[Dnsmasq-discuss] New smart --bind-dynamic is greedy (binds also to interface aliases)

[Simon Kelley]

On 13/05/13 01:00, Andrew Bartlett wrote:
> I've looked over the source code multiple times, and I can't see how it
> happens, but I've just filed
> https://bugzilla.redhat.com/show_bug.cgi?id=962246 with Fedora, and
> figured I would also work here to see how this can be fixed.
>
> I do agree that interface detection is some of the most crazy,
> OS-specific code ever.  Oddly Samba has much the same challenge, but
> seems to use a different set of APIs.
>
> In any case what happens is this:
>
> dnsmasq is being run by libvirt like this:
> /sbin/dnsmasq --strict-order --local=// --domain-needed
> --pid-file=/var/run/libvirt/network/default.pid --conf-file=
> --except-interface lo --bind-dynamic --interface virbr0 --dhcp-range
> 192.168.122.2,192.168.122.254
> --dhcp-leasefile=/var/lib/libvirt/dnsmasq/default.leases
> --dhcp-lease-max=253 --dhcp-no-override
>
> I then run:
> ifconfig virbr0:0 192.168.122.2
>
> And then I find dnsmasq has also chosen to bind to 192.168.122.2!
>
> eg this in netstat
>
> tcp        0      0 192.168.122.2:53        0.0.0.0:*
> LISTEN      1039/dnsmasq
> tcp        0      0 192.168.122.1:53        0.0.0.0:*               LISTEN      1039/dnsmasq
> udp        0      0 192.168.122.2:53        0.0.0.0:*                           1039/dnsmasq
> udp        0      0 192.168.122.1:53        0.0.0.0:*                           1039/dnsmasq
> udp        0      0 0.0.0.0:67              0.0.0.0:*                           1039/dnsmasq
> unix  2      [ ]         DGRAM                    21571    1039/dnsmasq
>
> I'm presuming somewhere we are comparing on a name without the alias (:0) bit, or doing a length-limited comparison, but I've looked and just can't find it!
>
> Andrew Bartlett
>

My guess about what's happening, (and I've not looked thoroughly, at 
least yet) is this.

Linux long ago moved past the idea of interface aliases and into 
complete support for multiple addresses of the same address family per 
interface.

There's backwards compatible support for aliases, so that

ifconfig virbr0:0 192.168.122.2

has been redefined to mean something like "add another IP address to 
eth0, and give it the backwards-compatible name eth0:0"

If you use the more modern "ip" command to look at things, this becomes 
obvious: after

ifconfig eth0:0 192.168.99.1

if I do

ip addr show

I get

2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP 
qlen 1000
     link/ether 00:11:25:a1:91:c5 brd ff:ff:ff:ff:ff:ff
     inet 192.168.0.193/24 brd 192.168.0.255 scope global eth0
     inet 192.168.99.1/24 brd 192.168.99.255 scope global eth0:0
........

So, there's just another address in eth0, but it has the wierd eth0:0 on 
the end.

Dnsmasq is using netlink to enumerate all the addresses associated with 
interfaces, passes the interface index to ioctl(..., SIOCGIFNAME, ...) 
to find the name of the interface associated with each address. I think 
that both addresses probably have the same index and therefore that 
always returns eth0 and not eth0:0 in my example. That would explain 
what you're seeing exactly.

There are various ways we could proceed:

1) I could tell you to stop using old-fashioned aliases, and specify 
which addresses you want dnsmasq to listen in directly using 
--listen-address

2) I could find out the API to get the "alias name" associated with an 
address and change the behaviour of dnsmasq.

3) We could check my wild guesses above, and find that something else 
entirely is happening.


I'm currently highly agnostic about which will come to pass.


Cheers,

Simon.