Re: [PATCH 12/18] shared mount handling: bind and rbind

[Rob Landley <rob@landley.net>]

On Wednesday 16 November 2005 02:19, Miklos Szeredi wrote:
> > > This is why we have "pivot_root()" and "chroot()", which can both be
> > > used to do what you want to do. You mount the new root somewhere else,
> > > and then you chroot (or pivot-root) to it. And THEN you do 'chdir("/")'
> > > to move the cwd into the new root too (and only at that point have you
> > > "lost" the old root - although you can actually get it back if you have
> > > some file descriptor open to it).
> >
> > Wouldn't this constitute a security flaw?
> >
> > Shouldn't chroot jail you?
>
> No, chroot should just change the root.
>
> If you don't want to be able to get back the old root, just close all
> file descriptors _in addition_ to chroot() and chdir().

If you try the chdir by filedescriptor trick on the stdin/stdout/stderr fed 
into PID 1 when it's started up by the kernel, which filesystem do you wind 
up in?  (rootfs?)

I ask because switch_root redoes those to point to /dev/console from the real 
root (presumably for security reasons), and this happens _before_ the init on 
the real root gets called, and thus before the real root gets to populate 
its' own dynamic /dev.

I suppose initramfs could make a temporary /dev, do the mknods for console and 
the real root, and then mount --move this tmpdir to the real root's /dev once 
that's available (and then let the real root's udev populate it the rest of 
the way).  Or the real root could have a hard /dev/console living in the 
directory that's going to get overmounted by tmpfs later.  Or just leave 
initramfs accessible until init can switch consoles...

Sigh.  I need to document the requirements here...

Rob