[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, <[email protected]>
 */

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

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

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

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

   That's all. Full PnP 8)

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

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

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

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

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

#define FMASK (IFF_BROADCAST|IFF_POINTOPOINT)

/* "teql*" qdisc routines */

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

	q->m = m;

	skb_queue_head_init(&q->q);

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


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

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

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

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

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

	if (dev->header_ops == NULL ||
	    skb_dst(skb) == NULL ||
	    skb_dst(skb)->neighbour == 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 netdev_queue *txq = netdev_get_tx_queue(dev, 0);
	struct Qdisc *start, *q;
	int busy;
	int nores;
	int subq = skb_get_queue_mapping(skb);
	struct sk_buff *skb_res = NULL;

	start = master->slaves;

restart:
	nores = 0;
	busy = 0;

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

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

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

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

				if (!netif_tx_queue_stopped(slave_txq) &&
				    !netif_tx_queue_frozen(slave_txq) &&
				    slave_ops->ndo_start_xmit(skb, slave) == NETDEV_TX_OK) {
					txq_trans_update(slave_txq);
					__netif_tx_unlock(slave_txq);
					master->slaves = NEXT_SLAVE(q);
					netif_wake_queue(dev);
					txq->tx_packets++;
					txq->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;
	}
	dev->stats.tx_errors++;

drop:
	txq->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 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 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_change_mtu	= teql_master_mtu,
};

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

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

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

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

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

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

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

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

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

		master = netdev_priv(dev);

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

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

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

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

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

		list_del(&master->master_list);

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

module_init(teql_init);
module_exit(teql_exit);

MODULE_LICENSE("GPL");
Commit	Line	Data
1da177e4 LT	1	/* net/sched/sch_teql.c "True" (or "trivial") link equalizer.
	2	*
	3	* This program is free software; you can redistribute it and/or
	4	* modify it under the terms of the GNU General Public License
	5	* as published by the Free Software Foundation; either version
	6	* 2 of the License, or (at your option) any later version.
	7	*
	8	* Authors: Alexey Kuznetsov, <[email protected]>
	9	*/
	10
	11	#include <linux/module.h>
1da177e4 LT	12	#include <linux/types.h>
1da177e4 LT	13	#include <linux/kernel.h>
1da177e4	14	#include <linux/string.h>
1da177e4	15	#include <linux/errno.h>
14c85021	16	#include <linux/if_arp.h>
1da177e4	17	#include <linux/netdevice.h>
1da177e4	18	#include <linux/init.h>
1da177e4 LT	19	#include <linux/skbuff.h>
1da177e4 LT	20	#include <linux/moduleparam.h>
0ba48053 PM	21	#include <net/dst.h>
0ba48053 PM	22	#include <net/neighbour.h>
1da177e4 LT	23	#include <net/pkt_sched.h>
	24
	25	/*
	26	How to setup it.
	27	----------------
	28
	29	After loading this module you will find a new device teqlN
	30	and new qdisc with the same name. To join a slave to the equalizer
	31	you should just set this qdisc on a device f.e.
	32
	33	# tc qdisc add dev eth0 root teql0
	34	# tc qdisc add dev eth1 root teql0
	35
	36	That's all. Full PnP 8)
	37
	38	Applicability.
	39	--------------
	40
	41	1. Slave devices MUST be active devices, i.e., they must raise the tbusy
	42	signal and generate EOI events. If you want to equalize virtual devices
	43	like tunnels, use a normal eql device.
	44	2. This device puts no limitations on physical slave characteristics
	45	f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
	46	Certainly, large difference in link speeds will make the resulting
	47	eqalized link unusable, because of huge packet reordering.
	48	I estimate an upper useful difference as ~10 times.
	49	3. If the slave requires address resolution, only protocols using
	50	neighbour cache (IPv4/IPv6) will work over the equalized link.
	51	Other protocols are still allowed to use the slave device directly,
	52	which will not break load balancing, though native slave
	53	traffic will have the highest priority. */
	54
	55	struct teql_master
	56	{
	57	struct Qdisc_ops qops;
	58	struct net_device *dev;
	59	struct Qdisc *slaves;
	60	struct list_head master_list;
1da177e4 LT	61	};
	62
	63	struct teql_sched_data
	64	{
	65	struct Qdisc *next;
	66	struct teql_master *m;
	67	struct neighbour *ncache;
	68	struct sk_buff_head q;
	69	};
	70
	71	#define NEXT_SLAVE(q) (((struct teql_sched_data*)qdisc_priv(q))->next)
	72
cc8fd14d	73	#define FMASK (IFF_BROADCAST\|IFF_POINTOPOINT)
1da177e4 LT	74
	75	/* "teql" qdisc routines /
	76
	77	static int
	78	teql_enqueue(struct sk_buff skb, struct Qdisc sch)
	79	{
5ce2d488	80	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	81	struct teql_sched_data *q = qdisc_priv(sch);
1da177e4 LT	82
4cd8c9e8 KK	83	if (q->q.qlen < dev->tx_queue_len) {
4cd8c9e8 KK	84	__skb_queue_tail(&q->q, skb);
0abf77e5	85	sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4 LT	86	sch->bstats.packets++;
	87	return 0;
	88	}
	89
1da177e4 LT	90	kfree_skb(skb);
	91	sch->qstats.drops++;
	92	return NET_XMIT_DROP;
	93	}
	94
1da177e4 LT	95	static struct sk_buff *
	96	teql_dequeue(struct Qdisc* sch)
	97	{
	98	struct teql_sched_data *dat = qdisc_priv(sch);
b0e1e646	99	struct netdev_queue *dat_queue;
1da177e4 LT	100	struct sk_buff *skb;
	101
	102	skb = __skb_dequeue(&dat->q);
e8a0464c	103	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
1da177e4	104	if (skb == NULL) {
b0e1e646	105	struct net_device *m = qdisc_dev(dat_queue->qdisc);
1da177e4 LT	106	if (m) {
	107	dat->m->slaves = sch;
	108	netif_wake_queue(m);
	109	}
	110	}
b0e1e646	111	sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
1da177e4 LT	112	return skb;
	113	}
	114
8e3af978 JP	115	static struct sk_buff *
	116	teql_peek(struct Qdisc* sch)
	117	{
	118	/* teql is meant to be used as root qdisc */
	119	return NULL;
	120	}
	121
1da177e4 LT	122	static __inline__ void
	123	teql_neigh_release(struct neighbour *n)
	124	{
	125	if (n)
	126	neigh_release(n);
	127	}
	128
	129	static void
	130	teql_reset(struct Qdisc* sch)
	131	{
	132	struct teql_sched_data *dat = qdisc_priv(sch);
	133
	134	skb_queue_purge(&dat->q);
	135	sch->q.qlen = 0;
	136	teql_neigh_release(xchg(&dat->ncache, NULL));
	137	}
	138
	139	static void
	140	teql_destroy(struct Qdisc* sch)
	141	{
	142	struct Qdisc q, prev;
	143	struct teql_sched_data *dat = qdisc_priv(sch);
	144	struct teql_master *master = dat->m;
	145
	146	if ((prev = master->slaves) != NULL) {
	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);
	167	teql_neigh_release(xchg(&dat->ncache, NULL));
	168	break;
	169	}
10297b99	170
1da177e4 LT	171	} while ((prev = q) != master->slaves);
	172	}
	173	}
	174
1e90474c	175	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt)
1da177e4	176	{
5ce2d488	177	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	178	struct teql_master m = (struct teql_master)sch->ops;
	179	struct teql_sched_data *q = qdisc_priv(sch);
	180
	181	if (dev->hard_header_len > m->dev->hard_header_len)
	182	return -EINVAL;
	183
	184	if (m->dev == dev)
	185	return -ELOOP;
	186
	187	q->m = m;
	188
	189	skb_queue_head_init(&q->q);
	190
	191	if (m->slaves) {
	192	if (m->dev->flags & IFF_UP) {
f64f9e71 JP	193	if ((m->dev->flags & IFF_POINTOPOINT &&
	194	!(dev->flags & IFF_POINTOPOINT)) \|\|
	195	(m->dev->flags & IFF_BROADCAST &&
	196	!(dev->flags & IFF_BROADCAST)) \|\|
	197	(m->dev->flags & IFF_MULTICAST &&
	198	!(dev->flags & IFF_MULTICAST)) \|\|
	199	dev->mtu < m->dev->mtu)
1da177e4 LT	200	return -EINVAL;
	201	} else {
	202	if (!(dev->flags&IFF_POINTOPOINT))
	203	m->dev->flags &= ~IFF_POINTOPOINT;
	204	if (!(dev->flags&IFF_BROADCAST))
	205	m->dev->flags &= ~IFF_BROADCAST;
	206	if (!(dev->flags&IFF_MULTICAST))
	207	m->dev->flags &= ~IFF_MULTICAST;
	208	if (dev->mtu < m->dev->mtu)
	209	m->dev->mtu = dev->mtu;
	210	}
	211	q->next = NEXT_SLAVE(m->slaves);
	212	NEXT_SLAVE(m->slaves) = sch;
	213	} else {
	214	q->next = sch;
	215	m->slaves = sch;
	216	m->dev->mtu = dev->mtu;
	217	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	218	}
	219	return 0;
	220	}
	221
1da177e4 LT	222
	223	static int
	224	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res, struct net_device *dev)
	225	{
e8a0464c DM	226	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
e8a0464c DM	227	struct teql_sched_data *q = qdisc_priv(dev_queue->qdisc);
adf30907	228	struct neighbour *mn = skb_dst(skb)->neighbour;
1da177e4 LT	229	struct neighbour *n = q->ncache;
	230
	231	if (mn->tbl == NULL)
	232	return -EINVAL;
	233	if (n && n->tbl == mn->tbl &&
	234	memcmp(n->primary_key, mn->primary_key, mn->tbl->key_len) == 0) {
	235	atomic_inc(&n->refcnt);
	236	} else {
	237	n = __neigh_lookup_errno(mn->tbl, mn->primary_key, dev);
	238	if (IS_ERR(n))
	239	return PTR_ERR(n);
	240	}
	241	if (neigh_event_send(n, skb_res) == 0) {
	242	int err;
0c4e8581	243
1da177e4	244	read_lock(&n->lock);
0c4e8581 SH	245	err = dev_hard_header(skb, dev, ntohs(skb->protocol),
0c4e8581 SH	246	n->ha, NULL, skb->len);
1da177e4	247	read_unlock(&n->lock);
0c4e8581	248
1da177e4 LT	249	if (err < 0) {
	250	neigh_release(n);
	251	return -EINVAL;
	252	}
	253	teql_neigh_release(xchg(&q->ncache, n));
	254	return 0;
	255	}
	256	neigh_release(n);
	257	return (skb_res == NULL) ? -EAGAIN : 1;
	258	}
	259
3b04ddde SH	260	static inline int teql_resolve(struct sk_buff *skb,
3b04ddde SH	261	struct sk_buff skb_res, struct net_device dev)
1da177e4	262	{
e8a0464c DM	263	struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
e8a0464c DM	264	if (txq->qdisc == &noop_qdisc)
4f9f8311 EP	265	return -ENODEV;
4f9f8311 EP	266
3b04ddde	267	if (dev->header_ops == NULL \|\|
adf30907 ED	268	skb_dst(skb) == NULL \|\|
adf30907 ED	269	skb_dst(skb)->neighbour == NULL)
1da177e4 LT	270	return 0;
	271	return __teql_resolve(skb, skb_res, dev);
	272	}
	273
6fef4c0c	274	static netdev_tx_t teql_master_xmit(struct sk_buff skb, struct net_device dev)
1da177e4	275	{
2941a486	276	struct teql_master *master = netdev_priv(dev);
ab35cd4b	277	struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
1da177e4 LT	278	struct Qdisc start, q;
	279	int busy;
	280	int nores;
4e3ab47a	281	int subq = skb_get_queue_mapping(skb);
1da177e4 LT	282	struct sk_buff *skb_res = NULL;
	283
	284	start = master->slaves;
	285
	286	restart:
	287	nores = 0;
	288	busy = 0;
	289
	290	if ((q = start) == NULL)
	291	goto drop;
	292
	293	do {
5ce2d488	294	struct net_device *slave = qdisc_dev(q);
61294e2e SH	295	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
61294e2e SH	296	const struct net_device_ops *slave_ops = slave->netdev_ops;
10297b99	297
e8a0464c	298	if (slave_txq->qdisc_sleeping != q)
1da177e4	299	continue;
fd2ea0a7	300	if (__netif_subqueue_stopped(slave, subq) \|\|
f25f4e44	301	!netif_running(slave)) {
1da177e4 LT	302	busy = 1;
	303	continue;
	304	}
	305
	306	switch (teql_resolve(skb, skb_res, slave)) {
	307	case 0:
c3f26a26	308	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	309	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	310
c3f26a26 DM	311	if (!netif_tx_queue_stopped(slave_txq) &&
c3f26a26 DM	312	!netif_tx_queue_frozen(slave_txq) &&
6fef4c0c	313	slave_ops->ndo_start_xmit(skb, slave) == NETDEV_TX_OK) {
08baf561	314	txq_trans_update(slave_txq);
c3f26a26	315	__netif_tx_unlock(slave_txq);
1da177e4 LT	316	master->slaves = NEXT_SLAVE(q);
1da177e4 LT	317	netif_wake_queue(dev);
ab35cd4b ED	318	txq->tx_packets++;
ab35cd4b ED	319	txq->tx_bytes += length;
6ed10654	320	return NETDEV_TX_OK;
1da177e4	321	}
c3f26a26	322	__netif_tx_unlock(slave_txq);
1da177e4 LT	323	}
	324	if (netif_queue_stopped(dev))
	325	busy = 1;
	326	break;
	327	case 1:
	328	master->slaves = NEXT_SLAVE(q);
6ed10654	329	return NETDEV_TX_OK;
1da177e4 LT	330	default:
	331	nores = 1;
	332	break;
	333	}
bbe735e4	334	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	335	} while ((q = NEXT_SLAVE(q)) != start);
	336
	337	if (nores && skb_res == NULL) {
	338	skb_res = skb;
	339	goto restart;
	340	}
	341
	342	if (busy) {
	343	netif_stop_queue(dev);
5b548140	344	return NETDEV_TX_BUSY;
1da177e4	345	}
ab35cd4b	346	dev->stats.tx_errors++;
1da177e4 LT	347
1da177e4 LT	348	drop:
ab35cd4b	349	txq->tx_dropped++;
1da177e4	350	dev_kfree_skb(skb);
6ed10654	351	return NETDEV_TX_OK;
1da177e4 LT	352	}
	353
	354	static int teql_master_open(struct net_device *dev)
	355	{
	356	struct Qdisc * q;
2941a486	357	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	358	int mtu = 0xFFFE;
	359	unsigned flags = IFF_NOARP\|IFF_MULTICAST;
	360
	361	if (m->slaves == NULL)
	362	return -EUNATCH;
	363
	364	flags = FMASK;
	365
	366	q = m->slaves;
	367	do {
5ce2d488	368	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	369
	370	if (slave == NULL)
	371	return -EUNATCH;
	372
	373	if (slave->mtu < mtu)
	374	mtu = slave->mtu;
	375	if (slave->hard_header_len > LL_MAX_HEADER)
	376	return -EINVAL;
	377
	378	/* If all the slaves are BROADCAST, master is BROADCAST
	379	If all the slaves are PtP, master is PtP
	380	Otherwise, master is NBMA.
	381	*/
	382	if (!(slave->flags&IFF_POINTOPOINT))
	383	flags &= ~IFF_POINTOPOINT;
	384	if (!(slave->flags&IFF_BROADCAST))
	385	flags &= ~IFF_BROADCAST;
	386	if (!(slave->flags&IFF_MULTICAST))
	387	flags &= ~IFF_MULTICAST;
	388	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	389
	390	m->dev->mtu = mtu;
	391	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	392	netif_start_queue(m->dev);
	393	return 0;
	394	}
	395
	396	static int teql_master_close(struct net_device *dev)
	397	{
	398	netif_stop_queue(dev);
	399	return 0;
	400	}
	401
1da177e4 LT	402	static int teql_master_mtu(struct net_device *dev, int new_mtu)
1da177e4 LT	403	{
2941a486	404	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	405	struct Qdisc *q;
	406
	407	if (new_mtu < 68)
	408	return -EINVAL;
	409
	410	q = m->slaves;
	411	if (q) {
	412	do {
5ce2d488	413	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4 LT	414	return -EINVAL;
	415	} while ((q=NEXT_SLAVE(q)) != m->slaves);
	416	}
	417
	418	dev->mtu = new_mtu;
	419	return 0;
	420	}
	421
61294e2e SH	422	static const struct net_device_ops teql_netdev_ops = {
	423	.ndo_open = teql_master_open,
	424	.ndo_stop = teql_master_close,
	425	.ndo_start_xmit = teql_master_xmit,
61294e2e SH	426	.ndo_change_mtu = teql_master_mtu,
	427	};
	428
1da177e4 LT	429	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	430	{
2941a486	431	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	432	struct Qdisc_ops *ops = &master->qops;
	433
	434	master->dev = dev;
	435	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	436
1da177e4 LT	437	ops->enqueue = teql_enqueue;
1da177e4 LT	438	ops->dequeue = teql_dequeue;
8e3af978	439	ops->peek = teql_peek;
1da177e4 LT	440	ops->init = teql_qdisc_init;
	441	ops->reset = teql_reset;
	442	ops->destroy = teql_destroy;
	443	ops->owner = THIS_MODULE;
	444
61294e2e	445	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	446	dev->type = ARPHRD_VOID;
	447	dev->mtu = 1500;
	448	dev->tx_queue_len = 100;
	449	dev->flags = IFF_NOARP;
	450	dev->hard_header_len = LL_MAX_HEADER;
1da177e4 LT	451	}
	452
	453	static LIST_HEAD(master_dev_list);
	454	static int max_equalizers = 1;
	455	module_param(max_equalizers, int, 0);
	456	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	457
	458	static int __init teql_init(void)
	459	{
	460	int i;
	461	int err = -ENODEV;
	462
	463	for (i = 0; i < max_equalizers; i++) {
	464	struct net_device *dev;
	465	struct teql_master *master;
	466
	467	dev = alloc_netdev(sizeof(struct teql_master),
	468	"teql%d", teql_master_setup);
	469	if (!dev) {
	470	err = -ENOMEM;
	471	break;
	472	}
	473
	474	if ((err = register_netdev(dev))) {
	475	free_netdev(dev);
	476	break;
	477	}
	478
2941a486	479	master = netdev_priv(dev);
1da177e4 LT	480
	481	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	482	err = register_qdisc(&master->qops);
	483
	484	if (err) {
	485	unregister_netdev(dev);
	486	free_netdev(dev);
	487	break;
	488	}
	489
	490	list_add_tail(&master->master_list, &master_dev_list);
	491	}
	492	return i ? 0 : err;
	493	}
	494
10297b99	495	static void __exit teql_exit(void)
1da177e4 LT	496	{
	497	struct teql_master master, nxt;
	498
	499	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	500
	501	list_del(&master->master_list);
	502
	503	unregister_qdisc(&master->qops);
	504	unregister_netdev(master->dev);
	505	free_netdev(master->dev);
	506	}
	507	}
	508
	509	module_init(teql_init);
	510	module_exit(teql_exit);
	511
	512	MODULE_LICENSE("GPL");