RFC: bring UP 'lo' by default after creating new netns

RFC: bring UP 'lo' by default after creating new netns

[Mahesh Bandewar (महेश बंडेवार)]

Creation of new network namespace is almost always followed up by
bringing up the loopback device.

        ip netns add foo
        ip -netns foo link set lo up

I'm not sure if there are any consequences of bringing the device UP
at the creation of network-namespace.

thanks,
--mahesh..

diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 2178db8e47cd..ac0e86c9a17f 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -428,6 +428,11 @@ struct net *copy_net_ns(unsigned long flags,
                net_drop_ns(net);
                return ERR_PTR(rv);
        }
+       /* Set the loopback device UP */
+       rtnl_lock();
+       dev_open(net->loopback_dev);
+       rtnl_unlock();
+
        return net;
 }

Re: RFC: bring UP 'lo' by default after creating new netns

[Eric W. Biederman]

"Mahesh Bandewar (महेश बंडेवार)" <maheshb@google.com> writes:

> Creation of new network namespace is almost always followed up by
> bringing up the loopback device.
>
>         ip netns add foo
>         ip -netns foo link set lo up
>
> I'm not sure if there are any consequences of bringing the device UP
> at the creation of network-namespace.

Hard coded in net/core/net_namespace.c:copy_net_ns is definitely the
wrong place in the code for something like this.  If this lives anywhere
it should live in driver/net/loopback.c, or possibly in
net/core/dev.c:net_dev_init.

If we want this we want to match what we do when we the primary network
namespace.  Just so that there are no unneeded surprises with network
namespaces.

Eric

Re: RFC: bring UP 'lo' by default after creating new netns

[Mahesh Bandewar (महेश बंडेवार)]

On Thu, Jun 29, 2017 at 6:08 PM, Eric W. Biederman
<ebiederm@xmission.com> wrote:
> "Mahesh Bandewar (महेश बंडेवार)" <maheshb@google.com> writes:
>
>> Creation of new network namespace is almost always followed up by
>> bringing up the loopback device.
>>
>>         ip netns add foo
>>         ip -netns foo link set lo up
>>
>> I'm not sure if there are any consequences of bringing the device UP
>> at the creation of network-namespace.jjh
>
> Hard coded in net/core/net_namespace.c:copy_net_ns is definitely the
> wrong place in the code for something like this.  If this lives anywhere
> it should live in driver/net/loopback.c, or possibly in
> net/core/dev.c:net_dev_init.
>
interestingly dev_open() needs RTNL while it's not recommended at
net_dev_init() time.
Also net_dev_init() is called at the sub-system registration, do you
mean loopback_net_init() time?

> If we want this we want to match what we do when we the primary network
> namespace.  Just so that there are no unneeded surprises with network
> namespaces.
>
Yes, correct, it should be same whether it's init-net or some other. I
can add the IFF_UP flag during
loopback_setup() but then it will not perform anything that is done
during dev_open() phase. I'll try
moving the code to loopback_net_init() to achieve the same result so
that I can avoid that hard-coding too.

Thanks for comments Eric.

--mahesh..

> Eric