Re: [PATCH v2 2/2] namespaces: add transparent user namespaces

[ebiederm@xmission.com (Eric W. Biederman)]

Mickaël Salaün <mic@digikod.net> writes:
>
 Hi,
>
> I have been looking for this kind of feature for StemJail [1]. One of
> the main idea is to being able to create mount points inside a jail as
> an unprivileged user but to keep as much as possible the same
> environment from outside the jail. For now, I can only create a
> mapping for the current user, so when a process list any files
> belonging to another user/group it get "nobody", which seems weird
> from a user point of view :)

Weird but I have never found it harmful.

The big thrashing about with adding s_user_ns is complete in my tree so
I can accept a reasonable patch, and transparent user namespaces is on
the edge.

If transparent user namespaces solve anything except for weird.  They
are simple enough that given a good quality patch I will merge them.

Still I want to make certain they solve something real or else I am not
certain the added complexity is worth the maintenance burden.
Especially in a part of the code where getting confused and making small
mistakes results in security issues.

On the flip side it isn't that much weirder than from_kuid_munged today
so it may not be any kind of problem at all.

>
> Regards,
> Mickaël
>
>
> 1. https://github.com/stemjail/stemjail
>
>
> On 27/06/2016 17:09, Eric W. Biederman wrote:
>> 
>> Added a few more relevant cc's.
>> 
>> Jann Horn <jannh@google.com> writes:
>> 
>>> This allows the admin of a user namespace to mark the namespace as
>>> transparent. All other namespaces, by default, are opaque.
>> 
>> 
>> I have just skimmed through this and at a high level this doesn't seem
>> too scary.  Having an identity mapped user namespace that just limits
>> you to using a subset of uids and gids while allowing displaying the
>> full range of uids and gids.
>> 
>> I don't quite get the use case and I would like to a little better
>> but in the long term this shouldn't cause any significant maintenance
>> issues, so I don't have any objects.
>> 
>> At the same time this isn't quite the time to merge this.  I am in the
>> process of slowly going through Seth's vfs changes to support things
>> such as truly unprivileged fuse support.  Those changes alter which
>> places can always be assumed to be init_user_ns (many fewer), and also
>> slightly change the set of from_kuid calls being made.
>> 
>> The changes that have made it through my review currently reside at:
>> 
>>     git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace.git for-next
>> 
>> Those vfs changes make it conceivable and simple from an infrastructure
>> standpoint to transition fileystems to unprivileged user namespace
>> mounts, with perhaps as little work as just setting FS_USER_NS.  At the
>> same time that won't be recommend because of the difficulty verifying
>> evil filesystem contents can't cause fs implementations to do bad things
>> is difficult.
>> 
>> That change means your first patch that just zaps all from_kuid_munged
>> users in init_user_ns isn't a particularly good idea.  I don't think it
>> is a good idea to have one set of rules for things that will always be
>> init_user_ns and another set of rules for code that will change.
>> 
>> The long and short of this is I am asking you to wait a week or so and
>> rebase this on my for-next branch so that we can confirm this change
>> interacts nicely will all of the other on-going work.
>> 
>> Thank you,
>> Eric Biederman

Eric