For anyone who wants to reproduce this - the magic commands you have to use to make libraries like libpcap-devel available on RHEL9 are (at least for me):

subscription-manager config --rhsm.manage_repos=1
subscription-manager repos --enable codeready-builder-for-rhel-9-$(arch)-rpms

This is the backtrace of the command line mentioned in the ticket, after enabling FIPS-mode with fips-mode-setup --enable:

#0  0x00007ffff6e1f54c in __pthread_kill_implementation () from /lib64/libc.so.6
#1  0x00007ffff6dd2ce6 in raise () from /lib64/libc.so.6
#2  0x00007ffff6da67f3 in abort () from /lib64/libc.so.6
#3  0x000000000108040b in zeek::Reporter::InternalError (this=<optimized out>, fmt=fmt@entry=0x23e832f "EVP_DigestInit failed") at ../src/Reporter.cc:236
#4  0x0000000000f5fc0e in zeek::detail::hash_init (alg=alg@entry=zeek::detail::Hash_MD5) at ../src/digest.cc:48
#5  0x0000000000f5fc77 in zeek::detail::calculate_digest (out=0x2bedab0 <zeek::detail::KeyedHash::shared_hmac_md5_key> "", len=64,
    data=0x7fffffffd980 "\231\255\"\212dx\303T\017\021,I\037w\254\264\371\355\376Іbo\244G\340Z\320N\\8ԃ\303\062t\356P\201A\227)\261\366\002\200YR\350\257\351\307\374\216h\250\322\065 l\270+|j\326\354\266m\273\221D\312\032W+`i+", <incomplete sequence \306>, alg=zeek::detail::Hash_MD5) at ../src/digest.cc:81
#6  zeek::detail::internal_md5 (
    data=data@entry=0x7fffffffd980 "\231\255\"\212dx\303T\017\021,I\037w\254\264\371\355\376Іbo\244G\340Z\320N\\8ԃ\303\062t\356P\201A\227)\261\366\002\200YR\350\257\351\307\374\216h\250\322\065 l\270+|j\326\354\266m\273\221D\312\032W+`i+", <incomplete sequence \306>, len=len@entry=64, out=out@entry=0x2bedab0 <zeek::detail::KeyedHash::shared_hmac_md5_key> "")
    at ../src/digest.cc:69
#7  0x000000000106b773 in zeek::detail::KeyedHash::InitializeSeeds (seed_data=...) at /usr/include/c++/11/array:261
#8  0x0000000000f627aa in zeek::util::detail::init_random_seed (read_file=<optimized out>, write_file=0x0, use_empty_seeds=<optimized out>) at ../src/util.cc:486
#9  0x0000000000f938c5 in zeek::detail::setup (argc=argc@entry=3, argv=argv@entry=0x7fffffffe4b8, zopts=zopts@entry=0x0) at ../src/zeek-setup.cc:634
#10 0x0000000000e38fe2 in main (argc=3, argv=0x7fffffffe4b8) at ../src/main.cc:23

I tried taking a stab at this today - and unless I am missing something, there might not be an easy workaround anymore to get MD5 to work with OpenSSL.

If true, I am not entirely sure what this means for us. We could add our own MD5 library. However, while that is a technical workaround, I am not sure if this might render the system no longer fips compliant - I am not sure there still is a caveout for using MD5 for non-security use in the current iteration of FIPS. If someone has any information on that, I would be curious.

The second alternative - which might be preferable - would be to make sure that Zeek is able to run without requiring the use of MD5.

The second alternative - which might be preferable - would be to make sure that Zeek is able to run without requiring the use of MD5.

I like this one. Are we losing anything by dropping it?

Replacing MD5 in the IP address anonymization routines would be a breaking change for users that supply their own HMAC seeds for predictable outputs for the same inputs across restarts. I've never seen anonymization used this way, only ever in anonymizing the output of a PCAP for sharing, but it's worth knowing.

I am not sure there still is a caveout for using MD5 for non-security use in the current iteration of FIPS.

It's my understanding that it's still permitted to use MD5 for non-security applications in 140-3. The above example, though, would qualify as a security application, as it's used to protect identifiable information.

Installing packages while FIPS is enabled can be addressed by using the -fips context flag, similar to how OpenSSL's own X509 implementation does it in src/digest.cc when OpenSSL >= 3.0.0 is in use.

Plugins that use MD5Vals or scripts that use the md5_hash methods would still be broken in FIPS mode, as those rely on MD5_Init for OpenSSL >= 3.0.0 because OpaqueVals need to have their entire state serializable.

I'm putting this on the radar again, for 7.0. It also strikes me as the right direction to strip MD5 functionality from Zeek in this setting. This doesn't look too daunting per se — MD5 isn't particularly entrenched in the code — but it certainly brings up other technical/maintainability/UX questions.

+1 interest in FIPS compliance for Zeek.

Turns out there's a quick way to reproduce on Fedora, since its OpenSSL supports toggling via an env var:

$ zeek -e 'print md5_hash("zeek"); print sha1_hash("zeek"); print sha256_hash("zeek")'
cfc17d3ae3f66d99e60b5f6a2a93c239
e274c2b93298a8c267b5442538911f456b7058d9
6e349f5b2a04d4f0c538ea1e8dd8a95067fd2c32f150a152816f14dcfdae0364
$ OPENSSL_FORCE_FIPS_MODE=1 zeek -e 'print md5_hash("zeek"); print sha1_hash("zeek"); print sha256_hash("zeek")'
internal error: EVP_DigestInit failed
Aborted (core dumped)