Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20191121070613.4286-1-hu1.chen@intel.com>
In-Reply-To: <20191121070613.4286-1-hu1.chen@intel.com>
From:   Miklos Szeredi <miklos@szeredi.hu>
Date:   Thu, 21 Nov 2019 11:03:18 +0100
Message-ID: <CAJfpegtK_S3K0j_qP6x3+qKBPdLag+ayCWHAakJvMtVXMdmXtw@mail.gmail.com>
Subject: Re: [PATCH] proc: align mnt_id in /proc/pid/fdinfo and /proc/pid/mountinfo
To:     "Chen, Hu" <hu1.chen@intel.com>
Cc:     Andrey Vagin <avagin@openvz.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Alexey Dobriyan <adobriyan@gmail.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Nov 21, 2019 at 8:28 AM Chen, Hu <hu1.chen@intel.com> wrote:
>
> For Android application process, we found that the mnt_id read from
> /proc/pid/fdinfo doesn't exist in /proc/pid/mountinfo. Thus CRIU fails
> to dump such process and it complains
>
> "(00.019206) Error (criu/files-reg.c:1299): Can't lookup mount=42 for
> fd=-3 path=/data/dalvik-cache/x86_64/system@framework@boot.art"
>
> This is due to how Android application is launched. In Android, there is
> a special process called Zygote which handles the forking of each new
> application process:
> 0. Zygote opens and maps some files, for example
>    "/data/dalvik-cache/x86_64/system@framework@boot.art" in its current
>    mount namespace, say "old mnt ns".
> 1. Zygote waits for the request to fork a new application.
> 2. Zygote gets a request, it forks and run the new process in a new
>    mount namespace, say "new mnt ns".
>
> The file opened in step 0 ties to the mount point in "old mnt ns". The
> mnt_id of that mount is listed in /proc/pid/fdinfo. However,
> /proc/pid/mountinfo points to current ns, i.e., "new mnt ns".
>
> Althgouh this issue is exposed in Android, we believe it's generic.
> Prcoess may open files and enter new mnt ns.
>
> To address it, this patch searches the mirror mount in current ns with
> MAJOR and MINOR and shows the mirror's mnt_id.

This is a hack.   I suggest instead to add a new line to fdinfo with
the MAJOR:MINOR number of the device.

Thanks,
Miklos