You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 30 Next »

The EOF macro represents a negative value that is used to indicate that the file is exhausted and no data remains when reading data from a file. EOF is an example of an in-band error indicator. In-band error indicators are problematic to work with, and the creation of new in-band-error indicators is discouraged by ERR02-C. Avoid in-band error indicators.

The byte I/O functions fgetc(), getc(), and getchar() all read a character from a stream and return it as an int. (See STR00-C. Represent characters using an appropriate type.) If the stream is at the end of the file, the end-of-file indicator for the stream is set and the function returns EOF. If a read error occurs, the error indicator for the stream is set and the function returns EOF. If these functions succeed, they cast the character returned into an unsigned char.

Because EOF is negative, it should not match any unsigned character value. However, this is only true for implementations where the int type is wider than char. On an implementation where int and char have the same width, a character-reading function can read and return a valid character that has the same bit-pattern as EOF. This could occur, for example, if an attacker inserted a value that looked like EOF into the file or data stream to alter the behavior of the program.

The C Standard requires only that the int type be able to represent a maximum value of +32767 and that a char type be no larger than an int. Although uncommon, this situation can result in the integer constant expression EOF being indistinguishable from a valid character; that is, (int)(unsigned char)65535 == -1. Consequently, failing to use feof() and ferror() to detect end-of-file and file errors can result in incorrectly identifying the EOF character on rare implementations where sizeof(int) == sizeof(char).

This problem is much more common when reading wide characters. The fgetwc(), getwc(), and getwchar() functions return a value of type wint_t. This value can represent the next wide character read, or it can represent WEOF, which indicates end-of-file for wide character streams. On most implementations, the wchar_t type has the same width as wint_t, and these functions can return a character indistinguishable from WEOF.

In the UTF-16 character set, 0xFFFF is guaranteed not to be a character, which allows WEOF to be represented as the value -1. Similarly, all UTF-32 characters are positive when viewed as a signed 32-bit integer. All widely used character sets are designed with at least one value that does not represent a character. Consequently, it would require a custom character set designed without consideration of the C programming language for this problem to occur with wide characters or with ordinary characters that are as wide as int.

The C Standard feof() and ferror() functions are not subject to the problems associated with character and integer sizes and should be used to verify end-of-file and file errors for susceptible implementations [Kettlewell 2002]. Calling both functions on each iteration of a loop adds significant overhead, so a good strategy is to temporarily trust EOF and WEOF within the loop but verify them with feof() and ferror() following the loop.

Noncompliant Code Example

This noncompliant code example loops while the character c is not EOF:

#include <stdio.h>
 
void func(void) {
  int c;
 
  do {
    c = getchar();
  } while (c != EOF);
}

Although EOF is guaranteed to be negative and distinct from the value of any unsigned character, it is not guaranteed to be different from any such value when converted to an int. Consequently, when int has the same width as char, this loop may terminate prematurely.

Compliant Solution (Portable)

This compliant solution uses feof() to test for end-of-file and ferror() to test for errors:

#include <stdio.h>

void func(void) {
  int c;

  do {
    c = getchar();
  } while (c != EOF);
  if (feof(stdin)) {
    /* Handle end of file */
  } else if (ferror(stdin)) {
    /* Handle file error */
  } else {
    /* Received a character that resembles EOF; handle error */
  }
}

Noncompliant Code Example (Nonportable)

This noncompliant code example uses an assertion to ensure that the code is executed only on architectures where int is wider than char and EOF is guaranteed not to be a valid character value. However, this code example is noncompliant because the variable c is declared as a char rather than an int, making it possible for a valid character value to compare equal to the value of the EOF macro when char is signed because of sign extension:

#include <assert.h>
#include <limits.h>
#include <stdio.h>

void func(void) {
  char c;
  static_assert(UCHAR_MAX < UINT_MAX, "FIO34-C violation");

  do {
    c = getchar();
  } while (c != EOF);
}

Assuming that a char is a signed 8-bit type and an int is a 32-bit type, if getchar() returns the character value '\xff (decimal 255), it will be interpreted as EOF because this value is sign-extended to 0xFFFFFFFF (the value of EOF) to perform the comparison. (See STR34-C. Cast characters to unsigned char before converting to larger integer sizes.)

Compliant Solution (Nonportable)

This compliant solution declares c to be an int. Consequently, the loop will terminate only when the file is exhausted.

#include <assert.h>
#include <stdio.h>
#include <limits.h>

void func(void) {
  int c;
  static_assert(UCHAR_MAX < UINT_MAX, "FIO34-C violation");

  do {
    c = getchar();
  } while (c != EOF);
}

Noncompliant Code Example (Wide Characters)

In this noncompliant example, the result of the call to the C standard library function getwc() is stored into a variable of type wchar_t and is subsequently compared with WEOF:

#include <stddef.h>
#include <stdio.h>
#include <wchar.h>

enum { BUFFER_SIZE = 32 };

void g(void) {
  wchar_t buf[BUFFER_SIZE];
  wchar_t wc;
  size_t i = 0;

  while ((wc = getwc(stdin)) != L'\n' && wc != WEOF) {
    if (i < (BUFFER_SIZE - 1)) {
      buf[i++] = wc;
    }
  }
  buf[i] = L'\0';
}

This code suffers from two problems. First, the value returned by getwc() is immediately converted to wchar_t before being compared with WEOF. Second, there is no check to ensure that wint_t is wider than wchar_t. Both of these problems make it possible for an attacker to terminate the loop prematurely by supplying the wide-character value matching WEOF in the file.

Compliant Solution (Portable)

This compliant solution declares c to be a wint_t to match the integer type returned by getwc(). Furthermore, it does not rely on WEOF to determine end-of-file definitively.

#include <stddef.h>
#include <stdio.h>
#include <wchar.h>
 
enum {BUFFER_SIZE = 32 }

void g(void) {
  wchar_t buf[BUFFER_SIZE];
  wint_t wc;
  size_t i = 0;
  
  while ((wc = getwc(stdin)) != L'\n' && wc != WEOF) {
    if (i < BUFFER_SIZE - 1) {
      buf[i++] = wc;
    }
  }

  if (feof(stdin) || ferror(stdin)) {
   buf[i] = L'\0';
  } else {
    /* Received a wide character that resembles WEOF; handle error */
  }
}

Exceptions

FIO34-C-EX1: A number of C functions do not return characters but can return EOF as a status code. These functions include fclose(), fflush(), fputs(), fscanf(), puts(), scanf(), sscanf(), vfscanf(), and vscanf(). These return values can be compared to EOF without validating the result.

Risk Assessment

Incorrectly assuming characters from a file cannot match EOF or WEOF has resulted in significant vulnerabilities, including command injection attacks. (See the *CA-1996-22 advisory.)

Rule

Severity

Likelihood

Remediation Cost

Priority

Level

FIO34-C

High

Probable

Medium

P12

L1

 

Automated Detection

Tool

Version

Checker

Description

CodeSonar8.1p0LANG.CAST.COERCECoercion alters value
Compass/ROSE

 

 

 

Coverity6.5

CHAR_IO

Identifies defects when the return value of fgetc()getc(), or getchar() is incorrectly assigned to a char instead of an int. Coverity Prevent cannot discover all violations of this rule, so further verification is necessary

ECLAIR1.2

CC2.FIO34

Partially implemented

Klocwork2024.1CWARN.CMPCHR.EOF 
LDRA tool suite9.7.1662 S
Fully implemented
Parasoft C/C++test9.5MISRA2004-10_1_dPartially implemented
Splint3.1.1

 

 

 PRQA QA-C++ 4.23051, 3001, 3010, 3137, 3717  

Related Vulnerabilities

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

Related Guidelines

Bibliography

[Kettlewell 2002]Section 1.2, "<stdio.h> and Character Types"
[NIST 2006]SAMATE Reference Dataset Test Case ID 000-000-088
[Summit 2005]Question 12.2

 


  • No labels