Re: Antw: [EXT] Re: [systemd-devel] Creating executable device nodes in /dev?

[Jarkko Sakkinen <jarkko@kernel.org>]

On Tue, Dec 15, 2020 at 06:19:09AM +0200, Jarkko Sakkinen wrote:
> On Mon, Dec 14, 2020 at 08:25:50AM +0100, Ulrich Windl wrote:
> > >>> Topi Miettinen <toiwoton@gmail.com> schrieb am 11.12.2020 um 12:46 in
> > Nachricht
> > <27796c04-249e-6cf0-c3e1-0fd657a82f9c@gmail.com>:
> > > On 11.12.2020 12.46, Jarkko Sakkinen wrote:
> > >> On Wed, Dec 09, 2020 at 10:35:21AM +0200, Topi Miettinen wrote:
> > >>> On 9.12.2020 2.15, Jarkko Sakkinen wrote:
> > >>>> On Wed, Dec 09, 2020 at 01:15:27AM +0200, Topi Miettinen wrote:
> > >>>>>>>> As  a further argument, I just did this on a Fedora system:
> > >>>>>>>> $ find /dev ‑perm /ugo+x ‑a \! ‑type d ‑a \! ‑type l
> > >>>>>>>> No results.  So making /dev noexec doesn't seem to have any benefit.
> > >>>>>>>
> > >>>>>>> It's no surprise that there aren't any executables in /dev since
> > >>>>>>> removing MAKEDEV ages ago. That's not the issue, which is that
> > >>>>>>> /dev is a writable directory (for UID=0 but no capabilities are
> > >>>>>>> needed) and thus a potential location for constructing unapproved
> > >>>>>>> executables if it is also mounted exec (W^X).
> > >>>>>>
> > >>>>>> UID 0 can just change mount options, though, unless SELinux or similar
> > is 
> > > used. And SELinux can protect /dev just fine without noexec.
> > >>>>>
> > >>>>> Well, mounting would need CAP_SYS_ADMIN in addition to UID 0. Also
> > SELinux
> > >>>>> is not universal and the policies might not contain all users or
> > services.
> > >>>>>
> > >>>>> ‑Topi
> > >>>>
> > >>>> What's the data that supports having noexec /dev anyway? With root
> > >>>> access I can then just use something else like /dev/shm mount.
> > >>>>
> > >>>> Has there been out in the wild real world cases that noexec mount
> > >>>> of would have prevented?
> > >>>>
> > >>>> For me this sounds a lot just something that "feels more secure"
> > >>>> without any measurable benefit. Can you prove me wrong?
> > >>>
> > >>> I don't think security works that way. An attacker has various methods to
> > >>> choose from, some are more interesting than others. The case where
> > rw,exec
> > >>> /dev would be interesting would imply that easier or more common avenues
> > >>> would be blocked, for example rw,exec /dev/shm, /tmp, /var/tmp, or
> > >>> /run/user/$UID/ for user. Also fileless malware with pure ROP/JOP
> > approach
> > >>> with no need for any file system access is getting more common. It does
> > not
> > >>> mean that it would not be prudent to block the relatively easy approaches
> > >>> too, including /dev.
> > >> 
> > >> What if we add a new mount option "chrexec", which allows exec
> > >> for character devices (S_IFCHR).
> > > 
> > > I think devices are a bad match for SGX because devices haven't been 
> > > executable and SGX is actually an operation for memory. So something 
> > > like memfd_create(, MFD_SGX) or mmap(,, PROT_READ|PROT_EXEC|PROT_SGX) 
> > > would be much more natural. Even better would be something that 
> > > conceptully also works for AMD version (either with the same flags or 
> > > MFD_SGX / MFD_whatever_the_AMD_version_is).
> > 
> > +1
> 
> SGX reserved memory from kernel's point of view is IO memory.
> 
> Mapping SGX to memfd would not be a great idea, as it does not map
> into concept of anonymous file backed by regular memory.
> 
> A device file is very natural match actually. We have ioctl API for
> uploading enclave pages during the build procedure to the enclave and
> custom #PF handler. Conceptually it's a lot like video memory or such
> special device specific memory area.
> 
> There's no AMD equivalent of this technology.

Anyway, I take a not on "PROT_SGX" as one of the ways sort this out in
the future. That would at least fit what we have. Thanks for all the
feedback.

/Jarkko