Nexus

Nexus is the Vitruvian kernel subsystem that bridges the Linux kernel with the BeOS/Haiku runtime. It’s implemented as a set of Linux kernel modules (loaded at boot) that emulate the IPC and filesystem APIs Haiku provides natively in its monolithic kernel.

Native V applications expect ports, semaphores, shared memory areas, thread control, virtual file references, and node monitoring. None of these exist in Linux in BeOS-compatible form. Nexus fills that gap.

Device layout

Each Nexus subsystem is exposed as a character device:

Device	Subsystem
`/dev/nexus`	ports, semaphores, threads
`/dev/nexus_area`	shared memory areas
`/dev/nexus_node_monitor`	filesystem watches

Every process opens its own file descriptors to these devices at startup via BKernelPrivate::Team::Init(). File descriptors are process-local and not inherited across fork; children re-initialize via a constructor.

IPC primitives

Ports

Haiku-style message ports: bounded queues with read/write operations.

NEXUS_PORT_CREATE: create a new port
NEXUS_PORT_CLOSE: close a port (no new writers)
NEXUS_PORT_DELETE: destroy a port
NEXUS_PORT_READ / NEXUS_PORT_WRITE: send and receive messages
NEXUS_PORT_FIND: look up a port by name
NEXUS_PORT_INFO / NEXUS_PORT_MESSAGE_INFO: query port state

Limits: max queue depth 4096, max message size 256 KB.

Semaphores

Standard counting semaphores:

NEXUS_SEM_CREATE / NEXUS_SEM_DELETE
NEXUS_SEM_ACQUIRE / NEXUS_SEM_RELEASE
NEXUS_SEM_COUNT / NEXUS_SEM_INFO / NEXUS_SEM_NEXT_INFO

Threads

Thread lifecycle and communication:

NEXUS_THREAD_SPAWN: create a new thread
NEXUS_THREAD_SET_NAME / NEXUS_THREAD_RESUME
NEXUS_THREAD_READ / NEXUS_THREAD_WRITE / NEXUS_THREAD_HAS_DATA
NEXUS_THREAD_WAITFOR / NEXUS_THREAD_WAIT_NEWBORN
NEXUS_THREAD_SET_RETURN_CODE

Note: Haiku spawns threads in suspended state; resume_thread() starts them. Vitruvian’s pthread_create starts immediately, so the run-state model differs.

Areas

Shared memory regions backed by memfd_create + mmap:

NEXUS_AREA_CREATE / NEXUS_AREA_CLONE / NEXUS_AREA_DELETE
NEXUS_AREA_FIND / NEXUS_AREA_GET_INFO / NEXUS_AREA_GET_NEXT
NEXUS_AREA_RESIZE / NEXUS_AREA_SET_PROTECTION / NEXUS_AREA_TRANSFER

Cloning an area maps the same memfd into the target process.

Node monitor

Provides filesystem event notifications compatible with the BeOS node_monitor API:

B_ENTRY_CREATED: file or directory created
B_ENTRY_REMOVED: file or directory removed
B_ENTRY_MOVED: file or directory renamed or moved
B_STAT_CHANGED: metadata changed (size, mtime, etc.)
B_ATTR_CHANGED: extended attribute added, modified, or removed
B_DEVICE_MOUNTED / B_DEVICE_UNMOUNTED: volume mounted or unmounted

Nexus hooks into Linux’s fsnotify subsystem and translates events into B_NODE_MONITOR messages. This is what allows Tracker to watch directories and respond to changes in real time, exactly as it would on Haiku.

Team registration

Every process that uses Nexus opens the devices and registers itself as a team. On process exit, the kernel module cleans up all resources owned by that team (ports, semaphores, areas, vrefs).

Deployment

Nexus ships as a DKMS package (nexus-dkms). It auto-rebuilds on kernel update. The DKMS source copy at debian/nexus-dkms/usr/src/nexus-1/nexus/ must be kept in sync with the main nexus/ tree: every ioctl addition and every fix goes in both places.