|
| 1 | +package agent |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "crypto/rand" |
| 6 | + "crypto/rsa" |
| 7 | + "errors" |
| 8 | + "fmt" |
| 9 | + "io" |
| 10 | + "net" |
| 11 | + "os/exec" |
| 12 | + "os/user" |
| 13 | + "sync" |
| 14 | + "time" |
| 15 | + |
| 16 | + "cdr.dev/slog" |
| 17 | + "github.com/coder/coder/agent/usershell" |
| 18 | + "github.com/coder/coder/peer" |
| 19 | + "github.com/coder/coder/peerbroker" |
| 20 | + "github.com/coder/coder/pty" |
| 21 | + "github.com/coder/retry" |
| 22 | + |
| 23 | + "github.com/gliderlabs/ssh" |
| 24 | + gossh "golang.org/x/crypto/ssh" |
| 25 | + "golang.org/x/xerrors" |
| 26 | +) |
| 27 | + |
| 28 | +func DialSSH(conn *peer.Conn) (net.Conn, error) { |
| 29 | + channel, err := conn.Dial(context.Background(), "ssh", &peer.ChannelOptions{ |
| 30 | + Protocol: "ssh", |
| 31 | + }) |
| 32 | + if err != nil { |
| 33 | + return nil, err |
| 34 | + } |
| 35 | + return channel.NetConn(), nil |
| 36 | +} |
| 37 | + |
| 38 | +func DialSSHClient(conn *peer.Conn) (*gossh.Client, error) { |
| 39 | + netConn, err := DialSSH(conn) |
| 40 | + if err != nil { |
| 41 | + return nil, err |
| 42 | + } |
| 43 | + sshConn, channels, requests, err := gossh.NewClientConn(netConn, "localhost:22", &gossh.ClientConfig{ |
| 44 | + Config: gossh.Config{ |
| 45 | + Ciphers: []string{"arcfour"}, |
| 46 | + }, |
| 47 | + // SSH host validation isn't helpful, because obtaining a peer |
| 48 | + // connection already signifies user-intent to dial a workspace. |
| 49 | + // #nosec |
| 50 | + HostKeyCallback: gossh.InsecureIgnoreHostKey(), |
| 51 | + }) |
| 52 | + if err != nil { |
| 53 | + return nil, err |
| 54 | + } |
| 55 | + return gossh.NewClient(sshConn, channels, requests), nil |
| 56 | +} |
| 57 | + |
| 58 | +type Options struct { |
| 59 | + Logger slog.Logger |
| 60 | +} |
| 61 | + |
| 62 | +type Dialer func(ctx context.Context) (*peerbroker.Listener, error) |
| 63 | + |
| 64 | +func New(dialer Dialer, options *Options) io.Closer { |
| 65 | + ctx, cancelFunc := context.WithCancel(context.Background()) |
| 66 | + server := &server{ |
| 67 | + clientDialer: dialer, |
| 68 | + options: options, |
| 69 | + closeCancel: cancelFunc, |
| 70 | + closed: make(chan struct{}), |
| 71 | + } |
| 72 | + server.init(ctx) |
| 73 | + return server |
| 74 | +} |
| 75 | + |
| 76 | +type server struct { |
| 77 | + clientDialer Dialer |
| 78 | + options *Options |
| 79 | + |
| 80 | + closeCancel context.CancelFunc |
| 81 | + closeMutex sync.Mutex |
| 82 | + closed chan struct{} |
| 83 | + |
| 84 | + sshServer *ssh.Server |
| 85 | +} |
| 86 | + |
| 87 | +func (s *server) init(ctx context.Context) { |
| 88 | + // Clients' should ignore the host key when connecting. |
| 89 | + // The agent needs to authenticate with coderd to SSH, |
| 90 | + // so SSH authentication doesn't improve security. |
| 91 | + randomHostKey, err := rsa.GenerateKey(rand.Reader, 2048) |
| 92 | + if err != nil { |
| 93 | + panic(err) |
| 94 | + } |
| 95 | + randomSigner, err := gossh.NewSignerFromKey(randomHostKey) |
| 96 | + if err != nil { |
| 97 | + panic(err) |
| 98 | + } |
| 99 | + sshLogger := s.options.Logger.Named("ssh-server") |
| 100 | + forwardHandler := &ssh.ForwardedTCPHandler{} |
| 101 | + s.sshServer = &ssh.Server{ |
| 102 | + ChannelHandlers: ssh.DefaultChannelHandlers, |
| 103 | + ConnectionFailedCallback: func(conn net.Conn, err error) { |
| 104 | + sshLogger.Info(ctx, "ssh connection ended", slog.Error(err)) |
| 105 | + }, |
| 106 | + Handler: func(session ssh.Session) { |
| 107 | + err := s.handleSSHSession(session) |
| 108 | + if err != nil { |
| 109 | + s.options.Logger.Debug(ctx, "ssh session failed", slog.Error(err)) |
| 110 | + _ = session.Exit(1) |
| 111 | + return |
| 112 | + } |
| 113 | + }, |
| 114 | + HostSigners: []ssh.Signer{randomSigner}, |
| 115 | + LocalPortForwardingCallback: func(ctx ssh.Context, destinationHost string, destinationPort uint32) bool { |
| 116 | + // Allow local port forwarding all! |
| 117 | + sshLogger.Debug(ctx, "local port forward", |
| 118 | + slog.F("destination-host", destinationHost), |
| 119 | + slog.F("destination-port", destinationPort)) |
| 120 | + return true |
| 121 | + }, |
| 122 | + PtyCallback: func(ctx ssh.Context, pty ssh.Pty) bool { |
| 123 | + return true |
| 124 | + }, |
| 125 | + ReversePortForwardingCallback: func(ctx ssh.Context, bindHost string, bindPort uint32) bool { |
| 126 | + // Allow reverse port forwarding all! |
| 127 | + sshLogger.Debug(ctx, "local port forward", |
| 128 | + slog.F("bind-host", bindHost), |
| 129 | + slog.F("bind-port", bindPort)) |
| 130 | + return true |
| 131 | + }, |
| 132 | + RequestHandlers: map[string]ssh.RequestHandler{ |
| 133 | + "tcpip-forward": forwardHandler.HandleSSHRequest, |
| 134 | + "cancel-tcpip-forward": forwardHandler.HandleSSHRequest, |
| 135 | + }, |
| 136 | + ServerConfigCallback: func(ctx ssh.Context) *gossh.ServerConfig { |
| 137 | + return &gossh.ServerConfig{ |
| 138 | + Config: gossh.Config{ |
| 139 | + // "arcfour" is the fastest SSH cipher. We prioritize throughput |
| 140 | + // over encryption here, because the WebRTC connection is already |
| 141 | + // encrypted. If possible, we'd disable encryption entirely here. |
| 142 | + Ciphers: []string{"arcfour"}, |
| 143 | + }, |
| 144 | + NoClientAuth: true, |
| 145 | + } |
| 146 | + }, |
| 147 | + } |
| 148 | + |
| 149 | + go s.run(ctx) |
| 150 | +} |
| 151 | + |
| 152 | +func (*server) handleSSHSession(session ssh.Session) error { |
| 153 | + var ( |
| 154 | + command string |
| 155 | + args = []string{} |
| 156 | + err error |
| 157 | + ) |
| 158 | + |
| 159 | + username := session.User() |
| 160 | + if username == "" { |
| 161 | + currentUser, err := user.Current() |
| 162 | + if err != nil { |
| 163 | + return xerrors.Errorf("get current user: %w", err) |
| 164 | + } |
| 165 | + username = currentUser.Username |
| 166 | + } |
| 167 | + |
| 168 | + // gliderlabs/ssh returns a command slice of zero |
| 169 | + // when a shell is requested. |
| 170 | + if len(session.Command()) == 0 { |
| 171 | + command, err = usershell.Get(username) |
| 172 | + if err != nil { |
| 173 | + return xerrors.Errorf("get user shell: %w", err) |
| 174 | + } |
| 175 | + } else { |
| 176 | + command = session.Command()[0] |
| 177 | + if len(session.Command()) > 1 { |
| 178 | + args = session.Command()[1:] |
| 179 | + } |
| 180 | + } |
| 181 | + |
| 182 | + signals := make(chan ssh.Signal) |
| 183 | + breaks := make(chan bool) |
| 184 | + defer close(signals) |
| 185 | + defer close(breaks) |
| 186 | + go func() { |
| 187 | + for { |
| 188 | + select { |
| 189 | + case <-session.Context().Done(): |
| 190 | + return |
| 191 | + // Ignore signals and breaks for now! |
| 192 | + case <-signals: |
| 193 | + case <-breaks: |
| 194 | + } |
| 195 | + } |
| 196 | + }() |
| 197 | + |
| 198 | + cmd := exec.CommandContext(session.Context(), command, args...) |
| 199 | + cmd.Env = session.Environ() |
| 200 | + |
| 201 | + sshPty, windowSize, isPty := session.Pty() |
| 202 | + if isPty { |
| 203 | + cmd.Env = append(cmd.Env, fmt.Sprintf("TERM=%s", sshPty.Term)) |
| 204 | + ptty, process, err := pty.Start(cmd) |
| 205 | + if err != nil { |
| 206 | + return xerrors.Errorf("start command: %w", err) |
| 207 | + } |
| 208 | + go func() { |
| 209 | + for win := range windowSize { |
| 210 | + err := ptty.Resize(uint16(win.Width), uint16(win.Height)) |
| 211 | + if err != nil { |
| 212 | + panic(err) |
| 213 | + } |
| 214 | + } |
| 215 | + }() |
| 216 | + go func() { |
| 217 | + _, _ = io.Copy(ptty.Input(), session) |
| 218 | + }() |
| 219 | + go func() { |
| 220 | + _, _ = io.Copy(session, ptty.Output()) |
| 221 | + }() |
| 222 | + _, _ = process.Wait() |
| 223 | + _ = ptty.Close() |
| 224 | + return nil |
| 225 | + } |
| 226 | + |
| 227 | + cmd.Stdout = session |
| 228 | + cmd.Stderr = session |
| 229 | + // This blocks forever until stdin is received if we don't |
| 230 | + // use StdinPipe. It's unknown what causes this. |
| 231 | + stdinPipe, err := cmd.StdinPipe() |
| 232 | + if err != nil { |
| 233 | + return xerrors.Errorf("create stdin pipe: %w", err) |
| 234 | + } |
| 235 | + go func() { |
| 236 | + _, _ = io.Copy(stdinPipe, session) |
| 237 | + }() |
| 238 | + err = cmd.Start() |
| 239 | + if err != nil { |
| 240 | + return xerrors.Errorf("start: %w", err) |
| 241 | + } |
| 242 | + _ = cmd.Wait() |
| 243 | + return nil |
| 244 | +} |
| 245 | + |
| 246 | +func (s *server) run(ctx context.Context) { |
| 247 | + var peerListener *peerbroker.Listener |
| 248 | + var err error |
| 249 | + // An exponential back-off occurs when the connection is failing to dial. |
| 250 | + // This is to prevent server spam in case of a coderd outage. |
| 251 | + for retrier := retry.New(50*time.Millisecond, 10*time.Second); retrier.Wait(ctx); { |
| 252 | + peerListener, err = s.clientDialer(ctx) |
| 253 | + if err != nil { |
| 254 | + if errors.Is(err, context.Canceled) { |
| 255 | + return |
| 256 | + } |
| 257 | + if s.isClosed() { |
| 258 | + return |
| 259 | + } |
| 260 | + s.options.Logger.Warn(context.Background(), "failed to dial", slog.Error(err)) |
| 261 | + continue |
| 262 | + } |
| 263 | + s.options.Logger.Debug(context.Background(), "connected") |
| 264 | + break |
| 265 | + } |
| 266 | + select { |
| 267 | + case <-ctx.Done(): |
| 268 | + return |
| 269 | + default: |
| 270 | + } |
| 271 | + |
| 272 | + for { |
| 273 | + conn, err := peerListener.Accept() |
| 274 | + if err != nil { |
| 275 | + if s.isClosed() { |
| 276 | + return |
| 277 | + } |
| 278 | + s.options.Logger.Debug(ctx, "peer listener accept exited; restarting connection", slog.Error(err)) |
| 279 | + s.run(ctx) |
| 280 | + return |
| 281 | + } |
| 282 | + go s.handlePeerConn(ctx, conn) |
| 283 | + } |
| 284 | +} |
| 285 | + |
| 286 | +func (s *server) handlePeerConn(ctx context.Context, conn *peer.Conn) { |
| 287 | + for { |
| 288 | + channel, err := conn.Accept(ctx) |
| 289 | + if err != nil { |
| 290 | + if errors.Is(err, peer.ErrClosed) || s.isClosed() { |
| 291 | + return |
| 292 | + } |
| 293 | + s.options.Logger.Debug(ctx, "accept channel from peer connection", slog.Error(err)) |
| 294 | + return |
| 295 | + } |
| 296 | + |
| 297 | + switch channel.Protocol() { |
| 298 | + case "ssh": |
| 299 | + s.sshServer.HandleConn(channel.NetConn()) |
| 300 | + default: |
| 301 | + s.options.Logger.Warn(ctx, "unhandled protocol from channel", |
| 302 | + slog.F("protocol", channel.Protocol()), |
| 303 | + slog.F("label", channel.Label()), |
| 304 | + ) |
| 305 | + } |
| 306 | + } |
| 307 | +} |
| 308 | + |
| 309 | +// isClosed returns whether the API is closed or not. |
| 310 | +func (s *server) isClosed() bool { |
| 311 | + select { |
| 312 | + case <-s.closed: |
| 313 | + return true |
| 314 | + default: |
| 315 | + return false |
| 316 | + } |
| 317 | +} |
| 318 | + |
| 319 | +func (s *server) Close() error { |
| 320 | + s.closeMutex.Lock() |
| 321 | + defer s.closeMutex.Unlock() |
| 322 | + if s.isClosed() { |
| 323 | + return nil |
| 324 | + } |
| 325 | + close(s.closed) |
| 326 | + s.closeCancel() |
| 327 | + _ = s.sshServer.Close() |
| 328 | + return nil |
| 329 | +} |
0 commit comments