[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/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;
	struct Qdisc *q;

	skb = __skb_dequeue(&dat->q);
	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
	q = rcu_dereference_bh(dat_queue->qdisc);

	if (skb == NULL) {
		struct net_device *m = qdisc_dev(q);
		if (m) {
			dat->m->slaves = sch;
			netif_wake_queue(m);
		}
	} else {
		qdisc_bstats_update(sch, skb);
	}
	sch->q.qlen = dat->q.qlen + q->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 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(rtnl_dereference(txq->qdisc));
						spin_lock_bh(root_lock);
						qdisc_reset(rtnl_dereference(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(tc_skb_protocol(skb)),
				      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 (rcu_access_pointer(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, slave_txq, false) ==
				    NETDEV_TX_OK) {
					__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;
	netif_keep_dst(dev);
}

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	98	struct sk_buff *skb;
46e5da40	99	struct Qdisc *q;
1da177e4 LT	100
1da177e4 LT	101	skb = __skb_dequeue(&dat->q);
e8a0464c	102	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
46e5da40 JF	103	q = rcu_dereference_bh(dat_queue->qdisc);
46e5da40 JF	104
1da177e4	105	if (skb == NULL) {
46e5da40	106	struct net_device *m = qdisc_dev(q);
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	}
46e5da40	114	sch->q.qlen = dat->q.qlen + q->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
1da177e4	125	static void
cc7ec456	126	teql_reset(struct Qdisc *sch)
1da177e4 LT	127	{
	128	struct teql_sched_data *dat = qdisc_priv(sch);
	129
	130	skb_queue_purge(&dat->q);
	131	sch->q.qlen = 0;
1da177e4 LT	132	}
	133
	134	static void
cc7ec456	135	teql_destroy(struct Qdisc *sch)
1da177e4 LT	136	{
	137	struct Qdisc q, prev;
	138	struct teql_sched_data *dat = qdisc_priv(sch);
	139	struct teql_master *master = dat->m;
	140
cc7ec456 ED	141	prev = master->slaves;
cc7ec456 ED	142	if (prev) {
1da177e4 LT	143	do {
	144	q = NEXT_SLAVE(prev);
	145	if (q == sch) {
	146	NEXT_SLAVE(prev) = NEXT_SLAVE(q);
	147	if (q == master->slaves) {
	148	master->slaves = NEXT_SLAVE(q);
	149	if (q == master->slaves) {
e8a0464c	150	struct netdev_queue *txq;
83874000	151	spinlock_t *root_lock;
e8a0464c DM	152
e8a0464c DM	153	txq = netdev_get_tx_queue(master->dev, 0);
1da177e4	154	master->slaves = NULL;
83874000	155
46e5da40	156	root_lock = qdisc_root_sleeping_lock(rtnl_dereference(txq->qdisc));
83874000	157	spin_lock_bh(root_lock);
46e5da40	158	qdisc_reset(rtnl_dereference(txq->qdisc));
83874000	159	spin_unlock_bh(root_lock);
1da177e4 LT	160	}
	161	}
	162	skb_queue_purge(&dat->q);
1da177e4 LT	163	break;
1da177e4 LT	164	}
10297b99	165
1da177e4 LT	166	} while ((prev = q) != master->slaves);
	167	}
	168	}
	169
1e90474c	170	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt)
1da177e4	171	{
5ce2d488	172	struct net_device *dev = qdisc_dev(sch);
cc7ec456	173	struct teql_master m = (struct teql_master )sch->ops;
1da177e4 LT	174	struct teql_sched_data *q = qdisc_priv(sch);
	175
	176	if (dev->hard_header_len > m->dev->hard_header_len)
	177	return -EINVAL;
	178
	179	if (m->dev == dev)
	180	return -ELOOP;
	181
	182	q->m = m;
	183
	184	skb_queue_head_init(&q->q);
	185
	186	if (m->slaves) {
	187	if (m->dev->flags & IFF_UP) {
f64f9e71 JP	188	if ((m->dev->flags & IFF_POINTOPOINT &&
	189	!(dev->flags & IFF_POINTOPOINT)) \|\|
	190	(m->dev->flags & IFF_BROADCAST &&
	191	!(dev->flags & IFF_BROADCAST)) \|\|
	192	(m->dev->flags & IFF_MULTICAST &&
	193	!(dev->flags & IFF_MULTICAST)) \|\|
	194	dev->mtu < m->dev->mtu)
1da177e4 LT	195	return -EINVAL;
	196	} else {
	197	if (!(dev->flags&IFF_POINTOPOINT))
	198	m->dev->flags &= ~IFF_POINTOPOINT;
	199	if (!(dev->flags&IFF_BROADCAST))
	200	m->dev->flags &= ~IFF_BROADCAST;
	201	if (!(dev->flags&IFF_MULTICAST))
	202	m->dev->flags &= ~IFF_MULTICAST;
	203	if (dev->mtu < m->dev->mtu)
	204	m->dev->mtu = dev->mtu;
	205	}
	206	q->next = NEXT_SLAVE(m->slaves);
	207	NEXT_SLAVE(m->slaves) = sch;
	208	} else {
	209	q->next = sch;
	210	m->slaves = sch;
	211	m->dev->mtu = dev->mtu;
	212	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	213	}
	214	return 0;
	215	}
	216
1da177e4 LT	217
1da177e4 LT	218	static int
f7e57044 ED	219	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res,
f7e57044 ED	220	struct net_device dev, struct netdev_queue txq,
dbedbe6d	221	struct dst_entry *dst)
1da177e4	222	{
dbedbe6d DM	223	struct neighbour *n;
dbedbe6d DM	224	int err = 0;
1da177e4	225
dbedbe6d DM	226	n = dst_neigh_lookup_skb(dst, skb);
	227	if (!n)
	228	return -ENOENT;
	229
	230	if (dst->dev != dev) {
	231	struct neighbour *mn;
	232
	233	mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
	234	neigh_release(n);
	235	if (IS_ERR(mn))
	236	return PTR_ERR(mn);
	237	n = mn;
1da177e4	238	}
dbedbe6d	239
1da177e4 LT	240	if (neigh_event_send(n, skb_res) == 0) {
1da177e4 LT	241	int err;
0ed8ddf4	242	char haddr[MAX_ADDR_LEN];
0c4e8581	243
0ed8ddf4	244	neigh_ha_snapshot(haddr, n, dev);
d8b9605d JP	245	err = dev_hard_header(skb, dev, ntohs(tc_skb_protocol(skb)),
d8b9605d JP	246	haddr, NULL, skb->len);
0c4e8581	247
dbedbe6d DM	248	if (err < 0)
	249	err = -EINVAL;
	250	} else {
	251	err = (skb_res == NULL) ? -EAGAIN : 1;
1da177e4 LT	252	}
1da177e4 LT	253	neigh_release(n);
dbedbe6d	254	return err;
1da177e4 LT	255	}
1da177e4 LT	256
3b04ddde	257	static inline int teql_resolve(struct sk_buff *skb,
f7e57044 ED	258	struct sk_buff *skb_res,
	259	struct net_device *dev,
	260	struct netdev_queue *txq)
1da177e4	261	{
f7e57044	262	struct dst_entry *dst = skb_dst(skb);
f7e57044 ED	263	int res;
f7e57044 ED	264
46e5da40	265	if (rcu_access_pointer(txq->qdisc) == &noop_qdisc)
4f9f8311 EP	266	return -ENODEV;
4f9f8311 EP	267
f7e57044	268	if (!dev->header_ops \|\| !dst)
1da177e4	269	return 0;
f7e57044 ED	270
f7e57044 ED	271	rcu_read_lock();
dbedbe6d	272	res = __teql_resolve(skb, skb_res, dev, txq, dst);
f7e57044 ED	273	rcu_read_unlock();
	274
	275	return res;
1da177e4 LT	276	}
1da177e4 LT	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	299	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
10297b99	300
e8a0464c	301	if (slave_txq->qdisc_sleeping != q)
1da177e4	302	continue;
73466498	303	if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) \|\|
f25f4e44	304	!netif_running(slave)) {
1da177e4 LT	305	busy = 1;
	306	continue;
	307	}
	308
f7e57044	309	switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
1da177e4	310	case 0:
c3f26a26	311	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	312	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	313
73466498	314	if (!netif_xmit_frozen_or_stopped(slave_txq) &&
fa2dbdc2 DM	315	netdev_start_xmit(skb, slave, slave_txq, false) ==
fa2dbdc2 DM	316	NETDEV_TX_OK) {
c3f26a26	317	__netif_tx_unlock(slave_txq);
1da177e4 LT	318	master->slaves = NEXT_SLAVE(q);
1da177e4 LT	319	netif_wake_queue(dev);
1ac9ad13 ED	320	master->tx_packets++;
1ac9ad13 ED	321	master->tx_bytes += length;
6ed10654	322	return NETDEV_TX_OK;
1da177e4	323	}
c3f26a26	324	__netif_tx_unlock(slave_txq);
1da177e4	325	}
73466498	326	if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
1da177e4 LT	327	busy = 1;
	328	break;
	329	case 1:
	330	master->slaves = NEXT_SLAVE(q);
6ed10654	331	return NETDEV_TX_OK;
1da177e4 LT	332	default:
	333	nores = 1;
	334	break;
	335	}
bbe735e4	336	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	337	} while ((q = NEXT_SLAVE(q)) != start);
	338
	339	if (nores && skb_res == NULL) {
	340	skb_res = skb;
	341	goto restart;
	342	}
	343
	344	if (busy) {
	345	netif_stop_queue(dev);
5b548140	346	return NETDEV_TX_BUSY;
1da177e4	347	}
1ac9ad13	348	master->tx_errors++;
1da177e4 LT	349
1da177e4 LT	350	drop:
1ac9ad13	351	master->tx_dropped++;
1da177e4	352	dev_kfree_skb(skb);
6ed10654	353	return NETDEV_TX_OK;
1da177e4 LT	354	}
	355
	356	static int teql_master_open(struct net_device *dev)
	357	{
cc7ec456	358	struct Qdisc *q;
2941a486	359	struct teql_master *m = netdev_priv(dev);
1da177e4	360	int mtu = 0xFFFE;
cc7ec456	361	unsigned int flags = IFF_NOARP \| IFF_MULTICAST;
1da177e4 LT	362
	363	if (m->slaves == NULL)
	364	return -EUNATCH;
	365
	366	flags = FMASK;
	367
	368	q = m->slaves;
	369	do {
5ce2d488	370	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	371
	372	if (slave == NULL)
	373	return -EUNATCH;
	374
	375	if (slave->mtu < mtu)
	376	mtu = slave->mtu;
	377	if (slave->hard_header_len > LL_MAX_HEADER)
	378	return -EINVAL;
	379
	380	/* If all the slaves are BROADCAST, master is BROADCAST
	381	If all the slaves are PtP, master is PtP
	382	Otherwise, master is NBMA.
	383	*/
	384	if (!(slave->flags&IFF_POINTOPOINT))
	385	flags &= ~IFF_POINTOPOINT;
	386	if (!(slave->flags&IFF_BROADCAST))
	387	flags &= ~IFF_BROADCAST;
	388	if (!(slave->flags&IFF_MULTICAST))
	389	flags &= ~IFF_MULTICAST;
	390	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	391
	392	m->dev->mtu = mtu;
	393	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	394	netif_start_queue(m->dev);
	395	return 0;
	396	}
	397
	398	static int teql_master_close(struct net_device *dev)
	399	{
	400	netif_stop_queue(dev);
	401	return 0;
	402	}
	403
1ac9ad13 ED	404	static struct rtnl_link_stats64 teql_master_stats64(struct net_device dev,
	405	struct rtnl_link_stats64 *stats)
	406	{
	407	struct teql_master *m = netdev_priv(dev);
	408
	409	stats->tx_packets = m->tx_packets;
	410	stats->tx_bytes = m->tx_bytes;
	411	stats->tx_errors = m->tx_errors;
	412	stats->tx_dropped = m->tx_dropped;
	413	return stats;
	414	}
	415
1da177e4 LT	416	static int teql_master_mtu(struct net_device *dev, int new_mtu)
1da177e4 LT	417	{
2941a486	418	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	419	struct Qdisc *q;
	420
	421	if (new_mtu < 68)
	422	return -EINVAL;
	423
	424	q = m->slaves;
	425	if (q) {
	426	do {
5ce2d488	427	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4	428	return -EINVAL;
cc7ec456	429	} while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4 LT	430	}
	431
	432	dev->mtu = new_mtu;
	433	return 0;
	434	}
	435
61294e2e SH	436	static const struct net_device_ops teql_netdev_ops = {
	437	.ndo_open = teql_master_open,
	438	.ndo_stop = teql_master_close,
	439	.ndo_start_xmit = teql_master_xmit,
1ac9ad13	440	.ndo_get_stats64 = teql_master_stats64,
61294e2e SH	441	.ndo_change_mtu = teql_master_mtu,
	442	};
	443
1da177e4 LT	444	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	445	{
2941a486	446	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	447	struct Qdisc_ops *ops = &master->qops;
	448
	449	master->dev = dev;
	450	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	451
1da177e4 LT	452	ops->enqueue = teql_enqueue;
1da177e4 LT	453	ops->dequeue = teql_dequeue;
8e3af978	454	ops->peek = teql_peek;
1da177e4 LT	455	ops->init = teql_qdisc_init;
	456	ops->reset = teql_reset;
	457	ops->destroy = teql_destroy;
	458	ops->owner = THIS_MODULE;
	459
61294e2e	460	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	461	dev->type = ARPHRD_VOID;
	462	dev->mtu = 1500;
	463	dev->tx_queue_len = 100;
	464	dev->flags = IFF_NOARP;
	465	dev->hard_header_len = LL_MAX_HEADER;
02875878	466	netif_keep_dst(dev);
1da177e4 LT	467	}
	468
	469	static LIST_HEAD(master_dev_list);
	470	static int max_equalizers = 1;
	471	module_param(max_equalizers, int, 0);
	472	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	473
	474	static int __init teql_init(void)
	475	{
	476	int i;
	477	int err = -ENODEV;
	478
	479	for (i = 0; i < max_equalizers; i++) {
	480	struct net_device *dev;
	481	struct teql_master *master;
	482
c835a677 TG	483	dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
c835a677 TG	484	NET_NAME_UNKNOWN, teql_master_setup);
1da177e4 LT	485	if (!dev) {
	486	err = -ENOMEM;
	487	break;
	488	}
	489
	490	if ((err = register_netdev(dev))) {
	491	free_netdev(dev);
	492	break;
	493	}
	494
2941a486	495	master = netdev_priv(dev);
1da177e4 LT	496
	497	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	498	err = register_qdisc(&master->qops);
	499
	500	if (err) {
	501	unregister_netdev(dev);
	502	free_netdev(dev);
	503	break;
	504	}
	505
	506	list_add_tail(&master->master_list, &master_dev_list);
	507	}
	508	return i ? 0 : err;
	509	}
	510
10297b99	511	static void __exit teql_exit(void)
1da177e4 LT	512	{
	513	struct teql_master master, nxt;
	514
	515	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	516
	517	list_del(&master->master_list);
	518
	519	unregister_qdisc(&master->qops);
	520	unregister_netdev(master->dev);
	521	free_netdev(master->dev);
	522	}
	523	}
	524
	525	module_init(teql_init);
	526	module_exit(teql_exit);
	527
	528	MODULE_LICENSE("GPL");