banger/internal/firecracker/client.go
Thales Maciel fba30f26d4
firecracker: chown API + vsock sockets inside the sudo shell
Bug: Firecracker creates its API and vsock sockets as root:root 0700
(enforced by the intentional umask 077 in buildProcessRunner). The
daemon, running as the invoking user, then can't connect(2) to
either — AF_UNIX connect needs write permission on the socket file
and 0700 root-owned leaves thales without any.

firecracker-go-sdk's Machine.Start() blocks on waitForSocket, which
probes the socket with both os.Stat (succeeds — parent dir is the
user's XDG_RUNTIME_DIR) and an HTTP GET over the socket (fails —
EACCES on connect). The SDK loops for 3 seconds then fails with
"Firecracker did not create API socket ... context deadline exceeded".

The daemon's EnsureSocketAccess chown was meant to fix permissions,
but it runs *after* Machine.Start returns — and Start never returns
because it's still looping on the SDK's probe. Chicken-and-egg.

Fix: inside the sudo'd shell that launches firecracker, spawn a
background subshell that polls for each expected socket (API + vsock,
when configured) and chowns it to $SUDO_UID:$SUDO_GID as soon as it
appears. The background polling is bounded at 1s (20 × 50ms) so a
broken firecracker invocation doesn't leak a waiting shell.

Post-fix: socket appears root-owned 0600 briefly, is chowned to the
invoking user within ~50ms, SDK's HTTP probe succeeds, Machine.Start
returns normally. EnsureSocketAccess's later chmod 600 remains the
belt-and-braces guarantee on final mode.

Verified: manual repro of the shell script produces a socket owned
by thales:thales that a non-root python socket.connect() accepts.
Without the fix the same setup gives "PermissionError: [Errno 13]
Permission denied".

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-22 16:09:02 -03:00

292 lines
7.7 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

package firecracker
import (
"context"
"io"
"log/slog"
"os"
"os/exec"
"strings"
"sync"
sdk "github.com/firecracker-microvm/firecracker-go-sdk"
models "github.com/firecracker-microvm/firecracker-go-sdk/client/models"
"github.com/sirupsen/logrus"
"banger/internal/vsockagent"
)
type MachineConfig struct {
BinaryPath string
VMID string
SocketPath string
LogPath string
MetricsPath string
KernelImagePath string
InitrdPath string
KernelArgs string
Drives []DriveConfig
TapDevice string
VSockPath string
VSockCID uint32
VCPUCount int
MemoryMiB int
Logger *slog.Logger
}
type DriveConfig struct {
ID string
Path string
ReadOnly bool
IsRoot bool
}
type Machine struct {
machine *sdk.Machine
logFile *os.File
closeOnce sync.Once
}
type Client struct {
client *sdk.Client
}
func NewMachine(ctx context.Context, cfg MachineConfig) (*Machine, error) {
logFile, err := openLogFile(cfg.LogPath)
if err != nil {
return nil, err
}
cmd := buildProcessRunner(cfg, logFile)
machine, err := sdk.NewMachine(
ctx,
buildConfig(cfg),
sdk.WithProcessRunner(cmd),
sdk.WithLogger(newLogger(cfg.Logger)),
)
if err != nil {
if logFile != nil {
_ = logFile.Close()
}
return nil, err
}
return &Machine{machine: machine, logFile: logFile}, nil
}
func (m *Machine) Start(ctx context.Context) error {
if err := m.machine.Start(ctx); err != nil {
m.closeLog()
return err
}
go func() {
_ = m.machine.Wait(context.Background())
m.closeLog()
}()
return nil
}
func (m *Machine) PID() (int, error) {
return m.machine.PID()
}
func New(apiSock string, logger *slog.Logger) *Client {
return &Client{client: sdk.NewClient(apiSock, newLogger(logger), false)}
}
func (c *Client) SendCtrlAltDel(ctx context.Context) error {
action := models.InstanceActionInfoActionTypeSendCtrlAltDel
_, err := c.client.CreateSyncAction(ctx, &models.InstanceActionInfo{
ActionType: &action,
})
return err
}
func openLogFile(path string) (*os.File, error) {
if path == "" {
return nil, nil
}
return os.OpenFile(path, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o644)
}
func buildConfig(cfg MachineConfig) sdk.Config {
rootDrive, extraDrives := splitDrives(cfg.Drives)
drivesBuilder := sdk.NewDrivesBuilder(rootDrive.Path).
WithRootDrive(rootDrive.Path, sdk.WithDriveID(defaultDriveID(rootDrive, "rootfs")), sdk.WithReadOnly(rootDrive.ReadOnly))
for _, drive := range extraDrives {
if strings.TrimSpace(drive.Path) == "" {
continue
}
drivesBuilder = drivesBuilder.AddDrive(drive.Path, drive.ReadOnly, sdk.WithDriveID(defaultDriveID(drive, "drive")))
}
drives := drivesBuilder.Build()
return sdk.Config{
SocketPath: cfg.SocketPath,
LogPath: cfg.LogPath,
MetricsPath: cfg.MetricsPath,
KernelImagePath: cfg.KernelImagePath,
InitrdPath: cfg.InitrdPath,
KernelArgs: cfg.KernelArgs,
Drives: drives,
NetworkInterfaces: sdk.NetworkInterfaces{{
StaticConfiguration: &sdk.StaticNetworkConfiguration{
HostDevName: cfg.TapDevice,
},
}},
VsockDevices: buildVsockDevices(cfg),
MachineCfg: models.MachineConfiguration{
VcpuCount: sdk.Int64(int64(cfg.VCPUCount)),
MemSizeMib: sdk.Int64(int64(cfg.MemoryMiB)),
Smt: sdk.Bool(false),
},
VMID: cfg.VMID,
}
}
func buildVsockDevices(cfg MachineConfig) []sdk.VsockDevice {
if strings.TrimSpace(cfg.VSockPath) == "" || cfg.VSockCID == 0 {
return nil
}
return []sdk.VsockDevice{{
ID: "vsock",
Path: cfg.VSockPath,
CID: cfg.VSockCID,
}}
}
func splitDrives(drives []DriveConfig) (DriveConfig, []DriveConfig) {
root := DriveConfig{ID: "rootfs"}
var extras []DriveConfig
for _, drive := range drives {
if strings.TrimSpace(drive.Path) == "" {
continue
}
if drive.IsRoot {
root = drive
if root.ID == "" {
root.ID = "rootfs"
}
continue
}
extras = append(extras, drive)
}
return root, extras
}
func defaultDriveID(drive DriveConfig, fallback string) string {
if strings.TrimSpace(drive.ID) != "" {
return drive.ID
}
return fallback
}
func buildProcessRunner(cfg MachineConfig, logFile *os.File) *exec.Cmd {
// Two moving parts, run inside a single sudo'd shell:
//
// 1. umask 077 + exec firecracker → the API and vsock sockets
// firecracker creates are born 0600 owned by root (sudo user),
// not 0755. Without the umask there's a real window where a
// local attacker could hit the control plane.
//
// 2. A background subshell polls for each expected socket and
// chowns it to $SUDO_UID:$SUDO_GID as soon as it appears.
//
// The chown is required *before* the firecracker-go-sdk's
// waitForSocket returns from Machine.Start — the SDK does both an
// os.Stat and an HTTP GET over the socket, and AF_UNIX connect(2)
// needs write permission on the socket file. With the socket at
// 0600 root:root, the daemon process (running as the invoking
// user) gets EACCES on connect and the SDK loops until its 3s
// timeout. The daemon's post-Start EnsureSocketAccess chown would
// fix it, but Start never returns to hand control back.
//
// Racing the chown inside sudo's shell closes the gap: by the
// time the SDK's HTTP probe fires, the socket is already owned by
// the invoking user.
chownWatcher := func(path string) string {
// Bounded poll: 20 × 50ms = 1s. Matches the SDK's 3s wait
// budget with headroom and bails quietly if firecracker
// never creates the socket (e.g. bad args — the error
// surfaces through firecracker's non-zero exit).
return `for _ in $(seq 1 20); do [ -S ` + shellQuote(path) + ` ] && break; sleep 0.05; done; ` +
`[ -S ` + shellQuote(path) + ` ] && chown "$SUDO_UID:$SUDO_GID" ` + shellQuote(path) + ` || true`
}
watchers := chownWatcher(cfg.SocketPath)
if strings.TrimSpace(cfg.VSockPath) != "" {
watchers += "; " + chownWatcher(cfg.VSockPath)
}
script := "umask 077 && (" + watchers + ") & exec " + shellQuote(cfg.BinaryPath) +
" --api-sock " + shellQuote(cfg.SocketPath) +
" --id " + shellQuote(cfg.VMID)
// sudo -E preserves SUDO_UID / SUDO_GID (sudo sets them itself
// regardless, but -E is already the convention in this codebase
// and the background subshell needs them).
cmd := exec.Command("sudo", "-n", "-E", "sh", "-c", script)
cmd.Stdin = nil
if logFile != nil {
cmd.Stdout = logFile
cmd.Stderr = logFile
}
return cmd
}
func shellQuote(value string) string {
return "'" + strings.ReplaceAll(value, "'", `'"'"'`) + "'"
}
func newLogger(base *slog.Logger) *logrus.Entry {
logger := logrus.New()
logger.SetOutput(io.Discard)
logger.SetLevel(logrus.DebugLevel)
logger.AddHook(slogHook{logger: base})
return logrus.NewEntry(logger)
}
func HealthVSock(ctx context.Context, logger *slog.Logger, socketPath string) error {
return vsockagent.Health(ctx, logger, socketPath)
}
func PingVSock(ctx context.Context, logger *slog.Logger, socketPath string) error {
return HealthVSock(ctx, logger, socketPath)
}
type slogHook struct {
logger *slog.Logger
}
func (h slogHook) Levels() []logrus.Level {
return logrus.AllLevels
}
func (h slogHook) Fire(entry *logrus.Entry) error {
if h.logger == nil {
return nil
}
level := slog.LevelDebug
switch entry.Level {
case logrus.PanicLevel, logrus.FatalLevel, logrus.ErrorLevel:
level = slog.LevelError
case logrus.WarnLevel:
level = slog.LevelWarn
default:
level = slog.LevelDebug
}
attrs := make([]any, 0, len(entry.Data)*2+2)
attrs = append(attrs, "component", "firecracker_sdk")
for key, value := range entry.Data {
attrs = append(attrs, key, value)
}
h.logger.Log(context.Background(), level, entry.Message, attrs...)
return nil
}
func (m *Machine) closeLog() {
m.closeOnce.Do(func() {
if m.logFile != nil {
_ = m.logFile.Close()
}
})
}