Re: ataraid(4) and VOP_OPENing the same block device

[Adam Hamsik <haaaad%gmail.com@localhost>]

On Aug,Tuesday 26 2008, at 2:53 PM, Juan Romero Pardines wrote:

> 2008/8/26 Quentin Garnier <cube%cubidou.net@localhost>:
>
> > I don't think there is any need of opening the devices multiple  
> > times.
> > The ataraid(4) driver should open all the necessary devices for all  
> > its
> > arrays, and then use them.  In your example, ataraid0 "owns" wd3 and
> > wd4, so why would it need to have them opened multiple times?
>
> Perhaps I didn't explain this correctly before. In my example I got  
> ataraid0
> and (like you said) it owns the disks wd3 and wd4, but my problem is  
> related
> to the fact that ld0 owns the same VBLK vnode on the same devices than
> ld1 wants to have access to... therefore once ld0 has taken (and  
> locked) the
> vnode via VOP_OPEN, subsequents openings to the vnode are failing  
> with EBUSY.
>
> At least that's my understanding. I don't know if you are allowed to
> bdevvp() on the same VBLK vnode either, but this call seems to  
> succeed...
> I'd assume it's allowed.

I think that it is not possible to VOP_OPEN block device more than once,
but you can open raw devices  as many times as you want.

In my device-mapper driver I created
struct pdev {
        struct vnode pdev_vnode;
        . . .
};

Which is in SLIST shared between all dm devices in system. When dm  
device
is created I VOP_OPEN selected disk device and add it's vnode to pdev  
SLIST.
When another dm device with same disk is created I search SLIST list  
for vnode
return it. You can look at this code at haad-dm branch in sys/dev/dm/ 
dm_pdev.c.

You probably need to do the same in ld0, or you can use device-mapper  
to do it
for you and only write own target.

Regards

Adam.