Re: [PATCH] mm/secretmem: use refcount_t instead of atomic_t

[Jordy Zomer]

Hi There!

Because this is a global variable, it appears to be exploitable. Either we generate a sufficient number of processes to achieve this counter, or you increase the open file limit with ulimit or sysctl. Unless the kernel has a hard restriction on the number of potential file descriptors that I'm not aware of.

In any case, it's probably a good idea to patch this to make it explicitly secure. If you discover a hard-limit in the kernel for open file descriptors, please let me know. I'm genuinely ​interested :D!

Best Regards,

Jordy

> On 08/20/2021 12:05 PM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> 
>  
> On Fri, Aug 20, 2021 at 07:57:25AM -0700, James Bottomley wrote:
> > On Fri, 2021-08-20 at 06:33 +0200, Jordy Zomer wrote:
> > > As you can see there's an `atomic_inc` for each `memfd` that is
> > > opened in the `memfd_secret` syscall. If a local attacker succeeds to
> > > open 2^32 memfd's, the counter will wrap around to 0. This implies
> > > that you may hibernate again, even though there are still regions of
> > > this secret memory, thereby bypassing the security check.
> > 
> > This isn't a possible attack, is it?  secret memory is per process and
> > each process usually has an open fd limit of 1024.  That's not to say
> > we shouldn't have overflow protection just in case, but I think today
> > we don't have a problem.
> 
> But it's a _global_ setting, so it's still possible, though likely
> impractical today. But refcount_t mitigates it and is a trivial change.
> :)
> 
> -- 
> Kees Cook