Re: [PATCH] net: make net namespace sysctls belong to container's owner

Dmitry Torokhov [off-list ref] writes:

If net namespace is attached to a user namespace let's make container's
root owner of sysctls affecting said network namespace instead of global
root.

This also allows us to clean up net_ctl_permissions() because we do not
need to fudge permissions anymore for the container's owner since it now
owns the objects in question.

Acked-by: "Eric W. Biederman" <redacted>

Overall this seems reasonable.  However I am not a fan of your error
handling.

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
---

This helps when running Android CTS in a container, but I think it makes
sense regardless.

+static void net_ctl_set_ownership(struct ctl_table_header *head,
+				  struct ctl_table *table,
+				  kuid_t *uid, kgid_t *gid)
+{
+	struct net *net = container_of(head->set, struct net, sysctls);
+
+	*uid = make_kuid(net->user_ns, 0);
+	if (!uid_valid(*uid))
+		*uid = GLOBAL_ROOT_UID;
+
+	*gid = make_kgid(net->user_ns, 0);
+	if (!gid_valid(*gid))
+		*gid = GLOBAL_ROOT_GID;

This code should eiter be:
	*uid = make_kuid(net->user_ns, 0);
        *gid = make_kgid(net->user_ns, 0);

Or it should be:
	tmp_uid = make_kuid(net->user_ns, 0);
        if (uid_valid(tmp_uid))
        	*uid = tmp_uid;

	tmp_gid = make_kgid(net->user_ns, 0);
        if (gid_valid(tmp_gid))
        	*gid = tmp_gid;

It is just very fragile to assume to know what uid and gid
would be if this code fails.

As of v4.8-rc1 INVALID_UID and INVALID_GID can be set in inode->i_uid
and inode->i_gid without causing horrible vfs confusion (making the
first option viable), but I expect with the mention of Android you want
to backport this so I will ask that you ask to implement the error
handling that doesn't assume you know better than the generic code.

If you don't have a better value to set something to it really should be
left alone.

Eric

On Mon, Aug 8, 2016 at 2:08 PM, Eric W. Biederman [off-list ref] wrote:

Dmitry Torokhov [off-list ref] writes:

quoted

If net namespace is attached to a user namespace let's make container's
root owner of sysctls affecting said network namespace instead of global
root.

This also allows us to clean up net_ctl_permissions() because we do not
need to fudge permissions anymore for the container's owner since it now
owns the objects in question.

Acked-by: "Eric W. Biederman" <redacted>

Overall this seems reasonable.  However I am not a fan of your error
handling.

quoted

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
---

This helps when running Android CTS in a container, but I think it makes
sense regardless.

quoted

+static void net_ctl_set_ownership(struct ctl_table_header *head,
+                               struct ctl_table *table,
+                               kuid_t *uid, kgid_t *gid)
+{
+     struct net *net = container_of(head->set, struct net, sysctls);
+
+     *uid = make_kuid(net->user_ns, 0);
+     if (!uid_valid(*uid))
+             *uid = GLOBAL_ROOT_UID;
+
+     *gid = make_kgid(net->user_ns, 0);
+     if (!gid_valid(*gid))
+             *gid = GLOBAL_ROOT_GID;

This code should eiter be:
        *uid = make_kuid(net->user_ns, 0);
        *gid = make_kgid(net->user_ns, 0);

Or it should be:
        tmp_uid = make_kuid(net->user_ns, 0);
        if (uid_valid(tmp_uid))
                *uid = tmp_uid;

        tmp_gid = make_kgid(net->user_ns, 0);
        if (gid_valid(tmp_gid))
                *gid = tmp_gid;

It is just very fragile to assume to know what uid and gid
would be if this code fails.

As of v4.8-rc1 INVALID_UID and INVALID_GID can be set in inode->i_uid
and inode->i_gid without causing horrible vfs confusion (making the
first option viable), but I expect with the mention of Android you want
to backport this so I will ask that you ask to implement the error
handling that doesn't assume you know better than the generic code.

If you don't have a better value to set something to it really should be
left alone.

OK, fair enough. I will adopt the 2nd option and will resubmit. I need
to also test without net namespaces support (my other change blows up
because we are getting half-initialized init_net structure when
namespaces are disabled).

Thanks.

-- 
Dmitry