metasploit-framework/lib/rex/arch/x86.rb

477 lines
11 KiB
Ruby
Raw Normal View History

#!/usr/bin/env ruby
module Rex
module Arch
#
# everything here is mostly stole from vlad's perl x86 stuff
#
module X86
#
# Register number constants
#
EAX = AL = AX = ES = 0
ECX = CL = CX = CS = 1
EDX = DL = DX = SS = 2
EBX = BL = BX = DS = 3
ESP = AH = SP = FS = 4
EBP = CH = BP = GS = 5
ESI = DH = SI = 6
EDI = BH = DI = 7
REG_NAMES32 = [ 'eax', 'ecx', 'edx', 'ebx',
'esp', 'ebp', 'esi', 'edi' ] # :nodoc:
# Jump tp a specific register
def self.jmp_reg(str)
reg = reg_number(str)
_check_reg(reg)
"\xFF" + [224 + reg].pack('C')
end
# This method returns the opcodes that compose a jump instruction to the
# supplied relative offset.
def self.jmp(addr)
"\xe9" + pack_dword(rel_number(addr))
end
#
# This method adds/subs a packed long integer
#
def self.dword_adjust(dword, amount=0)
[dword.unpack('V')[0] + amount].pack('V')
end
#
# This method returns the opcodes that compose a tag-based search routine
#
def self.searcher(tag)
"\xbe" + dword_adjust(tag,-1)+ # mov esi, Tag - 1
"\x46" + # inc esi
"\x47" + # inc edi (end_search:)
"\x39\x37" + # cmp [edi],esi
"\x75\xfb" + # jnz 0xa (end_search)
"\x46" + # inc esi
"\x4f" + # dec edi (start_search:)
"\x39\x77\xfc" + # cmp [edi-0x4],esi
"\x75\xfa" + # jnz 0x10 (start_search)
jmp_reg('edi') # jmp edi
end
#
# This method returns the opcodes that compose a short jump instruction to
# the supplied relative offset.
#
def self.jmp_short(addr)
"\xeb" + pack_lsb(rel_number(addr, -2))
end
#
# This method returns the opcodes that compose a relative call instruction
# to the address specified.
#
def self.call(addr)
"\xe8" + pack_dword(rel_number(addr, -5))
end
#
# This method returns a number offset to the supplied string.
#
def self.rel_number(num, delta = 0)
s = num.to_s
case s[0, 2]
when '$+'
num = s[2 .. -1].to_i
when '$-'
num = -1 * s[2 .. -1].to_i
when '0x'
num = s.hex
else
delta = 0
end
return num + delta
end
#
# This method returns the number associated with a named register.
#
def self.reg_number(str)
return self.const_get(str.upcase)
end
#
# This method returns the register named associated with a given register
# number.
#
def self.reg_name32(num)
_check_reg(num)
return REG_NAMES32[num].dup
end
#
# This method generates the encoded effective value for a register.
#
def self.encode_effective(shift, dst)
return (0xc0 | (shift << 3) | dst)
end
#
# This method generates the mod r/m character for a source and destination
# register.
#
def self.encode_modrm(dst, src)
_check_reg(dst, src)
return (0xc0 | src | dst << 3).chr
end
#
# This method generates a push byte instruction.
#
def self.push_byte(byte)
# push byte will sign extend...
if byte < 128 && byte >= -128
return "\x6a" + (byte & 0xff).chr
end
raise ::ArgumentError, "Can only take signed byte values!", caller()
end
#
# This method generates a push dword instruction.
#
def self.push_dword(val)
return "\x68" + [ val ].pack('V')
end
#
# This method generates a pop dword instruction into a register.
#
def self.pop_dword(dst)
_check_reg(dst)
return (0x58 | dst).chr
end
#
# This method generates an instruction that clears the supplied register in
# a manner that attempts to avoid bad characters, if supplied.
#
def self.clear(reg, badchars = '')
_check_reg(reg)
return set(reg, 0, badchars)
end
#
# This method generates the opcodes that set the low byte of a given
# register to the supplied value.
#
def self.mov_byte(reg, val)
_check_reg(reg)
# chr will raise RangeError if val not between 0 .. 255
return (0xb0 | reg).chr + val.chr
end
#
# This method generates the opcodes that set the low word of a given
# register to the supplied value.
#
def self.mov_word(reg, val)
_check_reg(reg)
if val < 0 || val > 0xffff
raise RangeError, "Can only take unsigned word values!", caller()
end
return "\x66" + (0xb8 | reg).chr + [ val ].pack('v')
end
#
# This method generates the opcodes that set the a register to the
# supplied value.
#
def self.mov_dword(reg, val)
_check_reg(reg)
return (0xb8 | reg).chr + [ val ].pack('V')
end
#
# This method is a general way of setting a register to a value. Depending
# on the value supplied, different sets of instructions may be used.
#
# TODO: Make this moderatly intelligent so it chain instructions by itself
# (ie. xor eax, eax + mov al, 4 + xchg ah, al)
def self.set(dst, val, badchars = '')
_check_reg(dst)
# If the value is 0 try xor/sub dst, dst (2 bytes)
if(val == 0)
opcodes = Rex::Text.remove_badchars("\x29\x2b\x31\x33", badchars)
if !opcodes.empty?
return opcodes[rand(opcodes.length)].chr + encode_modrm(dst, dst)
end
# TODO: SHL/SHR
# TODO: AND
end
# try push BYTE val; pop dst (3 bytes)
begin
return _check_badchars(push_byte(val) + pop_dword(dst), badchars)
rescue ::ArgumentError, ::RuntimeError, ::RangeError
end
# try clear dst, mov BYTE dst (4 bytes)
begin
# break if val == 0
return _check_badchars(clear(dst, badchars) + mov_byte(dst, val), badchars)
rescue ::ArgumentError, ::RuntimeError, ::RangeError
end
# try mov DWORD dst (5 bytes)
begin
return _check_badchars(mov_dword(dst, val), badchars)
rescue ::ArgumentError, ::RuntimeError, ::RangeError
end
# try push DWORD, pop dst (6 bytes)
begin
return _check_badchars(push_dword(val) + pop_dword(dst), badchars)
rescue ::ArgumentError, ::RuntimeError, ::RangeError
end
# try clear dst, mov WORD dst (6 bytes)
begin
# break if val == 0
return _check_badchars(clear(dst, badchars) + mov_word(dst, val), badchars)
rescue ::ArgumentError, ::RuntimeError, ::RangeError
end
raise RuntimeError, "No valid set instruction could be created!", caller()
end
#
# Builds a subtraction instruction using the supplied operand
# and register.
#
def self.sub(val, reg, badchars = '', add = false, adjust = false, bits = 0)
opcodes = []
shift = (add == true) ? 0 : 5
if (bits <= 8 and val >= -0x7f and val <= 0x7f)
opcodes <<
((adjust) ? '' : clear(reg, badchars)) +
"\x83" +
[ encode_effective(shift, reg) ].pack('C') +
[ val.to_i ].pack('C')
end
if (bits <= 16 and val >= -0xffff and val <= 0)
opcodes <<
((adjust) ? '' : clear(reg, badchars)) +
"\x66\x81" +
[ encode_effective(shift, reg) ].pack('C') +
[ val.to_i ].pack('v')
end
opcodes <<
((adjust) ? '' : clear(reg, badchars)) +
"\x81" +
[ encode_effective(shift, reg) ].pack('C') +
[ val.to_i ].pack('V')
# Search for a compatible opcode
opcodes.each { |op|
begin
_check_badchars(op, badchars)
rescue
next
end
return op
}
if opcodes.empty?
raise RuntimeError, "Could not find a usable opcode", caller()
end
end
#
# This method generates the opcodes equivalent to subtracting with a
# negative value from a given register.
#
def self.add(val, reg, badchars = '', adjust = false, bits = 0)
sub(val, reg, badchars, true, adjust, bits)
end
#
# This method wrappers packing an integer as a little-endian buffer.
#
def self.pack_dword(num)
[num].pack('V')
end
#
# This method returns the least significant byte of a packed dword.
#
def self.pack_lsb(num)
pack_dword(num)[0,1]
end
#
# This method adjusts the value of the ESP register by a given amount.
#
def self.adjust_reg(reg, adjustment)
if (adjustment > 0)
sub(adjustment, reg, '', false, false, 32)
else
add(adjustment, reg, '', true, 32)
end
end
def self._check_reg(*regs) # :nodoc:
regs.each { |reg|
if reg > 7 || reg < 0
raise ArgumentError, "Invalid register #{reg}", caller()
end
}
return nil
end
def self._check_badchars(data, badchars) # :nodoc:
idx = Rex::Text.badchar_index(data, badchars)
if idx
raise RuntimeError, "Bad character at #{idx}", caller()
end
return data
end
#
# This method returns an array of 'safe' FPU instructions
#
def self.fpu_instructions
fpus = []
0xe8.upto(0xee) { |x| fpus << "\xd9" + x.chr }
0xc0.upto(0xcf) { |x| fpus << "\xd9" + x.chr }
0xc0.upto(0xdf) { |x| fpus << "\xda" + x.chr }
0xc0.upto(0xdf) { |x| fpus << "\xdb" + x.chr }
0xc0.upto(0xc7) { |x| fpus << "\xdd" + x.chr }
fpus << "\xd9\xd0"
fpus << "\xd9\xe1"
fpus << "\xd9\xf6"
fpus << "\xd9\xf7"
fpus << "\xd9\xe5"
# This FPU instruction seems to fail consistently on Linux
#fpus << "\xdb\xe1"
fpus
end
#
# This method returns an array containing a geteip stub, a register, and an offset
# This method will return nil if the getip generation fails
#
def self.geteip_fpu(badchars)
#
# Default badchars to an empty string
#
badchars ||= ''
#
# Bail out early if D9 is restricted
#
return nil if badchars.index("\xd9")
#
# Create a list of FPU instructions
#
fpus = *self.fpu_instructions
bads = []
badchars.each_byte do |c|
fpus.each do |str|
bads << str if (str.index(c.chr))
end
end
bads.each { |str| fpus.delete(str) }
return nil if fpus.length == 0
#
# Create a list of registers to use for fnstenv
#
dsts = []
0.upto(7) do |c|
dsts << c if (not badchars.index( (0x70+c).chr ))
end
if (dsts.include?(ESP) and badchars.index("\x24"))
dsts.delete(ESP)
end
return nil if dsts.length == 0
#
# Grab a random FPU instruction
#
fpu = fpus[ rand(fpus.length) ]
#
# Grab a random register from dst
#
while(dsts.length > 0)
buf = ''
dst = dsts[ rand(dsts.length) ]
dsts.delete(dst)
# If the register is not ESP, copy ESP
if (dst != ESP)
next if badchars.index( (0x70 + dst).chr )
if (not (badchars.index("\x89") or badchars.index( (0xE0+dst).chr )))
buf << "\x89" + (0xE0 + dst).chr
else
next if badchars.index("\x54")
next if badchars.index( (0x58+dst).chr )
buf << "\x54" + (0x58 + dst).chr
end
end
pad = 0
while (pad < (128-12) and badchars.index( (256-12-pad)))
pad += 4
end
# Give up on finding a value to use here
if (pad == (128-12))
return nil
end
out = buf + fpu + "\xd9" + (0x70 + dst).chr
out << "\x24" if dst == ESP
out << (256-12-pad).chr
regs = [*(0..7)]
while (regs.length > 0)
reg = regs[ rand(regs.length) ]
regs.delete(reg)
next if reg == ESP
next if badchars.index( (0x58 + reg).chr )
# Pop the value back out
0.upto(pad / 4) { |c| out << (0x58 + reg).chr }
# Fix the value to point to self
gap = out.length - buf.length
return [out, REG_NAMES32[reg].upcase, gap]
end
end
return nil
end
end
end end