From mboxrd@z Thu Jan 1 00:00:00 1970 From: Kay Sievers Date: Tue, 04 Jan 2005 18:38:36 +0000 Subject: Re: Hotplugging ethernet cables. Message-Id: <1104863916.5258.32.camel@localhost.localdomain> List-Id: References: <41D74A49.2060607@are-b.org> In-Reply-To: <41D74A49.2060607@are-b.org> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-hotplug@vger.kernel.org On Tue, 2005-01-04 at 11:44 -0600, Linas Vepstas wrote: > A naive question re ethernet hotplug: what if its iSCSI? > > On Sun, Jan 02, 2005 at 01:43:32AM +0100, Kay Sievers was heard to remark: > > On Sun, 2005-01-02 at 02:11 +0100, oliver wrote: > > > They mention ifplugd and even though this seems to work reasonably, I > > > was wondering why hotplugging doesn't support ethernet hotplugging > > > nativly? > > > > Hotplug supports device add and remove, not state changes of hardware. > > > > > As I understand it the hotplugging daemon responds to events > > > created by several devices, where the ifplugd polls and the device. > > > > There is no such daemon involved in the hotplug handling. Hotplug is > > driven by more or less stateless scripts invoked by the kernel. > > > > The network link state changes can be received on a netlink socket from > > the kernel. These events are not hotplug events and don't really fit > > into the model of linux hotplug. > > > I can imagine a scenario where Linux has mounted a filesystem that > is sitting on an iSCSI disk device. Thus, in many respects, this is > analogous to a USB disk being plugged and unplugged. I also presume > that iSCSI devices need not be just disks, but could be "anything", > thus making the situation even more analogous to USB. Sure, it's similar in some aspects, but linux-hotplug is only a device "add" or "remove" notification. Every event corresponds to a created or removed directory in /sys. There is no such thing for a network link. If you add/remove the network device you get hotplug events. You don't get hotplug events if you press a button of an USB device, which is very similar to the network link state change. :) Some networking applications produce a very high event traffic for state changes - like the zebra routing daemon, this is definitely not possible to do with the event-by-forked-helper-application-hotplug system. They use the netlink socket for good reason. What do you think is the problem to get network layer events over netlink? It works nicely this way for a long time. For a application level feature rich example, you may look how HAL is collecting all available kernel events and offers a unified interface to applications to subscribe to these events: http://cvs.freedesktop.org/*checkout*/hal/hal/doc/spec/hal-spec.html?rev=1.36.2.5 > I'm also curious about plans for things like infiniband fabrics, > which might be plugged/unplugged in various ways, which the OS might > want to know about. Again, this seems clearly analogous to USB fabrics Never seen such a thing. :) Kay ------------------------------------------------------- The SF.Net email is sponsored by: Beat the post-holiday blues Get a FREE limited edition SourceForge.net t-shirt from ThinkGeek. It's fun and FREE -- well, almost....http://www.thinkgeek.com/sfshirt _______________________________________________ Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net Linux-hotplug-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel