Re: [RFC ABI V2 5/8] RDMA/core: Add new ioctl interface

[Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>]

On Wed, Jul 20, 2016 at 05:03:46AM -0500, Christoph Lameter wrote:
> > If I recall, Jason proposed that. I think the main reason here is for strace
> > and debugging. Since the ABI is now driver specific, this helps you parse the
> > structs via strace.
> 
> You specify the driver_id in the ioctl data structure. The driver in an
> ioctl is specified by the descriptor. strace etc would have an easier time
> if we would follow the standard conventions for devices and not add
> another device_id somewhere.

There isn't enough ioctl #'s for that. We probably need something like
over 500 ioctls by the time we are done all the drivers and
interfaces.. There is only about 100 reserved for RDMA today, and the
ioctl space is looking pretty full to steal another 4 blocks.

So the basic proposal is to use only a small number of ioctls and
have an 'extended ioctl #' in the struct.

The basic requirement is the same as in ioctl, the ioctl # and
extended # in the struct must be globally unique.

The device_id part of the extension allows each driver to have its own
globally unique number space without requireing another ioctl block.

> Why would there be an issue of sharing data between multiple descriptors?
> Data could be a subsystem specific state and not device specific if you
> want to share.

The state should be fd specific.

> Why do you need that information a second time if the descriptor already
> provides the device information?

Because the descriptor only indirectly implies a specific device.

It is the same reason overlapping ioctls are discouraged in the
kernel, even though I know I opened /dev/foo I still should send
globally unique 'FOO' ioctls to what I opened.

> and then run ioctls on that file to configure the device. That is very
> similar to the traditional use of ioctls. Security for each device can be
> controlled separately without inspecting the data being passed
> (which

It turns out that doesn't work today anyhow, since we have
multi-device requirements in the rdma_cm and you end up with a rdma_cm
descriptor that is world accessible and can interact with the network
even without device permissions.

netdev doesn't work in this screwy way, I think we should move away
from it as well...

> would require modifications to the security code). strace and
> other tools would just natively know that the descriptor refers to a device.

No, strace doesn't know what an open FD is, it just inspects the ioctl
# and content to do its decode. Requiring strace to decode differently
depending on what was opened would be the deviation from what we do
today.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html