[RFC PATCH v19 1/5] exec: Add a new AT_CHECK flag to execveat(2)

[Mickaël Salaün]

On Wed, Jul 17, 2024 at 06:51:11PM -0700, Jeff Xu wrote:
> On Wed, Jul 17, 2024 at 3:00 AM Mickaël Salaün <mic at digikod.net> wrote:
> >
> > On Wed, Jul 17, 2024 at 09:26:22AM +0100, Steve Dower wrote:
> > > On 17/07/2024 07:33, Jeff Xu wrote:
> > > > Consider those cases: I think:
> > > > a> relying purely on userspace for enforcement does't seem to be
> > > > effective,  e.g. it is trivial  to call open(), then mmap() it into
> > > > executable memory.
> > >
> > > If there's a way to do this without running executable code that had to pass
> > > a previous execveat() check, then yeah, it's not effective (e.g. a Python
> > > interpreter that *doesn't* enforce execveat() is a trivial way to do it).
> > >
> > > Once arbitrary code is running, all bets are off. So long as all arbitrary
> > > code is being checked itself, it's allowed to do things that would bypass
> > > later checks (and it's up to whoever audited it in the first place to
> > > prevent this by not giving it the special mark that allows it to pass the
> > > check).
> >
> We will want to define what is considered as "arbitrary code is running"
> 
> Using an example of ROP, attackers change the return address in stack,
> e.g. direct the execution flow to a gauge to call "ld.so /tmp/a.out",
> do you consider "arbitrary code is running" when stack is overwritten
> ? or after execve() is called.

Yes, ROP is arbitrary code execution (which can be mitigated with CFI).
ROP could be enough to interpret custom commands and create a small
interpreter/VM.

> If it is later, this patch can prevent "ld.so /tmp/a.out".
> 
> > Exactly.  As explained in the patches, one crucial prerequisite is that
> > the executable code is trusted, and the system must provide integrity
> > guarantees.  We cannot do anything without that.  This patches series is
> > a building block to fix a blind spot on Linux systems to be able to
> > fully control executability.
> 
> Even trusted executable can have a bug.

Definitely, but this patch series is dedicated to script execution
control.

> 
> I'm thinking in the context of ChromeOS, where all its system services
> are from trusted partitions, and legit code won't load .so from a
> non-exec mount.  But we want to sandbox those services, so even under
> some kind of ROP attack, the service still won't be able to load .so
> from /tmp. Of course, if an attacker can already write arbitrary
> length of data into the stack, it is probably already a game over.
> 

OK, you want to tie executable file permission to mmap.  That makes
sense if you have a consistent execution model.  This can be enforced by
LSMs.  Contrary to script interpretation which is a full user space
implementation (and then controlled by user space), mmap restrictions
should indeed be enforced by the kernel.