imagine the following implementation:
- NFS server on private LAN, vlan 5 on the switch's port
- Linux server, with system NFS share and getting kernel through tftp as part of the nfsroot process, vlan 5 for private IP address and vlan 15 and 20 on public IP ranges.
NFS client interfaces configuration is like :
- eth0 with private IP address
- vlan15 using raw-device-id eth0, with one public IP address
- vlan20 using raw-device-id eth0, with one public IP address on another range
From the Foundry point of view, the port of the nfs client is tagged on vlan 5, 15 and 20, dual-mode on vlan 5 (please translate to Cisco if needed)
The problem is that when the nfs client boots, he can't mount the NFS share to get its system space unless its switch's port is "no tagged" for vlan 15 and 20 (which needs to be tagged again in order to reach the world).
In fact, he can't even get DHCP answer at that time (well, the DHCP request doesn't reach the DHCP server).
Another nfs client which doesn't need vlan other than 5 (only on private network) can boot without any trouble.
I guess that's because there is no way to tell the kernel "hey dude, use vlan5" while booting.
Is that correct?
Why doesn't use the vlan specified on the dual-mode option of the switch's port?
Is there any workaround?
By the way, how do YOU use vlans on your network?
Thanks for your help.