Re: [PATCH linux-next 1/4] infiniband/ipoib: fix possible NULL pointer dereference in ipoib_get_iflink

[Jason Gunthorpe <jgunthorpe@obsidianresearch.com>]

On Wed, Apr 15, 2015 at 09:17:14AM +0300, Erez Shitrit wrote:
> >>+       /* parent interface */
> >>+       if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
> >>+               return dev->ifindex;
> >>+
> >>+       /* child/vlan interface */
> >>+       if (!priv->parent)
> >>+               return -1;

> >Like was said for other drivers, I can't see how parent can be null
> >while IPOIB_FLAG_SUBINTERFACE is set. Drop the last if.

> It can, at least for ipoib child interface (AKA "vlan"), you can't
> control the call for that ndo and it can be called before the parent
> was set.

If the ndo can be called before the netdev private structures are fully
prepared then we have another bug, and returning -1 or 0 is not the right
answer anyhow.

For safety, fold this into your patch.

diff --git a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
index 9fad7b5ac8b9..e62b007adf5d 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
@@ -58,6 +58,7 @@ int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
        /* MTU will be reset when mcast join happens */
        priv->dev->mtu   = IPOIB_UD_MTU(priv->max_ib_mtu);
        priv->mcast_mtu  = priv->admin_mtu = priv->dev->mtu;
+       priv->parent = ppriv->dev;
        set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
 
        result = ipoib_set_dev_features(priv, ppriv->ca);
@@ -84,8 +85,6 @@ int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
                goto register_failed;
        }
 
-       priv->parent = ppriv->dev;
-
        ipoib_create_debug_files(priv->dev);
 
        /* RTNL childs don't need proprietary sysfs entries */