Date: Wed, 19 Dec 2007 08:10:04 -0600
From: "Serge E. Hallyn" <serue@us.ibm.com>
To: Pavel Emelyanov <xemul@openvz.org>
Cc: Oren Laadan <orenl@cs.columbia.edu>, "Serge E. Hallyn" <serue@us.ibm.com>,
       Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
       linux-fsdevel@vger.kernel.org,
       Linux Containers <containers@lists.osdl.org>,
       linux-kernel@vger.kernel.org
Subject: Re: [patch 1/2] [RFC] Simple tamper-proof device filesystem.
Message-ID: <20071219141004.GA31694@sergelap.austin.ibm.com>
References: <20071216080628.061470932@I-love.SAKURA.ne.jp> <200712161944.HEI26071.MOtOFLVHFSQFOJ@I-love.SAKURA.ne.jp> <200712161956.BJE32406.FOOHtQJLMFOSVF@I-love.SAKURA.ne.jp> <20071217194802.GA14156@sergelap.austin.ibm.com> <200712180003.lBI03N7F092396@www262.sakura.ne.jp> <20071218003955.GA27048@sergelap.austin.ibm.com> <476724E3.6060901@cs.columbia.edu> <20071218020933.GA28745@sergelap.austin.ibm.com> <476738A0.9010601@cs.columbia.edu> <4768E7B2.40203@openvz.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <4768E7B2.40203@openvz.org>
User-Agent: Mutt/1.5.16 (2007-06-09)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2608
Lines: 56

Quoting Pavel Emelyanov (xemul@openvz.org):
> Oren Laadan wrote:
> > Serge E. Hallyn wrote:
> >> Quoting Oren Laadan (orenl@cs.columbia.edu):
> >>> I hate to bring this again, but what if the admin in the container
> >>> mounts an external file system (eg. nfs, usb, loop mount from a file,
> >>> or via fuse), and that file system already has a device that we would
> >>> like to ban inside that container ?
> >> Miklos' user mount patches enforced that if !capable(CAP_MKNOD),
> >> then mnt->mnt_flags |= MNT_NODEV.  So that's no problem.
> > 
> > Yes, that works to disallow all device files from a mounted file system.
> > 
> > But it's a black and white thing: either they are all banned or allowed;
> > you can't have some devices allowed and others not, depending on type
> > A scenario where this may be useful is, for instance, if we some apps in
> > the container to execute withing a pre-made chroot (sub)tree within that
> > container.
> > 
> >> But that's been pulled out of -mm! ?  Crap.
> >>
> >>> Since anyway we will have to keep a white- (or black-) list of devices
> >>> that are permitted in a container, and that list may change even change
> >>> per container -- why not enforce the access control at the VFS layer ?
> >>> It's safer in the long run.
> >> By that you mean more along the lines of Pavel's patch than my whitelist
> >> LSM, or you actually mean Tetsuo's filesystem (i assume you don't mean that
> >> by 'vfs layer' :), or something different entirely?
> > 
> > :)
> > 
> > By 'vfs' I mean at open() time, and not at mount(), or mknod() time.
> > Either yours or Pavel's; I tend to prefer not to use LSM as it may
> > collide with future security modules.
> 
> Oren, AFAIS you've seen my patches for device access controller, right?
> 
> Maybe we can revisit the issue then and try to come to agreement on what
> kind of model and implementation we all want?

That would be great, Pavel.  I do prefer your solution over my LSM, so
if we can get an elegant block device control right in the vfs code that
would be my preference.

The only thing that makes me keep wanting to go back to an LSM is the
fact that the code defining the whitelist seems out of place in the vfs.
But I guess that's actually separated into a modular cgroup, with the
actual enforcement built in at the vfs.  So that's really the best
solution.

-serge
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/