Sorry for the late response.
On 08/21/2014 08:34 AM, Baoquan He wrote:
On 08/20/14 at 10:42am, Vivek Goyal wrote:
> Hi Bao,
> Using "ip route get to" requires knowing the target. But that's little
> different than bringing up an interface. There can be multiple static
> routes going through an interface which lead to multiple targets.
> I think this makes the whole mechanism less generic. I liked the previous
> method better where if we bring an interface, all the static routes going
> through that interface will come up too. (Instead of only one static
> route coming up for the target).
> So I would rather stick to the mechanism of determining all the routes
> going through an interface and add the ones which are not present in
> second kernel's routing table.
I think you are right about this. Since Marc said other modules could
use network too, such as heartbeat. In kdump we need build the network
conf for it. I don't remember this clearly, need wait for Marc to check
The reason I tried the exact route is some duplicate routes may exist in
dracut. From dracut's point of view, it can handle route correctly if
user can pass it the correct route. If duplicate routes are added,
either we tolerate the warning message, or we require to change the
implementation of dracut handling. But from dracut's side, they don't do
anything wrong, why they need be changed.
I have found the following things that could each use a different route,
and should be fully supported:
* ssh/NFS/iscsi - but only one of these per configuration (as far as
* Name resolution (/etc/resolv.conf specifies this)
I am not sure if we support:
Additionally, customers could access the network through their own
scripts, but it would be reasonable to not set up networks specifically