Re: [RFC PATCH 4/4] bonding: Implement ndo_xmit_slave_get

[Maor Gottlieb <maorg@mellanox.com>]

On 1/29/2020 4:08 AM, Jay Vosburgh wrote:
> Maor Gottlieb <maorg@mellanox.com> wrote:
>
>> Add implementation of ndo_xmit_slave_get.
>> When user set the LAG_FLAGS_HASH_ALL_SLAVES bit and the xmit slave
>> result is based on the hash, then the slave will be selected from the
>> array of all the slaves.
>>
>> Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
>> ---
>> drivers/net/bonding/bond_main.c | 63 ++++++++++++++++++++++++++++++---
>> include/net/bonding.h           |  1 +
>> 2 files changed, 60 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>> index adab1e3549ff..c8f440d1b624 100644
>> --- a/drivers/net/bonding/bond_main.c
>> +++ b/drivers/net/bonding/bond_main.c
>> @@ -4098,7 +4098,8 @@ static void bond_skip_slave(struct bond_up_slave *slaves,
>>   */
>> int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
>> {
>> -	struct bond_up_slave *active_slaves, *old_active_slaves;
>> +	struct bond_up_slave *active_slaves = NULL, *all_slaves = NULL;
>> +	struct bond_up_slave *old_active_slaves, *old_all_slaves;
>> 	struct slave *slave;
>> 	struct list_head *iter;
>> 	int agg_id = 0;
>> @@ -4110,7 +4111,9 @@ int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
>>
>> 	active_slaves = kzalloc(struct_size(active_slaves, arr,
>> 					    bond->slave_cnt), GFP_KERNEL);
>> -	if (!active_slaves) {
>> +	all_slaves = kzalloc(struct_size(all_slaves, arr,
>> +					 bond->slave_cnt), GFP_KERNEL);
>> +	if (!active_slaves || !all_slaves) {
>> 		ret = -ENOMEM;
>> 		pr_err("Failed to build slave-array.\n");
>> 		goto out;
>> @@ -4141,14 +4144,17 @@ int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
>> 			if (!agg || agg->aggregator_identifier != agg_id)
>> 				continue;
>> 		}
>> -		if (!bond_slave_can_tx(slave))
>> +		if (!bond_slave_can_tx(slave)) {
>> +			all_slaves->arr[all_slaves->count++] = slave;
>> 			continue;
>> +		}
>> 		if (skipslave == slave)
>> 			continue;
>>
>> 		slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
>> 			  active_slaves->count);
>>
>> +		all_slaves->arr[all_slaves->count++] = slave;
>> 		active_slaves->arr[active_slaves->count++] = slave;
>> 	}
>>
>> @@ -4156,10 +4162,18 @@ int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
>> 	rcu_assign_pointer(bond->active_slaves, active_slaves);
>> 	if (old_active_slaves)
>> 		kfree_rcu(old_active_slaves, rcu);
>> +
>> +	old_all_slaves = rtnl_dereference(bond->all_slaves);
>> +	rcu_assign_pointer(bond->all_slaves, all_slaves);
>> +	if (old_all_slaves)
>> +		kfree_rcu(old_all_slaves, rcu);
>> out:
>> -	if (ret != 0 && skipslave)
>> +	if (ret != 0 && skipslave) {
>> 		bond_skip_slave(rtnl_dereference(bond->active_slaves),
>> 				skipslave);
>> +		kfree(all_slaves);
>> +		kfree(active_slaves);
>> +	}
> 	I'm still going through the patch set, but noticed this right
> away: the above will leak memory if !skipslave and the allocation for
> active_slaves succeeds, but the allocation for all_slaves fails.
>> 	return ret;
>> }
>> @@ -4265,6 +4279,46 @@ static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
>> 	return txq;
>> }
>>
>> +static struct net_device *bond_xmit_slave_get(struct net_device *master_dev,
>> +					      struct sk_buff *skb,
>> +					      int flags)
>> +{
>> +	struct bonding *bond = netdev_priv(master_dev);
>> +	struct bond_up_slave *slaves;
>> +	struct slave *slave;
>> +
>> +	switch (BOND_MODE(bond)) {
>> +	case BOND_MODE_ROUNDROBIN:
>> +		slave = bond_xmit_roundrobin_slave_get(bond, skb);
>> +		break;
>> +	case BOND_MODE_ACTIVEBACKUP:
>> +		slave = bond_xmit_activebackup_slave_get(bond, skb);
>> +		break;
>> +	case BOND_MODE_8023AD:
>> +	case BOND_MODE_XOR:
>> +		if (flags & LAG_FLAGS_HASH_ALL_SLAVES)
>> +			slaves = rcu_dereference(bond->all_slaves);
>> +		else
>> +			slaves = rcu_dereference(bond->active_slaves);
>> +		slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
>> +		break;
>> +	case BOND_MODE_BROADCAST:
>> +		return ERR_PTR(-EOPNOTSUPP);
>> +	case BOND_MODE_ALB:
>> +		slave = bond_xmit_alb_slave_get(bond, skb);
>> +		break;
>> +	case BOND_MODE_TLB:
>> +		slave = bond_xmit_tlb_slave_get(bond, skb);
>> +		break;
>> +	default:
>> +		return NULL;
> 	I would argue this should (a) return an error (not NULL), and,
> (b) ideally issue a netdev_err for this impossible situation, similar to
> the other switch statements in bonding.
>
> 	-J
> 	
>> +	}
>> +
>> +	if (slave)
>> +		return slave->dev;
>> +	return NULL;
>> +}
>> +
>> static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
>> {
>> 	struct bonding *bond = netdev_priv(dev);
>> @@ -4387,6 +4441,7 @@ static const struct net_device_ops bond_netdev_ops = {
>> 	.ndo_del_slave		= bond_release,
>> 	.ndo_fix_features	= bond_fix_features,
>> 	.ndo_features_check	= passthru_features_check,
>> +	.ndo_xmit_slave_get	= bond_xmit_slave_get,
>> };
>>
>> static const struct device_type bond_type = {
>> diff --git a/include/net/bonding.h b/include/net/bonding.h
>> index b77daffc1b52..6dd970eb9d3f 100644
>> --- a/include/net/bonding.h
>> +++ b/include/net/bonding.h
>> @@ -201,6 +201,7 @@ struct bonding {
>> 	struct   slave __rcu *current_arp_slave;
>> 	struct   slave __rcu *primary_slave;
>> 	struct   bond_up_slave __rcu *active_slaves; /* Array of usable slaves */
>> +	struct   bond_up_slave __rcu *all_slaves; /* Array of all slaves */
>> 	bool     force_primary;
>> 	s32      slave_cnt; /* never change this value outside the attach/detach wrappers */
>> 	int     (*recv_probe)(const struct sk_buff *, struct bonding *,
>> -- 
>> 2.17.2
>>
> ---
> 	-Jay Vosburgh, jay.vosburgh@canonical.com

Thanks Jay,
I will address the comments and submit this patch set along with the 
RoCE patches.