[Dnsmasq-discuss] Gotcha!

[Gene Czarcinski]

On 12/01/2012 04:02 PM, Gene Czarcinski wrote:
> On 12/01/2012 03:36 PM, Gene Czarcinski wrote:
>> Well, I seem to be able to create the problem at will!
>>
>> I created a small patch (attached) to instrument things.  The result 
>> was that BOTH autostarted networks got screwed up! The syslog is also 
>> attached.  Here is the output of ip addr which show not one but TWO 
>> bad addresses on it [10:6 and 10:8]:
>>
>> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue state UNKNOWN
>>     link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
>>     inet 127.0.0.1/8 scope host lo
>>     inet6 ::1/128 scope host
>>        valid_lft forever preferred_lft forever
>> 2: p33p1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast 
>> state UP qlen 1000
>>     link/ether 30:85:a9:8f:e9:82 brd ff:ff:ff:ff:ff:ff
>>     inet 192.168.17.2/24 brd 192.168.17.255 scope global p33p1
>>     inet6 fd00:beef:10:6:3285:a9ff:fe8f:e982/64 scope global dynamic
>>        valid_lft 7135sec preferred_lft 7135sec
>>     inet6 fd00:beef:10:8:3285:a9ff:fe8f:e982/64 scope global dynamic
>>        valid_lft 7172sec preferred_lft 7172sec
>>     inet6 fd00:dead:beef:17:1::2/128 scope global
>>        valid_lft forever preferred_lft forever
>>     inet6 fe80::3285:a9ff:fe8f:e982/64 scope link
>>        valid_lft forever preferred_lft forever
>>
>> Here is what I believe is the solution (it is pretty much in the 
>> patch).  In periodic_ra() [radv.c], a test needs to be added so that 
>> if the send_ra() is to go to an interface which is NOT ours, don't do 
>> it ... "ours" being defined by any interface=<ip6addr>
>>
>> My next step is to give this a try.
>>
>> I do not know if this is the absolutely right fix because I am not 
>> sure I understand how p33p1 got queued up for this dnsmasq instance.
>>
>> Any insight appreciated.
>>
> Getting close but not a winner yet.  I am not bothering to add any 
> more syslog, etc. as it is essentially the same.
>
> This time, the 10:6 network was OK but the 10:8 network was not. I 
> need to go back up the chain of things to find out what happens in the 
> first place.
>
> Oh yes, I detected the bad interface name but just not sending the RA 
> is not good enough.  Using ip addr shows the SLAAC 
> fd00:beef:10:8::3285:a9ff:fe8f:e982/64 address on the interface.
>
> I believe that checking the interface names against what is specified 
> is the correct thing to do, it just needs to be in exactly the "right" 
> place in the code.
>
> I do not want to touch the libvirt code so that I have something 
> consistent to test with (and rebuilding the dnsmasq rpm takes all of 
> 10 seconds).  But, I may add something which will force 
> bind-interfaces when bind-dynamic is specified just to see if that 
> changes anything.
Not conclusive but promising.  I will need to modify libvirt slightly so 
that it uses bind-interfaces, interface=<if_name>, except-interface=lo

Gene