In computer science, program analysis is the process of automatically analyzing the behavior of computer programs regarding a property such as correctness, robustness, safety and liveness. Program analysis focuses on two major areas: program optimization and program correctness. The first focuses on improving the program’s performance while reducing the resource usage while the latter focuses on ensuring that the program does what it is supposed to do.
Program analysis can be performed without executing the program (static program analysis), during runtime (dynamic program analysis) or in a combination of both.
Static program analysis
In the context of program correctness, static analysis can discover vulnerabilities during the development phase of the program. These vulnerabilities are easier to correct than the ones found during the testing phase since static analysis leads to the root of the vulnerability.
Due to many forms of static analysis being computationally undecidable, the mechanisms for doing it will in general not always terminate with the right answer â€" either because they sometimes return a false negative ("no problems found" when the code does in fact have problems) or a false positive, or because they never return the wrong answer but sometimes never terminate. Despite their limitations, the first type of mechanism might reduce the number of vulnerabilities, while the second can sometimes give strong assurance of the lack of a certain class of vulnerabilities.
Incorrect optimizations are highly undesirable. So, in the context of program optimization, there are two main strategies to handle computationally undecidable analysis:
- An optimizer that is expected to complete in a relatively short amount of time, such as the optimizer in an optimizing compiler, may use a truncated version of an analysis that is guaranteed to complete in a finite amount of time, and guaranteed to only find correct optimizations.
- A third-party optimization tool may be implemented in such a way as to never produce an incorrect optimization, but also so that it can, in some situations, continue running indefinitely until it finds one (which may never happen). In this case, the developer using the tool would have to stop the tool and avoid running the tool on that piece of code again (or possibly modify the code to avoid tripping up the tool).
However, there is also a third strategy that is sometimes applicable for languages that are not completely specified, such as C. An optimizing compiler is at liberty to generate code that does anything at runtime â€" even crashes â€" if it encounters source code whose semantics are unspecified by the language standard in use.
Control-flow
The purpose of control-flow analysis is to obtain information about which functions can be called at various points during the execution of a program. The collected information is represented by a control flow graph (CFG) where the nodes are instructions of the program and the edges represent the flow of control. By identifying code blocks and loops CFG becomes a starting point for compiler made optimizations.
Data-flow analysis
Data-flow analysis is a technique designed to gather information about the values at each point of the program and how they change over time. This technique is often used by compilers to optimize the code. One of the most known examples of data-flow analysis is taint checking which consists of considering all variables which contain user supplied data â€" which is considered "tainted", i.e. insecure â€" and preventing those variables from being used until they have been sanitized. This technique is often used to prevent SQL injection attacks. Taint checking can be done statically or dynamically.
Abstract interpretation
Abstract interpretation allows to extract information about a possible execution of a program without actually executing the program. This information can be used by compilers to look for possible optimizations or for certifying a program against certain classes of bugs.
Type systems
Type systems associate types to programs that fulfill certain requirements. Their purpose is to select a subset of programs of a language that are considered correct according to a property.
- Type checking â€" verify whether the program is accepted by the type system.
Type checking is used in programming to limit how a programming object is used and what can they do. This is done by the compiler or interpreter. Type checking can also help preventing vulnerabilities by ensuring that a signed value isn't attributed to an unsigned variable. Type checking can be done statically (at compile time), dynamically (at runtime) or a combination of both.
Static type information (either inferred, or explicitly provided by type annotations in the source code) can also be used to do optimizations, such as replacing boxed arrays with unboxed arrays.
Effect systems
Effect systems are formal systems designed to represent the effects executing a function or method can have. An effect codifies what is being done and with what it is being done â€" usually referred to as effect kind and region, respectively.
Model checking
Model checking refers to strict, formal and automated ways to check if a model (which in this context means a formal model of a piece of code, though in other contexts it can be a model of a piece of hardware) complies with a given specification. Due to the inherent finite state nature of code, and both the specification and the code being convertible into logical formulae, it is possible to check if the system violates the specification using efficient algorithmic methods.
Dynamic program analysis
Dynamic analysis can use runtime knowledge of the program to increase the precision of the analysis while also providing runtime protection but can only analyze a single execution of the problem and might degrade the program’s performance due to runtime checks.
Testing
Software should be tested to ensure its quality and that it performs as it is supposed to in a reliable manner, and that won’t create conflicts with other software that may function alongside it. The tests are performed by executing the program with an input and evaluating its behavior and the produced output. Even if no security requirements are specified, additional security testing should be performed to ensure that an attacker can’t tamper with the software and steal information, disrupt the software’s normal operations or use it as a pivot to attack its users.
Monitoring
Program monitoring records and logs different kinds of information about the program such as resource usage, events and interaction so it can be reviewed to find abnormal behavior or even pinpoint what caused the abnormal behavior. Furthermore it can be used to perform security audits. Automated monitoring of programs is sometimes referred to as runtime verification.
Program slicing
For a given subset of a program’s behavior, program slicing consists of reducing the program to the minimum form that still produces the selected behavior. The reduced program is called a “slice†and is a faithful representation of the original program within the domain of the specified behavior subset. Generally, finding a slice is an unsolvable problem, but by specifying the target behavior subset by the values of a set of variables, it is possible to obtain approximate slices using a data-flow algorithm. These slices are usually used by developers during debugging to locate the source of errors.
Further reading
- Hiralal Agrawal, Joseph R. Horgan, "Dynamic program slicing"
- Wang Chunlei, Zhao Gang, Dai Yiqi, "An Efficient Control Flow Security Analysis Approach for Binary Executables"
- Flemming Nielson, Hanne Riis Nielson, Chris Hankin (2005). "Principles of Program Analysis". Springer.
See also
- Automated code review
- Language-based security
- Profiling (computer programming)
- Program verification
