Support devices with multiple IPv4 addresses

Add a get_ipv4_addresses() method to ethtool.etherinfo to support devices
with multiple IPv4 addresses (rhbz#759150)

Previously, get_etherinfo() made queries to NETLINK with NLQRY_ADDR, and
callback_nl_address handled responses of family AF_INET (IPv4) by writing to
fields within a struct etherinfo.

If multiple AF_INET responses come back, each overwrote the last, and the
last one won.

This patch generalizes things by moving the relevant fields:
   char *ipv4_address;                 /**< Configured IPv4 address */
   int ipv4_netmask;                   /**< Configured IPv4 netmask */
   char *ipv4_broadcast;
from (struct etherinfo) into a new Python class, currently named
PyNetlinkIPv4Address.

This object has a sane repr():
>>> ethtool.get_interfaces_info('eth1')[0].get_ipv4_addresses()
[ethtool.NetlinkIPv4Address(address='192.168.1.10', netmask=24, broadcast='192.168.1.255')]

and attributes:
>>> print [iface.address for iface in ethtool.get_interfaces_info('eth1')[0].get_ipv4_addresses()]
['192.168.1.10']
>>> print [iface.netmask for iface in ethtool.get_interfaces_info('eth1')[0].get_ipv4_addresses()]
[24]
>>> print [iface.broadcast for iface in ethtool.get_interfaces_info('eth1')[0].get_ipv4_addresses()]
['192.168.1.255']

The (struct etherinfo) then gains a new field:
   PyObject *ipv4_addresses;        /**< list of PyNetlinkIPv4Address instances */

which is created before starting the query, and populated by the callback as
responses come in.

All direct usage of the old fields (which assumed a single IPv4 address)
are changed to use the last entry in the list (if any), to mimic the old
behavior.

dump_etherinfo() and _ethtool_etherinfo_str() are changed to loop over all
of the IPv4 addresses when outputting, rather than just outputting one.

Caveats:
* the exact terminology is probably incorrect: I'm not a networking
  specialist
* the relationship between each of devices, get_interfaces_info() results,
  and addresses seems both unclear and messy to me: how changable is the
  API?

>>> ethtool.get_interfaces_info('eth1')[0].get_ipv4_addresses()
[ethtool.NetlinkIPv4Address(address='192.168.1.10', netmask=24, broadcast='192.168.1.255')]

It seems that an etherinfo object relates to a device: perhaps it should be
named as such?  But it may be too late to make this change.

Notes:
The _ethtool_etherinfo_members array within python-ethtool/etherinfo_obj.c
was broken: it defined 4 attributes of type PyObject*, to be extracted from
etherinfo_py->data, which is of a completed different type.  If these
PyMemberDef fields were ever used, Python would segfault.  Thankfully
_ethtool_etherinfo_getter() has handlers for these attributes, and gets
called first.

This is a modified version of the patch applied downstream in RHEL 6.4
within python-ethtool-0.6-3.el6:
  python-ethtool-0.6-add-get_ipv4_addresses-method.patch
ported to take account of 508ffffbb3c48eeeb11eeab2bf971180fe4e1940

1 files changed, 2 insertions, 1 deletions

diff --git a/setup.py b/setup.py
index f081aa6..e1c3411 100644
--- a/setup.py
+++ b/setup.py
@@ -59,7 +59,8 @@ setup(name='ethtool',
                 'python-ethtool/ethtool.c',
                 'python-ethtool/etherinfo.c',
                 'python-ethtool/etherinfo_obj.c',
-                'python-ethtool/etherinfo_ipv6_obj.c'],
+                'python-ethtool/etherinfo_ipv6_obj.c',
+                'python-ethtool/netlink-address.c'],
             include_dirs = libnl['include'],
             library_dirs = libnl['libdirs'],
             libraries = libnl['libs'],