On Thu, Nov 13, 2008 at 10:07:35AM -0500, Dan Williams wrote: > On Thu, 2008-11-13 at 09:20 -0500, Chuck Anderson wrote: > > On Thu, Nov 13, 2008 at 08:12:24AM -0600, Les Mikesell wrote: > > > Chuck Anderson wrote: > > >> On Thu, Nov 13, 2008 at 02:27:42PM +0100, Adam Tkac wrote: > > >>> Crucial thing is static IPs which NM can't handle. > > >> > > >> False. > > > > > > If you bring up a mix of static and dynamically assigned interfaces, can > > > you control which gets to assign the default route and DNS servers? > > > > The last time I looked at the code, NM had a hard-coded policy for how > > it assigns the default route and DNS servers. If you only have one > > interface with a default route and the other interfaces don't have a > > default route, then the DNS and default route for that interface is > > set. If you have more than one, I believe it picks based on the > > hard-coded policy, which I believe is wired first, then wireless, then > > dialup/mobile broadband. In the face of multiple wired connections, > > I'm not sure what the policy is. > > > > Yes, NM needs to grow the ability to specify policy outside of the > > code. > > The current policy is this. If the device gets a gateway from anything > (either DHCP or the GUI or the ifcfg files) then it becomes a candidate > for the default route. If you don't enter a gateway for any of that > device's IP addresses in the GUI, it shouldn't ever get the default > route. > > Yes, there's room for improvement. If you check "Ignore automatically > provided routes" and you're using DHCP, NM should probably ignore the > gateway for that device when deciding which device gets the default > route. > > > It also needs IPv6 support (coming soon I hear), alias support, > > IPv6: target for next version > Alias: you can already add multiple IPs to an interface > bridging: planned > bonding: planned > VLANs: good to have, needs more investigation The combo of last 3 is key for many virtualization deployments. eg, we want to put eth0 + eth1 into a bonding device bond0, then create 10 vlans bond0.0, bond0.1.... devices ontop of bond0, and finally put each of these vlan devices into its own bridge. The guests' tap devices are then connected to various bridges according to which VLAN you want the guest to see. AFAIK, its not even possible to get this config working with regular networking initscripts in all cases, let alone networkmanager. So people currently use horrible shell scripts to configure it, eg as per this doc http://www.certifried.com/files/Xen_networking.pdf Daniel -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :| -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
