Up1: A Client-side Encrypted Image Host

Up1 is a simple host that client-side encrypts images, text, and other data, and stores them, with the server knowing nothing about the contents. It has the ability to view images, text with syntax highlighting, short videos, and arbitrary binaries as downloadables.

Public Server

We currently run a public, free to use server at https://up1.ca.

Example image: https://up1.ca/#hsd2mdSuIkzTUR6saZpn1Q - (source). Hint: Try double-clicking on the image, and dragging around the image to zoom in and out, RES-style!

Example text paste: https://up1.ca/#_H7Wy9atfl2CW_cYx6Xh9A - (source)

Example short video: https://up1.ca/#55s3dQnPjtcgstvspdYT5g - (source)

Example download file: https://up1.ca/#Nmo7tW7Qi8ga0UORVZ-zUw - (source)

This server is open to all users, so feel free to upload your own files.

####Client Utilities:

• ShareX, a popular screenshot and image uploader, now merged with Up1 support
• upclient, a command-line tool for uploading to Up1 servers

Quick start

To install and run the server with default settings:

apt-get install golang
git clone https://github.com/Upload/Up1
cd upload
go build server.go
./server

Server configuration is done through the server.conf file. For a quick start, simply move server.conf.example to server.conf.

listen is an address:port-formatted string, where either one are optional. Some examples include ":9000" to listen on any interface, port 9000; "1.2.3.4" to listen on localhost port 80; "1.1.1.1:8080" to listen on 1.1.1.1 port 8080; or even "" to listen on any interface, port 80.

api_key is a very basic security measure, requiring any client making an upload to know this key. This doesn't seem very useful and should be revamped; replace it with HTTP auth maybe?

delete_key is a key used to secure the deletion keys. Set this to something that only the server knows.

maximum_file_size is the largest file, in bytes, that's allowed to be uploaded to the server. The default here is a decimal 50MB.

There are three additional sections in the configuration file: http, https and cloudflare-cache-invalidate. The first two are fairly self-explanitory (and at least one must be enabled).

cloudflare-cache-invalidate is disabled by default and only useful if you choose to run the Up1 server behind Cloudflare. When this section is enabled, it ensures that when an upload is deleted, Cloudflare doesn't hold on to copies of the upload on its edge servers by sending an API call to invalidate it.

For the web application configuration, a config.js.example file is provided. Make sure the api_key here matches the one in server.conf.

External Tools

Currently, there are two external programs adapted to work with Up1: ShareX (relevant code, and upclient.

ShareX is a popular screenshot tool which supports tons of upload services, not just for images but also for text, video, documents, etc. ShareX includes a service which can send files to any Up1 server. It uses .NET BouncyCastle for the crypto.

Upclient is a CLI tool which can send files or data to Up1 servers either via unix pipe (ps aux | up), or via argument (up image.png), and returns a URL to the uploaded file on stdout. It runs on nodejs and uses SJCL for the crypto.

How it works

Before an image is uploaded, a "seed" is generated. This seed can be of any length (because really, the server will never be able to tell), but has a length of 25 characters by default. The seed is then run through SHA512, giving the AES key in bytes 0-256, the CCM IV in bytes 256-384, and the server's file identifier in bytes 384-512. Using this output, the image data is then encrypted using said AES key and IV using SJCL's AES-CCM methods, and sent to the server with an identifier. Within the encryption, there is also a prepended JSON object that contains metadata (currently just the filename and mime-type). The (decrypted) blob format starts with 2 bytes denoting the JSON character length, the JSON data itself, and then the file data at the end.

Image deletion functionality is also available. When an image is uploaded, a delete token is returned. Sending this delete token back to the server will delete the image. On the server side, HMAC-SHA256(static_delete_key, identifier) is used, where the key is a secret on the server.

Technologies

The browser-side is written in plain Javascript using SJCL for the AES-CCM encryption, with entropy obtained using the WebCrypto APIs and encryption performed within a Web Worker. The video and audio players just use the HTML5 players hopefully built into the browser. The paste viewer uses highlight.js for the syntax highlighting and line numbers.

Additionally, the repository copy of SJCL comes from the source at https://github.com/bitwiseshiftleft/sjcl, commit fb1ba931a46d41a7c238717492b66201b2995840 (Version 1.0.3), built with the command line ./configure --without-all --with-aes --with-sha512 --with-codecBytes --with-random --with-codecBase64 --with-ccm, and compressed using Closure Compiler. If all goes well, a self-built copy should match up byte-for-byte to the contents of static/deps/sjcl.min.js.

The server-side is written in Go and uses no dependencies outside of the standard library. The only cryptography it uses is for generating deletion keys, using HMAC and SHA256 in the built-in crypto/hmac and crypto/sha256 packages, respectively.

Caveats

• Encryption/Decryption are not streamed or chunked. This means that (at the time) any download must fit fully in memory, or the browser may crash. This is not a problem with sub-10MB images, but may be a problem if you want to share a long gameplay video or recorded meeting minutes. We would love help and contributions, even if they break backwards compatibilty. As this project is still relatively new, the API is not sacred, so break away!

• CCM is kinda slow. Compared to other authenticated encryption modes out there such as GCM or OCB, CCM is considered one of the slower modes (slightly slower than GCM, and almost twice as slow as OCB), isn't parallelizable and didn't make the best design decisions. The reason that we chose this algorithm, however, is twofold: primarily, this is the most-audited, oldest and most commonly used algorithm contained in SJCL; as this is used for viewing data, security there is important - and secondly, the other two mentioned algorithms in SJCL were actually slower than CCM. There are other crypto libraries which may be allegedly faster, such as asmcrypto.js, but it seems new, we don't know anything about it and currently prefer SJCL for its familiarity. With an audit from a trusted party, we may take a second look at asmcrypto.js.

• By its very nature, this uses cryptography in Javascript. There have been many reasons given as to why it's bad to use cryptography in Javascript, and some may be more valid than others. We're working on browser extensions to mitigate some of these reasons (and non-Javascript clients are always welcome!), but safe to say that if you unconditionally believe that Javascript crypto is bad, you probably won't want to use this.

• As a new project, this code hasn't been audited by a trusted party. Since this is brand new, there have been (to date) very few eyes on the code, and even fewer trusted eyes on the code. While we've put as much effort as possible into offloading the hard crypto stuff to SJCL, we still might have made a mistake somewhere (reading over static/js/encryption.js and letting us know if you find issues would be very helpful to us!), and so for that reason, using this software is at your own risk.

• The server will, in most cases, receive referrer headers. If a server decides to log requests, they will also be able to receive Referer headers. For private/protected websites and direct links sent via IM or email, this isn't a big deal. If the link is on a public website however, it means the server owner might be able to find the original image. Unfortunately there's nothing that the software or server owner can do about this (apart from hosting behind a CDN and offloading the Referer header to the edge), however when posting a link you have a couple of options:

  1. Put rel="noreferrer" into any <a> links that are directed at the Up1 server.
  2. If you don't have control over the link attributes, you can use a referrer breaker such as https://anon.click/ or https://href.li/, amongst many.

Contributing

Any contributions, whether to our existing code or as separate applications, are very welcome! Additionally, as we're in the early stages of this project, we don't mind breaking the API if it's for a good reason.

We don't ask for any CLAs - you don't have to give up copyright on your code - however we prefer that you contribute under the MIT license, just for consistency.

If you find serious security issues, please email us at security@up1.ca.

Some of us idle on irc.freenode.net in the #upload channel, if you would like to chat!

Thank you for you contributions!

License

The Up1 server and browser code are both licensed under MIT.

ShareX's base code is licensed under GPLv2, however the modifications (namely, the C# encryption code) is licensed under MIT.

Upclient is licensed fully under MIT.