[people/ms/linux.git] / net / sched / sch_teql.c

// SPDX-License-Identifier: GPL-2.0-or-later
/* net/sched/sch_teql.c	"True" (or "trivial") link equalizer.
 *
 * 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;

	if (!master)
		return;

	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(skb_protocol(skb, false)),
				      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
2874c5fd	1	// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4	2	/* net/sched/sch_teql.c "True" (or "trivial") link equalizer.
1da177e4 LT	3	*
	4	* Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
	5	*/
	6
	7	#include <linux/module.h>
1da177e4 LT	8	#include <linux/types.h>
1da177e4 LT	9	#include <linux/kernel.h>
5a0e3ad6	10	#include <linux/slab.h>
1da177e4	11	#include <linux/string.h>
1da177e4	12	#include <linux/errno.h>
14c85021	13	#include <linux/if_arp.h>
1da177e4	14	#include <linux/netdevice.h>
1da177e4	15	#include <linux/init.h>
1da177e4 LT	16	#include <linux/skbuff.h>
1da177e4 LT	17	#include <linux/moduleparam.h>
0ba48053 PM	18	#include <net/dst.h>
0ba48053 PM	19	#include <net/neighbour.h>
1da177e4 LT	20	#include <net/pkt_sched.h>
	21
	22	/*
	23	How to setup it.
	24	----------------
	25
	26	After loading this module you will find a new device teqlN
	27	and new qdisc with the same name. To join a slave to the equalizer
	28	you should just set this qdisc on a device f.e.
	29
	30	# tc qdisc add dev eth0 root teql0
	31	# tc qdisc add dev eth1 root teql0
	32
	33	That's all. Full PnP 8)
	34
	35	Applicability.
	36	--------------
	37
	38	1. Slave devices MUST be active devices, i.e., they must raise the tbusy
	39	signal and generate EOI events. If you want to equalize virtual devices
	40	like tunnels, use a normal eql device.
	41	2. This device puts no limitations on physical slave characteristics
	42	f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
	43	Certainly, large difference in link speeds will make the resulting
	44	eqalized link unusable, because of huge packet reordering.
	45	I estimate an upper useful difference as ~10 times.
	46	3. If the slave requires address resolution, only protocols using
	47	neighbour cache (IPv4/IPv6) will work over the equalized link.
	48	Other protocols are still allowed to use the slave device directly,
	49	which will not break load balancing, though native slave
	50	traffic will have the highest priority. */
	51
cc7ec456	52	struct teql_master {
1da177e4 LT	53	struct Qdisc_ops qops;
	54	struct net_device *dev;
	55	struct Qdisc *slaves;
	56	struct list_head master_list;
1ac9ad13 ED	57	unsigned long tx_bytes;
	58	unsigned long tx_packets;
	59	unsigned long tx_errors;
	60	unsigned long tx_dropped;
1da177e4 LT	61	};
1da177e4 LT	62
cc7ec456	63	struct teql_sched_data {
1da177e4 LT	64	struct Qdisc *next;
1da177e4 LT	65	struct teql_master *m;
1da177e4 LT	66	struct sk_buff_head q;
	67	};
	68
cc7ec456	69	#define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
1da177e4	70
cc7ec456	71	#define FMASK (IFF_BROADCAST \| IFF_POINTOPOINT)
1da177e4 LT	72
	73	/* "teql" qdisc routines /
	74
	75	static int
520ac30f	76	teql_enqueue(struct sk_buff skb, struct Qdisc sch, struct sk_buff **to_free)
1da177e4	77	{
5ce2d488	78	struct net_device *dev = qdisc_dev(sch);
1da177e4 LT	79	struct teql_sched_data *q = qdisc_priv(sch);
1da177e4 LT	80
4cd8c9e8 KK	81	if (q->q.qlen < dev->tx_queue_len) {
4cd8c9e8 KK	82	__skb_queue_tail(&q->q, skb);
9871e50e	83	return NET_XMIT_SUCCESS;
1da177e4 LT	84	}
1da177e4 LT	85
520ac30f	86	return qdisc_drop(skb, sch, to_free);
1da177e4 LT	87	}
1da177e4 LT	88
1da177e4	89	static struct sk_buff *
cc7ec456	90	teql_dequeue(struct Qdisc *sch)
1da177e4 LT	91	{
1da177e4 LT	92	struct teql_sched_data *dat = qdisc_priv(sch);
b0e1e646	93	struct netdev_queue *dat_queue;
1da177e4	94	struct sk_buff *skb;
46e5da40	95	struct Qdisc *q;
1da177e4 LT	96
1da177e4 LT	97	skb = __skb_dequeue(&dat->q);
e8a0464c	98	dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
46e5da40 JF	99	q = rcu_dereference_bh(dat_queue->qdisc);
46e5da40 JF	100
1da177e4	101	if (skb == NULL) {
46e5da40	102	struct net_device *m = qdisc_dev(q);
1da177e4 LT	103	if (m) {
	104	dat->m->slaves = sch;
	105	netif_wake_queue(m);
	106	}
9190b3b3 ED	107	} else {
9190b3b3 ED	108	qdisc_bstats_update(sch, skb);
1da177e4	109	}
46e5da40	110	sch->q.qlen = dat->q.qlen + q->q.qlen;
1da177e4 LT	111	return skb;
	112	}
	113
8e3af978	114	static struct sk_buff *
cc7ec456	115	teql_peek(struct Qdisc *sch)
8e3af978 JP	116	{
	117	/* teql is meant to be used as root qdisc */
	118	return NULL;
	119	}
	120
1da177e4	121	static void
cc7ec456	122	teql_reset(struct Qdisc *sch)
1da177e4 LT	123	{
	124	struct teql_sched_data *dat = qdisc_priv(sch);
	125
	126	skb_queue_purge(&dat->q);
	127	sch->q.qlen = 0;
1da177e4 LT	128	}
	129
	130	static void
cc7ec456	131	teql_destroy(struct Qdisc *sch)
1da177e4 LT	132	{
	133	struct Qdisc q, prev;
	134	struct teql_sched_data *dat = qdisc_priv(sch);
	135	struct teql_master *master = dat->m;
	136
1ffbc7ea PT	137	if (!master)
	138	return;
	139
cc7ec456 ED	140	prev = master->slaves;
cc7ec456 ED	141	if (prev) {
1da177e4 LT	142	do {
	143	q = NEXT_SLAVE(prev);
	144	if (q == sch) {
	145	NEXT_SLAVE(prev) = NEXT_SLAVE(q);
	146	if (q == master->slaves) {
	147	master->slaves = NEXT_SLAVE(q);
	148	if (q == master->slaves) {
e8a0464c	149	struct netdev_queue *txq;
83874000	150	spinlock_t *root_lock;
e8a0464c DM	151
e8a0464c DM	152	txq = netdev_get_tx_queue(master->dev, 0);
1da177e4	153	master->slaves = NULL;
83874000	154
46e5da40	155	root_lock = qdisc_root_sleeping_lock(rtnl_dereference(txq->qdisc));
83874000	156	spin_lock_bh(root_lock);
46e5da40	157	qdisc_reset(rtnl_dereference(txq->qdisc));
83874000	158	spin_unlock_bh(root_lock);
1da177e4 LT	159	}
	160	}
	161	skb_queue_purge(&dat->q);
1da177e4 LT	162	break;
1da177e4 LT	163	}
10297b99	164
1da177e4 LT	165	} while ((prev = q) != master->slaves);
	166	}
	167	}
	168
e63d7dfd AA	169	static int teql_qdisc_init(struct Qdisc sch, struct nlattr opt,
e63d7dfd AA	170	struct netlink_ext_ack *extack)
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);
d7bf2ebe	245	err = dev_hard_header(skb, dev, ntohs(skb_protocol(skb, false)),
d8b9605d	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
bc1f4470	404	static void teql_master_stats64(struct net_device *dev,
bc1f4470	405	struct rtnl_link_stats64 *stats)
1ac9ad13 ED	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;
1ac9ad13 ED	413	}
1ac9ad13 ED	414
1da177e4 LT	415	static int teql_master_mtu(struct net_device *dev, int new_mtu)
1da177e4 LT	416	{
2941a486	417	struct teql_master *m = netdev_priv(dev);
1da177e4 LT	418	struct Qdisc *q;
1da177e4 LT	419
1da177e4 LT	420	q = m->slaves;
	421	if (q) {
	422	do {
5ce2d488	423	if (new_mtu > qdisc_dev(q)->mtu)
1da177e4	424	return -EINVAL;
cc7ec456	425	} while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4 LT	426	}
	427
	428	dev->mtu = new_mtu;
	429	return 0;
	430	}
	431
61294e2e SH	432	static const struct net_device_ops teql_netdev_ops = {
	433	.ndo_open = teql_master_open,
	434	.ndo_stop = teql_master_close,
	435	.ndo_start_xmit = teql_master_xmit,
1ac9ad13	436	.ndo_get_stats64 = teql_master_stats64,
61294e2e SH	437	.ndo_change_mtu = teql_master_mtu,
	438	};
	439
1da177e4 LT	440	static __init void teql_master_setup(struct net_device *dev)
1da177e4 LT	441	{
2941a486	442	struct teql_master *master = netdev_priv(dev);
1da177e4 LT	443	struct Qdisc_ops *ops = &master->qops;
	444
	445	master->dev = dev;
	446	ops->priv_size = sizeof(struct teql_sched_data);
10297b99	447
1da177e4 LT	448	ops->enqueue = teql_enqueue;
1da177e4 LT	449	ops->dequeue = teql_dequeue;
8e3af978	450	ops->peek = teql_peek;
1da177e4 LT	451	ops->init = teql_qdisc_init;
	452	ops->reset = teql_reset;
	453	ops->destroy = teql_destroy;
	454	ops->owner = THIS_MODULE;
	455
61294e2e	456	dev->netdev_ops = &teql_netdev_ops;
1da177e4 LT	457	dev->type = ARPHRD_VOID;
1da177e4 LT	458	dev->mtu = 1500;
91572088 JW	459	dev->min_mtu = 68;
91572088 JW	460	dev->max_mtu = 65535;
1da177e4 LT	461	dev->tx_queue_len = 100;
	462	dev->flags = IFF_NOARP;
	463	dev->hard_header_len = LL_MAX_HEADER;
02875878	464	netif_keep_dst(dev);
1da177e4 LT	465	}
	466
	467	static LIST_HEAD(master_dev_list);
	468	static int max_equalizers = 1;
	469	module_param(max_equalizers, int, 0);
	470	MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
	471
	472	static int __init teql_init(void)
	473	{
	474	int i;
	475	int err = -ENODEV;
	476
	477	for (i = 0; i < max_equalizers; i++) {
	478	struct net_device *dev;
	479	struct teql_master *master;
	480
c835a677 TG	481	dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
c835a677 TG	482	NET_NAME_UNKNOWN, teql_master_setup);
1da177e4 LT	483	if (!dev) {
	484	err = -ENOMEM;
	485	break;
	486	}
	487
	488	if ((err = register_netdev(dev))) {
	489	free_netdev(dev);
	490	break;
	491	}
	492
2941a486	493	master = netdev_priv(dev);
1da177e4 LT	494
	495	strlcpy(master->qops.id, dev->name, IFNAMSIZ);
	496	err = register_qdisc(&master->qops);
	497
	498	if (err) {
	499	unregister_netdev(dev);
	500	free_netdev(dev);
	501	break;
	502	}
	503
	504	list_add_tail(&master->master_list, &master_dev_list);
	505	}
	506	return i ? 0 : err;
	507	}
	508
10297b99	509	static void __exit teql_exit(void)
1da177e4 LT	510	{
	511	struct teql_master master, nxt;
	512
	513	list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
	514
	515	list_del(&master->master_list);
	516
	517	unregister_qdisc(&master->qops);
	518	unregister_netdev(master->dev);
	519	free_netdev(master->dev);
	520	}
	521	}
	522
	523	module_init(teql_init);
	524	module_exit(teql_exit);
	525
	526	MODULE_LICENSE("GPL");