[illumos-Developer] NOW with zvol-only zfs diffs --> iSCSI UNMAP

[Garrett D'Amore]

On Sun, 2011-02-20 at 15:47 -0500, Eric Schrock wrote:

>         
> 
> 
> Right.  To get there you have go through devfs, where I assume these
> credentials are actually enforced.

Thats where they *should* be enforced.  I've not looked.

> 
> 
>         I take it if we want to do DKIOCFREE, we should also implement
>         this ioctl in
>         sd.c, and other places!?  Wow.  That's looking like a lot more
>         work.  Now I
>         don't mind that work, but it will take time for design, and
>         possibly have
>         other consequences.
> 
> 
> I don't think that's what Garrett (or myself) is suggesting.  I think
> the point is to get the right interface in place such that it could be
> used throughout the stack someday.  Putting something back with the
> right architecture and one tested consumer is good enough for me; we
> don't have to conquer the world right now :-)

*Right.  We just want an interface that can grow to do the other bits...
I'll add the bits to sd later. :-)

> 

>         
> 
> 
> Seems a little strange to use 'extpart_info'; we should probably do
> something like:
> 
> 
> typedef struct dkioc_free {
>         diskaddr_t df_start;
>         diskaddr_t df_length;
> } dkioc_free_t;
> 

Agreed. :-)  I'd also add a "flags" field for possible future expansion.
(I'm thinking that we might want a flag for"immediate" free vs. deferred
garbage collection, for example.  There might be other flags to indicate
whether a "data deletion" guarantee is required as well.  Those are all
future ideas though, which is why I propose to leave them as a flags
field.)

	- Garrett