Re: owner field in `struct fs'

[Rusty Russell]

In message <[e-mail]> you write:
> 
> [e-mail] said:
> >  Adding a field to every registerable structure is gross.  The
> > responsibility for getting modules right should be with the person
> > writing the modules.  The rules are simple; and people w/o module
> > support don't pay the penalty for the extra field everywhere... 
> 
> It's possible to have the module use count maintained from outside the 
> driver without having the extra structure.
> 
> Take a look at put_module_symbol() in kernel/module.c. It doesn't require 
> any argument other than the address of the function, and it can find the 
> module quite happily.

Cute, but paying that penalty for every open() on a filesystem which
isn't a module (true for 99%, ext2 root) seems a little harsh.  Hmm,
could be cached, but Al's solution has the advantage of not making me
want to barf...

> While we're at it - can we get rid of MOD_INC_USE_COUNT and 
> __MOD_INC_USE_COUNT altogether to force people to be aware of the need for 
> try_inc_mod_count and checking its return value? First thing in 2.5 along 
> with removing sleep_on() ?

Are you volunteering to go through net/ipv4/netfilter/ and change them
all?  Because that'd be really great!

For 2.5, I'd prefer the two-stage cleanup solution, because it is easy
for module writers to understand (`deactivate' deregisters so that
module counts won't increase after sys_del_module syncs bhs, `cleanup'
frees resources), allows module counts to increment inside interrupt
handlers, is simple to write assertions to check behaviour, and for
code like mine, avoids the ugly `loop and schedule()' in cleanup().

But that's a long way away, and getting furthur,
Rusty.
--
Hacking time.