On 12/3/21 13:50, James Bottomley wrote:

On Fri, 2021-12-03 at 13:06 -0500, Stefan Berger wrote:

quoted

On 12/3/21 12:03, James Bottomley wrote:

quoted

On Thu, 2021-12-02 at 21:31 -0500, Stefan Berger wrote:
[...]

quoted

   static int securityfs_init_fs_context(struct fs_context *fc)
   {
+	int rc;
+
+	if (fc->user_ns->ima_ns->late_fs_init) {
+		rc = fc->user_ns->ima_ns->late_fs_init(fc->user_ns);
+		if (rc)
+			return rc;
+	}
   	fc->ops = &securityfs_context_ops;
   	return 0;
   }

I know I suggested this, but to get this to work in general, it's
going to have to not be specific to IMA, so it's going to have to
become something generic like a notifier chain.  The other problem
is it's only working still by accident:

I had thought about this also but the rationale was:

securityfs is compiled due to CONFIG_IMA_NS and the user namespace
exists there and that has a pointer now to ima_namespace, which can
have that callback. I assumed that other namespaced subsystems could
also be  reached then via such a callback, but I don't know.

Well securityfs is supposed to exist for LSMs.  At some point each of
those is going to need to be namespaced, which may eventually be quite
a pile of callbacks, which is why I thought of a notifier.

quoted

I suppose any late filesystem init callchain would have to be
connected to the user_namespace somehow?

I don't think so; I think just moving some securityfs entries into the
user_namespace and managing the notifier chain from within securityfs
will do for now.  [although I'd have to spec this out in code before I
knew for sure].

It doesn't have to be right in the user_namespace. The IMA namespace is 
connected to the user namespace and holds the dentries now...

Please spec it out...

quoted

quoted

quoted

+int ima_fs_ns_init(struct ima_namespace *ns)
+{
+	ns->mount = securityfs_ns_create_mount(ns->user_ns);

This actually triggers on the call to securityfs_init_fs_context,
but nothing happens because the callback is null.  Every subsequent
use of fscontext will trigger this.  The point of a keyed supeblock
is that fill_super is only called once per key, that's the place we
should be doing this.   It should also probably be a blocking
notifier so anyconsumer of securityfs can be namespaced by
registering for this notifier.

What I don't like about the fill_super is that it gets called too
early:

[   67.058611] securityfs_ns_create_mount @ 102 target user_ns:
ffff95c010698c80; nr_extents: 0
[   67.059836] securityfs_fill_super @ 47  user_ns:
ffff95c010698c80;
nr_extents: 0

Right, it's being activated by securityfs_ns_create_mount which is
called as soon as the user_ns is created.

Well, that doesn't help us then...

quoted

We are switching to the target user namespace in
securityfs_ns_create_mount. The expected nr_extents at this point is
0, since user_ns hasn't been configured, yet. But then
security_fill_super is also called with nr_extents 0. We cannot use
that, it's too early!

Exactly, so I was thinking of not having a securityfs_ns_create_mount
at all.  All the securityfs_ns_create.. calls would be in the notifier

But we need to somehow have a call to get_tree_keyed() and have that 
user namespace switched out. I don't know how else to do this other than 
having some function that does that and that is now called 
securityfs_ns_create_mount().

get_tree_keyed() will also call the fill_super() which is called when 
securityfs_ns_create_mount() is called.

[  196.739071] ima_fs_ns_init @ 639 before securityfs_ns_create_mount()
[  196.740426] securityfs_init_fs_context @ 72  user_ns: 
ffffffff98a3cc60; nr_extents: 1
[  196.741519] securityfs_ns_create_mount @ 105 target user_ns: 
ffff9e239753eb80; nr_extents: 0
[  196.742657] securityfs_get_tree @ 60 before get_tree_keyed()
[  196.743418] securityfs_fill_super @ 47  user_ns: ffff9e239753eb80; 
nr_extents: 0
[  196.744467] ima_fs_ns_init @ 641 after securityfs_ns_create_mount()
[  196.745304] ima: Allocated hash algorithm: sha256
[  196.757650] securityfs_init_fs_context @ 72  user_ns: 
ffff9e239753eb80; nr_extents: 1
[  196.758759] securityfs_get_tree @ 60 before get_tree_keyed()

You said it works by 'accident'. I know it works because the function 
securityfs_init_fs_context() that now populates the filesystem via the 
late_fs_init() is getting called twice. Does 'accident' here mean the 
call sequence could change?

quoted

Where would the vfsmount pointer reside? For now it's in
ima_namespace, but it sounds like it should be in a more centralized
place? Should it also be  connected to the user_namespace so we can
pick it up using get_user_ns()?

exactly.  I think struct user_namespace should have two elements gated
by a #ifdef CONFIG_SECURITYFS which are the vfsmount and the
mount_count for passing into simple_pin_fs.

Also that we can do for as long as it flies beyond the conversation 
here... :-) Anyone else have an opinion ?

   Stefan

James