is a home automation and monitoring project for the RaspberryPi. I saw some articles about PrivateEyePi and saw that they even have their own store
on their website. Since I am the author of alertR
(another open-source alarm system), I thought I take a closer look at what they are doing. Before I started with alertR, I looked briefly at some open-source projects to see if some project is doing what I wanted to do. In most cases it was sufficient to only look at the website of the project. PrivateEyePi was quickly dismissed by me because they only work with one host. Hence, I never took a closer look into it. And I know that it looks a little bit pitiful that I am writing an article about "how insecure this project is". But after taking a closer look, I have to say that from a security perspective I strongly discourage anybody to use it. And the following will explain why ...
Let us start at the beginning. And in the beginning was ... the installation. As far as I can tell I analyzed version v12
of the PrivateEyePi software. For all versions, the project explains in a small tutorial how to install PrivateEyePi
. They describe that you should download the install.sh
file and execute it via sudo sh install.sh
. The install.sh
file looks like the following:
sudo rm -f pep.zip
sudo mkdir -p pep_backup
sudo mv alarm.py pep_backup 2>/dev/null
sudo mv restarter.py pep_backup 2>/dev/null
sudo mv alarmfunctionsr.py pep_backup 2>/dev/null
sudo mv dht22.py pep_backup 2>/dev/null
sudo mv dallas.py pep_backup 2>/dev/null
sudo mv rfsensor.py pep_backup 2>/dev/null
sudo mv globals.py pep_backup 2>/dev/null
sudo mv webcam.py pep_backup 2>/dev/null
sudo mv lcd_hd44780.py pep_backup 2>/dev/null
sudo mv lcd_nokia.py pep_backup 2>/dev/null
sudo mv lcdtest.py pep_backup 2>/dev/null
sudo mv publish.py pep_backup 2>/dev/null
sudo mv subscribe.py pep_backup 2>/dev/null
sudo wget www.privateeyepi.com/downloads/pep.zip
unzip -o pep.zip
sudo chmod 777 alarm.py
sudo chmod 777 dallas.py
sudo chmod 777 globals.py
sudo chmod 777 alarmfunctionsr.py
sudo chmod 777 dht22.py
sudo chmod 777 restarter.py
sudo chmod 777 pep_backup
sudo chmod 777 webcam.py
sudo chmod 777 lcd_hd44780.py
sudo chmod 777 lcd_nokia.py
sudo chmod 777 lcdtest.py
sudo chmod 777 publish.py
sudo chmod 777 subscribe.py
sudo apt-get install python-serial
Perhaps some of you saw my little Twitter rant
(actually, I do not think so but everyone can hope, right?
). First of all the install.sh
file is already executed with sudo. Therefore all the additional sudos are unnecessarily. Second, you should not execute wget or unzip with root permissions. You never know what nasty bug could be triggered. But what you certainly do not do is change the file permissions for a file owned by root (or in this matter for any other user) to 777 (world-read-exec-write)! To make this bug complete, in the next steps of the tutorial
the alarm system is started via the command sudo python alarm.py
, which means it is also run with root permissions.
Perhaps some of you ask now "Why shouldn't I have a file world-read-exec-writable? And why not as root?". Well, some other user on the same system can write into the file (it is writable by everyone, so it can be any user). And when the application is now executed with root permissions, this user can execute code with root privileges.
Well, this was the installation process. Let us switch to the next step in the installation tutorial, the registration at the central server of PrivateEyePi.
Central Server Webinterface
You have to register an account on the central server of PrivateEyePi in order to use your alarm system at home. Wait, what?! Yes, PrivateEyePi uses a central server which is owned by them to give user access to their home alarm system. Sounds suspicious you think? Well, we will come back to this later.
We go not into depth here, since I am not the owner of this server and probably have no permission to test the webinterface. But again, perhaps you saw my little rant on Twitter
You can see that when I want to change my user account information, I can see my used password in the source code of the page. What does this mean? First of all this means that the password is stored in plaintext. This means anyone of the PrivateEyePi guys can read your password (or any attacker that compromised the server). If you have used your default password and your default eMail address that you use everywhere: Sucks to be you
! Second, even if you have a magical reason to store the password in plaintext (there is none!), you certainly do not send it back to a web page. For example, if you are logged in to the webinterface and someone else uses your browser, this someone can see your password.
This is all I can say about the webinterface. But seeing the quality of the code so far I am pretty sure that you will find other security issues there like SQLi (which would probably allow you to get user informations like eMail addresses and passwords ... in plaintext).
What is next in the installation tutorial? The configuration of the alarm system.
Configuring the Alarm System
Now we come back to the central server part. The alarm system is mainly configured via the webinterface on the central PrivateEyePi server. Stuff like eMail addresses the alarms should be sent to, GPIOs of the sensors etc. are configured via the central server. The stuff that is configured locally can be found in the globals.py
. This is rather uninteresting stuff like user credentials for the central PrivateEyePi server, or if the temperature is shown in Fahrenheit, SMTP server settings, or the GPIO for a connected siren (I do not know why this is actually configured locally when the sensors are configured on the central server).
The architecture of the system can be seen in the following figure:
As can be seen, all sensors and other hardware is connected to the RaspberryPi. The home alarm system collects the data and sends it to the central server. Additionally, the home alarm system polls the server for commands it should execute. The webinterface of the central server can be used by the user to see the information the sensors collect and give commands like turning off the alarm system (hopefully, some of you already see how problematic this design is ... if not, I will come to it in a few seconds).
I think this design was chosen to circumvent the need to configure port forwarding on the local router. This way, a user is able to reach his home alarm system from his mobile phone without changing any Internet settings at home or setting up a service that is reachable from the Internet. But this design is quite problematic and also naive. Why naive? Well, imagine you have a home alarm system. Would you tell a stranger that nobody is at your home? And while we are on it, would you also give this stranger the key to disarm your home alarm system? No? Well, me neither. But if you use PrivateEyePi you do.
Let me explain why before I show concrete attack scenarios. PrivateEyePi sends the collected sensor data to the central server. If you have motion detection sensors and sensors at your windows and doors, their states are send to the central server and stored there. Not only you, but also the PrivateEyePi guys and an attacker that compromised their server are able to see this information. Therefore, they know when no one is at your home. Since the central server is able to send commands to your alarm system, the PrivateEyePi guys and an attacker is also able to do it. Hence, these strangers are able to disarm your alarm system.
Now let us take a look at concrete attack scenarios.
We are secure. We used encryption!
First we take a look at the connection between the home alarm system and the central server. These connections can be found in the alarmfunctionsr.py
file. Both are done in the same way. A https
URL is visited with GET parameters that either update the data on the central server or request data. This URL also contains the username and password of your user account. Well, since they are using an encrypted connection to their central server, everything is fine and we can go to the next section? Not exactly. Yes, they are setting up an encrypted connection. But they are using urllib2
for it. And urllib2 does not validate the certificate offered by the server by default. This option has to be manually set. And yes, there is a proposal
to activate this check by default. But I do not know if it is already implemented. I checked it with Python 2.7.6 and it does not verify the certificate. Hence, when an attacker is able to hijack the connection he can just offer any certificate and the home alarm system will accept it. Therefore, an attacker does not necessarily have to compromise the central server, he just needs to hijack the connection (although I guess the first option might be easier).
I spy with my little eye ... uff, why am I suddenly blind?
The home alarm system gets its sensor configuration from the central server. Therefore, the central server can just send an empty or false configuration to the home alarm system. And suddenly ... the home alarm system is blind. This way an attacker is able to disable the alarm system completely or just remove specific sensors (to be a little bit sneakier).
Why does this work? Well, in the file alarm.py
function enters an infinity loop which polls the GPIO states of the sensors. And this infinity loop polls also the current configuration from the central server and applies it locally any 600 seconds.
Look mum no hands!
One interesting command is the /PHOTO
command. This attack only works if a webcam is connected to the home alarm system and the SMTP server is configured to send eMails. Well, what does it do? The command tells the home alarm system to send an eMail to the configured eMail addresses with a picture taken by the webcam. The fun part is, the eMail addresses this message is sent to are also configured via the central server. Before the home alarm system sends this message, it asks the central server for the eMail addresses. Therefore, the attacker can just send his eMail address to the home alarm system and gets a picture taken by the webcam. No one else is getting this message. So it is kind of sneaky
If you want to see it for yourself, you can find it in the alarmfunctionsr.py
is called when a message is received from the central server. This calls SendEmailAlertFromRule()
with all the necessary arguments to take a picture. A thread is started that executes SendEmailAlertThread()
. This function gets the eMail addresses from the central server with GetDataFromHost()
and the used arguments. Then it builds the message and sends it to the received eMail addresses.
This is a really cool attack, because it can actually damage the hardware. The command /RELAYON
is responsible for it. With this command, a GPIO pin number and the option HIGH or LOW is also received. What does it do? Well, it sets the given GPIO pin number as output pin and sets it to either HIGH or LOW. In the case of RaspberryPi, HIGH means it sets up 3.3V. This can be used for two attacks. First, we can trigger a sensor that is configured on our home alarm system. This can be done by setting the GPIO pin to a HIGH state that is also used as an input GPIO pin of a sensor. Second, if the sensor (or other hardware) that is connected to a GPIO pin that is set to a HIGH output is not built well, it can be damaged (I asked an electronics guy because I was not sure, and he told me it can happen). Not so often that we see a software attack that can actually damage hardware
Again, this can be found in the alarmfunctionsr.py
is called when a message is received from the central server. This calls SwitchRelay()
which sets the GPIO pin to an output port and to HIGH/LOW.
I am not a professional programmer. Therefore, I do not like to say stuff like "Uh, look at this shady code!". But there are some constructs in the PrivateEyePi code I wanted to show. Let us just look at the function ProcessActions()
for x in ActionList:
The argument ActionList
has the type list
. Well, what does this function do? It iterates over the ActionList
elements and checks if they contain a specific known command (more specifically, if the first element of this list contains a specific command). In the end, the function returns FalseInd
Let us take a closer look at FalseInd
. Do you see it? In the beginning this variable is set to True
. If the command /FALSE
is received, it is set to False
. When the command /SIREN
is received, it is set to 3
. Dafuq?! I know that Python allows you to do this. But this does not mean that you should do it (like goto
in C). If you use a variable as a boolean, stick to it and do not change it to an integer.
Also, each list element x
should be removed from the ActionList
(at least, I think this is what it should do). Hence, the line x.remove
. Well, x
is a list and has therefore the function remove()
. But first of all, it is not invoked by the line x.remove
. This just returns you the address of the function. And second, remove()
takes an argument. Therefore, this line does actually nothing and does not make any sense.
Another thing you often see in the code is something like the following:
# Routine to continuously poll the IO ports on the Raspberry Pi
def NotifyHostEvent(z, status):
I do not know why they are using global variables for everything. Hell, even the whole configuration is done with global variables. And the file is called globals.py
. And because of the import globals
in the beginning of each file, this has strange effects on the syntax highlighting of an editor. Because globals()
is a built-in
function in Python, the editor highlights any use of the configuration variables like globals.LCDAlarmActivity
. Definitely another thing you do not do when programming in Python.
This was a non-exhaustive list of security issues I found in PrivateEyePi. Note that I did non take a detailed dive into any file. Only the parts that work directly with data from the central server I looked into (and I did not really track the complete data-flow ... there could be a lot more to find). Well, as you have seen these are the reasons I would not recommend using PrivateEyePi. In my opinion, it is totally broken by design. The complete alarm system can be controlled by the PrivateEyePi guys and if the server is compromised, an attacker can so too. Additionally, the design allows an attacker to actually damage the connected hardware.
The PrivateEyePi authors contacted me and said that they fixed a lot of issues in their current version. The following is a summary of changes they send me. However, I like to stress that this is only a list of things the authors send to me. I did not check the code if these changes were actually made:
- User password encryption in database and password are no longer passed from server to client.
- Devices and sensors connecting to PrivateEyePi server now use a token to authenticate, not email and password. The token cannot be use used to login to the website. The token is only displayed to the user once and never transmitted from the server to the web page after the initial creation.
- Alarm activation/deactivation can only be done through the website.
- Remote control for GPIO, RF devices and web cam photo function configured off by default through a new setting AllowExternalControl.
- Installation script does not use chmod 777 anymore and removed sudo where it is not necessary.
- Python 2.7.9. now validates certificates by default.
- Cleaned up use of FalseInd Boolean variable to store status beyond True/False.
- Cleaned up remove function from ActionList.