[PATCH v3 0/3] introduce a uid/gid shifting bind mount
christian.brauner at ubuntu.com
Wed Feb 19 13:26:03 UTC 2020
On Tue, Feb 18, 2020 at 03:43:18PM -0800, James Bottomley wrote:
> On Tue, 2020-02-18 at 21:03 +0100, Christian Brauner wrote:
> > On Tue, Feb 18, 2020 at 11:05:48AM -0800, James Bottomley wrote:
> > > On Tue, 2020-02-18 at 18:26 +0100, Christian Brauner wrote:
> > > > But way more important: what Amir got right is that your approach
> > > > and fsid mappings don't stand in each others way at all. Shiftfed
> > > > bind-mounts can be implemented completely independent of fsid
> > > > mappings after the fact on top of it.
> > > >
> > > > Your example, does this:
> > > >
> > > > nsfd = open("/proc/567/ns/user", O_RDONLY); /* note: not O_PATH
> > > > */
> > > > configfd_action(fd, CONFIGFD_SET_FD, "ns", NULL, nsfd);
> > > >
> > > > as the ultimate step. Essentially marking a mountpoint as shifted
> > > > relative to that user namespace. Once fsid mappings are in all
> > > > that you need to do is replace your
> > > > make_kuid()/from_kuid()/from_kuid_munged() calls and so on in
> > > > your patchset with
> > > > make_kfsuid()/from_kfsuid()/from_kfsuid_munged() and you're done.
> > > > So I honestly don't currently see any need to block the patchsets
> > > > on each other.
> > >
> > > Can I repeat: there's no rush to get upstream on this. Let's pause
> > > to get the kernel implementation (the thing we have to maintain)
> > > right. I realise we could each work around the other and get our
> > > implementations bent around each other so they all work
> > > independently thus making our disjoint user cabals happy but I
> > > don't think that would lead to the best outcome for kernel
> > > maintainability.
> > We have had the discussion with all major stakeholders in a single
> > room on what we need at LPC 2019.
> Well, you didn't invite me, so I think "stakeholders" means people we
I'm confused as you were in the room with everyone. It's even on the
schedule under your name:
> selected because we like their use case. More importantly:
> "stakeholders" traditionally means not only people who want to consume
> the feature but also people who have to maintain it ... how many VFS
> stakeholders were present?
Again, I'm confused you were in the room with at least David and Eric.
In any case, the patches are on the relevant lists. They are actively
being discussed and visible to everyone and we'll have time for proper
review which is already happening.
More information about the Containers