Re: [PATCH 16/43] userns: Simplify the user_namespace by making userns->creator a kuid.

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

Quoting Eric W. Biederman (ebiederm@xmission.com):
> "Serge E. Hallyn" <serge@hallyn.com> writes:
> 
> > Quoting Eric W. Beiderman (ebiederm@xmission.com):
> >> From: Eric W. Biederman <ebiederm@xmission.com>
> >> 
> >> - Transform userns->creator from a user_struct reference to a simple
> >>   kuid_t, kgid_t pair.
> >> 
> >>   In cap_capable this allows the check to see if we are the creator of
> >>   a namespace to become the classic suser style euid permission check.
> >> 
> >>   This allows us to remove the need for a struct cred in the mapping
> >>   functions and still be able to dispaly the user namespace creators
> >>   uid and gid as 0.
> >> 
> >> - Remove the now unnecessary delayed_work in free_user_ns.
> >> 
> >>   All that is left for free_user_ns to do is to call kmem_cache_free
> >>   and put_user_ns.  Those functions can be called in any context
> >>   so call them directly from free_user_ns removing the need for delayed work.
> >> 
> >> Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
> >> ---
> >>  include/linux/user_namespace.h |    4 ++--
> >>  kernel/user.c                  |    7 ++++---
> >>  kernel/user_namespace.c        |   39 ++++++++++++++++++---------------------
> >>  security/commoncap.c           |    5 +++--
> >>  4 files changed, 27 insertions(+), 28 deletions(-)
> >> 
> >> diff --git a/include/linux/user_namespace.h b/include/linux/user_namespace.h
> >> index d767508..8a391bd 100644
> >> --- a/include/linux/user_namespace.h
> >> +++ b/include/linux/user_namespace.h
> >> @@ -9,8 +9,8 @@
> >>  struct user_namespace {
> >>  	struct kref		kref;
> >>  	struct user_namespace	*parent;
> >> -	struct user_struct	*creator;
> >> -	struct work_struct	destroyer;
> >> +	kuid_t			owner;
> >> +	kgid_t			group;
> >>  };
> >>  
> >>  extern struct user_namespace init_user_ns;
> >> diff --git a/kernel/user.c b/kernel/user.c
> >> index 025077e..cff3856 100644
> >> --- a/kernel/user.c
> >> +++ b/kernel/user.c
> >> @@ -25,7 +25,8 @@ struct user_namespace init_user_ns = {
> >>  	.kref = {
> >>  		.refcount	= ATOMIC_INIT(3),
> >>  	},
> >> -	.creator = &root_user,
> >> +	.owner = GLOBAL_ROOT_UID,
> >> +	.group = GLOBAL_ROOT_GID,
> >>  };
> >>  EXPORT_SYMBOL_GPL(init_user_ns);
> >>  
> >> @@ -54,9 +55,9 @@ struct hlist_head uidhash_table[UIDHASH_SZ];
> >>   */
> >>  static DEFINE_SPINLOCK(uidhash_lock);
> >>  
> >> -/* root_user.__count is 2, 1 for init task cred, 1 for init_user_ns->user_ns */
> >> +/* root_user.__count is 1, for init task cred */
> >>  struct user_struct root_user = {
> >> -	.__count	= ATOMIC_INIT(2),
> >> +	.__count	= ATOMIC_INIT(1),
> >>  	.processes	= ATOMIC_INIT(1),
> >>  	.files		= ATOMIC_INIT(0),
> >>  	.sigpending	= ATOMIC_INIT(0),
> >> diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
> >> index 898e973..f69741a 100644
> >> --- a/kernel/user_namespace.c
> >> +++ b/kernel/user_namespace.c
> >> @@ -27,6 +27,16 @@ int create_user_ns(struct cred *new)
> >>  {
> >>  	struct user_namespace *ns, *parent_ns = new->user_ns;
> >>  	struct user_struct *root_user;
> >> +	kuid_t owner = make_kuid(new->user_ns, new->euid);
> >> +	kgid_t group = make_kgid(new->user_ns, new->egid);
> >> +
> >> +	/* The creator needs a mapping in the parent user namespace
> >> +	 * or else we won't be able to reasonably tell userspace who
> >> +	 * created a user_namespace.
> >> +	 */
> >> +	if (!kuid_has_mapping(parent_ns, owner) ||
> >> +	    !kgid_has_mapping(parent_ns, group))
> >> +		return -EPERM;
> >>  
> >>  	ns = kmem_cache_alloc(user_ns_cachep, GFP_KERNEL);
> >>  	if (!ns)
> >> @@ -43,7 +53,9 @@ int create_user_ns(struct cred *new)
> >>  
> >>  	/* set the new root user in the credentials under preparation */
> >>  	ns->parent = parent_ns;
> >
> > I think in the past the creator cred pinned the ns->parent.  Do you now
> > need to explicitly pin ns->parent (and release it in free_user_ns())?
> 
> Yes we do have to explicitly reference count the parent namespace.
> But that happened in the patch 7:
> "userns: Add an explicit reference to the parent user namespace"

Perhaps that suffices, but I'm not convinced.  The struct cred is
pinning it's own ns, but if t1 does clone(CLONE_NEWUSER) to produce
t2, which does the same to procduce t3, and then t2 exits, I'm not
seeing what will pin t2's userns.

-serge