Putting Amix on your real LAN (Amiberry)¶

This is a complete, first-hand-verified recipe (2026-06: Amiberry 8.1.6, Amix 2.1 / SVR4.0, Linux host) for turning an emulated Amix into a real, reachable host on your physical LAN — with a static IP, a working default gateway, DNS name resolution and internet access, reachable inbound (telnet/ftp/rlogin), and coming up automatically at boot. ✅

The result, with the example values used throughout:


Amix IP	`192.168.2.38/24` (`aen0`)
Gateway	`192.168.2.1`
DNS	`192.168.2.14`, `192.168.2.23`
From the LAN	`ping 192.168.2.38`, `telnet 192.168.2.38` reach Amix
From Amix	`ping www.google.com`, internet, NFS, etc.

It builds on the conceptual networking page (the SVR4 STREAMS stack, aen0, static-IP/no-DHCP, DNS-off-by-default) and the Amiberry page: Amiberry 8.x fully runs and installs Amix — its A3000 on-board SCSI emulates the disk (ID 6) and tape (ID 4) — and adds real A2065 networking. ✅

Substitute your own IP/gateway/DNS throughout. The host-side helpers in tools/host-net/ default to exactly these values and autodetect your LAN interface.

Why this works on Amiberry (and the design choice)¶

Amiberry ≥ 8.0 added a TAP backend for the emulated A2065 card: set a2065=tap0 and Amiberry attaches the card to a Linux TAP via TUNSETIFF. This uses /dev/net/tun (no cap_net_admin, no libpcap), which is why it works even in locked-down environments where pcap capture is blocked. ✅
We then run the host as a proxy-ARP router between that TAP and the LAN NIC, routing the guest's /32 at the TAP. This makes Amix a real LAN host without ever touching the host's own IP, default route, or NetworkManager — so it can't knock the host off the network. ✅
A true L2 bridge (br0 enslaving the NIC + tap0, host IP moved onto br0) also works and is more "native", but it momentarily blips the host's connectivity and risks the uplink for no functional gain here. The Amiberry/guest config is identical either way (a2065=tap0); only the host plumbing differs. 🟡

Step 1 — Host side (proxy-ARP routed TAP)¶

Run the helper (idempotent; defaults to GUEST_IP=192.168.2.38, autodetects the NIC):

sudo sh tools/host-net/amix-lan-up.sh

That is exactly equivalent to (LAN NIC shown as $ETH):

# user-owned TAP Amiberry attaches to (no caps needed)
sudo ip tuntap add dev tap0 mode tap user "$USER"
sudo ip link set tap0 up
sudo ip addr replace 192.168.2.2/32 dev tap0          # host-side anchor (host<->guest)

# route the guest's /32 at the tap, with a same-subnet source
sudo ip route replace 192.168.2.38/32 dev tap0 src 192.168.2.2

# make the host a proxy-ARP router
sudo sysctl -wq net.ipv4.ip_forward=1
sudo sysctl -wq net.ipv4.conf.$ETH.proxy_arp=1
sudo sysctl -wq net.ipv4.conf.tap0.proxy_arp=1

# allow forwarding past Docker's FORWARD=DROP policy (DOCKER-USER if Docker present)
sudo iptables -I DOCKER-USER 1 -i tap0 -o $ETH -j ACCEPT
sudo iptables -I DOCKER-USER 1 -i $ETH -o tap0 -j ACCEPT

Gotcha — Docker. If Docker is installed it sets the iptables FORWARD policy to DROP; without the two ACCEPT rules above, the host will answer ARP for the guest but silently drop its forwarded packets, and ping <gateway> from Amix returns "no answer". This is the single most confusing host-side failure. ✅

tap0 shows state DOWN / NO-CARRIER until Amiberry attaches to it — that is normal; the carrier comes up when the emulator opens the device. ✅

Step 2 — Amiberry config¶

In your machine's .uae (a complete working example is committed at tools/amix-amiberry.uae), the networking lines are:

a2065=tap0
a2065_rom_options=mac=00:80:10:00:00:38
bsdsocket_emu=false

plus the standard, mandatory Amix machine settings (see Amiberry status and hardware):

cpu_model=68030
fpu_model=68882
mmu_model=68030          ; MMU ON  — required or Amix won't boot
cpu_compatible=false     ; "More compatible" OFF
cachesize=0              ; JIT OFF
chipset_compatible=A3000
scsi_a3000=true
kickstart_rom_file=Kickstart v2.04 r37.175 (1991)(Commodore)(A3000).rom
hardfile2=rw,DH0:/path/to/Amix.hdf,32,1,2,512,0,,scsi6_a3000

Gotcha — the MAC must be a Commodore OUI (00:80:10:xx:xx:xx). The Amix aen0/LANCE path expects it; an arbitrary MAC misbehaves. The host-side TAP keeps its own random MAC (correct — the bridge/host learns the Amiga MAC from frames). ✅

Launching headless is possible (Xvfb + a controlling TTY + xdotool); see Headless operation. Otherwise just launch Amiberry normally and use its window.

Step 3 — Inside Amix (one-time, persistent)¶

Three files make the network come up automatically at every boot. Edit them as root (over the console, or once networking is up, over telnet/ftp).

3a. `/etc/inet/hosts` — the key to boot-time bring-up¶

At boot, /etc/init.d/inetinit sources /etc/inet/network-config, which runs:

/usr/amiga/bin/aen -S && /usr/sbin/slink addaen /dev/aen0 aen0 \
    && /usr/sbin/ifconfig aen0 `uname -n` up -trailers

So aen0's address is whatever `uname -n` (here: amix) resolves to in /etc/inet/hosts. If that file is empty, you get ifconfig: amix: bad value at boot and no interface. Populate it (/etc/hosts is a symlink to this file):

127.0.0.1     localhost loghost
192.168.2.38  amix amix.localdomain
192.168.2.1   gateway
192.168.2.14  ns1
192.168.2.23  ns2

No explicit netmask is needed: 192.168.2.x takes the class-C default (ffffff00 = /24) automatically. ✅

3b. Default route — in `/etc/inet/rc.inet`¶

rc.inet (run right after network-config) is where the persistent default route lives. Add (note the mandatory metric — 1):

/usr/sbin/route add default 192.168.2.1 1

3c. DNS — library swap + transport table + resolv.conf¶

Amix resolves from /etc/hosts only until you enable DNS. Two libraries are involved — swap both:

# 1. gethostbyname() path (ping, telnet, ftp clients): swap libsocket -> libsockdns
cp /usr/lib/libsocket.so /usr/lib/libsocket.so.orig
ln -f /usr/lib/libsockdns.so /usr/lib/libsocket.so

# 2. TLI/netdir path: activate the DNS-enabled transport table
cp /etc/netconfig /etc/netconfig.TCP
cp /etc/netconfig.DNS /etc/netconfig          # adds /usr/lib/resolv.so to tcp/udp/...

# 3. resolver config — nameservers ONLY, no `domain`/`search` (see the gotcha)
cat > /etc/resolv.conf <<'EOF'
nameserver 192.168.2.14
nameserver 192.168.2.23
EOF

⚠️ Also fix the boot-time ifconfigs, or every boot gets ~3 min slower. As soon as DNS is enabled (the libsockdns link above), the two ifconfig calls that bring up lo0 and aen0 at boot each hang ~90 s (~180 s total), because they take hostnames, and gethostbyname() now tries DNS first — but they run before the default route is installed (route add default … is later, in /etc/inet/rc.inet), so with no route to the nameservers each lookup burns the full resolver retransmit schedule before falling back to /etc/hosts. Stock (DNS-off) Amix resolves these names from /etc/hosts instantly, which is exactly why the trap appears only after you enable DNS, and why the Step 3a /etc/inet/hosts recipe is not enough on its own once DNS is on. The fix is to give both boot-time ifconfigs a literal IP so no resolver call happens before the network is up (boot 210 s → 29 s; runtime DNS for clients/mail is untouched). ✅

# /etc/rc2.d/S69inet        -- the loopback bring-up
-   /usr/sbin/ifconfig lo0 localhost up
+   /usr/sbin/ifconfig lo0 127.0.0.1 up

# /etc/inet/network-config  -- the aen0 bring-up (replaces the `uname -n` from Step 3a)
-   /usr/sbin/ifconfig aen0 `uname -n` up -trailers
+   /usr/sbin/ifconfig aen0 192.168.2.38 up -trailers   # this host's static IP, literal

lo0 is always 127.0.0.1; aen0 uses this host's own static address (the same .38 already in /etc/inet/hosts). Configuring your own interface should never depend on a name service, so this is the correct design regardless. ✅

Editing gotchas for /etc/rc2.d/S69inet — read before you sed:
S69inet is a hardlink. The same inode is also /etc/init.d/inetinit. Edit it in place so both names stay in sync — sed to a temp then cat > the original, or use ed. Do not mv/unlink and recreate, or you break the link and inetinit keeps the old (slow) behaviour:
sed 's/ifconfig lo0 localhost up/ifconfig lo0 127.0.0.1 up/' \
    /etc/rc2.d/S69inet > /tmp/S69inet.new && cat /tmp/S69inet.new > /etc/rc2.d/S69inet
Never leave a backup named S* inside /etc/rc2.d/. rc2 runs for f in /etc/rc2.d/S* at boot, so a copy like S69inet.bak or S69inet.orig gets executed too (verified against the live /etc/rc2 glob). Park backups outside that directory:
cp /etc/rc2.d/S69inet /root/S69inet.orig     # NOT /etc/rc2.d/S69inet.orig
This no-route boot hang is independent of the /etc/domain "general weirdness" below — a box with an already-empty /etc/domain still hangs the full 180 s. They are two different problems; fix both. ✅
Gotcha — leave the domain UNSET (/etc/domain). This is the documented Amix "general weirdness": if a domain is set, the resolver appends it to every lookup — even fully-qualified names — so www.google.com is queried as www.google.com.<yourdomain>. With a wildcard in your local zone this resolves to the wrong address and the lookup never falls back. The domain is set at sysinit from /etc/domain (via domainname \cat /etc/domain`). Empty it:
cp /etc/domain /etc/domain.orig
cp /dev/null /etc/domain          # or: echo nodomain > /etc/domain
domainname ""                     # apply now, without a reboot
options ndots:1inresolv.confdoes **not** help — this old resolver ignores it. Cosmetic side effect: the console banner becomesNode: amixinstead ofamix.(the FQDN still resolves via/etc/inet/hosts`). ✅

Verification checklist¶

# From the host:
ping 192.168.2.38                       # -> replies
python3 tools/host-net/amixsh.py 'ifconfig aen0' 'netstat -rn' 'domainname' \
        'ping 192.168.2.1' 'ping www.google.com'
# Expect: aen0 inet 192.168.2.38 netmask ffffff00 ; default -> 192.168.2.1 via aen0 ;
#         domainname empty ; "192.168.2.1 is alive" ; "www.google.com is alive"

# From any other LAN machine:
telnet 192.168.2.38                     # -> Amix login: prompt

SVR4 ping host is single-shot ("host is alive" / "no answer"), not a continuous stream — handy for scripting. ✅

Persistence (survives reboot)¶

Guest: everything in Step 3 is on-disk (/etc/inet/hosts, rc.inet, /etc/resolv.conf, /etc/netconfig, the libsocket.so link, empty /etc/domain) and is re-applied by the normal SVR4 boot. Reboot-verified: after shutdown -i6, aen0 comes up at .38, the route is set, domainname is empty, and ping www.google.com works — with zero manual steps. ✅

Host: the ip/sysctl/iptables state is not persistent. Install the service so it's recreated at boot:

sudo install -m 755 tools/host-net/amix-lan-up.sh tools/host-net/amix-lan-down.sh /usr/local/sbin/
sudo install -m 644 tools/host-net/amix-lan.service /etc/systemd/system/
sudo systemctl daemon-reload && sudo systemctl enable --now amix-lan.service

Headless operation (optional)¶

Useful for automation / agents driving Amix without a visible window:

Run Amiberry under Xvfb with a controlling TTY and software rendering: DISPLAY=:77 SDL_RENDER_DRIVER=software script -qfc "amiberry -f amix_net.uae" /dev/null (start Xvfb :77 -screen 0 1280x1024x24 -nolisten tcp first; drive the console with xdotool windowfocus <id> + xdotool type/key, screenshot with import -window root). ✅
Once aen0 is up, skip the console entirely and use the LAN: scripted shell via tools/host-net/amixsh.py (telnet) and file uploads via amixput.py (FTP). ✅

Sources¶

First-hand reproduction, 2026-06-02: Amiberry 8.1.6 (a2065=tap0, A2065 attached to a host TAP, MAC 00:80:10:00:00:38), Amix 2.1 on Amix.hdf at scsi6_a3000; host = Linux on 192.168.2.0/24 with Docker. Verified host↔guest, gateway/DNS/internet, inbound telnet/ftp, and reboot persistence (aen0/route/DNS auto-up after shutdown -i6).
Amiberry 8.0.0 release + PRs #1784 (TAP backend, a2065=tap0/a2065=br0), #1777 (PCAP bridged rewrite), #1774 (config parse fix). Source: src/cfgfile.cpp, src/ethernet.cpp, src/osdep/amiberry_uaenet.cpp, src/a2065.cpp.
Amix boot path: /etc/inittab (S2::sysinit:/etc/sysinit), /etc/sysinit (domainname \cat /etc/domain`),/etc/init.d/inetinit,/etc/inet/network-config(ifconfig aen0 `uname -n` up -trailers),/etc/inet/rc.inet` — read directly off the running system.
DNS enable: /etc/netconfig.DNS (adds /usr/lib/resolv.so), /usr/lib/libsockdns.so — present on the install; the domain-append behaviour matches amigaunix.com networking notes and was confirmed via a gethostbyname() probe.
DNS-enable slow-boot + literal-IP fix: The A4091-on-Amix project — networking investigation, 2026-06-07 (reproduced locally ✅) — instrumented /etc/rc2 per-script timing on Amix 2.1c under Amiberry, isolating the ~180 s hang to the two boot-time gethostbyname() calls; boot 210 s → 29 s after the fix. /etc/rc2.d/S69inet (hardlinked to /etc/init.d/inetinit), /etc/inet/network-config (ifconfig aen0 \uname -n`→ literal IP),/etc/inet/rc.inet(default route runs *after*network-config) — read directly off the running system;for f in /etc/rc2.d/S*` glob confirms backup S* files execute at boot.
Research brief §11 (networking) and the conceptual networking page.