|
| 1 | +// SPDX-FileCopyrightText: Copyright 2025 Stacklok, Inc. |
| 2 | +// SPDX-License-Identifier: Apache-2.0 |
| 3 | + |
| 4 | +//go:build linux |
| 5 | + |
| 6 | +// thv-vm-init is the PID 1 init process for ToolHive go-microvm guest VMs. |
| 7 | +// It boots the guest (mounts, DHCP, SSH), reads the original OCI entrypoint |
| 8 | +// from /etc/thv-entrypoint.json, starts the MCP server as a child process, |
| 9 | +// and forwards termination signals. |
| 10 | +package main |
| 11 | + |
| 12 | +import ( |
| 13 | + "context" |
| 14 | + "log/slog" |
| 15 | + "os" |
| 16 | + "os/exec" |
| 17 | + "os/signal" |
| 18 | + "syscall" |
| 19 | + "time" |
| 20 | + |
| 21 | + "github.com/stacklok/go-microvm/guest/boot" |
| 22 | + "github.com/stacklok/go-microvm/guest/reaper" |
| 23 | +) |
| 24 | + |
| 25 | +const shutdownTimeout = 5 * time.Second |
| 26 | + |
| 27 | +func main() { |
| 28 | + logger := slog.New(slog.NewTextHandler(os.Stderr, nil)) |
| 29 | + |
| 30 | + // PID 1 must reap orphaned children. |
| 31 | + stopReaper := reaper.Start(logger) |
| 32 | + defer stopReaper() |
| 33 | + |
| 34 | + // Boot: essential mounts, DHCP networking, SSH server. |
| 35 | + shutdown, err := boot.Run(logger, |
| 36 | + boot.WithSSHKeysPath("/root/.ssh/authorized_keys"), |
| 37 | + boot.WithEnvFilePath("/etc/environment"), |
| 38 | + ) |
| 39 | + if err != nil { |
| 40 | + logger.Error("boot failed", "error", err) |
| 41 | + halt() |
| 42 | + return |
| 43 | + } |
| 44 | + |
| 45 | + // Load the original OCI entrypoint that was captured by the |
| 46 | + // InjectEntrypoint rootfs hook before WithInitOverride replaced it. |
| 47 | + ep, err := loadEntrypoint(DefaultEntrypointPath) |
| 48 | + if err != nil { |
| 49 | + logger.Error("failed to load entrypoint", "error", err) |
| 50 | + gracefulHalt(logger, shutdown) |
| 51 | + return |
| 52 | + } |
| 53 | + |
| 54 | + // Start the MCP server as a child process. |
| 55 | + cmd := exec.Command(ep.Cmd[0], ep.Cmd[1:]...) //nolint:gosec // cmd comes from trusted OCI config |
| 56 | + cmd.Stdout = os.Stdout |
| 57 | + cmd.Stderr = os.Stderr |
| 58 | + cmd.Env = append(os.Environ(), ep.Env...) |
| 59 | + if ep.WorkingDir != "" { |
| 60 | + cmd.Dir = ep.WorkingDir |
| 61 | + } |
| 62 | + |
| 63 | + if err := cmd.Start(); err != nil { |
| 64 | + logger.Error("failed to start MCP server", "error", err, "cmd", ep.Cmd) |
| 65 | + gracefulHalt(logger, shutdown) |
| 66 | + return |
| 67 | + } |
| 68 | + logger.Info("MCP server started", "pid", cmd.Process.Pid, "cmd", ep.Cmd) |
| 69 | + |
| 70 | + // Forward SIGTERM/SIGINT to the child process. |
| 71 | + sig := make(chan os.Signal, 1) |
| 72 | + signal.Notify(sig, syscall.SIGTERM, syscall.SIGINT) |
| 73 | + go func() { |
| 74 | + for received := range sig { |
| 75 | + logger.Info("received signal, forwarding to child", "signal", received) |
| 76 | + _ = cmd.Process.Signal(received) |
| 77 | + } |
| 78 | + }() |
| 79 | + |
| 80 | + // Wait for the child to exit. |
| 81 | + if err := cmd.Wait(); err != nil { |
| 82 | + if exitErr, ok := err.(*exec.ExitError); ok { |
| 83 | + logger.Info("MCP server exited", "code", exitErr.ExitCode()) |
| 84 | + } else { |
| 85 | + logger.Error("error waiting for MCP server", "error", err) |
| 86 | + } |
| 87 | + } |
| 88 | + |
| 89 | + gracefulHalt(logger, shutdown) |
| 90 | +} |
| 91 | + |
| 92 | +// gracefulHalt shuts down the boot services with a timeout and then halts the VM. |
| 93 | +func gracefulHalt(logger *slog.Logger, shutdown func()) { |
| 94 | + ctx, cancel := context.WithTimeout(context.Background(), shutdownTimeout) |
| 95 | + defer cancel() |
| 96 | + |
| 97 | + done := make(chan struct{}) |
| 98 | + go func() { |
| 99 | + shutdown() |
| 100 | + close(done) |
| 101 | + }() |
| 102 | + |
| 103 | + select { |
| 104 | + case <-done: |
| 105 | + logger.Info("shutdown complete") |
| 106 | + case <-ctx.Done(): |
| 107 | + logger.Warn("shutdown timed out") |
| 108 | + } |
| 109 | + |
| 110 | + halt() |
| 111 | +} |
| 112 | + |
| 113 | +// halt powers off the VM. As PID 1 inside a VM, Reboot with POWER_OFF |
| 114 | +// is the clean way to stop — calling os.Exit() would cause a kernel panic. |
| 115 | +func halt() { |
| 116 | + _ = syscall.Reboot(syscall.LINUX_REBOOT_CMD_POWER_OFF) |
| 117 | +} |
0 commit comments