Re: [PATCH] block: convert tasklets to use new tasklet_setup() API

[Kees Cook]

On Tue, Aug 18, 2020 at 01:00:33PM -0700, James Bottomley wrote:
> On Mon, 2020-08-17 at 13:02 -0700, Jens Axboe wrote:
> > On 8/17/20 12:48 PM, Kees Cook wrote:
> > > On Mon, Aug 17, 2020 at 12:44:34PM -0700, Jens Axboe wrote:
> > > > On 8/17/20 12:29 PM, Kees Cook wrote:
> > > > > On Mon, Aug 17, 2020 at 06:56:47AM -0700, Jens Axboe wrote:
> > > > > > On 8/17/20 2:15 AM, Allen Pais wrote:
> > > > > > > From: Allen Pais <allen.lkml@xxxxxxxxx>
> > > > > > > 
> > > > > > > In preparation for unconditionally passing the
> > > > > > > struct tasklet_struct pointer to all tasklet
> > > > > > > callbacks, switch to using the new tasklet_setup()
> > > > > > > and from_tasklet() to pass the tasklet pointer explicitly.
> > > > > > 
> > > > > > Who came up with the idea to add a macro 'from_tasklet' that
> > > > > > is just container_of? container_of in the code would be
> > > > > > _much_ more readable, and not leave anyone guessing wtf
> > > > > > from_tasklet is doing.
> > > > > > 
> > > > > > I'd fix that up now before everything else goes in...
> > > > > 
> > > > > As I mentioned in the other thread, I think this makes things
> > > > > much more readable. It's the same thing that the timer_struct
> > > > > conversion did (added a container_of wrapper) to avoid the
> > > > > ever-repeating use of typeof(), long lines, etc.
> > > > 
> > > > But then it should use a generic name, instead of each sub-system 
> > > > using some random name that makes people look up exactly what it
> > > > does. I'm not huge fan of the container_of() redundancy, but
> > > > adding private variants of this doesn't seem like the best way
> > > > forward. Let's have a generic helper that does this, and use it
> > > > everywhere.
> > > 
> > > I'm open to suggestions, but as things stand, these kinds of
> > > treewide
> > 
> > On naming? Implementation is just as it stands, from_tasklet() is
> > totally generic which is why I objected to it. from_member()? Not
> > great with naming... But I can see this going further and then we'll
> > suddenly have tons of these. It's not good for readability.
> 
> Since both threads seem to have petered out, let me suggest in
> kernel.h:
> 
> #define cast_out(ptr, container, member) \
> 	container_of(ptr, typeof(*container), member)
> 
> It does what you want, the argument order is the same as container_of
> with the only difference being you name the containing structure
> instead of having to specify its type.

I like this! Shall I send this to Linus to see if this can land in -rc2
for use going forward?

-- 
Kees Cook