recognize incoming multicast packages about aiocoap

[chrysn]

if someone came up with an idea on how to fix that without monkey-patching asyncio, that would be much appreciated; i don't see an elegant solution right now.

[HRogge]

Has there been any work on this topic since this issue-report was last touched?

Receiving multicast traffic should not be difficult with the lowlevel socket API... you need one socket to bound to the unicast address (to receive unicast and to send unicast/multicast) and one bound to the multicast IP (to receive multicast).

[HRogge]

Here is a code snippet I used for experimenting with coapthon (3) until I discovered it does not support messages without ACKs:

The main issue with this approach is that you cannot easily bind to "0.0.0.0" (because you would get the multicast there too), but you have to register pair of sockets per interface.

serverinfo = socket.getaddrinfo(server_address)[0]

self._ucsocket = socket.socket(serverinfo[0], serverinfo[1], serverinfo[2])
self._ucsocket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self._ucsocket.bind(serverinfo[-1])

if serverinfo[0] == socket.AF_INET:  # IPv4
    multicast_address = (defines.ALL_COAP_NODES) + server_address[1:]
elif serverinfo[-1][-1] != 0: # Linklocal IPv6
    multicast_address = (defines.ALL_COAP_NODES_LINKLOCAL_IPV6) + server_address[1:]
else: # Site-local IPv6
    multicast_address = (defines.ALL_COAP_NODES_SITELOCAL_IPV6) + server_address[1:]
mcinfo = socket.getaddrinfo(multicast_address)[0]

self._mcsocket = socket.socket(mcinfo[0], mcinfo[1], mcinfo[2])
self._mcsocket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self._mcsocket.bind(mcinfo[-1])

# Join group
mreq = socket.inet_pton(mcinfo[0], mcinfo[-1][0]) + socket.inet_pton(serverinfo[0], serverinfo[-1][0])
if serverinfo[0] == socket.AF_INET:  # IPv4
    self._mcsocket.setsockopt(socket.IPPROTO_IP, socket.IP_ADD_MEMBERSHIP, mreq)
else:
    self._mcsocket.setsockopt(socket.IPPROTO_IPV6, socket.IPV6_JOIN_GROUP, mreq)

[chrysn]

Much of this has definitely been solved, I'm going through the code to ensure no points of this are left (especially since aiocoap has been using recvmsg for quite some time now; this alleviates the need to bind to a socket per IP address, which in dynamic situations would mean I'd need (operating system specific) means of seeing them appear and disappear).

[HRogge]

Yes, works for me (after the no_response thing in #170 is fixed).

[chrysn]

The server's behavior in multicast cases is now enhanced (the server still answered RST to an exotic class of cases which it shouldn't have, and showed harmless warnings) and tested for since 5055bd5.

Thanks @HRogge for bringing my attention to this long-standing issue.