lkml.org 
[lkml]   [2020]   [Feb]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v2 00/28] user_namespace: introduce fsid mappings
And re-sending, this time hopefully actually in plain text mode.
Sorry about that, my e-mail client isn't behaving today...

Stéphane

On Mon, Feb 17, 2020 at 4:57 PM Stéphane Graber <stgraber@ubuntu.com> wrote:
>
> On Mon, Feb 17, 2020 at 4:12 PM James Bottomley <James.Bottomley@hansenpartnership.com> wrote:
>>
>> On Fri, 2020-02-14 at 19:35 +0100, Christian Brauner wrote:
>> [...]
>> > With this patch series we simply introduce the ability to create fsid
>> > mappings that are different from the id mappings of a user namespace.
>> > The whole feature set is placed under a config option that defaults
>> > to false.
>> >
>> > In the usual case of running an unprivileged container we will have
>> > setup an id mapping, e.g. 0 100000 100000. The on-disk mapping will
>> > correspond to this id mapping, i.e. all files which we want to appear
>> > as 0:0 inside the user namespace will be chowned to 100000:100000 on
>> > the host. This works, because whenever the kernel needs to do a
>> > filesystem access it will lookup the corresponding uid and gid in the
>> > idmapping tables of the container.
>> > Now think about the case where we want to have an id mapping of 0
>> > 100000 100000 but an on-disk mapping of 0 300000 100000 which is
>> > needed to e.g. share a single on-disk mapping with multiple
>> > containers that all have different id mappings.
>> > This will be problematic. Whenever a filesystem access is requested,
>> > the kernel will now try to lookup a mapping for 300000 in the id
>> > mapping tables of the user namespace but since there is none the
>> > files will appear to be owned by the overflow id, i.e. usually
>> > 65534:65534 or nobody:nogroup.
>> >
>> > With fsid mappings we can solve this by writing an id mapping of 0
>> > 100000 100000 and an fsid mapping of 0 300000 100000. On filesystem
>> > access the kernel will now lookup the mapping for 300000 in the fsid
>> > mapping tables of the user namespace. And since such a mapping
>> > exists, the corresponding files will have correct ownership.
>>
>> How do we parametrise this new fsid shift for the unprivileged use
>> case? For newuidmap/newgidmap, it's easy because each user gets a
>> dedicated range and everything "just works (tm)". However, for the
>> fsid mapping, assuming some newfsuid/newfsgid tool to help, that tool
>> has to know not only your allocated uid/gid chunk, but also the offset
>> map of the image. The former is easy, but the latter is going to vary
>> by the actual image ... well unless we standardise some accepted shift
>> for images and it simply becomes a known static offset.
>
>
> For unprivileged runtimes, I would expect images to be unshifted and be
> unpacked from within a userns. So your unprivileged user would be allowed
> a uid/gid range through /etc/subuid and /etc/subgid and allowed to use
> them through newuidmap/newgidmap.In that namespace, you can then pull
> and unpack any images/layers you may want and the resulting fs tree will
> look correct from within that namespace.
>
> All that is possible today and is how for example unprivileged LXC works
> right now.
>
> What this patchset then allows is for containers to have differing
> uid/gid maps while still being based off the same image or layers.
> In this scenario, you would carve a subset of your main uid/gid map for
> each container you run and run them in a child user namespace while
> setting up a fsuid/fsgid map such that their filesystem access do not
> follow their uid/gid map. This then results in proper isolation for
> processes, networks, ... as everything runs as different kuid/kgid but
> the VFS view will be the same in all containers.
>
> Shared storage between those otherwise isolated containers would also
> work just fine by simply bind-mounting the same path into two or more
> containers.
>
>
> Now one additional thing that would be safe for a setuid wrapper to
> allow would be for arbitrary mapping of any of the uid/gid that the user
> owns to be used within the fsuid/fsgid map. One potential use for this
> would be to create any number of user namespaces, each with their own
> mapping for uid 0 while still having all VFS access be mapped to the
> user that spawned them (say uid=1000, gid=1000).
>
>
> Note that in our case, the intended use for this is from a privileged runtime
> where our images would be unshifted as would be the container storage
> and any shared storage for containers. The security model effectively relying
> on properly configured filesystem permissions and mount namespaces such
> that the content of those paths can never be seen by anyone but root outside
> of those containers (and therefore avoids all the issues around setuid/setgid/fscaps).
>
> We will then be able to allocate distinct, random, ranges of 65536 uids/gids (or more)
> for each container without ever having to do any uid/gid shifting at the filesystem layer
> or run into issues when having to setup shared storage between containers or attaching
> external storage volumes to those containers.
>
>> James
>
>
> Stéphane

\
 
 \ /
  Last update: 2020-02-17 23:03    [W:0.188 / U:3.008 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site