The buffer in receive_netlink_message() is shared between all threads,
which poses a problem when several CM instances start accessing it at
the same time.

That can create issues e.g. with libnice where each NiceAgent instance
runs its own GUPnPContextManager in a separate thread.

Let each GUPnPLinuxContextManager have its own recvbuf.

Apparently the scope id is only global, link or host, but never site
even with ULAs

If the wireless card's name is exactly IFNAMSIZ bytes long (without the
terminating NUL), then iwr.ifr_name will lack the terminating NUL
character. Only copy IFNAMSIZ - 1 bytes so we can be sure that the last
character will terminate the string properly.

Use parentheses to make logic operator clear from bitwise

Signed-off-by: Jens Georg <mail@jensge.org>

Looks like it was a recent addition.
Signed-off-by: Jens Georg <mail@jensge.org>

https://bugzilla.gnome.org/show_bug.cgi?id=760217

Signed-off-by: Jens Georg <mail@jensge.org>

Use if we don't have ESSID use the network as a name, as we don't have
GSSDPClient not looking it up anymore.
Signed-off-by: Jens Georg <mail@jensge.org>

 - Some DHCP clients assign a link-local address and then switch back to the
   new address, causing loss of context
 - Some DHCP renews cause RTM_NEWADDR with the current address, causing
   needless loss of context
Signed-off-by: Jens Georg <mail@jensge.org>

https://bugzilla.gnome.org/show_bug.cgi?id=736365

Add the possibility to dump the received netlink packets.
Signed-off-by: Jens Georg <mail@jensge.org>

https://bugzilla.gnome.org/show_bug.cgi?id=736365

We get it "for free" with the ILF_* attributes.
Signed-off-by: Jens Georg <mail@jensge.org>

https://bugzilla.gnome.org/show_bug.cgi?id=736365

RTM_*ADDRESS rt_attr was wrongly parsed using IFLA_* macros instead of
IFA_* macros, probably causing some rt_attrs to be missed.

https://bugzilla.gnome.org/show_bug.cgi?id=728889

Sometimes it might happen that netlink sockets aren't available on
runtime (QEmu for example).

Previously the Linux CM could not cope with them since GSSDP lacks the
understanding that a device can have more than one address.

This works around the fact by treating the aliases (eth0:1) as different
devices.

Fixes:
    https://bugzilla.gnome.org/show_bug.cgi?id=682005

These where not leaked before because their initial
declaration declares them as static, but make the
actual function declaration static too just because
it's nice to keep declarations in sync.

https://bugzilla.gnome.org/show_bug.cgi?id=673933

When building gupnp with the linux context manager
the library gets another public symbol.
It should be relatively safe to simply make the
function declaration static to hide it.

https://bugzilla.gnome.org/show_bug.cgi?id=673933

Might happen e.g. on usbnet devices when cable is unplugged.

https://bugzilla.gnome.org/show_bug.cgi?id=655970

Use g_main_context_get_thread_default()

https://bugzilla.gnome.org/show_bug.cgi?id=654990

This is based on RT Netlink sockets, see rtnetlink(7)