Skip to main content

CLE Loads Everything (at least, many binary formats!) and provides a pythonic interface to analyze what they are and what they would look like in memory.

Project description

CLE

Latest Release Python Version PyPI Statistics License

CLE loads binaries and their associated libraries, resolves imports and provides an abstraction of process memory the same way as if it was loader by the OS's loader.

Project Links

Project repository: https://github.com/angr/cle

Documentation: https://api.angr.io/projects/cle/en/latest/

Installation

pip install cle

Usage example

>>> import cle
>>> ld = cle.Loader("/bin/ls")
>>> hex(ld.main_object.entry)
'0x4048d0'
>>> ld.shared_objects
{'ld-linux-x86-64.so.2': <ELF Object ld-2.21.so, maps [0x5000000:0x522312f]>,
 'libacl.so.1': <ELF Object libacl.so.1.1.0, maps [0x2000000:0x220829f]>,
 'libattr.so.1': <ELF Object libattr.so.1.1.0, maps [0x4000000:0x4204177]>,
 'libc.so.6': <ELF Object libc-2.21.so, maps [0x3000000:0x33a1a0f]>,
 'libcap.so.2': <ELF Object libcap.so.2.24, maps [0x1000000:0x1203c37]>}
>>> ld.addr_belongs_to_object(0x5000000)
<ELF Object ld-2.21.so, maps [0x5000000:0x522312f]>
>>> libc_main_reloc = ld.main_object.imports['__libc_start_main']
>>> hex(libc_main_reloc.addr)       # Address of GOT entry for libc_start_main
'0x61c1c0'
>>> import pyvex
>>> some_text_data = ld.memory.load(ld.main_object.entry, 0x100)
>>> irsb = pyvex.lift(some_text_data, ld.main_object.entry, ld.main_object.arch)
>>> irsb.pp()
IRSB {
   t0:Ity_I32 t1:Ity_I32 t2:Ity_I32 t3:Ity_I64 t4:Ity_I64 t5:Ity_I64 t6:Ity_I32 t7:Ity_I64 t8:Ity_I32 t9:Ity_I64 t10:Ity_I64 t11:Ity_I64 t12:Ity_I64 t13:Ity_I64 t14:Ity_I64

   15 | ------ IMark(0x4048d0, 2, 0) ------
   16 | t5 = 32Uto64(0x00000000)
   17 | PUT(rbp) = t5
   18 | t7 = GET:I64(rbp)
   19 | t6 = 64to32(t7)
   20 | t2 = t6
   21 | t9 = GET:I64(rbp)
   22 | t8 = 64to32(t9)
   23 | t1 = t8
   24 | t0 = Xor32(t2,t1)
   25 | PUT(cc_op) = 0x0000000000000013
   26 | t10 = 32Uto64(t0)
   27 | PUT(cc_dep1) = t10
   28 | PUT(cc_dep2) = 0x0000000000000000
   29 | t11 = 32Uto64(t0)
   30 | PUT(rbp) = t11
   31 | PUT(rip) = 0x00000000004048d2
   32 | ------ IMark(0x4048d2, 3, 0) ------
   33 | t12 = GET:I64(rdx)
   34 | PUT(r9) = t12
   35 | PUT(rip) = 0x00000000004048d5
   36 | ------ IMark(0x4048d5, 1, 0) ------
   37 | t4 = GET:I64(rsp)
   38 | t3 = LDle:I64(t4)
   39 | t13 = Add64(t4,0x0000000000000008)
   40 | PUT(rsp) = t13
   41 | PUT(rsi) = t3
   42 | PUT(rip) = 0x00000000004048d6
   43 | t14 = GET:I64(rip)
   NEXT: PUT(rip) = t14; Ijk_Boring
}

Valid options

For a full listing and description of the options that can be provided to the loader and the methods it provides, please examine the docstrings in cle/loader.py. If anything is unclear or poorly documented (there is much) please complain through whatever channel you feel appropriate.

Loading Backends

CLE's loader is implemented in the Loader class. There are several backends that can be used to load a single file:

  • ELF, as its name says, loads ELF binaries. ELF files loaded this way are statically parsed using PyElfTools.

  • PE is a backend to load Microsoft's Portable Executable format, effectively Windows binaries. It uses the (optional) pefile module.

  • Mach-O is a backend to load, you guessed it, Mach-O binaries. Support is limited for this backend.

  • Blob is a backend to load unknown data. It requires that you specify the architecture it would be run on, in the form of a class from ArchInfo.

Which backend you use can be specified as an argument to Loader. If left unspecified, the loader will pick a reasonable default.

Finding shared libraries

  • If the auto_load_libs option is set to False, the Loader will not automatically load libraries requested by loaded objects. Otherwise...

  • The loader determines which shared objects are needed when loading binaries, and searches for them in the following order:

    • in the current working directory
    • in folders specified in the ld_path option
    • in the same folder as the main binary
    • in the system (in the corresponding library path for the architecture of the binary, e.g., /usr/arm-linux-gnueabi/lib for ARM, note that you need to install cross libraries for this, e.g., libc6-powerpc-cross on Debian - needs emdebian repos)
    • in the system, but with mismatched version numbers from what is specified as a dependency, if the ignore_import_version_numbers option is True
  • If no binary is found with the correct architecture, the loader raises an exception if except_missing_libs option is True. Otherwise it simply leaves the dependencies unresolved.

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cle-9.2.89.tar.gz (179.4 kB view details)

Uploaded Source

Built Distribution

cle-9.2.89-py3-none-any.whl (184.2 kB view details)

Uploaded Python 3

File details

Details for the file cle-9.2.89.tar.gz.

File metadata

  • Download URL: cle-9.2.89.tar.gz
  • Upload date:
  • Size: 179.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.8.18

File hashes

Hashes for cle-9.2.89.tar.gz
Algorithm Hash digest
SHA256 5ffbe592caf6f4f9dd81e4eceb43b64c2f498ecac4289fedc2869a41b5ef3694
MD5 e62d34a14cd982e22f33bb5b2d5a6aba
BLAKE2b-256 ec443b9fdf20d5ad44ff2c4983397e88c028c0120949ac51d48f2c05c461fad3

See more details on using hashes here.

File details

Details for the file cle-9.2.89-py3-none-any.whl.

File metadata

  • Download URL: cle-9.2.89-py3-none-any.whl
  • Upload date:
  • Size: 184.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.8.18

File hashes

Hashes for cle-9.2.89-py3-none-any.whl
Algorithm Hash digest
SHA256 d7559282775bf4004a64e3f26c14fec6c6991595da23d9d6331f4d896dca345b
MD5 5121cba7d0fdecbd5a7511c0dc0915dd
BLAKE2b-256 38ebeb519daaafb600a05978a6bb10f03f7f4a0f49bec930ee21a6dc2927bc01

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page