Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230502160839.361544-1-roman.gushchin@linux.dev> <20230502160839.361544-2-roman.gushchin@linux.dev>
In-Reply-To: <20230502160839.361544-2-roman.gushchin@linux.dev>
From:   Yosry Ahmed <yosryahmed@google.com>
Date:   Tue, 2 May 2023 13:21:30 -0700
Message-ID: <CAJD7tkbPUNv+tNDwnsmyXdOStCoA91dBCX+bRDQx5pDQx4bmpw@mail.gmail.com>
Subject: Re: [PATCH v2 2/2] mm: memcg: use READ_ONCE()/WRITE_ONCE() to access stock->cached
To:     Roman Gushchin <roman.gushchin@linux.dev>
Cc:     linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Michal Hocko <mhocko@kernel.org>,
        Shakeel Butt <shakeelb@google.com>,
        Muchun Song <muchun.song@linux.dev>,
        linux-kernel@vger.kernel.org, Dmitry Vyukov <dvyukov@google.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, May 2, 2023 at 9:09=E2=80=AFAM Roman Gushchin <roman.gushchin@linux=
.dev> wrote:
>
> A memcg pointer in the percpu stock can be accessed by drain_all_stock()
> from another cpu in a lockless way.
> In theory it might lead to an issue, similar to the one which has been
> discovered with stock->cached_objcg, where the pointer was zeroed
> between the check for being NULL and dereferencing.
> In this case the issue is unlikely a real problem, but to make it
> bulletproof and similar to stock->cached_objcg, let's annotate all
> accesses to stock->cached with READ_ONCE()/WTRITE_ONCE().

Is it time to rename that to cached_memcg? :)

Anyway, same comment as patch 1 about annotating all reads with
READ_ONCE() vs. singling out the racy read.

>
> Signed-off-by: Roman Gushchin <roman.gushchin@linux.dev>
> Cc: Dmitry Vyukov <dvyukov@google.com>
> Cc: Yosry Ahmed <yosryahmed@google.com>
> Cc: Shakeel Butt <shakeelb@google.com>
> ---
>  mm/memcontrol.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index c823c35c2ed4..1e364ad495a3 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -2275,7 +2275,7 @@ static bool consume_stock(struct mem_cgroup *memcg,=
 unsigned int nr_pages)
>         local_lock_irqsave(&memcg_stock.stock_lock, flags);
>
>         stock =3D this_cpu_ptr(&memcg_stock);
> -       if (memcg =3D=3D stock->cached && stock->nr_pages >=3D nr_pages) =
{
> +       if (memcg =3D=3D READ_ONCE(stock->cached) && stock->nr_pages >=3D=
 nr_pages) {
>                 stock->nr_pages -=3D nr_pages;
>                 ret =3D true;
>         }
> @@ -2290,7 +2290,7 @@ static bool consume_stock(struct mem_cgroup *memcg,=
 unsigned int nr_pages)
>   */
>  static void drain_stock(struct memcg_stock_pcp *stock)
>  {
> -       struct mem_cgroup *old =3D stock->cached;
> +       struct mem_cgroup *old =3D READ_ONCE(stock->cached);
>
>         if (!old)
>                 return;
> @@ -2303,7 +2303,7 @@ static void drain_stock(struct memcg_stock_pcp *sto=
ck)
>         }
>
>         css_put(&old->css);
> -       stock->cached =3D NULL;
> +       WRITE_ONCE(stock->cached, NULL);

Is it me or can we call drain_stock() from memcg_hotplug_cpu_dead()
without holding the lock, unlike all other callers. Is this a problem?

>  }
>
>  static void drain_local_stock(struct work_struct *dummy)
> @@ -2338,10 +2338,10 @@ static void __refill_stock(struct mem_cgroup *mem=
cg, unsigned int nr_pages)
>         struct memcg_stock_pcp *stock;
>
>         stock =3D this_cpu_ptr(&memcg_stock);
> -       if (stock->cached !=3D memcg) { /* reset if necessary */
> +       if (READ_ONCE(stock->cached) !=3D memcg) { /* reset if necessary =
*/
>                 drain_stock(stock);
>                 css_get(&memcg->css);
> -               stock->cached =3D memcg;
> +               WRITE_ONCE(stock->cached, memcg);
>         }
>         stock->nr_pages +=3D nr_pages;
>
> @@ -2383,7 +2383,7 @@ static void drain_all_stock(struct mem_cgroup *root=
_memcg)
>                 bool flush =3D false;
>
>                 rcu_read_lock();
> -               memcg =3D stock->cached;
> +               memcg =3D READ_ONCE(stock->cached);
>                 if (memcg && stock->nr_pages &&
>                     mem_cgroup_is_descendant(memcg, root_memcg))
>                         flush =3D true;
> --
> 2.40.1
>