Re: [PATCH RESEND] init/initramfs.c: make initramfs support pivot_root

From: Menglong Dong <menglong8.dong@gmail.com>
To: Luis Chamberlain <mcgrof@kernel.org>
Cc: Jan Kara <jack@suse.cz>, Jens Axboe <axboe@kernel.dk>,
	hare@suse.de, gregkh@linuxfoundation.org, tj@kernel.org,
	Menglong Dong <dong.menglong@zte.com.cn>,
	song@kernel.org, NeilBrown <neilb@suse.de>,
	Andrew Morton <akpm@linux-foundation.org>,
	wangkefeng.wang@huawei.com, f.fainelli@gmail.com, arnd@arndb.de,
	Barret Rhoden <brho@google.com>,
	Rasmus Villemoes <linux@rasmusvillemoes.dk>,
	mhiramat@kernel.org, Steven Rostedt <rostedt@goodmis.org>,
	Kees Cook <keescook@chromium.org>,
	vbabka@suse.cz, Alexander Potapenko <glider@google.com>,
	pmladek@suse.com, Chris Down <chris@chrisdown.name>,
	ebiederm@xmission.com, jojing64@gmail.com,
	LKML <linux-kernel@vger.kernel.org>,
	palmerdabbelt@google.com, linux-fsdevel@vger.kernel.org,
	Alexander Viro <viro@zeniv.linux.org.uk>
Subject: Re: [PATCH RESEND] init/initramfs.c: make initramfs support pivot_root
Date: Tue, 25 May 2021 14:09:08 +0800	[thread overview]
Message-ID: <CADxym3bbx7pXahoRb98ocQb1JYQMdagYDJ+XKe_RwD=7c6MCug@mail.gmail.com> (raw)
In-Reply-To: <20210525014304.GH4332@42.do-not-panic.com>

On Tue, May 25, 2021 at 9:43 AM Luis Chamberlain <mcgrof@kernel.org> wrote:
>
> >
> > This change seems transparent to users, which don't change the behavior
> > of initramfs.
>
> Are we sure there nothing in the kernel that can regress with this
> change? Are you sure? How sure?
>
> > However, it seems more reasonable to make it a kconfig option.
> > I'll do it in the v2 of the three patches I sended.
>
> I'm actually quite convinced now this is a desirable default *other*
> than the concern if this could regress. I recently saw some piece of
> code fetching for the top most mount, I think it was on the
> copy_user_ns() path or something like that, which made me just
> consider possible regressions for heuristics we might have forgotten
> about.
>
> I however have't yet had time to review the path I was concerned for
> yet.

Yeah, I'm sure...probably. The way I create and mount 'user root' is
almost the same to block root device. When it comes to block
device, such as hda, what kernel do is:

/* This will mount block device on '/root' and chdir to '/root' */
prepare_namespace->mount_root->mount_block_root->do_mount_root;

/* This will move the block device mounted on '/root' to '/' */
init_mount(".", "/", NULL, MS_MOVE, NULL);

/* This will change the root to current dir, which is the root of block
 * device.
 */
init_chroot(".")

And these steps are exactly what I do with 'user root'. However, I'm
not totally sure. For safety, I'll make it into a kconfig option. Is
it acceptable to make it enabled by default?

Thanks!
Menglong Dong