[RFC next 2/5] usernamespace: make usernamespace rcu safe

[Chen Ridong]

From: Chen Ridong <chenridong@xxxxxxxxxx>

To ensure a safe top-down iteration, the user namespace should be made
RCU safe. This way, it is safe to iterate over all the child namespaces
of a root namespace while holding an RCU read lock.

Signed-off-by: Chen Ridong <chenridong@xxxxxxxxxx>
---
 include/linux/user_namespace.h |  1 +
 kernel/user_namespace.c        | 12 ++++++++++--
 2 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/include/linux/user_namespace.h b/include/linux/user_namespace.h
index 7b1e180227c8..d84b2703caab 100644
--- a/include/linux/user_namespace.h
+++ b/include/linux/user_namespace.h
@@ -80,6 +80,7 @@ struct user_namespace {
 	struct user_namespace	*parent;
 	struct list_head	ns_node;
 	struct list_head	children;
+	struct rcu_head		rcu;
 	int			level;
 	kuid_t			owner;
 	kgid_t			group;
diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
index b570536934cc..cbe8f96c3e60 100644
--- a/kernel/user_namespace.c
+++ b/kernel/user_namespace.c
@@ -196,6 +196,15 @@ int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
 	return err;
 }
 
+static void __free_user_ns(struct rcu_head *p)
+{
+	struct user_namespace *ns =
+		container_of(p, struct user_namespace, rcu);
+
+	list_del_rcu(&ns->ns_node);
+	kmem_cache_free(user_ns_cachep, ns);
+}
+
 static void free_user_ns(struct work_struct *work)
 {
 	struct user_namespace *parent, *ns =
@@ -220,10 +229,9 @@ static void free_user_ns(struct work_struct *work)
 		kfree(ns->binfmt_misc);
 #endif
 		retire_userns_sysctls(ns);
-		list_del_rcu(&ns->ns_node);
 		key_free_user_ns(ns);
 		ns_free_inum(&ns->ns);
-		kmem_cache_free(user_ns_cachep, ns);
+		call_rcu(&ns->rcu, __free_user_ns);
 		dec_user_namespaces(ucounts);
 		ns = parent;
 	} while (refcount_dec_and_test(&parent->ns.count));
-- 
2.34.1