jewel / clearskies

Open source btsync clone

ClearSkies

ClearSkies is a sync program similar to DropBox, except it does not require a monthly fee. Instead, you set up shares between two or more computers and the sharing happens amongst them directly.

ClearSkies is inspired by BitTorrent Sync, but it has an open protocol that can be audited for security.

The protocol is layered in such a way that other applications can take advantage of it for purposes other than file sync.

The Protocol

The ClearSkies protocol has been documented and is in a draft state. It can be found in the protocol/ directory. protocol/core.md is a good starting place.

The protocol features:

• Simple-to-share access codes
• Read-write sync
• Read-only sharing
• Encrypted connections
• Shallow copy (do not sync certain files from peer)
• Subtree copy (only sync certain directories from peer)
• Streaming support
• Rsync file transfer (extension)
• Gzip compression (extension)
• Encrypted backup sharing to an untrusted peer (future extension)
• Media streaming (future extension)
• Photo thumbnails (future extension)

The protocol is designed to be a common base for other sync programs, so that they can interoperate with each other. For example, a hypothetical wifi-enabled MIDI piano could speak the protocol and thereby sync its saved files to the owner's computer or tablet, without the piano manufacturer needing to write any PC or tablet software.

Where's the code?

We are focusing our effort on making a C++ implementation, clearskies_core. The C++ library will be portable to a wide variety of operating systems, including Windows, Android and iOS.

The C++ daemon is being built for Android in this repository.

There is a proof-of-concept of the protocol that is written in ruby. It is currently out-of-date in relation to the latest protocol changes.

There is a python control library and also a Desktop GUI.

Debian/Ubuntu packages of the ruby proof-of-concept are also available.

Security

The software does not attempt to provide anonymity. Access code sharing is designed to reduce the impact of surveillance by using one-time codes by default, and using perfect forward secrecy on the wire.

Setup of a share is vulnerable to an active man-in-the-middle attack if the channel used to send the access code is insecure.

For example, if Bob sends Alice an access code over SMS, Eve can try to connect to Bob before Alice does. Alice will not be able to connect to the share. Eve can even create another share and issue the same access code to fool Alice into thinking she has connected to Bob.

It is believed that security-conscious users will automatically avoid this problem by sharing the access codes over secure channels.

Contributing

If you are a professional cryptographer with interest in this project, any feedback on the protocol is very welcome.

A major area that needs work is creating GUIs for each platform, such as GTK, Cocoa, QT, Android, iOS, browser-based, and a Windows program. GUIs do not need to be written in any particular language, since they can control the daemon using a simple JSON protocol, which is documented in protocol/control.md.

Issues and pull requests are welcome.

The project mailing list is on google groups. (It is possible to participate via email if you do not have a google account.)

clearskies's Issues

[feature] support time difference accounting in peers to make it more robust

It's a bit of a hassle to assume that all peers need to have the clocks synchronized.

Status of project

Is the protocol being worked on any more? I note that clearskies_core is now abandoned.

Daemon not starting after recent commit

Commit 9c2f4 (Bind TCP to IPv6) broke one of my Clearskies setups: Raspberry Pi on IPv4-only network.
My other setup on a desktop in a dual IPv4+IPv6 network works fine, so I can't say if the problem is the architecture of the network.

The previous commit runs fine.
Here's the error log.

12:35:21.287118246 error main> Exiting due to exception in 'main' thread: Address family not supported by protocol - socket(2)
12:35:21.287557228 error main> /home/cachapa/clearskies/lib/network.rb:20:in initialize' 12:35:21.287898214 error main> /home/cachapa/clearskies/lib/network.rb:20:innew'
12:35:21.288262200 error main> /home/cachapa/clearskies/lib/network.rb:20:in start' 12:35:21.288597186 error main> /home/cachapa/clearskies/lib/daemon.rb:62:inrun'
12:35:21.288930173 error main> /home/cachapa/clearskies/lib/daemon.rb:41:in daemonize' 12:35:21.289513149 error main> ./clearskies:147:in

'

uTP doesn't support IPv6

The uTP code needs to try and open an IPv6 SharedUDPSocket, as well as as another STUNClient and UTPSocket.

Also, the tracker protocol needs to support a separate port number for IPv6. It almost certainly will be a different port number than the IPv4 port, especially since the IPv4 port is assigned by the NAT table on the router.

Collaborate with Syncthing?

See discussion at https://github.com/calmh/syncthing/issues/105

Wire protocol chunk size

Size of chunks in the wire protocol be limited to a maximum to avoid DOS attacks on clients that reserve a buffer with the size of the chunk.

segfault on ARM in debian stable

On a sheevaplug running debian wheezy, the gnutls code segfaults. I have yet to determine if this is ARM specific.

Cannot load GNUTLS (Windows)

I was wondering how to get this section loaded on windows, I have tried installing gnutls (libgnutls26) with cygwin, have I missed anything?

https://github.com/jewel/clearskies/blob/master/lib/gnutls.rb

minor passphrase clarifications

The Password section starts with "Instead of generating the master key". It might be clearer to say "Instead of generating a random read-write PSK". (I don't think the term "master key" is used anywhere else?)

Second, "The SHA256 algorithm is applied twenty million times to generate the 256-bit encryption key". Does this mean the read-write PSK? Is the PSK actually 256-bits in the passphrase case, but 128-bits for random keys?

Also, I was thinking it might be good practice for users to include a unique(ish) identifier as part of their passphrase, such as an email or domain name, in order to avoid accidentally colliding with someone else's passphrase.

This is looking pretty cool -- keep it up!
-greg

Debian/Ubuntu packaging

I'd like to take a stab at packaging this for Debian/Ubuntu if nobody is working on that.
Sounds good?

Silly Question

Is there any documentation on how to use the cli? I can't seem to get the daemon to start. Every time I try to run any of the commands I get 'Daemon not running'. I ran the tests and I get:

test_0004_should detect deleted files(Scanner::finds files)
test_0002_should detect new files(Scanner::finds files)

I am not sure what to do. I am not too familiar with ruby but I would still like to contribute.

ipv6 not working correctly

Looks like host B is trying to connect to an ipv6 port but host A only opened it for ipv4:

HOST A:

tracker> Tracking with http://clearskies.tuxng.com/clearskies/track?id=ab7f710316924fed38f493aa14413442bf2db28c8542054eb50988dc1f596fef%40b44ddbaf4d241e348568a515810654c0&id=10e53688a92370ced39427d8af929217b5445aa40c3cb69c890906231f8bbd33%40b44ddbaf4d241e348568a515810654c0&tcp_port=48555&utp_port=46327

root@cereza:0:/home/piotr/devel/clearskies# lsof -Pni | grep 48555
ruby 19347 piotr 9u IPv4 1376349 0t0 TCP *:48555 (LISTEN)

HOST B:

tracker> New outgoing connection with A
tracker> Tracker discovered 10e53688a92370ced39427d8af929217b5445aa40c3cb69c890906231f8bbd33 b44ddbaf4d241e348568a515810654c0 utp A 46327
tracker> New outgoing connection with A
connection410> Opening socket to utp A 46327
connection410> Could not connect via utp to 2a02:8109:8440:8b8:f2de:f1ff:fe00:441b:46327: Address family not supported by protocol - sendto(2)
connection409> Opening socket to tcp A 48555
connection409> Could not connect via tcp to A:48555: Connection refused - connect(2)

I can look more into it tomorrow.

separate implementation and protocol in different repositories

Hi

as agreed we should separate the protocol spec and implementation in different repositories.

Improvement: use rolling checksums for files

If big files change often recalculating the sha256 hash and resending the whole file will be very inefficient. A rolling checksum mechanism like the one rsync uses would be much more efficient in case a large file is being modified constantly, for example when having on sync a directory where torrents are being downloaded.

http://rsync.samba.org/tech_report/node2.html

Local IP tracking, broadcast limitations

The broadcast mechanism is not sufficient for networks which have different disjoint broadcast domains. The peer should report to the tracker it's list of known local addresses, so other peers in the same private network can readily find them, even their broadcast domains are different.

GUI

Hi there,
I just found a quite nice Linux GUI for Bittorrent Sync released under GPLv2, I think you guys might use parts of the code:

https://github.com/tuxpoldo/btsync-deb

Just in case you didn't know about it already.

Do not use SHA-256 hashing

Because of bitcoin miners, we now have dedicated devices that can hash SHA-256 in the order of Terahash/s. Since this rate is rising fast, SHA-256 will soon be obsolete.

I don't know enough to suggest any algorithm with confidance, but I'd also avoid s-crypt as dedicated devices for that algorithm are coming this year. S-crypt-n should be fine, however.

Die properly if two instances are ran

If you run two clearskies instances you get a crashed thread, but clearskies continues to run. It should exit.

control> Thread crash: Daemon already running
control> /home/jake/repos/clearskies/lib/control_server.rb:23:in run' control> /home/jake/repos/clearskies/lib/control_server.rb:13:inblock in start'
control> /home/jake/repos/clearskies/lib/simple_thread.rb:72:in block (2 levels) in initialize' control> /home/jake/repos/clearskies/lib/simple_thread.rb:38:inglock'
control> /home/jake/repos/clearskies/lib/simple_thread.rb:70:in `block in initialize'

"incompatible character encodings" error with Ruby 2.0

$ ./clearskies debug
....
control_connection> Control error: incompatible character encodings: UTF-8 and ASCII-8BIT
control_connection> /home/badboy/code/clearskies/lib/access_code.rb:59:in `to_s'
/home/badboy/code/clearskies/lib/control_server.rb:115:in `handle_command'
/home/badboy/code/clearskies/lib/control_server.rb:59:in `block in serve'
/home/badboy/code/clearskies/lib/control_server.rb:53:in `loop'
/home/badboy/code/clearskies/lib/control_server.rb:53:in `serve'
/home/badboy/code/clearskies/lib/control_server.rb:40:in `block (2 levels) in run'
/home/badboy/code/clearskies/lib/simple_thread.rb:72:in `block (2 levels) in initialize'
/home/badboy/code/clearskies/lib/simple_thread.rb:38:in `glock'
/home/badboy/code/clearskies/lib/simple_thread.rb:70:in `block in initialize'

$ ./clearskies share $(pwd)/shared read-write
/home/badboy/code/clearskies/lib/control_client.rb:24:in `issue': server can't :create_access_code, says: Encoding::CompatibilityError incompatible character encodings: UTF-8 and ASCII-8BIT (RuntimeError)
    from ./clearskies:143:in `<main>'

In Ruby 2.0 the default encoding is UTF-8 for all files (if not specified differently). This way access_code fails expecting ASCII-8BIT where it is UTF-8.
You should specifiy that you're handling binary data.

Small issue with installation instructions

I had a bit of trouble installing Clear Skies on my Debian Unstable and Raspbian distos.

The Ubuntu packages suggested don't seem to exist in Debian, so I tried to use the gem installation method, but that method requires the "ruby-dev" package.
Googling the problem shows that this is also true for Ubuntu, so I propose adding that to the instructions to simplify the installation process for others.

My abridged instructions:
[sudo] apt-get install ruby ruby-dev
[sudo] gem install rb-inotify ffi

Issue with file deletion policy

https://github.com/jewel/clearskies/blob/master/protocol/core.md#deleted-files explains the file deletion policy which is used to avoid ghost copies.
However, that policy has a flaw where a file which has been deleted cannot be re-added to that share.

Testing with the latest client reveals this problem (machines A and B):

1. Add file test.txt to A
2. See file being correctly copied to B
3. Delete file on A
4. File is deleted on B
5. Re-add the file on A
6. File is not copied to B

Furthermore:
7. Copy the file manually to B
8. File is automatically deleted from A and B

In my view, this problem is as severe, if not worse than ghost copies, as it results in silent loss of data.

My proposal here would be to store the date of the file which was deleted along with the file name, and use it to decide whether any given file is a ghost or a new version.

Need to list required ports for firewalls

Could you add the required ports for opening up a firewall somewhere?

Detach command throws error

./clearskies detach ../aaaaa
clearskies/lib/control_client.rb:14:in []=': can't convert Symbol into Integer (TypeError) from clearskies/lib/control_client.rb:14:inissue'
from ./clearskies:246:in `

'

Donations

Hi, I think you should display a donation option more prominently. I found a bitcoin address in the clearskies_core repo, but not many people go there, and there isn't even a bitcoin logo or a colorful donate button, or something to catch the eyes.

I think this project is pretty awesome, and I think many people might support the efforts if you made it a bit easier. Adding a paypal option might not hurt either.

restart command throws error

./clearskies restart
./clearskies:149:in `

': uninitialized constant Daemon (NameError)

rb-inotify is broken in older ubuntus

The system version of rb-inotify keeps the global-vm-lock in older versions of Ubuntu. This may be a problem with FFI instead of rb-inotify.

The symptom is that the test as ran from rake locks up completely.

Parsing JSON before authentication opens for certain vulnerabilities

Very large JSON objects in the handshake could consume non-trivial amounts of memory and CPU on the server side.

Unclear what to do when access code or club is not found

In core.md:

The server then uses the username to see if it has a corresponding access code or club. If it does, it completes the connection. The password or key is given to the TLS library as lowercase hexadecimal.

It isn't clear what is supposed to happen if it does not have a corresponding access code or club.

No fork in JRuby

The following test cases fail because there is no fork in Jruby.

1. Error:
   test_0001_can_connect_and_send_data(GnuTLS::Session::acts as a client):
   NotImplementedError: fork is not available on this platform
   org/jruby/RubyKernel.java:1855:in fork' clearskies/test/gnutls.rb:16:inrun_test'
   org/jruby/RubyBasicObject.java:1735:in `instance_eval'

2. Error:
   test_0002_won_t_connect_with_wrong_password(GnuTLS::Session::acts as a client):
   NotImplementedError: fork is not available on this platform
   org/jruby/RubyKernel.java:1855:in fork' clearskies/test/gnutls.rb:16:inrun_test'
   org/jruby/RubyBasicObject.java:1735:in `instance_eval'

3. Error:
   test_0001_can_connect_and_send_data(GnuTLS::Session::acts as a server):
   NotImplementedError: fork is not available on this platform
   org/jruby/RubyKernel.java:1855:in fork' clearskies/test/gnutls.rb:16:inrun_test'
   org/jruby/RubyBasicObject.java:1735:in `instance_eval'

4. Error:
   test_0002_won_t_connect_with_wrong_password(GnuTLS::Session::acts as a server):
   NotImplementedError: fork is not available on this platform
   org/jruby/RubyKernel.java:1855:in fork' clearskies/test/gnutls.rb:16:inrun_test'
   org/jruby/RubyBasicObject.java:1735:in `instance_eval'

Request for clarification

Hello.

While reading the protocol draft, I didn't catch the following idea. When a user want to give full access to a share to a peer, does he have to generate a temporary access code or it's enough to share only a read-write secret (like in BT Sync)?

Thanks in advance.

Tracker protocol unclear text

"The ID and peer ID are combined into a single string separated by an "@" character, and sent as the "id" parameter."

It's not clear which ID is referring to. The ids should be clarified with an abbreviation consistent in the whole text.