Re: HEADS UP: Source-specific multicast code committed to tree


HEADS UP: Source-specific multicast code commit...
\ Bruce M Simpson (12 Jun 2007)
. \ Julian Elischer (12 Jun 2007)
. . \ Bruce M Simpson (12 Jun 2007)

2 msg	Build failure on amd64
1 msg	HEADS UP: nfe(4), Default NVIDIA nForce MCP Eth...
33 msg	[head tinderbox] failure on i386/i386
6 msg	nfe(4) strangeness on the 7-CURRENT
3 msg	IBM serveraid ipsstat physical drive state values
3 msg	'rtfree: %p has 1 refs'
1 msg	make release on current ends with 'No space lef...
7 msg	new wpa_supplcant and hostapd coming
5 msg	HEADSUP: new 802.11 work committed
4 msg	sun4v Tinderbox !
3 msg	yacc/gdb issues with 7.0-current
3 msg	zfs problem
1 msg	Testing changeset for i386 on amd64?
4 msg	AMD64 panic on boot
4 msg	Don't know what I was doing but...
3 msg	Archives of this list
2 msg	panic: mutex process slock not owned ...
4 msg	snd_es137x & -CURRENT, /dev/dsp: Invalid ar...
20 msg	GPT - (last) call for action

Subject:	Re: HEADS UP: Source-specific multicast code committed to tree
Group:	Freebsd-current
From:	Bruce M Simpson
Date:	12 Jun 2007

Julian Elischer wrote:
>> * Strict socket membership is now the default; the legacy 4.4BSD
>> behaviour of delivering a multicast datagram to all sockets matching
>> the 4-tuple has been removed.
> this seems a but counterintuitive
This particular change only affects UDP.

Strictly speaking, multicast datagrams should only be delivered locally
to sockets which are members of the multicast group. Source multicast
makes this definition stricter - multicast datagrams should only be
delivered if the source is permitted to deliver to the destination.

In the absence of IGMPv3/MLDv2 support from upstream routers, it is
possible for datagrams from blocked sources to reach the end host. For
the time being FreeBSD merely enforces these semantics by dropping such
datagrams. When IGMPv3/MLDv2 support is added, the source filter list is
relayed to upstream multicast routers whenever it changes, and they
SHOULD NOT deliver datagrams blocked by source filters to the end host.

Other implementations have done this for many years and it's time
FreeBSD followed suit. A sysctl to enable the strict behaviour has been
in place for a few years now. It is preferred by design in a source
multicast environment.

regards,
BMS
_______________________________________________
freebsd-current mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe"