[sword-devel] How to validate a Sword module unlock key?

Wed Jan 15 11:10:03 MST 2020

Hi!

As has been said before, if somebody really wants to crack an unlock key there are still ways to do that ... so, not having the validation function is then more a lack of usability than actual security.
What's the conclusion to this discussion?
Given the feedback that I saw I would still suggest to move forward with implementing the functionality based on Jaak's proposal.

Best regards,
Tobias

Am 13. Januar 2020 19:41:36 MEZ schrieb Greg Hellings <greg.hellings at gmail.com>:
>On Mon, Jan 13, 2020, 11:28 Troy A. Griffitts <scribe at crosswire.org>
>wrote:
>
>> Hello all,
>>
>> While the request at face value seems reasonable, let me explain a
>bit of
>> the history behind not having a method like this.
>>
>> The way we've recommended in the past for frontends to build a UI
>> interface for unlocking is to show some entry in the Bible to the
>user
>> after deciphering and ask them if it looks OK.  This might seem
>kludgy, but
>> it has the same effect as one of those stupid captchas-- it require
>human
>> interaction upon each decipher confirmation.
>>
>> The reasoning, as you've probably guessed, is to make it more
>difficult to
>> brute force an unlock key.
>>
>> This may or may not be important or effective, but it has been the
>policy
>> up until now.
>>
>> Practically, what happens these days is that deciphering has been
>> relocated to before decompression, and most enciphered modules use
>> compression, so the end result is that that the decompression filter
>will
>> throw a 'corrupt data during decompression' error and return a 0
>length
>> entry-- which can be used for brute force key guessing.
>>
>
>The last time I looked into this, it wasn't just practice. Enciphering
>just
>plain didn't work on uncompressed modules. But that was a while ago.
>
>So, we're in a state where we don't officially provide a means to
>> programmatically test an unlock key (for reasons stated above), but
>in
>> practice, an empty buffer returned for, say John.3.16, would give a
>> reasonable check for an invalid unlock key, or a check for a valid
>UTF-8
>> stream would also be a reasonable programmatic check.
>>
>
>What about a module like LXX or WLC? I realize those aren't under lock
>and
>key, but they're informative that checking for zero length in a
>particular
>reference isn't foolhardy. Conversely if it's Genesis 1:1, what about
>NA28?
>Particularly of note in commentary modules which could be limited to
>one
>book, alone.
>
>So, while in principle, I believe it's a good thing to force user input
>> into unlock confirmation to discourage brute force guessing, we don't
>do a
>> good job with the implementation right now.
>>
>
>Being a programmatic API, I'm not sure we can provide anything to
>prevent
>brute force attacks. If I know even a single word of any reference in
>the
>work, it becomes trivial to attack by just searching for the word's
>presence in the resulting output. As long as we are using a
>cryptographic
>algorithm that is strong against known plain text attacks, we can leave
>attacks up to the field of crypt analysis and be happy that we've done
>our
>due diligence.
>
>Jaak's proposal is reasonable and prevents known plain text attacks by
>resorting to a hash, but it makes maintaining the config file more of a
>pain. Storing the text in a module file field solves the maintenance
>problem with the conf file but could cause problems with existing
>content,
>depending on how exactly the file format is laid out - an area you're
>the
>expert in.
>
>--Greg
>
>> Thoughts?
>>
>> Troy
>> On 1/12/20 11:42 PM, Tobias Klein wrote:
>>
>> I like this idea, Jaak! :)
>>
>> Can we implement this in the Sword engine with the next release that
>also
>> delivers the "individualized unlock key function"? Ideally directly
>with a
>> convenient API function that has the purpose to validate a given
>unlock
>> key, with a signature like this:
>>
>> *bool isSwordUnlockKeyValid(std::string key)*
>>
>> In my view, having a mechanism for validating the unlock key is
>essential
>> for having a professional unlock frontend. Without the availability
>of such
>> a mechanism I see the following issues:
>>
>> - Users need to go through full installation of a module before
>knowing
>> that the unlock key they entered works. This is a rather lengthy
>feedback
>> loop.
>>
>> - Since there is a possibility for input errors when entering the
>key, the
>> frontend must provide extra functions to "correct the key" after the
>> installation has already happened (this wouldn't be necessary with a
>> validation function).
>>
>> Best regards,
>> Tobias
>> On 1/12/20 11:46 PM, Jaak Ristioja wrote:
>>
>> Hi!
>>
>> On 12.01.20 20:53, Greg Hellings wrote:
>>
>> On Sun, Jan 12, 2020 at 10:32 AM Tobias Klein <contact at tklein.info>
><contact at tklein.info> wrote:
>>
>>
>> Hi,
>>
>> I'm adding Sword module unlock support to Ezra Project and I've been
>> wondering how you would validate a given unlock key?
>>
>> Basically the dialog for entering the unlock key is shown when a
>locked
>> module is selected for installation. Before going through the effort
>of
>> installing a module I would like to make sure that the given unlock
>key
>> actually works with the selected module. Is there something in the
>SWORD
>> API that supports the validation of the unlock key entered by the
>user?
>>
>>
>> The last time this came up, I believe the answer was that you just
>have to
>> try it and display it to the user and they have to decide if the
>results
>> are human readable.
>>
>> It would be possible to include a field in modules with a known-good
>value,
>> then the API could test if that value matched what was expected when
>it was
>> decrypted. Unless that functionality already exists, I don't know of
>any
>> other way you could accomplish this.
>>
>> I've thought about this many times myself and as far as I know Greg
>is
>> right that there is currently no other way besides trial and error to
>> verify the unlock key.
>>
>> Greg: Do I understand you correctly, that there would need to be an
>> extra field in every such module, and extra logic must be added to
>SWORD
>> so that this extra field does not show up in frontends? If this is
>so,
>> it might slightly break compatiblity of modules with older versions
>of
>> SWORD which do not contain such enhancements.
>>
>> As an alternative, I suggest for consideration the following
>approach:
>>
>>
>> Add in the module configuration file the two extra pieces of
>> information (presented here as two configuration options with bad
>names):
>>
>>   UnlockKeyVerifyValue=<Some sufficiently long random ASCII string>
>>   UnlockKeyVerifyHash=<Hash of field value>
>>
>> When a newer version of SWORD detects these configuration options in
>the
>> module configuration, it can verify the unlock key using the
>following
>> algorithm:
>>
>>   1) Decrypt the value of the UnlockKeyVerifyValue configuration
>option
>> (after whitespace trimming) with the unlock key
>>   2) Verify that the hash of the value decrypted in step 1 matches
>the
>> value of the UnlockKeyVerifyHash configuration option.
>>
>>
>> Pros:
>>  * Modules can easily be amended by adding new entries to their
>> configuration files.
>>  * No extra field in the module text is needed, so modules amended
>with
>> these configuration options will continue to work with older versions
>of
>> SWORD.
>>  * Anyone with the key can generate this verification information.
>>  * Only access to the module configuration file is needed to verify
>the
>> unlock key, so no expensive seeking/reading/parsing the encrypted
>module
>> content is necessary.
>>  * Doesn't too leak much about the key.
>>
>> Cons:
>>  * A hash function must be implemented, but I think this would not
>need
>> to be cryptographically secure, but would act more like checksum, so
>> even something as simple as CRC-32 might do.
>>
>>
>> Notes:
>>  * Another alternative would be to use a ciphertext/plaintext pair
>> instead so that no checksum/has must be implemented at all, but this
>> might potentially leak too much about the key, and will likely
>require
>> the configuration options to include binary values (i.e.
>> escaping/encoding would be needed).
>>  * Another alternative would be to decrypt and verify a field from
>the
>> encrypted module itself, but reading the ciphertext from the module
>file
>> might be a more expensive operation.
>>
>>
>> Hope this helps.
>>
>>
>> Best regards,
>> J
>>
>> _______________________________________________
>> sword-devel mailing list:
>sword-devel at crosswire.orghttp://www.crosswire.org/mailman/listinfo/sword-devel
>> Instructions to unsubscribe/change your settings at above page
>>
>>
>> _______________________________________________
>> sword-devel mailing list:
>sword-devel at crosswire.orghttp://www.crosswire.org/mailman/listinfo/sword-devel
>> Instructions to unsubscribe/change your settings at above page
>>
>> _______________________________________________
>> sword-devel mailing list: sword-devel at crosswire.org
>> http://www.crosswire.org/mailman/listinfo/sword-devel
>> Instructions to unsubscribe/change your settings at above page

-- 
Message sent from my phone. Please excuse brevity.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.crosswire.org/pipermail/sword-devel/attachments/20200115/3289b308/attachment-0001.html>