[PATCH] ovs: queue_userspace_packet: bail out if nla_nest_start returns NULL

[PATCH] ovs: queue_userspace_packet: bail out if nla_nest_start returns NULL

[Daniel Borkmann]

The return value of nla_nest_start() is not checked, and can be NULL,
which is then being dereferenced in nla_nest_end(). Add a check to
prevent such situations.

Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
---
 On top of "net-next" tree.

 net/openvswitch/datapath.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index d12d6b8..51a3c66 100644
--- a/net/openvswitch/datapath.c
+++ b/net/openvswitch/datapath.c
@@ -427,6 +427,10 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
 	upcall->dp_ifindex = dp_ifindex;
 
 	nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
+	if (unlikely(!nla)) {
+		err = -ENOMEM;
+		goto out;
+	}
 	ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 	nla_nest_end(user_skb, nla);
 
-- 
1.7.11.7

Re: [PATCH] ovs: queue_userspace_packet: bail out if nla_nest_start returns NULL

[Jesse Gross]

On Wed, Jul 3, 2013 at 1:49 AM, Daniel Borkmann <dborkman@redhat.com> wrote:
> The return value of nla_nest_start() is not checked, and can be NULL,
> which is then being dereferenced in nla_nest_end(). Add a check to
> prevent such situations.
>
> Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
> ---
>  On top of "net-next" tree.
>
>  net/openvswitch/datapath.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
> index d12d6b8..51a3c66 100644
> --- a/net/openvswitch/datapath.c
> +++ b/net/openvswitch/datapath.c
> @@ -427,6 +427,10 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
>         upcall->dp_ifindex = dp_ifindex;
>
>         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
> +       if (unlikely(!nla)) {
> +               err = -ENOMEM;
> +               goto out;
> +       }

nla_nest_start() doesn't allocate any memory, it only fails if there
isn't enough space, so the correct error code would be -EMSGSIZE.
However, we just calculated the correct size a few lines earlier and
we don't double check the size anywhere else, so this doesn't make a
lot of sense to me.

Re: [PATCH] ovs: queue_userspace_packet: bail out if nla_nest_start returns NULL

[Daniel Borkmann]

On 07/03/2013 06:32 PM, Jesse Gross wrote:
> On Wed, Jul 3, 2013 at 1:49 AM, Daniel Borkmann <dborkman@redhat.com> wrote:
>> The return value of nla_nest_start() is not checked, and can be NULL,
>> which is then being dereferenced in nla_nest_end(). Add a check to
>> prevent such situations.
>>
>> Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
>> ---
>>   On top of "net-next" tree.
>>
>>   net/openvswitch/datapath.c | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
>> index d12d6b8..51a3c66 100644
>> --- a/net/openvswitch/datapath.c
>> +++ b/net/openvswitch/datapath.c
>> @@ -427,6 +427,10 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
>>          upcall->dp_ifindex = dp_ifindex;
>>
>>          nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
>> +       if (unlikely(!nla)) {
>> +               err = -ENOMEM;
>> +               goto out;
>> +       }
>
> nla_nest_start() doesn't allocate any memory, it only fails if there
> isn't enough space, so the correct error code would be -EMSGSIZE.
> However, we just calculated the correct size a few lines earlier and
> we don't double check the size anywhere else, so this doesn't make a
> lot of sense to me.

Fair enough, please drop this then.