IPv6 multicast in 2.4.25 broken?

IPv6 multicast in 2.4.25 broken?

[Ignacy Gawedzki]

Hi,

It seems that simple multicast support in IPv6 has ceased working
starting at kernel version 2.4.25-pre4.  More specifically, receiving
multicast packets (those with destination address ff05::1 or ff02::1) is
broken (read: sending is okay).  The same binary run on kernels 2.4.24
and earlier works with no problem.

I tried as many combinations of USAGI patches and kernel versions as I
could, but cannot figure the exact source of the problem.  I did not
inspect the diffs with enough attention, though.

Another funny thing is that the changelog for version 2.4.25-pre4
doesn't mention any change to IPv6 code.

Did anyone have similar experience and or suggestions?

I have not subscribed to the linux-kernel list, because of its high
traffic, so please CC me the replies.

Thanks,

Ignacy

PS: The detailed way in which I prepare the socket is the following:

  Open a UDP socket in IPv6 protocol family.

  Bind it to network interface using SO_BINDTODEVICE.

  Set the multicast network interface using IPV6_MULTICAST_IF.
  
  Bind it to the network interface's address (which is set manually to
  fec0::1).
  
  Set the multicast hops with IPV6_MULTICAST_HOPS to 1.
  
  Set the multicast membership to ff02::1 (or ff05::1) using
  IPV6_ADD_MEMBERSHIP.
  
  Unset the multicast loop using IPV6_MULTICAST_LOOP.

Note: the same socket is used to send and receive multicast packets.

-- 
To err is human, to moo bovine.

Re: (usagi-users 02870) Re: IPv6 multicast in 2.4.25 broken?

[Ignacy Gawedzki]

On Tue, Mar 30, 2004 at 01:47:06PM -0800, thus spake David Stevens:
>    Can you reproduce this in a small test program and send me
>    the source? Also, the contents of /dev/proc/igmp6 and any
>    arguments you ran the program with would be helpful.

Not at this time, as I would have to reinstall a 2.4.25 kernel.  But I
plan to do the testing soon.

>    At first I thought it might be that you have an MLD-snooping
>    switch that doesn't understand MLDv2 packets, but ff02::1 is
>    the all-nodes address which all hosts join, and which is not
>    advertised at all. That should work if you're receiving any
>    multicasts at all.

Well, I see those packets with tcpdump on both the sending and receiving
interfaces, so no ethernet equipment is involved.  Moreover, the same
code works perfectly on 2.4.24, which should indicate that there are
some changes between the two (more precisely between 2.4.24 and
2.4.25-pre4) that break something.  Somehow the multicast packets are
received but not relayed to the socket...

-- 
 "The whole problem with the world is that fools and fanatics are
   always so certain of themselves, and wiser people so full of doubts."
                                                 - Bertrand Russell

Re: (usagi-users 02872) Re: IPv6 multicast in 2.4.25 broken?

[Ignacy Gawedzki]

On Wed, Mar 31, 2004 at 10:01:29PM +0900, thus spake Takashi Hibi:
> Isn't it related to the patch (usagi-users 02627)?
> http://www2.linux-ipv6.org/ml/usagi-users/msg02626.html
> 
> If the socket is bound to non-multicast address, it can't receive
> multicast packets.

This is why it doesn't work anymore, as it shouldn't have in the first
place.  Thanks for pointing this out.  I did not know that source
address selection is automatic, depending on the multicast address
prefix (ff05::/16, ff02::/16, etc) and the IPV6_MULTICAST_IF option, so
I instinctively did bind the socket to the unicast address and not the
multicast address.

Now it seems to work like a charm. =)

Thanks,

Ignacy

-- 
To err is human, to moo bovine.