This is a fork of the public repository of ID2T (Intrusion Detection Dasaset Toolkit). ID2T aims at creating labeled IT network datasets that contain user defined synthetic attacks.
|
6 years ago | |
---|---|---|
code | 6 years ago | |
code_boost | 6 years ago | |
logo | 7 years ago | |
resources | 6 years ago | |
.gitignore | 6 years ago | |
.gitmodules | 6 years ago | |
LICENSE.md | 7 years ago | |
README.md | 6 years ago | |
build.sh | 6 years ago |
A toolkit for injecting synthetic attacks into PCAP files.
As Intrusion Detection Systems encounter growing importance in the area of network security, the need of high quality network datasets for evaluation against real-world attacks rises.
Comparability of the results must be ensured by use of publicly available datasets. Existing datasets, however, suffer from several disadvantages. Often they do not provide ground truth knowledge, consist of outdated traffic and do not contain any payload because of privacy reasons. Moreover, frequently datasets do not contain latest attacks and missing attack labels make it difficult to identify existing attacks and enable a transparent comparison of Intrusion Detection Systems.
The ID2T toolkit was first proposed in [1] and [2] and targets the injection of attacks into existing network datasets. At first, it analyzes a given dataset and collects statistics from it. These statistics are stored into a local database. Next, these statistics can be used to define attack parameters for the injection of one or multiple attacks. Finally, the application creates the required attack packets and injects them into the existing file. Resulting in a new PCAP with the injected attacks and a label file indicating the position (timestamps) of the first and last attack packet.
[1] Garcia Cordero et al. (2015) ID2T: a DIY Dataset Creation Toolkit for Intrusion Detection System
ID2T is written using Python 3 and C++ 11. The main logic is programmed in Python whereas performance critical components are programmed in C++11. The C++11 module uses the Libtins library. The python and c++ modules interact with each other through the Boost.Python library .
The following packages/libraries are required to compile the ID2T C++ modules
cmake
(minimum version 2.8)
boost
with the python
component (minimum version 1.54)
libtins
(minimum version 3.4)
python
development libraries
sqlite
(minimum version 3.0)
The following python packages are required to run ID2T. Install the packages with your preferred package manager. For example, you can use pip3 (pip for python 3). Install pip3 in ubuntu with apt install python3-pip
and install the packages with sudo pip3 install <packagename>
.
scapy
(make sure its the python3 version)lea
numpy
matplotlib
SciPy Stack
(see installation instructions)The minimum version stated in the previous requirements are the versions we have used in the development of ID2T. Other (older) versions might also work; however, we cannot guarantee nor support them. Furthermore, some compilation scripts would need to be manually modified to accommodate these older versions.
Once you satisfy all dependencies, clone the repository to get started with the installation:
git clone https://git.tk.informatik.tu-darmstadt.de/SPIN/ID2T-toolkit
After cloning the repository, initialize its submodules with
git submodule init
git submodule update
Build the C++ modules and create the ID2T executable:
./build.sh
Run ID2T with the command ./id2t
.
In this section, we provide examples on how ID2T is used.
In the following we inject the PortscanAttack into the dataset pcap_capture.pcap:
./id2t -i /home/user/pcap_capture.pcap -a PortscanAttack ip.src=192.168.178.2 mac.src=32:08:24:DC:8D:27 inject.at-timestamp=1476301843
Explanation: The parameter -i/--input
takes the path to the PCAP file. This triggers the statistics calculation of the file. After the calculation, the statistics are stored into a SQLite database. If the statistics were already computed in an earlier run, the data is retrieved from the generated database. This saves time as the calculation of the statistics may take long time - depending on the PCAP file size.
An attack can be injected by providing -a/--attack
followed by the attack name and the attack parameters. The available attacks and the allowed attack parameters vary, see the attack-specific wiki articles for a reference of supported attack parameters. The parameter -a/--attack
can be provided multiple times for injection of multiple attacks. In this case the attacks are injected sequentially.
After injecting the attack, the application generates a XML label file containing the timestamps of the first and last attack packet. The file name is equal to the output file, except with _labels.xml
as suffix.
The toolkit recognizes if the input dataset has an associated label file. This requires a file naming according to the aforementioned scheme, e.g., mydataset.pcap and mydataset_labels.xml. In this case ID2T parses the label file and the resulting output label file contains the labels from the input label file plus the labels from the recently added attack(s).
Whenever ID2T processes a pcap file, it creates a database detailing many things related to the network traffic it has processed. These details can be seen using the query mode of ID2T. To specify a query against a pcap file, use the option `-q/--query
. For example, if we want to know the IP address with the most activity in the pcap file 'test.pcap' we can issue the command:
./id2t -i test.pcap -q 'most_used(ipAddress);'
The query mode serves as a place where standard SQL queries (known as user-defined queries) can be issued against the database created for a pcap file. Furthermore, the most commonly used queries are provided with special keywords known as named queries.
SELECT ipAddress FROM ip_statistics WHERE pktsSent > 1000
most_used(ipAddress)
, random(all(ipAddress))
The named queries can be further divided into two classes:
all(ipAddress)
random(...)
returns a randomly chosen element of the listA complete list of supported named queries can be found in section Named Queries. The database scheme, required for building SQL queries, is documented in the wiki article DB Tables and Fields
If -q/--query
is called without an argument, the application enters into REPL query mode. This mode is like a standard read-eval-print-loop (REPL) for SQL queries. In this mode, the user can repeatedly input queries (each query must finish with a ";" (semicolon)); send the query by pressing ENTER and see the response in the terminal:
Example query mode usage: ./id2t -i test.pcap -q
Example output:
Input file: /home/user/pcap_capture.pcap
Located statistics database at: /home/pjattke/ID2T_data/db/99/137/81a0a71b0f36.sqlite3
Loaded file statistics in 0.00 sec from statistics database.
Entering into query mode...
Enter statement ending by ';' and press ENTER to send query. Exit by sending an empty query..
most_used(ipAddress);
Query 'most_used(ipAddress);' returned:
203.114.236.243
avg(ttlValue);
Query 'avg(pktsSent);' returned:
5.322
By calling ./id2t -h
, a list of available application arguments with a short description is shown.
Selectors are named queries which return a single element or a list of elements, depending on the values in the database and the query.
For example, the named query most_used(ipAddress)
may return a single IP address if the most used IP address, based on the sum of packets sent and received, is unique. If there are multiple IP addresses with the same number of packets sent plus packets received, a list of IP addresses is returned. As the user cannot know how many values are returned, the extractors are ignored if the result is a single element.
most_used(ipAddress | macAddress | portNumber | protocolName | ttlValue)
least_used(ipAddress | macAddress | portNumber | protocolName | ttlValue)
avg(pktsReceived | pktsSent | kbytesSent | kbytesReceived | ttlValue | mss)
all(ipAddress | ttlValue | mss | macAddress | portNumber | protocolName)
There are also parameterizable selectors which take conditions as input. Following two examples to show the syntax by example:
ipAddress(macAddress=AA:BB:CC:DD:EE:FF, pktsSent > 1000, kbytesReceived < 1000)
-> returns one or multiple IP addresses matching the given criterias
Supports the fields: macAddress, ttlValue, ttlCount, portName, portNumber, portDirection, kbytesSent, kbytesReceived, pktsSent, pktsReceived,
macAddress(ipAddress=192.168.178.2)
-> returns the MAC address matching the given criteria
Supports the field: ipAddress
Extractors are to be used on the result of a named query. If the result is a list, applying an extractor reduces the result set to a single element. If the result is already a single element, the extractor is ignored.
random(...) -> returns a random element from a list
first(...) -> returns the first element from a list
last(...) -> returns the last element from a list
Attention: Named queries are designed to be combined with extractors, like random(all(ipAddress))
. But it is currently NOT possible to encapsulate multiple named queries, like macAddress(ipAddress=most_used(ipAddress))
. This can be circumvented by first querying most_used(ipAddress)
and then inserting the result as argument in macAddress(…)
.
The SemVer is used for versioning. For currently available versions of ID2T, see page releases.
Dr. Emmanouil Vasilomanolakis - ID2T idea, guidance and suggestions during development
Carlos Garcia - ID2T idea, guidance and suggestions during development
Nikolay Milanov - development of first prototype within his Master Thesis
Patrick Jattke - development of first public release as part of his Bachelor Thesis
Distributed under the MIT license. See LICENSE for more information.