Re: [dm-devel] [PATCH RFC 0/2] dm-mp netlink events

[Mike Anderson <andmike us ibm com>]

Christophe Varoqui <christophe varoqui free fr> wrote:
> Would uevents work instead of yet another Netlink wire ?
> The multipath daemon already listens for uevents ...

I general similar information could have been sent using the uevents
netlink. Though changes would be needed in kobject_uevent to allow more
event data to be passed in the events environmental data. Also while we
could generate events against /sys/block/dm-* it might be more interesting
if we had more detailed kobjects in the sysfs tree to generate events
against, but that is a larger change.

The reason I decide to use a different netlink number was that I was trying
to use a event method that would have a higher chance of working under low
memory conditions (i.e. the skb mempool). I also used unicast instead of
broadcast. When using netlink_broadcast I could not control the memory
allocation in the skb_clone. While skb_clone just clones the header so it
would me a smaller allocation and you can pass in the flags for the
allocation I was still concerned that this might not work in low memory
cases.

-andmike
--
Michael Anderson
andmike us ibm com