String to Number Conversion in C Takes its Toll

Converting a string to a number in C is no simple affair. Many of you may have heard of atoi(3); one of the ways to convert a string to a number. Unfortunately, modern thinking says that atoi should never be used, and so it’s use is discouraged. Instead, we now have strtol. 1

What’s wrong with atoi?

The reason for discouraging use of atoi stems from the fact that there is no way to detect if overflow or underflow has occurred, and no way to check if the entire string has been converted (aka there’s no way to detect atoi("123garbage")). Consider the following code:

// 2^32+1, assuming 32-bit int
const char *uintmax_plus_one = "4294967297";
printf("%d\n", atoi(uintmax_plus_one));

When run instead of printing 4294967296 as expected, this program will print 1! The vast majority of programs do not check for or properly handle this case, and so you can end up with situations like the following:

// 2^32+1, assuming 32-bit int
const char *uintmax_plus_one = "4294967297";
malloc(atoi(uintmax_plus_one));

Now we’re allocating far less memory than we expected. This problem can quickly become the source of an integer overflow vulnerability.

Enter strtol!

In order to do the conversion safely, we instead should use strtol. It is unfortunately quite difficult to call this function properly. Consider the following documentation pulled from the BSD Library Functions Manual’s section on strtol:

The strtol(), strtoll(), strtoimax(), and strtoq() functions return the result of the conversion, unless the value would underflow or overflow. If no conversion could be performed, 0 is returned and the global variable errno is set to EINVAL (the last feature is not portable across all platforms). If an overflow or underflow occurs, errno is set to ERANGE and the function return value is clamped according to the following table.

Functionunderflowoverflow
strtol()LONG_MINLONG_MAX
strtoll()LLONG_MINLLONG_MAX
strtoimax()INTMAX_MININTMAX_MAX
strtoq()LLONG_MINLLONG_MAX

Based on this, the two ways to check for an overflow are to check if strtol returns 0 or to check if errno is set to ERANGE. There’s another simple case where strtol returns 0 specifically if the input string to strtol is "0", so in order to accurately detect range errors, we must check for ERANGE. This isn’t quite so simple either however, as if no error has occurred, strtol will not change the value of errno. If this happens, and some previous code has set errno to ERANGE you will erroneously think that a range error has occurred. So now in order to check for range errors you must reset errno to a value that indicates that no error has occurred. Now, what value is that? Thankfully, POSIX.1-2008 has considered this possibility, and defined that “No function in this volume of POSIX.1-2008 shall set errno to 0,” meaning that no error in all of POSIX will have the value 0. So now we can do the following:

errno = 0;
long val = strtol(str, NULL, 10);
if (errno == ERANGE) {
	switch(val) {
	case LONG_MIN:
		// underflow
		break;
	case LONG_MAX:
		// overflow
		break;
	default:
		assert(false); // impossible
	}
} else if (errno != 0) {
	// something else happened. die die die
}

But wait, there’s more! This works fine for detecting range errors, but fails to detect garbage at the end of the string. Thankfully, strtol lets us handle this too via it’s char **endptr argument.

If endptr is not NULL, strtol() stores the address of the first invalid character in *endptr. If there were no digits at all, however, strtol() stores the original value of str in *endptr. (Thus, if *str is not `\0’ but **endptr is `\0’ on return, the entire string was valid.)

Using this, we can detect if there was garbage at the end of the string by passing in a char ** value for endptr.

errno = 0;
char *endptr;
long val = strtol(str, &endptr, 10);
if (errno == ERANGE) {
	switch(val) {
	case LONG_MIN:
		// underflow
		break;
	case LONG_MAX:
		// overflow
		break;
	default:
		assert(false); // impossible
	}
} else if (errno != 0) {
	// something else happened. die die die
} else if (*endptr != '\0') {
	// garbage at end of string
}

And now we’ve turned a relatively simple one-line call to atoi into twenty lines of code. FML

Enter strtonum

The great folks over at OpenBSD have made a nice replacement for strtol which fixes all of the issues discussed. Called strtonum, the function doesn’t allow trailing characters and makes it easy to determine if a range error has occurred. The one drawback is that strtonum is an OpenBSD extension, and so is not found in any standard. If you want to use strtonum on other platforms, you can grab the source here.

  1. strtol, short for “string to long” is only one of a few of such functions for converting from a string to a number. Also in this family are strtoimax, strtoll and strtoq.

blog comments powered by Disqus