1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234
|
08 - Transforming an ELF executable into a library
--------------------------------------------------
In this tutorial, we will see how to convert a **PIE** executable into a library
Scripts and materials are available here: `materials <https://github.com/lief-project/tutorials/tree/master/08_ELF_bin2lib>`_
By Romain Thomas - `@rh0main <https://twitter.com/rh0main>`_
------
Introduction
~~~~~~~~~~~~
Actually, if we look at the header of a ELF PIE executable one can notice that it has the same type as a shared object (i.e. library)
.. code-block:: console
$ readelf -h /usr/bin/ssh|grep Type
Type: DYN (Shared object file)
$ readelf -h /usr/lib/libm.so|grep Type
Type: DYN (Shared object file)
Using LIEF we can access this information through the :attr:`~lief.ELF.Header.file_type` attribute
.. code-block:: python
>>> libm = lief.parse("/usr/lib/libm.so.6")
>>> print(libm.header.file_type)
E_TYPE.DYNAMIC
>>> ssh = lief.parse("/usr/bin/ssh")
>>> print(ss.header.file_type)
E_TYPE.DYNAMIC
The main difference between a PIE binaries and a shared libraries is how symbols are exported.
A shared library aims to expose functions so that executable can bind to it whereas executables shouldn't not expose functions [1]_
It's confirmed with the number of exported functions in the two different objects:
.. code-block:: python
>>> print(len(libm.exported_functions))
572
>>> print(len(ssh.exported_functions))
10
In this tutorial we will see how we can transform raw function addresses into exported functions associated with a symbol,
thus thus exposing internal functions of the executable.
Exporting functions
~~~~~~~~~~~~~~~~~~~
Such transformation can be useful if we found a function at a given address and want to instrument it (using ``dlopen``/``dlsym`` for example).
Once the target function is exported we can link it as we would do for a *normal* library.
For example in a fuzzing scenario if one identifies a function that is a parser, we can export it and then we can feed its inputs with AFL. Thus we cut the path from the normal entrypoint to reach the function.
Let's see how it works on a basic *crackme*:
.. code-block:: cpp
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#define LOCAL __attribute__ ((visibility ("hidden")))
#define NOINLINE __attribute__ ((noinline))
NOINLINE LOCAL int check(char* input) {
if (strcmp(input, "easy") == 0) {
return 1;
}
return 0;
}
int main(int argc, char** argv) {
if (argc != 2) {
printf("Usage: %s flag\n", argv[0]);
exit(-1);
}
if (check(argv[1])) {
printf("Well done!\n");
} else {
printf("Wrong!\n");
}
return 0;
}
This code takes a string as input and call the **check** function on this string, then it returns ``1`` if the input is ``easy``. ``0`` otherwise.
The ``__attribute__ ((visibility ("hidden")))`` attribute is used to avoid that the compiler export automatically the **check** and the ``__attribute__ ((noinline))`` one
to disable the inline optimization. If the function check is inlined, there won't be an address associated to this function.
This figure sump-up the execution flow:
.. figure:: ../_static/tutorial/08/bin2lib_a.png
:scale: 25%
:align: center
The *crackme* can be compiled with:
.. code-block:: console
$ gcc crackme101.c -O0 -fPIE -pie -Wl,-strip-all -o crackme101.bin
$ ./crackme101.bin foo
Wrong!
$ ./crackme101.bin easy
Well done!
By opening ``crackme101.bin`` with LIEF we can check that no functions are exported:
.. code-block:: python
>>> import lief
>>> crackme101 = lief.parse("./crackme101.bin")
>>> print(len(crackme101.exported_functions))
0
Using a disassembler we can quickly identify the check function address:
.. figure:: ../_static/tutorial/08/crackme101_ida.png
:scale: 100%
:align: center
In this case, the **check** function is located at the address: ``0x72A`` [2]_
Now that we identified the address we can export it as a named function: ``check_found``
.. code-block:: python
>>> crackme101.add_exported_function(0x72A, "check_found")
>>> crackme101.write("libcrackme101.so")
And that all!
``libcrackme101.so`` is now a **library** that export one function: ``check_found``.
.. code-block:: python
>>> import lief
>>> libcrackme101 = lief.parse("./libcrackme101.so")
>>> print(len(crackme101.exported_functions))
1
>>> print(crackme101.exported_functions[0])
check_found
It turns out that ``libcrackme101.so`` is still an executable:
.. code-block:: console
$ ./libcrackme101.so foo
Wrong!
$ ./libcrackme101.so easy
Well done!
Since we have exported a function we can now use ``dlopen`` on ``libcrackme101.so`` and
``dlsym`` on ``check_found``
.. code-block:: cpp
:emphasize-lines: 9,10
#include <dlfcn.h>
#include <stdio.h>
#include <stdlib.h>
typedef int(*check_t)(char*);
int main (int argc, char** argv) {
void* handler = dlopen("./libcrackme101.so", RTLD_LAZY);
check_t check_found = (check_t)dlsym(handler, "check_found");
int output = check_found(argv[1]);
printf("Output of check_found('%s'): %d\n", argv[1], output);
return 0;
}
Running the code above should give a similar output:
.. code-block:: console
$ gcc instrument.c -O0 -fPIE -pie -o instrument.bin -ldl
$ ./instrument.bin test
Output of check('test'): 0
$ ./instrument.bin easy
Output of check('easy'): 1
The transformation of the execution flow can be represented as follow:
.. figure:: ../_static/tutorial/08/bin2lib_b.png
:scale: 25%
:align: center
Conclusion
~~~~~~~~~~
Because PIE executables aim to be mapped at a random base address, they globally behave as a library. We only need to export the *interesting* functions.
For non-PIE executables such transformation would be very difficult because it requires to transform first
the executable into a *relocatable* executable. It means creating relocations, patching absolute *jump*, ...
LIEF only support this transformation for ELF and we need to investigate the PE and Mach-O cases [3]_.
.. rubric:: Notes
.. [1] Some functions can be exported by the linker such as ``_init``
.. [2] The mapped virtual address will be ``BASE + 0x72A`` where ``BASE`` is randomly choosed by the ASLR
.. [3] In OSX all executables are compiled with the PIE flag.
:API:
* :meth:`lief.ELF.Binary.add_exported_function`
* :meth:`lief.ELF.Binary.export_symbol`
* :attr:`lief.ELF.Symbol.visibility`
* :attr:`lief.ELF.Symbol.name`
* :attr:`lief.ELF.Symbol.value`
|