2012-06-29 05:18:28 +00:00
|
|
|
# -*- coding: binary -*-
|
2009-11-18 06:35:44 +00:00
|
|
|
|
|
|
|
require 'rexml/document'
|
|
|
|
|
|
|
|
module Rex
|
|
|
|
module Parser
|
|
|
|
|
|
|
|
#
|
|
|
|
# Stream parser for nmap -oX xml output
|
|
|
|
#
|
|
|
|
# Yields a hash representing each host found in the xml stream. Each host
|
|
|
|
# will look something like the following:
|
|
|
|
# {
|
|
|
|
# "status" => "up",
|
|
|
|
# "addrs" => { "ipv4" => "192.168.0.1", "mac" => "00:0d:87:a1:df:72" },
|
2010-02-14 18:32:37 +00:00
|
|
|
# "ports" => [
|
2009-11-18 06:35:44 +00:00
|
|
|
# { "portid" => "22", "state" => "closed", ... },
|
|
|
|
# { "portid" => "80", "state" => "open", ... },
|
2010-02-14 18:32:37 +00:00
|
|
|
# ...
|
2009-11-18 06:35:44 +00:00
|
|
|
# ]
|
|
|
|
# }
|
|
|
|
#
|
|
|
|
# Usage:
|
2011-04-04 17:06:59 +00:00
|
|
|
# parser = NmapXMLStreamParser.new { |host|
|
|
|
|
# # do stuff with the host
|
|
|
|
# }
|
|
|
|
# REXML::Document.parse_stream(File.new(nmap_xml), parser)
|
2009-11-18 06:35:44 +00:00
|
|
|
# -- or --
|
2011-04-04 17:06:59 +00:00
|
|
|
# parser = NmapXMLStreamParser.new
|
|
|
|
# parser.on_found_host = Proc.new { |host|
|
|
|
|
# # do stuff with the host
|
|
|
|
# }
|
|
|
|
# REXML::Document.parse_stream(File.new(nmap_xml), parser)
|
2009-11-18 06:35:44 +00:00
|
|
|
#
|
|
|
|
# This parser does not maintain state as well as a tree parser, so malformed
|
|
|
|
# xml will trip it up. Nmap shouldn't ever output malformed xml, so it's not
|
|
|
|
# a big deal.
|
|
|
|
#
|
|
|
|
class NmapXMLStreamParser
|
|
|
|
|
2013-08-30 21:28:33 +00:00
|
|
|
#
|
|
|
|
# Callback for processing each found host
|
|
|
|
#
|
|
|
|
attr_accessor :on_found_host
|
2009-11-18 06:35:44 +00:00
|
|
|
|
2013-08-30 21:28:33 +00:00
|
|
|
#
|
|
|
|
# Create a new stream parser for NMAP XML output
|
|
|
|
#
|
|
|
|
# If given a block, it will be stored in +on_found_host+, otherwise you
|
|
|
|
# need to set it explicitly, e.g.:
|
|
|
|
# parser = NmapXMLStreamParser.new
|
|
|
|
# parser.on_found_host = Proc.new { |host|
|
|
|
|
# # do stuff with the host
|
|
|
|
# }
|
|
|
|
# REXML::Document.parse_stream(File.new(nmap_xml), parser)
|
|
|
|
#
|
|
|
|
def initialize(&block)
|
|
|
|
reset_state
|
|
|
|
on_found_host = block if block
|
|
|
|
end
|
2009-11-18 06:35:44 +00:00
|
|
|
|
2013-08-30 21:28:33 +00:00
|
|
|
def reset_state
|
|
|
|
@host = { "status" => nil, "addrs" => {}, "ports" => [], "scripts" => {} }
|
|
|
|
@state = nil
|
|
|
|
end
|
2009-11-18 06:35:44 +00:00
|
|
|
|
2013-08-30 21:28:33 +00:00
|
|
|
def tag_start(name, attributes)
|
|
|
|
begin
|
|
|
|
case name
|
|
|
|
when "address"
|
|
|
|
@host["addrs"][attributes["addrtype"]] = attributes["addr"]
|
|
|
|
if (attributes["addrtype"] =~ /ipv[46]/)
|
|
|
|
@host["addr"] = attributes["addr"]
|
|
|
|
end
|
|
|
|
when "osclass"
|
|
|
|
# If there is more than one, take the highest accuracy. In case of
|
|
|
|
# a tie, this will have the effect of taking the last one in the
|
|
|
|
# list. Last is really no better than first but nmap appears to
|
|
|
|
# put OSes in chronological order, at least for Windows.
|
|
|
|
# Accordingly, this will report XP instead of 2000, 7 instead of
|
|
|
|
# Vista, etc, when each has the same accuracy.
|
|
|
|
if (@host["os_accuracy"].to_i <= attributes["accuracy"].to_i)
|
|
|
|
@host["os_vendor"] = attributes["vendor"]
|
|
|
|
@host["os_family"] = attributes["osfamily"]
|
|
|
|
@host["os_version"] = attributes["osgen"]
|
|
|
|
@host["os_accuracy"] = attributes["accuracy"]
|
|
|
|
end
|
|
|
|
when "osmatch"
|
|
|
|
if(attributes["accuracy"].to_i == 100)
|
|
|
|
@host["os_match"] = attributes["name"]
|
|
|
|
end
|
|
|
|
when "uptime"
|
|
|
|
@host["last_boot"] = attributes["lastboot"]
|
|
|
|
when "hostname"
|
|
|
|
if(attributes["type"] == "PTR")
|
|
|
|
@host["reverse_dns"] = attributes["name"]
|
|
|
|
end
|
|
|
|
when "status"
|
|
|
|
# <status> refers to the liveness of the host; values are "up" or "down"
|
|
|
|
@host["status"] = attributes["state"]
|
|
|
|
@host["status_reason"] = attributes["reason"]
|
|
|
|
when "port"
|
|
|
|
@host["ports"].push(attributes)
|
|
|
|
when "state"
|
|
|
|
# <state> refers to the state of a port; values are "open", "closed", or "filtered"
|
|
|
|
@host["ports"].last["state"] = attributes["state"]
|
|
|
|
when "service"
|
|
|
|
# Store any service and script info with the associated port. There shouldn't
|
|
|
|
# be any collisions on attribute names here, so just merge them.
|
|
|
|
@host["ports"].last.merge!(attributes)
|
|
|
|
when "script"
|
|
|
|
# Associate scripts under a port tag with the appropriate port.
|
|
|
|
# Other scripts from <hostscript> tags can only be associated with
|
|
|
|
# the host and scripts from <postscript> tags don't really belong
|
|
|
|
# to anything, so ignore them
|
|
|
|
if @state == :in_port_tag
|
|
|
|
@host["ports"].last["scripts"] ||= {}
|
|
|
|
@host["ports"].last["scripts"][attributes["id"]] = attributes["output"]
|
|
|
|
elsif @host
|
|
|
|
@host["scripts"] ||= {}
|
|
|
|
@host["scripts"][attributes["id"]] = attributes["output"]
|
|
|
|
else
|
|
|
|
# post scripts are used for things like comparing all the found
|
|
|
|
# ssh keys to see if multiple hosts have the same key
|
|
|
|
# fingerprint. Ignore them.
|
|
|
|
end
|
|
|
|
when "trace"
|
|
|
|
@host["trace"] = {"port" => attributes["port"], "proto" => attributes["proto"], "hops" => [] }
|
|
|
|
when "hop"
|
|
|
|
if @host["trace"]
|
|
|
|
@host["trace"]["hops"].push(attributes)
|
|
|
|
end
|
|
|
|
end
|
|
|
|
rescue NoMethodError => err
|
|
|
|
raise err unless err.message =~ /NilClass/
|
|
|
|
end
|
|
|
|
end
|
2009-11-18 06:35:44 +00:00
|
|
|
|
2013-08-30 21:28:33 +00:00
|
|
|
def tag_end(name)
|
|
|
|
case name
|
|
|
|
when "port"
|
|
|
|
@state = nil
|
|
|
|
when "host"
|
|
|
|
on_found_host.call(@host) if on_found_host
|
|
|
|
reset_state
|
|
|
|
end
|
|
|
|
end
|
2009-11-18 06:35:44 +00:00
|
|
|
|
2013-08-30 21:28:33 +00:00
|
|
|
# We don't need these methods, but they're necessary to keep REXML happy
|
|
|
|
def text(str) # :nodoc:
|
|
|
|
end
|
|
|
|
def xmldecl(version, encoding, standalone) # :nodoc:
|
|
|
|
end
|
|
|
|
def cdata # :nodoc:
|
|
|
|
end
|
|
|
|
def comment(str) # :nodoc:
|
|
|
|
end
|
|
|
|
def instruction(name, instruction) # :nodoc:
|
|
|
|
end
|
|
|
|
def attlist # :nodoc:
|
|
|
|
end
|
2009-11-18 06:35:44 +00:00
|
|
|
end
|
|
|
|
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|