Re: [PATCH 03/33] mm: slub: add knowledge of reserve pages

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

[-- Attachment #1: Type: text/plain, Size: 2278 bytes --]

On Wed, 2007-10-31 at 21:46 +1100, Nick Piggin wrote:
> On Wednesday 31 October 2007 21:42, Peter Zijlstra wrote:
> > On Wed, 2007-10-31 at 14:37 +1100, Nick Piggin wrote:
> > > On Wednesday 31 October 2007 03:04, Peter Zijlstra wrote:
> > > > Restrict objects from reserve slabs (ALLOC_NO_WATERMARKS) to allocation
> > > > contexts that are entitled to it.
> > > >
> > > > Care is taken to only touch the SLUB slow path.
> > > >
> > > > This is done to ensure reserve pages don't leak out and get consumed.
> > >
> > > I think this is generally a good idea (to prevent slab allocators
> > > from stealing reserve). However I naively think the implementation
> > > is a bit overengineered and thus has a few holes.
> > >
> > > Humour me, what was the problem with failing the slab allocation
> > > (actually, not fail but just call into the page allocator to do
> > > correct waiting  / reclaim) in the slowpath if the process fails the
> > > watermark checks?
> >
> > Ah, we actually need slabs below the watermarks.
> 
> Right, I'd still allow those guys to allocate slabs. Provided they
> have the right allocation context, right?
> 
> 
> > Its just that once I 
> > allocated those slabs using __GFP_MEMALLOC/PF_MEMALLOC I don't want
> > allocation contexts that do not have rights to those pages to walk off
> > with objects.
> 
> And I'd prevent these ones from doing so.
> 
> Without keeping track of "reserve" pages, which doesn't feel
> too clean.

The problem with that is that once a slab was allocated with the right
allocation context, anybody can get objects from these slabs.


low memory, and empty slab:

task A                        task B

kmem_cache_alloc() = NULL

                              current->flags |= PF_MEMALLOC
                              kmem_cache_alloc() = obj
                              (slab != NULL)

kmem_cache_alloc() = obj
kmem_cache_alloc() = obj
kmem_cache_alloc() = obj


And now task A, who doesn't have the right permissions walks
away with all our reserve memory.

So we either reserve a page per object, which for 32 byte objects is a
large waste, or we stop anybody who doesn't have the right permissions
from obtaining objects. I took the latter approach.


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]