[LTP] [PATCH 1/1] net/route: Rewrite route-rmmod to new API

[Alexey Kodanev]

Hi Petr,
On 02.09.2021 19:45, Petr Vorel wrote:
> Hi Alexey,
> 
> <snip>
>>>> BTW, why not using add_macvlan() in route_lib.sh (or gre, vxlan, etc.)
>>>> and remove that driver, so that this test can be run with custom setup, and
>>>> with remote host setup?
> 
>>> Looking into this old patch, it looked to me quite bad approach to move
>>> add_macvlan() into tst_net.sh to be reused (you didn't suggested that, that's
>>> what I'd do to prevent the duplicity).
> 
>> Why not move add_macvlan() to the virt_lib.sh, I think this lib is better
>> suited for creating macvlan?
> Well, route-change-if.sh and route-change-netlink-if.sh depends on helpers from
> route-lib.sh. Using both libraries would require little changes, but sure it's
> possible (+ add obviously add $virt_type).
> 
> It's just a bit strange to add this special purpose function when there is
> virt_add() and virt_add_rhost(). Best will be probably to add macvlan to these
> functions and migrate router tests which needs macvlan to use virt_lib.sh before
> route-rmmod.sh using it.

Yes, it would be better to have the generic ones, moreover macvlan01/macvlan02
tests are already calling virt_add().

> 
> In the long term I'd really prefer to add some TST_NET* variable (due doc
> generation via docparse), but that can be postponed as another effort after the
> release (I thought it'd have to be in tst_net.sh so that tests does not care
> about including virt_lib.sh when declaring it, but variable could also have
> prefix TST_NET_VIRT_ to make it obvious, that virt_lib.sh must be loaded. But I
> still prefer moving to tst_net.sh when implementing this approach).
> 
> BTW you consider ok to use macvlan for testing this? I suppose this test was
> intended to be used on the real hardware not on virtualization. But I don't have
> proper setup and give up on this approach.

I think we are not changing the purpose of this test using a virtual
driver. Using the real drivers would complicate the test setup
unnecessarily considering what it does with the driver...

> 
> Kind regards,
> Petr
> 
>>> But much better approach would be IMHO to move virt_add() and virt_add_rhost()
>>> from virt_lib.sh to tst_net.sh and adjust it not to be too tight to virt_lib.sh.
>>> I suppose $virt_type should became $2 (second parameter).
> 
>>> Also there could be moved from virt_lib.sh to tst_net.sh: e.g. add flag
>>> TST_NET_ADD_VIRT_TYPE (e.g. macvlan, gre, ...) and doing setup and cleanup
>>> there. We could reduce code and document which virt drivers are used.
> 
>>> route-change-netlink-if is the only test which needs to call tst_init_iface()
>>> (to add routes), virt_lib.sh does not need it.
> 
>>> Kind regards,
>>> Petr
> 
>