Re: [PATCH v2 1/4] userns: introduce capability bounding set for user namespaces

From: Hallgrimur H. Gunnarsson

Date: Thu Nov 19 2026 - 14:20:03 EST


On Thu, Nov 19, 2026 at 05:05:26PM +0200, Jarkko Sakkinen wrote:
> On Thu, Nov 19, 2026 at 10:42:05AM +0000, Hallgrimur H. Gunnarsson wrote:
> > static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
>
> I read the diff through but this led me to dead end. Cannot find this
> function from anywhere. I look the changes again once this is sorted
> out becaus this is interesting.
>
> I wanted to look at the call site to understand how cap_bset gets set,
> which is not now possible but I do get the gist and agree with it.

Thanks for taking a look. The series is based on v7.4-rc1 and
set_cred_user_ns() lives in kernel/user_namespace.c, where it is called
from create_user_ns() and userns_install(). ns->cap_bset itself is
initialized in create_user_ns() right before calling set_cred_user_ns().

> I'd figure getting this relates to my rootns patches (perhas). One key
> mechanism for me resolving the unprivileged in my use case is
> MOVE_MOUNT_T_MNTNS [1], which is orthogonal but just pointing out
> the problem space.

>From a quick look, they appear complementary. Being able to do more
setup operations from the outside generally aligns with bounding
capabilities inside the user namespace.

> I read through https://lwn.net/Articles/978846/ to understand the topic
> overall. Just mentionining as a possible addition to the cover letter
> (not a priority/necessity).

That LWN article is actually linked in the cover letter as [2].

> Hope these help and can be kept in the loop :-)

Will do, I'll keep you CC'd on future revisions.