Mascali
/
piston
mirror of https://github.com/engineer-man/piston.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
72 Commits 11 Branches 1 Tag 2.5 MiB
JavaScript 80.7%
Shell 11.8%
Python 3.4%
Dockerfile 3%
Makefile 0.9%
Other 0.2%
Go to file
Brian Seymour 37ba22baa1 jdk instead of jre 2018-10-05 21:18:46 -05:00
api java support 2018-10-05 21:11:12 -05:00
docker jdk instead of jre 2018-10-05 21:18:46 -05:00
var update limits 2018-09-21 12:05:55 -05:00
.gitignore mvp done 2018-09-20 00:49:02 -05:00
license poc work 2018-09-19 21:58:50 -05:00
readme.md asm support 2018-09-28 15:24:24 -05:00

readme.md

Piston

Piston is the underlying engine for running untrusted and possibly malicious code that originates from from EMKC contests and challenges. It's also used in the Engineer Man Discord server via felix bot.

Installation

Usage

  • docker/execute <lang> <path to file>

Supported Languages

Currently python2, python3, c, c++, go, node, ruby, r, c#, nasm, and php is supported.

Principle of Operation

Piston utilizes Docker as the primary mechanism for sandboxing. There is a small API written in Go which takes in execution requests and spawns new containers to execute the source from that request. High level, the API writes a temporary source file to /tmp and that gets mounted read-only along with the execution scripts into the container. The source file is either ran or compiled and ran (in the case of languages like c, c++, c#, go, etc.).

Security

Docker provides a great deal of security out of the box. Piston takes additional steps to make it resistant to various privilege escalation, denial-of-service, and resource saturation threats. These steps include:

  • Disabling outgoing network interaction
  • Capping memory at 64mb (resists RAM saturation)
  • Capping max processes at 16 (resists :(){ :|: &}:;, while True: os.fork(), etc.)
  • Capping max files at 256 (resists various file based attacks)
  • Mounting all resources read-only (resists sudo rm -rf --no-preserve-root /)
  • Capping runtime execution at 3 seconds
  • Capping stdout to 65536 characters (resists yes/no bombs and runaway output)
  • SIGKILLing misbehaving code
  • Disabling journald logs (resists log flood)

Performance

One thing that needs investigation is how to spawn containers faster. The Docker daemon is synchronous in its container spawning. This means the bottleneck for code execution is how fast containers can start. Environments vary, but, in ours they start at a rate of no more than 1-2 per second. One possibility is Docker in Docker where by X number of containers stay running all the time and then requests are delivered to each in a round robin and spawn new Piston containers.

License

Piston is licensed under the MIT license.