Re: Connecting to sockets on MNT_READONLY mounts?

On Thu, May 1, 2014 at 4:51 PM, Al Viro [off-list ref] wrote:

On Thu, May 01, 2014 at 04:00:49PM -0700, Andy Lutomirski wrote:

quoted

On Thu, May 1, 2014 at 3:34 PM, Al Viro [off-list ref] wrote:

quoted

On Thu, May 01, 2014 at 03:20:00PM -0700, Andy Lutomirski wrote:

quoted

Is it supposed to work?

Why the hell not?  Same as opening a device node on r/o filesystem for
write, or doing the same with FIFO.

You can't bind a socket on a read-only fs, so I thought it was a fair question.

I'll write a patch to add MS_NOIPCCONNECT and MNT_NOIPCCONNECT to
block connect on unix sockets and open on fifos.  This will be useful
for sandboxes that want to prevent sandboxed programs from accessing
undesirable parts of the outside world.

Sigh...  Don't expose those FIFOs et.al. to them, then.
mount --bind /dev/null <pathname>
as part of setting the sucker up.  And if you *are* blindly exposing the
host filesystems to them wholesale, sockets and fifos are the least of your
problems, even if you do that read-only.

Why?

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.

--Andy