Scallion - GPU-based Onion Addresses Hash Generator - Hacking Land

Scallion lets you create vanity GPG keys and .onion addresses (for Tor's hidden services) using OpenCL.
Scallion runs on Mono (tested in Arch Linux) and .NET 3.5+ (tested on Windows 7 and Server 2008).
Scallion is currently in beta stage and under active development. Nevertheless, we feel that it is ready for use. Improvements are expected primarily in performance, user interface, and ease of installation, not in the overall algorithm used to generate keys.

FAQ
Here are some frequently asked questions and their answers:

Why generate GPG keys?
Scallion was used to find collisions for every 32bit key id in the Web of Trust's strong set demonstrating how insecure 32bit key ids are. There was/is a talk at DEFCON (video) and additional info can be found at https://evil32.com/.
What are valid characters?
Tor .onion addresses use Base32, consisting of all letters and the digits 2 through 7, inclusive. They are case-insensitive.
GPG fingerprints use hexadecimal, consisting of the digits 0-9 and the letters A-F.
Can you use Bitcoin ASICs (e.g. Jalapeno, KnC) to accelerate this process?
Sadly, no. While the process Scallion uses is conceptually similar (increment a nonce and check the hash), the details are different (SHA-1 vs double SHA-256 for Bitcoin). Furthermore, Bitcoin ASICs are as fast as they are because they are extremely tailored to Bitcoin mining applications. For example, here's the datasheet for the CoinCraft A-1, an ASIC that never came out, but is probably indicitive of the general approach. The microcontroller sends work in the form of the final 128-bits of a Bitcoin block, the hash midstate of the previous bits, a target difficulty, and the maximum nonce to try. The ASIC chooses the location to insert the nonce, and it chooses what blocks meet the hash. Scallion has to insert the nonce in a different location, and it checks for a pattern match rather than just "lower than XXXX".
How can you use multiple devices?
Run multiple Scallion instances. Scallion searches are probabilistic, so you won't be repeating work with the second device. True multi-device support wouldn't be too difficult, but it also wouldn't add much. I've run several scallion instances in tmux or screen with great success. You'll just need to manually abort all the jobs when one finds a pattern (or write a shell script to monitor the output file and kill them all when it sees results).

Dependencies

OpenCL and relevant drivers installed and configured. Refer to your distribution's documentation.
OpenSSL. For Windows, the prebuilt x86 DLLs are included
On windows only, VC++ Redistributable 2008

Binary Download
Just want the latest binary version? Grab it here.

Build Linux
Prerequisites

Get the latest mono for your linux distribution:
http://www.mono-project.com/download/

Install Common dependencies:

sudo apt-get update
 sudo apt-get install libssl-dev mono-devel

AMD/OpenSource build sudo apt-get install ocl-icd-opencl-dev
Nvidia build sudo apt-get install nvidia-opencl-dev nvidia-opencl-icd
Finally msbuild scallion.sln

Docker Linux (nvidia GPUs only)

Have the nvidia-docker container runtime

Build the container:

docker build -t scallion -f Dockerfile.nvidia .

Run:

docker run --runtime=nvidia -ti --rm scallion -l

screenshot of expected output

Build Windows

Open 'scallion.sln' in VS Express for Desktop 2012
Build the solution, I did everything in debug mode.

Usage
Restarting Scallion during a search will not lose "progress". It is a probabilistic search and Scallion does not make "progress"
List devices

$ mono scallion/bin/Debug/scallion.exe -l

Generate a hash

$ mono scallion/bin/Debug/scallion.exe -d 0 prefix
 Cooking up some delicious scallions...
 Using kernel optimized from file kernel.cl (Optimized4)
 Using work group size 128
 Compiling kernel... done.
 Testing SHA1 hash...
 CPU SHA-1: d3486ae9136e7856bc42212385ea797094475802
 GPU SHA-1: d3486ae9136e7856bc42212385ea797094475802
 Looks good!
 LoopIteration:40  HashCount:671.09MH  Speed:9.5MH/s  Runtime:00:01:10  Predicted:00:00:56  Found new key! Found 1 unique keys.
 <XmlMatchOutput>
   <GeneratedDate>2014-08-05T07:14:50.329955Z</GeneratedDate>
   <Hash>prefix64kxpwmzdz.onion</Hash>
   <PrivateKey>-----BEGIN RSA PRIVATE KEY-----
 MIICXAIBAAKBgQCmYmTnwGOCpsPOqvs5mZQbIM1TTqOHK1r6zGvpk61ZaT7z2BCE
 FPvdTdkZ4tQ3/95ufjhPx7EVDjeJ/JUbT0QAW/YflzUfFJuBli0J2eUJzhhiHpC/
 1d3rb6Uhnwvv3xSnfG8m7LeI/Ao3FLtyZFgGZPwsw3BZYyJn3sD1mJIJrQIEB/ZP
 ZwKBgCTUQTR4zcz65zSOfo95l3YetVhfmApYcQQd8HTxgTqEsjr00XzW799ioIWt   
 vaKMCtJlkWLz4N1EqflOH3WnXsEkNA5AVFe1FTirijuaH7e46fuaPJWhaSq1qERT
 eQT1jY2jytnsJT0VR7e2F83FKINjLeccnkkiVknsjrOPrzkXAkEA0Ky+vQdEj64e
 iP4Rxc1NreB7oKor40+w7XSA0hyLA3JQjaHcseg/bqYxPZ5J4JkCNmjavGdM1v6E
 OsVVaMWQ7QJBAMweWSWtLp6rVOvTcjZg+l5+D2NH+KbhHbNLBcSDIvHNmD9RzGM1
 Xvt+rR0FA0wUDelcdJt0R29v2t19k2IBA8ECQFMDRoOQ+GBSoDUs7PUWdcXtM7Nt
 QW350QEJ1hBJkG2SqyNJuepH4PIktjfytgcwQi9w7iFafyxcAAEYgj4HZw8CQAUI
 3xXEA2yZf9/wYax6/Gm67cpKc3sgKVczFxsHhzEml6hi5u0FG7aNs7jQTRMW0aVF
 P8Ecx3l7iZ6TeakqGhcCQGdhCaEb7bybAmwQ520omqfHWSte2Wyh+sWZXNy49EBg
 d1mBig/w54sOBCUHjfkO9gyiANP/uBbR6k/bnmF4dMc=
 -----END RSA PRIVATE KEY-----
 </PrivateKey>
   <PublicModulusBytes>pmJk58BjgqbDzqr7OZmUGyDNU06jhyta+sxr6ZOtWWk+89gQhBT73U3ZGeLUN//ebn44T8exFQ43ifyVG09EAFv2H5c1HxSbgZYtCdnlCc4YYh6Qv9Xd62+lIZ8L798Up3xvJuy3iPwKNxS7cmRYBmT8LMNwWWMiZ97A9ZiSCa0=</PublicModulusBytes>
   <PublicExponentBytes>B/ZPZw==</PublicExponentBytes>
 </XmlMatchOutput>
 ini   t: 491ms / 1 (491ms, 2.04/s)
 generate key: 1193ms / 6 (198.83ms, 5.03/s)
 cpu precompute: 10ms / 6 (1.67ms, 600/s)
 total without init: 70640ms / 1 (70640ms, 0.01/s)
 set buffers: 0ms / 40 (0ms, 0/s)
 write buffers: 3ms / 40 (0.08ms, 13333.33/s)
 read results: 67442ms / 40 (1686.05ms, 0.59/s)
 check results: 185ms / 40 (4.63ms, 216.22/s)
 
 9.50 million hashes per second
 
 Stopping the GPU and shutting down...

Multipattern Hashing
Scallion supports finding one or more of multiple patterns through a primitive regex syntax. Only character classes (ex. [abcd]) are supported. The . character represents any character. Onion addresses are always 16 characters long and GPG fingerprints are always 40 characters. You can find a suffix by putting $ at the end of the match (ex. DEAD$). Finally, the pipe syntax (ex. pattern1|pattern2) can be used to find multiple patterns. Searching for multible patterns (within reason) will NOT produce a significant decrease in speed. Many regexps will produce a single pattern on the GPU and result in no speed reduction.
Some use cases with examples:

Generate a prefix followed by a number for better readability:
```
  mono scallion.exe prefix[234567]
```
Search for several patterns at once (n.b. -c causes scallion to continue generating even once it gets a hit)
```
  mono scallion.exe -c prefix scallion hashes
   mono scallion.exe -c "prefix|scallion|hashes"
```

Search for the suffix "badbeef"

  mono scallion.exe .........badbeef
   mono scallion.exe --gpg badbeef$ # Generate GPG key

Complicated self explanatory example:

  mono scallion.exe "suffixa$|suffixb$|prefixa|prefixb|a.suffix$|a.test.$"

How does Scallion work?
At a high level Scallion works as follows:

Generate RSA key using OpenSSL on the CPU
Send the key to the GPU
Increase the key's public exponent
Hash the key
If the hashed key is not a partial collision go to step 3
If the key does not pass the sanity checks recommended by PKCS #1 v2.1 (checked on the CPU) go to step 3
Brand new key with partial collision!

The basic algorithm is described above. Speed / performance is the result of massive parallelization, both on the GPU and the CPU.

Speed / Performance
It is important to realize that Scallion preforms a probabilistic search. Actual times may very significantly from predicated
The inital RSA key generation is done the CPU. An ivybridge i7 can generate 51 keys per second using a single core. Each key can provide 1 gigahash worth of exponents to mine and a decent CPU can keep up with several GPUs as it is currently implemented.
SHA1 hashing is done on the GPU. The hashrates for several GPUs we have tested are below (grouped by manufacturer and sorted by power):

GPU	Speed
Intel i7-2620M	9.9 MH/s
Intel i5-5200U	118 MH/s
NVIDIA GT 520	38.7 MH/s
NVIDIA Quadro K2000M	90 MH/s
NVIDIA GTS 250	128 MH/s
NVIDIA GTS 450	144 MH/s
NVIDIA GTX 670	480 MH/s
NVIDIA GTX 970	2350 MH/s
NVIDIA GTX 980	3260 MH/s
NVIDIA GTX 1050 (M)	1400 MH/s
NVIDIA GTX 1070	4140 MH/s
NVIDIA GTX 1070 TI	5100 MH/s
NVIDIA GTX TITAN X	4412 MH/s
NVIDIA GTX 1080	5760 MH/s
NVIDIA Tesla V100	11646 MH/s
AMD A8-7600 APU	120 MH/s
AMD Radeon HD5770	520 MH/s
AMD Radeon HD6850	600 MH/s
AMD Radeon RX 460	840 MH/s
AMD Radeon RX 470	957 MH/s
AMD Radeon R9 380X	2058 MH/s
AMD FirePro W9100	2566 MH/s
AMD Radeon RX 480	2700 MH/s
AMD Radeon RX 580	3180 MH/s
AMD Radeon R9 Nano	3325 MH/s
AMD Vega Frontier Edition	7119 MH/s

MH/s = million hashes per second
Its worth noting that Intel has released OpenCL drivers for its processors and short collisions can be found on the CPU.
To calculate the number of seconds required for a given partial collision (on average), use the formula:

Type	Estimated time
GPG Key	2^(4*length-1) / hashspeed
.onion Address	2^(5*length-1) / hashspeed

For example on my nVidia Quadro K2000M, I see around 90 MH/s. With those speed I can generate an eight character .onion prefix in about 1h 41m, 2^(5*8-1)/90 million = 101 minutes.

Workgroup Size
Scallion will use your devices reported preferred work group size by default. This is a reasonable default but experimenting with the workgroup may increase performance.

Security
The keys generated by Scallion are quite similar to those generated by shallot. They have unusually large public exponents, but they are put through the full set of sanity checks recommended by PKCS #1 v2.1 via openssl's RSA_check_key function. Scallion supports several RSA key sizes, with optimized kernels for 1024b, 2048b, and 4096b. Other key sizes may work, but have not been tested.

Thanks / References