linux/drivers/net/eql.c

612 lines
15 KiB
C
Raw Normal View History

/*
* Equalizer Load-balancer for serial network interfaces.
*
* (c) Copyright 1995 Simon "Guru Aleph-Null" Janes
* NCM: Network and Communications Management, Inc.
*
* (c) Copyright 2002 David S. Miller (davem@redhat.com)
*
* This software may be used and distributed according to the terms
* of the GNU General Public License, incorporated herein by reference.
*
* The author may be reached as simon@ncm.com, or C/O
* NCM
* Attn: Simon Janes
* 6803 Whittier Ave
* McLean VA 22101
* Phone: 1-703-847-0040 ext 103
*/
/*
* Sources:
* skeleton.c by Donald Becker.
* Inspirations:
* The Harried and Overworked Alan Cox
* Conspiracies:
* The Alan Cox and Mike McLagan plot to get someone else to do the code,
* which turned out to be me.
*/
/*
* $Log: eql.c,v $
* Revision 1.2 1996/04/11 17:51:52 guru
* Added one-line eql_remove_slave patch.
*
* Revision 1.1 1996/04/11 17:44:17 guru
* Initial revision
*
* Revision 3.13 1996/01/21 15:17:18 alan
* tx_queue_len changes.
* reformatted.
*
* Revision 3.12 1995/03/22 21:07:51 anarchy
* Added capable() checks on configuration.
* Moved header file.
*
* Revision 3.11 1995/01/19 23:14:31 guru
* slave_load = (ULONG_MAX - (ULONG_MAX / 2)) -
* (priority_Bps) + bytes_queued * 8;
*
* Revision 3.10 1995/01/19 23:07:53 guru
* back to
* slave_load = (ULONG_MAX - (ULONG_MAX / 2)) -
* (priority_Bps) + bytes_queued;
*
* Revision 3.9 1995/01/19 22:38:20 guru
* slave_load = (ULONG_MAX - (ULONG_MAX / 2)) -
* (priority_Bps) + bytes_queued * 4;
*
* Revision 3.8 1995/01/19 22:30:55 guru
* slave_load = (ULONG_MAX - (ULONG_MAX / 2)) -
* (priority_Bps) + bytes_queued * 2;
*
* Revision 3.7 1995/01/19 21:52:35 guru
* printk's trimmed out.
*
* Revision 3.6 1995/01/19 21:49:56 guru
* This is working pretty well. I gained 1 K/s in speed.. now it's just
* robustness and printk's to be diked out.
*
* Revision 3.5 1995/01/18 22:29:59 guru
* still crashes the kernel when the lock_wait thing is woken up.
*
* Revision 3.4 1995/01/18 21:59:47 guru
* Broken set-bit locking snapshot
*
* Revision 3.3 1995/01/17 22:09:18 guru
* infinite sleep in a lock somewhere..
*
* Revision 3.2 1995/01/15 16:46:06 guru
* Log trimmed of non-pertinent 1.x branch messages
*
* Revision 3.1 1995/01/15 14:41:45 guru
* New Scheduler and timer stuff...
*
* Revision 1.15 1995/01/15 14:29:02 guru
* Will make 1.14 (now 1.15) the 3.0 branch, and the 1.12 the 2.0 branch, the one
* with the dumber scheduler
*
* Revision 1.14 1995/01/15 02:37:08 guru
* shock.. the kept-new-versions could have zonked working
* stuff.. shudder
*
* Revision 1.13 1995/01/15 02:36:31 guru
* big changes
*
* scheduler was torn out and replaced with something smarter
*
* global names not prefixed with eql_ were renamed to protect
* against namespace collisions
*
* a few more abstract interfaces were added to facilitate any
* potential change of datastructure. the driver is still using
* a linked list of slaves. going to a heap would be a bit of
* an overkill.
*
* this compiles fine with no warnings.
*
* the locking mechanism and timer stuff must be written however,
* this version will not work otherwise
*
* Sorry, I had to rewrite most of this for 2.5.x -DaveM
*/
#include <linux/capability.h>
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/timer.h>
#include <linux/netdevice.h>
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 18:56:21 +00:00
#include <net/net_namespace.h>
#include <linux/if.h>
#include <linux/if_arp.h>
#include <linux/if_eql.h>
#include <asm/uaccess.h>
static int eql_open(struct net_device *dev);
static int eql_close(struct net_device *dev);
static int eql_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd);
static netdev_tx_t eql_slave_xmit(struct sk_buff *skb, struct net_device *dev);
#define eql_is_slave(dev) ((dev->flags & IFF_SLAVE) == IFF_SLAVE)
#define eql_is_master(dev) ((dev->flags & IFF_MASTER) == IFF_MASTER)
static void eql_kill_one_slave(slave_queue_t *queue, slave_t *slave);
static void eql_timer(unsigned long param)
{
equalizer_t *eql = (equalizer_t *) param;
struct list_head *this, *tmp, *head;
spin_lock_bh(&eql->queue.lock);
head = &eql->queue.all_slaves;
list_for_each_safe(this, tmp, head) {
slave_t *slave = list_entry(this, slave_t, list);
if ((slave->dev->flags & IFF_UP) == IFF_UP) {
slave->bytes_queued -= slave->priority_Bps;
if (slave->bytes_queued < 0)
slave->bytes_queued = 0;
} else {
eql_kill_one_slave(&eql->queue, slave);
}
}
spin_unlock_bh(&eql->queue.lock);
eql->timer.expires = jiffies + EQL_DEFAULT_RESCHED_IVAL;
add_timer(&eql->timer);
}
static const char version[] __initconst =
"Equalizer2002: Simon Janes (simon@ncm.com) and David S. Miller (davem@redhat.com)\n";
static const struct net_device_ops eql_netdev_ops = {
.ndo_open = eql_open,
.ndo_stop = eql_close,
.ndo_do_ioctl = eql_ioctl,
.ndo_start_xmit = eql_slave_xmit,
};
static void __init eql_setup(struct net_device *dev)
{
equalizer_t *eql = netdev_priv(dev);
init_timer(&eql->timer);
eql->timer.data = (unsigned long) eql;
eql->timer.expires = jiffies + EQL_DEFAULT_RESCHED_IVAL;
eql->timer.function = eql_timer;
spin_lock_init(&eql->queue.lock);
INIT_LIST_HEAD(&eql->queue.all_slaves);
eql->queue.master_dev = dev;
dev->netdev_ops = &eql_netdev_ops;
/*
* Now we undo some of the things that eth_setup does
* that we don't like
*/
dev->mtu = EQL_DEFAULT_MTU; /* set to 576 in if_eql.h */
dev->flags = IFF_MASTER;
dev->type = ARPHRD_SLIP;
dev->tx_queue_len = 5; /* Hands them off fast */
dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
}
static int eql_open(struct net_device *dev)
{
equalizer_t *eql = netdev_priv(dev);
/* XXX We should force this off automatically for the user. */
printk(KERN_INFO "%s: remember to turn off Van-Jacobson compression on "
"your slave devices.\n", dev->name);
BUG_ON(!list_empty(&eql->queue.all_slaves));
eql->min_slaves = 1;
eql->max_slaves = EQL_DEFAULT_MAX_SLAVES; /* 4 usually... */
add_timer(&eql->timer);
return 0;
}
static void eql_kill_one_slave(slave_queue_t *queue, slave_t *slave)
{
list_del(&slave->list);
queue->num_slaves--;
slave->dev->flags &= ~IFF_SLAVE;
dev_put(slave->dev);
kfree(slave);
}
static void eql_kill_slave_queue(slave_queue_t *queue)
{
struct list_head *head, *tmp, *this;
spin_lock_bh(&queue->lock);
head = &queue->all_slaves;
list_for_each_safe(this, tmp, head) {
slave_t *s = list_entry(this, slave_t, list);
eql_kill_one_slave(queue, s);
}
spin_unlock_bh(&queue->lock);
}
static int eql_close(struct net_device *dev)
{
equalizer_t *eql = netdev_priv(dev);
/*
* The timer has to be stopped first before we start hacking away
* at the data structure it scans every so often...
*/
del_timer_sync(&eql->timer);
eql_kill_slave_queue(&eql->queue);
return 0;
}
static int eql_enslave(struct net_device *dev, slaving_request_t __user *srq);
static int eql_emancipate(struct net_device *dev, slaving_request_t __user *srq);
static int eql_g_slave_cfg(struct net_device *dev, slave_config_t __user *sc);
static int eql_s_slave_cfg(struct net_device *dev, slave_config_t __user *sc);
static int eql_g_master_cfg(struct net_device *dev, master_config_t __user *mc);
static int eql_s_master_cfg(struct net_device *dev, master_config_t __user *mc);
static int eql_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
{
if (cmd != EQL_GETMASTRCFG && cmd != EQL_GETSLAVECFG &&
!capable(CAP_NET_ADMIN))
return -EPERM;
switch (cmd) {
case EQL_ENSLAVE:
return eql_enslave(dev, ifr->ifr_data);
case EQL_EMANCIPATE:
return eql_emancipate(dev, ifr->ifr_data);
case EQL_GETSLAVECFG:
return eql_g_slave_cfg(dev, ifr->ifr_data);
case EQL_SETSLAVECFG:
return eql_s_slave_cfg(dev, ifr->ifr_data);
case EQL_GETMASTRCFG:
return eql_g_master_cfg(dev, ifr->ifr_data);
case EQL_SETMASTRCFG:
return eql_s_master_cfg(dev, ifr->ifr_data);
default:
return -EOPNOTSUPP;
};
}
/* queue->lock must be held */
static slave_t *__eql_schedule_slaves(slave_queue_t *queue)
{
unsigned long best_load = ~0UL;
struct list_head *this, *tmp, *head;
slave_t *best_slave;
best_slave = NULL;
/* Make a pass to set the best slave. */
head = &queue->all_slaves;
list_for_each_safe(this, tmp, head) {
slave_t *slave = list_entry(this, slave_t, list);
unsigned long slave_load, bytes_queued, priority_Bps;
/* Go through the slave list once, updating best_slave
* whenever a new best_load is found.
*/
bytes_queued = slave->bytes_queued;
priority_Bps = slave->priority_Bps;
if ((slave->dev->flags & IFF_UP) == IFF_UP) {
slave_load = (~0UL - (~0UL / 2)) -
(priority_Bps) + bytes_queued * 8;
if (slave_load < best_load) {
best_load = slave_load;
best_slave = slave;
}
} else {
/* We found a dead slave, kill it. */
eql_kill_one_slave(queue, slave);
}
}
return best_slave;
}
static netdev_tx_t eql_slave_xmit(struct sk_buff *skb, struct net_device *dev)
{
equalizer_t *eql = netdev_priv(dev);
slave_t *slave;
spin_lock(&eql->queue.lock);
slave = __eql_schedule_slaves(&eql->queue);
if (slave) {
struct net_device *slave_dev = slave->dev;
skb->dev = slave_dev;
skb->priority = 1;
slave->bytes_queued += skb->len;
dev_queue_xmit(skb);
dev->stats.tx_packets++;
} else {
dev->stats.tx_dropped++;
dev_kfree_skb(skb);
}
spin_unlock(&eql->queue.lock);
return NETDEV_TX_OK;
}
/*
* Private ioctl functions
*/
/* queue->lock must be held */
static slave_t *__eql_find_slave_dev(slave_queue_t *queue, struct net_device *dev)
{
struct list_head *this, *head;
head = &queue->all_slaves;
list_for_each(this, head) {
slave_t *slave = list_entry(this, slave_t, list);
if (slave->dev == dev)
return slave;
}
return NULL;
}
static inline int eql_is_full(slave_queue_t *queue)
{
equalizer_t *eql = netdev_priv(queue->master_dev);
if (queue->num_slaves >= eql->max_slaves)
return 1;
return 0;
}
/* queue->lock must be held */
static int __eql_insert_slave(slave_queue_t *queue, slave_t *slave)
{
if (!eql_is_full(queue)) {
slave_t *duplicate_slave = NULL;
duplicate_slave = __eql_find_slave_dev(queue, slave->dev);
if (duplicate_slave)
eql_kill_one_slave(queue, duplicate_slave);
list_add(&slave->list, &queue->all_slaves);
queue->num_slaves++;
slave->dev->flags |= IFF_SLAVE;
return 0;
}
return -ENOSPC;
}
static int eql_enslave(struct net_device *master_dev, slaving_request_t __user *srqp)
{
struct net_device *slave_dev;
slaving_request_t srq;
if (copy_from_user(&srq, srqp, sizeof (slaving_request_t)))
return -EFAULT;
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 18:56:21 +00:00
slave_dev = dev_get_by_name(&init_net, srq.slave_name);
if (slave_dev) {
if ((master_dev->flags & IFF_UP) == IFF_UP) {
/* slave is not a master & not already a slave: */
if (!eql_is_master(slave_dev) &&
!eql_is_slave(slave_dev)) {
slave_t *s = kmalloc(sizeof(*s), GFP_KERNEL);
equalizer_t *eql = netdev_priv(master_dev);
int ret;
if (!s) {
dev_put(slave_dev);
return -ENOMEM;
}
memset(s, 0, sizeof(*s));
s->dev = slave_dev;
s->priority = srq.priority;
s->priority_bps = srq.priority;
s->priority_Bps = srq.priority / 8;
spin_lock_bh(&eql->queue.lock);
ret = __eql_insert_slave(&eql->queue, s);
if (ret) {
dev_put(slave_dev);
kfree(s);
}
spin_unlock_bh(&eql->queue.lock);
return ret;
}
}
dev_put(slave_dev);
}
return -EINVAL;
}
static int eql_emancipate(struct net_device *master_dev, slaving_request_t __user *srqp)
{
equalizer_t *eql = netdev_priv(master_dev);
struct net_device *slave_dev;
slaving_request_t srq;
int ret;
if (copy_from_user(&srq, srqp, sizeof (slaving_request_t)))
return -EFAULT;
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 18:56:21 +00:00
slave_dev = dev_get_by_name(&init_net, srq.slave_name);
ret = -EINVAL;
if (slave_dev) {
spin_lock_bh(&eql->queue.lock);
if (eql_is_slave(slave_dev)) {
slave_t *slave = __eql_find_slave_dev(&eql->queue,
slave_dev);
if (slave) {
eql_kill_one_slave(&eql->queue, slave);
ret = 0;
}
}
dev_put(slave_dev);
spin_unlock_bh(&eql->queue.lock);
}
return ret;
}
static int eql_g_slave_cfg(struct net_device *dev, slave_config_t __user *scp)
{
equalizer_t *eql = netdev_priv(dev);
slave_t *slave;
struct net_device *slave_dev;
slave_config_t sc;
int ret;
if (copy_from_user(&sc, scp, sizeof (slave_config_t)))
return -EFAULT;
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 18:56:21 +00:00
slave_dev = dev_get_by_name(&init_net, sc.slave_name);
if (!slave_dev)
return -ENODEV;
ret = -EINVAL;
spin_lock_bh(&eql->queue.lock);
if (eql_is_slave(slave_dev)) {
slave = __eql_find_slave_dev(&eql->queue, slave_dev);
if (slave) {
sc.priority = slave->priority;
ret = 0;
}
}
spin_unlock_bh(&eql->queue.lock);
dev_put(slave_dev);
if (!ret && copy_to_user(scp, &sc, sizeof (slave_config_t)))
ret = -EFAULT;
return ret;
}
static int eql_s_slave_cfg(struct net_device *dev, slave_config_t __user *scp)
{
slave_t *slave;
equalizer_t *eql;
struct net_device *slave_dev;
slave_config_t sc;
int ret;
if (copy_from_user(&sc, scp, sizeof (slave_config_t)))
return -EFAULT;
[NET]: Make the device list and device lookups per namespace. This patch makes most of the generic device layer network namespace safe. This patch makes dev_base_head a network namespace variable, and then it picks up a few associated variables. The functions: dev_getbyhwaddr dev_getfirsthwbytype dev_get_by_flags dev_get_by_name __dev_get_by_name dev_get_by_index __dev_get_by_index dev_ioctl dev_ethtool dev_load wireless_process_ioctl were modified to take a network namespace argument, and deal with it. vlan_ioctl_set and brioctl_set were modified so their hooks will receive a network namespace argument. So basically anthing in the core of the network stack that was affected to by the change of dev_base was modified to handle multiple network namespaces. The rest of the network stack was simply modified to explicitly use &init_net the initial network namespace. This can be fixed when those components of the network stack are modified to handle multiple network namespaces. For now the ifindex generator is left global. Fundametally ifindex numbers are per namespace, or else we will have corner case problems with migration when we get that far. At the same time there are assumptions in the network stack that the ifindex of a network device won't change. Making the ifindex number global seems a good compromise until the network stack can cope with ifindex changes when you change namespaces, and the like. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-09-17 18:56:21 +00:00
slave_dev = dev_get_by_name(&init_net, sc.slave_name);
if (!slave_dev)
return -ENODEV;
ret = -EINVAL;
eql = netdev_priv(dev);
spin_lock_bh(&eql->queue.lock);
if (eql_is_slave(slave_dev)) {
slave = __eql_find_slave_dev(&eql->queue, slave_dev);
if (slave) {
slave->priority = sc.priority;
slave->priority_bps = sc.priority;
slave->priority_Bps = sc.priority / 8;
ret = 0;
}
}
spin_unlock_bh(&eql->queue.lock);
dev_put(slave_dev);
return ret;
}
static int eql_g_master_cfg(struct net_device *dev, master_config_t __user *mcp)
{
equalizer_t *eql;
master_config_t mc;
if (eql_is_master(dev)) {
eql = netdev_priv(dev);
mc.max_slaves = eql->max_slaves;
mc.min_slaves = eql->min_slaves;
if (copy_to_user(mcp, &mc, sizeof (master_config_t)))
return -EFAULT;
return 0;
}
return -EINVAL;
}
static int eql_s_master_cfg(struct net_device *dev, master_config_t __user *mcp)
{
equalizer_t *eql;
master_config_t mc;
if (copy_from_user(&mc, mcp, sizeof (master_config_t)))
return -EFAULT;
if (eql_is_master(dev)) {
eql = netdev_priv(dev);
eql->max_slaves = mc.max_slaves;
eql->min_slaves = mc.min_slaves;
return 0;
}
return -EINVAL;
}
static struct net_device *dev_eql;
static int __init eql_init_module(void)
{
int err;
printk(version);
dev_eql = alloc_netdev(sizeof(equalizer_t), "eql", eql_setup);
if (!dev_eql)
return -ENOMEM;
err = register_netdev(dev_eql);
if (err)
free_netdev(dev_eql);
return err;
}
static void __exit eql_cleanup_module(void)
{
unregister_netdev(dev_eql);
free_netdev(dev_eql);
}
module_init(eql_init_module);
module_exit(eql_cleanup_module);
MODULE_LICENSE("GPL");