Re: VKernel progress update - 10 Jan 2006

["Sepherosa Ziehau" <sepherosa@xxxxxxxxx>]

On 1/12/07, Matthew Dillon <dillon@apollo.backplane.com> wrote:
>     I would suggest that VKE's be numbered from zero, because then we could
>     have the vkernel automatically allocate a free TAP interface and not
>     create confusion with rc.conf in the virtual environment.
>
>     Also, the vkernel can ifconfig up the TAP side of the interface as well
>     or do the basic bridge tie-in.  That way the sysop wouldn't have to
>     worry about which tap interface was allocated.
>
>     Bridging is easy mode but not paricularly efficient because broadcasts
>     would wind up being spammed to all the running vkernels (not fun!).
>     routing a LAN IP might be a better solution.  I can see a need for both
>     solutions.
>
>     So, e.g. leave the -I option intact but add some features and optional
>     real-kernel-side IP configuration.  like this:
>
>     -I auto:10.0.0.1:10.0.0.2
>     -I auto:10.0.0.1:10.0.0.2:24
>     -I auto:bridge0
>     -I tap0:bridge0
>
>        auto        - automatically find a free TAP interface and use it
>        10.0.0.1    - specify TAP interface IP
>        10.0.0.2    - specify VKE interface IP
>        24          - specify CIDR bits for TAP/BKE (e.g. would be 10.0.0.0/24)
>                      if not specified a /30 would be used.
>
>        bridgeN     - instead of specifying IPs, just tie the TAP interface
>                      into the specified bridge.
>
>                      (we have to make sure that disconnecting the TAP also
>                      disconnects it from the bridge)

I think following patch does these:
http://leaf.dragonflybsd.org/~sephe/vke_multi_syntax.diff

NOTE:
1) bridge will not be brought up automaticly
2) non-existent bridge will not be created

Except above syntax, I have extended it a little bit:
-I tap0:x.x.x.x:y
 will only set real kernel tap0's address and netmask
-I tap0:x.x.x.x
 will only set real kernel tap0's address

-I tap0:bridge0:x.x.x.x
 will tie tap0 to bridge0 in real kernel, and x.x.x.x will be vkeX's address
-I tap0:bridge0:x.x.x.x:y
 will tie tap0 to bridge0 in real kernel, and x.x.x.x will be vkeX's
address and y will be vke0's netmask

If you have assigned vkeX's address/netmask on vkernel command line,
then in vkernel you only need 'ifconfig vkeX up', vkeX will have the
address/netmask supplied on command line.

Two sysctls are added for vkeX:
hw.vkeX.intr_rate (rw), #intr/second, default is 20, valid range is [1, hz]
hw.vkeX.tap_unit (ro), backend tap(4) interface unit in real kernel
And now 'ifconfig vkeX' will print backend tap(4) interface in real kernel

Make sure you have pulled in two tap(4) bug fixing in HAED, before
trying this patch.

Please test/review it.  Hope I can drive it into repo before 1.8 branching.

Best Regards,
sephe

--
Live Free or Die