GH-161739: Don't test PBKDF2 HMAC on FIPS with len(salt) < 16 - #161748

Does it only fail on OpenSSL 4.x

Actually, I only reproduced it on one specific version of 4.x, but that is my understanding, yes.

Mmh. That's weird that it's only in OpenSSL 4.x and not in others, because this looks like it means that the salt validation was not done previously.

I'm double checking.

My testing is heavily based on using OPENSSL_FORCE_FIPS_MODE=1. There, it only happens with OpenSSL 4+ but that does not necessarily mean that a real FIPS environment would not failed even on older SSL. Let's just remove the version information from news then.

On Fedora 45 with OpenSSL 4.0.3 29 Sep 2026, OPENSSL_FORCE_FIPS_MODE=1 ./python -m test -v test_hashlib -m test_pbkdf2_hmac_c fails with:

ERROR: test_pbkdf2_hmac_c (test.test_hashlib.KDFTests.test_pbkdf2_hmac_c)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/vstinner/python/main/Lib/test/test_hashlib.py", line 1362, in test_pbkdf2_hmac_c
    self._test_pbkdf2_hmac(openssl_hashlib.pbkdf2_hmac, openssl_md_meth_names)
    ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/vstinner/python/main/Lib/test/test_hashlib.py", line 1316, in _test_pbkdf2_hmac
    out = pbkdf2(digest_name, password, salt, rounds, dklen)
ValueError: [Provider routines] invalid iteration count

The test pass with sha1 b'passwordPASSWORDpassword' b'saltSALTsaltSALTsaltSALTsaltSALTsalt' 4096 25, but fails with sha1 b'password' b'saltsaltsaltsalt' 1 None. It seems like dklen=None is rejected in FIPS mode.

NIST SP 800-132 actually specifies 1000 as iteration counts as well I think and kLen should be 112 (i think dklen is klen in the NIST guidelines), see https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-132.pdf