[mirror_ubuntu-jammy-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 sk_buff **to_free)
{
	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, to_free);
}

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 netlink_ext_ack *extack)
{
	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 void 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;
}

static int teql_master_mtu(struct net_device *dev, int new_mtu)
{
	struct teql_master *m = netdev_priv(dev);
	struct Qdisc *q;

	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->min_mtu		= 68;
	dev->max_mtu		= 65535;
	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
520ac30f	80	teql_enqueue(struct sk_buff skb, struct Qdisc sch, struct sk_buff **to_free)
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
520ac30f	90	return qdisc_drop(skb, sch, to_free);
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
e63d7dfd AA	170	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt,
e63d7dfd AA	171	struct netlink_ext_ack *extack)
1da177e4	172	{
5ce2d488	173	struct net_device *dev = qdisc_dev(sch);
cc7ec456	174	struct teql_master m = (struct teql_master )sch->ops;
1da177e4 LT	175	struct teql_sched_data *q = qdisc_priv(sch);
	176
	177	if (dev->hard_header_len > m->dev->hard_header_len)
	178	return -EINVAL;
	179
	180	if (m->dev == dev)
	181	return -ELOOP;
	182
	183	q->m = m;
	184
	185	skb_queue_head_init(&q->q);
	186
	187	if (m->slaves) {
	188	if (m->dev->flags & IFF_UP) {
f64f9e71 JP	189	if ((m->dev->flags & IFF_POINTOPOINT &&
	190	!(dev->flags & IFF_POINTOPOINT)) \|\|
	191	(m->dev->flags & IFF_BROADCAST &&
	192	!(dev->flags & IFF_BROADCAST)) \|\|
	193	(m->dev->flags & IFF_MULTICAST &&
	194	!(dev->flags & IFF_MULTICAST)) \|\|
	195	dev->mtu < m->dev->mtu)
1da177e4 LT	196	return -EINVAL;
	197	} else {
	198	if (!(dev->flags&IFF_POINTOPOINT))
	199	m->dev->flags &= ~IFF_POINTOPOINT;
	200	if (!(dev->flags&IFF_BROADCAST))
	201	m->dev->flags &= ~IFF_BROADCAST;
	202	if (!(dev->flags&IFF_MULTICAST))
	203	m->dev->flags &= ~IFF_MULTICAST;
	204	if (dev->mtu < m->dev->mtu)
	205	m->dev->mtu = dev->mtu;
	206	}
	207	q->next = NEXT_SLAVE(m->slaves);
	208	NEXT_SLAVE(m->slaves) = sch;
	209	} else {
	210	q->next = sch;
	211	m->slaves = sch;
	212	m->dev->mtu = dev->mtu;
	213	m->dev->flags = (m->dev->flags&~FMASK)\|(dev->flags&FMASK);
	214	}
	215	return 0;
	216	}
	217
1da177e4 LT	218
1da177e4 LT	219	static int
f7e57044 ED	220	__teql_resolve(struct sk_buff skb, struct sk_buff skb_res,
f7e57044 ED	221	struct net_device dev, struct netdev_queue txq,
dbedbe6d	222	struct dst_entry *dst)
1da177e4	223	{
dbedbe6d DM	224	struct neighbour *n;
dbedbe6d DM	225	int err = 0;
1da177e4	226
dbedbe6d DM	227	n = dst_neigh_lookup_skb(dst, skb);
	228	if (!n)
	229	return -ENOENT;
	230
	231	if (dst->dev != dev) {
	232	struct neighbour *mn;
	233
	234	mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
	235	neigh_release(n);
	236	if (IS_ERR(mn))
	237	return PTR_ERR(mn);
	238	n = mn;
1da177e4	239	}
dbedbe6d	240
1da177e4 LT	241	if (neigh_event_send(n, skb_res) == 0) {
1da177e4 LT	242	int err;
0ed8ddf4	243	char haddr[MAX_ADDR_LEN];
0c4e8581	244
0ed8ddf4	245	neigh_ha_snapshot(haddr, n, dev);
d8b9605d JP	246	err = dev_hard_header(skb, dev, ntohs(tc_skb_protocol(skb)),
d8b9605d JP	247	haddr, NULL, skb->len);
0c4e8581	248
dbedbe6d DM	249	if (err < 0)
	250	err = -EINVAL;
	251	} else {
	252	err = (skb_res == NULL) ? -EAGAIN : 1;
1da177e4 LT	253	}
1da177e4 LT	254	neigh_release(n);
dbedbe6d	255	return err;
1da177e4 LT	256	}
1da177e4 LT	257
3b04ddde	258	static inline int teql_resolve(struct sk_buff *skb,
f7e57044 ED	259	struct sk_buff *skb_res,
	260	struct net_device *dev,
	261	struct netdev_queue *txq)
1da177e4	262	{
f7e57044	263	struct dst_entry *dst = skb_dst(skb);
f7e57044 ED	264	int res;
f7e57044 ED	265
46e5da40	266	if (rcu_access_pointer(txq->qdisc) == &noop_qdisc)
4f9f8311 EP	267	return -ENODEV;
4f9f8311 EP	268
f7e57044	269	if (!dev->header_ops \|\| !dst)
1da177e4	270	return 0;
f7e57044 ED	271
f7e57044 ED	272	rcu_read_lock();
dbedbe6d	273	res = __teql_resolve(skb, skb_res, dev, txq, dst);
f7e57044 ED	274	rcu_read_unlock();
	275
	276	return res;
1da177e4 LT	277	}
1da177e4 LT	278
6fef4c0c	279	static netdev_tx_t teql_master_xmit(struct sk_buff skb, struct net_device dev)
1da177e4	280	{
2941a486	281	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	282	struct Qdisc start, q;
	283	int busy;
	284	int nores;
4e3ab47a	285	int subq = skb_get_queue_mapping(skb);
1da177e4 LT	286	struct sk_buff *skb_res = NULL;
	287
	288	start = master->slaves;
	289
	290	restart:
	291	nores = 0;
	292	busy = 0;
	293
cc7ec456 ED	294	q = start;
cc7ec456 ED	295	if (!q)
1da177e4 LT	296	goto drop;
	297
	298	do {
5ce2d488	299	struct net_device *slave = qdisc_dev(q);
61294e2e	300	struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
10297b99	301
e8a0464c	302	if (slave_txq->qdisc_sleeping != q)
1da177e4	303	continue;
73466498	304	if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) \|\|
f25f4e44	305	!netif_running(slave)) {
1da177e4 LT	306	busy = 1;
	307	continue;
	308	}
	309
f7e57044	310	switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
1da177e4	311	case 0:
c3f26a26	312	if (__netif_tx_trylock(slave_txq)) {
c0f84d0d ED	313	unsigned int length = qdisc_pkt_len(skb);
c0f84d0d ED	314
73466498	315	if (!netif_xmit_frozen_or_stopped(slave_txq) &&
fa2dbdc2 DM	316	netdev_start_xmit(skb, slave, slave_txq, false) ==
fa2dbdc2 DM	317	NETDEV_TX_OK) {
c3f26a26	318	__netif_tx_unlock(slave_txq);
1da177e4 LT	319	master->slaves = NEXT_SLAVE(q);
1da177e4 LT	320	netif_wake_queue(dev);
1ac9ad13 ED	321	master->tx_packets++;
1ac9ad13 ED	322	master->tx_bytes += length;
6ed10654	323	return NETDEV_TX_OK;
1da177e4	324	}
c3f26a26	325	__netif_tx_unlock(slave_txq);
1da177e4	326	}
73466498	327	if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
1da177e4 LT	328	busy = 1;
	329	break;
	330	case 1:
	331	master->slaves = NEXT_SLAVE(q);
6ed10654	332	return NETDEV_TX_OK;
1da177e4 LT	333	default:
	334	nores = 1;
	335	break;
	336	}
bbe735e4	337	__skb_pull(skb, skb_network_offset(skb));
1da177e4 LT	338	} while ((q = NEXT_SLAVE(q)) != start);
	339
	340	if (nores && skb_res == NULL) {
	341	skb_res = skb;
	342	goto restart;
	343	}
	344
	345	if (busy) {
	346	netif_stop_queue(dev);
5b548140	347	return NETDEV_TX_BUSY;
1da177e4	348	}
1ac9ad13	349	master->tx_errors++;
1da177e4 LT	350
1da177e4 LT	351	drop:
1ac9ad13	352	master->tx_dropped++;
1da177e4	353	dev_kfree_skb(skb);
6ed10654	354	return NETDEV_TX_OK;
1da177e4 LT	355	}
	356
	357	static int teql_master_open(struct net_device *dev)
	358	{
cc7ec456	359	struct Qdisc *q;
2941a486	360	struct teql_master *m = netdev_priv(dev);
1da177e4	361	int mtu = 0xFFFE;
cc7ec456	362	unsigned int flags = IFF_NOARP \| IFF_MULTICAST;
1da177e4 LT	363
	364	if (m->slaves == NULL)
	365	return -EUNATCH;
	366
	367	flags = FMASK;
	368
	369	q = m->slaves;
	370	do {
5ce2d488	371	struct net_device *slave = qdisc_dev(q);
1da177e4 LT	372
	373	if (slave == NULL)
	374	return -EUNATCH;
	375
	376	if (slave->mtu < mtu)
	377	mtu = slave->mtu;
	378	if (slave->hard_header_len > LL_MAX_HEADER)
	379	return -EINVAL;
	380
	381	/* If all the slaves are BROADCAST, master is BROADCAST
	382	If all the slaves are PtP, master is PtP
	383	Otherwise, master is NBMA.
	384	*/
	385	if (!(slave->flags&IFF_POINTOPOINT))
	386	flags &= ~IFF_POINTOPOINT;
	387	if (!(slave->flags&IFF_BROADCAST))
	388	flags &= ~IFF_BROADCAST;
	389	if (!(slave->flags&IFF_MULTICAST))
	390	flags &= ~IFF_MULTICAST;
	391	} while ((q = NEXT_SLAVE(q)) != m->slaves);
	392
	393	m->dev->mtu = mtu;
	394	m->dev->flags = (m->dev->flags&~FMASK) \| flags;
	395	netif_start_queue(m->dev);
	396	return 0;
	397	}
	398
	399	static int teql_master_close(struct net_device *dev)
	400	{
	401	netif_stop_queue(dev);
	402	return 0;
	403	}
	404
bc1f4470	405	static void teql_master_stats64(struct net_device *dev,
bc1f4470	406	struct rtnl_link_stats64 *stats)
1ac9ad13 ED	407	{
	408	struct teql_master *m = netdev_priv(dev);
	409
	410	stats->tx_packets = m->tx_packets;
	411	stats->tx_bytes = m->tx_bytes;
	412	stats->tx_errors = m->tx_errors;
	413	stats->tx_dropped = m->tx_dropped;
1ac9ad13 ED	414	}
1ac9ad13 ED	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;
1da177e4 LT	420
1da177e4 LT	421	q = m->slaves;
	422	if (q) {
	423	do {
5ce2d488	424	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4	425	return -EINVAL;
cc7ec456	426	} while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4 LT	427	}
	428
	429	dev->mtu = new_mtu;
	430	return 0;
	431	}
	432
61294e2e SH	433	static const struct net_device_ops teql_netdev_ops = {
	434	.ndo_open = teql_master_open,
	435	.ndo_stop = teql_master_close,
	436	.ndo_start_xmit = teql_master_xmit,
1ac9ad13	437	.ndo_get_stats64 = teql_master_stats64,
61294e2e SH	438	.ndo_change_mtu = teql_master_mtu,
	439	};
	440
1da177e4 LT	441	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	442	{
2941a486	443	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	444	struct Qdisc_ops *ops = &master->qops;
	445
	446	master->dev = dev;
	447	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	448
1da177e4 LT	449	ops->enqueue = teql_enqueue;
1da177e4 LT	450	ops->dequeue = teql_dequeue;
8e3af978	451	ops->peek = teql_peek;
1da177e4 LT	452	ops->init = teql_qdisc_init;
	453	ops->reset = teql_reset;
	454	ops->destroy = teql_destroy;
	455	ops->owner = THIS_MODULE;
	456
61294e2e	457	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	458	dev->type = ARPHRD_VOID;
1da177e4 LT	459	dev->mtu = 1500;
91572088 JW	460	dev->min_mtu = 68;
91572088 JW	461	dev->max_mtu = 65535;
1da177e4 LT	462	dev->tx_queue_len = 100;
	463	dev->flags = IFF_NOARP;
	464	dev->hard_header_len = LL_MAX_HEADER;
02875878	465	netif_keep_dst(dev);
1da177e4 LT	466	}
	467
	468	static LIST_HEAD(master_dev_list);
	469	static int max_equalizers = 1;
	470	module_param(max_equalizers, int, 0);
	471	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	472
	473	static int __init teql_init(void)
	474	{
	475	int i;
	476	int err = -ENODEV;
	477
	478	for (i = 0; i < max_equalizers; i++) {
	479	struct net_device *dev;
	480	struct teql_master *master;
	481
c835a677 TG	482	dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
c835a677 TG	483	NET_NAME_UNKNOWN, teql_master_setup);
1da177e4 LT	484	if (!dev) {
	485	err = -ENOMEM;
	486	break;
	487	}
	488
	489	if ((err = register_netdev(dev))) {
	490	free_netdev(dev);
	491	break;
	492	}
	493
2941a486	494	master = netdev_priv(dev);
1da177e4 LT	495
	496	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	497	err = register_qdisc(&master->qops);
	498
	499	if (err) {
	500	unregister_netdev(dev);
	501	free_netdev(dev);
	502	break;
	503	}
	504
	505	list_add_tail(&master->master_list, &master_dev_list);
	506	}
	507	return i ? 0 : err;
	508	}
	509
10297b99	510	static void __exit teql_exit(void)
1da177e4 LT	511	{
	512	struct teql_master master, nxt;
	513
	514	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	515
	516	list_del(&master->master_list);
	517
	518	unregister_qdisc(&master->qops);
	519	unregister_netdev(master->dev);
	520	free_netdev(master->dev);
	521	}
	522	}
	523
	524	module_init(teql_init);
	525	module_exit(teql_exit);
	526
	527	MODULE_LICENSE("GPL");