[mirror_ubuntu-hirsute-kernel.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 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(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(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 (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;
	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	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
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;
1da177e4 LT	139	}
	140
	141	static void
cc7ec456	142	teql_destroy(struct Qdisc *sch)
1da177e4 LT	143	{
	144	struct Qdisc q, prev;
	145	struct teql_sched_data *dat = qdisc_priv(sch);
	146	struct teql_master *master = dat->m;
	147
cc7ec456 ED	148	prev = master->slaves;
cc7ec456 ED	149	if (prev) {
1da177e4 LT	150	do {
	151	q = NEXT_SLAVE(prev);
	152	if (q == sch) {
	153	NEXT_SLAVE(prev) = NEXT_SLAVE(q);
	154	if (q == master->slaves) {
	155	master->slaves = NEXT_SLAVE(q);
	156	if (q == master->slaves) {
e8a0464c	157	struct netdev_queue *txq;
83874000	158	spinlock_t *root_lock;
e8a0464c DM	159
e8a0464c DM	160	txq = netdev_get_tx_queue(master->dev, 0);
1da177e4	161	master->slaves = NULL;
83874000	162
46e5da40	163	root_lock = qdisc_root_sleeping_lock(rtnl_dereference(txq->qdisc));
83874000	164	spin_lock_bh(root_lock);
46e5da40	165	qdisc_reset(rtnl_dereference(txq->qdisc));
83874000	166	spin_unlock_bh(root_lock);
1da177e4 LT	167	}
	168	}
	169	skb_queue_purge(&dat->q);
1da177e4 LT	170	break;
1da177e4 LT	171	}
10297b99	172
1da177e4 LT	173	} while ((prev = q) != master->slaves);
	174	}
	175	}
	176
1e90474c	177	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt)
1da177e4	178	{
5ce2d488	179	struct net_device *dev = qdisc_dev(sch);
cc7ec456	180	struct teql_master m = (struct teql_master )sch->ops;
1da177e4 LT	181	struct teql_sched_data *q = qdisc_priv(sch);
	182
	183	if (dev->hard_header_len > m->dev->hard_header_len)
	184	return -EINVAL;
	185
	186	if (m->dev == dev)
	187	return -ELOOP;
	188
	189	q->m = m;
	190
	191	skb_queue_head_init(&q->q);
	192
	193	if (m->slaves) {
	194	if (m->dev->flags & IFF_UP) {
f64f9e71 JP	195	if ((m->dev->flags & IFF_POINTOPOINT &&
	196	!(dev->flags & IFF_POINTOPOINT)) \|\|
	197	(m->dev->flags & IFF_BROADCAST &&
	198	!(dev->flags & IFF_BROADCAST)) \|\|
	199	(m->dev->flags & IFF_MULTICAST &&
	200	!(dev->flags & IFF_MULTICAST)) \|\|
	201	dev->mtu < m->dev->mtu)
1da177e4 LT	202	return -EINVAL;
	203	} else {
	204	if (!(dev->flags&IFF_POINTOPOINT))
	205	m->dev->flags &= ~IFF_POINTOPOINT;
	206	if (!(dev->flags&IFF_BROADCAST))
	207	m->dev->flags &= ~IFF_BROADCAST;
	208	if (!(dev->flags&IFF_MULTICAST))
	209	m->dev->flags &= ~IFF_MULTICAST;
	210	if (dev->mtu < m->dev->mtu)
	211	m->dev->mtu = dev->mtu;
	212	}
	213	q->next = NEXT_SLAVE(m->slaves);
	214	NEXT_SLAVE(m->slaves) = sch;
	215	} else {
	216	q->next = sch;
	217	m->slaves = sch;
	218	m->dev->mtu = dev->mtu;
	219	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	220	}
	221	return 0;
	222	}
	223
1da177e4 LT	224
1da177e4 LT	225	static int
f7e57044 ED	226	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res,
f7e57044 ED	227	struct net_device dev, struct netdev_queue txq,
dbedbe6d	228	struct dst_entry *dst)
1da177e4	229	{
dbedbe6d DM	230	struct neighbour *n;
dbedbe6d DM	231	int err = 0;
1da177e4	232
dbedbe6d DM	233	n = dst_neigh_lookup_skb(dst, skb);
	234	if (!n)
	235	return -ENOENT;
	236
	237	if (dst->dev != dev) {
	238	struct neighbour *mn;
	239
	240	mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
	241	neigh_release(n);
	242	if (IS_ERR(mn))
	243	return PTR_ERR(mn);
	244	n = mn;
1da177e4	245	}
dbedbe6d	246
1da177e4 LT	247	if (neigh_event_send(n, skb_res) == 0) {
1da177e4 LT	248	int err;
0ed8ddf4	249	char haddr[MAX_ADDR_LEN];
0c4e8581	250
0ed8ddf4 ED	251	neigh_ha_snapshot(haddr, n, dev);
	252	err = dev_hard_header(skb, dev, ntohs(skb->protocol), haddr,
	253	NULL, skb->len);
0c4e8581	254
dbedbe6d DM	255	if (err < 0)
	256	err = -EINVAL;
	257	} else {
	258	err = (skb_res == NULL) ? -EAGAIN : 1;
1da177e4 LT	259	}
1da177e4 LT	260	neigh_release(n);
dbedbe6d	261	return err;
1da177e4 LT	262	}
1da177e4 LT	263
3b04ddde	264	static inline int teql_resolve(struct sk_buff *skb,
f7e57044 ED	265	struct sk_buff *skb_res,
	266	struct net_device *dev,
	267	struct netdev_queue *txq)
1da177e4	268	{
f7e57044	269	struct dst_entry *dst = skb_dst(skb);
f7e57044 ED	270	int res;
f7e57044 ED	271
46e5da40	272	if (rcu_access_pointer(txq->qdisc) == &noop_qdisc)
4f9f8311 EP	273	return -ENODEV;
4f9f8311 EP	274
f7e57044	275	if (!dev->header_ops \|\| !dst)
1da177e4	276	return 0;
f7e57044 ED	277
f7e57044 ED	278	rcu_read_lock();
dbedbe6d	279	res = __teql_resolve(skb, skb_res, dev, txq, dst);
f7e57044 ED	280	rcu_read_unlock();
	281
	282	return res;
1da177e4 LT	283	}
1da177e4 LT	284
6fef4c0c	285	static netdev_tx_t teql_master_xmit(struct sk_buff skb, struct net_device dev)
1da177e4	286	{
2941a486	287	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	288	struct Qdisc start, q;
	289	int busy;
	290	int nores;
4e3ab47a	291	int subq = skb_get_queue_mapping(skb);
1da177e4 LT	292	struct sk_buff *skb_res = NULL;
	293
	294	start = master->slaves;
	295
	296	restart:
	297	nores = 0;
	298	busy = 0;
	299
cc7ec456 ED	300	q = start;
cc7ec456 ED	301	if (!q)
1da177e4 LT	302	goto drop;
	303
	304	do {
5ce2d488	305	struct net_device *slave = qdisc_dev(q);
61294e2e	306	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
10297b99	307
e8a0464c	308	if (slave_txq->qdisc_sleeping != q)
1da177e4	309	continue;
73466498	310	if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) \|\|
f25f4e44	311	!netif_running(slave)) {
1da177e4 LT	312	busy = 1;
	313	continue;
	314	}
	315
f7e57044	316	switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
1da177e4	317	case 0:
c3f26a26	318	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	319	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	320
73466498	321	if (!netif_xmit_frozen_or_stopped(slave_txq) &&
fa2dbdc2 DM	322	netdev_start_xmit(skb, slave, slave_txq, false) ==
fa2dbdc2 DM	323	NETDEV_TX_OK) {
c3f26a26	324	__netif_tx_unlock(slave_txq);
1da177e4 LT	325	master->slaves = NEXT_SLAVE(q);
1da177e4 LT	326	netif_wake_queue(dev);
1ac9ad13 ED	327	master->tx_packets++;
1ac9ad13 ED	328	master->tx_bytes += length;
6ed10654	329	return NETDEV_TX_OK;
1da177e4	330	}
c3f26a26	331	__netif_tx_unlock(slave_txq);
1da177e4	332	}
73466498	333	if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
1da177e4 LT	334	busy = 1;
	335	break;
	336	case 1:
	337	master->slaves = NEXT_SLAVE(q);
6ed10654	338	return NETDEV_TX_OK;
1da177e4 LT	339	default:
	340	nores = 1;
	341	break;
	342	}
bbe735e4	343	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	344	} while ((q = NEXT_SLAVE(q)) != start);
	345
	346	if (nores && skb_res == NULL) {
	347	skb_res = skb;
	348	goto restart;
	349	}
	350
	351	if (busy) {
	352	netif_stop_queue(dev);
5b548140	353	return NETDEV_TX_BUSY;
1da177e4	354	}
1ac9ad13	355	master->tx_errors++;
1da177e4 LT	356
1da177e4 LT	357	drop:
1ac9ad13	358	master->tx_dropped++;
1da177e4	359	dev_kfree_skb(skb);
6ed10654	360	return NETDEV_TX_OK;
1da177e4 LT	361	}
	362
	363	static int teql_master_open(struct net_device *dev)
	364	{
cc7ec456	365	struct Qdisc *q;
2941a486	366	struct teql_master *m = netdev_priv(dev);
1da177e4	367	int mtu = 0xFFFE;
cc7ec456	368	unsigned int flags = IFF_NOARP \| IFF_MULTICAST;
1da177e4 LT	369
	370	if (m->slaves == NULL)
	371	return -EUNATCH;
	372
	373	flags = FMASK;
	374
	375	q = m->slaves;
	376	do {
5ce2d488	377	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	378
	379	if (slave == NULL)
	380	return -EUNATCH;
	381
	382	if (slave->mtu < mtu)
	383	mtu = slave->mtu;
	384	if (slave->hard_header_len > LL_MAX_HEADER)
	385	return -EINVAL;
	386
	387	/* If all the slaves are BROADCAST, master is BROADCAST
	388	If all the slaves are PtP, master is PtP
	389	Otherwise, master is NBMA.
	390	*/
	391	if (!(slave->flags&IFF_POINTOPOINT))
	392	flags &= ~IFF_POINTOPOINT;
	393	if (!(slave->flags&IFF_BROADCAST))
	394	flags &= ~IFF_BROADCAST;
	395	if (!(slave->flags&IFF_MULTICAST))
	396	flags &= ~IFF_MULTICAST;
	397	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	398
	399	m->dev->mtu = mtu;
	400	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	401	netif_start_queue(m->dev);
	402	return 0;
	403	}
	404
	405	static int teql_master_close(struct net_device *dev)
	406	{
	407	netif_stop_queue(dev);
	408	return 0;
	409	}
	410
1ac9ad13 ED	411	static struct rtnl_link_stats64 teql_master_stats64(struct net_device dev,
	412	struct rtnl_link_stats64 *stats)
	413	{
	414	struct teql_master *m = netdev_priv(dev);
	415
	416	stats->tx_packets = m->tx_packets;
	417	stats->tx_bytes = m->tx_bytes;
	418	stats->tx_errors = m->tx_errors;
	419	stats->tx_dropped = m->tx_dropped;
	420	return stats;
	421	}
	422
1da177e4 LT	423	static int teql_master_mtu(struct net_device *dev, int new_mtu)
1da177e4 LT	424	{
2941a486	425	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	426	struct Qdisc *q;
	427
	428	if (new_mtu < 68)
	429	return -EINVAL;
	430
	431	q = m->slaves;
	432	if (q) {
	433	do {
5ce2d488	434	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4	435	return -EINVAL;
cc7ec456	436	} while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4 LT	437	}
	438
	439	dev->mtu = new_mtu;
	440	return 0;
	441	}
	442
61294e2e SH	443	static const struct net_device_ops teql_netdev_ops = {
	444	.ndo_open = teql_master_open,
	445	.ndo_stop = teql_master_close,
	446	.ndo_start_xmit = teql_master_xmit,
1ac9ad13	447	.ndo_get_stats64 = teql_master_stats64,
61294e2e SH	448	.ndo_change_mtu = teql_master_mtu,
	449	};
	450
1da177e4 LT	451	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	452	{
2941a486	453	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	454	struct Qdisc_ops *ops = &master->qops;
	455
	456	master->dev = dev;
	457	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	458
1da177e4 LT	459	ops->enqueue = teql_enqueue;
1da177e4 LT	460	ops->dequeue = teql_dequeue;
8e3af978	461	ops->peek = teql_peek;
1da177e4 LT	462	ops->init = teql_qdisc_init;
	463	ops->reset = teql_reset;
	464	ops->destroy = teql_destroy;
	465	ops->owner = THIS_MODULE;
	466
61294e2e	467	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	468	dev->type = ARPHRD_VOID;
	469	dev->mtu = 1500;
	470	dev->tx_queue_len = 100;
	471	dev->flags = IFF_NOARP;
	472	dev->hard_header_len = LL_MAX_HEADER;
fa68a782	473	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1da177e4 LT	474	}
	475
	476	static LIST_HEAD(master_dev_list);
	477	static int max_equalizers = 1;
	478	module_param(max_equalizers, int, 0);
	479	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	480
	481	static int __init teql_init(void)
	482	{
	483	int i;
	484	int err = -ENODEV;
	485
	486	for (i = 0; i < max_equalizers; i++) {
	487	struct net_device *dev;
	488	struct teql_master *master;
	489
c835a677 TG	490	dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
c835a677 TG	491	NET_NAME_UNKNOWN, teql_master_setup);
1da177e4 LT	492	if (!dev) {
	493	err = -ENOMEM;
	494	break;
	495	}
	496
	497	if ((err = register_netdev(dev))) {
	498	free_netdev(dev);
	499	break;
	500	}
	501
2941a486	502	master = netdev_priv(dev);
1da177e4 LT	503
	504	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	505	err = register_qdisc(&master->qops);
	506
	507	if (err) {
	508	unregister_netdev(dev);
	509	free_netdev(dev);
	510	break;
	511	}
	512
	513	list_add_tail(&master->master_list, &master_dev_list);
	514	}
	515	return i ? 0 : err;
	516	}
	517
10297b99	518	static void __exit teql_exit(void)
1da177e4 LT	519	{
	520	struct teql_master master, nxt;
	521
	522	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	523
	524	list_del(&master->master_list);
	525
	526	unregister_qdisc(&master->qops);
	527	unregister_netdev(master->dev);
	528	free_netdev(master->dev);
	529	}
	530	}
	531
	532	module_init(teql_init);
	533	module_exit(teql_exit);
	534
	535	MODULE_LICENSE("GPL");