Hi David:
Le Vendredi 8 Novembre 2002 01:12, David Stevens a écrit :
> Don,
> Multicast group membership is per-interface, not per-port. If an
> application binds to INADDR_ANY:9462 (UDP), it should receive any multicast
> UDP packets sent to port 9462, whether or not that application did the join
> for that multicast group.
You are correct, but are talking about a lower level. The problem with the
current implementation, which differs from the IPv6 one btw, is that an
application that binds INADDR_ANY:9462, as in your example, and subscribes to
a multicast group, will still receive all the groups that were subscribed to
by all the applications on the box.
For example, if applicaiton binds a socket S1 to INADDR_ANY:9462 and
subscribes to 239.255.0.1 and it is the only application on the box that has
subscribed to a multicast group on port 9462, it will only receive traffic on
addressed to 239.255.0.1:9462. On the other hand, if any application
(including this one) binds another socket S2 to INADDR_ANY:9462 and
subscribes to 239.255.0.2, both sockets will receive the traffic destined for
both groups.
Similarly, if an application joins a group, but
> never binds to a UDP port, the host should still answer, for example, ICMP
> echo requests sent to that multicast address.
But if it doesn't bind a port, it won't receive anything. But that's not the
issue. The issue is that the IPv4 udp layer does not perform perfect
filtering of multicast groups for sockets that have bound a port and
subscribed to one or more groups. The socket should only receive traffice
for groups it has joined.
Please see Stevens' discussion of filtering in UNPv1 page 492 for a much
better explaination than I can provide.
I have a test program I added to the ACE distribution that tests this and has
been run on several different platforms, e.g., Solaris, OpenBDS, Windows,
etc, and only Linux has a problem with receiving traffic on a socket for a
group that was not joined. The current work around is to bind the multicast
group and only subscribe to a single group per socket--but that wastes a lot
of sockets...
take care...
don
>
> +-DLS
>
>
> Don Hinton <dhinton@ieee.org>@vger.kernel.org on 11/07/2002 03:55:05 PM
>
> Please respond to dhinton@ieee.org
>
> Sent by: linux-net-owner@vger.kernel.org
>
>
> To: linux-net@vger.kernel.org
> cc:
> Subject: [PATCH] (IPv4) multicast group filtering doesn't work if socket
> only bound to a port
>
>
>
> Hi:
>
> The IPv4 stack, unlike IPv6, doesn't filter mcast groups for sockets only
> bound to a port, but not an address, based on subscriptions. So even
> if you only subscribe to a single group, but don't bind an address to the
> socket, you'll receive all mcast dgrams delivered to the udp layer, i.e.,
> all subscribed to groups no matter which application actually made the
> subscription.
>
> The following patch fixes the problem in a manner similar to the way the
> IPv6 stack does it. It's against 2.4.19:
>
> --- linux-2.4.19/net/ipv4/udp.c.old Wed Nov 6 19:16:03 2002
> +++ linux-2.4.19/net/ipv4/udp.c Thu Nov 7 20:43:00 2002
> @@ -82,6 +82,7 @@
> #include <linux/mm.h>
> #include <linux/config.h>
> #include <linux/inet.h>
> +#include <linux/igmp.h>
> #include <linux/netdevice.h>
> #include <net/snmp.h>
> #include <net/ip.h>
> @@ -267,6 +268,7 @@
> int dif)
> {
> struct sock *s = sk;
> + struct ip_mc_socklist *mc;
> unsigned short hnum = ntohs(loc_port);
> for(; s; s = s->next) {
> if ((s->num != hnum) ||
> @@ -275,6 +277,14 @@
> (s->rcv_saddr && s->rcv_saddr != loc_addr) ||
> (s->bound_dev_if && s->bound_dev_if != dif))
> continue;
> + else {
> + for (mc=s->protinfo.af_inet.mc_list; mc;
> mc=mc->next) {
> + if (mc->multi.imr_multiaddr.s_addr ==
> loc_addr)
> + break;
> + }
> + if (!mc)
> + continue;
> + }
> break;
> }
> return s;
>
>
> take care...
> don
--
Don Hinton <dhinton @ ieee.org>
Software Engineer
P.O. Box 23524, Alexandria, Virginia 22304
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
