[CRIU] Introspecting userns relationships to other namespaces?

[Andrew Vagin]

On Thu, Jul 07, 2016 at 08:20:05PM -0700, James Bottomley wrote:
> On Thu, 2016-07-07 at 19:16 -0700, Andrew Vagin wrote:
> > On Thu, Jul 07, 2016 at 12:17:35PM -0700, James Bottomley wrote:
> > > On Thu, 2016-07-07 at 20:21 +0200, Michael Kerrisk (man-pages)
> > > wrote:
> > > > On 7 July 2016 at 17:01, James Bottomley
> > > > <James.Bottomley at hansenpartnership.com> wrote:
> > > [Serge already answered the parenting issue]
> > > > > On Thu, 2016-07-07 at 08:36 -0500, Serge E. Hallyn wrote:
> > > > > > Hm.  Probably best-effort based on the process hierarchy.  So
> > > > > > yeah you could probably get a tree into a state that would be
> > > > > > wrongly recreated. Create a new netns, bind mount it, exit; 
> > > > > >  Have 
> > > > > > another task create a new user_ns, bind mount it, exit; 
> > > > > >  Third 
> > > > > > task setns()s first to the new netns then to the new user_ns.
> > > > > >   I 
> > > > > > suspect criu will recreate that wrongly.
> > > > > 
> > > > > This is a bit pathological, and you have to be root to do it:
> > > > > so 
> > > > > root can set up a nesting hierarchy, bind it and destroy the
> > > > > pids 
> > > > > but I know of no current orchestration system which does this.
> > > > > 
> > > > > Actually, I have to back pedal a bit: the way I currently set
> > > > > up
> > > > > architecture emulation containers does precisely this: I set up
> > > > > the
> > > > > namespaces unprivileged with child mount namespaces, but then I
> > > > > ask
> > > > > root to bind the userns and kill the process that created it so
> > > > > I 
> > > > > have a permanent handle to enter the namespace by, so I suspect
> > > > > that when our current orchestration systems get more
> > > > > sophisticated, 
> > > > > they might eventually want to do something like this as well.
> > > > > 
> > > > > In theory, we could get nsfs to show this information as an
> > > > > option
> > > > > (just add a show_options entry to the superblock ops), but the 
> > > > > problem is that although each namespace has a parent user_ns, 
> > > > > there's no way to get it without digging in the namespace
> > > > > specific 
> > > > > structure.  Probably we should restructure to move it into 
> > > > > ns_common, then we could display it (and enforce all namespaces
> > > > > having owning user_ns) but it would be a
> > > > 
> > > > I'm missing something here. Is it not already the case that all
> > > > namespaces have an owning user_ns?
> > > 
> > > Um, yes, I don't believe I said they don't.  The problem I thought
> > > you
> > > were having is that there's no way of seeing what it is.
> > > 
> > > nsfs is the Namespace fileystem where bound namespaces appear to a
> > > cat
> > > of /proc/self/mounts.  It can display any information that's in
> > > ns_common (the common core of namespaces) but the owning user_ns
> > > pointer currently isn't in this structure.  Every user namespace
> > > has a
> > > pointer to it, but they're all privately embedded in the individual
> > > namespace specific structures.  What I was proposing was that since
> > > every current namespace has a pointer somewhere to the owning user
> > > namespace, we could abstract this out into ns_common so it's now
> > > accessible to be displayed by nsfs, probably as a mount option.
> > 
> > James, I am not sure that I understood you correctly. We have one
> > file system for all namespace files, how we can show per-file 
> > properties in mount options.
> 
> We have two ways of getting information.  For a namespace that only
> exists as a bind mount we only have what the mount/mountinfo shows, so
> you see something like this:
> 
> jejb at jarvis:~> mount|grep nsfs
> nsfs on /run/build-container/userns type nsfs (rw)
> nsfs on /run/build-container/ppc64 type nsfs (rw)
> 
> the (rw) are the mount options.  We could add the ability to add other
> mount options to this via the superblock .show_options callback.  We
> could make it show the type and parent user namespace.

Yes, we could. But this way works only for bind-mounted ns files, fdinfo
works for any ns files (e.g: /proc/PID/ns/X).
fdinfo show information about one namespace, when /proc/pid/mountinfo
shows infromation about all mounts, so we can parse fdinfo faster and
easier.

> 
> >  I think we can show all required information in fdinfo. We open a
> > namespaces file (/proc/pid/ns/N) and then read /proc/pid/fdinfo/X for
> > it.
> 
> Not if we don't have an extant process in the namespace, we can't use
> these files because they don't exist, plus fdinfo on the
> /proc/<pid>/ns/X doesn't tell you what the parent user_ns of X is
> (again, we could add this information somewhere ... not sure where
> yet).

we can read fdinfo for any ns file.

For example,

fd = open("/run/build-container/userns", O_PATH);

then read fdinfo for this "fd" (/proc/self/fdinfo/[fd])

Thanks,
Andrew

> 
> James
>