On Thu, 2009-04-02 at 21:28 +0100, Al Viro wrote: > On Thu, Apr 02, 2009 at 01:17:47PM -0700, Linus Torvalds wrote: > > > > > > On Thu, 2 Apr 2009, Al Viro wrote: > > > > > > *growl* > > > > > > Right, we are out of clone() flags, thanks to namespace-of-the-week > > > insanity by openvz bunch ;-/ > > > > I really think "clone()" is the wrong kind of granularity, and much too > > big a hammer. > > > > You never answered my suggestion about passing in namespace details to the > > open routines in nameidata. I don't think it actually needs the whole > > thing, the only thing (I _think_) it cares about ends up being just the > > root/cwd. > > The thing is, it wants more - it wants to mount stuff there. Which means > that we really want current switched to it, or the normal check_mnt() > logics will screw you. > > I don't have too serious objections against having a reference to namespace > in nameidata, but I really doubt that it'll be enough here. The NFS client is constructing the namespace as it walks the path using the ->follow_link() trick to automount new filesystems. As long as we can add a mount namespace argument to do_add_mount(), then we should be able to make use of a namespace in struct nameidata. The other thing we'd need is a 'root' path argument to ensure that __vfs_follow_link() does the right thing when presented with an absolute symlink. Given those two, plus a method for creating an empty namespace, I think we should have all we need. -- Trond Myklebust Linux NFS client maintainer NetApp Trond.Myklebust@xxxxxxxxxx www.netapp.com -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
