Date: Tue, 20 Nov 2012 23:53:53 +0000
From: Alan Cox <alan@lxorguk.ukuu.org.uk>
To: Kees Cook <keescook@chromium.org>
Cc: linux-kernel@vger.kernel.org,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        ellyjones@chromium.org, Kay Sievers <kay@vrfy.org>,
        Roland Eggner <edvx1@systemanalysen.net>
Subject: Re: [PATCH v2] devtmpfs: mount with noexec and nosuid
Message-ID: <20121120235353.3092c8d2@pyramind.ukuu.org.uk>
In-Reply-To: <CAGXu5jJv58UGwsWM=+ZSnaNpymh6Sm6xSjc0JuT+QoZ9U2++rw@mail.gmail.com>
References: <20121120204238.GA19554@www.outflux.net>
	<20121120211312.57f1b63d@pyramind.ukuu.org.uk>
	<CAGXu5jJv58UGwsWM=+ZSnaNpymh6Sm6xSjc0JuT+QoZ9U2++rw@mail.gmail.com>
Face: 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
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1949
Lines: 47

> > 1. Why not just have your userspace mount -o remount the file system this
> > way already in early boot. (and if you trojanned boot that early then any
> > supposed security gain is already lost)
> 
> That's certainly possible, but I am hoping to avoid adding any extra
> boot time. The kernel is responsible for this mount, so its flags
> should be configurable, resulting in no time penalty anywhere.

You just broke my bullshitometer

It's a single syscall from your init binary, its microseconds.

> > 2. If you want to do this right then you need to work out what you are
> > trying to prevent. Your devtmpfs can force file permissions on the
> > underlying device nodes by having its own operation handling for chmod.
> >
> > At that point you can force permissions on anything that you want to
> > avoid floating around that filesystem with other rights, while not
> > touching it on device or directory nodes where the meaning is different.
> >
> > In its current form however it appears to be a kernel implementation of
> > "mount is too hard".
> 
> This change also stops mmap() with PROT_EXEC which a chmod handler
> wouldn't be able to do.

You don't want to stop mmap with PROT_EXEC on /dev/mem as that breaks a
load of stuff, you want to stop people adding stuff to that file system
and executing it.

Think about it more carefully - if I've got access to /dev/mem you
already lost.

The patch is as is nonsense. It's doable in all kernels (including
existing legacy ones) as a single syscall in your init code. That's
portable back compatible and works all over the place. Your patch
introduces a pointless config option to make things less compatible than
before. It's bogus.

Alan


--
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/