[Xorp-hackers] Linux VLAN config if interface is down

[Ben Greear]

Pavlin Radoslavov wrote:
> [Note: this is not exactly XORP-specific question, but it affects
> the XORP FEA]
> 
> Ben,
> 
> I have a Linux VLAN-related question which hopefully you can answer,
> because I believe you were the primary author of the VLAN code
> in the Linux kernel.

I did most of the original work, but Patrick is now the maintainer
and knows the code better.

I would like to see this constraint you mention removed as well.
 From what I recall, the main problem is that some HW-Accel drivers
might not handle adding vlans properly while the NIC is disabled.

This seems like something that the NIC driver should fix, if that
actually is the case.

Thanks,
Ben



> 
> While testing the XORP VLAN configuration on Linux, I noticed that
> the kernel will refuse creating a VLAN if the corresponding
> interface is down. Once the interface is UP the VLAN can be created:
> 
> OS: Ubuntu 8.04 Server (running under VMware Fusion)
> Kernel: 2.6.24-19-server
> 
> root at vm-ubuntu[3] ip link set eth1 down
> root at vm-ubuntu[4] vconfig add eth1 10
> ERROR: trying to add VLAN #10 to IF -:eth1:-  error: Network is down
> Exit 3
> root at vm-ubuntu[5] ip link set eth1 up
> root at vm-ubuntu[6] vconfig add eth1 10
> Added VLAN with VID == 10 to IF -:eth1:-
> root at vm-ubuntu[7] ip addr
> ...
> 3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
> qlen 1000
>     link/ether 00:0c:29:c4:42:3f brd ff:ff:ff:ff:ff:ff
>     inet6 fe80::20c:29ff:fec4:423f/64 scope link 
>        valid_lft forever preferred_lft forever
> 4: vlan10 at eth1: <BROADCAST,MULTICAST> mtu 1500 qdisc noop 
>     link/ether 00:0c:29:c4:42:3f brd ff:ff:ff:ff:ff:ff
> 
> Though, once the VLAN is created, the interface can be taken
> down and the VLAN is still there:
> 
> root at vm-ubuntu[8] ip link set eth1 down
> root at vm-ubuntu[9] ip addr
> ...
> 3: eth1: <BROADCAST,MULTICAST> mtu 1500 qdisc pfifo_fast qlen 1000
>     link/ether 00:0c:29:c4:42:3f brd ff:ff:ff:ff:ff:ff
> 4: vlan10 at eth1: <BROADCAST,MULTICAST,M-DOWN> mtu 1500 qdisc noop 
>     link/ether 00:0c:29:c4:42:3f brd ff:ff:ff:ff:ff:ff
> 
> 
> My question is whether this semantic is intentional or is a bug in
> the kernel.
> FYI, when setting the MAC address or the MTU on an interface, the
> Linux kernel requires that the interface is DOWN, but with VLANs
> I see just the opposite requirement.
> 
> The reason I ask this question is because it affects the mechanism
> used the XORP FEA to set network interfaces information.
> Currently, when pushing the state to the kernel, all information
> (MAC addresses, MTU, IP addresses, VLANs, etc) is pushed first and
> the last operation is to enable the interface.
> Operationally this is the correct behavior because we don't want to
> enable half-configured interface.
> With the VLAN issue mentioned above all this logic breaks, and we
> need to do preliminary interface enabling which adds more
> complexity.
> 
> Thanks,
> Pavlin


-- 
Ben Greear <greearb at candelatech.com>
Candela Technologies Inc  http://www.candelatech.com