Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 3 Jun 2021 12:33:58 +0100
From:   Chris Down <chris@chrisdown.name>
To:     legion@kernel.org
Cc:     LKML <linux-kernel@vger.kernel.org>,
        Linux Containers <containers@lists.linux.dev>,
        Linux Containers <containers@lists.linux-foundation.org>,
        Linux FS Devel <linux-fsdevel@vger.kernel.org>,
        linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
        Christian Brauner <christian.brauner@ubuntu.com>,
        "Eric W . Biederman" <ebiederm@xmission.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Michal Hocko <mhocko@kernel.org>
Subject: Re: [PATCH v1] proc: Implement /proc/self/meminfo
Message-ID: <YLi+JoBwfLtqVGiP@chrisdown.name>
References: <ac070cd90c0d45b7a554366f235262fa5c566435.1622716926.git.legion@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Disposition: inline
In-Reply-To: <ac070cd90c0d45b7a554366f235262fa5c566435.1622716926.git.legion@kernel.org>
User-Agent: Mutt/2.0.7 (481f3800) (2021-05-04)
Precedence: bulk

Hi Alexey,

legion@kernel.org writes:
>From: Alexey Gladkov <legion@kernel.org>
>The /proc/meminfo contains information regardless of the cgroups
>restrictions. This file is still widely used [1]. This means that all
>these programs will not work correctly inside container [2][3][4]. Some
>programs try to respect the cgroups limits, but not all of them
>implement support for all cgroup versions [5].
>
>Correct information can be obtained from cgroups, but this requires the
>cgroups to be available inside container and the correct version of
>cgroups to be supported.

Then they should add support for it. We already export these metrics as part of 
cgroups and plenty of applications like Docker, podman, containerd, systemd, 
runc, etc already support it.

Putting stuff in /proc to get around the problem of "some other metric I need 
might not be exported to a container" is not a very compelling argument. If 
they want it, then export it to the container...

Ultimately, if they're going to have to add support for a new 
/proc/self/meminfo file anyway, these use cases should just do it properly 
through the already supported APIs.

>+	for_each_online_node(nid)
>+		mem_cgroup_nr_pages(memcg, nid, mi->pages);
>+
>+	mi->slab_reclaimable = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B);
>+	mi->slab_unreclaimable = memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B);
>+	mi->cached = memcg_page_state(memcg, NR_FILE_PAGES);
>+	mi->swapcached = memcg_page_state(memcg, NR_SWAPCACHE);
>+	mi->anon_pages = memcg_page_state(memcg, NR_ANON_MAPPED);
>+	mi->mapped = memcg_page_state(memcg, NR_FILE_MAPPED);
>+	mi->nr_pagetable = memcg_page_state(memcg, NR_PAGETABLE);
>+	mi->dirty_pages = memcg_page_state(memcg, NR_FILE_DIRTY);
>+	mi->writeback_pages = memcg_page_state(memcg, NR_WRITEBACK);
>+}

This presents an extraordinarily confusing API. A cgroup can contain more than 
one process, so it's not right to present this information as "meminfo" in 
/proc/self when these statistics may not have any relation to the current task 
under question.