I recall I was looking into this ~ two months back ... not sure I covered (all of) these, could you be more specific in terms on Ruby OpenSSL code ? ... also try out a snapshot of a gem from: https://oss.sonatype.org/content/repositories/snapshots/rubygems/jruby-openssl/ ... although it's a bit outdated (still should include most of the related changes) but I'm sure @mkristian will push us another or let me know if I can ... thanks!

just pushed a new snapshot. @kares you can push a snapshot anytime -
sonatype will cleanup things after some time.

I tried the version from this snapshot. It seems to produce the same padding inconsistencies. If I turn padding off explicitly it does work so it's really just about the consistent default with MRI.

@iamasmith thanks ... we would appreciate concrete code samples (cipher names) that are inconsistent

Sure, here's the POC Blowfish wrapper where I discovered the inconsistency, it's part of a closed network RPC transport that I developed so it's designed to process presentation protocol frames and is paired against each session.

The module provides a pluggable abstraction of the encryption setup and use to my RPC pieces so I can tell my RPC to use any encryption engine.

Points of note, resetting the Cipher I found necessary when interfacing this to a Java client class that does the same thing since java's doFinal() resets the Cipher, that's more around Ruby to Java differences, is O/T but I thought I would explain that reset.

The critical thing to make this work between Ruby and JRuby in test was to have the .padding = 0 for the Cipher since Ruby assumes padding off for Blowfish OFB and JRuby assumes it on.

# Blowfish cryptography wrapper.

require 'openssl'
require 'securerandom'

class BlowfishCrypto

  def initialize(key)
    @key = key
    @c_enc = nil
    @c_dec = nil
  end

  # Produces encrypted serial stream as string
  def encrypt(data)
    begin
      # On first encrypt we generate the salt, cipher and key and we send the salt
      salt = ''
      unless @c_enc
        salt = SecureRandom.random_bytes(8)
        @c_enc = OpenSSL::Cipher::Cipher.new('bf-ofb').send(:encrypt)
        # Padding is NOT significant for MRI Ruby, it is always turn off for OFB
        # It IS significant is using JRuby which will default it to be on.
        @c_enc.padding = 0
        @c_enc.pkcs5_keyivgen(@key,salt)
      end
      # Cipher must be reset because final doesn't reset it
      @c_enc.reset
      salt + @c_enc.update(data) << @c_enc.final
    rescue OpenSSL::Cipher::CipherError
      nil
    rescue TypeError
      nil
    end
  end

  # Decrypts data with passed key, traps and returns nil if key could not be used
  def decrypt(data)
    begin
      unless @c_dec
        salt, data = data.unpack('a8a*')
        @c_dec = OpenSSL::Cipher::Cipher.new('bf-ofb').send(:decrypt)
        @c_dec.padding = 0
        @c_dec.pkcs5_keyivgen(@key,salt)
      end
      @c_dec.reset
      @c_dec.update(data) << @c_dec.final
    rescue OpenSSL::Cipher::CipherError => e
      nil
    rescue TypeError => e
      nil
    end
  end

end

I just ran into this as well. I believe the issue is that OFB, CFB, CFB8, and CTR modes should use padding type of NoPadding. There's already a TODO to this effect at https://github.com/jruby/jruby-openssl/blob/master/src/main/java/org/jruby/ext/openssl/Cipher.java#L559-L562

I believe that ECB SHOULD have padding, despite the comment there already.

I opened a bug with BouncyCastle discussing this as well (bcgit/bc-java#291), but I believe the true failure is here in JRuby OpenSSL. Note that this is a incompatibility with MRI -- there's no way in MRI to add padding to CFB/OFB/CTR modes, since OpenSSL won't do so even if asked (cipher.set_padding = 1).

I'm happy to work on a fix if you agree with this.

@dgolombek makes sense - feel free to go for it