[RFC PATCH] vfs: Delay root FS switch after UMH completion

[Michal Koutný <mkoutny@xxxxxxxx>]

We want to make sure no UMHs started with an old root survive into the
world with the new root (they may fail when it is not expected).
Therefore, insert a wait for existing UMHs termination (this assumes UMH
runtime is finite).

A motivation are asynchronous module loads that start in initrd and they
may be (un)intentionally terminated by a userspace cleanup during rootfs
transition.

This is also inspired by an ancient problem [1].

This is just a rough idea, only superficially tested, no broader context
(VFS locking et al) is considered.

[1] https://kernelnewbies.org/KernelProjects/usermode-helper-enhancements#Filesystem_suspend

Signed-off-by: Michal Koutný <mkoutny@xxxxxxxx>
---

I'm not amused by this patch. I'm sending it to get some NAck reasons
(besides indefinite UMH lifetime) and get it off my head. Thanks.

 fs/namespace.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/fs/namespace.c b/fs/namespace.c
index ab467ee58341..48cb658ae10c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2931,6 +2931,7 @@ static int do_move_mount(struct path *old_path, struct path *new_path)
 	struct mount *old;
 	struct mount *parent;
 	struct mountpoint *mp, *old_mp;
+	struct path ns_root;
 	int err;
 	bool attached;
 
@@ -2985,6 +2986,14 @@ static int do_move_mount(struct path *old_path, struct path *new_path)
 		if (p == old)
 			goto out;
 
+	ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
+	ns_root.dentry = ns_root.mnt->mnt_root;
+	path_get(&ns_root);
+	/* See argument in pivot_root() */
+	if (path_equal(new_path, &ns_root))
+		usermodehelper_disable();
+
+
 	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
 				   attached);
 	if (err)
@@ -2996,6 +3005,9 @@ static int do_move_mount(struct path *old_path, struct path *new_path)
 	if (attached)
 		put_mountpoint(old_mp);
 out:
+	if (path_equal(new_path, &ns_root))
+		usermodehelper_enable();
+
 	unlock_mount(mp);
 	if (!err) {
 		if (attached)
@@ -3997,6 +4009,8 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out2;
 
 	get_fs_root(current->fs, &root);
+	/* UMHs started from old root should finish before we switch root under */
+	usermodehelper_disable(); // XXX error handling
 	old_mp = lock_mount(&old);
 	error = PTR_ERR(old_mp);
 	if (IS_ERR(old_mp))
@@ -4058,6 +4072,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	error = 0;
 out4:
 	unlock_mount(old_mp);
+	usermodehelper_enable();
 	if (!error)
 		mntput_no_expire(ex_parent);
 out3:
-- 
2.39.1