microsoft/openvmm
Publicmirrored from https://github.com/microsoft/openvmmAvailable
openhcl/diag_client/src/lib.rs
928lines · modecode
| 1 | // Copyright (c) Microsoft Corporation. |
| 2 | // Licensed under the MIT License. |
| 3 | |
| 4 | //! The client for connecting to the Underhill diagnostics server. |
| 5 | |
| 6 | #![warn(missing_docs)] |
| 7 | |
| 8 | pub mod kmsg_stream; |
| 9 | |
| 10 | use anyhow::Context; |
| 11 | use diag_proto::network_packet_capture_request::OpData; |
| 12 | use diag_proto::network_packet_capture_request::Operation; |
| 13 | use diag_proto::ExecRequest; |
| 14 | use diag_proto::WaitRequest; |
| 15 | use diag_proto::WaitResponse; |
| 16 | use futures::AsyncReadExt; |
| 17 | use futures::AsyncWrite; |
| 18 | use futures::AsyncWriteExt; |
| 19 | use inspect::Node; |
| 20 | use inspect::ValueKind; |
| 21 | use inspect_proto::InspectResponse2; |
| 22 | use inspect_proto::UpdateResponse2; |
| 23 | use kmsg_stream::KmsgStream; |
| 24 | use mesh_rpc::service::Status; |
| 25 | use pal_async::driver::Driver; |
| 26 | use pal_async::socket::PolledSocket; |
| 27 | use pal_async::task::Spawn; |
| 28 | use std::io::ErrorKind; |
| 29 | use std::path::Path; |
| 30 | use std::path::PathBuf; |
| 31 | use std::time::Duration; |
| 32 | use thiserror::Error; |
| 33 | |
| 34 | #[cfg(windows)] |
| 35 | /// Functions for Hyper-V |
| 36 | pub mod hyperv { |
| 37 | use super::ConnectError; |
| 38 | use anyhow::Context; |
| 39 | use guid::Guid; |
| 40 | use pal_async::driver::Driver; |
| 41 | use pal_async::socket::PolledSocket; |
| 42 | use pal_async::windows::pipe::NamedPipeServer; |
| 43 | use std::fs::File; |
| 44 | use std::io::Write; |
| 45 | use std::process::Command; |
| 46 | use std::time::Duration; |
| 47 | use vmsocket::VmAddress; |
| 48 | use vmsocket::VmSocket; |
| 49 | use vmsocket::VmStream; |
| 50 | |
| 51 | /// Defines how to access the serial port |
| 52 | pub enum ComPortAccessInfo { |
| 53 | /// Access by number |
| 54 | PortNumber(u32), |
| 55 | /// Access through a named pipe |
| 56 | PortPipePath(String), |
| 57 | } |
| 58 | |
| 59 | /// Get ID from name |
| 60 | pub fn vm_id_from_name(name: &str) -> anyhow::Result<Guid> { |
| 61 | let output = Command::new("hvc.exe") |
| 62 | .arg("id") |
| 63 | .arg(name) |
| 64 | .output() |
| 65 | .context("failed to launch hvc")?; |
| 66 | |
| 67 | if output.status.success() { |
| 68 | let stdout = std::str::from_utf8(&output.stdout) |
| 69 | .context("failed to parse hvc output")? |
| 70 | .trim(); |
| 71 | Ok(stdout |
| 72 | .parse() |
| 73 | .with_context(|| format!("failed to parse VM ID '{}'", &stdout))?) |
| 74 | } else { |
| 75 | anyhow::bail!( |
| 76 | "{}", |
| 77 | std::str::from_utf8(&output.stderr).context("failed to parse hvc error output")? |
| 78 | ) |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | /// Connect to Hyper-V socket |
| 83 | pub async fn connect_vsock( |
| 84 | driver: &(impl Driver + ?Sized), |
| 85 | vm_id: Guid, |
| 86 | port: u32, |
| 87 | ) -> Result<VmStream, ConnectError> { |
| 88 | let socket = VmSocket::new() |
| 89 | .context("failed to create AF_HYPERV socket") |
| 90 | .map_err(ConnectError::other)?; |
| 91 | |
| 92 | socket |
| 93 | .set_connect_timeout(Duration::from_secs(1)) |
| 94 | .context("failed to set connect timeout") |
| 95 | .map_err(ConnectError::other)?; |
| 96 | |
| 97 | socket |
| 98 | .set_high_vtl(true) |
| 99 | .context("failed to set socket for VTL2") |
| 100 | .map_err(ConnectError::other)?; |
| 101 | |
| 102 | let mut socket: PolledSocket<socket2::Socket> = PolledSocket::new(driver, socket.into()) |
| 103 | .context("failed to create polled socket") |
| 104 | .map_err(ConnectError::other)?; |
| 105 | |
| 106 | socket |
| 107 | .connect(&VmAddress::hyperv_vsock(vm_id, port).into()) |
| 108 | .await |
| 109 | .map_err(ConnectError::connect)?; |
| 110 | |
| 111 | Ok(socket.convert().into_inner()) |
| 112 | } |
| 113 | |
| 114 | /// Opens a serial port on a Hyper-V VM. |
| 115 | /// |
| 116 | /// If the VM is not running, it will act as a server for the pipe. Hyper-V |
| 117 | /// will connect to the server once the VM starts running. |
| 118 | pub async fn open_serial_port( |
| 119 | driver: &(impl Driver + ?Sized), |
| 120 | vm: &str, |
| 121 | port: ComPortAccessInfo, |
| 122 | ) -> anyhow::Result<File> { |
| 123 | let path = match port { |
| 124 | ComPortAccessInfo::PortNumber(num) => { |
| 125 | let output = Command::new("powershell.exe") |
| 126 | .arg(format!( |
| 127 | r#"$x = Get-VMComPort "{vm}" -Number {num} -ErrorAction Stop; $x.Path"#, |
| 128 | )) |
| 129 | .output() |
| 130 | .context("failed to query VM com port")?; |
| 131 | |
| 132 | if !output.status.success() { |
| 133 | let _ = std::io::stderr().write_all(&output.stderr); |
| 134 | anyhow::bail!( |
| 135 | "failed to query VM com port: exit status {}", |
| 136 | output.status.code().unwrap() |
| 137 | ); |
| 138 | } |
| 139 | String::from_utf8(output.stdout)? |
| 140 | } |
| 141 | ComPortAccessInfo::PortPipePath(path) => path, |
| 142 | }; |
| 143 | |
| 144 | let path = path.trim(); |
| 145 | if path.is_empty() { |
| 146 | anyhow::bail!("Requested VM COM port is not configured"); |
| 147 | } |
| 148 | |
| 149 | let pipe = match fs_err::OpenOptions::new().read(true).write(true).open(path) { |
| 150 | Ok(pipe) => pipe.into(), |
| 151 | Err(err) if err.kind() == std::io::ErrorKind::NotFound => { |
| 152 | let server = |
| 153 | NamedPipeServer::create(path).context("failed to create pipe server")?; |
| 154 | |
| 155 | server.accept(driver)?.await? |
| 156 | } |
| 157 | Err(err) => Err(err)?, |
| 158 | }; |
| 159 | |
| 160 | Ok(pipe) |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | /// Connect to a vsock with port and path |
| 165 | pub async fn connect_hybrid_vsock( |
| 166 | driver: &(impl Driver + ?Sized), |
| 167 | path: &Path, |
| 168 | port: u32, |
| 169 | ) -> Result<PolledSocket<socket2::Socket>, ConnectError> { |
| 170 | let socket = unix_socket::UnixStream::connect(path).map_err(ConnectError::connect)?; |
| 171 | let mut socket = PolledSocket::new(driver, socket).map_err(ConnectError::other)?; |
| 172 | socket |
| 173 | .write_all(format!("CONNECT {port}\n").as_bytes()) |
| 174 | .await |
| 175 | .map_err(ConnectError::other)?; |
| 176 | |
| 177 | let mut ok = [0; 3]; |
| 178 | socket |
| 179 | .read_exact(&mut ok) |
| 180 | .await |
| 181 | .map_err(ConnectError::other)?; |
| 182 | if &ok != b"OK " { |
| 183 | // FUTURE: consider returning an error that can be retried. This may |
| 184 | // require some changes to the hybrid vsock protocol, unclear. |
| 185 | return Err(ConnectError::other(anyhow::anyhow!( |
| 186 | "missing hybrid vsock response" |
| 187 | ))); |
| 188 | } |
| 189 | |
| 190 | for _ in 0.."4294967295\n".len() { |
| 191 | let mut b = [0]; |
| 192 | socket |
| 193 | .read_exact(&mut b) |
| 194 | .await |
| 195 | .map_err(ConnectError::other)?; |
| 196 | if b[0] == b'\n' { |
| 197 | // Don't need to parse the host port number. |
| 198 | return Ok(socket.convert()); |
| 199 | } |
| 200 | } |
| 201 | Err(ConnectError::other(anyhow::anyhow!( |
| 202 | "invalid hybrid vsock response" |
| 203 | ))) |
| 204 | } |
| 205 | |
| 206 | enum SocketType<'a> { |
| 207 | #[cfg(windows)] |
| 208 | VmId { |
| 209 | vm_id: guid::Guid, |
| 210 | port: u32, |
| 211 | }, |
| 212 | HybridVsock { |
| 213 | path: &'a Path, |
| 214 | port: u32, |
| 215 | }, |
| 216 | } |
| 217 | |
| 218 | async fn new_data_connection( |
| 219 | driver: &(impl Driver + ?Sized), |
| 220 | typ: SocketType<'_>, |
| 221 | ) -> anyhow::Result<(u64, PolledSocket<socket2::Socket>)> { |
| 222 | let mut socket = match typ { |
| 223 | #[cfg(windows)] |
| 224 | SocketType::VmId { vm_id, port } => { |
| 225 | let socket = hyperv::connect_vsock(driver, vm_id, port).await?; |
| 226 | PolledSocket::new(driver, socket2::Socket::from(socket))? |
| 227 | } |
| 228 | SocketType::HybridVsock { path, port } => connect_hybrid_vsock(driver, path, port).await?, |
| 229 | }; |
| 230 | |
| 231 | // Read the 8 byte connection id which is always sent first on the connection. |
| 232 | let mut id = [0; 8]; |
| 233 | socket |
| 234 | .read_exact(&mut id) |
| 235 | .await |
| 236 | .context("reading connection id")?; |
| 237 | let id = u64::from_ne_bytes(id); |
| 238 | Ok((id, socket)) |
| 239 | } |
| 240 | |
| 241 | /// Represents different VM types. |
| 242 | enum VmType { |
| 243 | /// A Hyper-V VM represented by a VM ID GUID, which uses a VmSocket to connect. |
| 244 | #[cfg(windows)] |
| 245 | HyperV(guid::Guid), |
| 246 | /// A VM which uses hybrid vsock over Unix sockets. |
| 247 | HybridVsock(PathBuf), |
| 248 | /// A VM that cannot be used for data connections. |
| 249 | None, |
| 250 | } |
| 251 | |
| 252 | /// The diagnostics client. |
| 253 | pub struct DiagClient { |
| 254 | vm: VmType, |
| 255 | ttrpc: mesh_rpc::Client, |
| 256 | driver: Box<dyn Driver>, |
| 257 | } |
| 258 | |
| 259 | /// Defines packet capture operations. |
| 260 | #[derive(PartialEq)] |
| 261 | pub enum PacketCaptureOperation { |
| 262 | /// Query details. |
| 263 | Query, |
| 264 | /// Start packet capture. |
| 265 | Start, |
| 266 | /// Stop packet capture. |
| 267 | Stop, |
| 268 | } |
| 269 | |
| 270 | /// An error connecting to the diagnostics server. |
| 271 | #[derive(Debug, Error)] |
| 272 | #[error("failed to connect")] |
| 273 | pub struct ConnectError { |
| 274 | #[source] |
| 275 | err: anyhow::Error, |
| 276 | kind: ConnectErrorKind, |
| 277 | } |
| 278 | |
| 279 | #[derive(Debug)] |
| 280 | enum ConnectErrorKind { |
| 281 | Other, |
| 282 | VmNotStarted, |
| 283 | ServerTimedOut, |
| 284 | } |
| 285 | |
| 286 | impl ConnectError { |
| 287 | /// Returns the time to wait before retrying the connection. If `None`, the |
| 288 | /// connection should not be retried. |
| 289 | pub fn retry_timeout(&self) -> Option<Duration> { |
| 290 | match self.kind { |
| 291 | ConnectErrorKind::VmNotStarted => Some(Duration::from_secs(1)), |
| 292 | ConnectErrorKind::ServerTimedOut => { |
| 293 | // The socket infrastructure has an internal timeout. |
| 294 | Some(Duration::ZERO) |
| 295 | } |
| 296 | _ => None, |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | fn other(err: impl Into<anyhow::Error>) -> Self { |
| 301 | Self { |
| 302 | err: err.into(), |
| 303 | kind: ConnectErrorKind::Other, |
| 304 | } |
| 305 | } |
| 306 | |
| 307 | fn connect(err: std::io::Error) -> Self { |
| 308 | let kind = match err.kind() { |
| 309 | ErrorKind::AddrNotAvailable => ConnectErrorKind::VmNotStarted, |
| 310 | ErrorKind::TimedOut => ConnectErrorKind::ServerTimedOut, |
| 311 | _ => match err.raw_os_error() { |
| 312 | #[cfg(windows)] |
| 313 | Some(windows_sys::Win32::Networking::WinSock::WSAENETUNREACH) => { |
| 314 | ConnectErrorKind::VmNotStarted |
| 315 | } |
| 316 | _ => ConnectErrorKind::Other, |
| 317 | }, |
| 318 | }; |
| 319 | Self { |
| 320 | err: anyhow::Error::from(err).context("failed to connect"), |
| 321 | kind, |
| 322 | } |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | impl DiagClient { |
| 327 | /// Creates a client from Hyper-V VM name. |
| 328 | #[cfg(windows)] |
| 329 | pub async fn from_hyperv_name( |
| 330 | driver: impl Driver + Spawn, |
| 331 | name: &str, |
| 332 | ) -> Result<Self, ConnectError> { |
| 333 | Self::from_hyperv_id( |
| 334 | driver, |
| 335 | hyperv::vm_id_from_name(name).map_err(ConnectError::other)?, |
| 336 | ) |
| 337 | .await |
| 338 | } |
| 339 | |
| 340 | /// Creates a client from a Hyper-V or HCS VM ID. |
| 341 | #[cfg(windows)] |
| 342 | pub async fn from_hyperv_id( |
| 343 | driver: impl Driver + Spawn, |
| 344 | vm_id: guid::Guid, |
| 345 | ) -> Result<Self, ConnectError> { |
| 346 | let socket = hyperv::connect_vsock(&driver, vm_id, diag_proto::VSOCK_CONTROL_PORT).await?; |
| 347 | Ok(Self { |
| 348 | vm: VmType::HyperV(vm_id), |
| 349 | ttrpc: mesh_rpc::Client::new(&driver, socket), |
| 350 | driver: Box::new(driver), |
| 351 | }) |
| 352 | } |
| 353 | |
| 354 | /// Creates a client from a Unix socket path. |
| 355 | pub async fn from_socket( |
| 356 | driver: impl Driver + Spawn, |
| 357 | path: &Path, |
| 358 | ) -> Result<Self, ConnectError> { |
| 359 | let socket = unix_socket::UnixStream::connect(path).map_err(ConnectError::other)?; |
| 360 | Ok(Self { |
| 361 | vm: VmType::None, |
| 362 | ttrpc: mesh_rpc::Client::new(&driver, socket), |
| 363 | driver: Box::new(driver), |
| 364 | }) |
| 365 | } |
| 366 | |
| 367 | /// Creates a client from a hybrid vsock Unix socket path. |
| 368 | pub async fn from_hybrid_vsock( |
| 369 | driver: impl Driver + Spawn, |
| 370 | path: &Path, |
| 371 | ) -> Result<Self, ConnectError> { |
| 372 | let socket = connect_hybrid_vsock(&driver, path, diag_proto::VSOCK_CONTROL_PORT) |
| 373 | .await? |
| 374 | .into_inner(); |
| 375 | Ok(Self { |
| 376 | vm: VmType::HybridVsock(path.into()), |
| 377 | ttrpc: mesh_rpc::Client::new(&driver, socket), |
| 378 | driver: Box::new(driver), |
| 379 | }) |
| 380 | } |
| 381 | |
| 382 | /// Creates a client from an existing connection. |
| 383 | /// |
| 384 | /// This client won't be usable with operations that require additional connections. |
| 385 | pub fn from_conn<T>(driver: impl Driver + Spawn, conn: T) -> Self |
| 386 | where |
| 387 | T: 'static + Send + Sync + pal_async::socket::AsSockRef + std::io::Read + std::io::Write, |
| 388 | { |
| 389 | Self { |
| 390 | vm: VmType::None, |
| 391 | ttrpc: mesh_rpc::Client::new(&driver, conn), |
| 392 | driver: Box::new(driver), |
| 393 | } |
| 394 | } |
| 395 | |
| 396 | /// Creates a builder for execing a command. |
| 397 | pub fn exec(&self, command: impl AsRef<str>) -> ExecBuilder<'_> { |
| 398 | ExecBuilder { |
| 399 | client: self, |
| 400 | with_stdin: false, |
| 401 | with_stdout: false, |
| 402 | with_stderr: false, |
| 403 | request: ExecRequest { |
| 404 | command: command.as_ref().to_owned(), |
| 405 | ..Default::default() |
| 406 | }, |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | /// Creates a new data connection socket. |
| 411 | /// |
| 412 | /// This can be used with [`DiagClient::custom_call`]. |
| 413 | pub async fn connect_data(&self) -> anyhow::Result<(u64, PolledSocket<socket2::Socket>)> { |
| 414 | let socket_type = match &self.vm { |
| 415 | #[cfg(windows)] |
| 416 | VmType::HyperV(guid) => SocketType::VmId { |
| 417 | vm_id: *guid, |
| 418 | port: diag_proto::VSOCK_DATA_PORT, |
| 419 | }, |
| 420 | VmType::HybridVsock(path) => SocketType::HybridVsock { |
| 421 | path, |
| 422 | port: diag_proto::VSOCK_DATA_PORT, |
| 423 | }, |
| 424 | VmType::None => { |
| 425 | anyhow::bail!("cannot make additional connections with this client") |
| 426 | } |
| 427 | }; |
| 428 | new_data_connection(self.driver.as_ref(), socket_type).await |
| 429 | } |
| 430 | |
| 431 | /// Sends an inspection request to the server. |
| 432 | pub async fn inspect( |
| 433 | &self, |
| 434 | path: impl Into<String>, |
| 435 | depth: Option<usize>, |
| 436 | timeout: Option<Duration>, |
| 437 | ) -> anyhow::Result<Node> { |
| 438 | let response = self.ttrpc.start_call( |
| 439 | inspect_proto::InspectService::Inspect, |
| 440 | inspect_proto::InspectRequest { |
| 441 | path: path.into(), |
| 442 | // It would be better to pass an Option<u32> in the proto, but that would break backcompat. |
| 443 | depth: depth.unwrap_or(u32::MAX as usize) as u32, |
| 444 | }, |
| 445 | timeout, |
| 446 | ); |
| 447 | |
| 448 | let response = response.upcast::<Result<InspectResponse2, Status>>(); |
| 449 | let response = response.await?.map_err(grpc_status)?; |
| 450 | |
| 451 | Ok(response.result) |
| 452 | } |
| 453 | |
| 454 | /// Updates an inspectable value. |
| 455 | pub async fn update( |
| 456 | &self, |
| 457 | path: impl Into<String>, |
| 458 | value: impl Into<String>, |
| 459 | ) -> anyhow::Result<inspect::Value> { |
| 460 | let response = self.ttrpc.start_call( |
| 461 | inspect_proto::InspectService::Update, |
| 462 | inspect_proto::UpdateRequest { |
| 463 | path: path.into(), |
| 464 | value: value.into(), |
| 465 | }, |
| 466 | None, |
| 467 | ); |
| 468 | |
| 469 | let response = response.upcast::<Result<UpdateResponse2, Status>>(); |
| 470 | let response = response.await?.map_err(grpc_status)?; |
| 471 | |
| 472 | Ok(response.new_value) |
| 473 | } |
| 474 | |
| 475 | /// Get PID of a given process |
| 476 | pub async fn get_pid(&self, name: &str) -> anyhow::Result<i32> { |
| 477 | let hosts = self.inspect("mesh/hosts", Some(1), None).await?; |
| 478 | let mut plist = Vec::new(); |
| 479 | |
| 480 | let Node::Dir(processes) = hosts else { |
| 481 | anyhow::bail!("Hosts node is not a dir"); |
| 482 | }; |
| 483 | for process in processes { |
| 484 | let Node::Dir(pnode) = process.node else { |
| 485 | anyhow::bail!("Process node is not a dir"); |
| 486 | }; |
| 487 | for entry in pnode { |
| 488 | if entry.name == "name" { |
| 489 | let Node::Value(value) = entry.node else { |
| 490 | anyhow::bail!("Name node is not a value"); |
| 491 | }; |
| 492 | let ValueKind::String(strval) = value.kind else { |
| 493 | anyhow::bail!("Name node is not a string"); |
| 494 | }; |
| 495 | if strval == name { |
| 496 | return Ok(process.name.parse()?); |
| 497 | } |
| 498 | plist.push(strval); |
| 499 | } |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | anyhow::bail!("PID of {name} not found. Processes: {:?}", plist) |
| 504 | } |
| 505 | |
| 506 | /// Starts the VM. |
| 507 | pub async fn start( |
| 508 | &self, |
| 509 | env: impl IntoIterator<Item = (String, Option<String>)>, |
| 510 | args: impl IntoIterator<Item = String>, |
| 511 | ) -> anyhow::Result<()> { |
| 512 | let request = diag_proto::StartRequest { |
| 513 | env: env |
| 514 | .into_iter() |
| 515 | .map(|(name, value)| diag_proto::EnvPair { name, value }) |
| 516 | .collect(), |
| 517 | args: args.into_iter().collect(), |
| 518 | }; |
| 519 | self.ttrpc |
| 520 | .call(diag_proto::UnderhillDiag::Start, request) |
| 521 | .await? |
| 522 | .map_err(grpc_status)?; |
| 523 | |
| 524 | Ok(()) |
| 525 | } |
| 526 | |
| 527 | /// Gets the contents of /dev/kmsg |
| 528 | pub async fn kmsg(&self, follow: bool) -> anyhow::Result<KmsgStream> { |
| 529 | let (conn, socket) = self.connect_data().await?; |
| 530 | |
| 531 | self.ttrpc |
| 532 | .call( |
| 533 | diag_proto::UnderhillDiag::Kmsg, |
| 534 | diag_proto::KmsgRequest { follow, conn }, |
| 535 | ) |
| 536 | .await? |
| 537 | .map_err(grpc_status)?; |
| 538 | |
| 539 | Ok(KmsgStream::new(socket)) |
| 540 | } |
| 541 | |
| 542 | /// Gets the contents of the file |
| 543 | pub async fn read_file( |
| 544 | &self, |
| 545 | follow: bool, |
| 546 | file_path: String, |
| 547 | ) -> anyhow::Result<PolledSocket<socket2::Socket>> { |
| 548 | let (conn, socket) = self.connect_data().await?; |
| 549 | |
| 550 | self.ttrpc |
| 551 | .call( |
| 552 | diag_proto::UnderhillDiag::ReadFile, |
| 553 | diag_proto::FileRequest { |
| 554 | follow, |
| 555 | conn, |
| 556 | file_path, |
| 557 | }, |
| 558 | ) |
| 559 | .await? |
| 560 | .map_err(grpc_status)?; |
| 561 | |
| 562 | Ok(socket) |
| 563 | } |
| 564 | |
| 565 | /// Issues a call to the server using a custom RPC. |
| 566 | /// |
| 567 | /// This can be used to support extension RPCs that are not part of the main |
| 568 | /// diagnostics service. |
| 569 | pub fn custom_call<F, R, T, U>( |
| 570 | &self, |
| 571 | rpc: F, |
| 572 | input: T, |
| 573 | timeout: Option<Duration>, |
| 574 | ) -> mesh::OneshotReceiver<Result<U, Status>> |
| 575 | where |
| 576 | F: FnOnce(T, mesh::OneshotSender<Result<U, Status>>) -> R, |
| 577 | R: mesh_rpc::service::ServiceRpc, |
| 578 | U: mesh::MeshPayload, |
| 579 | { |
| 580 | self.ttrpc.start_call(rpc, input, timeout) |
| 581 | } |
| 582 | |
| 583 | /// Crashes the VM. |
| 584 | pub async fn crash(&self, pid: i32) -> anyhow::Result<()> { |
| 585 | self.ttrpc |
| 586 | .call( |
| 587 | diag_proto::UnderhillDiag::Crash, |
| 588 | diag_proto::CrashRequest { pid }, |
| 589 | ) |
| 590 | .await? |
| 591 | .map_err(grpc_status)?; |
| 592 | |
| 593 | Ok(()) |
| 594 | } |
| 595 | |
| 596 | /// Sets up network packet capture trace. |
| 597 | pub async fn packet_capture( |
| 598 | &self, |
| 599 | op: PacketCaptureOperation, |
| 600 | num_streams: u32, |
| 601 | snaplen: u16, |
| 602 | ) -> anyhow::Result<(Vec<PolledSocket<socket2::Socket>>, u32)> { |
| 603 | let mut sockets = Vec::new(); |
| 604 | let op_data = match op { |
| 605 | PacketCaptureOperation::Start => { |
| 606 | let mut conns = Vec::new(); |
| 607 | for _ in 0..num_streams { |
| 608 | let (conn, socket) = self.connect_data().await?; |
| 609 | conns.push(conn); |
| 610 | sockets.push(socket); |
| 611 | } |
| 612 | Some(OpData::StartData(diag_proto::StartPacketCaptureData { |
| 613 | snaplen: snaplen.into(), |
| 614 | conns, |
| 615 | })) |
| 616 | } |
| 617 | _ => None, |
| 618 | }; |
| 619 | |
| 620 | let operation = match op { |
| 621 | PacketCaptureOperation::Query => Operation::Query, |
| 622 | PacketCaptureOperation::Start => Operation::Start, |
| 623 | PacketCaptureOperation::Stop => Operation::Stop, |
| 624 | }; |
| 625 | |
| 626 | let response = self |
| 627 | .ttrpc |
| 628 | .call( |
| 629 | diag_proto::UnderhillDiag::PacketCapture, |
| 630 | diag_proto::NetworkPacketCaptureRequest { |
| 631 | operation: operation.into(), |
| 632 | op_data, |
| 633 | }, |
| 634 | ) |
| 635 | .await? |
| 636 | .map_err(grpc_status)?; |
| 637 | |
| 638 | Ok((sockets, response.num_streams)) |
| 639 | } |
| 640 | |
| 641 | /// Saves a core dump file being streamed from Underhill |
| 642 | pub async fn core_dump( |
| 643 | &self, |
| 644 | pid: i32, |
| 645 | mut writer: impl AsyncWrite + Unpin, |
| 646 | mut stderr: impl AsyncWrite + Unpin, |
| 647 | verbose: bool, |
| 648 | ) -> anyhow::Result<()> { |
| 649 | // Launch hcl-dump to dump the target process. Use raw_socket_io so that |
| 650 | // the diagnostics process does not have to be running during the core |
| 651 | // dump process; this ensures that we can dump the diagnostics process, |
| 652 | // too. |
| 653 | let mut process = self.exec("/bin/underhill-dump"); |
| 654 | if verbose { |
| 655 | process.args(["-v"]); |
| 656 | } |
| 657 | let mut process = process |
| 658 | .args([pid.to_string()]) |
| 659 | .stdin(false) |
| 660 | .stdout(true) |
| 661 | .stderr(true) |
| 662 | .raw_socket_io(true) |
| 663 | .spawn() |
| 664 | .await |
| 665 | .context("failed to launch underhill-dump")?; |
| 666 | |
| 667 | let process_stdout = PolledSocket::new(&self.driver, process.stdout.take().unwrap())?; |
| 668 | let process_stderr = PolledSocket::new(&self.driver, process.stderr.take().unwrap())?; |
| 669 | |
| 670 | let out = futures::io::copy(process_stdout, &mut writer); |
| 671 | let err = futures::io::copy(process_stderr, &mut stderr); |
| 672 | |
| 673 | futures::try_join!(out, err)?; |
| 674 | |
| 675 | let status = process |
| 676 | .wait() |
| 677 | .await |
| 678 | .context("failed to wait for underhill-dump")?; |
| 679 | |
| 680 | if !status.success() { |
| 681 | anyhow::bail!( |
| 682 | "underhill-dump failed with exit code {}", |
| 683 | status.exit_code() |
| 684 | ); |
| 685 | } |
| 686 | Ok(()) |
| 687 | } |
| 688 | |
| 689 | /// Restarts the Underhill worker. |
| 690 | pub async fn restart(&self) -> anyhow::Result<()> { |
| 691 | self.ttrpc |
| 692 | .call(diag_proto::UnderhillDiag::Restart, ()) |
| 693 | .await? |
| 694 | .map_err(grpc_status)?; |
| 695 | |
| 696 | Ok(()) |
| 697 | } |
| 698 | |
| 699 | /// Pause the VM (including all devices). |
| 700 | pub async fn pause(&self) -> anyhow::Result<()> { |
| 701 | self.ttrpc |
| 702 | .call(diag_proto::UnderhillDiag::Pause, ()) |
| 703 | .await? |
| 704 | .map_err(grpc_status)?; |
| 705 | |
| 706 | Ok(()) |
| 707 | } |
| 708 | |
| 709 | /// Resume the VM. |
| 710 | pub async fn resume(&self) -> anyhow::Result<()> { |
| 711 | self.ttrpc |
| 712 | .call(diag_proto::UnderhillDiag::Resume, ()) |
| 713 | .await? |
| 714 | .map_err(grpc_status)?; |
| 715 | |
| 716 | Ok(()) |
| 717 | } |
| 718 | |
| 719 | /// Dumps the VM's VTL2 saved state. |
| 720 | pub async fn dump_saved_state(&self) -> anyhow::Result<Vec<u8>> { |
| 721 | let state = self |
| 722 | .ttrpc |
| 723 | .call(diag_proto::UnderhillDiag::DumpSavedState, ()) |
| 724 | .await? |
| 725 | .map_err(grpc_status)?; |
| 726 | |
| 727 | Ok(state.data) |
| 728 | } |
| 729 | } |
| 730 | |
| 731 | fn grpc_status(status: Status) -> anyhow::Error { |
| 732 | anyhow::anyhow!(status.message) |
| 733 | } |
| 734 | |
| 735 | /// A builder for launching a command in VTL2. |
| 736 | pub struct ExecBuilder<'a> { |
| 737 | client: &'a DiagClient, |
| 738 | with_stdin: bool, |
| 739 | with_stdout: bool, |
| 740 | with_stderr: bool, |
| 741 | request: ExecRequest, |
| 742 | } |
| 743 | |
| 744 | impl ExecBuilder<'_> { |
| 745 | /// Adds `args` to the argument list. |
| 746 | pub fn args<T: AsRef<str>>(&mut self, args: impl IntoIterator<Item = T>) -> &mut Self { |
| 747 | self.request |
| 748 | .args |
| 749 | .extend(args.into_iter().map(|s| s.as_ref().to_owned())); |
| 750 | self |
| 751 | } |
| 752 | |
| 753 | /// Sets whether the process is spawned with a TTY. |
| 754 | pub fn tty(&mut self, tty: bool) -> &mut Self { |
| 755 | self.request.tty = tty; |
| 756 | self |
| 757 | } |
| 758 | |
| 759 | /// Specifies whether a stdin socket should be opened. |
| 760 | pub fn stdin(&mut self, stdin: bool) -> &mut Self { |
| 761 | self.with_stdin = stdin; |
| 762 | self |
| 763 | } |
| 764 | |
| 765 | /// Specifies whether a stdout socket should be opened. |
| 766 | pub fn stdout(&mut self, stdout: bool) -> &mut Self { |
| 767 | self.with_stdout = stdout; |
| 768 | self |
| 769 | } |
| 770 | |
| 771 | /// Specifies whether a stderr socket should be opened. |
| 772 | pub fn stderr(&mut self, stderr: bool) -> &mut Self { |
| 773 | self.with_stderr = stderr; |
| 774 | self |
| 775 | } |
| 776 | |
| 777 | /// Specifies whether the processes's stdout and stderr should be combined |
| 778 | /// into a single stream (the stdout socket). |
| 779 | pub fn combine_stderr(&mut self, combine_stderr: bool) -> &mut Self { |
| 780 | self.request.combine_stderr = combine_stderr; |
| 781 | self |
| 782 | } |
| 783 | |
| 784 | /// Specifies whether the vsock sockets used for stdio should be passed |
| 785 | /// directly to the launched process instead of going through relays. |
| 786 | pub fn raw_socket_io(&mut self, raw_socket_io: bool) -> &mut Self { |
| 787 | self.request.raw_socket_io = raw_socket_io; |
| 788 | self |
| 789 | } |
| 790 | |
| 791 | /// Clears the default environment. |
| 792 | pub fn env_clear(&mut self) -> &mut Self { |
| 793 | self.request.clear_env = true; |
| 794 | self |
| 795 | } |
| 796 | |
| 797 | /// Removes an environment variable. |
| 798 | pub fn env_remove(&mut self, name: impl AsRef<str>) -> &mut Self { |
| 799 | self.request.env.push(diag_proto::EnvPair { |
| 800 | name: name.as_ref().to_owned(), |
| 801 | value: None, |
| 802 | }); |
| 803 | self |
| 804 | } |
| 805 | |
| 806 | /// Sets an environment variable. |
| 807 | pub fn env(&mut self, name: impl AsRef<str>, value: impl AsRef<str>) -> &mut Self { |
| 808 | self.request.env.push(diag_proto::EnvPair { |
| 809 | name: name.as_ref().to_owned(), |
| 810 | value: Some(value.as_ref().to_owned()), |
| 811 | }); |
| 812 | self |
| 813 | } |
| 814 | |
| 815 | /// Spawns the process. |
| 816 | pub async fn spawn(&self) -> anyhow::Result<Process> { |
| 817 | let mut request = self.request.clone(); |
| 818 | |
| 819 | let stdin = if self.with_stdin { |
| 820 | let (id, stdin) = self |
| 821 | .client |
| 822 | .connect_data() |
| 823 | .await |
| 824 | .context("failed to connect stdin")?; |
| 825 | request.stdin = id; |
| 826 | |
| 827 | Some(stdin.into_inner()) |
| 828 | } else { |
| 829 | None |
| 830 | }; |
| 831 | |
| 832 | let stdout = if self.with_stdout { |
| 833 | let (id, stdout) = self |
| 834 | .client |
| 835 | .connect_data() |
| 836 | .await |
| 837 | .context("failed to connect stdout")?; |
| 838 | request.stdout = id; |
| 839 | |
| 840 | Some(stdout.into_inner()) |
| 841 | } else { |
| 842 | None |
| 843 | }; |
| 844 | |
| 845 | let stderr = if self.with_stdout { |
| 846 | let (id, stderr) = self |
| 847 | .client |
| 848 | .connect_data() |
| 849 | .await |
| 850 | .context("failed to connect stderr")?; |
| 851 | request.stderr = id; |
| 852 | |
| 853 | Some(stderr.into_inner()) |
| 854 | } else { |
| 855 | None |
| 856 | }; |
| 857 | |
| 858 | let response = self |
| 859 | .client |
| 860 | .ttrpc |
| 861 | .call(diag_proto::UnderhillDiag::Exec, request) |
| 862 | .await? |
| 863 | .map_err(grpc_status)?; |
| 864 | |
| 865 | let wait = self.client.ttrpc.start_call( |
| 866 | diag_proto::UnderhillDiag::Wait, |
| 867 | WaitRequest { pid: response.pid }, |
| 868 | None, |
| 869 | ); |
| 870 | |
| 871 | Ok(Process { |
| 872 | stdin, |
| 873 | stdout, |
| 874 | stderr, |
| 875 | wait, |
| 876 | pid: response.pid, |
| 877 | }) |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | /// A process running in VTL2. |
| 882 | #[derive(Debug)] |
| 883 | pub struct Process { |
| 884 | /// The standard input stream. |
| 885 | pub stdin: Option<socket2::Socket>, |
| 886 | /// The standard output stream. |
| 887 | pub stdout: Option<socket2::Socket>, |
| 888 | /// The standard error stream. |
| 889 | pub stderr: Option<socket2::Socket>, |
| 890 | pid: i32, |
| 891 | wait: mesh::OneshotReceiver<Result<WaitResponse, Status>>, |
| 892 | } |
| 893 | |
| 894 | impl Process { |
| 895 | /// Returns the process ID. |
| 896 | pub fn id(&self) -> i32 { |
| 897 | self.pid |
| 898 | } |
| 899 | |
| 900 | /// Waits for the process to exit. |
| 901 | pub async fn wait(self) -> anyhow::Result<ExitStatus> { |
| 902 | let response = self |
| 903 | .wait |
| 904 | .await |
| 905 | .context("disconnected")? |
| 906 | .map_err(|err| anyhow::anyhow!("{}", err.message))?; |
| 907 | |
| 908 | Ok(ExitStatus { response }) |
| 909 | } |
| 910 | } |
| 911 | |
| 912 | /// Process exit status. |
| 913 | #[derive(Debug)] |
| 914 | pub struct ExitStatus { |
| 915 | response: WaitResponse, |
| 916 | } |
| 917 | |
| 918 | impl ExitStatus { |
| 919 | /// The exit code. |
| 920 | pub fn exit_code(&self) -> i32 { |
| 921 | self.response.exit_code |
| 922 | } |
| 923 | |
| 924 | /// Whether the process successfully terminated. |
| 925 | pub fn success(&self) -> bool { |
| 926 | self.response.exit_code == 0 |
| 927 | } |
| 928 | } |
| 929 | |