Saturday, May 31, 2014

Loading IP Test Data Into Postgres

Recently, I was trolling around the internet looking for some IP address data to play with.  Fortunately, I stumbled across MaxMind's Geolite Database, which is available for free.    All I have to do is include this notice:

This product includes GeoLite2 data created by MaxMind, available from <a href="http://www.maxmind.com">http://www.maxmind.com</a>.

That's easy enough.  The next question is how to get this into my local Postgres database.  A bit over a year ago, I very happily gave up on my Dell computer and opted for a Mac.  One downside to a Mac is that SQL Server doesn't run on it (obviously my personal opinion).  Happily, Postgres does and it is extremely easy to install by going to postgresapp.com.   An interface similar enough to SQL Server Management Studio (called pgadmin3) is almost as easy to install by going here.

So, the next problem is getting the MaxMind data into Postgres.  Getting the two tables into Postgres is easy, using the copy command.  The challenge is IPV6 versus IPV4 addresses.  The data is in IPV6 format with a subnet mask to represent ranges.  Most of us who are familiar with IP addresses are familiar with IPV4 addresses.  These are 32 bits and look something like this:  173.194.121.17 (this happens to be an address for www.google.com attained by running ping www.google.com in a terminal window).  Alas, the data from MaxMind uses IPV6 values rather than IPV4.

In IPV6, the above would look like: ::ffff:173.194.121.17 (to be honest, this is a hybrid format for representing IPV4 addresses in IPV6 address space).  And the situation is a bit worse, because these records contain address ranges.  So the address range is really:  ::ffff:173.194.0.0/112.

The "112" is called a subnet mask.  And, IPV4 also uses them.  In IPV4, they represent the initial range of digits, so they range from 1-32, with a number like "24" being very common.  "112" doesn't make sense in a 32-bit addressing scheme.   To fast forward, the "112" subnet mask for IPV6 corresponds to 16 in the IPV4 world.  This means that the first 16 bits are for the main network and the last 16 bits are for the subnet.  That is, the addresses range from 173.194.0.0 to 173.194.255.255.  The relationship between the subnet mask for IPV6 and IPV4 is easy to express:  the IPV4 subnet mask is the IPV6 subnet mask minus 96.

I have to credit this blog for helping me understand this, even though it doesn't give the exact formula.  Here, I am going to shamelessly reproduce a figure from that blog (along with its original attribution):
ipv6-address
Courtesy of ls-a.org

This figure says  the following.  The last 64 bits are new to IPV6, so they can be automatically subtracted out of the subnet mask.  Then, bits 0-32 also seem to be new, so they can also be subtracted out.  That totals 96 bits in the new version not in the old version.  To be honest, I am not 100% positive about my interpretation.  But it does seem to work.  Google does indeed own exactly this address range.

The Postgres code for creating the table then goes as follows:

create table ipcity_staging (
    network_start_ip varchar(255),
    network_mask_length int,
    geoname_id int,
    registered_country_geoname_id int,
    represented_country_geoname_id int,
    postal_code varchar(255),
    latitude decimal(15, 10),
    longitude decimal(15, 10),
    is_anonymous_proxy int,
    is_satellite_provider int
);

copy public.ipcity_staging
    from '...data/MaxMind IP/GeoLite2-City-CSV_20140401/GeoLite2-City-Blocks.csv'
    with CSV HEADER;

create table ipcity (
    IPCityId serial not null,
    IPStart int not null,
    IPEnd int not null,
    IPStartStr varchar(255) not null,
    IPEndStr varchar(255) not null,
    GeoNameId int,
    GeoNameId_RegisteredCountry int,
    GeoNameId_RepresentedCountry int,
    PostalCode varchar(255),
    Latitude decimal(15, 10),
    Longitude decimal(15, 10),
    IsAnonymousProxy int,
    IsSatelliteProvider int,
    unique (IPStart, IPEnd),
    unique (IPStartStr, IPEndStr)
);

insert into ipcity(IPStart, IPEnd, IPStartStr, IPEndStr, GeoNameId, GeoNameId_RegisteredCountry, GeoNameId_RepresentedCountry,
                   PostalCode, Latitude, Longitude, IsAnonymousProxy, IsSatelliteProvider
                  ) 
    select IPStart, IPEnd, IPStartStr, IPEndStr, GeoName_Id, registered_country_geoname_id, represented_country_geoname_id,
           Postal_Code, Latitude, Longitude, Is_Anonymous_Proxy, Is_Satellite_Provider
    from (select network_mask_length - 96,
                 hostmask(inet (substr(network_start_ip, 8) || '/' || network_mask_length - 96)) ,
                 inet(host(inet (substr(network_start_ip, 8) || '/' || network_mask_length - 96) )) |
                 hostmask(inet (substr(network_start_ip, 8) || '/' || network_mask_length - 96)
                ) as ipend_inet,
                substr(network_start_ip, 8) || '/' || network_mask_length - 96,
                ((split_part(IPStartStr, '.', 1)::int << 24) +
                 (split_part(IPStartStr, '.', 2)::int << 16) +
                 (split_part(IPStartStr, '.', 3)::int << 8) +
                 (split_part(IPStartStr, '.', 4)::int)
                ) as IPStart,
                ((split_part(IPEndStr, '.', 1)::int << 24) +
                 (split_part(IPEndStr, '.', 2)::int << 16) +
                 (split_part(IPEndStr, '.', 3)::int << 8) +
                 (split_part(IPEndStr, '.', 4)::int)
                ) as IPEnd,
                st.*
          from (select st.*,
                       host(inet (substr(network_start_ip, 8) || '/' || network_mask_length - 96)) as IPStartStr,
                       host(inet(host(inet (substr(network_start_ip, 8) || '/' || network_mask_length - 96) )) |
                            hostmask(inet (substr(network_start_ip, 8) || '/' || network_mask_length - 96))
                           ) as IPEndStr
                from ipcity_staging st 
                where network_start_ip like '::ffff:%'
               ) st
         ) st;



1 comment:

  1. Why not use the built-in network address types (http://www.postgresql.org/docs/9.3/static/datatype-net-types.html) instead of VARCHARs or TEXTs? There's a rich set of operators and functions available for those types (http://www.postgresql.org/docs/9.3/static/functions-net.html) meaning you don't have to roll your own with string processing.

    ReplyDelete

Your comment will appear when it has been reviewed by the moderators.