Ipv6 - problem with multicast addresses

[Brecht Vermeulen]

Hi all again,

Still more information:
we've tested now with different cards and it works directly (without
using ifconfig ethX allmulti) for all the other ones !!
3com 3c905  (driver: 3c59x)
dlink538tx (driver 8139too)
dlink570tx (4port card, driver tulip)

so only with the dlink 530tx (driver via-rhine) promiscuous mode doesn't
include ethernet multicast packets. (and I think it is related to the
card and not to the driver, but not sure).

Just bad luck that we used those cards :-).

regards,
Brecht


Brecht Vermeulen wrote:
> 
> Hi all again,
> 
> sorry for replying to my own mail, but I've found the solution, so maybe
> anyone is interested :-).
> This was a tough one :-).
> 
> The key is lying in the use of multicast MAC addresses on the ethernet
> layer. For IPv6 neighbour discovery, hosts send multicasts on ipv6 layer
> which are mapped onto multicast L2 addresses (starting with 33:33:...).
> 
> Now, if you load the ipv6 module, then auto-magically the ethernet card
> listens to some multicast ethernet addresses, which can be found in
> /proc/net/dev_mcast.
> 
> 2    eth0            1     0     3333ffca2b8e
> 2    eth0            1     0     333300000001
> 2    eth0            1     0     01005e000001
> 3    eth1            1     0     3333ff07acd4
> 3    eth1            1     0     333300000001
> 3    eth1            1     0     01005e000001
> 4    eth2            1     0     3333ffc81c97
> 4    eth2            1     0     333300000001
> 4    eth2            1     0     01005e000001
> 
> the 3333 addresses are added by ipv6. So, this means that the card
> receives also these ethernet packets.
> So in the setup:
> host 1 (eth1) ------- (eth1) host 2
> 
> if you run click on host 2 without ipv6 with a configuration:
> echo "FromDevice(eth1,1)->Print(yes)->Discard;"> /proc/click/config
> 
> then it doesn't receive the neighbour sollicitation messages from host 1
> (ipv6 loaded) which are destined for 33:33...
> 
> However if you load ipv6 on host 2, then the packets are received
> because the loading of ipv6 adds the needed multicast groups to the NIC.
> (well, ipv6 uses link local addresses which are based on the MAC
> address, so if you ping6 -I eth1 to the virtual ipv6 link local address
> of host 2, if you ping6 to another address than host 2 isn't listening
> to the right multicast address).
> 
> So, I missly assumed that promiscuous mode would capture all packets.
> The ethernet multicast packets aren't captured. However, with ifconfig
> you can put the interface in 'allmulti' mode :
> [-]allmulti
>               Enable or disable all-multicast mode.  If selected,
>               all  multicast  packets  on  the  network  will  be
>               received by the interface.
> 
> so, a ifconfig eth1 allmulti
> does the job on host2. (maybe with putting the interface down and up,
> but this wasn't needed for me)
> 
> So, now it is possible without loading ipv6 on host 2 and with the click
> config as above to capture all packets from host 1.
> 
> I don't know if this is true for all interface cards (ours are DLINK
> DFE530TX, via-rhine driver, which are cheap ones) or if promiscuous mode
> is more promiscous on other cards (will test this tomorrow).
> 
> This parameter makes it also possible to simply run tcpdump on a host
> which isn't ipv6 enabled while capturing *all* packets.
> 
> regards,
> Brecht
> 
> Brecht Vermeulen wrote:
> >
> > Hi all,
> >
> > I did some further testing on the problem with IPv6 and multicast
> > addresses for sollicitation messages (see mail of Frederik Scholaert).
> >
> > So the problem is further compressed to:
> >
> > host 1 (eth1)  ----    host 2 (eth 2)
> >
> > I have 3 scenarios (with a fresh reboot of both machines in between)
> >
> > 1) if we load on both hosts the ipv6 module, then it is possible to ping
> > from host 1 to the link scope address of host 2. (everything is
> > auto-magically configured, we do not nothing)
> > ping6 fe80::250:baff:feca:2cae -I eth1
> >
> > on host 1, we see
> > tcpdump: listening on eth1
> > 04:42:52.504557 :: > ff02::1:ff07:acd4: icmp6: neighbor sol: who has
> > fe80::250:baff:fe07:acd4
> > 04:42:56.504584 fe80::250:baff:fe07:acd4 > ip6-allrouters: icmp6: router
> > solicitation
> > 04:43:00.504555 fe80::250:baff:fe07:acd4 > ip6-allrouters: icmp6: router
> > solicitation
> > 04:43:01.045766 fe80::250:baff:fe07:acd4 > ff02::1:ffca:2cae: icmp6:
> > neighbor sol: who has fe80::250:baff:feca:2cae
> > 04:43:01.045896 fe80::250:baff:feca:2cae > fe80::250:baff:fe07:acd4:
> > icmp6: neighbor adv: tgt is fe80::250:baff:feca:2cae
> > 04:43:01.045934 fe80::250:baff:fe07:acd4 > fe80::250:baff:feca:2cae:
> > icmp6: echo request
> > 04:43:01.046038 fe80::250:baff:feca:2cae > fe80::250:baff:fe07:acd4:
> > icmp6: echo reply
> > 04:43:02.044621 fe80::250:baff:fe07:acd4 > fe80::250:baff:feca:2cae:
> > icmp6: echo request
> > 04:43:02.044727 fe80::250:baff:feca:2cae > fe80::250:baff:fe07:acd4:
> > icmp6: echo reply
> > 04:43:03.044605 fe80::250:baff:fe07:acd4 > fe80::250:baff:feca:2cae:
> > icmp6: echo request
> >
> > So, this works.
> >
> > 2) on host 1, we load the ipv6 module, on host 2 we load click with the
> > following config:
> > echo "FromDevice(eth1,1)->Print(yes)->Discard;"> /proc/click/config
> >
> > and we do the same ping from host 1 to host 2 : nothing is printed by
> > click (a tcpdump on host 1 shows
> > tcpdump: listening on eth1
> > 04:51:06.010989 fe80::250:baff:fe07:acd4 > ff02::1:ffca:2cae: icmp6:
> > neighbor sol: who has fe80::250:baff:feca:2cae
> > 04:51:07.004349 fe80::250:baff:fe07:acd4 > ff02::1:ffca:2cae: icmp6:
> > neighbor sol: who has fe80::250:baff:feca:2cae
> > 04:51:08.004341 fe80::250:baff:fe07:acd4 > ff02::1:ffca:2cae: icmp6:
> > neighbor sol: who has fe80::250:baff:feca:2cae
> >
> > If we then however load ipv6 on host2, then the packets are printed in
> > the syslog (without stopping click).
> > We use the kernel level click module of Jan. 11th, 2002 with linux
> > 2.2.19. The ethernet cards are DLINK DFE 530TX with via-rhine
> > (non-polling) driver. I haven't tried with other cards yet.
> >
> > 3) if we start on host 2 a tcpdump, load the ipv6 modules on both hosts
> > and do the ping from host1 to host2, then it is the same as in scenario
> > 2, no packets at host2, but tcpdump on host 1 sees the packets. If we
> > stop tcpdump on host 2 and start tcpdump again, then we see the packets.
> > (tried with libpcap 0.4a6/tcpdump 3.7.1 and libpcap 0.6.2/tcpdump 3.6.2,
> > hope these are the correct version numbers).
> >
> > If we look at the ethernet headers, we see
> > tcpdump: listening on eth1
> > 04:51:20.932240 0:50:ba:7:ac:d4 33:33:ff:ca:2c:ae 86dd 86:
> > fe80::250:baff:fe07:acd4 > ff02::1:ffca:2cae: icmp6: neighbor sol: who
> > has fe80::250:baff:feca:2cae
> >
> > so, the destination MAC address is 33:33:... which is an ethernet
> > multicast.
> >
> > So, scenario 2 seems very strange to me as I thought that click just
> > took over the networking from Linux, but how comes then that it doesn't
> > see any packet and that the loading of the ipv6 module makes that this
> > works.
> > Or could there be something broken with the multicast ethernet addresses
> > and the card (although click is in promiscuous mode) ?
> >
> > Our system is debian 2.2 (potato), kernel 2.2.19 with Click patch. (for
> > all scenarios, Click is however not loaded in the 1st and 3rd scenario).
> >
> > anyone any good idea (next possibilities I see is testing with other
> > cards, 2.4.x kernels) ?
> >
> > regards,
> > Brecht
> >
> > Hi,
> >
> > I'm experimenting with IPv6 in click. It works pretty well except for
> > the
> > fact that i'm not able to receive all-nodes or all-router multicast
> > addresses in my click config.
> >
> > My configuration is like this:
> >
> > |HOST|-ETH1-fec0:0:0:1::65/64---ETH1-|CLICK
> > ROUTER|-ETH2---fec0:0:0:2::66/64-ETH1-|HOST|
> >                           [fec0:0:0:1::64/64]  [fec0:0:0:2::64/64]
> >
> > The IPv6-addresses fec0:0:0:1::64 and fec0:0:0:2::64 are not really
> > assigned because IPv6 is not loaded on the click router.
> >
> > When for instance the IPv6 Click config is loaded and i load IPv6 on the
> > left host, the all-routers sollicitation packets don't seem to arrive at
> > the click router. Or, when i do a ping fec0:0:0:2::66 on the left host,
> > i
> > see 'fec0:0:0:1::65 > ff02::1:ff00:64: icmp6: neighbor sol: who has
> > fec0:0:0:1::64' -packets (leaving?) on eth1 of the left host, but they
> > also do not seem to arrive at the click router. (I have a Print-element
> > there after a FromDevice(eth1,1)-element and see nothing).
> >
> > But when i first load the IPv6-module on the click-router and actually
> > assign the fec0:0:0:1::64 and fec0:0:0:2::64 addresses, and then
> > afterwards load click, i can ping through the click router.
> >
> > I don't think it should have to be this way. Shouldn't i be able to run
> > a
> > IPv6 click config completely indepedent of IPv6 in the Linux kernel?
> >
> > I use a click-version from january (from cvs), and kernel 2.2.19.
> >
> > Can somebody help me on this matter?
> >
> > regards,
> > Frederik