Re: [PATCH] implement in-kernel keys & keyring management

[David Howells <dhowells@redhat.com>]

Greg KH <greg@kroah.com> wrote:
> I think that if the /proc interface was moved over to sysfs (which is
> where it should be), a number of these syscalls would go away.

Well, I could move these two files into /sysfs. But just doing that wouldn't
get rid of any of the system calls. To move these files into sysfs, should I
create a "keys" subsystem?

Can you elaborate as to what you envision? I wonder if you'd thinking that I
should make every key a kobject and fan-out them out in a directory in sysfs
somewhere. I really don't want to do that, though... kobject seems to add
quite a large overhead that I'd rather avoid (a directory in sysfs for
instance).

I could a keyfs filesystem, fan the keys out in there, but this would spawn
more code than just a few new syscalls or prctls. However, I can't just
pretend all keyrings are directories and all keys files and then use link()
and unlink(). I'd need to be able to link() and unlink() directories. I could
do it by representing two keyrings, as two adjacent directories, and then use
symlink() to create a link between them.

The main advantage of doing this, however, is that shell scripts would be able
to modify their own keyrings without a utility program such as keyctl.c that I
put up for download.

David