I agree that preserving an existing registrymodifications.xcu is desirable.

There's no problem so long as the AOO 3.4 install is set to have the UseSHA1 
and UseBlowfish options by default when there is no setting in 
registrymodifications.xcu.

If you are confident patching Common.xcs is sufficient for all of the paths 
where the default is not over-ridden, let's do it.  It is probably easy to test 
whether there is an uncovered path to the default setting in the Save Options 
class constructor.

 - Dennis

-----Original Message-----
From: Juergen Schmidt [mailto:[email protected]] 
Sent: Tuesday, March 27, 2012 00:46
To: [email protected]
Subject: Re: [RELEASE,CODE]: Bug 119090 - Default Encryption Fails for 
Down-Level Implementations

On 3/27/12 3:48 AM, Dennis E. Hamilton wrote:
> I confirmed my hypothesis.  When AOO 3.4 is installed over the top of an 
> existing (i.e., OpenOffice.org 3.3.0) installation, it does not updated the 
> registrymodifications.xcu that is already there.  Since there are no settings 
> of options for Save As Password use of SHA1 and Blowfish there, none are 
> there after the AOO 3.4 install.  That means only the program-set defaults 
> will kick in and the user will be converted to "Save with Password" using 
> SHA256/K checksums and AES256 CBC encryption.
>

The update installs over an existing user installation where a user have 
made some local changes that we don't overwrite. Everything else would 
be probably surprising to the users who have potentially changed a lot 
of default values.

I think the behaviour is want the user expect. And the config entries 
are read from the configuration anyway independent from the 
initialization in the code. So either the default values from Common.xcs 
(finallly main.xcd) are used or the overwritten value in the user config 
layer.

I don't see a problem here.

Juergen


> I verified this with AOO 3.4 r1303653 atop OO.o 3.3.0.
>
>   - Dennis
>
> PS: I also confirmed that LibreOffice 3.5.0rc3 is adding chaff to XML files 
> that are compressed and encrypted, preventing easy access to known plaintexts 
> for attacking the encryptions in the ODF package.  (There is a discussion of 
> chaff, among other technicalities 
> at<https://issues.apache.org/ooo/show_bug.cgi?id=119090>.)
>
> -----Original Message-----
> From: Dennis E. Hamilton [mailto:[email protected]]
> Sent: Monday, March 26, 2012 17:16
> To: [email protected]
> Cc: [email protected]; [email protected]
> Subject: RE: [RELEASE,CODE]: Bug 119090 - Default Encryption Fails for 
> Down-Level Implementations
>
> I did more experiments with AOO-dev 3.4 and LO 3.5.0rc3 which I happened to 
> have installed where it was easy to test them together.
>
> TEST RESULTS
>
>   1. It is possible to change the default behavior of AOO-dev 3.4 and LO 
> 3.5.0rc3 (both of which produce AES 256 CBC and SHA256-1k encryptions by 
> default) by setting options in registrymodifications.xcu.
>
>   2. If registrymodifications.xcu is deleted, a new one is created *but* it 
> has *no* settings for the SHA1 and Blowfish in ODF12 and these installations 
> *revert* to AES256 CBC and SHA256-1k even if their last use was with options 
> set for Blowfish CFB and SHA1/1K.
>
> HYPOTHESIS **CONFIRMED**: If an install is done on top of a previous 
> installation not supporting AES to update to a later version, no settings for 
> this will be added to the "legacy" registrymodifications.xcu and the default 
> will go into effect: encryptions will start being done in AES256, surprise, 
> surprise.
>
> RECOMMENDATION:
>
>   1. It looks like registrymodification.xcu is the place where a tool or 
> script can do the job when it comes to setting/changing the desired option.
>
>   2. It looks like there must be code changes to set the default to Blowfish 
> and SHA1/1K within the application to cover the case where 
> registrymodification.xcu doesn't specify an option either way.
>
>   This last may be in Common.xcs but I am betting that the assured default 
> setting is in the constructor initial values in savopt.cxx.  Why?  Because 
> that class holds the options and setters and getters for them.  Other 
> software uses the setters when processing configuration parameters from 
> elsewhere, with the default value delivered by the getter when no 
> configuration parameter provides a change.  My money is on that being the 
> place.
>
>   - Dennis
>
> -----Original Message-----
> From: Dennis E. Hamilton [mailto:[email protected]]
> Sent: Sunday, March 25, 2012 18:15
> To: [email protected]
> Cc: [email protected]
> Subject: RE: [RELEASE,CODE]: Bug 119090 - Default Encryption Fails for 
> Down-Level Implementations
>
> TJ,
>
> I was doing some nosing around and, based on some information on the 
> Community Forums (thank you Hagar), it looks like the settings are controlled 
> in a file called registrymodifications.xcu, at least on Windows.  The 
> location will vary with different versions of windows.
>
> On windows, you can find one under the installed-user profile, such as 
> Documents&  Settings\orcmid\Application Data [a hidden file], 
> OpenOffice/3/user/registrymodification.xcu for any install since the AES256 
> has been instituted as default.  the *.xcu is actually an XML file and you 
> can find the settings by searching for "blowfish" and for "SHA1".
>
> How this works for Mac, Solaris, OS/2, and the various Linus and BSD builds, 
> I have no idea.
>
>   - Dennis
>
> -----Original Message-----
> From: TJ Frazier [mailto:[email protected]]
> Sent: Friday, March 23, 2012 11:26
> To: [email protected]
> Subject: Re: [RELEASE,CODE]: Bug 119090 - Default Encryption Fails for 
> Down-Level Implementations
>
> [ ... ]
>
> ... options to consider:
>
> 3. User change to config file, to use the new option.
>
> I have suggested a writeup on this, but such instructions are much
> better aimed at the (few?) users who want the "latest and greatest"
> security option, and will do a little work to get it. (Does anybody know
> what that file name is? Given that, I volunteer to update the Release
> Notes.)
>
> 4. Macro to toggle the settings.
>
> This could be distributed in a BASIC library (new or existing); no
> extension necessary. User instructions to find and run the macro are
> simple. I may be able to write this; preliminary investigation is
> promising but not certain. I volunteer to try. There are several real
> experts on this list, whom I might ask for help.
>
> /tj/
>>
>>
>>
>> [1] https://issues.apache.org/ooo/show_bug.cgi?id=119090
>>
>> On 19.03.2012 14:48, Jürgen Schmidt wrote:
>>> On 3/19/12 2:16 PM, TJ Frazier wrote:
>>>> On 3/19/2012 08:48, Jürgen Schmidt wrote:
>>>>> Hi,
>>>>>
>>>>> I think issue 119090 is no show stopper from my point of view. The new
>>>>> default provides a better security than before when I understand it
>>>>> correct. And if people detect potential problems they can save the
>>>>> document again with other settings.
>>>>>
>>>>> I agree that this is important for interoperability but no show
>>>>> stopper.
>>>>>
>>>>> Any other opinion?
>>>>>
>>>>> Juergen
>>>>>
>>>>>
>>>> Hi, Jürgen,
>>>>
>>>> Like Dennis, I'm nervous about this. Perhaps we can handle it with a
>>>> mention in the Release Notes; something like,
>>>>
>>>> PLEASE NOTE: the default options for [technical details here] should
>>>> provide your best /individual/ security. However, if you intend to share
>>>> the document in secure fashion, the default mode cannot be read by
>>>> * previous versions of OpenOffice.org
>>>> * current versions of LibreOffice, at least through [version]
>>>> * Ms Office [version info]
>>>> For compatibility, use the options [details here].
>>>>
>>>
>>> I agree that it make sense to mention it in the release notes.
>>>
>>> Any volunteer for updating the release notes?
>>>
>>> Juergen
>>
>>
>
>

Reply via email to