Re: [PATCH] nbd: use an idr to keep track of nbd devices

[Josef Bacik <jbacik@fb.com>]

On Mon, Jan 16, 2017 at 3:29 PM, Sagi Grimberg <sagi@grimberg.me> wrote:
> Hey Josef,
> 
>> I'm going to use it the same way loop does, there will be a
>> /dev/nbd-control where you can say ADD, REMOVE, and GET_NEXT. I need 
>> the
>> search functionality to see if we are adding something that already
>> exists, and to see what is the next unused device that can be used 
>> for a
>> connection.  Looking at the ida api it does not appear I can do 
>> that. If
>> I'm wrong then please point out an example I can look at, because I
>> haven't been able to find one.  Thanks,
> 
> Nope, ida doesn't have search functionality. Having said that, will it
> suffice to have the idr tree without a separate data structure? 
> because
> the tree mutation under idr_for_each is not allowed. For example,
> I'd expect that nbd module unload will iterate over the existing
> devices and destroy them which requires a separate tracking of
> them, also, I think that the idr user needs to take care of locking
> (unlike ida).

So I don't plan on messing with it under idr_for_each.  I'm basically 
copying+pasting what loop is doing and doing :1,$s/loop/nbd/g and 
hoping for the best.  I'm going to add reference counting to the nbd 
devices obviously to keep us from removing in use nbd devices.  The 
idr_for_each in the unload doesn't actually remove the devices from the 
idr, it just free's them.

Also so I don't have to find the other thread I did manage to test with 
per-device workqueues and a global workqueue and the performance was 
the same, I'll make that change to the other patch and resend it when I 
finish all this work so you can see the full picture.  Thanks!

Josef