linux-kernel - Re: [PATCH 1/3] mm: memcg: remote memcg charging for kmem allocations

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 19 Jun 2018 12:24:29 -0400
From:   Johannes Weiner <hannes@...xchg.org>
To:     Shakeel Butt <shakeelb@...gle.com>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Michal Hocko <mhocko@...nel.org>,
        Vladimir Davydov <vdavydov.dev@...il.com>,
        Jan Kara <jack@...e.com>, Greg Thelen <gthelen@...gle.com>,
        linux-kernel@...r.kernel.org, cgroups@...r.kernel.org,
        linux-fsdevel@...r.kernel.org, linux-mm@...ck.org,
        Jan Kara <jack@...e.cz>, Amir Goldstein <amir73il@...il.com>,
        Christoph Lameter <cl@...ux.com>,
        Pekka Enberg <penberg@...nel.org>,
        David Rientjes <rientjes@...gle.com>,
        Joonsoo Kim <iamjoonsoo.kim@....com>,
        Mel Gorman <mgorman@...e.de>, Vlastimil Babka <vbabka@...e.cz>,
        Alexander Viro <viro@...iv.linux.org.uk>
Subject: Re: [PATCH 1/3] mm: memcg: remote memcg charging for kmem allocations

On Mon, Jun 18, 2018 at 10:13:25PM -0700, Shakeel Butt wrote:
> @@ -248,6 +248,30 @@ static inline void memalloc_noreclaim_restore(unsigned int flags)
>  	current->flags = (current->flags & ~PF_MEMALLOC) | flags;
>  }
>  
> +#ifdef CONFIG_MEMCG
> +static inline struct mem_cgroup *memalloc_memcg_save(struct mem_cgroup *memcg)
> +{
> +	struct mem_cgroup *old_memcg = current->target_memcg;
> +
> +	current->target_memcg = memcg;
> +	return old_memcg;
> +}
> +
> +static inline void memalloc_memcg_restore(struct mem_cgroup *memcg)
> +{
> +	current->target_memcg = memcg;
> +}

The use_mm() and friends naming scheme would be better here:
memalloc_use_memcg(), memalloc_unuse_memcg(), current->active_memcg

> @@ -375,6 +376,27 @@ static __always_inline void kfree_bulk(size_t size, void **p)
>  	kmem_cache_free_bulk(NULL, size, p);
>  }
>  
> +/*
> + * Calling kmem_cache_alloc_memcg implicitly assumes that the caller wants
> + * a __GFP_ACCOUNT allocation. However if memcg is NULL then
> + * kmem_cache_alloc_memcg is same as kmem_cache_alloc.
> + */
> +static __always_inline void *kmem_cache_alloc_memcg(struct kmem_cache *cachep,
> +						    gfp_t flags,
> +						    struct mem_cgroup *memcg)
> +{
> +	struct mem_cgroup *old_memcg;
> +	void *ptr;
> +
> +	if (!memcg)
> +		return kmem_cache_alloc(cachep, flags);
> +
> +	old_memcg = memalloc_memcg_save(memcg);
> +	ptr = kmem_cache_alloc(cachep, flags | __GFP_ACCOUNT);
> +	memalloc_memcg_restore(old_memcg);
> +	return ptr;

I'm not a big fan of these functions as an interface because it
implies that kmem_cache_alloc() et al wouldn't charge a memcg - but
they do, just using current's memcg.

It's also a lot of churn to duplicate all the various slab functions.

Can you please inline the save/restore (or use/unuse) functions into
the callsites? If you make them handle NULL as parameters, it merely
adds two bracketing lines around the allocation call in the callsites,
which I think would be better to understand - in particular with a
comment on why we are charging *that* group instead of current's.

> +static __always_inline struct mem_cgroup *get_mem_cgroup(
> +				struct mem_cgroup *memcg, struct mm_struct *mm)
> +{
> +	if (unlikely(memcg)) {
> +		rcu_read_lock();
> +		if (css_tryget_online(&memcg->css)) {
> +			rcu_read_unlock();
> +			return memcg;
> +		}
> +		rcu_read_unlock();
> +	}
> +	return get_mem_cgroup_from_mm(mm);
> +}
> +
>  /**
>   * mem_cgroup_iter - iterate over memory cgroup hierarchy
>   * @root: hierarchy root
> @@ -2260,7 +2274,7 @@ struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
>  	if (current->memcg_kmem_skip_account)
>  		return cachep;
>  
> -	memcg = get_mem_cgroup_from_mm(current->mm);
> +	memcg = get_mem_cgroup(current->target_memcg, current->mm);

get_mem_cgroup_from_current(), which uses current->active_memcg if set
and current->mm->memcg otherwise, would be a nicer abstraction IMO.