Difference between revisions of "Command Injection"

From OWASP
Jump to: navigation, search
m (minor spelling mistakes)
 
(29 intermediate revisions by 4 users not shown)
Line 1: Line 1:
 
{{Template:Attack}}
 
{{Template:Attack}}
 +
<br>
 +
[[Category:OWASP ASDR Project]]
  
==Description==
+
Last revision (mm/dd/yy): '''{{REVISIONMONTH}}/{{REVISIONDAY}}/{{REVISIONYEAR}}'''
  
Purpose of the command injection attack is to inject and execute commands specified by the attacker in the vulnerable application. In situation like this, application which executes unwanted system commands is like a pseudo system shell and the attacker may use it as any authorized system user. However commands are executed with the same privileges and environment as the application has. Command injection attacks are possible in most cases because of lack of correct input data validation, whose in addition can be manipulated by the attacker (forms, cookies, HTTP headers etc.).
+
==Description==
 +
The purpose of the command injection attack is to inject and execute commands specified by the attacker in the vulnerable application. In situation like this, the application, which executes unwanted system commands, is like a pseudo system shell, and the attacker may use it as any authorized system user. However, commands are executed with the same privileges and environment as the application has. Command injection attacks are possible in most cases because of lack of correct input data validation, which can be manipulated by the attacker (forms, cookies, HTTP headers etc.).
  
There is also different variant of the injection attack called "code injection".
+
There is a variant of the [[Code Injection]] attack. The difference with code injection is that the attacker adds his own code to the existing code. In this way, the attacker extends the  default functionality of the application without the necessity of executing system commands. Injected code is executed with the same privileges and environment as the application has.
The difference in code injection is that the attacker adds his own code to the existing one. The attacker extends this way the default functionality of the application without necessity of executing system commands. Injected code is executed with the same privileges and environment as application has.
+
  
 +
An OS command injection attack occurs when an attacker attempts to execute system level commands through a vulnerable application. Applications are considered vulnerable to the OS command injection attack if they utilize user input in a system level command.
  
 +
<!--==Risk Factors==
 +
TBD
 +
-->
 
==Examples ==
 
==Examples ==
  
Example 1
+
===Example 1===
  
The following code is wrapper around the UNIX command cat which prints the contents of a file to standard out. It is also injectable:
+
The following code is a wrapper around the UNIX command ''cat'' which prints the contents of a file to standard output. It is also injectable:
  
 
<pre>
 
<pre>
Line 52: Line 58:
 
unstosig.c              www*                    a.out*
 
unstosig.c              www*                    a.out*
 
format.c                strlen.c                useFree*
 
format.c                strlen.c                useFree*
catWrapper*            misnull.c              strlength.c               useFree.c               commandinjection.c      nodefault.c            trunc.c                writeWhatWhere.c
+
catWrapper*            misnull.c              strlength.c             useFree.c
 +
commandinjection.c      nodefault.c            trunc.c                writeWhatWhere.c
 +
</pre>
  
</pre>
 
 
If catWrapper had been set to have a higher privilege level than the standard user, arbitrary commands could be executed with that higher privilege.
 
If catWrapper had been set to have a higher privilege level than the standard user, arbitrary commands could be executed with that higher privilege.
  
Example 2
+
===Example 2===
  
The following simple program accepts a filename as a command line argument and displays the contents of the file back to the user. The program is installed setuid root because it is intended for use as a learning tool to allow system administrators in-training to inspect privileged system files without giving them the ability to modify them or damage the system.
+
The following simple program accepts a filename as a command line argument, and displays the contents of the file back to the user. The program is installed setuid root because it is intended for use as a learning tool to allow system administrators in-training to inspect privileged system files without giving them the ability to modify them or damage the system.
  
 
<pre>
 
<pre>
Line 72: Line 79:
 
Because the program runs with root privileges, the call to system() also executes with root privileges. If a user specifies a standard filename, the call works as expected. However, if an attacker passes a string of the form ";rm -rf /", then the call to system() fails to execute cat due to a lack of arguments and then plows on to recursively delete the contents of the root partition.
 
Because the program runs with root privileges, the call to system() also executes with root privileges. If a user specifies a standard filename, the call works as expected. However, if an attacker passes a string of the form ";rm -rf /", then the call to system() fails to execute cat due to a lack of arguments and then plows on to recursively delete the contents of the root partition.
  
Example 3
+
===Example 3===
  
The following code from a privileged program uses the environment variable $APPHOME to determine the application's installation directory and then executes an initialization script in that directory.
+
The following code from a privileged program uses the environment variable $APPHOME to determine the application's installation directory, and then executes an initialization script in that directory.
  
 
<pre>
 
<pre>
Line 89: Line 96:
 
</pre>
 
</pre>
  
As in Example 2, the code in this example allows an attacker to execute arbitrary commands with the elevated privilege of the application. In this example, the attacker can modify the environment variable $APPHOME to specify a different path containing a malicious version of INITCMD. Because the program does not validate the value read from the environment, by controlling the environment variable the attacker can fool the application into running malicious code.
+
As in Example 2, the code in this example allows an attacker to execute arbitrary commands with the elevated privilege of the application. In this example, the attacker can modify the environment variable $APPHOME to specify a different path containing a malicious version of INITCMD. Because the program does not validate the value read from the environment, by controlling the environment variable, the attacker can fool the application into running malicious code.
  
The attacker is using the environment variable to control the command that the program invokes, so the effect of the environment is explicit in this example. We will now turn our attention to what can happen when the attacker can change the way the command is interpreted.
+
The attacker is using the environment variable to control the command that the program invokes, so the effect of the environment is explicit in this example. We will now turn our attention to what can happen when the attacker changes the way the command is interpreted.
  
Example 4
+
===Example 4===
  
The code below is from a web-based CGI utility that allows users to change their passwords. The password update process under NIS includes running make in the /var/yp directory. Note that since the program updates password records, it has been installed setuid root.
+
The code below is from a web-based CGI utility that allows users to change their passwords. The password update process under NIS includes running ''make'' in the /var/yp directory. Note that since the program updates password records, it has been installed setuid root.
  
 
The program invokes make as follows:
 
The program invokes make as follows:
Line 102: Line 109:
 
</pre>
 
</pre>
  
Unlike the previous examples, the command in this example is hardcoded, so an attacker cannot control the argument passed to system(). However, since the program does not specify an absolute path for make and does not scrub any environment variables prior to invoking the command, the attacker can modify their $PATH variable to point to a malicious binary named make and execute the CGI script from a shell prompt. And since the program has been installed setuid root, the attacker's version of make now runs with root privileges.
+
Unlike the previous examples, the command in this example is hardcoded, so an attacker cannot control the argument passed to system(). However, since the program does not specify an absolute path for make, and does not scrub any environment variables prior to invoking the command, the attacker can modify their $PATH variable to point to a malicious binary named make and execute the CGI script from a shell prompt. And since the program has been installed setuid root, the attacker's version of make now runs with root privileges.
  
 
The environment plays a powerful role in the execution of system commands within programs. Functions like system() and exec() use the environment of the program that calls them, and therefore attackers have a potential opportunity to influence the behavior of these calls.
 
The environment plays a powerful role in the execution of system commands within programs. Functions like system() and exec() use the environment of the program that calls them, and therefore attackers have a potential opportunity to influence the behavior of these calls.
Line 108: Line 115:
 
There are many sites that will tell you that Java's Runtime.exec is exactly the same as C's system function. This is not true. Both allow you to invoke a new program/process. However, C's system function passes its arguments to the shell (/bin/sh) to be parsed, whereas Runtime.exec tries to split the string into an array of words, then executes the first word in the array with the rest of the words as parameters. Runtime.exec does NOT try to invoke the shell at any point. The key difference is that much of the functionality provided by the shell that could be used for mischief (chaining commands using "&", "&&", "|", "||", etc, redirecting input and output) would simply end up as a parameter being passed to the first command, and likely causing a syntax error, or being thrown out as an invalid parameter.
 
There are many sites that will tell you that Java's Runtime.exec is exactly the same as C's system function. This is not true. Both allow you to invoke a new program/process. However, C's system function passes its arguments to the shell (/bin/sh) to be parsed, whereas Runtime.exec tries to split the string into an array of words, then executes the first word in the array with the rest of the words as parameters. Runtime.exec does NOT try to invoke the shell at any point. The key difference is that much of the functionality provided by the shell that could be used for mischief (chaining commands using "&", "&&", "|", "||", etc, redirecting input and output) would simply end up as a parameter being passed to the first command, and likely causing a syntax error, or being thrown out as an invalid parameter.
  
==Related Threats==
+
===Example 5===
 +
The following trivial code snippets are vulnerable to OS command injection on the Unix/Linux platform:
  
==Related Attacks==
+
:* C:
  
Code Injection
+
#include <stdlib.h>
Blind SQL Injection
+
#include <stdio.h>
Blind XPath Injection
+
#include <string.h>
LDAP Injection
+
Relative Path Traversal
+
int main(int argc, char **argv)
Absolute Path Traversal
+
{
 +
      char command[256];
 +
 +
      if(argc != 2) {
 +
          printf("Error: Please enter a program to time!\n");
 +
          return -1;
 +
      }
 +
 +
      memset(&command, 0, sizeof(command));
 +
 +
      strcat(command, "time ./");
 +
      strcat(command, argv[1]);
 +
 +
      system(command);
 +
      return 0;
 +
}
  
 +
:* If this were a suid binary, consider the case when an attacker enters the following: 'ls; cat /etc/shadow'. In the Unix environment, shell commands are separated by a semi-colon. We now can execute system commands at will!
  
==Related Vulnerabilities==
+
:* Java:
 
+
 
+
 
+
==Related Countermeasures==
+
 
+
*[[:Category:Input Validation]]
+
 
+
Using black and/or white lists which defines valid input data. Such approach is more accurate and provides better risk analysis, when there is need of modification of the lists.
+
 
+
E.g. When we expect digits as an input, then we should perform accurate input data validation.
+
 
+
<pre>
+
 
+
#include <stdio.h>
+
#include <ctype.h>
+
#include <string.h>
+
 
+
int main(int argc, char **argv)
+
{
+
      char a[256];
+
      strncpy(a, argv[1], sizeof(a)-1);
+
 
+
      int b=0;
+
 
+
      for(b=0; b<strlen(a); b++) {
+
              if(isdigit((int)a[b])) printf("%c", a[b]);
+
      }
+
 
+
      printf("\n");
+
      return 0;
+
}
+
</pre>
+
 
+
In PHP for input data validation we may use e.g. preg_match() function:
+
 
+
 
+
<pre>
+
<?php
+
  $clean = array();
+
  if (preg_match("/^[0-9]+:[X-Z]+$/D", $_GET['var'])) {
+
    $clean['var'] = $_GET['var'];
+
  }
+
?>
+
</pre>
+
  
For special attention deserves modifier "/D", which additionally protects against HTTP Response Splitting type of attacks.
+
'''There are many sites that will tell you that Java's Runtime.exec is exactly the same as C's system function. This is not true. Both allow you to invoke a new program/process. However, C's system function passes its arguments to the shell (/bin/sh) to be parsed, whereas Runtime.exec tries to split the string into an array of words, then executes the first word in the array with the rest of the words as parameters. Runtime.exec does NOT try to invoke the shell at any point. The key difference is that much of the functionality provided by the shell that could be used for mischief (chaining commands using "&", "&&", "|", "||", etc, redirecting input and output) would simply end up as a parameter being passed to the first command, and likely causing a syntax error, or being thrown out as an invalid parameter.'''
  
 +
<!--==Related [[Threat Agents]]==
 +
TBD
 +
-->
 +
==Related [[Attacks]]==
 +
* [[Code Injection]]
 +
* [[Blind SQL Injection]]
 +
* [[Blind XPath Injection]]
 +
* [[LDAP injection]]
 +
* [[Relative Path Traversal]]
 +
* [[Absolute Path Traversal]]
  
Avoid using of environment variables if the attacker may alter their values.
+
<!--==Related [[Vulnerabilities]]==
 +
TBD
 +
-->
 +
==Related [[Controls]]==
 +
* [[:Category:Input Validation]]
  
 +
Ideally, a developer should use existing API for their language. For example (Java): Rather than use Runtime.exec() to issue a 'mail' command, use the available Java API located at javax.mail.*
  
References:
+
If no such available API exists, the developer should scrub all input for malicious characters. Implementing a positive security model would be most efficient. Typically, it is much easier to define the legal characters than the illegal characters.
http://blog.php-security.org/archives/76-Holes-in-most-preg_match-filters.html
+
  
==Categories==
+
==References==
 +
* [http://cwe.mitre.org/data/definitions/77.html CWE-77: Command Injection]
 +
* [http://cwe.mitre.org/data/definitions/78.html CWE-78: OS Command Injection]
 +
* http://blog.php-security.org/archives/76-Holes-in-most-preg_match-filters.html
  
 
[[Category:Injection Attack]]
 
[[Category:Injection Attack]]
 
[[Category:Injection]]
 
[[Category:Injection]]
 
[[Category:Attack]]
 
[[Category:Attack]]
 +
[[Category:Externally Linked Page]]

Latest revision as of 09:19, 4 December 2012

This is an Attack. To view all attacks, please see the Attack Category page.



Last revision (mm/dd/yy): 12/4/2012

Description

The purpose of the command injection attack is to inject and execute commands specified by the attacker in the vulnerable application. In situation like this, the application, which executes unwanted system commands, is like a pseudo system shell, and the attacker may use it as any authorized system user. However, commands are executed with the same privileges and environment as the application has. Command injection attacks are possible in most cases because of lack of correct input data validation, which can be manipulated by the attacker (forms, cookies, HTTP headers etc.).

There is a variant of the Code Injection attack. The difference with code injection is that the attacker adds his own code to the existing code. In this way, the attacker extends the default functionality of the application without the necessity of executing system commands. Injected code is executed with the same privileges and environment as the application has.

An OS command injection attack occurs when an attacker attempts to execute system level commands through a vulnerable application. Applications are considered vulnerable to the OS command injection attack if they utilize user input in a system level command.

Examples

Example 1

The following code is a wrapper around the UNIX command cat which prints the contents of a file to standard output. It is also injectable:


#include <stdio.h>
#include <unistd.h>

int main(int argc, char **argv) {
 char cat[] = "cat ";
 char *command;
 size_t commandLength;

 commandLength = strlen(cat) + strlen(argv[1]) + 1;
 command = (char *) malloc(commandLength);
 strncpy(command, cat, commandLength);
 strncat(command, argv[1], (commandLength - strlen(cat)) );

 system(command);
 return (0);
}

Used normally, the output is simply the contents of the file requested:


$ ./catWrapper Story.txt
When last we left our heroes...

However, if we add a semicolon and another command to the end of this line, the command is executed by catWrapper with no complaint:

$ ./catWrapper "Story.txt; ls"
When last we left our heroes...
Story.txt               doubFree.c              nullpointer.c
unstosig.c              www*                    a.out*
format.c                strlen.c                useFree*
catWrapper*             misnull.c               strlength.c             useFree.c
commandinjection.c      nodefault.c             trunc.c                 writeWhatWhere.c

If catWrapper had been set to have a higher privilege level than the standard user, arbitrary commands could be executed with that higher privilege.

Example 2

The following simple program accepts a filename as a command line argument, and displays the contents of the file back to the user. The program is installed setuid root because it is intended for use as a learning tool to allow system administrators in-training to inspect privileged system files without giving them the ability to modify them or damage the system.

       int main(char* argc, char** argv) {
               char cmd[CMD_MAX] = "/usr/bin/cat ";
               strcat(cmd, argv[1]);
               system(cmd);
       }

Because the program runs with root privileges, the call to system() also executes with root privileges. If a user specifies a standard filename, the call works as expected. However, if an attacker passes a string of the form ";rm -rf /", then the call to system() fails to execute cat due to a lack of arguments and then plows on to recursively delete the contents of the root partition.

Example 3

The following code from a privileged program uses the environment variable $APPHOME to determine the application's installation directory, and then executes an initialization script in that directory.

       ...
       char* home=getenv("APPHOME");
       char* cmd=(char*)malloc(strlen(home)+strlen(INITCMD));
       if (cmd) {
               strcpy(cmd,home);
               strcat(cmd,INITCMD);
               execl(cmd, NULL);
       }
       ...

As in Example 2, the code in this example allows an attacker to execute arbitrary commands with the elevated privilege of the application. In this example, the attacker can modify the environment variable $APPHOME to specify a different path containing a malicious version of INITCMD. Because the program does not validate the value read from the environment, by controlling the environment variable, the attacker can fool the application into running malicious code.

The attacker is using the environment variable to control the command that the program invokes, so the effect of the environment is explicit in this example. We will now turn our attention to what can happen when the attacker changes the way the command is interpreted.

Example 4

The code below is from a web-based CGI utility that allows users to change their passwords. The password update process under NIS includes running make in the /var/yp directory. Note that since the program updates password records, it has been installed setuid root.

The program invokes make as follows:

       system("cd /var/yp && make &> /dev/null");

Unlike the previous examples, the command in this example is hardcoded, so an attacker cannot control the argument passed to system(). However, since the program does not specify an absolute path for make, and does not scrub any environment variables prior to invoking the command, the attacker can modify their $PATH variable to point to a malicious binary named make and execute the CGI script from a shell prompt. And since the program has been installed setuid root, the attacker's version of make now runs with root privileges.

The environment plays a powerful role in the execution of system commands within programs. Functions like system() and exec() use the environment of the program that calls them, and therefore attackers have a potential opportunity to influence the behavior of these calls.

There are many sites that will tell you that Java's Runtime.exec is exactly the same as C's system function. This is not true. Both allow you to invoke a new program/process. However, C's system function passes its arguments to the shell (/bin/sh) to be parsed, whereas Runtime.exec tries to split the string into an array of words, then executes the first word in the array with the rest of the words as parameters. Runtime.exec does NOT try to invoke the shell at any point. The key difference is that much of the functionality provided by the shell that could be used for mischief (chaining commands using "&", "&&", "|", "||", etc, redirecting input and output) would simply end up as a parameter being passed to the first command, and likely causing a syntax error, or being thrown out as an invalid parameter.

Example 5

The following trivial code snippets are vulnerable to OS command injection on the Unix/Linux platform:

  • C:
#include <stdlib.h>
#include <stdio.h>
#include <string.h>

int main(int argc, char **argv)
{
     char command[256];

     if(argc != 2) {
          printf("Error: Please enter a program to time!\n");
          return -1;
     }

     memset(&command, 0, sizeof(command));

     strcat(command, "time ./");
     strcat(command, argv[1]);

     system(command);
     return 0;
}
  • If this were a suid binary, consider the case when an attacker enters the following: 'ls; cat /etc/shadow'. In the Unix environment, shell commands are separated by a semi-colon. We now can execute system commands at will!
  • Java:

There are many sites that will tell you that Java's Runtime.exec is exactly the same as C's system function. This is not true. Both allow you to invoke a new program/process. However, C's system function passes its arguments to the shell (/bin/sh) to be parsed, whereas Runtime.exec tries to split the string into an array of words, then executes the first word in the array with the rest of the words as parameters. Runtime.exec does NOT try to invoke the shell at any point. The key difference is that much of the functionality provided by the shell that could be used for mischief (chaining commands using "&", "&&", "|", "||", etc, redirecting input and output) would simply end up as a parameter being passed to the first command, and likely causing a syntax error, or being thrown out as an invalid parameter.

Related Attacks

Related Controls

Ideally, a developer should use existing API for their language. For example (Java): Rather than use Runtime.exec() to issue a 'mail' command, use the available Java API located at javax.mail.*

If no such available API exists, the developer should scrub all input for malicious characters. Implementing a positive security model would be most efficient. Typically, it is much easier to define the legal characters than the illegal characters.

References