banger/internal/daemon/imagebuild.go

package daemon

import (
	"bytes"
	"context"
	"crypto/sha256"
	"errors"
	"fmt"
	"io"
	"os"
	"path/filepath"
	"strings"
	"time"

	"banger/internal/firecracker"
	"banger/internal/guest"
	"banger/internal/hostnat"
	"banger/internal/model"
	"banger/internal/system"
)

type imageBuildSpec struct {
	ID            string
	Name          string
	BaseRootfs    string
	RootfsPath    string
	BuildLog      io.Writer
	KernelPath    string
	InitrdPath    string
	ModulesDir    string
	PackagesPath  string
	InstallDocker bool
	Size          string
}

type imageBuildVM struct {
	Name      string
	GuestIP   string
	TapDevice string
	APISock   string
	PID       int
}

func (d *Daemon) runImageBuild(ctx context.Context, spec imageBuildSpec) error {
	if d.imageBuild != nil {
		return d.imageBuild(ctx, spec)
	}
	return d.runImageBuildNative(ctx, spec)
}

func (d *Daemon) runImageBuildNative(ctx context.Context, spec imageBuildSpec) (err error) {
	packages, err := system.ReadNormalizedLines(spec.PackagesPath)
	if err != nil {
		return err
	}
	if err := system.CopyFilePreferClone(spec.BaseRootfs, spec.RootfsPath); err != nil {
		return err
	}
	if spec.Size != "" {
		if err := resizeRootfs(spec.BaseRootfs, spec.RootfsPath, spec.Size); err != nil {
			return err
		}
	}

	vm, cleanup, err := d.startImageBuildVM(ctx, spec)
	if err != nil {
		return err
	}
	defer func() {
		cleanupErr := cleanup(context.Background())
		if cleanupErr != nil {
			err = errors.Join(err, cleanupErr)
		}
	}()

	sshAddress := vm.GuestIP + ":22"
	if _, err := fmt.Fprintf(spec.BuildLog, "[image.build] waiting for ssh on %s\n", sshAddress); err != nil {
		return err
	}
	waitCtx, cancel := context.WithTimeout(ctx, 60*time.Second)
	defer cancel()
	if err := guest.WaitForSSH(waitCtx, sshAddress, d.config.SSHKeyPath, time.Second); err != nil {
		return err
	}

	client, err := guest.Dial(ctx, sshAddress, d.config.SSHKeyPath)
	if err != nil {
		return err
	}
	defer client.Close()

	if err := writeBuildLog(spec.BuildLog, "configuring guest"); err != nil {
		return err
	}
	if err := client.RunScript(ctx, buildProvisionScript(vm.Name, d.config.DefaultDNS, packages, spec.InstallDocker), spec.BuildLog); err != nil {
		return err
	}
	if strings.TrimSpace(spec.ModulesDir) != "" {
		if err := writeBuildLog(spec.BuildLog, "copying kernel modules"); err != nil {
			return err
		}
		if err := client.StreamTar(ctx, spec.ModulesDir, buildModulesCommand(filepath.Base(spec.ModulesDir)), spec.BuildLog); err != nil {
			return err
		}
	}
	if err := writeBuildLog(spec.BuildLog, "shutting down guest"); err != nil {
		return err
	}
	if err := client.RunScript(ctx, "set -e\nsync\n", spec.BuildLog); err != nil {
		return err
	}
	return d.shutdownImageBuildVM(ctx, vm)
}

func resizeRootfs(baseRootfs, rootfsPath, sizeSpec string) error {
	sizeBytes, err := model.ParseSize(sizeSpec)
	if err != nil {
		return err
	}
	info, err := os.Stat(baseRootfs)
	if err != nil {
		return err
	}
	if sizeBytes < info.Size() {
		return fmt.Errorf("size must be >= base image size")
	}
	return system.ResizeExt4Image(context.Background(), system.NewRunner(), rootfsPath, sizeBytes)
}

func (d *Daemon) startImageBuildVM(ctx context.Context, spec imageBuildSpec) (imageBuildVM, func(context.Context) error, error) {
	if err := d.ensureBridge(ctx); err != nil {
		return imageBuildVM{}, nil, err
	}
	if err := d.ensureSocketDir(); err != nil {
		return imageBuildVM{}, nil, err
	}
	fcPath, err := d.firecrackerBinary()
	if err != nil {
		return imageBuildVM{}, nil, err
	}

	shortID := system.ShortID(spec.ID)
	guestIP, err := d.store.NextGuestIP(ctx, bridgePrefix(d.config.BridgeIP))
	if err != nil {
		return imageBuildVM{}, nil, err
	}
	vm := imageBuildVM{
		Name:      "image-build-" + shortID,
		GuestIP:   guestIP,
		TapDevice: "tap-img-" + shortID,
		APISock:   filepath.Join(d.layout.RuntimeDir, "img-"+shortID+".sock"),
	}
	if err := os.RemoveAll(vm.APISock); err != nil && !os.IsNotExist(err) {
		return imageBuildVM{}, nil, err
	}
	if err := d.createTap(ctx, vm.TapDevice); err != nil {
		return imageBuildVM{}, nil, err
	}
	if err := hostnat.Ensure(ctx, d.runner, vm.GuestIP, vm.TapDevice, true); err != nil {
		_, _ = d.runner.RunSudo(ctx, "ip", "link", "del", vm.TapDevice)
		return imageBuildVM{}, nil, err
	}

	firecrackerCtx := context.Background()
	machine, err := firecracker.NewMachine(firecrackerCtx, firecracker.MachineConfig{
		BinaryPath:      fcPath,
		VMID:            spec.ID,
		SocketPath:      vm.APISock,
		LogPath:         spec.RootfsPath + ".firecracker.log",
		MetricsPath:     filepath.Join(filepath.Dir(spec.RootfsPath), "metrics.json"),
		KernelImagePath: spec.KernelPath,
		InitrdPath:      spec.InitrdPath,
		KernelArgs:      system.BuildBootArgs(vm.Name, vm.GuestIP, d.config.BridgeIP, d.config.DefaultDNS),
		RootDrivePath:   spec.RootfsPath,
		TapDevice:       vm.TapDevice,
		VCPUCount:       model.DefaultVCPUCount,
		MemoryMiB:       model.DefaultMemoryMiB,
		Logger:          d.logger,
	})
	if err != nil {
		_ = hostnat.Ensure(ctx, d.runner, vm.GuestIP, vm.TapDevice, false)
		_, _ = d.runner.RunSudo(ctx, "ip", "link", "del", vm.TapDevice)
		return imageBuildVM{}, nil, err
	}
	if err := machine.Start(firecrackerCtx); err != nil {
		_ = hostnat.Ensure(ctx, d.runner, vm.GuestIP, vm.TapDevice, false)
		_, _ = d.runner.RunSudo(ctx, "ip", "link", "del", vm.TapDevice)
		return imageBuildVM{}, nil, err
	}
	vm.PID = d.resolveFirecrackerPID(firecrackerCtx, machine, vm.APISock)
	if err := d.ensureSocketAccess(ctx, vm.APISock); err != nil {
		_ = d.killVMProcess(context.Background(), vm.PID)
		_ = hostnat.Ensure(ctx, d.runner, vm.GuestIP, vm.TapDevice, false)
		_, _ = d.runner.RunSudo(ctx, "ip", "link", "del", vm.TapDevice)
		return imageBuildVM{}, nil, err
	}

	cleanup := func(cleanupCtx context.Context) error {
		if vm.PID > 0 && system.ProcessRunning(vm.PID, vm.APISock) {
			_ = d.killVMProcess(cleanupCtx, vm.PID)
			_ = d.waitForExit(cleanupCtx, vm.PID, vm.APISock, 10*time.Second)
		}
		_ = hostnat.Ensure(cleanupCtx, d.runner, vm.GuestIP, vm.TapDevice, false)
		if vm.TapDevice != "" {
			_, _ = d.runner.RunSudo(cleanupCtx, "ip", "link", "del", vm.TapDevice)
		}
		if vm.APISock != "" {
			_ = os.Remove(vm.APISock)
		}
		return nil
	}
	return vm, cleanup, nil
}

func (d *Daemon) shutdownImageBuildVM(ctx context.Context, vm imageBuildVM) error {
	buildVM := model.VMRecord{Runtime: model.VMRuntime{APISockPath: vm.APISock}}
	if err := d.sendCtrlAltDel(ctx, buildVM); err != nil {
		return err
	}
	return d.waitForExit(ctx, vm.PID, vm.APISock, 15*time.Second)
}

func buildProvisionScript(vmName, dnsServer string, packages []string, installDocker bool) string {
	var script bytes.Buffer
	script.WriteString("set -euo pipefail\n")
	fmt.Fprintf(&script, "printf 'nameserver %%s\\n' %s > /etc/resolv.conf\n", shellQuote(dnsServer))
	fmt.Fprintf(&script, "printf '%%s\\n' %s > /etc/hostname\n", shellQuote(vmName))
	fmt.Fprintf(&script, "printf '127.0.0.1 localhost\\n127.0.1.1 %%s\\n' %s > /etc/hosts\n", shellQuote(vmName))
	script.WriteString("touch /etc/fstab\n")
	script.WriteString("sed -i '\\|^/dev/vdb[[:space:]]\\+/home[[:space:]]|d; \\|^/dev/vdc[[:space:]]\\+/var[[:space:]]|d' /etc/fstab\n")
	script.WriteString("if ! grep -q '^tmpfs /run ' /etc/fstab; then echo 'tmpfs /run tmpfs defaults,nodev,nosuid,mode=0755 0 0' >> /etc/fstab; fi\n")
	script.WriteString("if ! grep -q '^tmpfs /tmp ' /etc/fstab; then echo 'tmpfs /tmp tmpfs defaults,nodev,nosuid,mode=1777 0 0' >> /etc/fstab; fi\n")
	script.WriteString("apt-get update\n")
	script.WriteString("DEBIAN_FRONTEND=noninteractive apt-get -y upgrade\n")
	fmt.Fprintf(&script, "PACKAGES=%s\n", shellArray(packages))
	script.WriteString("DEBIAN_FRONTEND=noninteractive apt-get -y install \"${PACKAGES[@]}\"\n")
	if installDocker {
		script.WriteString("DEBIAN_FRONTEND=noninteractive apt-get -y remove containerd || true\n")
		script.WriteString("if ! DEBIAN_FRONTEND=noninteractive apt-get -y install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin; then\n")
		script.WriteString("  DEBIAN_FRONTEND=noninteractive apt-get -y install docker.io\n")
		script.WriteString("fi\n")
		script.WriteString("if command -v systemctl >/dev/null 2>&1; then systemctl enable --now docker || true; fi\n")
	}
	script.WriteString("git config --system init.defaultBranch main\n")
	return script.String()
}

func buildModulesCommand(modulesBase string) string {
	return fmt.Sprintf("bash -se <<'EOF'\nset -euo pipefail\nmkdir -p /lib/modules\ntar -C /lib/modules -xf -\ndepmod -a %s\nmkdir -p /etc/modules-load.d\nprintf 'nf_tables\\nnft_chain_nat\\nveth\\nbr_netfilter\\noverlay\\n' > /etc/modules-load.d/docker-netfilter.conf\nmkdir -p /etc/sysctl.d\ncat > /etc/sysctl.d/99-docker.conf <<'SYSCTL'\nnet.bridge.bridge-nf-call-iptables = 1\nnet.bridge.bridge-nf-call-ip6tables = 1\nnet.ipv4.ip_forward = 1\nSYSCTL\nsysctl --system >/dev/null 2>&1 || true\nEOF", shellQuote(modulesBase))
}

func shellArray(values []string) string {
	quoted := make([]string, 0, len(values))
	for _, value := range values {
		quoted = append(quoted, shellQuote(value))
	}
	return "(" + strings.Join(quoted, " ") + ")"
}

func shellQuote(value string) string {
	return "'" + strings.ReplaceAll(value, "'", `'"'"'`) + "'"
}

func writeBuildLog(w io.Writer, message string) error {
	if w == nil {
		return nil
	}
	_, err := fmt.Fprintf(w, "[image.build] %s\n", message)
	return err
}

func packagesHash(lines []string) string {
	sum := sha256.Sum256([]byte(strings.Join(lines, "\n") + "\n"))
	return fmt.Sprintf("%x", sum)
}