[RFC PATCH v6 14/26] userland: Add USERLAND_SHARE_NSFD ioctl
From: Jarkko Sakkinen
Date: Sat Nov 21 2026 - 20:20:27 EST
userland_open() isolates every namespace by default rather than relying
on unshare() deny-listing, which prevents workloads such as
host-networked sandboxes or sidecars sharing a network or IPC namespace
from opting into specific existing namespaces.
Add the build-phase USERLAND_SHARE_NSFD ioctl to replace the userland's
cgroup, IPC, UTS, network, or time namespace with a supplied nsfs file
descriptor in-place under userland->lock while the userland is still in
USERLAND_INIT, rejecting calls with -EBUSY once CLONE_INTO_USERLAND
begins. Require CAP_SYS_ADMIN in the userland's user namespace, and
require CAP_SYS_ADMIN in the target namespace's owning user namespace
when the caller is not already a member of that namespace.
Signed-off-by: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
---
include/uapi/linux/userland.h | 14 ++++++
kernel/userland.c | 92 +++++++++++++++++++++++++++++++++++
2 files changed, 106 insertions(+)
diff --git a/include/uapi/linux/userland.h b/include/uapi/linux/userland.h
index e81a06d6e9eb..addb84e996f5 100644
--- a/include/uapi/linux/userland.h
+++ b/include/uapi/linux/userland.h
@@ -41,4 +41,18 @@ enum userland_flags {
*/
#define USERLAND_GET_NSFD _IO(USERLAND_IOCTL_MAGIC, 0x01)
+/**
+ * DOC: USERLAND_SHARE_NSFD
+ *
+ * ``ioctl(userland_fd, USERLAND_SHARE_NSFD, ns_fd)`` replaces the corresponding
+ * member namespace of ``userland_fd`` with ``ns_fd`` (``CLONE_NEWCGROUP``,
+ * ``CLONE_NEWIPC``, ``CLONE_NEWUTS``, ``CLONE_NEWNET``, or ``CLONE_NEWTIME``)
+ * while ``userland_fd`` is in ``USERLAND_INIT``.
+ *
+ * Requires ``CAP_SYS_ADMIN`` in the userland's user namespace, plus
+ * ``CAP_SYS_ADMIN`` in ``ns_fd``'s owning user namespace when the caller is
+ * not already a member of ``ns_fd``.
+ */
+#define USERLAND_SHARE_NSFD _IO(USERLAND_IOCTL_MAGIC, 0x02)
+
#endif /* _UAPI_LINUX_USERLAND_H */
diff --git a/kernel/userland.c b/kernel/userland.c
index 7b44f400dc23..efac9e4f2fa5 100644
--- a/kernel/userland.c
+++ b/kernel/userland.c
@@ -24,6 +24,8 @@
#include <linux/nsfs.h>
#include <linux/pid.h>
#include <linux/pid_namespace.h>
+#include <linux/proc_fs.h>
+#include <linux/proc_ns.h>
#include <linux/time_namespace.h>
#include <linux/user_namespace.h>
#include <linux/uts_namespace.h>
@@ -118,6 +120,94 @@ static long userland_get_nsfd(struct userland *userland, unsigned long nstype)
return open_namespace(ns);
}
+static long userland_share_nsfd(struct userland *c, unsigned long arg)
+{
+ struct ns_common *old_time_children = NULL;
+ struct ns_common *old_ns = NULL;
+ struct ns_common *ns;
+
+ if (arg > INT_MAX)
+ return -EBADF;
+
+ CLASS(fd, f)((int)arg);
+ if (fd_empty(f))
+ return -EBADF;
+ if (!proc_ns_file(fd_file(f)))
+ return -EINVAL;
+ if (unlikely(!current->nsproxy))
+ return -EOPNOTSUPP;
+
+ ns = get_proc_ns(file_inode(fd_file(f)));
+ switch (ns->ns_type) {
+ case CLONE_NEWCGROUP:
+ case CLONE_NEWIPC:
+ case CLONE_NEWUTS:
+ case CLONE_NEWNET:
+ break;
+ case CLONE_NEWTIME:
+ if (!IS_ENABLED(CONFIG_TIME_NS))
+ return -EOPNOTSUPP;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ if (!ns_capable(c->user_ns, CAP_SYS_ADMIN))
+ return -EPERM;
+ if (!is_current_namespace(ns) &&
+ !ns_capable(ns->ops->owner(ns), CAP_SYS_ADMIN))
+ return -EPERM;
+
+ scoped_guard(spinlock, &c->lock) {
+ if (c->state != USERLAND_INIT)
+ return -EBUSY;
+
+ __ns_ref_inc(ns);
+ __ns_ref_active_get(ns);
+
+ switch (ns->ns_type) {
+ case CLONE_NEWCGROUP:
+ old_ns = to_ns_common(c->ns->cgroup_ns);
+ c->ns->cgroup_ns = to_cg_ns(ns);
+ break;
+ case CLONE_NEWIPC:
+ old_ns = to_ns_common(c->ns->ipc_ns);
+ c->ns->ipc_ns = to_ipc_ns(ns);
+ break;
+ case CLONE_NEWUTS:
+ old_ns = to_ns_common(c->ns->uts_ns);
+ c->ns->uts_ns = to_uts_ns(ns);
+ break;
+ case CLONE_NEWNET:
+ old_ns = to_ns_common(c->ns->net_ns);
+ c->ns->net_ns = to_net_ns(ns);
+ break;
+#ifdef CONFIG_TIME_NS
+ case CLONE_NEWTIME:
+ __ns_ref_inc(ns);
+ __ns_ref_active_get(ns);
+ old_ns = to_ns_common(c->ns->time_ns);
+ old_time_children =
+ to_ns_common(c->ns->time_ns_for_children);
+ c->ns->time_ns = to_time_ns(ns);
+ c->ns->time_ns_for_children = to_time_ns(ns);
+ break;
+#endif
+ default:
+ WARN_ON_ONCE(1);
+ break;
+ }
+ }
+
+ __ns_ref_active_put(old_ns);
+ old_ns->ops->put(old_ns);
+ if (old_time_children) {
+ __ns_ref_active_put(old_time_children);
+ old_time_children->ops->put(old_time_children);
+ }
+ return 0;
+}
+
static long userland_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
struct userland *userland = file->private_data;
@@ -125,6 +215,8 @@ static long userland_ioctl(struct file *file, unsigned int cmd, unsigned long ar
switch (cmd) {
case USERLAND_GET_NSFD:
return userland_get_nsfd(userland, arg);
+ case USERLAND_SHARE_NSFD:
+ return userland_share_nsfd(userland, arg);
default:
return -ENOIOCTLCMD;
}
--
2.47.3