On 04/13/2011 05:53 PM, Michael Chan wrote:
On Wed, 2011-04-13 at 10:00 -0700, Mike Christie wrote:
ccing Broadcom devs, Some questions for you guys below
On 04/02/2011 01:34 PM, vikas.chaudhary@xxxxxxxxxx wrote:
From: Vikas Chaudhary<vikas.chaudhary@xxxxxxxxxx>
To support multiple network addresses per adapter need to have a new way to
represent network interface (net iface) in sysfs.
Currently only one ipaddress and hwaddress is displayed
\# ls /sys/class/iscsi_host/host18
device hwaddress initiatorname ipaddress power subsystem uevent
In this patch the net iface is presented as a separate class device.
The one that can be added/removed dynamically or statically, based on how
the user configures the multiple net iface on the adapter.
The new sysfs directory would look like this
\# /sys/class/iscsi_iface/
|
|- ipv4-iface-<host_no>-<iface_no>/<-- for ipv4
|- ipaddress
|- subnet
|- gateway
|- bootproto
|- state
|- ipv6-iface-<host_no>-<iface_no>/<-- for ipv6
|- ipaddress
|- link_local_addr
|- router_addr
|- ipaddr_autocfg
|- linklocal_autocfg
|- state
With patch "[RFC-V2 PATCH 1/5] iscsi_transport: add support for
set_net_config" userspace would send down the vlan info.
If we add a vlan sysfs file to the iscsi_iface, to export the info was
bnx2i going call iscsi_create_iface for each vlan? If so I am not sure
what bnx2i will use for the iface_num. It is supposed to be persistent,
right?
bnx2i doesn't need to know the IP addresses and VLANs for each HBA. It
is all handled in userspace by uIP. So if we add this sysfs,
iscsi_transport can handle sysfs creation for devices generically,
right?
With these patches the driver need to call iface_create_iface and
implement a callout on the iscsi_transport template so
scsi_transport_iscsi.c can call into the driver to get the values to
display in sysfs like is done for the connection and session settings.
We can add some helpers in libiscsi for this like is done for the
conn/session stuff.
For bnx2i, when doing iscsi offload and vlans, do you have to have a
netdev like ethX.Y setup for each vlan or can bnx2i operate without it
(the call to cnic_get_vlan always throws me and I cannot remember if we
were going to still do that or change something in the driver so you did
not need it).
For now, cnic_get_vlan will try to get a matching VLAN from netdev with
a route to the target. But it is not required. Once we have VLAN
attribute in iface, uIP can override and use the VLAN specified in
iface.
Ok.
And just to confirm for vlans and bnx2i, when we make a ep and session,
cnic_get_route/cnic_cm_select_dev will do the magic to figure out what
vlan to use?
Yes for now. The plan is for uIP to override if a VLAN is specified in
iface.
Is the idea to have uIP read the userspace iface config files in
/etc/iscsi/iface or /var/lib/iscsi/ifaces or iscsid (basically something
in userspace) then pass down in a msg_set_path msg?
This is the flow:
-> ep_connect
-> cnic_cm_connect
/* Try to get route/MTU/VLAN from netdev if available */
-> cnic_get_route
-> send nl message to uIP to get next hop
-> recv nl message from uIP
/* at this point we may get a new VLAN ID from uIP */
-> make connection to target
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
