Re: [PATCH net-next 0/3] net: introduce IFF_PROTO_DOWN flag.

On Fri, Mar 20, 2015 at 1:28 PM, David Miller [off-list ref] wrote:

From: anuradhak@cumulusnetworks.com
Date: Fri, 20 Mar 2015 08:11:55 -0700

quoted

Applications can detect errors in the network that would require
disabling the device independent of the admin state.

I hate changes like this.

The only reason it exists, is because you don't want to put together
the infrastructure and framework necessary for applications managing
this state in userspace to _work_ _together_.  So you make it a kernel
problem.

UP/DOWN is a boolean state, that's not changing.

So you need to design your stuff such that one entity takes all of the
collective decisions together and calculates the final up/down state,
and then asks the kernel to do that.

Nothing is more bullshit than having someone ask the kernel to up the
interface and the thing doesn't come up because of this auxiliary
crap.  That's broken and nobody is going to expect nor be happy with
that behavior.

I understand your position on this. And I agree .... ..however, in
this particular case ....

Unfortunately there is no way to distinguish between admin state and
error state in the user space. Currently administrators are directly
using “ip link set up/down”/IFF_UP as a means to admin-enable/disable
a device. So even if we were to consolidate all the errors in the user
space there is no way to error-disable/enable a device without
overriding the administrator’s directive.

What we are looking for is a way for user space to hold the device
down on detecting incorrect config/topology and for enabling the
device once the error condition is removed provided the administrator
didn’t intentionally disable it. And we don't see a way for a user
protocol to do this without involving the kernel.

Thanks for the review.