Re: [PATCH] libsemanage: treat module data as bytes in the Python bindings

On Thu, Aug 13, 2026 at 2:41 PM Stephen Smalley
<[email protected]> wrote:
>
> semanage_module_install() / semanage_module_install_info() take a
> (char *data, size_t data_len) pair, and semanage_module_extract()
> returns (void **mapped_data, size_t *data_len). In both directions
> the payload is an opaque module blob (.pp / bzip2 / CIL / HLL), not a
> NUL-terminated C string.
>
> On Python 3 the current SWIG bindings get this wrong on both sides:
>
> - install()/install_info(): the SWIG default for a bare "char *"
>   only accepts str objects and UTF-8-encodes them, so a .pp blob
>   cannot be passed at all, and even for CIL text the separately
>   supplied data_len is wrong for any non-ASCII content since
>   Python len(str) counts code points, not UTF-8 bytes.
>
> - extract(): %cstring_output_allocate_size turns the mapped buffer
>   into a str via UTF-8 decoding, which for a binary module either
>   raises UnicodeDecodeError or (with surrogateescape) yields a str
>   that cannot be fed straight back into install().
>
> Add a (char *module_data, size_t data_len) input typemap that
> accepts any object exposing the buffer protocol (bytes, bytearray,
> mmap, ...) as a single argument and derives the length from it, and
> apply it to the (char *data, size_t data_len) pair of
> semanage_module_install_info() as well.
>
> Replace the extract() output typemap with an explicit multi-argument
> typemap that returns a bytes object built from the mapped buffer and
> then unmaps it, so extract() -> install() round-trips.  Drop the
> now-unused <cstring.i> include.
>
> The Python signatures become
>   semanage_module_install(sh, data, name, ext_lang)
>   semanage_module_install_info(sh, modinfo, data)
> There are no in-tree Python callers.
>
> Fixes: https://github.com/SELinuxProject/selinux/issues/502
> Signed-off-by: Stephen Smalley <[email protected]>

Acked-by: James Carter <[email protected]>

> ---
>  libsemanage/src/semanageswig_python.i | 43 +++++++++++++++++++++++++--
>  1 file changed, 40 insertions(+), 3 deletions(-)
>
> diff --git a/libsemanage/src/semanageswig_python.i b/libsemanage/src/semanageswig_python.i
> index 0ca09763..1729e965 100644
> --- a/libsemanage/src/semanageswig_python.i
> +++ b/libsemanage/src/semanageswig_python.i
> @@ -102,9 +102,46 @@
>  %apply int *OUTPUT { unsigned int * };
>  %apply int *OUTPUT { uint16_t * };
>
> -%include <cstring.i>
> -/* This is needed to properly mmap binary data in SWIG */
> -%cstring_output_allocate_size(void **mapped_data, size_t *data_len, munmap(*$1, *$2));
> +/*
> + * semanage_module_extract(): the (void **mapped_data, size_t *data_len) pair
> + * is an opaque module blob (.pp / bzip2 / CIL / HLL).  Return it to Python
> + * as a bytes object; %cstring_output_allocate_size would UTF-8-decode it
> + * into a str, which corrupts binary modules.
> + */
> +%typemap(in, numinputs=0) (void **mapped_data, size_t *data_len)
> +       (void *temp_data = NULL, size_t temp_len = 0) {
> +       $1 = &temp_data;
> +       $2 = &temp_len;
> +}
> +%typemap(argout) (void **mapped_data, size_t *data_len) {
> +       if (*$1) {
> +               $result = SWIG_AppendOutput(
> +                       $result,
> +                       PyBytes_FromStringAndSize((const char *)*$1, *$2));
> +               munmap(*$1, *$2);
> +       } else {
> +               $result = SWIG_AppendOutput($result, SWIG_Py_Void());
> +       }
> +}
> +
> +/*
> + * semanage_module_install() and semanage_module_install_info() take a
> + * (char *data, size_t data_len) pair that is an opaque module blob
> + * (.pp / bzip2 / CIL / HLL), not a NUL-terminated C string.  Accept any
> + * Python object exposing the buffer protocol (bytes, bytearray, mmap, ...)
> + * as a single argument and derive the length from it.
> + */
> +%typemap(in) (char *module_data, size_t data_len) (Py_buffer view) {
> +       view.obj = NULL;
> +       if (PyObject_GetBuffer($input, &view, PyBUF_SIMPLE) < 0)
> +               SWIG_fail;
> +       $1 = (char *)view.buf;
> +       $2 = (size_t)view.len;
> +}
> +%typemap(freearg) (char *module_data, size_t data_len) {
> +       PyBuffer_Release(&view$argnum);
> +}
> +%apply (char *module_data, size_t data_len) { (char *data, size_t data_len) };
>
>  %typemap(in, numinputs=0) char **(char *temp=NULL) {
>         $1 = &temp;
> --
> 2.55.0
>