Labka-cz / hadoop-dns-mining

hadoop-dns-mining

This is a small framework for performing large amounts of DNS lookups using Hadoop. This is a work in progress, pull requests are welcome.

Here are the steps for getting it working:

Download, compile and install the Maxmind JAR into maven

wget http://geolite.maxmind.com/download/geoip/api/java/GeoIPJava-1.2.5.zip
unzip GeoIPJava-1.2.5.zip
cd GeoIPJava-1.2.5/source/com/maxmind/geoip/
javac *.java
cd ../../../
zip -r maxmind.jar com/
mvn install:install-file -Dfile=maxmind.jar -DgroupId=com.maxmind -DartifactId=geo-ip -Dversion=1.2.5 -Dpackaging=jar

Obtain the Maxmind IP Geo Database

wget http://geolite.maxmind.com/download/geoip/database/GeoLiteCity.dat.gz
gzip -d GeoLiteCity.dat.gz

Obtain the Maxmind ASN Database

wget http://www.maxmind.com/download/geoip/database/asnum/GeoIPASNum.dat.gz
gzip -d GeoIPASNum.dat.gz

Create/obtain large lists of domain names (e.g. domains.txt) and copy them into HDFS

# you may want to split these domain files before placing in HDFS in order to use more mappers
split -a 5 -d -l 100000  domains.txt domains_
hadoop fs -put domains_* /data/domains/

Download and build this project

git clone https://[e-mail]/jt6211/hadoop-dns-mining.git
cd hadoop-dns-mining
mvn package assembly:assembly

Run the various MapReduce jobs

# These are the records that will be requested
REC_TYPES=A,MX,NS,TXT

JAR=target/hadoop-dns-mining-1.0-SNAPSHOT-job.jar

# performs A record, MX record, and NS record lookups on each domain provided using 50 
# resolving threads per Mapper using the nameserver of 8.8.8.8 and store the results in 
# HDFS in /data/dns-mining/01_raw
# Note: choose the nameserver wisely, otherwise you may overload it.  In testing I mainly 
#  used a bind server deployed on each hadoop node so my nameserver was 127.0.0.1
time hadoop jar $JAR io.covert.dns.collection.CollectionJob \
    -D dns.collection.num.resolvers=50 \
    -D dns.collection.nameservers=8.8.8.8 \
    IN \
    "$REC_TYPES" \
    /data/domains/ \
    /data/dns-mining/01_raw

# parse the raw responses into JSON (one record per RR in the DNS responses)
time hadoop jar $JAR io.covert.dns.parse.ParseJob \
    /data/dns-mining/01_raw \
    /data/dns-mining/02_parsed

# lookup any IP addresses in the results in the maxmind DBs and enrich the records
time hadoop jar $JAR io.covert.dns.geo.GeoJob \
    -files /usr/local/lib/maxmind/GeoLiteCity.dat,/usr/local/lib/maxmind/GeoIPASNum.dat \
    GeoLiteCity.dat \
    GeoIPASNum.dat \
    /data/dns-mining/02_parsed  \
    /data/dns-mining/03_enriched

# run a filter job for the rec types requested as well as for rec types that commonly occur in 
# the results as part of normal queries.  This will separate the various DNS records into their
# own directories in HDFS
for X REC `echo "$REC_TYPES,SOA,NS,CNAME" | sed 's/,/\n/g'| sort -u`; 
do
    time hadoop jar $JAR io.covert.dns.filtering.FilterJob \
        "type == '$REC'" \
        /data/dns-mining/03_enriched \
        /data/dns-mining/04_filtered-type=$REC; 
done

# This is a JEXL expression that filters out target fields that are IP addresses 
# and returns the target field lowercased
TARGET_EXPR='if(target !~ "^\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}\.$")return target.toLowerCase()'

# extract the 'target' field from the MX records
time hadoop jar $JAR io.covert.dns.extract.ExtractorJob "$TARGET_EXPR" \
    /data/dns-mining/04_filtered-type=MX /data/dns-mining/05_extracted-mailservers

# extract the 'target' field from the NS records
time hadoop jar $JAR io.covert.dns.extract.ExtractorJob "$TARGET_EXPR" \
    /data/dns-mining/04_filtered-type=NS /data/dns-mining/05_extracted-nameservers

HOST_EXPR='if(host !~ "^\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}\.$")return host.toLowerCase()'

# extract the 'host' field from the SOA records
time hadoop jar $JAR io.covert.dns.extract.ExtractorJob "$HOST_EXPR" \
    /data/dns-mining/04_filtered-type=SOA /data/dns-mining/05_extracted-nameservers-SOA