VULINX
Free software · GPL-2.0-or-later
VULINX runs mainframe workloads in isolated Execution Spaces on ordinary servers: Linux on s390x, and System/370 operating systems such as MVS 3.8j. An engine built on QEMU, and a console that makes it easy to use and to deploy.
What it is
A fork of QEMU that completes and extends s390x virtualization and adds a System/370 machine of its own: processors, channel subsystems, and the devices a mainframe system expects. Every change to upstream is recorded, so merging a newer QEMU stays possible.
A daemon, a CLI and a web console over a versioned REST API. Define Nodes and Execution Spaces, keep profiles as YAML, deploy from them, manage volumes and networks, run the lifecycle, watch live state, and reach a guest's console in the browser.
VULINX is not an IBM mainframe, an IBM Z replacement or a LinuxONE emulator, and claims none of it. An Execution Space is comparable to a logical partition in spirit; it is a VULINX construct. A laboratory, a development platform, and a way to keep old systems running.
Two machines, one engine
Unmodified distributions boot and run — emulated under TCG on any host, accelerated by KVM on an s390x host. IPL from an ECKD volume with no other disk attached.
A System/370 processor with DAT, key protection and its own channel subsystem. The TK5 distribution IPLs from its unmodified volumes, and its batch jobs run 2.5 to 3.4 times faster than on the reference emulator.
What works today
CKD and compressed CKD volumes are used as they already exist — read in place, written through an overlay per Execution Space, so one set of volumes serves several machines.
$ vulinx volume convert mvsres.cckd mvsres.ckd
3350, 555 cylinders · compressed → plain · verifiedA profile is a YAML document for your version control. Nothing in it knows where it runs: the Node decides how to execute it when it starts.
An OSD adapter the Linux qeth driver binds to, beside virtio. Internal networks ride a user-space switch per network.
SCLP line mode, 3270 and 3215 — proxied over WebSocket, each with its own permission.
Sign-in with the host's accounts through PAM, permissions per route, grants scoped to named Execution Spaces, and an audit trail of what was done and refused.
A security model that says what is enforced, where in the source, and which test holds it in place — from disk images that try to read host files to a channel program that never ends.
How it fits together
The daemon links no engine library and shares no address space with a guest. It starts engine processes and talks to them over QMP — a management outage cannot stall a running Execution Space.
Get started
git clone --recurse-submodules \
https://gitlab.com/kaypel1-group/vulinx/open.git vulinx
cd vulinxcd engine && ./configure \
--target-list=s390x-softmmu,s370x-softmmu
make -j"$(nproc)" && cd ..make -C vulinx # Go 1.25, Node 20./vulinx/vulinxd \
-vulinx engine/build/vulinx-system-s390x \
-state-dir ~/.local/state/vulinxOpen http://127.0.0.1:8676. On loopback it asks for no sign-in; anywhere else it serves HTTPS and requires one. Packages for Debian and Ubuntu build from vulinx/packaging.
Community
Report a problem or propose an idea, read the code, send a merge request. If you run an operating system we have not tried, tell us what happened — that is the most useful report there is.
CONTRIBUTING.md says what a change needs: a contributor agreement, the licence tag, provenance, and documentation in the same change. No code comes from Hercules or Hyperion — their licence is not compatible with ours.
Free software under the GNU General Public License, version 2 or later — the engine, which carries QEMU's licence, and the daemon, the CLI and the console beside it.
The source is one command away.