177e4 LT	259	}
1da177e4 LT	260
3b04ddde	261	static inline int teql_resolve(struct sk_buff *skb,
f7e57044 ED	262	struct sk_buff *skb_res,
	263	struct net_device *dev,
	264	struct netdev_queue *txq)
1da177e4	265	{
f7e57044	266	struct dst_entry *dst = skb_dst(skb);
f7e57044 ED	267	int res;
f7e57044 ED	268
e8a0464c	269	if (txq->qdisc == &noop_qdisc)
4f9f8311 EP	270	return -ENODEV;
4f9f8311 EP	271
f7e57044	272	if (!dev->header_ops \|\| !dst)
1da177e4	273	return 0;
f7e57044 ED	274
f7e57044 ED	275	rcu_read_lock();
dbedbe6d	276	res = __teql_resolve(skb, skb_res, dev, txq, dst);
f7e57044 ED	277	rcu_read_unlock();
	278
	279	return res;
1da177e4 LT	280	}
1da177e4 LT	281
6fef4c0c	282	static netdev_tx_t teql_master_xmit(struct sk_buff skb, struct net_device dev)
1da177e4	283	{
2941a486	284	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	285	struct Qdisc start, q;
	286	int busy;
	287	int nores;
4e3ab47a	288	int subq = skb_get_queue_mapping(skb);
1da177e4 LT	289	struct sk_buff *skb_res = NULL;
	290
	291	start = master->slaves;
	292
	293	restart:
	294	nores = 0;
	295	busy = 0;
	296
cc7ec456 ED	297	q = start;
cc7ec456 ED	298	if (!q)
1da177e4 LT	299	goto drop;
	300
	301	do {
5ce2d488	302	struct net_device *slave = qdisc_dev(q);
61294e2e	303	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
10297b99	304
e8a0464c	305	if (slave_txq->qdisc_sleeping != q)
1da177e4	306	continue;
73466498	307	if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) \|\|
f25f4e44	308	!netif_running(slave)) {
1da177e4 LT	309	busy = 1;
	310	continue;
	311	}
	312
f7e57044	313	switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
1da177e4	314	case 0:
c3f26a26	315	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	316	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	317
73466498	318	if (!netif_xmit_frozen_or_stopped(slave_txq) &&
4798248e	319	netdev_start_xmit(skb, slave) == NETDEV_TX_OK) {
08baf561	320	txq_trans_update(slave_txq);
c3f26a26	321	__netif_tx_unlock(slave_txq);
1da177e4 LT	322	master->slaves = NEXT_SLAVE(q);
1da177e4 LT	323	netif_wake_queue(dev);
1ac9ad13 ED	324	master->tx_packets++;
1ac9ad13 ED	325	master->tx_bytes += length;
6ed10654	326	return NETDEV_TX_OK;
1da177e4	327	}
c3f26a26	328	__netif_tx_unlock(slave_txq);
1da177e4	329	}
73466498	330	if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
1da177e4 LT	331	busy = 1;
	332	break;
	333	case 1:
	334	master->slaves = NEXT_SLAVE(q);
6ed10654	335	return NETDEV_TX_OK;
1da177e4 LT	336	default:
	337	nores = 1;
	338	break;
	339	}
bbe735e4	340	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	341	} while ((q = NEXT_SLAVE(q)) != start);
	342
	343	if (nores && skb_res == NULL) {
	344	skb_res = skb;
	345	goto restart;
	346	}
	347
	348	if (busy) {
	349	netif_stop_queue(dev);
5b548140	350	return NETDEV_TX_BUSY;
1da177e4	351	}
1ac9ad13	352	master->tx_errors++;
1da177e4 LT	353
1da177e4 LT	354	drop:
1ac9ad13	355	master->tx_dropped++;
1da177e4	356	dev_kfree_skb(skb);
6ed10654	357	return NETDEV_TX_OK;
1da177e4 LT	358	}
	359
	360	static int teql_master_open(struct net_device *dev)
	361	{
cc7ec456	362	struct Qdisc *q;
2941a486	363	struct teql_master *m = netdev_priv(dev);
1da177e4	364	int mtu = 0xFFFE;
cc7ec456	365	unsigned int flags = IFF_NOARP \| IFF_MULTICAST;
1da177e4 LT	366
	367	if (m->slaves == NULL)
	368	return -EUNATCH;
	369
	370	flags = FMASK;
	371
	372	q = m->slaves;
	373	do {
5ce2d488	374	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	375
	376	if (slave == NULL)
	377	return -EUNATCH;
	378
	379	if (slave->mtu < mtu)
	380	mtu = slave->mtu;
	381	if (slave->hard_header_len > LL_MAX_HEADER)
	382	return -EINVAL;
	383
	384	/* If all the slaves are BROADCAST, master is BROADCAST
	385	If all the slaves are PtP, master is PtP
	386	Otherwise, master is NBMA.
	387	*/
	388	if (!(slave->flags&IFF_POINTOPOINT))
	389	flags &= ~IFF_POINTOPOINT;
	390	if (!(slave->flags&IFF_BROADCAST))
	391	flags &= ~IFF_BROADCAST;
	392	if (!(slave->flags&IFF_MULTICAST))
	393	flags &= ~IFF_MULTICAST;
	394	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	395
	396	m->dev->mtu = mtu;
	397	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	398	netif_start_queue(m->dev);
	399	return 0;
	400	}
	401
	402	static int teql_master_close(struct net_device *dev)
	403	{
	404	netif_stop_queue(dev);
	405	return 0;
	406	}
	407
1ac9ad13 ED	408	static struct rtnl_link_stats64 teql_master_stats64(struct net_device dev,
	409	struct rtnl_link_stats64 *stats)
	410	{
	411	struct teql_master *m = netdev_priv(dev);
	412
	413	stats->tx_packets = m->tx_packets;
	414	stats->tx_bytes = m->tx_bytes;
	415	stats->tx_errors = m->tx_errors;
	416	stats->tx_dropped = m->tx_dropped;
	417	return stats;
	418	}
	419
1da177e4 LT	420	static int teql_master_mtu(struct net_device *dev, int new_mtu)
1da177e4 LT	421	{
2941a486	422	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	423	struct Qdisc *q;
	424
	425	if (new_mtu < 68)
	426	return -EINVAL;
	427
	428	q = m->slaves;
	429	if (q) {
	430	do {
5ce2d488	431	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4	432	return -EINVAL;
cc7ec456	433	} while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4 LT	434	}
	435
	436	dev->mtu = new_mtu;
	437	return 0;
	438	}
	439
61294e2e SH	440	static const struct net_device_ops teql_netdev_ops = {
	441	.ndo_open = teql_master_open,
	442	.ndo_stop = teql_master_close,
	443	.ndo_start_xmit = teql_master_xmit,
1ac9ad13	444	.ndo_get_stats64 = teql_master_stats64,
61294e2e SH	445	.ndo_change_mtu = teql_master_mtu,
	446	};
	447
1da177e4 LT	448	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	449	{
2941a486	450	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	451	struct Qdisc_ops *ops = &master->qops;
	452
	453	master->dev = dev;
	454	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	455
1da177e4 LT	456	ops->enqueue = teql_enqueue;
1da177e4 LT	457	ops->dequeue = teql_dequeue;
8e3af978	458	ops->peek = teql_peek;
1da177e4 LT	459	ops->init = teql_qdisc_init;
	460	ops->reset = teql_reset;
	461	ops->destroy = teql_destroy;
	462	ops->owner = THIS_MODULE;
	463
61294e2e	464	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	465	dev->type = ARPHRD_VOID;
	466	dev->mtu = 1500;
	467	dev->tx_queue_len = 100;
	468	dev->flags = IFF_NOARP;
	469	dev->hard_header_len = LL_MAX_HEADER;
fa68a782	470	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1da177e4 LT	471	}
	472
	473	static LIST_HEAD(master_dev_list);
	474	static int max_equalizers = 1;
	475	module_param(max_equalizers, int, 0);
	476	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	477
	478	static int __init teql_init(void)
	479	{
	480	int i;
	481	int err = -ENODEV;
	482
	483	for (i = 0; i < max_equalizers; i++) {
	484	struct net_device *dev;
	485	struct teql_master *master;
	486
c835a677 TG	487	dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
c835a677 TG	488	NET_NAME_UNKNOWN, teql_master_setup);
1da177e4 LT	489	if (!dev) {
	490	err = -ENOMEM;
	491	break;
	492	}
	493
	494	if ((err = register_netdev(dev))) {
	495	free_netdev(dev);
	496	break;
	497	}
	498
2941a486	499	master = netdev_priv(dev);
1da177e4 LT	500
	501	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	502	err = register_qdisc(&master->qops);
	503
	504	if (err) {
	505	unregister_netdev(dev);
	506	free_netdev(dev);
	507	break;
	508	}
	509
	510	list_add_tail(&master->master_list, &master_dev_list);
	511	}
	512	return i ? 0 : err;
	513	}
	514
10297b99	515	static void __exit teql_exit(void)
1da177e4 LT	516	{
	517	struct teql_master master, nxt;
	518
	519	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	520
	521	list_del(&master->master_list);
	522
	523	unregister_qdisc(&master->qops);
	524	unregister_netdev(master->dev);
	525	free_netdev(master->dev);
	526	}
	527	}
	528
	529	module_init(teql_init);
	530	module_exit(teql_exit);
	531
	532	MODULE_LICENSE("GPL");