Re: [PATCH 4.9 27/75] net: igmp: Use correct source address on IGMPv3 reports

[Kevin Cernekee]

On Mon, Jan 15, 2018 at 8:44 PM, Sebastian Gottschall
<s.gottschall@xxxxxxxxxx> wrote:
> Am 16.01.2018 um 05:32 schrieb Kevin Cernekee:
>>
>> On Mon, Jan 15, 2018 at 8:26 PM, Sebastian Gottschall
>> <s.gottschall@xxxxxxxxxx> wrote:
>>>
>>> havent check the source addresses right now. i basicly discovered that
>>> this
>>> patch breaks the igmp routing and all traffic stops
>>> this here is from a working system with the reverted patch. if you really
>>> need that i break it again using the patch you need to wait a little bit
>>>
>>> 05:14:22.697962 IP 10.88.195.138 > 239.35.100.8: igmp v2 report
>>> 239.35.100.8
>>
>> The patch should only affect IGMPv3 behavior.  I did not intend to
>> change IGMPv2 behavior.  If it does, that might be a bug.
>
> it does change the behaviour indeed. i dont know the reason. but i while
> discovering the issue on 4.14 last week and newly on 4.9 this week while
> testing
> (my latest firmware i builded was from 30. december and worked) i got
> tracked it down to this small patch and it immediatly worked after reverting
> it
>>
>> Is it possible that the kernel is using a source IP of 0.0.0.0, but
>> another host does not recognize it because it does not comply with RFC
>> 3376?
>
> this is possible yes, but i cannot look into the "deutsche telekom" host
>>
>>
>> Before/after packet traces would be the best way to see if the kernel
>> change is causing it to violate the standard.
>
> let me just take a look into our patch
> + for_ifa(in_dev) {
> + if (inet_ifa_match(fl4->saddr, ifa))
> + return fl4->saddr;
> + } endfor_ifa(in_dev);
> this looks like you're checking if the source address matches to a local
> interface, if not you return 0.0.0.0 instead of the source address
>
> (193.158.35.251, 239.35.20.4)    Iif: ppp0       Oifs: briptv
>
> our first source address here 193.158.35.251 is from a remote network. so
> your patch also will change the behaviour since the source address will get
> ignored

According to my understanding of igmpv3_newpack(), the destination
address should always be IGMPV3_ALL_MCR = 224.0.0.22.  That is what I
see in my testing.

However, your packet trace says 239.35.100.8.  I don't know how the
code that we touched would be generating an IGMPv2 packet with that
destination address.

Would it be possible to get a stack trace for the case where the
source address is being cleared to 0.0.0.0 in your configuration?