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: <20220722162934.1888835-1-mcgrof@kernel.org> <20220722162934.1888835-3-mcgrof@kernel.org>
In-Reply-To: <20220722162934.1888835-3-mcgrof@kernel.org>
From:   Muchun Song <songmuchun@bytedance.com>
Date:   Sat, 23 Jul 2022 22:17:12 +0800
Message-ID: <CAMZfGtV4wC-8uT47v-nYhEi82cqNVNf1CFMM8qRU7Nd3SMyv5w@mail.gmail.com>
Subject: Re: [PATCH 2/2] Documentation/filesystems/proc.rst: document procfs
 inode timestamps
To:     Luis Chamberlain <mcgrof@kernel.org>
Cc:     "Eric W. Biederman" <ebiederm@xmission.com>,
        Jonathan Corbet <corbet@lwn.net>,
        Kees Cook <keescook@chromium.org>,
        Iurii Zaikin <yzaikin@google.com>,
        Zhang Yuchen <zhangyuchen.lcr@bytedance.com>,
        David Howells <dhowells@redhat.com>,
        Deepa Dinamani <deepa.kernel@gmail.com>,
        Christoph Hellwig <hch@lst.de>,
        Linux Doc Mailing List <linux-doc@vger.kernel.org>,
        linux-api@vger.kernel.org,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Sat, Jul 23, 2022 at 12:29 AM Luis Chamberlain <mcgrof@kernel.org> wrote:
>
> The timestamps for procfs files are not well understood and can
> confuse users and developers [0] in particular for the timestamp
> for the start time or a process. Clarify what they mean and that
> they are a reflection of the ephemeral nature of the filesystem
> inodes.
>
> The procfs inodes are created when you first read them and then
> stuffed in the page cache. If the page cache and indodes are
> reclaimed they can be removed, and re-created with a new timestamp
> after read again. Document this little bit of tribal knowledge.
>
> [0] https://lkml.kernel.org/r/20220721081617.36103-1-zhangyuchen.lcr@bytedance.com
> Reported-by: Zhang Yuchen <zhangyuchen.lcr@bytedance.com>
> Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
> ---
>  Documentation/filesystems/proc.rst | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
>
> diff --git a/Documentation/filesystems/proc.rst b/Documentation/filesystems/proc.rst
> index 9fd5249f1a5f..9defe9af683a 100644
> --- a/Documentation/filesystems/proc.rst
> +++ b/Documentation/filesystems/proc.rst
> @@ -59,6 +59,15 @@ The proc  file  system acts as an interface to internal data structures in the
>  kernel. It  can  be  used to obtain information about the system and to change
>  certain kernel parameters at runtime (sysctl).
>
> +The proc files are dynamic in nature and allow for developers to make the
> +content to be changed each time a file is read. The proc files and directories
> +inodes are created when someone first reads a respective proc file or directory,
> +as such the timestamps of the proc files reflect this time. As with other
> +filesystems, these proc inodes can be removed through reclaim under memory
> +pressure and so the timestamps of the proc files can change if the proc files
> +are destroyed and re-created (echo 3 > /proc/sys/vm/drop_caches forces and
> +illustrate the reclaim of inodes and page cache).

Thanks for fixing this.

> +
>  First, we'll  take  a  look  at the read-only parts of /proc. In Chapter 2, we
>  show you how you can use /proc/sys to change settings.
>
> @@ -328,6 +337,13 @@ It's slow but very precise.
>                 system call
>    ============= ===============================================================
>
> +Note that the start_time inside the stat file is different than the timestamp
> +of the stat file itself. The timestamp of the stat file simply reflects the
> +first time the stat file was read. The proc inode for this file can be reclaimed
> +under memory pressure and be recreated after this and so the timestamp can
> +change. Userspace should rely on the start_time entry in the the stat file to
> +get a process start time.
> +

I'm not sure the value of those comments since the above description
is already enough to tell people the timestamp of /proc files or directories
can be changed in some cases, which already includes the case of
/proc/PID/stat.

If we really want to take /proc/PID/stat as an example to show the
timestamp is unstable, I think it is better to move those comments to the
above section where you explain why the timestamp can be changed .

Thanks.

>  The /proc/PID/maps file contains the currently mapped memory regions and
>  their access permissions.
>
> --
> 2.35.1
>