_hashlib.c — OpenSSL hash module

CPython's _hashlib module wraps OpenSSL's EVP (envelope) digest API to implement hashlib. A thin Python layer in Lib/hashlib.py dispatches to this extension for all named algorithms that OpenSSL knows about, falling back to _sha3 and _blake2 for algorithms CPython ships its own C code for.

Map

Lines	Symbol	Role
1–120	includes / defines	OpenSSL and Python headers, Py_buffer helpers
121–310	EVPobject	C struct and type slots for a single hash context
311–480	EVP_new_impl	Allocate and initialise an EVP_MD_CTX
481–560	EVPXOF path	Extendable-output variant for shake_128 / shake_256
561–720	hashlib_new	Top-level dispatch by algorithm name
721–860	EVP_digest_impl / EVP_hexdigest_impl	Copy context, finalise, return bytes or hex
861–960	EVP_update_impl	Feed data into an existing context
961–1080	usedforsecurity handling	FIPS guard wired through EVP_DigestInit_ex flags
1081–1300	hashlib_file_digest	Chunked file hashing
1301–2500	method tables, module init	PyMethodDef, PyModuleDef, PyInit__hashlib

Reading

EVP_new_impl: creating a context

Every hash object starts here. The function allocates an EVP_MD_CTX, looks up the digest by name via EVP_get_digestbyname, and calls EVP_DigestInit_ex to bind the algorithm.

// CPython: Modules/_hashlib.c:311 EVP_new_impl
static PyObject *
EVP_new_impl(PyTypeObject *type, PyObject *name_obj, PyObject *data_obj,
             int usedforsecurity)
{
    EVP_MD_CTX *ctx = EVP_MD_CTX_new();
    if (ctx == NULL) return _setException(PyExc_MemoryError, NULL);
    const EVP_MD *digest = py_digest_by_name(state, name_obj, usedforsecurity);
    if (!EVP_DigestInit_ex(ctx, digest, NULL)) { ... }
    ...
}

EVPXOF: shake_128 and shake_256

Extendable-output functions do not have a fixed digest length. CPython exposes them through a separate code path that calls EVP_DigestFinalXOF instead of EVP_DigestFinal_ex, and requires the caller to pass an explicit length argument to digest().

// CPython: Modules/_hashlib.c:490 EVPXOF_digest_impl
static PyObject *
EVPXOF_digest_impl(EVPobject *self, Py_ssize_t length)
{
    EVP_MD_CTX *ctx = EVP_MD_CTX_new();
    EVP_MD_CTX_copy_ex(ctx, self->ctx);
    unsigned char *digest = PyMem_Malloc(length);
    EVP_DigestFinalXOF(ctx, digest, (size_t)length);
    ...
}

usedforsecurity and FIPS mode

Python 3.9 added the usedforsecurity keyword argument. When False, the digest is initialised with EVP_MD_CTX_FLAG_NON_FIPS_ALLOW, letting callers use MD5 or SHA-1 in non-security contexts (e.g. checksums) even on a FIPS-enabled system.

// CPython: Modules/_hashlib.c:975 py_digest_by_name
    if (!usedforsecurity) {
        EVP_MD_CTX_set_flags(ctx, EVP_MD_CTX_FLAG_NON_FIPS_ALLOW);
    }
    if (!EVP_DigestInit_ex(ctx, digest, NULL)) {
        _setException(PyExc_ValueError, "unknown hash name");
        return NULL;
    }

hashlib.file_digest: chunked hashing

hashlib.file_digest(fileobj, digest) reads the file in HASHXOF_DIGESTSIZE-sized chunks (8 KB) and feeds each chunk into the hash. It releases the GIL around the read() call when the file object is a true C-level FILE*.

// CPython: Modules/_hashlib.c:1100 hashlib_file_digest_impl
    while (1) {
        Py_BEGIN_ALLOW_THREADS
        n = fread(buf, 1, sizeof(buf), fp);
        Py_END_ALLOW_THREADS
        if (n == 0) break;
        if (!EVP_DigestUpdate(ctx, buf, n)) { ... }
    }

gopy notes

• hashlib is not yet ported. The gopy equivalent will wrap Go's crypto/* packages rather than OpenSSL directly, so usedforsecurity will be a no-op on non-FIPS builds.
• shake_128 / shake_256 need the XOF calling convention; mark as deferred until the standard fixed-length digests are green.
• file_digest can reuse Go's io.Reader loop with a 32 KB buffer.

CPython 3.14 changes

• 3.9: usedforsecurity keyword added (bpo-9216).
• 3.11: hashlib.file_digest() added (gh-87071).
• 3.13: Internal OpenSSL 3.x legacy provider wiring; EVP_MD_CTX_FLAG_NON_FIPS_ALLOW behaviour updated for OpenSSL 3.
• 3.14: No functional changes to _hashlib.c; build-system drop of OpenSSL 1.1.x support.