On 14/10/2018 06:21, Li RongQing wrote:
> After per-port vlan stats, vlan stats should be released
> when fail to add vlan
>
> Fixes: 9163a0fc1f0c0 ("net: bridge: add support for per-port vlan stats")
> cc: Nikolay Aleksandrov <nikolay@xxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Zhang Yu <zhangyu31@xxxxxxxxx>
> Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx>
> ---
> net/bridge/br_vlan.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/net/bridge/br_vlan.c b/net/bridge/br_vlan.c
> index 9b707234e4ae..e08e269041dd 100644
> --- a/net/bridge/br_vlan.c
> +++ b/net/bridge/br_vlan.c
> @@ -305,6 +305,10 @@ static int __vlan_add(struct net_bridge_vlan *v, u16 flags)
> if (masterv) {
> br_vlan_put_master(masterv);
> v->brvlan = NULL;
> +
> + if (masterv->stats != v->stats && v->stats)
> + free_percpu(v->stats);
> + v->stats = NULL;
> }
> } else {
> br_switchdev_port_vlan_del(dev, v->vid);
>
Hi,
Good catch, but the patch doesn't fix the bug entirely. The problem is that masterv can be
created just for this vlan and the br_vlan_put_master() above can free it, so we can
check a pointer that's not really up-to-date (and thus again leak memory).
You should move the new code above the br_vlan_put_master() call.
Also please tag the proper branch, this is for net-next, and CC all bridge
maintainers (added Roopa).
Thank you,
Nik
