> On Tue, Dec 21, 2010 at 06:45:21PM -0500, J. Bruce Fields wrote:
> > On Wed, Dec 22, 2010 at 01:32:15AM +0200, Kirill A. Shutemov wrote:
> > > On Mon, Dec 20, 2010 at 09:46:44AM -0500, J. Bruce Fields wrote:
> > > > By the way, was there ever a resolution to Trond's question?:
> > > >
> > > >
http://marc.info/?l=linux-nfs&m=128655758712817&w=2
> > > >
> > > > "The keyring upcalls are currently initiated through the same
> > > > mechanism as module_request and therefore get started with the
> > > > init_nsproxy namespace. We'd really like them to run inside the
> > > > same container as the process. As part of the same problem,
> > > > there is the issue of what to do with the dns resolver and
> > > > Bryan's new keyring based idmapper code."
> > >
> > > I'm not sure that I understand the problem correctly.
> > >
> > > Currently, idmap uses dentry taken from client's cl_rpcclient->cl_path
> > > (see nfs_idmap_new()). cl_rpcclient (and cl_path) is initialized with
> > > rpcmount resolved against mount namespace of mount process (see
> > > nfs_create_rpc_client()).
> > > I assume it's correct.
> >
> > There's actually two separate sets of idmapper code; look at
> > fs/nfs/idmapper.c, the first part of the file (between #ifdef
> > CONFIG_NFS_USE_NEW_IDMAPPER and #else) is idmapping code that uses
> > request_key(). The code you're looking at (including nfs_idmap_new())
> > is later in the file, and deprecated.
>
> IIUC, we need to save nsproxy of mount process in struct nfs_client and
> pass it down to request_key(). I think it's outside of this patchset.
