rinse

rinse

Web service that converts untrusted documents to image-based PDF:s in a sandbox.

Provides both a Web UI and a Swagger REST API.

Requirements

• podman is required.

Running

You should start the container in rootless mode. Inside the container we use gVisor to further sandbox operations, and gVisor requires the container to be started with --cap-add SYS_ADMIN and -v /proc:/newproc:ro.

Rinse will run as the container's root user which will translate to the user that started the container, so by default it will listen on either port 80 or 443. Since you will be starting the container as a non-privileged user, you'll need to forward HTTP requests to it from a non-privileged host port to a privileged port inside the container.

If you want the service to remember it's settings between runs, you'll need to mount a volume at /etc/rinse inside the container.

podman run --rm -d -p 8080:80 --cap-add SYS_ADMIN -v /proc:/newproc:ro -v $HOME:/etc/rinse ghcr.io/linkdata/rinse

Running it with HTTPS requires you to provide valid certificates. Rinse will look for fullchain.pem and privkey.pem at /etc/certs inside the container, and if found start in HTTPS mode.

podman run --rm -d -p 8443:443 --cap-add SYS_ADMIN -v /proc:/newproc:ro -v $HOME:/etc/rinse -v $HOME/certs:/etc/certs ghcr.io/linkdata/rinse

Process

First, a temporary directory is created for the job. This will be mounted in the gVisor container as /var/rinse. If we were given an URL, we download the document and place it here.

Then, each of these stages run in their own gVisor container, which is destroyed as soon as the stage is complete or fails. When the job is removed, all it's files are overwritten before they are deleted from the container filesystem.

The original document is renamed to input with it's extension preserved and made read-only before invoking the next stage.

• If the language is to be auto-detected, Apache Tika is used to do so.

• If the document is not a PDF, LibreOffice is used to try to covert it to one, and if successful, the original document is deleted.

• The input.pdf file is converted to a set of PNG files using pdftoppm.

• The set of PNG files is OCR-ed and processed into a PDF named output.pdf using tesseract.

• Finally the output.pdf file is renamed to the original filename (without extension) with -rinsed.pdf appended.