[PATCH 4/6] user namespaces: add user_ns to super block

Serge E. Hallyn serue at us.ibm.com
Fri Aug 1 17:06:09 PDT 2008

Quoting Eric W. Biederman (ebiederm at xmission.com):
> "Serge E. Hallyn" <serue at us.ibm.com> writes:
> >
> > The filesystem can figure that out based on current's context, no?
> >
> > With the per-sb user_ns, the default behavior is indeed very limited,
> > but since you want to move all the user_ns functionality into the
> > filesystem, the fs can tag vfsmounts based on the "new remount" you
> > had talked about before.
> I guess I want the filesystem to coordinate.
> >> 	Would this require passing the vfsmount to the filesystems themselves,
> >> or would they be within the VFS code only? If not wholly within the VFS
> >> I wonder if Al Viro would object to this. He's resisted past attempts to
> >> pass the vfsmount structs into more filesystem code paths and I'm
> >> guessing that could affect whether or not this approach can be
> >> implemented.
> >
> > Right, that's the main reason we might want to pursue the per-sb
> > approach.  Otherwise I would prefer the per-vfsmount approach.
> >
> > Eric, if you think the per-vfsmount fight is worth fighting, then by all
> > means let's do it and see what happens.  So in that case ignore patches
> > 3-5 from this set :)
> My intuitive sense is that the treating the handling of different
> user namespaces in the same filesystem is a trivial case of the
> superblock merging that nfs performs.  And that we will preserve
> existing semantics much better if the user namespace is stored
> in the vfsmount.  This allows mount propagation and friends to work
> without surprises.
> The practical limitation I see of storing things outside of the
> vfsmount is when do you setup the mapping to handle a new user
> namespace?
> So yes.  I think it is worth the discussion.  Let's not
> move the vfsmount down, and just move the user namespace pointer
> down as that is fundamentally what we care about.
> Eric

Ok I wasn't thinking right.  We still can't get to a user_ns from
an inode *.

So playing with this a bit tonight, it seems like the best way
to pass the user_namespace up to the fs is just to define new
super_operations which handle the conversions.  Something like
the following.

(This sits on top of the two patches I'm about to send out as
replacements for patches 1 and 2 from my previous posting.)


More information about the Containers mailing list