On Mon, Aug 08, 2005 at 01:03:44AM +0200, Patrick McHardy ([EMAIL PROTECTED])
wrote:
> Thomas Graf wrote:
> >Anyways, your idea of having 32bit identifiers rather than a
> >bitmask makes sense to me, however I think we should not lose
> >the ability to do multicasting for event notifications so we
> >might need a slightly more complex solution. I guess Patrick
> >is taking care of this though.
>
> Yes, the patches are almost ready but I could use some ideas
> for one remaining problem. The idea was that a subsystem which
> sends broadcast messages can specify how many groups are present,
> probably as a parameter to netlink_kernel_create. This information
> is needed to properly size the groups mask in struct netlink_sock
> (this mask is still needed). The problem is that sockets can bind
> to groups before a kernel socket is created, or in the case of
> a userspace implemented protocol, none is ever created. To solve
> this we could forbid binding to groups before either a kernel
> socket is created or userspace has registered how many groups
> are present using setsockopt. The questions regarding this are
> if the bind-change could cause problems for userspace and whether
> we should take care of userspace implemented protocols at all or
> simply return -ESOCKTNOSUPPORT in netlink_create if no kernel
> socket exists.
Should it be possible to specify remote groups in sending time?
I.e. netlink_broadcast(..., u32 *array_of_groups, int number_of_groups);
This will not hurt existing userspace applications.
--
Evgeniy Polyakov
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
