Re: SLUB defrag pull request?

[Miklos Szeredi]

On Wed, 22 Oct 2008, Christoph Lameter wrote:
> On Wed, 22 Oct 2008, Miklos Szeredi wrote:
> 
> > On Tue, 21 Oct 2008, Christoph Lameter wrote:
> >> The only way that a secure reference can be established is if the
> >> slab page is locked. That requires a spinlock. The slab allocator
> >> calls the get() functions while the slab lock guarantees object
> >> existence. Then locks are dropped and reclaim actions can start with
> >> the guarantee that the slab object will not suddenly vanish.
> >
> > Yes, you've made up your mind, that you want to do it this way.  But
> > it's the _wrong_ way, this "want to get a secure reference for use
> > later" leads to madness when applied to dentries or inodes.  Try for a
> > minute to think outside this template.
> >
> > For example dcache_lock will protect against dentries moving to/from
> > d_lru.  So you can do this:
> >
> >  take dcache_lock
> >  check if d_lru is non-empty
> 
> The dentry could have been freed even before we take the dcache_lock. We 
> cannot access d_lru without a stable reference to the dentry.

Why?  The kmem_cache_free() doesn't touch the contents of the object,
does it?

> >  take sb->s_umount
> >  free dentry
> >  release sb->s_umount
> >  release dcache_lock
> >
> > Yeah, locking will be more complicated in reality.  Still, much less
> > complicated than trying to do the same across two separate phases.
> >
> > Why can't something like that work?
> 
> Because the slab starts out with a series of objects left in a slab. It 
> needs to do build a list of objects etc in a way that is independent as 
> possible from the user of the slab page. It does that by locking the slab 
> page so that free operations stall until the reference has been 
> established. If it would not be shutting off frees then the objects could 
> vanish under us.

It doesn't matter.  All we care about is that the dentry is on the
lru: it's cached but unused.  Every other state (being created,
active, being freed, freed) is uninteresting.

> We could also avoid frees by calling some cache specific method that locks 
> out frees before and after. But then frees would stall everywhere and 
> every slab cache would have to check a global lock before freeing objects 
> (there would be numerous complications with RCU free etc etc).
> 
> Slab defrag only stops frees on a particular slab page.
> 
> The slab defrag approach also allows the slab cache (dentry or inodes 
> here) to do something else than free the object. It would be possible f.e. 
> to move the object by allocating a new entry and moving the information to 
> the new dentry. That would actually be better since it would preserve the 
> objects and just move them into the same slab page.

Sure, and all that is possible without doing this messy 2 phase thing.
Unless I'm still missing something obvious...

Miklos
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/