linux-kernel - Re: [kernel-hardening] Re: [PATCH v7 2/2] security: tty: make TIOCSTI ioctl require CAP_SYS

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20170530235106.11aab25c@alans-desktop>
Date:   Tue, 30 May 2017 23:51:06 +0100
From:   Alan Cox <gnomes@...rguk.ukuu.org.uk>
To:     Matt Brown <matt@...tt.com>
Cc:     Casey Schaufler <casey@...aufler-ca.com>,
        Boris Lukashev <blukashev@...pervictus.com>,
        Greg KH <gregkh@...uxfoundation.org>,
        "Serge E. Hallyn" <serge@...lyn.com>,
        Kees Cook <keescook@...omium.org>,
        kernel-hardening@...ts.openwall.com,
        linux-security-module <linux-security-module@...r.kernel.org>,
        linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [kernel-hardening] Re: [PATCH v7 2/2] security: tty: make
 TIOCSTI ioctl require CAP_SYS_ADMIN

On Tue, 30 May 2017 12:28:59 -0400
Matt Brown <matt@...tt.com> wrote:

> On 5/30/17 8:24 AM, Alan Cox wrote:
> > Look there are two problems here
> > 
> > 1. TIOCSTI has users  
> 
> I don't see how this is a problem.

Which is unfortunate. To start with if it didn't have users we could just
delete it.

> > 
> > 2. You don't actually fix anything
> > 
> > The underlying problem is that if you give your tty handle to another
> > process which you don't trust you are screwed. It's fundamental to the
> > design of the Unix tty model and it's made worse in Linux by the fact
> > that we use the tty descriptor to access all sorts of other console state
> > (which makes a ton of sense).
> > 
> > Many years ago a few people got this wrong. All those apps got fixes back
> > then. They allocate a tty/pty pair and create a new session over that.
> > The potentially hostile other app only gets to screw itself.
> >   
> 
> Many years ago? We already got one in 2017, as well as a bunch last year.
> See: https://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=tiocsti

All the apps got fixed at the time. The fact the next generation of
forgot to learn from it is unfortunate but hardly new. Also every single
one of those that exposes a tty in that way allows other annoying
behaviours via other ioctl interfaces so none of them would have been
properly mitigated.

If you really want to do that particular bit of snake oiling then you can
use the existing SELinux, seccomp and related interfaces. They can even
do the job properly by whitelisting or blocking long lists of ioctls.

> This protections seeks to protect users from programs that don't do things
> correctly. Rather than killing bugs, this feature attempts to kill an entire
> bug class that shows little sign of slowing down in the world of containers and
> sandboxes.

Well maybe the people writing them need to learn what they are doing and
stop passing random file descriptors into their container (I've even seen
people handing X file handles into their 'container').

The kernel can do some things to help programmers but it can't stop
people writing crap. Anyone writing code that crosses security boundaries
should have at least a vague idea of what they are doing.

The only way you'd actually really prevent this would be to magically
open a new pty/tty pair and substitute the file handlers plus a data
copying thread when someone created a namespace.

Now you can actually do that with the ptrace functionality in seccomp but
it would still be fairly insane to expect the kernel to handle.

Alan
[Actually even more sensible would be to revert the entire sorry
container mess and use VMs but it's a bit late for that ;-)]