Re: Connecting to sockets on MNT_READONLY mounts?

[Al Viro <viro@ZenIV.linux.org.uk>]

On Thu, May 01, 2014 at 04:57:13PM -0700, Andy Lutomirski wrote:

> Suppose I bind-mount /usr into a private namespace with
> nosuid,nodev,ro.  How can you use it to attack anything?  The only
> thing I've thought of is to open fifos and connect to sockets.  I'm
> assuming that there's a pid namespace blocking ptrace and such and a
> network namespace blocking abstract sockets.

How many FIFOs and sockets are there in your /usr?  Here all I see
outside of /dev, /run and /tmp (across seven boxen; I can check more, but
I really doubt it'll catch anything) is the grand total of 4:
/lib/cryptsetup/passfifo
/var/lib/oprofile/opd_pipe
/var/lib/nfs/rpc_pipefs/nfs/clnt0/idmap
/var/lib/nfs/rpc_pipefs/gssd/clntXX/gssd

None of those in /usr and I don't believe that you seriously propose to
bind e.g. /lib/cryptsetup into your sandbox.  And while we are at it,
exposing host /usr is *not* a good idea - if nothing else, it gives
quite a bit of information about the versions of software installed on
the host.  Ability to watch atime of /usr/bin/* also might be interesting,
etc.

Do you, by any chance, plan to expose the host /tmp or /run?  Or
rpc_pipefs, for that matter...