Security for Web Applications

In the modern web environment, we must assume that any service or resource that is made available on the web will be subject to attack by malicious people. Any web server will be tested for weaknesses, any web application will have a range of attempts at getting past security and any data that we store will be the target of someone trying to find interesting or valuable information. What is more, the people doing the attacking are clever and well funded - often cleverer and better funded than the people writing and maintaining the web applications. As web developers we need to be aware of the vulnerabilities that make our applications and servers susceptible to attack. We need to use tools and learn coding techniques to help prevent at least those exploits that are well understood. We need to understand how vulnerabilities come about so that we have a chance of seeing what the next wave of attacks might target. This chapter gives a brief introduction to the most common attacks on web applications and some ideas of how to prevent these.

There are many ways to classify security issues. I've chosen three categories here but there would be other ways to group these. The categories are:

Server Security

A web server is a computer connected to the Internet that is running software to interpret HTTP requests and generate responses. Any computer connected to the Internet is a target for attack since it might hold interesting or valuable information or it might provide a gateway to other computers that hold interesting or valuable information. If you are running a web server, you need to be aware of how your system might be attacked and what the appropriate responses are.

In a sense, all vulnerabilities for web applications come under this banner of Server Security. However it is useful to discuss problems with web applications separately since this is what affects the code that you write. The problems discussed here are in system and server software that you will probably install from third parties - or which might be run for you by your service provider.

Attack Vectors - Open Channels

Any attack on a server must happen over some kind of communication channel. The easiest way to secure a server would be to disconnect it from the network, however this makes it a lot less useful. A web server will have at least one open Internet port - usually port 80 - and will have software listening for requests on that port. Quite often, the same server has other open ports for other services; for example, ftp for file upload, smtp for mail transfer, ssh for secure communication. Each of these open ports presents a possible security risk, a means by which an attacker could try to gain access to the server.

Each open port that provides a service has a piece of software running on the server listening for requests and handling communications. For the web server this is HTTP traffic on port 80, for mail, there might be an SMTP server running on port 25 and so on. Each of these pieces of software accepts requests and processes them in some way and they all have some kind of access to the local machine, for example to read and write files, access a database or run jobs that take up CPU time. It is these programs that provide the way in for an attacker.

Since running software listening to open ports on our server makes it vulnerable it makes sense that we should run as few of these as is necessary. This means that if we don't need a mail service we shouldn't run the SMTP server etc. This might sound obvious but until recently, computers would come configured with many services running by default making them very vulnerable. Common practice now is that the default installation of an operating system has only the minimum services exposed and the system administrator needs to explicitly install and enable any further services. Note that this goes for desktop computers too if they are directly connected to the Internet.

Vulnerabilities

Once there is an open channel of communication with the server, we still need something to go wrong before an attacker can gain some kind of access to the system. In the normal course of events, the SMTP server does its job, transferring mail to clients elsewhere on the Internet. To gain access to the system, an attacker must take advantage of a bug in this software that allows them to do something outside of its design parameters.

All server software receives input from the internet and processes it in some way. This leads to the source of all vulnerabilities which must stem from some property of that input that is outside of the expectations of the programmer.

A very simple kind of attack is a Denial of Service (DOS) attack where the aim is to overload the server so that legitimate clients can't make use of it. There are many ways to achieve this but a common approach is to flood the server with input so that it either blocks the network due to too much traffic or keeps the server so busy processing requests that it can't keep up with the flow. For example, we could send thousands of large email messages to an SMTP server in very quick succession, this might fill up the available temporary storage on the server or force the server to spend lots of time parsing and processing each message. Another kind of DOS attack is to try to crash the server software, for example by sending input that is badly formed. It could be, for example, that if the SMTP server gets binary data when it is expecting text, it will crash rendering the server unusable.

Another common kind of attack is a Buffer Overflow attack. In this case, the server software has been written to use a fixed size buffer to process input; for example, it has set aside an array of 100 characters to receive the email address of a message recipient. In a langauge like C or C++, it has been common to allocate such fixed-size arrays in code directly and the result is that the storage for the email address sits next to the storage used for the program code. Another way to allocate space, which can be used in C/C++ and is always used in Java and C#, is to make space in a different area of memory (called the heap), away from program code. The problem arises when the software reads in data from the input and stores it in this buffer. If the size of the input isn't checked it is possible for the input to overflow the size of the buffer that was allocated. If the buffer is located in the same place as program code, the overflow bytes are written on top of the program code. In this way, the attacker is able to send input containing program code (which has to be machine code for the computer the software is running on) which is written into the running program and which may then be executed. Once the attacker can get their own code executed, they can start to take advantage of the server and find a way to get themselves in or get data out of the system.

An early example of a buffer overflow attack was the Morris Worm which was perhaps the first major Internet security breach back in 1988. The Worm was a piece of self-replicating code which would send a copy of itself to any computer it could gain access to. One of the ways that it propogated itself was via a network service called fingerd which was used to find out information about users on other systems. The fingerd server was a simple application that listened for requests and processed them but it was written with a fixed size buffer of 512 bytes and the program used the gets system call to read input without checking the size. The worm would send 536 characters, the last 24 of which would be instructions for the remote computer that allowed it to run an arbitrary script. Once it had this access it would upload a copy of itself to the remote machine and the cycle was repeated. The Morris Worm quickly brought the Internet to a crawl because it kept re-infecting the same machines and overloading the processor. It was therefore a kind of Denial of Service attack that was perpitrated via a Buffer Overflow vulnerability.

Exercises

  1. The Code Red Worm is another example of a security vulnerability caused by a buffer overflow bug. Find out what systems were affected by this attack, how the attack would be carried out and what effect the worm had on systems that it infected.
  2. A Trojan Horse is a piece of software that appears to do one job but in fact does something else behind the scenes. Find an example of a security exploit that was carried out via a Trojan Horse. What systems were affected by this exploit and how did it operate?

The buffer overflow attack allows code to be executed on the target machine but there are other ways to achieve this. For example, if the attacker can get an application installed on the server that seems to do something useful but in fact serves to provide access for them (a Trojan Horse). Once the attacker can run code on the server, they can start to look for something to do. One target might be the passwords of legitimate users, another might be to launch an attack on another system.