On Tue, 2009-01-06 at 19:20 -0500, Trond Myklebust wrote:
> On Tue, 2009-01-06 at 16:08 -0800, Matt Helsley wrote:
> > IMHO This seems more incorrect than trying to use a more proximal
> > namespace.
>
> You have yet to explain why.
It's "more proximal" -- it's closer to the container that we expect to
cause (directly or otherwise) the bulk of the RPC calls for that mount.
If the container does not wind up sharing that mount with other
containers then the reported node name matches. If the container winds
up sharing the mount with other containers then at least we can learn
which container originated the mount.
I imagine an NFS administrator trying to determine the source of a bunch
of RPC calls. If we just report the initial namespace then that
administrator has to do lots more digging to determine which container
sent the calls (assuming they aren't in different network namespaces).
By not always reporting the initial namespace we may give the
administrator one way to narrow down the search. Even if the reported
node name does not perfectly match the source of all RPC traffic related
to the mount at least the administrator gets something more specific.
Cheers,
-Matt Helsley
On Tue, 2009-01-06 at 16:43 -0800, Matt Helsley wrote:
> On Tue, 2009-01-06 at 19:20 -0500, Trond Myklebust wrote:
> > On Tue, 2009-01-06 at 16:08 -0800, Matt Helsley wrote:
> > > IMHO This seems more incorrect than trying to use a more proximal
> > > namespace.
> >
> > You have yet to explain why.
>
> It's "more proximal" -- it's closer to the container that we expect to
> cause (directly or otherwise) the bulk of the RPC calls for that mount.
> If the container does not wind up sharing that mount with other
> containers then the reported node name matches. If the container winds
> up sharing the mount with other containers then at least we can learn
> which container originated the mount.
>
> I imagine an NFS administrator trying to determine the source of a bunch
> of RPC calls. If we just report the initial namespace then that
> administrator has to do lots more digging to determine which container
> sent the calls (assuming they aren't in different network namespaces).
As I said elsewhere, the network namespaces do not matter, since we
always create the sockets via the rpciod workqueue.
> By not always reporting the initial namespace we may give the
> administrator one way to narrow down the search. Even if the reported
> node name does not perfectly match the source of all RPC traffic related
> to the mount at least the administrator gets something more specific.
The administrator would have to be running wireshark or something in
order to track the rpc calls. That's hardly a convenient interface, or a
compelling reason to add namespace info into RPC credentials. We
certainly haven't added any other information to the RPC calls in order
to allow the administrator to identify which process it originated from.
Trond
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com