Here, you can see the details of all components of the NetSetEnvironment and their usage. These components are located in game_components.py.
The following classes are used in the game to hold information about the state of the game. They are used both in the Actions and GameState.
IP is immutable object that represents an IPv4 object in the NetSecGame. It has a single parameter of the address in a dot-decimal notation (4 octet represeted as decimal value separeted by dots).
Example: ip = IP("192.168.1.1")
Network is immutable object that represents an IPv4 network object in the NetSecGame. It has 2 parameters:
network_ip:str
representing the IPv4 address of the network.mask:int
representing the mask in the CIDR notation.
Example: net = Network("192.168.1.0", 24)
Service class holds information about services running in hosts. Each Service has four parameters:
name
:str - Name of the service (e.g., "SSH")type
:str -passive
oractive
. Currently not being used.version
:str - version of the service.is_local
:bool - flag specifying if the service is local only. (ifTrue
, service is NOT visible without controlling the host).
Example: s = Service('postgresql', 'passive', '14.3.0', False)
Data class holds information about datapoints (files) present in the NetSecGame. Datapoints DO NOT hold the content of files. Each data instance has two parameters:
owner
:str - specifying the user who owns this datapointid
: str - unique identifier of the datapoint in a hostsize
: int - size of the datapoint (optional, default=0)type
: str - identification of a type of the file (optional, default="")
Examples:Data("User1", "DatabaseData")
, Data("User1", "DatabaseData", size=42, type="txt")
GameState is an object that represents a view of the NetSecGame environment in a given state. It is constructed as a collection of 'assets' available to the agent. GameState has following parts:
known_networks
: Set of Network objects that the agent is aware ofknown_hosts
: Set of IP objects that the agent is aware ofcontrolled_hosts
: Set of IP objetcs that the agent has control over. Note thatcontrolled_hosts
is a subset ofknown_hosts
.known_services
: Dictionary of services that the agent is aware of. The dictionary format: {IP
: {Service
}} where IP object is a key and the value is a set of Service objects located in theIP
.known_data
: Dictionary of data instances that the agent is aware of. The dictionary format: {IP
: {Data
}} where IP object is a key and the value is a set of Data objects located in theIP
.known_blocks
: Dictionary of firewall blocks the agent is aware of. It is a dictionary with format: {target_IP
: {blocked_IP
,blocked_IP
}}. Wheretarget_IP
is the IP where the FW rule was applied (usually a router) andblocked_IP
is the IP address that is blocked. For now the blocks happen in both input and output direction simultaneously.
Actions are the objects sent by the agents to the environment. Each action is evaluated by AIDojo and executed if
- Is a valid Action
- Can be processed in the current state of the environment
In all cases, when an agent sends an action to AIDojo, it is given a response.
The Action class is defined in env.game_components.py
. It has two basic parts:
- ActionType:Enum
- parameters:dict
ActionType is unique Enum that determines what kind of action is agent playing. Parameters are passed in a dictionary as follows.
- JoinGame, params={
agent_info
:AgentInfo(<name>, <role>)}: Used to register agent in a game with a given <role>. - QuitGame, params={}: Used for termination of agent's interaction.
- ResetGame, params={}: Used for requesting reset of the game to it's initial position.
- ScanNetwork, params{
source_host
:<IP>,target_network
:<Network>}: Scans the given <Network> from a specified source host. Discovers ALL hosts in a network that are accessible from <IP>. If successful, returns set of discovered <IP> objects. - FindServices, params={
source_host
:<IP>,target_host
:<IP>}: Used to discover ALL services running in thetarget_host
if the host is accessible fromsource_host
. If successful, returns a set of all discovered <Service> objects. - FindData, params={
source_host
:<IP>,target_host
:<IP>}: Searchestarget_host
for data. Ifsource_host
differs fromtarget_host
, success depends on accessability from thesource_host
. If successful, returns a set of all discovered <Data> objects. - ExploitService, params={
source_host
:<IP>,target_host
:<IP>,taget_service
:<Service>}: Exploitstarget_service
in a specifiedtarget_host
. If successful, the attacker gains control of thetarget_host
. - ExfiltrateData, params{
source_host
:<IP>,target_host
:<IP>,data
:<Data>}: Copiesdata
from thesource_host
totarget_host
IF both are controlled andtarget_host
is accessible fromsource_host
.
In the following table, we describe the effects of selected actions and their preconditions. Note that if the preconditions are not satisfied, the actions's effects are not applied.
Action | Params | Preconditions | Effects |
---|---|---|---|
ScanNetwork | source_host , target_network |
source_host ∈ controlled_hosts |
extends known_networks |
FindServices | source_host , target_host |
source_host ∈ controlled_hosts |
extends known_services AND known_hosts |
FindData | source_host , target_host |
source_host , target_host ∈ controlled_hosts |
extends known_data |
Exploit Service | source_host , target_host , target_service |
source_host ∈ controlled_hosts |
extends controlled_hosts with target_host |
ExfiltrateData | source_host ,target_host , data |
source_host , target_host ∈ controlled_hosts AND data ∈ known_data |
extends known_data[target_host] with data |
- When playing the
ExploitService
action, it is expected that the agent has discovered this service before (by playingFindServices
in thetarget_host
before this action) - The
Find Data
action finds all the available data in the host if successful. - The
Find Data
action requires ownership of the target host. - Playing
ExfiltrateData
requires controlling BOTH source and target hosts - Playing
Find Services
can be used to discover hosts (if those have any active services) - Parameters of
ScanNetwork
andFindServices
can be chosen arbitrarily (they don't have to be listed inknown_newtworks
/known_hosts
)
After submitting Action a
to the environment, agents receive an Observation
in return. Each observation consists of 4 parts:
state
:Gamestate
- with the current view of the environment statereward
:int
- with the immediate reward agent gets for playing Actiona
end
:bool
- indicating if the interaction can continue after playing Actiona
info
:dict
- placeholder for any information given to the agent (e.g., the reason whyend is True
)