RE: Passing ioctls between front-end and back-end drivers

RE: Passing ioctls between front-end and back-end drivers

[Ian Pratt]

> I see that the block driver supports requests for READ, WRITE 
> and PROBE. 
> Is there any recommended way to pass ioctl requests from the 
> front-end to be executed by the back-end driver?

Currently, no.

Creating a new request operation type that allows the front end to pass
through ioctls would be useful, but it ends up being rather Linux
specific. Further, for security reasons the backend is going to need to
be able to decode the ioctl and decide whether its safe to issue or not.
It would be better to have a 'wire protocol' for this.

A mechanism for passing scsi commands through would be better, but even
then there are a bunch of vendor-specific ones its not entirely easy to
tell whether they should be allowed by a guest or not. I suspect we need
to draw up a list of what "ioctls" should be allowed. What one do you
want to perform?

Ian 

Re: Passing ioctls between front-end and back-end drivers

[Nick Logan]

Ah, that explains why I couldn't find out how to do it. The ioctl's that 
I am thinking of are vendor specific. I'm looking at supporting devices 
in Dom0 that can be accessed from the guest domains and part of the 
driver functionality requires ioctls to get/set attributes etc. It would 
be up to the backend to determine if the ioctls were valid or not. I 
agree that this is rather Linux/Unix specific but in my case it's a 
fundamental part of the driver.

Maybe some of the attributes can be passed via Xenstore, I'll investigate.

Nick

 Ian Pratt wrote:

>>I see that the block driver supports requests for READ, WRITE 
>>and PROBE. 
>>Is there any recommended way to pass ioctl requests from the 
>>front-end to be executed by the back-end driver?
>>    
>>
>
>Currently, no.
>
>Creating a new request operation type that allows the front end to pass
>through ioctls would be useful, but it ends up being rather Linux
>specific. Further, for security reasons the backend is going to need to
>be able to decode the ioctl and decide whether its safe to issue or not.
>It would be better to have a 'wire protocol' for this.
>
>A mechanism for passing scsi commands through would be better, but even
>then there are a bunch of vendor-specific ones its not entirely easy to
>tell whether they should be allowed by a guest or not. I suspect we need
>to draw up a list of what "ioctls" should be allowed. What one do you
>want to perform?
>
>Ian 
>  
>

Passing ioctls between front-end and back-end drivers

[Nick Logan]

I see that the block driver supports requests for READ, WRITE and PROBE. 
Is there any recommended way to pass ioctl requests from the front-end 
to be executed by the back-end driver?

Thanks,

Nick