Floating-point numbers can take on two classes of exceptional values; infinity and NaN (not-a-number). These values are returned as the result of exceptional or otherwise unresolvable floating-point operations. (See also FLP32-C. Prevent or detect domain and range errors in math functions.) Additionally, they can be directly input by a user by
scanf or similar functions. Failure to detect and handle such values can result in undefined behavior.
NaN values are particularly problematic because the expression NaN == NaN (for every possible value of NaN) returns false. Any comparisons made with NaN as one of the arguments returns false, and all arithmetic functions on NaNs simply propagate them through the code. Hence, a NaN entered in one location in the code and not properly handled could potentially cause problems in other, more distant sections.
Formatted-input functions such as
scanf will accept the values
NAN (case insensitive) as valid inputs for the
%f format specification, allowing malicious users to feed them directly to a program. Programs should therefore check to ensure that all input floating-point values (especially those controlled by the user) have neither of these values if doing so would be inappropriate. The
<math.h> library provides two macros for this purpose:
isinf macro tests an input floating-point value for infinity.
isinf(val) is nonzero if
val is an infinity (positive or negative), and 0 otherwise.
isnan tests if an input is NaN.
isnan(val) is nonzero if
val is a NaN, and 0 otherwise.
If infinity or NaN values are not acceptable inputs in a program, these macros should be used to ensure they are not passed to vulnerable functions.
Noncompliant Code Example
This noncompliant code example accepts user data without first validating it:
This can be a problem if an invalid value is entered for
val and subsequently used for calculations or as control values. The user could, for example, input the strings
"NAN" (case insensitive) on the command line, which would be parsed by
scanf into the floating-point representations of infinity and NaN. All subsequent calculations using these values would be invalid, possibly crashing the program and enabling a denial-of-service attack.
Here, for example, entering
val would force
currentBalance to also equal
"nan", corrupting its value. If this value is used elsewhere for calculations, every resulting value would also be a NaN, possibly destroying important data.
The following code was run on 32-bit GNU Linux using the GCC 3.4.6 compiler. On this platform,
FLT_MAX has the value 340282346638528859811704183484516925440.000000.
The following table shows the value of
currentBalance returned for various arguments:
As this example demonstrates, the user can enter the exceptional values
NaN, as well as force a float's value to be infinite, by entering out-of-range floats. These entries subsequently corrupt the value of
currentBalance. So by entering exceptional floats, an attacker can corrupt the program data, possibly leading to a crash.
This compliant solution first validates the input float before using it. The value is tested to ensure that it is neither an infinity nor a NaN.
Occasionally, NaN or infinity may be acceptable or expected inputs to a program. If this is the case, then explicit checks may not be necessary. Such programs must, however, be prepared to handle these inputs gracefully and not blindly use them in mathematical expressions where they are not appropriate.
Inappropriate floating-point inputs can result in invalid calculations and unexpected results, possibly leading to crashing and providing a denial-of-service opportunity.
|Supported: Astrée reports potential runtime error resulting from missing checks for exceptional values.|
|SEI CERT C++ Coding Standard||FLP04-CPP. Check floating-point inputs for exceptional values|
|CERT Oracle Secure Coding Standard for Java||FLP06-J. Check floating-point inputs for exceptional values|