Quoting Mahesh Bandewar (mah...@bandewar.net):
> From: Mahesh Bandewar <mahe...@google.com>
>
> With this new notion of "controlled" user-namespaces, the controlled
> user-namespaces are marked at the time of their creation while the
> capabilities of processes that belong to them are controlled using the
> global mask.
>
> Init-user-ns is always uncontrolled and a process that has SYS_ADMIN
> that belongs to uncontrolled user-ns can create another (child) user-
> namespace that is uncontrolled. Any other process (that either does
> not have SYS_ADMIN or belongs to a controlled user-ns) can only
> create a user-ns that is controlled.
>
> global-capability-whitelist (controlled_userns_caps_whitelist) is used
> at the capability check-time and keeps the semantics for the processes
> that belong to uncontrolled user-ns as it is. Processes that belong to
> controlled user-ns however are subjected to different checks-
>
> (a) if the capability in question is controlled and process belongs
> to controlled user-ns, then it's always denied.
> (b) if the capability in question is NOT controlled then fall back
> to the traditional check.
>
> Signed-off-by: Mahesh Bandewar <mahe...@google.com>
Acked-by: Serge Hallyn <se...@hallyn.com>
Although a few comment addition requests below:
> ---
> v2:
> Don't recalculate user-ns flags for every setns() call.
> v1:
> Initial submission.
>
> include/linux/capability.h | 1 +
> include/linux/user_namespace.h | 20 ++++++++++++++++++++
> kernel/capability.c | 5 +++++
> kernel/user_namespace.c | 4 ++++
> security/commoncap.c | 8 ++++++++
> 5 files changed, 38 insertions(+)
>
> diff --git a/include/linux/capability.h b/include/linux/capability.h
> index 7d79a4689625..a1fd9e460379 100644
> --- a/include/linux/capability.h
> +++ b/include/linux/capability.h
> @@ -251,6 +251,7 @@ extern bool ptracer_capable(struct task_struct *tsk,
> struct user_namespace *ns);
> extern int get_vfs_caps_from_disk(const struct dentry *dentry, struct
> cpu_vfs_cap_data *cpu_caps);
> int proc_douserns_caps_whitelist(struct ctl_table *table, int write,
> void __user *buff, size_t *lenp, loff_t *ppos);
Here and at the definition below, please add a comment explaining
that a controlled cap is defined as not being in the sysctl.
> +bool is_capability_controlled(int cap);
>
> extern int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t
> size);
>
> diff --git a/include/linux/user_namespace.h b/include/linux/user_namespace.h
> index 3fe714da7f5a..647f825c7b5f 100644
> --- a/include/linux/user_namespace.h
> +++ b/include/linux/user_namespace.h
> @@ -23,6 +23,7 @@ struct uid_gid_map { /* 64 bytes -- 1 cache line */
> };
>
> #define USERNS_SETGROUPS_ALLOWED 1UL
> +#define USERNS_CONTROLLED 2UL
>
> #define USERNS_INIT_FLAGS USERNS_SETGROUPS_ALLOWED
>
> @@ -103,6 +104,16 @@ static inline void put_user_ns(struct user_namespace *ns)
> __put_user_ns(ns);
> }
>
Please add a comment explaining that a controlled ns
is one created by a user which did not have CAP_SYS_ADMIN
(or descended from such an ns).
> +static inline bool is_user_ns_controlled(const struct user_namespace *ns)
> +{
> + return ns->flags & USERNS_CONTROLLED;
> +}
> +
> +static inline void mark_user_ns_controlled(struct user_namespace *ns)
> +{
> + ns->flags |= USERNS_CONTROLLED;
> +}
> +
> struct seq_operations;
> extern const struct seq_operations proc_uid_seq_operations;
> extern const struct seq_operations proc_gid_seq_operations;
> @@ -161,6 +172,15 @@ static inline struct ns_common *ns_get_owner(struct
> ns_common *ns)
> {
> return ERR_PTR(-EPERM);
> }
> +
> +static inline bool is_user_ns_controlled(const struct user_namespace *ns)
> +{
> + return false;
> +}
> +
> +static inline void mark_user_ns_controlled(struct user_namespace *ns)
> +{
> +}
> #endif
>
> #endif /* _LINUX_USER_H */
> diff --git a/kernel/capability.c b/kernel/capability.c
> index 4a859b7d4902..bffe249922de 100644
> --- a/kernel/capability.c
> +++ b/kernel/capability.c
> @@ -511,6 +511,11 @@ bool ptracer_capable(struct task_struct *tsk, struct
> user_namespace *ns)
> }
>
> /* Controlled-userns capabilities routines */
> +bool is_capability_controlled(int cap)
> +{
> + return !cap_raised(controlled_userns_caps_whitelist, cap);
> +}
> +
> #ifdef CONFIG_SYSCTL
> int proc_douserns_caps_whitelist(struct ctl_table *table, int write,
> void __user *buff, size_t *lenp, loff_t *ppos)
> diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
> index c490f1e4313b..600c7dcb9ff7 100644
> --- a/kernel/user_namespace.c
> +++ b/kernel/user_namespace.c
> @@ -139,6 +139,10 @@ int create_user_ns(struct cred *new)
> goto fail_keyring;
>
> set_cred_user_ns(new, ns);
> + if (!ns_capable(parent_ns, CAP_SYS_ADMIN) ||
> + is_user_ns_controlled(parent_ns))
> + mark_user_ns_controlled(ns);
> +
> return 0;
> fail_keyring:
> #ifdef CONFIG_PERSISTENT_KEYRINGS
> diff --git a/security/commoncap.c b/security/commoncap.c
> index fc46f5b85251..89103f16ac37 100644
> --- a/security/commoncap.c
> +++ b/security/commoncap.c
> @@ -73,6 +73,14 @@ int cap_capable(const struct cred *cred, struct
> user_namespace *targ_ns,
> {
> struct user_namespace *ns = targ_ns;
>
> + /* If the capability is controlled and user-ns that process
> + * belongs-to is 'controlled' then return EPERM and no need
> + * to check the user-ns hierarchy.
> + */
> + if (is_user_ns_controlled(cred->user_ns) &&
> + is_capability_controlled(cap))
> + return -EPERM;
I'd be curious to see the performance impact on this on a regular
workload (kernel build?) in a controlled ns.
> +
> /* See if cred has the capability in the target user namespace
> * by examining the target user namespace and all of the target
> * user namespace's parents.
> --
> 2.15.0.448.gf294e3d99a-goog
