Re: [PATCH v1 2/4] clone: add CLONE_PIDFD

[Christian Brauner <christian@brauner.io>]

On Wed, Apr 17, 2019 at 04:22:54PM +0200, Oleg Nesterov wrote:
> On 04/16, Christian Brauner wrote:
> >
> > +	if (clone_flags & CLONE_PIDFD) {
> > +		retval = pidfd_create(pid, &pidfdf);
> > +		if (retval < 0)
> > +			goto bad_fork_free_pid;
> > +		pidfd = retval;
> > +	}
> 
> ...
> 
> > +	if (clone_flags & CLONE_PIDFD) {
> > +		fd_install(pidfd, pidfdf);
> > +		put_user(pidfd, parent_tidptr);
> 
> put_user() can fail, I don't think this error should be silently ignored,
> this can lead to the hard-to-trigger/debug problems.
> 
> Why can't we do put_user-with-check along with pidfd_create() above?

I've moved put_user() right were pidfd_create() is called but I think
then it makes sense to change pidfd_create() to also do the fd_install()
such that the following sequence creates the pidfd, installs it, and
calls put_user() and calls ksys_close() on error. Any objections Oleg?

+       if (clone_flags & CLONE_PIDFD) {
+               retval = pidfd_create(pid);
+               if (retval < 0)
+                       goto bad_fork_free_pid;
+
+               pidfd = retval;
+               retval = put_user(pidfd, parent_tidptr);
+               if (retval)
+                       goto bad_fork_put_pidfd;
+       }

Christian