This patch fixes a hang when enslaving a new slave while incoming traffic
is running, that looks like a deadlock between the BR_NETPROTO_LOCK,
dev->xmit_lock and the bond lock (happens on quad processor machines,
but KDB back trace wasn't clear enough).
This patch is against bonding 2.4.20-20030317.
diff -Nuarp linux-2.4.20-bonding-20030317/drivers/net/bonding.c linux-2.4.20-bonding-20030317-devel/drivers/net/bonding.c
--- linux-2.4.20-bonding-20030317/drivers/net/bonding.c 2003-03-18 17:03:25.000000000 +0200
+++ linux-2.4.20-bonding-20030317-devel/drivers/net/bonding.c 2003-03-18 17:03:26.000000000 +0200
@@ -295,6 +295,10 @@
* - Fixed hang in bond_release() while traffic is running.
* netdev_set_master() must not be called from within the bond lock.
*
+ * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
+ * Shmulik Hen <shmulik.hen at intel dot com>
+ * - Fixed hang in bond_enslave(): netdev_set_master() must not be
+ * called from within the bond lock while traffic is running.
*/
#include <linux/config.h>
@@ -1066,14 +1070,12 @@ static int bond_enslave(struct net_devic
"Warning : no link monitoring support for %s\n",
slave_dev->name);
}
- write_lock_irqsave(&bond->lock, flags);
/* not running. */
if ((slave_dev->flags & IFF_UP) != IFF_UP) {
#ifdef BONDING_DEBUG
printk(KERN_CRIT "Error, slave_dev is not running\n");
#endif
- write_unlock_irqrestore(&bond->lock, flags);
return -EINVAL;
}
@@ -1082,12 +1084,10 @@ static int bond_enslave(struct net_devic
#ifdef BONDING_DEBUG
printk(KERN_CRIT "Error, Device was already enslaved\n");
#endif
- write_unlock_irqrestore(&bond->lock, flags);
return -EBUSY;
}
if ((new_slave = kmalloc(sizeof(slave_t), GFP_ATOMIC)) == NULL) {
- write_unlock_irqrestore(&bond->lock, flags);
return -ENOMEM;
}
memset(new_slave, 0, sizeof(slave_t));
@@ -1100,9 +1100,7 @@ static int bond_enslave(struct net_devic
#ifdef BONDING_DEBUG
printk(KERN_CRIT "Error %d calling netdev_set_master\n", err);
#endif
- kfree(new_slave);
- write_unlock_irqrestore(&bond->lock, flags);
- return err;
+ goto err_free;
}
new_slave->dev = slave_dev;
@@ -1121,6 +1119,8 @@ static int bond_enslave(struct net_devic
dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
}
+ write_lock_irqsave(&bond->lock, flags);
+
bond_attach_slave(bond, new_slave);
new_slave->delay = 0;
new_slave->link_failure_count = 0;
@@ -1259,7 +1259,11 @@ static int bond_enslave(struct net_devic
new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
new_slave->link == BOND_LINK_UP ? "n up" : " down");
+ //enslave is successfull
return 0;
+err_free:
+ kfree(new_slave);
+ return err;
}
/*
@@ -1607,6 +1611,9 @@ static int bond_release_all(struct net_d
kfree(our_slave);
+ /* Can be safely called from inside the bond lock
+ since traffic and timers have already stopped
+ */
netdev_set_master(slave_dev, NULL);
/* only restore its RUNNING flag if monitoring set it down */
-- | Shmulik Hen | | Israel Design Center (Jerusalem) | | LAN Access Division | | Intel Communications Group, Intel corp. | | | | Anti-Spam: shmulik dot hen at intel dot com |- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Mar 23 2003 - 22:00:30 EST