Re: procfs: open("/proc/self/fd/...") allows bypassing O_RDONLY

[Christian Brauner <brauner@kernel.org>]

On Thu, May 12, 2022 at 02:56:22PM +0200, Miklos Szeredi wrote:
> On Thu, 12 May 2022 at 14:41, Simon Ser <contact@emersion.fr> wrote:
> >
> > On Thursday, May 12th, 2022 at 12:37, Simon Ser <contact@emersion.fr> wrote:
> >
> > > what would be a good way to share a FD to another
> > > process without allowing it to write to the underlying file?
> >
> > (I'm reminded that memfd + seals exist for this purpose. Still, I'd be
> > interested to know whether that O_RDONLY/O_RDWR behavior is intended,
> > because it's pretty surprising. The motivation for using O_RDONLY over
> > memfd seals is that it isn't Linux-specific.)
> 
> Yes, this is intended.   The /proc/$PID/fd/$FD file represents the
> inode pointed to by $FD.   So the open flags for $FD are irrelevant
> when operating on the proc fd file.

Fwiw, the original openat2() patchset contained upgrade masks which we
decided to split it out into a separate patchset.

The idea is that struct open_how would be extended with an upgrade mask
field which allows the opener to specify with which permissions a file
descriptor is allowed to be re-opened. This has quite a lot of
use-cases, especially in container runtimes. So one could open an fd and
restrict it from being re-opened with O_WRONLY. For container runtimes
this is a huge security win and for userspace in general it would
provide a backwards compatible way of e.g., making O_PATH fds
non-upgradable. The plan is to resend the extension at some point in the
not too distant future.

Christian