Re: [PATCH vfs 2/2] {block|char}_dev: remove inode->i_devices

[Tejun Heo <tj@kernel.org>]

Hey, Boaz.

On Tue, Nov 18, 2014 at 02:10:50PM +0200, Boaz Harrosh wrote:
> Right? well if so than those special inodes do
> not hold any data and never do IO of any kind
> so all the inode members that are needed for IO
> are candidates.

It may not carry dirty pages but the inode itself still can get
dirtied tho.

> example: i_wb_list; i_lru; i_dio_count i_writecount

So, neither i_wb_list or i_lru can be used.  It could be that some of
the atomic counters can be used but that requires collecting four such
counters consecutively.

> i_dquot (when QUOTA is on) i_private and more

If quota is off?  i_private maybe but it's not big enough.

> Even union with the "cgroup writback support" you
> want to add.

Again, these inodes can get dirtied.

I think unions are okay when lifetime rules clearly separate how the
field is used or the usages are contained in a logical unit but
overloading random fields which may be used across lifetime in a data
structure which is as widely used and abused as inode is likely to
lead to later headaches.

This is really something special and local {block|char}_dev are doing
which doens't have to interfere with anything else.  I think it's a
better approach to confine it to {block|char}_dev in the long term
even if that means carrying a bit more code.

Thanks.

-- 
tejun