All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH net-next] bonding: move bond-specific init after enslave happens
@ 2013-10-20 12:47 Veaceslav Falico
  2013-10-21  1:35 ` Ding Tianhong
  0 siblings, 1 reply; 3+ messages in thread
From: Veaceslav Falico @ 2013-10-20 12:47 UTC (permalink / raw)
  To: netdev; +Cc: jiri, Veaceslav Falico, Jay Vosburgh, Andy Gospodarek

As Jiri noted, currently we first do all bonding-specific initialization
(specifically - bond_select_active_slave(bond)) before we actually attach
the slave (so that it becomes visible through bond_for_each_slave() and
friends). This might result in bond_select_active_slave() not seeing the
first/new slave and, thus, not actually selecting an active slave.

Fix this by moving all the bond-related init part after we've actually
completely initialized and linked (via bond_master_upper_dev_link()) the
new slave.

After this we have all the initialization of the new slave *before*
linking, and all the stuff that needs to be done on bonding *after* it. It
has also a bonus effect - we can remove the locking on the new slave init
completely, and only use it for bond_select_active_slave().

Reported-by: Jiri Pirko <jiri@resnulli.us>
CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
---
 drivers/net/bonding/bond_main.c | 29 ++++++++++-------------------
 1 file changed, 10 insertions(+), 19 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index d90734f..047c0fb 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -1471,22 +1471,14 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
 		goto err_close;
 	}
 
-	write_lock_bh(&bond->lock);
-
 	prev_slave = bond_last_slave(bond);
 	bond_attach_slave(bond, new_slave);
 
 	new_slave->delay = 0;
 	new_slave->link_failure_count = 0;
 
-	write_unlock_bh(&bond->lock);
-
-	bond_compute_features(bond);
-
 	bond_update_speed_duplex(new_slave);
 
-	read_lock(&bond->lock);
-
 	new_slave->last_arp_rx = jiffies -
 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
 	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
@@ -1547,12 +1539,9 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
 		}
 	}
 
-	write_lock_bh(&bond->curr_slave_lock);
-
 	switch (bond->params.mode) {
 	case BOND_MODE_ACTIVEBACKUP:
 		bond_set_slave_inactive_flags(new_slave);
-		bond_select_active_slave(bond);
 		break;
 	case BOND_MODE_8023AD:
 		/* in 802.3ad mode, the internal mechanism
@@ -1578,7 +1567,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
 	case BOND_MODE_ALB:
 		bond_set_active_slave(new_slave);
 		bond_set_slave_inactive_flags(new_slave);
-		bond_select_active_slave(bond);
 		break;
 	default:
 		pr_debug("This slave is always active in trunk mode\n");
@@ -1596,10 +1584,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
 		break;
 	} /* switch(bond_mode) */
 
-	write_unlock_bh(&bond->curr_slave_lock);
-
-	bond_set_carrier(bond);
-
 #ifdef CONFIG_NET_POLL_CONTROLLER
 	slave_dev->npinfo = bond->dev->npinfo;
 	if (slave_dev->npinfo) {
@@ -1614,8 +1598,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
 	}
 #endif
 
-	read_unlock(&bond->lock);
-
 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
 					 new_slave);
 	if (res) {
@@ -1629,6 +1611,16 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
 		goto err_unregister;
 	}
 
+	bond_compute_features(bond);
+	bond_set_carrier(bond);
+
+	if (USES_PRIMARY(bond->params.mode)) {
+		read_lock(&bond->lock);
+		write_lock_bh(&bond->curr_slave_lock);
+		bond_select_active_slave(bond);
+		write_unlock_bh(&bond->curr_slave_lock);
+		read_unlock(&bond->lock);
+	}
 
 	pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
 		bond_dev->name, slave_dev->name,
@@ -1686,7 +1678,6 @@ err_free:
 	kfree(new_slave);
 
 err_undo_flags:
-	bond_compute_features(bond);
 	/* Enslave of first slave has failed and we need to fix master's mac */
 	if (!bond_has_slaves(bond) &&
 	    ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
-- 
1.8.1.4

^ permalink raw reply related	[flat|nested] 3+ messages in thread
* Re: [PATCH net-next] bonding: move bond-specific init after enslave happens
  2013-10-20 12:47 [PATCH net-next] bonding: move bond-specific init after enslave happens Veaceslav Falico
@ 2013-10-21  1:35 ` Ding Tianhong
  2013-10-21  7:03   ` Veaceslav Falico
  0 siblings, 1 reply; 3+ messages in thread
From: Ding Tianhong @ 2013-10-21  1:35 UTC (permalink / raw)
  To: Veaceslav Falico; +Cc: netdev, jiri, Jay Vosburgh, Andy Gospodarek

On 2013/10/20 20:47, Veaceslav Falico wrote:
> As Jiri noted, currently we first do all bonding-specific initialization
> (specifically - bond_select_active_slave(bond)) before we actually attach
> the slave (so that it becomes visible through bond_for_each_slave() and
> friends). This might result in bond_select_active_slave() not seeing the
> first/new slave and, thus, not actually selecting an active slave.
> 
> Fix this by moving all the bond-related init part after we've actually
> completely initialized and linked (via bond_master_upper_dev_link()) the
> new slave.
> 
> After this we have all the initialization of the new slave *before*
> linking, and all the stuff that needs to be done on bonding *after* it. It
> has also a bonus effect - we can remove the locking on the new slave init
> completely, and only use it for bond_select_active_slave().
> 
> Reported-by: Jiri Pirko <jiri@resnulli.us>
> CC: Jay Vosburgh <fubar@us.ibm.com>
> CC: Andy Gospodarek <andy@greyhouse.net>
> Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
> ---
>  drivers/net/bonding/bond_main.c | 29 ++++++++++-------------------
>  1 file changed, 10 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> index d90734f..047c0fb 100644
> --- a/drivers/net/bonding/bond_main.c
> +++ b/drivers/net/bonding/bond_main.c
> @@ -1471,22 +1471,14 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>  		goto err_close;
>  	}
>  
> -	write_lock_bh(&bond->lock);
> -
>  	prev_slave = bond_last_slave(bond);
>  	bond_attach_slave(bond, new_slave);
>  
>  	new_slave->delay = 0;
>  	new_slave->link_failure_count = 0;
>  
> -	write_unlock_bh(&bond->lock);
> -
> -	bond_compute_features(bond);
> -
>  	bond_update_speed_duplex(new_slave);
>  
> -	read_lock(&bond->lock);
> -
>  	new_slave->last_arp_rx = jiffies -
>  		(msecs_to_jiffies(bond->params.arp_interval) + 1);
>  	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
> @@ -1547,12 +1539,9 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>  		}
>  	}
>  
> -	write_lock_bh(&bond->curr_slave_lock);
> -
>  	switch (bond->params.mode) {
>  	case BOND_MODE_ACTIVEBACKUP:
>  		bond_set_slave_inactive_flags(new_slave);
> -		bond_select_active_slave(bond);
>  		break;
>  	case BOND_MODE_8023AD:
>  		/* in 802.3ad mode, the internal mechanism
> @@ -1578,7 +1567,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>  	case BOND_MODE_ALB:
>  		bond_set_active_slave(new_slave);
>  		bond_set_slave_inactive_flags(new_slave);
> -		bond_select_active_slave(bond);
>  		break;
>  	default:
>  		pr_debug("This slave is always active in trunk mode\n");
> @@ -1596,10 +1584,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>  		break;
>  	} /* switch(bond_mode) */
>  
> -	write_unlock_bh(&bond->curr_slave_lock);
> -
> -	bond_set_carrier(bond);
> -
>  #ifdef CONFIG_NET_POLL_CONTROLLER
>  	slave_dev->npinfo = bond->dev->npinfo;
>  	if (slave_dev->npinfo) {
> @@ -1614,8 +1598,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>  	}
>  #endif
>  
> -	read_unlock(&bond->lock);
> -
>  	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
>  					 new_slave);
>  	if (res) {
> @@ -1629,6 +1611,16 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>  		goto err_unregister;
>  	}
>  
> +	bond_compute_features(bond);
> +	bond_set_carrier(bond);
> +
> +	if (USES_PRIMARY(bond->params.mode)) {
> +		read_lock(&bond->lock);
> +		write_lock_bh(&bond->curr_slave_lock);
> +		bond_select_active_slave(bond);
> +		write_unlock_bh(&bond->curr_slave_lock);
> +		read_unlock(&bond->lock);
> +	}
>  

agree to move the lock, and I think bond_attach_slave() should add here,
as it look more logical, the slave_cnt should not add before the slave truly
add to the bond.

Regards.
Ding

>  	pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
>  		bond_dev->name, slave_dev->name,
> @@ -1686,7 +1678,6 @@ err_free:
>  	kfree(new_slave);
>  
>  err_undo_flags:
> -	bond_compute_features(bond);
>  	/* Enslave of first slave has failed and we need to fix master's mac */
>  	if (!bond_has_slaves(bond) &&
>  	    ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
> 

^ permalink raw reply	[flat|nested] 3+ messages in thread
* Re: [PATCH net-next] bonding: move bond-specific init after enslave happens
  2013-10-21  1:35 ` Ding Tianhong
@ 2013-10-21  7:03   ` Veaceslav Falico
  0 siblings, 0 replies; 3+ messages in thread
From: Veaceslav Falico @ 2013-10-21  7:03 UTC (permalink / raw)
  To: Ding Tianhong; +Cc: netdev, jiri, Jay Vosburgh, Andy Gospodarek

On Mon, Oct 21, 2013 at 09:35:09AM +0800, Ding Tianhong wrote:
>On 2013/10/20 20:47, Veaceslav Falico wrote:
>> As Jiri noted, currently we first do all bonding-specific initialization
>> (specifically - bond_select_active_slave(bond)) before we actually attach
>> the slave (so that it becomes visible through bond_for_each_slave() and
>> friends). This might result in bond_select_active_slave() not seeing the
>> first/new slave and, thus, not actually selecting an active slave.
>>
>> Fix this by moving all the bond-related init part after we've actually
>> completely initialized and linked (via bond_master_upper_dev_link()) the
>> new slave.
>>
>> After this we have all the initialization of the new slave *before*
>> linking, and all the stuff that needs to be done on bonding *after* it. It
>> has also a bonus effect - we can remove the locking on the new slave init
>> completely, and only use it for bond_select_active_slave().
>>
>> Reported-by: Jiri Pirko <jiri@resnulli.us>
>> CC: Jay Vosburgh <fubar@us.ibm.com>
>> CC: Andy Gospodarek <andy@greyhouse.net>
>> Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
>> ---
>>  drivers/net/bonding/bond_main.c | 29 ++++++++++-------------------
>>  1 file changed, 10 insertions(+), 19 deletions(-)
>>
>> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>> index d90734f..047c0fb 100644
>> --- a/drivers/net/bonding/bond_main.c
>> +++ b/drivers/net/bonding/bond_main.c
>> @@ -1471,22 +1471,14 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>>  		goto err_close;
>>  	}
>>
>> -	write_lock_bh(&bond->lock);
>> -
>>  	prev_slave = bond_last_slave(bond);
>>  	bond_attach_slave(bond, new_slave);
>>
>>  	new_slave->delay = 0;
>>  	new_slave->link_failure_count = 0;
>>
>> -	write_unlock_bh(&bond->lock);
>> -
>> -	bond_compute_features(bond);
>> -
>>  	bond_update_speed_duplex(new_slave);
>>
>> -	read_lock(&bond->lock);
>> -
>>  	new_slave->last_arp_rx = jiffies -
>>  		(msecs_to_jiffies(bond->params.arp_interval) + 1);
>>  	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
>> @@ -1547,12 +1539,9 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>>  		}
>>  	}
>>
>> -	write_lock_bh(&bond->curr_slave_lock);
>> -
>>  	switch (bond->params.mode) {
>>  	case BOND_MODE_ACTIVEBACKUP:
>>  		bond_set_slave_inactive_flags(new_slave);
>> -		bond_select_active_slave(bond);
>>  		break;
>>  	case BOND_MODE_8023AD:
>>  		/* in 802.3ad mode, the internal mechanism
>> @@ -1578,7 +1567,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>>  	case BOND_MODE_ALB:
>>  		bond_set_active_slave(new_slave);
>>  		bond_set_slave_inactive_flags(new_slave);
>> -		bond_select_active_slave(bond);
>>  		break;
>>  	default:
>>  		pr_debug("This slave is always active in trunk mode\n");
>> @@ -1596,10 +1584,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>>  		break;
>>  	} /* switch(bond_mode) */
>>
>> -	write_unlock_bh(&bond->curr_slave_lock);
>> -
>> -	bond_set_carrier(bond);
>> -
>>  #ifdef CONFIG_NET_POLL_CONTROLLER
>>  	slave_dev->npinfo = bond->dev->npinfo;
>>  	if (slave_dev->npinfo) {
>> @@ -1614,8 +1598,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>>  	}
>>  #endif
>>
>> -	read_unlock(&bond->lock);
>> -
>>  	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
>>  					 new_slave);
>>  	if (res) {
>> @@ -1629,6 +1611,16 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
>>  		goto err_unregister;
>>  	}
>>
>> +	bond_compute_features(bond);
>> +	bond_set_carrier(bond);
>> +
>> +	if (USES_PRIMARY(bond->params.mode)) {
>> +		read_lock(&bond->lock);
>> +		write_lock_bh(&bond->curr_slave_lock);
>> +		bond_select_active_slave(bond);
>> +		write_unlock_bh(&bond->curr_slave_lock);
>> +		read_unlock(&bond->lock);
>> +	}
>>
>
>agree to move the lock, and I think bond_attach_slave() should add here,
>as it look more logical, the slave_cnt should not add before the slave truly
>add to the bond.

bond_(de)attach_slave() should be removed completely, actually. we don't
need special functions for ++/--.

OTOH, the whole slave_cnt is flawed a bit, whilst using RCU - we can never
guarantee that it's the actual value if we don't hold rtnl lock (we do in
ioctl, but we don't in the hash functions).

I'll take a closer look and send v2.

>
>Regards.
>Ding
>
>>  	pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
>>  		bond_dev->name, slave_dev->name,
>> @@ -1686,7 +1678,6 @@ err_free:
>>  	kfree(new_slave);
>>
>>  err_undo_flags:
>> -	bond_compute_features(bond);
>>  	/* Enslave of first slave has failed and we need to fix master's mac */
>>  	if (!bond_has_slaves(bond) &&
>>  	    ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
>>
>
>

^ permalink raw reply	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2013-10-21  7:06 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-10-20 12:47 [PATCH net-next] bonding: move bond-specific init after enslave happens Veaceslav Falico
2013-10-21  1:35 ` Ding Tianhong
2013-10-21  7:03   ` Veaceslav Falico

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.