Re: [PATCH] vfs: move_mount: reject moving kernel internal mounts

[Eric Biggers]

On Mon, Jul 01, 2019 at 07:22:39PM +0100, Al Viro wrote:
> On Mon, Jul 01, 2019 at 09:45:37AM -0700, Eric Biggers wrote:
> > On Sat, Jun 29, 2019 at 01:27:44PM -0700, Eric Biggers wrote:
> > > 
> > > Reproducer:
> > > 
> > >     #include <unistd.h>
> > > 
> > >     #define __NR_move_mount         429
> > >     #define MOVE_MOUNT_F_EMPTY_PATH 0x00000004
> > > 
> > >     int main()
> > >     {
> > >     	  int fds[2];
> > > 
> > >     	  pipe(fds);
> > >         syscall(__NR_move_mount, fds[0], "", -1, "/", MOVE_MOUNT_F_EMPTY_PATH);
> > >     }
> > 
> > David, I'd like to add this as a regression test somewhere.
> > 
> > Can you point me to the tests for the new mount syscalls?
> > 
> > I checked LTP, kselftests, and xfstests, but nothing to be found.
> 
> FWIW, it's not just move_mount(2) - I'd expect
> 
> 	int fds[2];
> 	char s[80];
> 
> 	pipe(fds);
> 	sprintf(s, "/dev/fd/%d", fds[0]);
> 	mount(s, "/dev/null", NULL, MS_MOVE, 0);
> 
> to step into exactly the same thing.  mount(2) does follow symlinks -
> always had...

Sure, but the new mount syscalls still need tests.  Where are the tests?

Also, since the case of a fd with an internal mount was overlooked, probably the
man page needs to be updated clarify that move_mount(2) fails with EINVAL in
this case.  Where is the man page?

- Eric