[dm-devel] [RFC] Message ioctl direct to device

[Daniel Phillips]

On Friday 09 July 2004 05:36, Stefan Bader wrote:
> >   - Why bother ioctling the dm control device when we can ioctl
> >     the virtual device directly?
>
> Using the control device is usefull if you want to react on messages
> from the control interface without the need to open virtual devices
> at that time or keep filehandles open all the time.
> Userspace recovery of failed multipath devices has to save as much
> memory since it must pin itself into memory in order to handle
> low-mem situations.
> So I would like to see a interface through the control device as
> well.

So this is only about saving memory on open?  This is not a problem for 
my application, but perhaps it's more of a problem for multi-pathing.  
What traffic do you expect on the message interface. and by the way, 
how large are the messages?

Alasdair did mention that he'd be interested in seeing the whole ioctl 
interface available through either the control device or directly to 
the virtual device.  That would be a rather more elaborate patch than 
the 100 liner I just posted, but I'll have a look at it if everybody 
thinks it's worth doing.

I do not think it would be a good idea to graft the message interface 
onto the dm_ioctl struct though, there's no advantage and the code to 
call it from a user program would get uglier. That would be fine if the 
only callers are dmsetup, lvm2 or evms, but it's undesirable if other 
programs are expected to use this interface, which they are.

Regards,

Daniel