Re: [PATCH net] selftests: forwarding: Fix failing tests with old libnet

[Petr Machata <petrm@nvidia.com>]

Jiri Pirko <jiri@nvidia.com> writes:

> Tue, Aug 09, 2022 at 01:33:20PM CEST, idosch@nvidia.com wrote:
>>diff --git a/tools/testing/selftests/net/forwarding/custom_multipath_hash.sh b/tools/testing/selftests/net/forwarding/custom_multipath_hash.sh
>>index a15d21dc035a..56eb83d1a3bd 100755
>>--- a/tools/testing/selftests/net/forwarding/custom_multipath_hash.sh
>>+++ b/tools/testing/selftests/net/forwarding/custom_multipath_hash.sh
>>@@ -181,37 +181,43 @@ ping_ipv6()
>> 
>> send_src_ipv4()
>> {
>>-	$MZ $h1 -q -p 64 -A "198.51.100.2-198.51.100.253" -B 203.0.113.2 \
>>+	ip vrf exec v$h1 $MZ $h1 -q -p 64 \
>
> Not directly related to this, but I was wondering, if it would be
> possible to use $IP and allow user to replace the system-wide "ip" for
> testing purposes...

A typical forwarding test builds a whole topology with a number of IP
addresses and assumptions about them. E.g. routes need network address
derived from address of the remote endpoint, directly attached hosts
need addresses from the same network, flower rules etc. might match
pieces of endpoint addresses, the above code fragment actually walks a
range of IPs... So it's not just $IP. At the minimum it's a map of IPs
plus some helpers to derive the rest, and conversion of the selftests to
correctly use the helpers. Doable, sure, but I suspect a fairly messy
deal.

That's why the selftests use reserved ranges: 192.0.2.0/24,
198.51.100.0/24 etc. and 2001:db8::/32. They are not supposed to
conflict with whatever addresses the user uses.