Lead Image © Yewkeo, 123RF.com

Lead Image © Yewkeo, 123RF.com

Endlessh and tc tarpits slow down attackers

Sticky Fingers

Article from ADMIN 61/2021
By
Keep an attacker's connections open in an Endlessh "tarpit" or delay incoming connections with the more traditional rate-limiting approach of tc.

A number of methods can stop attackers from exhausting your server resources, such as filtering inbound traffic with a variety of security appliances locally or by utilizing commercial, online traffic-scrubbing services to catch upstream traffic for mitigating denial-of-service attacks. Equally, honeypots can be used to draw attackers in, so you get a flavor of the attacks that your production servers might be subjected to in the future.

In this article, I look at a relatively unusual technique for slowing attackers down. First, Endlessh, a natty piece of open source software, can consume an attacker's resources by keeping their connections open (so that they have less ability themselves to attack your online services), leaving them in a "tarpit." Second, to achieve similar results a more traditional rate-limiting approach, courtesy of advanced Linux networking and traffic control (tc), is investigated with the kernel's built-in Netfilter packet filter controlled by its iptables frontend.

As surely as night follows day, automated attacks will target the default Secure Shell port (TCP port 22), so I will use SSH as the guinea pig test case with the knowledge that I can move the real SSH service to an alternative port without noticeable disruption.

Sticky Connections

If you visit the GitHub page for Endlessh [1], you are greeted with a brief description of its purpose: "Endlessh is an SSH tarpit that very slowly sends an endless, random SSH banner. It keeps SSH clients locked up for hours or even days at a time."

The documentation goes on to explain that if you choose a non-standard port for your SSH server and leave Endlessh running on TCP port 22, it's possible to tie attackers in knots, reducing their ability to do actual harm. One relatively important caveat, though, is that if you commit too much capacity to what is known as tarpitting (i.e., bogging something down), it is possible to cause a denial of service unwittingly to your own services. Therefore, you should never blindly deploy security tools like these in production environments without massive amounts of testing first.

The particular tarpit I build here on my Ubuntu 20.04 (Focal Fossa, with the exceptionally aesthetically pleasing Linux Mint 20 Ulyana sitting atop) will catch the connection when the SSH banner is displayed before SSH keys are exchanged. By keeping things simple, you don't have to worry about the complexities involved in solving encryption issues.

In true DevOps fashion, I fire up Endlessh with a Docker container:

$ git clone https://github.com/skeeto/endlessh

If you look at the Dockerfile within the repository, you can see an image that uses Alpine Linux as its base (Listing 1).

Listing 1

Endlessh Dockerfile

FROM alpine:3.9 as builder
RUN apk add --no-cache build-base
ADD endlessh.c Makefile /
RUN make
FROM alpine:3.9
COPY --from=builder /endlessh /
EXPOSE 2222/tcp
ENTRYPOINT ["/endlessh"]
CMD ["-v"]

Assuming Docker is installed correctly (following the installation process for Ubuntu 20.04 in my case, or as directed otherwise [2]), you can build a container with the command:

$ cd endlessh/
$ docker build -t endlessh .
[...]
Successfully built 6fc5221548db
Successfully tagged endlessh:latest

Next, check that the container image exists with docker images (Listing 2). Now you can spawn a container. If you are au fait with Dockerfiles, you will have spotted that TCP port 2222 will be exposed, as shown in the output:

$ docker run -it endlessh
2020-11-09T15:38:03.585Z Port 2222
2020-11-09T15:38:03.586Z Delay 10000
2020-11-09T15:38:03.586Z MaxLineLength 32
2020-11-09T15:38:03.586Z MaxClients 4096
2020-11-09T15:38:03.586Z BindFamily IPv4 Mapped IPv6

Listing 2

docker images

$ docker images
REPOSITORY TAG        IMAGE ID         CREATED              SIZE
endlessh   latest     6fc5221548db     58 seconds ago       5.67MB
<none>     <none>     80dc7d447a48     About a minute ago   167MB
alpine     3.9        78a2ce922f86     5 months ago         5.55MB

The command you really want to use, however, will expose that container port on the underlying host, too:

$ docker run -d --name endlessh -p 2222:2222 endlessh

You can, of course, adjust the first 2222 entry and replace it with 22, the standard TCP port. Use the docker ps command to make sure that the container started as hoped.

In another terminal, you can check to see whether the port is open as hoped by the Docker container (Listing 3). Next, having proven that you have a working Endlessh instance, you can put it through its paces.

Listing 3

Checking for Open Port

$ lsof -i :2222
COMMAND     PID USER   FD  TYPE  DEVICE SIZE/OFF NODE NAME
docker-pr 13330 root   4u  IPv4   91904      0t0  TCP *:2222 (LISTEN)

A simple test is to use the unerring, ultra-reliable netcat to see what is coming back from port 2222 on the local machine:

$ nc -v localhost 2222
Connection to localhost 2222 port [tcp/*] succeeded!
vc"06m6rKE"S40rSE2l
&Noq1>p&DurlvJh84S
 bHzlY
mTj-(!EP_Ta|B]CJu;s'1^:m7/PrYF
LA%jF#vxZnN3Ai

Each line of output takes 10 seconds to appear after the succeeded line and is clearly designed to confuse whoever is connecting to the port into thinking something is about to respond with useful, sane commands. Simple but clever.

If you need more information to deploy Endlessh yourself, use the docker run command with the -h option at the end to see the help output (Listing 4).

Listing 4

Endlessh Help Output

$ docker run endlessh -h
Usage: endlessh [-vh] [-46] [-d MS] [-f CONFIG] [-l LEN] [-m LIMIT] [-p PORT]
  -4        Bind to IPv4 only
  -6        Bind to IPv6 only
  -d INT    Message millisecond delay [10000]
  -f        Set and load config file [/etc/endlessh/config]
  -h        Print this help message and exit
  -l INT    Maximum banner line length (3-255) [32]
  -m INT    Maximum number of clients [4096]
  -p INT    Listening port [2222]
  -v        Print diagnostics to standard output (repeatable)
  -V        Print version information and exit

As the help output demonstrates, it is very easy to alter ports (from a container perspective), but you should edit the docker run command as discussed above. Endlessh allows you to specify how much gobbledygook is displayed with the -l (length) option. To prevent an unexpected, self-induced denial of service on your own servers, you can hard-code the maximum number of client connections permitted so that your network stack doesn't start to creak at the seams.

Finally, it is possible to expose only the desired connection port on IPv4, IPv6, or both and, with the -d setting, alter the length of the gobbledygook display delays (which by default is currently set at 10,000ms or 10s).

Colonel Control

If you don't want to use a prebuilt solution like Endlessh for creating a tarpit, you can achieve similar results with tools available to Linux by default, or more accurately, with the correct kernel modules installed. As mentioned, this approach is more rate-limiting than tarpitting but ultimately much more powerful and well worth discovering. Much of the following was inspired by a blog post from 2017 [3]. An introduction that quotes Wikipedia notes that a tarpit "is a service on a computer system (usually a server) that purposely delays incoming connections" [4].

Having looked at that post, followed by a little more reading, I realized that I'd obviously missed the fact that iptables (the kernel's network firewall, courtesy of the Netfilter project's component [5]) has included by design a tarpit feature of its very own. A look at the online manual [6] (or the output of the man iptables command) has some useful information to help you get started. The premise to using the TARPIT target module in iptables, as you'd expect from such a sophisticated piece of software, looks well considered and refined.

The docs state that the iptables target, "Captures and holds incoming TCP connections using no local per-connection resources." Note that this reassuring opening sentence apparently improves the Endlessh chance of unwittingly causing a local denial-of-service attack. The manual goes on to say, "Attempts to close the connection are ignored, forcing the remote side to time out the connection in 12-24 minutes." That sounds pretty slick. A careful reading of the manual reveals more tips. To open a "sticky" port to torment attackers, you can use the command:

$ iptables -A INPUT -p tcp -m tcp --dport 22 -j TARPIT

Noted in the documentation is that you can prevent the connection tracking (conntrack) functionality in iptables from tracking such connections with the NOTRACK target. Should you not do this, the kernel will unnecessarily use up resources for those connections that are stuck in a tarpit, which is clearly unwelcome behavior.

To get started with the advanced Linux networking approach, according to the aforementioned blog post, you need to make sure your kernel supplies the tc traffic control utility (stripped down kernels might not have it enabled). Thankfully, on my Linux Mint version it is. To determine whether your system has the tool, enter the tc command (Listing 5). In Listing 6, you can see that no rules currently are loaded in iptables.

Listing 5

tc

$ tc
Usage:  tc [ OPTIONS ] OBJECT { COMMAND | help }
        tc [-force] -batch filename
where  OBJECT := { qdisc | class | filter | chain | action | monitor | exec }
       OPTIONS := { -V[ersion] | -s[tatistics] | -d[etails] | -r[aw] | -o[neline] | -j[son] | -p[retty] | -c[olor] -b[atch] [filename] | -n[etns] name | -N[umeric] | -nm | -nam[es] | { -cf | -conf } path }

Listing 6

iptables -nvL

$ iptables -nvL
Chain INPUT (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source      destination
Chain FORWARD (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source      destination
Chain OUTPUT (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source      destination

On my laptop, an SSH server isn't installed automatically so I have to add it to the system:

$ apt install openssh-server
The following NEW packages will be installed
  ncurses-term openssh-server openssh-sftp-server ssh-import-id

After some testing, you can uninstall those exact packages to keep your system trim. The commands in Listing 7 start up the SSH daemon (sshd) and tell you that it is listening on port 22. IPv6 and IPv4 connections are open on the default port, so you can continue.

Listing 7

Starting sshd

$ systemctl start sshd
$ lsof -i :22
COMMAND  PID USER   FD   TYPE DEVICE SIZE/OFF NODE NAME
sshd    5122 root   3u   IPv4  62113      0t0  TCP *:ssh (LISTEN)
sshd    5122 root   4u   IPv6  62115      0t0  TCP *:ssh (LISTEN)

At this stage, if you are testing on a server, make sure that you have altered your main SSH port and restarted your SSH daemon, or potentially you will be locked out of your server. The steps you aim to achieve are:

  1. With iptables, mark packets to all traffic hitting the SSH port, TCP port 22, with MARK as an option.
  2. Use tc to set up the hierarchy traffic bucket (HTB) qdisc, to catch traffic to be filtered.
  3. Create an HTB rule that will be used for normal traffic (allowing the use of loads of bandwidth), calling it 1:0 .
  4. Create a second HTB rule that will only be allowed a tiny amount of traffic and call it 1:5 .
  5. Use tc to create a filter and get it to match the MARK option set in step 1; then, watch it get allocated to the 1:5 traffic flows.
  6. Check the output of both the HTB traffic class rules to look for overlimits.

Now it's time to put those steps in action.

To begin, add an iptables rule with a "mangle" table that helps you manipulate connections to your heart's content. For step 1, you may need to adjust the -A options to -I if you have rules in place, MARK your connections to the SSH port (with two rules – one as the source port and one as the destination – because you'll test to see whether it works by using connections from another local machine):

$ iptables -A OUTPUT -t mangle -p tcp --sport 22 -j MARK --set-mark 10
$ iptables -A OUTPUT -t mangle -p tcp --dport 22 -j MARK --set-mark 10

As you can see, one source port and one destination port have been configured to mark the packets with the label 10 . Now, check that traffic is hitting the rules, or chains, created by checking the mangle table (Listing 8).

Listing 8

Checking the Mangle Table

$ iptables -t mangle -nvL
Chain OUTPUT (policy ACCEPT 15946 packets, 7814K bytes)
 pkts bytes target     prot opt in     out        source     destination
  192 35671   MARK    tcp  --  *       *     0.0.0.0/0       0.0.0.0/0        tcp spt:22 MARK set 0xa
   31  4173   MARK    tcp  --  *       *     0.0.0.0/0       0.0.0.0/0        tcp dpt:22 MARK set 0xa

For step 2, run a new tc command to add the HTB qdisc (or the scheduler) to your network interface. First, however, you need to know the name of your machine's network interfaces with ip a. In my case, I can ignore the lo localhost interface, and I can see the wireless interface named wlpls0 , as seen in a line of the output:

wlp1s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 state UP group default qlen 1000

From now on, I can simply alter the wlp1s0 entry for the network interface.

Now, add the HTB qdisc scheduler to the network interface as your parent rule (which is confusingly referenced as 1: , an abbreviated form of 1:0 ):

$ tc qdisc add dev wlp1s0 root handle 1: htb

As per step 3, you need to create a rule for all your network interface traffic, with the exception of your tarpit or throttled traffic, by dutifully naming this classifier 1:0 (I prefer to think of such rules as a "traffic class" for simplicity):

$ tc class add dev wlp1s0 parent 1: classid 1:0 htb rate 8000Mbit

For step 4, instead of adding loads of traffic allowance add just 10bps of available throughput and call the entry 1:5 for later reference:

$ tc class add dev wlp1s0 parent 1: classid 1:5 htb rate 80bit prio 1

The filter for step 5 picks up all marked packets courtesy of the iptables rules in step 1 and matches the 1:5 traffic class entry in HTB:

$ tc filter add dev wlp1s0 parent 1: prio 1 protocol ip handle 10 fw flowid 5

Note the flowid 5 to match the 1:5 traffic class and the ip handle 10 to match the iptables rules.

For step 6, you can see your qdisc in action:

$ watch -n1 tc -s -g class show dev wlp1s0

Figure 1 shows the hierarchical explanation of the two child classifiers, which are sitting under the parent qdisc.

Figure 1: One large and one small traffic class, in terms of throughput.

For more granular information on qdisc, traffic class, or filter level, you can use the commands:

$ tc qdisc show dev wlp1s0
$ tc class show dev wlp1s0
$ tc filter show dev wlp1s0

Next, you should look for your local network interface's IP address (make sure you replace my network interface name with your own):

$ ip a | grep wlp1s0 | grep inet
inet 192.168.0.16/24 brd 192.168.0.255 scope global dynamic noprefixroute wlp1s0

Now log in to another machine on your local network with SSH access (or open up SSH access for a remote machine) and run the command,

$ ssh 192.168.0.16

replacing your IP address for mine. After watching a very slow login prompt, you can generate some arbitrary noise in the terminal (use any command that you like), such as the following, which on my machine will push screeds of data up the screen:

$ find / -name chrisbinnie

If all goes well, you should see the output move very, very slowly indeed, and if you're serving an application over another port (I'm not on my laptop), other services should be absolutely fine, running as usual at the normal speed.

The proof in the pudding – that your second class is matching traffic to the filter that references the iptables rules – is seen in Figure 1 for the 1.5 class. If you run a few commands on your other machine that is SSH'd into your throttled SSH server, you should see the overlimits increase steadily. In Figure 1, that is showing 36 packets.

If you need to triple-check that it is working as hoped, you can remove the running tc configuration:

$ tc qdisc del dev wlp1s0 root

SSH should magically return to being nice and responsive again.

Although you are not strictly tarpitting connections, you are limiting them significantly. The exceptional tc will only allow that tiny upper limit of bandwidth to all SSH connections on TCP port 22, so rest assured that with multiple attackers vying for such a small amount of traffic, it's not going to be an enjoyable experience.

The End Is Nigh

If you're keen to learn more about the genuinely outstanding tc and its collection of various qdiscs, which you should, you can find a lot more information online to alter your network traffic in almost any way you imagine. The manual page for Ubuntu [7] is a good start.

For further reading on tarpits, refer to the Server Fault forum [8], which discusses the good and bad elements of tarpits in detail and offers some useful insights into how the ordering of your iptable chains, or rules, can be set up to be the most efficient. Also, pay attention to the comments about accidentally filling up logs and causing a different type of denial of service that you might not have been expecting.

As mentioned before, be certain you know what you are switching on when it comes to tarpit functionality, whichever route you take. On occasion, honeypots and tarpits can be an invaluable addition to your security setup, but without some forethought, it is quite possible to tie your shoelaces across your shoes and trip yourself up, irritate your users, and cause a whole heap of extra work for yourself.

Buy this article as PDF

Express-Checkout as PDF
Price $2.95
(incl. VAT)

Buy ADMIN Magazine

SINGLE ISSUES
 
SUBSCRIPTIONS
 
TABLET & SMARTPHONE APPS
Get it on Google Play

US / Canada

Get it on Google Play

UK / Australia

comments powered by Disqus