Re: [PATCH 4/5] AF_RXRPC: Key facility changes for AF_RXRPC [try #2]

[David Howells <dhowells@redhat.com>]

Christoph Hellwig <hch@infradead.org> wrote:

> You wrote the code so there must be some reason for this, but these
> changes look rather odd to me :)

The union is for use by the type in any way it sees fit, but it may not want to
use it as a list_head.  So all I've done is to make it available as a pair of
arbitrary pointers or a pair of arbitrary numbers.

Maybe a better way would be to have an overlay struct that's private to the
type, sort of how sk_buff::cb works.

> Having a type exported is really odd, how is this supposed to be a public API?

Keyrings are a special type.

It occurred to me whilst doing this that the best way to achieve what I wanted
was by dealing with rings of keys.  What I needed was for the server app to
give the kernel a key for each security type it wanted to support, which the
kernel would then have to retain.  It seems natural to use a keyring to do the
retension as that's its purpose.

Another way to look at it is that in the client I need just one key at once,
and I can get that from the process as it's setting up the connection.
However, in the server I need to have several keys, and I need them available
up front because the server app doesn't set up a connection, the kernel does,
and it needs the keys immediately.

I'll add a mention to Documentation/keys.txt to record this exportation.

David