Use FIPS-safe hashes for program cache keys (#2087)

* Use FIPS-safe hashes for program cache keys

Signed-off-by: Aryan <aryansputta@gmail.com>

* Fix FIPS cache test linting

Signed-off-by: Aryan <aryansputta@gmail.com>

* Use benchmarked FIPS-safe cache hashing

Signed-off-by: Aryan <aryansputta@gmail.com>

* Clarify SHA-384 cache digest comment

Signed-off-by: Aryan <aryansputta@gmail.com>

* Switch program cache hashing back to sha256

* Remove benchmark helper script

---------

Signed-off-by: Aryan <aryansputta@gmail.com>
Co-authored-by: Leo Fang <leof@nvidia.com>

Author: aryanputta

cuda_core/cuda/core/utils/_program_cache/_file_stream.py

@@ -422,11 +422,17 @@ def _path_for_key(self, key: object) -> Path:
         k = _as_key_bytes(key)
         # Hash the key to a fixed-length identifier so arbitrary-length user
         # keys never exceed per-component filename limits (typically 255 on
-        # ext4 / NTFS). With a 256-bit blake2b digest, the cache relies on
-        # cryptographic collision resistance for key uniqueness -- two
-        # distinct keys hashing to the same path is astronomically unlikely
-        # (~2^-128 with the 32-byte digest in use here).
-        digest = hashlib.blake2b(k, digest_size=32).hexdigest()
+        # ext4 / NTFS).
+        #
+        # FIPS: must use a FIPS-approved hash algorithm. FIPS-enforcing
+        # systems can disable non-approved hashlib algorithms (for example
+        # blake2b) at the OpenSSL level. See #2043.
+        #
+        # With a 256-bit SHA-256 digest, the cache relies on collision
+        # resistance for key uniqueness -- two distinct keys hashing to the
+        # same path is astronomically unlikely (~2^128 practical collision
+        # work).
+        digest = hashlib.sha256(k, usedforsecurity=False).hexdigest()
         return self._entries / digest[:2] / digest[2:]
 
     # -- mapping API ---------------------------------------------------------

cuda_core/cuda/core/utils/_program_cache/_keys.py

@@ -35,7 +35,7 @@
 )
 
 # Bump when the key schema changes in a way that invalidates existing caches.
-_KEY_SCHEMA_VERSION = 1
+_KEY_SCHEMA_VERSION = 2
 
 _VALID_CODE_TYPES = frozenset({"c++", "ptx", "nvvm"})
 _VALID_TARGET_TYPES = frozenset({"ptx", "cubin", "ltoir"})
@@ -768,7 +768,10 @@ def make_program_cache_key(
     option_bytes = backend.option_fingerprint(options, target_type)
     name_tags = backend.encode_name_expressions(name_expressions)
 
-    hasher = hashlib.blake2b(digest_size=32)
+    # IMPORTANT: Must use a FIPS-approved hash algorithm (SHA-2 family).
+    # FIPS-enforcing systems can disable non-approved hashlib algorithms
+    # (for example blake2b) at the OpenSSL level. See #2043.
+    hasher = hashlib.sha256(usedforsecurity=False)
 
     def _update(label: str, payload: bytes) -> None:
         hasher.update(label.encode("ascii"))

cuda_core/tests/test_program_cache.py

@@ -1773,7 +1773,7 @@ def test_filestream_cache_size_cap_counts_tmp_files(tmp_path):
 
 def test_filestream_cache_handles_long_keys(tmp_path):
     """Arbitrary-length keys must not overflow per-component filename limits.
-    The filename is a fixed-length 256-bit blake2b digest; key uniqueness
+    The filename is a fixed-length 256-bit digest; key uniqueness
     relies on the digest's collision resistance."""
     from cuda.core.utils import FileStreamProgramCache