CERT
Skip to end of metadata
Go to start of metadata

The size_t type is the unsigned integer type of the result of the sizeof operator. Variables of type size_t are guaranteed to be of sufficient precision to represent the size of an object. The limit of size_t is specified by the SIZE_MAX macro.

The type size_t generally covers the entire address space. The C Standard, Annex K (normative), "Bounds-checking interfaces," introduces a new type, rsize_t, defined to be size_t but explicitly used to hold the size of a single object [Meyers 2004]. In code that documents this purpose by using the type rsize_t, the size of an object can be checked to verify that it is no larger than RSIZE_MAX, the maximum size of a normal single object, which provides additional input validation for library functions. See STR07-C. Use the bounds-checking interfaces for string manipulation for additional discussion of C11 Annex K.

Any variable that is used to represent the size of an object, including integer values used as sizes, indices, loop counters, and lengths, should be declared rsize_t, if available. Otherwise, it should be declared size_t.

Noncompliant Code Example

In this noncompliant code example, the dynamically allocated buffer referenced by p overflows for values of n > INT_MAX:

Signed integer overflow causes undefined behavior. The following are two possible conditions under which this code constitutes a serious vulnerability:

sizeof(size_t) == sizeof(int)

The unsigned n may contain a value greater than INT_MAX. Assuming quiet wraparound on signed overflow, the loop executes n times because the comparison i < n is an unsigned comparison. Once i is incremented beyond INT_MAX, i takes on negative values starting with (INT_MIN). Consequently, the memory locations referenced by p[i] precede the memory referenced by p, and a write outside array bounds occurs.

sizeof(size_t) > sizeof(int)

For values of n where 0 < n <= INT_MAX, the loop executes n times, as expected.

For values of n where INT_MAX < n <= (size_t)INT_MIN, the loop executes INT_MAX times. Once i becomes negative, the loop stops, and i remains in the range 0 through INT_MAX.

For values of n where (size_t)INT_MIN < n <= SIZE_MAX, i wraps and takes the values INT_MIN to INT_MIN + (n - (size_t)INT_MIN - 1). Execution of the loop overwrites memory from p[INT_MIN] through p[INT_MIN + (n - (size_t)INT_MIN - 1)].

Compliant Solution (C11, Annex K)

Declaring i to be of type rsize_t eliminates the possible integer overflow condition (in this example). Also, the argument n is changed to be of type rsize_t to document additional validation in the form of a check against RSIZE_MAX:

Noncompliant Code Example

In this noncompliant code example, the value of length is read from a network connection and passed as an argument to a wrapper to malloc() to allocate the appropriate data block. Provided that the size of an unsigned long is equal to the size of an unsigned int, and both sizes are equal to or smaller than the size of size_t, this code runs as expected. However, if the size of an unsigned long is greater than the size of an unsigned int, the value stored in length may be truncated when passed as an argument to alloc().

Compliant Solution (C11, Annex K)

Declaring both length and the blocksize argument to alloc() as rsize_t eliminates the possibility of truncation. This compliant solution assumes that read_integer_from_network() and read_network_data() can also be modified to accept a length argument of type pointer to rsize_t and rsize_t, respectively. If these functions are part of an external library that cannot be updated, care must be taken when casting length into an unsigned long to ensure that integer truncation does not occur.

Risk Assessment

The improper calculation or manipulation of an object's size can result in exploitable vulnerabilities.

Recommendation

Severity

Likelihood

Remediation Cost

Priority

Level

INT01-C

Medium

Probable

Medium

P8

L2

Automated Detection

Tool

Version

Checker

Description

CodeSonar4.4

LANG.TYPE.BASIC

Basic numerical type used
Compass/ROSE

 

 

Can detect violations of this recommendation. In particular, it catches comparisons and operations where one operand is of type size_t or rsize_t and the other is not

Splint3.1.1

 

 

Related Vulnerabilities

Search for vulnerabilities resulting from the violation of this rule on the CERT website.

Related Guidelines

Bibliography

 


13 Comments

  1. Does it matter that there is no guarantee from the first sample compliant code that the returned string is null terminated, because there is no guarantee that the input string is the length specified and so on?

    Also, doesn't the call: p = copy(20, "hi there"); invoke undefined behaviour?  It is accessing 11 bytes beyond the end of the string.  This is the converse of the 'not null terminated' problem in some ways.

     The second compliant solution again has a null-termination problem.  In the fragment where the result is not used, this is "OK", but the example might be used elsewhere, and then it would matter.

    The third example is also vulnerable on an LP32 implementation (as well as LP64).  It also allocates space for a null terminator but does not visibly ensure that it is present. 

  2. >> Once i >> > INT_MAX, i takes on
    >> negative values starting with (INT_MIN). Consequently, the memory locations

    Not really. Once i overflows, UB happens and there is nothing guaranteed.

    Update: okay, in that same page (INT01-A) I found this code which
    claims to be secure & conforming

    from: http://groups.google.com/group/comp.lang.c/browse_thread/thread/1ad64a62aa6ec2d7/84c2ed6e56338aad?hl=en#84c2ed6e56338aad

    1. OK, I think I addressed both of these comments.

  3. (1) rsize_t is like errno_t (see my comments there) in that some work is needed to use the name in portable code.

    (2) rsize_t is pointless anyway: the purpose of size_t is merely to provide a type that is guaranteed to be big enough to index every byte of an object, and rsize_t says that  that is also its purpose.  The programmer only needs one type for use in declaring indices etc. and size_t is meant for that.  The idea that functions can perform better validation is bogus since if RSIZE_MAX is the largest that an object can be, then that value should be used for SIZE_MAX, and really useful bounds checking must be for the actual object size, not for the largest possible object size.  rsize_t was a mistake and should not be blessed by these guidelines.

  4. Well, if you want it to be bashed(smile)

    Both non-compliant examples are silly because several listed problems are buggy independently of the use of the variables for sizes/indexes, and thus independently of this recommendation:

    Example #1: In for(i = 0; i<n; ++i), 'i' should normally be of the same type as n and must be able to hold the max possible value of n.  Also it looks like a string so you should likely allocate one more byte and append a \0.

    Example #2: Passing unsigned long to unsigned int truncates the value, unless you know it's <= UINT_MAX.  And the integer wraparound test 'length + 1 == 0' is unnecessarily hacky and depends on the type being unsigned, length == ULONG_MAX would be more readable.

    If the network functions called in example #2 take unsigned longs, the "compliant solution" is broken when it instead passes rsize_t*, and when passing rsize_t without first checking that length <= ULONG_MAX.

    Which also illustrates one reason I'd personally sometimes let another guideline trump this one (and some others): Avoid integer type conversions, implicit as well as explicit, by declaring variables what will "meet" each other to have the same type.  Integer conversions often need to be checked for max/min limits to be safe.  So if some size value will have to be passed/declared as int (for printf("%*..")) or ssize_t (returned from Unix write()), the code may be cleaner if it uses those types for sizes.  There may be no point it having part of it support bigger objects anyway.  Though even then (r)size_t is sometimes just as simple to use since one in any case must ensure that the sizes involved can fit in the smallest integer type they will be stored/passed in.

    The text compares sizeof(long) with sizeof(int) when it should compare ULONG_MAX with UINT_MAX.  The results don't have to be the same, though of course they normally are.

    And just to bring up a different mess, with the first para (about size_t): It has been argued on comp.std.c that an object can be larger than the max value size_t:  size_t is merely defined as the return type of sizeof, which must return the correct size of an object.  Thus an object may be larger than SIZE_MAX as long as you don't use sizeof on it.  (I think there is/was a defect report on that.)

    1. I was mostly concerned with the the order/structure of the wording, not the meaning (which I have not changed)

      Ex #1 strings: fixed
      Ex #2 length +1: fixed
      Ex #2 rsize_t: not sure what you mean here
      Ex #2 sizeof vs ULONG_MAX: fixed

      1. Err, actually I picked on the "wrong error" with ex#1 length - being a string copy function I expected it to take a length rather than size argument, so the problem was just that the API is surprising.  And length rather than lenght+1 may be OK for unsigned char* (ex#2).

        Ex #2 rsize_t: Point is, the noncompliant code passes (pointer to) unsigned long to the network functions, which is hopefully because these functions take (pointer to) unsigned long.  The compliant solution changes the caller but there is no mention of changing the network functions to match.  If you can change the network functions without too much hassle, certainly do that (and change all callers). But if it's an external API you can't.  In either case, it's the network functions and not the example functions which are the source of the trouble and which originally break this recommendation.{{
        }}

        1. Ex #2 rsize_t: added a note about needing to update the network functions, thanks (smile)

  5. Is this rule supposed to apply to array indices, too?

    eg, good or bad:

    This code is definitely bad:

    because i is always nonnegative.

  6. [...]
    sizeof(size_t) > sizeof(int)

    Similar behavior as the case above occurs for values of n <= UINT_MAX. For values of n > UINT_MAX, the expression ++i will wrap around to zero before the condition i < n ever evaluates to false. This causes all memory within [INT_MIN, INT_MAX] from the beginning of the output buffer to be overwritten in an infinite loop.

    This is wrong. For 0<n<=INT_MAX, the loop executes n times, as expected. For n>INT_MAX, the loop executes INT_MAX+1 times. For each iteration, i is in the range 0 through INT_MAX. Once i becomes negative, the loop will stop, because a negative value for i converts into SIZE_MAX+1+i. Since sizeof(size_t)>sizeof(int), SIZE_MAX is well above INT_MAX*2, as well as SIZE_MAX+1+INT_MIN, therefore the size_t values i converts to are always greater than values of n in the range INT_MAX+1 to INT_MAX*2. I've verified the above on a compiler that has 16-bit int and 32-bit size_t.

    1. Good catch, thanks. I've revised the text to closely match what you have stated above.

    2. Suppose n == SIZE_MAX + 1 + INT_MIN + 50. Then when i wraps around and takes the values INT_MIN to INT_MIN + 49, the loop will scribble on p[INT_MIN] through p[INT_MIN + 49]. Then the loop will terminate.

      1. I've revised this one last time (I hope).