go-network-configurator/README.md
James Coleman fae43247de Switch unit tests to testify
Bring an administratively down interface up in AddAddress before writing
its addresses and routes. The kernel installs an address's connected
route only while its link is up, so adding a gateway to a down interface
was rejected as unreachable. A link raised this way is part of the
pre-change state AddAddress restores, so it is returned to down on any
path that does not complete.

Also fix the rollback masking its own error: deleting the address it
added takes the default route with it, since that route resolved its
nexthop through the address's connected subnet. The following RouteDel
then answered ESRCH and was reported in place of "aborted operation due
to loss of internet".
2026-07-10 08:52:52 -05:00

12 KiB

go-network-configurator

Go Reference

A Go library for inspecting and changing a host's IP addresses and static routes at runtime and persisting those changes to whatever on-disk network configuration backend the system actually uses — so the change survives a reboot.

It is designed for hosting environments where an automated system needs to add, swap, or remove IPs on a server without knowing in advance whether that server is running netplan, systemd-networkd, NetworkManager, RHEL network-scripts, ifupdown, or cloud-init — and without leaving a control panel (cPanel, Plesk, InterWorx) out of sync.

import "github.com/grmrgecko/go-network-configurator"

How it works

A single Configurator applies every change in two layers:

  1. Runtime — the live kernel state is changed immediately (via netlink on Linux, the IP Helper API on Windows). Address and gateway changes are verified against an internet-reachability test and rolled back automatically if connectivity is lost.
  2. Persistence — the same change is written to every detected configuration backend so it survives a reboot, and registered control panels are told to re-read the system's IPs.

Backends are auto-detected at construction time. A host running both netplan and cloud-init, for example, has both files kept in sync; a failure writing one backend is logged but does not abort the others.

Supported backends

Network configuration (persistence)

Backend Detection
netplan netplan binary on PATH, and netplan info succeeds
cloud-init cloud-init network config present
NetworkManager NetworkManager service running or enabled, or its pid file present
systemd-networkd systemd-networkd service running or enabled
RHEL network-scripts network service running or enabled
ifupdown /etc/network/interfaces, and the networking service running or enabled

Control panels

Panel Detection
cPanel /usr/local/cpanel/bin/whmapi1
Plesk /usr/sbin/plesk
InterWorx /usr/bin/nodeworx

How service-managed backends are detected

A service counts as managing the network when it is running now or enabled to start at boot. Enablement matters on its own: this library writes configuration that must survive a reboot, and a manager that is enabled but not yet started still owns the network after the next boot.

Platforms

  • Linux — full support (all backends and panels above).
  • Windows — runtime changes via the IP Helper API, persisted to cloud-init, with Plesk panel support.

API

type Configurator interface {
    // Enumerate interfaces with their addresses, gateways, static routes,
    // DNS servers, search domains, and DHCP client state.
    GetInterfaces(ctx context.Context) ([]*Interface, error)

    // Add an IP address (optionally setting/replacing the default gateway).
    AddAddress(ctx context.Context, iface string, addr *net.IPNet, gateway net.IP) error

    // Promote an address already on the interface to be the primary of its family.
    SetPrimaryAddress(ctx context.Context, iface string, addr *net.IPNet) error

    // Remove an IP address.
    RemoveAddress(ctx context.Context, iface string, addr *net.IPNet) error

    // Add or remove a static route.
    AddRoute(ctx context.Context, iface string, dst *net.IPNet, gateway net.IP, metric int) error
    RemoveRoute(ctx context.Context, iface string, dst *net.IPNet, gateway net.IP) error

    // Set the DNS servers and search domains for an interface.
    SetDNS(ctx context.Context, iface string, servers []net.IP, searchDomains []string) error

    // Turn each address family's DHCP client on or off.
    SetDHCP(ctx context.Context, iface string, dhcp4, dhcp6 bool) error
}

// Construct the configurator for the current OS, auto-detecting backends.
// ctx bounds the detection; behaviour is tuned with Option values (see
// Options below).
func NewConfigurator(ctx context.Context, opts ...Option) (Configurator, error)

// Resolve an interface by name, or by the special "public-internet" /
// "public-internet-6" selectors.
func FindInterfaceByName(name string, ifaces []*Interface) *Interface

// Filter to the hardware-backed interfaces, best candidate for internet
// configuration first.
func FindPhysicalInterfaces(ifaces []*Interface) []*Interface

Every operation takes a context.Context. The slow steps — the ICMP probe of a new gateway and the internet-reachability test — honour cancellation and deadlines, so a caller can bound how long a change may block.

DNS

SetDNS applies an interface's resolvers and search domains to the live resolver so they take effect immediately, and persists them through whichever network management backends are detected on the host so they survive a reboot.

DHCP

SetDHCP turns each address family's DHCP client on or off independently — moving an interface to DHCPv4 while keeping a static IPv6 address is SetDHCP(ctx, "eth0", true, false). Enabling a family acquires a lease now through the detected manager; disabling only rewrites the configuration, leaving an existing lease to expire so a caller connected over the leased address is not cut off. AddAddress never changes DHCP state — it adds static addresses only. Interface.DHCP4 and Interface.DHCP6 report the state back.

Logging

The package uses a single, package-wide logger for non-fatal diagnostics from backends and control panels. It defaults to the logrus standard logger. Replace it with SetLogger before constructing any Configurator:

netconfig.SetLogger(myLogger) // anything with Printf/Println, e.g. *log.Logger

Options

NewConfigurator accepts functional options:

c, err := netconfig.NewConfigurator(ctx,
    netconfig.WithTestAddress("http://my-canary/health"), // connectivity-test URL
    netconfig.WithConnectivityCheck(false),                // skip the post-change probe + rollback
    netconfig.WithSkipConnectivityCheck(),                 // same as WithConnectivityCheck(false)
    netconfig.WithConnectivityTimeout(5 * time.Second),    // HTTP reachability timeout
    netconfig.WithPingCount(3),                            // ICMP probes per ping test
    netconfig.WithPingTimeout(10 * time.Second),           // overall ping-run timeout
    netconfig.WithBackupRetention(10),                     // .bak.* copies kept per config file (0 = keep all)
    netconfig.WithAllowPrimaryRemoval(true),               // let RemoveAddress remove the primary IP
    netconfig.WithSkipPanels(true),                        // skip all panel interaction (add/primary/remove)
    netconfig.WithAllowNoBackends(true),                   // don't fail when no persistence backend is detected
    netconfig.WithServiceReadyTimeout(90 * time.Second),   // how long to wait for a daemon-backed backend (default 60s, 0 = don't wait)
)

WithConnectivityCheck(false) (or the convenience WithSkipConnectivityCheck()) is useful on hosts with no outbound internet access, where the default reachability test would always fail and roll changes back. WithConnectivityTimeout, WithPingCount, and WithPingTimeout tune the probes used after gateway changes. WithBackupRetention limits how many .bak.<timestamp> copies each backend keeps per original config file; the oldest backups beyond this number are pruned on every save.

WithAllowPrimaryRemoval(true) overrides the default refusal to remove an address that is the primary of its family. The intended safe flow is to call SetPrimaryAddress on another address first and then remove the old one; enable the override only when you are deliberately tearing down the current primary.

WithSkipPanels(true) skips all control-panel interaction: AddAddress, SetPrimaryAddress, and RemoveAddress do not tell cPanel, Plesk, or InterWorx to reload, set the main IP, or release an IP. Only the running system and the network-manager configuration files are changed.

Choosing an interface

FindInterfaceByName accepts two well-known selectors in addition to a literal interface name:

  • "public-internet" — the interface that carries the IPv4 default gateway.
  • "public-internet-6" — the interface that carries the IPv6 default gateway.

Those answer "which interface is already on the internet". When the host is not on the internet yet — a freshly provisioned VM, an image whose NIC name is not known in advance — FindPhysicalInterfaces answers "which interface should be". It drops the devices that should never carry a public address (bridges, bonds, VLANs, tunnels, container veth pairs) and returns what is left in the order a caller should try them:

  1. Interfaces that are up, before those that are down.
  2. Interfaces already carrying a default gateway, IPv4 ahead of IPv6-only.
  3. Wired ahead of wireless.
  4. By name as a person reads it, so eth0 precedes eth1 and eth2 precedes eth10.

Paravirtual NICs (virtio, vmxnet, Xen) count as physical: a VM's only real interface is still the one to configure. The result is a ranking, not a decision — a caller with a further requirement, such as an interface not already holding a public address, filters the slice and takes the first survivor:

ifaces, err := c.GetInterfaces(ctx)
for _, iface := range netconfig.FindPhysicalInterfaces(ifaces) {
    if !hasPublicAddress(iface) {
        return iface.Name
    }
}

Interface.Up and Interface.Physical carry the two facts this ranking rests on, and are readable on their own. Both describe the running system, so they are only set by GetInterfaces.

Usage

package main

import (
    "context"
    "log"
    "net"

    netconfig "github.com/grmrgecko/go-network-configurator"
)

func main() {
    ctx := context.Background()

    c, err := netconfig.NewConfigurator(ctx)
    if err != nil {
        log.Fatal(err)
    }

    ifaces, err := c.GetInterfaces(ctx)
    if err != nil {
        log.Fatal(err)
    }
    iface := netconfig.FindInterfaceByName(netconfig.Public, ifaces)
    if iface == nil {
        log.Fatal("no public interface found")
    }

    // Swap the primary IP without dropping connectivity:
    // add a new address, switch the system over to it, then remove the old one.
    _, newIP, _ := net.ParseCIDR("203.0.113.20/24")
    _, oldIP, _ := net.ParseCIDR("203.0.113.10/24")

    if err := c.AddAddress(ctx, iface.Name, newIP, nil); err != nil {
        log.Fatal(err)
    }
    if err := c.SetPrimaryAddress(ctx, iface.Name, newIP); err != nil {
        log.Fatal(err)
    }
    if err := c.RemoveAddress(ctx, iface.Name, oldIP); err != nil {
        log.Fatal(err)
    }
}

The primary address is the first address listed on an interface. SetPrimaryAddress reorders an interface's addresses so the chosen IP is first within its family, both in the running kernel and in the written configuration. The address must already be present on the interface (compose it with AddAddress); IPv4 and IPv6 each have their own primary.

Safety

Operations that can break connectivity are guarded:

  • Adding an IP that already answers ping is refused.
  • After an address/gateway change, an internet-reachability test runs; if it fails, the previous runtime state (addresses and default route) is restored and the operation returns an error.
  • Removing an address that is the only route to the default gateway is refused.
  • Removing the primary address of its family is refused by default (use WithAllowPrimaryRemoval(true) to override). The safe flow is to promote another address with SetPrimaryAddress first, then remove the old one.

Most operations require elevated privileges (root on Linux, Administrator on Windows) to modify network state.

Development

go test ./...

Some tests exercise real kernel networking inside a throwaway network namespace and are skipped unless run as root:

sudo go test ./...