Re: [PATCH/RFC] New module refcounting for net_proto_family
Max Krasnyansky (firstname.lastname@example.org)
Thu, 20 Feb 2003 09:52:10 -0800
At 11:04 PM 2/18/2003, David S. Miller wrote:
>> From: Rusty Russell <email@example.com>
>> Date: Wed, 19 Feb 2003 14:54:21 +1100
>> Firstly, the owner field should probably be in struct proto_ops not
>> struct socket, where the function pointers are.
>>I think this is one of Alexey's main problems with the patch.
>This is a bit more informative than "oh it's an ugly hack" ;-)
>Ok. I got at least three reasons why I think owner field should be in struct
> - struct proto_ops doesn't exists without struct socket.
> It cannot be registered or otherwise used on it's own.
> - struct sock might inherit (when needed see my explanation about different families)
> its owner from struct socket. In which case sk_set_owner(sk, socket->ops->owner) doesn't
> look right.
> - we might want to protect something else besides socket->ops.
>None of those reasons are critical. If you guys still feel that ->owner must be in struct
>proto_ops be that way, I'm ok with it.
Ok. I'll take that back :).
The thing is that socket->ops is set from the protocol itself not in the generic socket code.
Here is what sock_create() does
if (!(sock = sock_alloc()))
printk(KERN_WARNING "socket: no more sockets\n");
i = -ENFILE; /* Not exactly a match, but its the
closest posix thing */
sock->type = type;
if ((i = net_families[family]->create(sock, protocol)) < 0)
It simply calls net_family->create() which then sets its private struct proto_ops.
So I think owner field should be in the struct socket because it needs to be
accessible from net/socket.c:sock_create()/sock_release().
Dave, Alexey, do you guys still strongly believe that it's a hack ?
If yes what do I need to do to convince otherwise ? ;-)
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to firstname.lastname@example.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/