avoid-v4-udp-ports ineffective? (BIND 9.8.1-P1)

[David Coulthart]

On Nov 17, 2011, at 6:28 PM, Mark Andrews wrote:
> In message <201111171600.pAHG0uCW011627 at scramble.Princeton.EDU>, Irwin Tillman writes:
>> It appears that named is trying to use ports I've mentioned in avoid-v4-udp-ports.
>> 
>> Platform: BIND 9.8.1-P1 on Solaris 10 / SPARC
>> 
>> On some of the ports which BIND might otherwise choose to use, 
>> I have other daemons running and/or the OS treats the ports
>> as privileged.  To keep named from trying to use those ports, I have
>> in named.conf:
>> 
>> options {
>>    ...
>>    # there is no use-v4-udp-ports statement.
>>    avoid-v4-udp-ports { 1812; 1813; 2049; 4045; };
>>    # I don't speak v6.
>> };
>> 
>> When I upgraded from BIND 9.7.3-P3 to 9.8.1-P1, I began seeing in the log:
>> 
>> named[9185]: dispatch: warning: dispatch 42d950: open_socket(::#2049) -> permission denied: continuing
>> named[9185]: dispatch: warning: dispatch 42d950: open_socket(::#4045) -> permission denied: continuing
>> 
>> ...which suggests to me that BIND is trying to use ports I specified in avoid-v4-udp-ports.
>> 
>> 
>> Checking get_dispsocket() in ./lib/dns/dispatch.c, I see that a difference
>> between 9.7.3-P3 and 9.8.1-P1 is that 9.8.1-P1 logs a warning when an attempt
>> to open the socket returns ISC_R_NOPERM (perhaps the result of bind() returning EACCESS ?),
>> while 9.7.3-P3 didn't log the warning.  The warning is new.
>> When confronted with the error, both versions proceed to pick another port to try again. 
>> So I don't know if the older version was also trying to use these ports and encountering
>> the same error.
>> 
>> I imagine Solaris might return EACCESS because:
>> 
>> % ndd /dev/udp udp_extra_priv_ports
>> 2049 
>> 4045 
>> 
>> 
>> I don't understand why named would try to use these ports in the first
>> place as they appear in avoid-v4-udp-ports.
> 
> 	The "::" in the log message is the IPv6 equivalent of 0.0.0.0 in IPv4.
> 	You machine *is* dual stacked even if it only has IPv6 on loopback.

I noticed this same log message on our Solaris 9 servers and I don't see IPv6 enabled on any of our interfaces, including loopback:

$ /sbin/ifconfig -a inet6
$ /sbin/ifconfig -a inet 
lo0: flags=1000849<UP,LOOPBACK,RUNNING,MULTICAST,IPv4> mtu 8232 index 1
	inet 127.0.0.1 netmask ff000000 
bge0: flags=1000843<UP,BROADCAST,RUNNING,MULTICAST,IPv4> mtu 1500 index 2
	inet 128.59.59.92 netmask ffffff00 broadcast 128.59.59.255
bge0:10: flags=1000843<UP,BROADCAST,RUNNING,MULTICAST,IPv4> mtu 1500 index 2
	inet 128.59.59.70 netmask ffffff00 broadcast 128.59.59.255

It's not a big deal for me as it doesn't happen very often (& can be easily avoided by setting a matching avoid-v6-udp-ports), but it did surprise me.

Thanks Irwin for the tip on ndd.  I had to avoid those exact same ports & I couldn't find a process bound to port 2049 causing it.

-- 
Dave Coulthart