Re: [PATCH net-next] selftests: net: add IPv4 and IPv6 same scope address order check

[Ido Schimmel <idosch@nvidia.com>]

On Tue, Sep 01, 2026 at 10:22:00AM +0200, Martin Jabůrek wrote:
> Introduce two new tests `ipv4_verify_addr_order` and
> `ipv6_verify_addr_order`, to check the ordering of
> a set of IP addresses after insertion.
> 
> The implementations of these protocols are inconsistent
> in regard to address ordering. IPv4 addresses stay in the
> same order as inserted while IPv6 addresses appear in
> reverse order. This incosistency has prompted attempts

s/incosistency/inconsistency/

> to unify the ordering, so both protocols act the same
> (as IPv4). This however caused user-space regressions in
> certain applications, which relied on the order as it
> was prior to the change (particularly NetworkManager).
> 
> The addition of these tests aims to consolidate the
> behaviour to prevent regressions in the future.
> The expected behaviour is the initial one, where each
> protocol acts differently.
> 
> Tests were verified on a recent commit with the expected
> behaviour (61eb236c41c2) and a commit making both protocols
> act the same way (cb3de96eea66). Tests respectively pass
> and not pass as expected.
> 
> Conversations detailing the decision process for creating
> these tests are linked below.
> 
> Link: https://lore.kernel.org/netdev/20260521135310.GC977@cmadams.net/
> Link: https://lore.kernel.org/netdev/20260529112357.5079-1-fmancera@suse.de/
> Suggested-by: Fernando Fernandez Mancera <fmancera@suse.de>
> Signed-off-by: Martin Jabůrek <martin.jaburek@suse.com>
> ---
>  tools/testing/selftests/net/rtnetlink.py | 58 +++++++++++++++++++++++-
>  1 file changed, 57 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/testing/selftests/net/rtnetlink.py b/tools/testing/selftests/net/rtnetlink.py
> index 5cc3ebdcf08d..1ee604e4e5f6 100755
> --- a/tools/testing/selftests/net/rtnetlink.py
> +++ b/tools/testing/selftests/net/rtnetlink.py
> @@ -314,11 +314,67 @@ def ipv6_route_del_reason_absent() -> None:
>                          "user deletion must not carry del-reason")
>  
>  
> +def ipv4_verify_addr_order() -> None:
> +    """
> +    After inserting multiple same scope IPv4 addresses, their order
> +    must be the same as the insertion order.
> +
> +    See function ipv6_verify_addr_order in this file for further details.
> +    """

Please also test the inter-scope order. It also differs between IPv4 and
IPv6 and recently someone tried to align IPv4 with IPv6:

https://lore.kernel.org/netdev/20260721090114.GA2510713@shredder/

> +
> +    DEV_NAME = "dummy_dev"
> +    TEST_ADDRESSES = ["192.0.2.1", "192.0.2.2", "192.0.2.3"]

All the addresses are in the same subnet so this only tests the ordering
between a primary address and its secondary addresses. Please add
another address from a different subnet.

> +
> +    with NetNS() as ns:
> +        with NetNSEnter(str(ns)):
> +            ip(f"link add name {DEV_NAME} type dummy", ns=str(ns))
> +            for addr in TEST_ADDRESSES:
> +                ip(f"address add {addr}/24 dev {DEV_NAME}", ns=str(ns))
> +            ip(f"link set dev {DEV_NAME} up", ns=str(ns))
> +
> +            rtnl = RtnlAddrFamily()
> +            addrs = rtnl.getaddr({"ifa-family": socket.AF_INET}, dump=True)
> +            address_list = [addr["address"] for addr in addrs]
> +
> +            ksft_eq(TEST_ADDRESSES, address_list, "Incorrect IPv4 address order")
> +
> +
> +def ipv6_verify_addr_order() -> None:
> +    """
> +    After inserting multiple same scope IPv6 addresses, their order
> +    must be the _reverse_ of the insertion order.
> +
> +    While this behaviour is different from how IPv4 acts,
> +    updating the IPv6 implementation to act the same way
> +    has proved to cause user-space application regressions
> +    (particularly in NetworkManager). This behaviour is being
> +    tested for to consolidate it as being expected and correct.
> +    """
> +
> +    DEV_NAME = "dummy_dev"
> +    TEST_ADDRESSES = ["2001:db8::1", "2001:db8::2", "2001:db8::3"]
> +
> +    with NetNS() as ns:
> +        with NetNSEnter(str(ns)):
> +            ip(f"link add name {DEV_NAME} type dummy", ns=str(ns))
> +            for addr in TEST_ADDRESSES:
> +                ip(f"address add {addr}/32 dev {DEV_NAME}", ns=str(ns))

/64 is the more conventional choice.

> +            ip(f"link set dev {DEV_NAME} up", ns=str(ns))
> +
> +            rtnl = RtnlAddrFamily()
> +            addrs = rtnl.getaddr({"ifa-family": socket.AF_INET6}, dump=True)
> +            address_list = [addr["address"] for addr in addrs]
> +
> +            # We ignore the link-local address present by default.
> +            ksft_eq(TEST_ADDRESSES[::-1], address_list[:3], "Incorrect IPv6 address order")
> +
> +
>  def main() -> None:
>      ksft_run([dump_mcaddr_check, dump_mcaddr6_check, ipv4_devconf_notify,
>                ipv6_route_del_reason_expired,
>                ipv6_route_del_reason_ra_withdrawn,
> -              ipv6_route_del_reason_absent])
> +              ipv6_route_del_reason_absent,
> +              ipv4_verify_addr_order, ipv6_verify_addr_order])
>      ksft_exit()
>  
>  if __name__ == "__main__":
> -- 
> 2.55.0
>