December 18 meeting minutes

Attendees:

Old Business

• Open hours (Jon)

  • I’m hoping to get this started in January. The basic principle is sound, it’s just a matter of getting the details set for an initial process (which could be modified over time).

New Business

• Removing deprecated features from OpenSSL 4.0 (Matt)

  • ENGINEs #28548  

    • Announcement on the Library blog 

    • Adds a lot of complexity to the code, which is a maintenance burden and complexity is the enemy of security. (Matt)

    • Stubs can be included at application build time which allows applications that gracefully handle ENGINE failure to compile. (Matt)

    • Already merged to master

  • Custom ciphers #29299

    • Three types of custom ciphers:

      • Providers (unchanged)

      • Global EVP ciphers (unchanged)

      • EVP_CIPHER_meth allows people to construct their own custom ciphers (removed)

    • Removes more code than you might imagine which can only be reached when using EVP_CIPHER_meth, which is a fairly niche thing to do.

    • Already merged to master

  • Custom MDs #29344

    • Pretty much the same as the cipher methods, but for message digest (MD) functions

    • Merged to master

  • Custom PKEY #29384

    • Similar to the above but for PKEY

  • Custom ASN1

    • Not yet removed from master, but there is a PR

    • Remove ASN1 from the public API, but not internally where it’s still used.

    • This doesn’t introduce as much complexity (no #ifdef LEGACY, etc.)

    • So the plan is to remove this eventually? (Dmitry) Maybe (Matt)

    • Plans to get things ready to remove entirely next year. (Richard)

  • atexit cleanup #29385

    • libcrypto uses this to free up memory on exit. It was added early on, but causes all sorts of problems.

      • Some applications might dynamically load libcrypto and unload libcrypto later, which causes a crash on exit.

    • Plan is to not register an atexit handler at all.

      • Memory won’t be freed when libcrypto is unloaded (on exiting the program, for instance).

      • Some memory checker (Valgrind, for instance) complain, though it’s not really a problem.

      • If an application wants to clean up memory, it can call OPENSSL_cleanup.

    • Dmitry has seen reports of this problem and this seems reasonable.

    • There are still potential problems with OPENSSL_cleanup, but it can be useful if you know what your are doing. (Tomas)