Re: [PATCH 09/10] userns: check user namespace for task->file uid equivalence checks

["Serge E. Hallyn" <serge.hallyn@canonical.com>]

Quoting Nathan Lynch (ntl@pobox.com):
> On Thu, 2011-02-24 at 15:03 +0000, Serge E. Hallyn wrote:
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -1446,8 +1446,13 @@ enum {
> >  #define put_fs_excl() atomic_dec(&current->fs_excl)
> >  #define has_fs_excl() atomic_read(&current->fs_excl)
> >  
> > -#define is_owner_or_cap(inode)	\
> > -	((current_fsuid() == (inode)->i_uid) || capable(CAP_FOWNER))
> > +/*
> > + * until VFS tracks user namespaces for inodes, just make all files
> > + * belong to init_user_ns
> > + */
> > +extern struct user_namespace init_user_ns;
> 
> init_user_ns gets declared in fs.h in this patch, utsname.h in patch #1,
> capability.h in #2, ipc_namespace.h in #7.  Could this declaration be
> kept to a single header?
> 

ipc/msgutil.c includes security.h which includes fs.h, so we should be
able to drop the one in ipc_namespace.h.  The one in utsname.h is
there for init/version.c and needed AFAICS.

The one in capability.h should be able to go when has_capability* are
turned into functions.  They couldn't be turned into static functions
in capability.h (left as exercise for reader), but they can be made
full-fledged functions in kernel/capability.c.  I will do that in a
follow-on patch and try to remove the extra init_user_ns defines as
well.

thanks,
-serge