Encrypted repositories

Encrypted repositories

[Enrico Weigelt]

Hi,

I'm currently planning to implement an strong encryption in git
(not like gitcrypt, but with encrypted blobs, directories, etc,
directly in the core).

The idea goes like this:

* blobs are encrypted with their (original) content hash as
  encryption keys
* directory objects only hold randomized filenames and pointers
  to the encrypted blob (content hash of the encrypted data)
* new ext-directory objects are holding a mapping of the
  randomized file names to the real ones and the encryption keys,
  stored encrypted similar to the blobs
* ext-directory object is referenced by a special filename in
  the directory object.
* commit objects also hold an encrypted section (eg. uuencoded)
  with the ext-directory node's key, additional commit text, etc,
  itself encrypted with the repository key

This way, the lowlevel / bare repository operations (including
remote sync and gc) should continue to work, while only actual
access (eg. checkout or commit) need to be changed and have the
repository key available.


What do you think about this approach ?


cu
-- 
Mit freundlichen Grüßen / Kind regards 

Enrico Weigelt 
VNC - Virtual Network Consult GmbH 
Head Of Development 

Pariser Platz 4a, D-10117 Berlin
Tel.: +49 (30) 3464615-20
Fax: +49 (30) 3464615-59

enrico.weigelt@vnc.biz; www.vnc.de 

Re: Encrypted repositories

[Junio C Hamano]

Enrico Weigelt <enrico.weigelt@vnc.biz> writes:

> * blobs are encrypted with their (original) content hash as
>   encryption keys

What does this even mean?

Is it expected that anybody who has access to the repository can
learn names of objects (e.g. by running "ls .git/objects/??/")? If
so, from whom are you protecting your repository?

How does this encryption interact with delta compression employed in
pack generation?

Re: Encrypted repositories

[Enrico Weigelt]

Hi,

> Enrico Weigelt <enrico.weigelt@vnc.biz> writes:
> 
> > * blobs are encrypted with their (original) content hash as
> >   encryption keys
> 
> What does this even mean?
> 
> Is it expected that anybody who has access to the repository can
> learn names of objects (e.g. by running "ls .git/objects/??/")? If
> so, from whom are you protecting your repository?

Well, everybody can access the objects, but they're encrypted,
so you need the repo key (which, of course isn't contained in
the repo itself ;-p) to decrypt them.

The whole tree will still be consistent even without encryption
support (so, gc etc shouldn't break), but to actually _use_ the
repo (eg. checkout or adding new commits), you'll need the
encryption support and the repo key (well, committing should
theoretically even work with diffrent repo key, even this doesn't
make much sense ;-)).

> How does this encryption interact with delta compression employed
> in pack generation?

Probably not at all ;-o
For the usecases I have in mind (backups, filesharing, etc) this
wouldn't hurt so much, if the objects are compressed before encryption.


cu
-- 
Mit freundlichen Grüßen / Kind regards 

Enrico Weigelt 
VNC - Virtual Network Consult GmbH 
Head Of Development 

Pariser Platz 4a, D-10117 Berlin
Tel.: +49 (30) 3464615-20
Fax: +49 (30) 3464615-59

enrico.weigelt@vnc.biz; www.vnc.de 

Re: Encrypted repositories

[Junio C Hamano]

Enrico Weigelt <enrico.weigelt@vnc.biz> writes:

>> Enrico Weigelt <enrico.weigelt@vnc.biz> writes:
>> 
>> > * blobs are encrypted with their (original) content hash as
>> >   encryption keys
>> 
>> What does this even mean?
>> 
>> Is it expected that anybody who has access to the repository can
>> learn names of objects (e.g. by running "ls .git/objects/??/")? If
>> so, from whom are you protecting your repository?
>
> Well, everybody can access the objects, but they're encrypted,
> so you need the repo key (which, of course isn't contained in
> the repo itself ;-p) to decrypt them.

So, in short, blobs are not encrypted with the hash of their
contents as encryption keys at all.

>> How does this encryption interact with delta compression employed
>> in pack generation?
>
> Probably not at all ;-o
>
> For the usecases I have in mind (backups, filesharing, etc) this
> wouldn't hurt so much, if the objects are compressed before encryption.

For that kind of usage pattern, you are better off looking at
encrypted tarballs or zip archives.

Re: Encrypted repositories

[Enrico Weigelt]

> > Well, everybody can access the objects, but they're encrypted,
> > so you need the repo key (which, of course isn't contained in
> > the repo itself ;-p) to decrypt them.
> 
> So, in short, blobs are not encrypted with the hash of their
> contents as encryption keys at all.

No, the blobs are encrypted with their content hash as key, and the
encrypted blob will be stored with it's content hash as object id.

> > For the usecases I have in mind (backups, filesharing, etc) this
> > wouldn't hurt so much, if the objects are compressed before
> > encryption.
> 
> For that kind of usage pattern, you are better off looking at
> encrypted tarballs or zip archives.

No, that doesn't give us anything like history, incremental
synchronization, etc, etc.

What I finnaly wanna has is a usual git, just with encryption,
but I can live with loosing differential compression.


cu
-- 
Mit freundlichen Grüßen / Kind regards 

Enrico Weigelt 
VNC - Virtual Network Consult GmbH 
Head Of Development 

Pariser Platz 4a, D-10117 Berlin
Tel.: +49 (30) 3464615-20
Fax: +49 (30) 3464615-59

enrico.weigelt@vnc.biz; www.vnc.de 

Re: Encrypted repositories

[David Aguilar]

On Fri, Sep 7, 2012 at 8:34 PM, Enrico Weigelt <enrico.weigelt@vnc.biz> wrote:
>
>> > Well, everybody can access the objects, but they're encrypted,
>> > so you need the repo key (which, of course isn't contained in
>> > the repo itself ;-p) to decrypt them.
>>
>> So, in short, blobs are not encrypted with the hash of their
>> contents as encryption keys at all.
>
> No, the blobs are encrypted with their content hash as key, and the
> encrypted blob will be stored with it's content hash as object id.
>
>> > For the usecases I have in mind (backups, filesharing, etc) this
>> > wouldn't hurt so much, if the objects are compressed before
>> > encryption.
>>
>> For that kind of usage pattern, you are better off looking at
>> encrypted tarballs or zip archives.
>
> No, that doesn't give us anything like history, incremental
> synchronization, etc, etc.
>
> What I finnaly wanna has is a usual git, just with encryption,
> but I can live with loosing differential compression.

Something like this?

https://gist.github.com/873637

I've never tried it myself, who knows if it works, but google found it
when I searched for "git clean smudge filter encryption".

I hope that helps,
-- 
David