Also adding Karen, because I think we were talking about this before.
On 11/03/2010 02:36 PM, Mike Christie wrote:
ccing Ben Li from Broadcom.
On 11/02/2010 06:42 PM, Anish Bhatt wrote:
MAC addresses specified in iface files are currently case sensitive.
This can be an issue on vlan interfaces, as isciadm -m iface does not
create config files for vlan interfaces and they need to be written
manually. Attached patch will convert uppercase MAC addresses to lower
case after reading them (iscsiadm expects them in lower case only).
There seems to be another issue with VLAN interfaces. Interface files
for offloaded vlan interfaces (on chelsio cards atleast) need a hardware
address to be specified to work. As of now, iscsiadm does not catch this
So for software iscsi and vlans, you just need to specify the netdev
vlan name (for example eth0.1 for the iface.net_ifacename). You do not
need to set the hwaddress/mac in the iface.
iscsiadm does not catch what you need for offload because I had no idea
you were doing this.
For chelsio offload how are you guys specifying things when vlans are
used and what does cxgb3i export? Do you have to create a vlan with the
normal linux tools like vconfig first, then for the net_ifacename are
you specifying the vlan name created by vconfig?
cxgbi_device_find_by_netdev then matches the names with the real netdev?
I guess I am asking what does the kernel driver need? So right now
userspace would match a scsi_host with whatever is specified in the
iface. It then passes down the shost to cxgbi_ep_connect.
cxgbi_check_route then finds a netdev and we check if it matches the
shost hba passed in.
For vlans, then does cxgbi_check_route and cxgbi_device_find_by_netdev
just find the first vlan that matches the route and you use that one?
If we want to support something like EQL (where you commonly have
multiple ports on the same subnet and may want traffic going out
specific devices to specific target ports) plus vlans do we need to
specify the specific vlan id in userspace then pass that down to the
kernel to force it to use it?
Why do you need to pass in the hardware address?
So if we use this:
There was talk about adding a vlan id:
http://groups.google.com/group/open-iscsi/browse_thread/thread/7ee3b5bb5502871b/37fe62da7d1ffa4e?lnk=gst&q=Using+IFACE_SUBNET_MASK+and+IFACE_VLAN#37fe62da7d1ffa4e
and pass down the vlan id, would that allow the kernel driver to match
ids? Would it just allow us to match ids, but is there still a problem
where find_route_ipv4 might not return the vlan we wanted to use?
so we could have a more common way of specifying the vlan to be used.
and will simply fail. This is already noted in the readme, but would
What note in the readme are you referring to?
there be any interest in adding functionality to 'iscsiadm -m iface' to
handle vlan interfaces, or warn about missing hwaddress keys when it
comes to offloaded vlan interfaces ?
-Anish
--
You received this message because you are subscribed to the Google Groups
"open-iscsi" group.
To post to this group, send email to open-is...@googlegroups.com.
To unsubscribe from this group, send email to
open-iscsi+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/open-iscsi?hl=en.
