1 // Copyright © 2021 Intel Corporation 2 // 3 // SPDX-License-Identifier: Apache-2.0 4 // 5 6 #![allow(clippy::undocumented_unsafe_blocks)] 7 8 use once_cell::sync::Lazy; 9 use serde_json::Value; 10 use ssh2::Session; 11 use std::env; 12 use std::ffi::OsStr; 13 use std::fmt::Display; 14 use std::io; 15 use std::io::{Read, Write}; 16 use std::net::TcpListener; 17 use std::net::TcpStream; 18 use std::os::unix::fs::PermissionsExt; 19 use std::os::unix::io::{AsRawFd, FromRawFd}; 20 use std::path::Path; 21 use std::process::{Child, Command, ExitStatus, Output, Stdio}; 22 use std::str::FromStr; 23 use std::sync::Mutex; 24 use std::thread; 25 use std::time::Duration; 26 use std::{fmt, fs}; 27 use vmm_sys_util::tempdir::TempDir; 28 use wait_timeout::ChildExt; 29 30 #[derive(Debug)] 31 pub enum WaitTimeoutError { 32 Timedout, 33 ExitStatus, 34 General(std::io::Error), 35 } 36 37 #[derive(Debug)] 38 pub enum Error { 39 Parsing(std::num::ParseIntError), 40 SshCommand(SshCommandError), 41 WaitForBoot(WaitForBootError), 42 EthrLogFile(std::io::Error), 43 EthrLogParse, 44 FioOutputParse, 45 Iperf3Parse, 46 Spawn(std::io::Error), 47 WaitTimeout(WaitTimeoutError), 48 } 49 50 impl From<SshCommandError> for Error { 51 fn from(e: SshCommandError) -> Self { 52 Self::SshCommand(e) 53 } 54 } 55 56 pub struct GuestNetworkConfig { 57 pub guest_ip: String, 58 pub l2_guest_ip1: String, 59 pub l2_guest_ip2: String, 60 pub l2_guest_ip3: String, 61 pub host_ip: String, 62 pub guest_mac: String, 63 pub l2_guest_mac1: String, 64 pub l2_guest_mac2: String, 65 pub l2_guest_mac3: String, 66 pub tcp_listener_port: u16, 67 } 68 69 pub const DEFAULT_TCP_LISTENER_MESSAGE: &str = "booted"; 70 pub const DEFAULT_TCP_LISTENER_PORT: u16 = 8000; 71 pub const DEFAULT_TCP_LISTENER_TIMEOUT: i32 = 120; 72 73 #[derive(Debug)] 74 pub enum WaitForBootError { 75 EpollWait(std::io::Error), 76 Listen(std::io::Error), 77 EpollWaitTimeout, 78 WrongGuestAddr, 79 Accept(std::io::Error), 80 } 81 82 impl GuestNetworkConfig { 83 pub fn wait_vm_boot(&self, custom_timeout: Option<i32>) -> Result<(), WaitForBootError> { 84 let start = std::time::Instant::now(); 85 // The 'port' is unique per 'GUEST' and listening to wild-card ip avoids retrying on 'TcpListener::bind()' 86 let listen_addr = format!("0.0.0.0:{}", self.tcp_listener_port); 87 let expected_guest_addr = self.guest_ip.as_str(); 88 let mut s = String::new(); 89 let timeout = match custom_timeout { 90 Some(t) => t, 91 None => DEFAULT_TCP_LISTENER_TIMEOUT, 92 }; 93 94 let mut closure = || -> Result<(), WaitForBootError> { 95 let listener = 96 TcpListener::bind(listen_addr.as_str()).map_err(WaitForBootError::Listen)?; 97 listener 98 .set_nonblocking(true) 99 .expect("Cannot set non-blocking for tcp listener"); 100 101 // Reply on epoll w/ timeout to wait for guest connections faithfully 102 let epoll_fd = epoll::create(true).expect("Cannot create epoll fd"); 103 // Use 'File' to enforce closing on 'epoll_fd' 104 let _epoll_file = unsafe { fs::File::from_raw_fd(epoll_fd) }; 105 epoll::ctl( 106 epoll_fd, 107 epoll::ControlOptions::EPOLL_CTL_ADD, 108 listener.as_raw_fd(), 109 epoll::Event::new(epoll::Events::EPOLLIN, 0), 110 ) 111 .expect("Cannot add 'tcp_listener' event to epoll"); 112 let mut events = [epoll::Event::new(epoll::Events::empty(), 0); 1]; 113 loop { 114 let num_events = match epoll::wait(epoll_fd, timeout * 1000_i32, &mut events[..]) { 115 Ok(num_events) => Ok(num_events), 116 Err(e) => match e.raw_os_error() { 117 Some(libc::EAGAIN) | Some(libc::EINTR) => continue, 118 _ => Err(e), 119 }, 120 } 121 .map_err(WaitForBootError::EpollWait)?; 122 if num_events == 0 { 123 return Err(WaitForBootError::EpollWaitTimeout); 124 } 125 break; 126 } 127 128 match listener.accept() { 129 Ok((_, addr)) => { 130 // Make sure the connection is from the expected 'guest_addr' 131 if addr.ip() != std::net::IpAddr::from_str(expected_guest_addr).unwrap() { 132 s = format!( 133 "Expecting the guest ip '{}' while being connected with ip '{}'", 134 expected_guest_addr, 135 addr.ip() 136 ); 137 return Err(WaitForBootError::WrongGuestAddr); 138 } 139 140 Ok(()) 141 } 142 Err(e) => { 143 s = "TcpListener::accept() failed".to_string(); 144 Err(WaitForBootError::Accept(e)) 145 } 146 } 147 }; 148 149 match closure() { 150 Err(e) => { 151 let duration = start.elapsed(); 152 eprintln!( 153 "\n\n==== Start 'wait_vm_boot' (FAILED) ==== \ 154 \n\nduration =\"{duration:?}, timeout = {timeout}s\" \ 155 \nlisten_addr=\"{listen_addr}\" \ 156 \nexpected_guest_addr=\"{expected_guest_addr}\" \ 157 \nmessage=\"{s}\" \ 158 \nerror=\"{e:?}\" \ 159 \n\n==== End 'wait_vm_boot' outout ====\n\n" 160 ); 161 162 Err(e) 163 } 164 Ok(_) => Ok(()), 165 } 166 } 167 } 168 169 pub enum DiskType { 170 OperatingSystem, 171 CloudInit, 172 } 173 174 pub trait DiskConfig { 175 fn prepare_files(&mut self, tmp_dir: &TempDir, network: &GuestNetworkConfig); 176 fn prepare_cloudinit(&self, tmp_dir: &TempDir, network: &GuestNetworkConfig) -> String; 177 fn disk(&self, disk_type: DiskType) -> Option<String>; 178 } 179 180 #[derive(Clone)] 181 pub struct UbuntuDiskConfig { 182 osdisk_path: String, 183 cloudinit_path: String, 184 image_name: String, 185 } 186 187 impl UbuntuDiskConfig { 188 pub fn new(image_name: String) -> Self { 189 UbuntuDiskConfig { 190 image_name, 191 osdisk_path: String::new(), 192 cloudinit_path: String::new(), 193 } 194 } 195 } 196 197 pub struct WindowsDiskConfig { 198 image_name: String, 199 osdisk_path: String, 200 loopback_device: String, 201 windows_snapshot_cow: String, 202 windows_snapshot: String, 203 } 204 205 impl WindowsDiskConfig { 206 pub fn new(image_name: String) -> Self { 207 WindowsDiskConfig { 208 image_name, 209 osdisk_path: String::new(), 210 loopback_device: String::new(), 211 windows_snapshot_cow: String::new(), 212 windows_snapshot: String::new(), 213 } 214 } 215 } 216 217 impl Drop for WindowsDiskConfig { 218 fn drop(&mut self) { 219 // dmsetup remove windows-snapshot-1 220 std::process::Command::new("dmsetup") 221 .arg("remove") 222 .arg(self.windows_snapshot.as_str()) 223 .output() 224 .expect("Expect removing Windows snapshot with 'dmsetup' to succeed"); 225 226 // dmsetup remove windows-snapshot-cow-1 227 std::process::Command::new("dmsetup") 228 .arg("remove") 229 .arg(self.windows_snapshot_cow.as_str()) 230 .output() 231 .expect("Expect removing Windows snapshot CoW with 'dmsetup' to succeed"); 232 233 // losetup -d <loopback_device> 234 std::process::Command::new("losetup") 235 .args(["-d", self.loopback_device.as_str()]) 236 .output() 237 .expect("Expect removing loopback device to succeed"); 238 } 239 } 240 241 impl DiskConfig for UbuntuDiskConfig { 242 fn prepare_cloudinit(&self, tmp_dir: &TempDir, network: &GuestNetworkConfig) -> String { 243 let cloudinit_file_path = 244 String::from(tmp_dir.as_path().join("cloudinit").to_str().unwrap()); 245 246 let cloud_init_directory = tmp_dir.as_path().join("cloud-init").join("ubuntu"); 247 248 fs::create_dir_all(&cloud_init_directory) 249 .expect("Expect creating cloud-init directory to succeed"); 250 251 let source_file_dir = std::env::current_dir() 252 .unwrap() 253 .join("test_data") 254 .join("cloud-init") 255 .join("ubuntu") 256 .join("ci"); 257 258 ["meta-data"].iter().for_each(|x| { 259 rate_limited_copy(source_file_dir.join(x), cloud_init_directory.join(x)) 260 .expect("Expect copying cloud-init meta-data to succeed"); 261 }); 262 263 let mut user_data_string = String::new(); 264 fs::File::open(source_file_dir.join("user-data")) 265 .unwrap() 266 .read_to_string(&mut user_data_string) 267 .expect("Expected reading user-data file in to succeed"); 268 user_data_string = user_data_string.replace( 269 "@DEFAULT_TCP_LISTENER_MESSAGE", 270 DEFAULT_TCP_LISTENER_MESSAGE, 271 ); 272 user_data_string = user_data_string.replace("@HOST_IP", &network.host_ip); 273 user_data_string = 274 user_data_string.replace("@TCP_LISTENER_PORT", &network.tcp_listener_port.to_string()); 275 276 fs::File::create(cloud_init_directory.join("user-data")) 277 .unwrap() 278 .write_all(user_data_string.as_bytes()) 279 .expect("Expected writing out user-data to succeed"); 280 281 let mut network_config_string = String::new(); 282 283 fs::File::open(source_file_dir.join("network-config")) 284 .unwrap() 285 .read_to_string(&mut network_config_string) 286 .expect("Expected reading network-config file in to succeed"); 287 288 network_config_string = network_config_string.replace("192.168.2.1", &network.host_ip); 289 network_config_string = network_config_string.replace("192.168.2.2", &network.guest_ip); 290 network_config_string = network_config_string.replace("192.168.2.3", &network.l2_guest_ip1); 291 network_config_string = network_config_string.replace("192.168.2.4", &network.l2_guest_ip2); 292 network_config_string = network_config_string.replace("192.168.2.5", &network.l2_guest_ip3); 293 network_config_string = 294 network_config_string.replace("12:34:56:78:90:ab", &network.guest_mac); 295 network_config_string = 296 network_config_string.replace("de:ad:be:ef:12:34", &network.l2_guest_mac1); 297 network_config_string = 298 network_config_string.replace("de:ad:be:ef:34:56", &network.l2_guest_mac2); 299 network_config_string = 300 network_config_string.replace("de:ad:be:ef:56:78", &network.l2_guest_mac3); 301 302 fs::File::create(cloud_init_directory.join("network-config")) 303 .unwrap() 304 .write_all(network_config_string.as_bytes()) 305 .expect("Expected writing out network-config to succeed"); 306 307 std::process::Command::new("mkdosfs") 308 .args(["-n", "CIDATA"]) 309 .args(["-C", cloudinit_file_path.as_str()]) 310 .arg("8192") 311 .output() 312 .expect("Expect creating disk image to succeed"); 313 314 ["user-data", "meta-data", "network-config"] 315 .iter() 316 .for_each(|x| { 317 std::process::Command::new("mcopy") 318 .arg("-o") 319 .args(["-i", cloudinit_file_path.as_str()]) 320 .args(["-s", cloud_init_directory.join(x).to_str().unwrap(), "::"]) 321 .output() 322 .expect("Expect copying files to disk image to succeed"); 323 }); 324 325 cloudinit_file_path 326 } 327 328 fn prepare_files(&mut self, tmp_dir: &TempDir, network: &GuestNetworkConfig) { 329 let mut workload_path = dirs::home_dir().unwrap(); 330 workload_path.push("workloads"); 331 332 let mut osdisk_base_path = workload_path; 333 osdisk_base_path.push(&self.image_name); 334 335 let osdisk_path = String::from(tmp_dir.as_path().join("osdisk.img").to_str().unwrap()); 336 let cloudinit_path = self.prepare_cloudinit(tmp_dir, network); 337 338 rate_limited_copy(osdisk_base_path, &osdisk_path) 339 .expect("copying of OS source disk image failed"); 340 341 self.cloudinit_path = cloudinit_path; 342 self.osdisk_path = osdisk_path; 343 } 344 345 fn disk(&self, disk_type: DiskType) -> Option<String> { 346 match disk_type { 347 DiskType::OperatingSystem => Some(self.osdisk_path.clone()), 348 DiskType::CloudInit => Some(self.cloudinit_path.clone()), 349 } 350 } 351 } 352 353 impl DiskConfig for WindowsDiskConfig { 354 fn prepare_cloudinit(&self, _tmp_dir: &TempDir, _network: &GuestNetworkConfig) -> String { 355 String::new() 356 } 357 358 fn prepare_files(&mut self, tmp_dir: &TempDir, _network: &GuestNetworkConfig) { 359 let mut workload_path = dirs::home_dir().unwrap(); 360 workload_path.push("workloads"); 361 362 let mut osdisk_path = workload_path; 363 osdisk_path.push(&self.image_name); 364 365 let osdisk_blk_size = fs::metadata(osdisk_path) 366 .expect("Expect retrieving Windows image metadata") 367 .len() 368 >> 9; 369 370 let snapshot_cow_path = 371 String::from(tmp_dir.as_path().join("snapshot_cow").to_str().unwrap()); 372 373 // Create and truncate CoW file for device mapper 374 let cow_file_size: u64 = 1 << 30; 375 let cow_file_blk_size = cow_file_size >> 9; 376 let cow_file = std::fs::File::create(snapshot_cow_path.as_str()) 377 .expect("Expect creating CoW image to succeed"); 378 cow_file 379 .set_len(cow_file_size) 380 .expect("Expect truncating CoW image to succeed"); 381 382 // losetup --find --show /tmp/snapshot_cow 383 let loopback_device = std::process::Command::new("losetup") 384 .arg("--find") 385 .arg("--show") 386 .arg(snapshot_cow_path.as_str()) 387 .output() 388 .expect("Expect creating loopback device from snapshot CoW image to succeed"); 389 390 self.loopback_device = String::from_utf8_lossy(&loopback_device.stdout) 391 .trim() 392 .to_string(); 393 394 let random_extension = tmp_dir.as_path().file_name().unwrap(); 395 let windows_snapshot_cow = format!( 396 "windows-snapshot-cow-{}", 397 random_extension.to_str().unwrap() 398 ); 399 400 // dmsetup create windows-snapshot-cow-1 --table '0 2097152 linear /dev/loop1 0' 401 std::process::Command::new("dmsetup") 402 .arg("create") 403 .arg(windows_snapshot_cow.as_str()) 404 .args([ 405 "--table", 406 format!("0 {} linear {} 0", cow_file_blk_size, self.loopback_device).as_str(), 407 ]) 408 .output() 409 .expect("Expect creating Windows snapshot CoW with 'dmsetup' to succeed"); 410 411 let windows_snapshot = format!("windows-snapshot-{}", random_extension.to_str().unwrap()); 412 413 // dmsetup mknodes 414 std::process::Command::new("dmsetup") 415 .arg("mknodes") 416 .output() 417 .expect("Expect device mapper nodes to be ready"); 418 419 // dmsetup create windows-snapshot-1 --table '0 41943040 snapshot /dev/mapper/windows-base /dev/mapper/windows-snapshot-cow-1 P 8' 420 std::process::Command::new("dmsetup") 421 .arg("create") 422 .arg(windows_snapshot.as_str()) 423 .args([ 424 "--table", 425 format!( 426 "0 {} snapshot /dev/mapper/windows-base /dev/mapper/{} P 8", 427 osdisk_blk_size, 428 windows_snapshot_cow.as_str() 429 ) 430 .as_str(), 431 ]) 432 .output() 433 .expect("Expect creating Windows snapshot with 'dmsetup' to succeed"); 434 435 // dmsetup mknodes 436 std::process::Command::new("dmsetup") 437 .arg("mknodes") 438 .output() 439 .expect("Expect device mapper nodes to be ready"); 440 441 self.osdisk_path = format!("/dev/mapper/{windows_snapshot}"); 442 self.windows_snapshot_cow = windows_snapshot_cow; 443 self.windows_snapshot = windows_snapshot; 444 } 445 446 fn disk(&self, disk_type: DiskType) -> Option<String> { 447 match disk_type { 448 DiskType::OperatingSystem => Some(self.osdisk_path.clone()), 449 DiskType::CloudInit => None, 450 } 451 } 452 } 453 454 pub fn rate_limited_copy<P: AsRef<Path>, Q: AsRef<Path>>(from: P, to: Q) -> io::Result<u64> { 455 for i in 0..10 { 456 let free_bytes = unsafe { 457 let mut stats = std::mem::MaybeUninit::zeroed(); 458 let fs_name = std::ffi::CString::new("/tmp").unwrap(); 459 libc::statvfs(fs_name.as_ptr(), stats.as_mut_ptr()); 460 461 let free_blocks = stats.assume_init().f_bfree; 462 let block_size = stats.assume_init().f_bsize; 463 464 free_blocks * block_size 465 }; 466 467 // Make sure there is at least 6 GiB of space 468 if free_bytes < 6 << 30 { 469 eprintln!("Not enough space on disk ({free_bytes}). Attempt {i} of 10. Sleeping."); 470 thread::sleep(std::time::Duration::new(60, 0)); 471 continue; 472 } 473 474 match fs::copy(&from, &to) { 475 Err(e) => { 476 if let Some(errno) = e.raw_os_error() { 477 if errno == libc::ENOSPC { 478 eprintln!("Copy returned ENOSPC. Attempt {i} of 10. Sleeping."); 479 thread::sleep(std::time::Duration::new(60, 0)); 480 continue; 481 } 482 } 483 return Err(e); 484 } 485 Ok(i) => return Ok(i), 486 } 487 } 488 Err(io::Error::last_os_error()) 489 } 490 491 pub fn handle_child_output( 492 r: Result<(), std::boxed::Box<dyn std::any::Any + std::marker::Send>>, 493 output: &std::process::Output, 494 ) { 495 use std::os::unix::process::ExitStatusExt; 496 if r.is_ok() && output.status.success() { 497 return; 498 } 499 500 match output.status.code() { 501 None => { 502 // Don't treat child.kill() as a problem 503 if output.status.signal() == Some(9) && r.is_ok() { 504 return; 505 } 506 507 eprintln!( 508 "==== child killed by signal: {} ====", 509 output.status.signal().unwrap() 510 ); 511 } 512 Some(code) => { 513 eprintln!("\n\n==== child exit code: {code} ===="); 514 } 515 } 516 517 eprintln!( 518 "\n\n==== Start child stdout ====\n\n{}\n\n==== End child stdout ====", 519 String::from_utf8_lossy(&output.stdout) 520 ); 521 eprintln!( 522 "\n\n==== Start child stderr ====\n\n{}\n\n==== End child stderr ====", 523 String::from_utf8_lossy(&output.stderr) 524 ); 525 526 panic!("Test failed") 527 } 528 529 #[derive(Debug)] 530 pub struct PasswordAuth { 531 pub username: String, 532 pub password: String, 533 } 534 535 pub const DEFAULT_SSH_RETRIES: u8 = 6; 536 pub const DEFAULT_SSH_TIMEOUT: u8 = 10; 537 538 #[derive(Debug)] 539 pub enum SshCommandError { 540 Connection(std::io::Error), 541 Handshake(ssh2::Error), 542 Authentication(ssh2::Error), 543 ChannelSession(ssh2::Error), 544 Command(ssh2::Error), 545 ExitStatus(ssh2::Error), 546 NonZeroExitStatus(i32), 547 FileRead(std::io::Error), 548 FileMetadata(std::io::Error), 549 ScpSend(ssh2::Error), 550 WriteAll(std::io::Error), 551 SendEof(ssh2::Error), 552 WaitEof(ssh2::Error), 553 } 554 555 fn scp_to_guest_with_auth( 556 path: &Path, 557 remote_path: &Path, 558 auth: &PasswordAuth, 559 ip: &str, 560 retries: u8, 561 timeout: u8, 562 ) -> Result<(), SshCommandError> { 563 let mut counter = 0; 564 loop { 565 let closure = || -> Result<(), SshCommandError> { 566 let tcp = 567 TcpStream::connect(format!("{ip}:22")).map_err(SshCommandError::Connection)?; 568 let mut sess = Session::new().unwrap(); 569 sess.set_tcp_stream(tcp); 570 sess.handshake().map_err(SshCommandError::Handshake)?; 571 572 sess.userauth_password(&auth.username, &auth.password) 573 .map_err(SshCommandError::Authentication)?; 574 assert!(sess.authenticated()); 575 576 let content = fs::read(path).map_err(SshCommandError::FileRead)?; 577 let mode = fs::metadata(path) 578 .map_err(SshCommandError::FileMetadata)? 579 .permissions() 580 .mode() 581 & 0o777; 582 583 let mut channel = sess 584 .scp_send(remote_path, mode as i32, content.len() as u64, None) 585 .map_err(SshCommandError::ScpSend)?; 586 channel 587 .write_all(&content) 588 .map_err(SshCommandError::WriteAll)?; 589 channel.send_eof().map_err(SshCommandError::SendEof)?; 590 channel.wait_eof().map_err(SshCommandError::WaitEof)?; 591 592 // Intentionally ignore these results here as their failure 593 // does not precipitate a repeat 594 let _ = channel.close(); 595 let _ = channel.wait_close(); 596 597 Ok(()) 598 }; 599 600 match closure() { 601 Ok(_) => break, 602 Err(e) => { 603 counter += 1; 604 if counter >= retries { 605 eprintln!( 606 "\n\n==== Start scp command output (FAILED) ====\n\n\ 607 path =\"{path:?}\"\n\ 608 remote_path =\"{remote_path:?}\"\n\ 609 auth=\"{auth:#?}\"\n\ 610 ip=\"{ip}\"\n\ 611 error=\"{e:?}\"\n\ 612 \n==== End scp command outout ====\n\n" 613 ); 614 615 return Err(e); 616 } 617 } 618 }; 619 thread::sleep(std::time::Duration::new((timeout * counter).into(), 0)); 620 } 621 Ok(()) 622 } 623 624 pub fn scp_to_guest( 625 path: &Path, 626 remote_path: &Path, 627 ip: &str, 628 retries: u8, 629 timeout: u8, 630 ) -> Result<(), SshCommandError> { 631 scp_to_guest_with_auth( 632 path, 633 remote_path, 634 &PasswordAuth { 635 username: String::from("cloud"), 636 password: String::from("cloud123"), 637 }, 638 ip, 639 retries, 640 timeout, 641 ) 642 } 643 644 pub fn ssh_command_ip_with_auth( 645 command: &str, 646 auth: &PasswordAuth, 647 ip: &str, 648 retries: u8, 649 timeout: u8, 650 ) -> Result<String, SshCommandError> { 651 let mut s = String::new(); 652 653 let mut counter = 0; 654 loop { 655 let mut closure = || -> Result<(), SshCommandError> { 656 let tcp = 657 TcpStream::connect(format!("{ip}:22")).map_err(SshCommandError::Connection)?; 658 let mut sess = Session::new().unwrap(); 659 sess.set_tcp_stream(tcp); 660 sess.handshake().map_err(SshCommandError::Handshake)?; 661 662 sess.userauth_password(&auth.username, &auth.password) 663 .map_err(SshCommandError::Authentication)?; 664 assert!(sess.authenticated()); 665 666 let mut channel = sess 667 .channel_session() 668 .map_err(SshCommandError::ChannelSession)?; 669 channel.exec(command).map_err(SshCommandError::Command)?; 670 671 // Intentionally ignore these results here as their failure 672 // does not precipitate a repeat 673 let _ = channel.read_to_string(&mut s); 674 let _ = channel.close(); 675 let _ = channel.wait_close(); 676 677 let status = channel.exit_status().map_err(SshCommandError::ExitStatus)?; 678 679 if status != 0 { 680 Err(SshCommandError::NonZeroExitStatus(status)) 681 } else { 682 Ok(()) 683 } 684 }; 685 686 match closure() { 687 Ok(_) => break, 688 Err(e) => { 689 counter += 1; 690 if counter >= retries { 691 eprintln!( 692 "\n\n==== Start ssh command output (FAILED) ====\n\n\ 693 command=\"{command}\"\n\ 694 auth=\"{auth:#?}\"\n\ 695 ip=\"{ip}\"\n\ 696 output=\"{s}\"\n\ 697 error=\"{e:?}\"\n\ 698 \n==== End ssh command outout ====\n\n" 699 ); 700 701 return Err(e); 702 } 703 } 704 }; 705 thread::sleep(std::time::Duration::new((timeout * counter).into(), 0)); 706 } 707 Ok(s) 708 } 709 710 pub fn ssh_command_ip( 711 command: &str, 712 ip: &str, 713 retries: u8, 714 timeout: u8, 715 ) -> Result<String, SshCommandError> { 716 ssh_command_ip_with_auth( 717 command, 718 &PasswordAuth { 719 username: String::from("cloud"), 720 password: String::from("cloud123"), 721 }, 722 ip, 723 retries, 724 timeout, 725 ) 726 } 727 728 pub fn exec_host_command_with_retries(command: &str, retries: u32, interval: Duration) -> bool { 729 for _ in 0..retries { 730 let s = exec_host_command_output(command).status; 731 if !s.success() { 732 eprintln!("\n\n==== retrying in {:?} ===\n\n", interval); 733 thread::sleep(interval); 734 } else { 735 return true; 736 } 737 } 738 739 false 740 } 741 742 pub fn exec_host_command_status(command: &str) -> ExitStatus { 743 exec_host_command_output(command).status 744 } 745 746 pub fn exec_host_command_output(command: &str) -> Output { 747 let output = std::process::Command::new("bash") 748 .args(["-c", command]) 749 .output() 750 .unwrap_or_else(|e| panic!("Expected '{command}' to run. Error: {:?}", e)); 751 752 if !output.status.success() { 753 let stdout = String::from_utf8_lossy(&output.stdout); 754 let stderr = String::from_utf8_lossy(&output.stderr); 755 eprintln!( 756 "\n\n==== Start 'exec_host_command' failed ==== \ 757 \n\n---stdout---\n{stdout}\n---stderr---{stderr} \ 758 \n\n==== End 'exec_host_command' failed ====", 759 ); 760 } 761 762 output 763 } 764 765 pub fn check_lines_count(input: &str, line_count: usize) -> bool { 766 if input.lines().count() == line_count { 767 true 768 } else { 769 eprintln!( 770 "\n\n==== Start 'check_lines_count' failed ==== \ 771 \n\ninput = {input}\nline_count = {line_count} \ 772 \n\n==== End 'check_lines_count' failed ====", 773 ); 774 775 false 776 } 777 } 778 779 pub fn check_matched_lines_count(input: &str, keywords: Vec<&str>, line_count: usize) -> bool { 780 let mut matches = String::new(); 781 for line in input.lines() { 782 if keywords.iter().all(|k| line.contains(k)) { 783 matches += line; 784 } 785 } 786 787 if matches.lines().count() == line_count { 788 true 789 } else { 790 eprintln!( 791 "\n\n==== Start 'check_matched_lines_count' failed ==== \ 792 \nkeywords = {keywords:?}, line_count = {line_count} \ 793 \n\ninput = {input} matches = {matches} \ 794 \n\n==== End 'check_matched_lines_count' failed ====", 795 ); 796 797 false 798 } 799 } 800 801 pub const PIPE_SIZE: i32 = 32 << 20; 802 803 static NEXT_VM_ID: Lazy<Mutex<u8>> = Lazy::new(|| Mutex::new(1)); 804 805 pub struct Guest { 806 pub tmp_dir: TempDir, 807 pub disk_config: Box<dyn DiskConfig>, 808 pub network: GuestNetworkConfig, 809 } 810 811 // Safe to implement as we know we have no interior mutability 812 impl std::panic::RefUnwindSafe for Guest {} 813 814 impl Guest { 815 pub fn new_from_ip_range(mut disk_config: Box<dyn DiskConfig>, class: &str, id: u8) -> Self { 816 let tmp_dir = TempDir::new_with_prefix("/tmp/ch").unwrap(); 817 818 let network = GuestNetworkConfig { 819 guest_ip: format!("{class}.{id}.2"), 820 l2_guest_ip1: format!("{class}.{id}.3"), 821 l2_guest_ip2: format!("{class}.{id}.4"), 822 l2_guest_ip3: format!("{class}.{id}.5"), 823 host_ip: format!("{class}.{id}.1"), 824 guest_mac: format!("12:34:56:78:90:{id:02x}"), 825 l2_guest_mac1: format!("de:ad:be:ef:12:{id:02x}"), 826 l2_guest_mac2: format!("de:ad:be:ef:34:{id:02x}"), 827 l2_guest_mac3: format!("de:ad:be:ef:56:{id:02x}"), 828 tcp_listener_port: DEFAULT_TCP_LISTENER_PORT + id as u16, 829 }; 830 831 disk_config.prepare_files(&tmp_dir, &network); 832 833 Guest { 834 tmp_dir, 835 disk_config, 836 network, 837 } 838 } 839 840 pub fn new(disk_config: Box<dyn DiskConfig>) -> Self { 841 let mut guard = NEXT_VM_ID.lock().unwrap(); 842 let id = *guard; 843 *guard = id + 1; 844 845 Self::new_from_ip_range(disk_config, "192.168", id) 846 } 847 848 pub fn default_net_string(&self) -> String { 849 format!( 850 "tap=,mac={},ip={},mask=255.255.255.0", 851 self.network.guest_mac, self.network.host_ip 852 ) 853 } 854 855 pub fn default_net_string_w_iommu(&self) -> String { 856 format!( 857 "tap=,mac={},ip={},mask=255.255.255.0,iommu=on", 858 self.network.guest_mac, self.network.host_ip 859 ) 860 } 861 862 pub fn default_net_string_w_mtu(&self, mtu: u16) -> String { 863 format!( 864 "tap=,mac={},ip={},mask=255.255.255.0,mtu={}", 865 self.network.guest_mac, self.network.host_ip, mtu 866 ) 867 } 868 869 pub fn ssh_command(&self, command: &str) -> Result<String, SshCommandError> { 870 ssh_command_ip( 871 command, 872 &self.network.guest_ip, 873 DEFAULT_SSH_RETRIES, 874 DEFAULT_SSH_TIMEOUT, 875 ) 876 } 877 878 #[cfg(target_arch = "x86_64")] 879 pub fn ssh_command_l1(&self, command: &str) -> Result<String, SshCommandError> { 880 ssh_command_ip( 881 command, 882 &self.network.guest_ip, 883 DEFAULT_SSH_RETRIES, 884 DEFAULT_SSH_TIMEOUT, 885 ) 886 } 887 888 #[cfg(target_arch = "x86_64")] 889 pub fn ssh_command_l2_1(&self, command: &str) -> Result<String, SshCommandError> { 890 ssh_command_ip( 891 command, 892 &self.network.l2_guest_ip1, 893 DEFAULT_SSH_RETRIES, 894 DEFAULT_SSH_TIMEOUT, 895 ) 896 } 897 898 #[cfg(target_arch = "x86_64")] 899 pub fn ssh_command_l2_2(&self, command: &str) -> Result<String, SshCommandError> { 900 ssh_command_ip( 901 command, 902 &self.network.l2_guest_ip2, 903 DEFAULT_SSH_RETRIES, 904 DEFAULT_SSH_TIMEOUT, 905 ) 906 } 907 908 #[cfg(target_arch = "x86_64")] 909 pub fn ssh_command_l2_3(&self, command: &str) -> Result<String, SshCommandError> { 910 ssh_command_ip( 911 command, 912 &self.network.l2_guest_ip3, 913 DEFAULT_SSH_RETRIES, 914 DEFAULT_SSH_TIMEOUT, 915 ) 916 } 917 918 pub fn api_create_body(&self, cpu_count: u8, kernel_path: &str, kernel_cmd: &str) -> String { 919 format! {"{{\"cpus\":{{\"boot_vcpus\":{},\"max_vcpus\":{}}},\"payload\":{{\"kernel\":\"{}\",\"cmdline\": \"{}\"}},\"net\":[{{\"ip\":\"{}\", \"mask\":\"255.255.255.0\", \"mac\":\"{}\"}}], \"disks\":[{{\"path\":\"{}\"}}, {{\"path\":\"{}\"}}]}}", 920 cpu_count, 921 cpu_count, 922 kernel_path, 923 kernel_cmd, 924 self.network.host_ip, 925 self.network.guest_mac, 926 self.disk_config.disk(DiskType::OperatingSystem).unwrap().as_str(), 927 self.disk_config.disk(DiskType::CloudInit).unwrap().as_str(), 928 } 929 } 930 931 pub fn get_cpu_count(&self) -> Result<u32, Error> { 932 self.ssh_command("grep -c processor /proc/cpuinfo")? 933 .trim() 934 .parse() 935 .map_err(Error::Parsing) 936 } 937 938 pub fn get_total_memory(&self) -> Result<u32, Error> { 939 self.ssh_command("grep MemTotal /proc/meminfo | grep -o \"[0-9]*\"")? 940 .trim() 941 .parse() 942 .map_err(Error::Parsing) 943 } 944 945 #[cfg(target_arch = "x86_64")] 946 pub fn get_total_memory_l2(&self) -> Result<u32, Error> { 947 self.ssh_command_l2_1("grep MemTotal /proc/meminfo | grep -o \"[0-9]*\"")? 948 .trim() 949 .parse() 950 .map_err(Error::Parsing) 951 } 952 953 pub fn get_numa_node_memory(&self, node_id: usize) -> Result<u32, Error> { 954 self.ssh_command( 955 format!( 956 "grep MemTotal /sys/devices/system/node/node{node_id}/meminfo \ 957 | cut -d \":\" -f 2 | grep -o \"[0-9]*\"" 958 ) 959 .as_str(), 960 )? 961 .trim() 962 .parse() 963 .map_err(Error::Parsing) 964 } 965 966 pub fn wait_vm_boot(&self, custom_timeout: Option<i32>) -> Result<(), Error> { 967 self.network 968 .wait_vm_boot(custom_timeout) 969 .map_err(Error::WaitForBoot) 970 } 971 972 pub fn check_numa_node_cpus(&self, node_id: usize, cpus: Vec<usize>) -> Result<(), Error> { 973 for cpu in cpus.iter() { 974 let cmd = format!("[ -d \"/sys/devices/system/node/node{node_id}/cpu{cpu}\" ]"); 975 self.ssh_command(cmd.as_str())?; 976 } 977 978 Ok(()) 979 } 980 981 pub fn check_numa_node_distances( 982 &self, 983 node_id: usize, 984 distances: &str, 985 ) -> Result<bool, Error> { 986 let cmd = format!("cat /sys/devices/system/node/node{node_id}/distance"); 987 if self.ssh_command(cmd.as_str())?.trim() == distances { 988 Ok(true) 989 } else { 990 Ok(false) 991 } 992 } 993 994 pub fn check_numa_common( 995 &self, 996 mem_ref: Option<&[u32]>, 997 node_ref: Option<&[Vec<usize>]>, 998 distance_ref: Option<&[&str]>, 999 ) { 1000 if let Some(mem_ref) = mem_ref { 1001 // Check each NUMA node has been assigned the right amount of 1002 // memory. 1003 for (i, &m) in mem_ref.iter().enumerate() { 1004 assert!(self.get_numa_node_memory(i).unwrap_or_default() > m); 1005 } 1006 } 1007 1008 if let Some(node_ref) = node_ref { 1009 // Check each NUMA node has been assigned the right CPUs set. 1010 for (i, n) in node_ref.iter().enumerate() { 1011 self.check_numa_node_cpus(i, n.clone()).unwrap(); 1012 } 1013 } 1014 1015 if let Some(distance_ref) = distance_ref { 1016 // Check each NUMA node has been assigned the right distances. 1017 for (i, &d) in distance_ref.iter().enumerate() { 1018 assert!(self.check_numa_node_distances(i, d).unwrap()); 1019 } 1020 } 1021 } 1022 1023 #[cfg(target_arch = "x86_64")] 1024 pub fn check_sgx_support(&self) -> Result<(), Error> { 1025 self.ssh_command( 1026 "cpuid -l 0x7 -s 0 | tr -s [:space:] | grep -q 'SGX: \ 1027 Software Guard Extensions supported = true'", 1028 )?; 1029 self.ssh_command( 1030 "cpuid -l 0x7 -s 0 | tr -s [:space:] | grep -q 'SGX_LC: \ 1031 SGX launch config supported = true'", 1032 )?; 1033 self.ssh_command( 1034 "cpuid -l 0x12 -s 0 | tr -s [:space:] | grep -q 'SGX1 \ 1035 supported = true'", 1036 )?; 1037 1038 Ok(()) 1039 } 1040 1041 pub fn get_pci_bridge_class(&self) -> Result<String, Error> { 1042 Ok(self 1043 .ssh_command("cat /sys/bus/pci/devices/0000:00:00.0/class")? 1044 .trim() 1045 .to_string()) 1046 } 1047 1048 pub fn get_pci_device_ids(&self) -> Result<String, Error> { 1049 Ok(self 1050 .ssh_command("cat /sys/bus/pci/devices/*/device")? 1051 .trim() 1052 .to_string()) 1053 } 1054 1055 pub fn get_pci_vendor_ids(&self) -> Result<String, Error> { 1056 Ok(self 1057 .ssh_command("cat /sys/bus/pci/devices/*/vendor")? 1058 .trim() 1059 .to_string()) 1060 } 1061 1062 pub fn does_device_vendor_pair_match( 1063 &self, 1064 device_id: &str, 1065 vendor_id: &str, 1066 ) -> Result<bool, Error> { 1067 // We are checking if console device's device id and vendor id pair matches 1068 let devices = self.get_pci_device_ids()?; 1069 let devices: Vec<&str> = devices.split('\n').collect(); 1070 let vendors = self.get_pci_vendor_ids()?; 1071 let vendors: Vec<&str> = vendors.split('\n').collect(); 1072 1073 for (index, d_id) in devices.iter().enumerate() { 1074 if *d_id == device_id { 1075 if let Some(v_id) = vendors.get(index) { 1076 if *v_id == vendor_id { 1077 return Ok(true); 1078 } 1079 } 1080 } 1081 } 1082 1083 Ok(false) 1084 } 1085 1086 pub fn check_vsock(&self, socket: &str) { 1087 // Listen from guest on vsock CID=3 PORT=16 1088 // SOCKET-LISTEN:<domain>:<protocol>:<local-address> 1089 let guest_ip = self.network.guest_ip.clone(); 1090 let listen_socat = thread::spawn(move || { 1091 ssh_command_ip("sudo socat - SOCKET-LISTEN:40:0:x00x00x10x00x00x00x03x00x00x00x00x00x00x00 > vsock_log", &guest_ip, DEFAULT_SSH_RETRIES, DEFAULT_SSH_TIMEOUT).unwrap(); 1092 }); 1093 1094 // Make sure socat is listening, which might take a few second on slow systems 1095 thread::sleep(std::time::Duration::new(10, 0)); 1096 1097 // Write something to vsock from the host 1098 assert!(exec_host_command_status(&format!( 1099 "echo -e \"CONNECT 16\\nHelloWorld!\" | socat - UNIX-CONNECT:{socket}" 1100 )) 1101 .success()); 1102 1103 // Wait for the thread to terminate. 1104 listen_socat.join().unwrap(); 1105 1106 assert_eq!( 1107 self.ssh_command("cat vsock_log").unwrap().trim(), 1108 "HelloWorld!" 1109 ); 1110 } 1111 1112 #[cfg(target_arch = "x86_64")] 1113 pub fn check_nvidia_gpu(&self) { 1114 assert!(self.ssh_command("nvidia-smi").unwrap().contains("Tesla T4")); 1115 } 1116 1117 pub fn reboot_linux(&self, current_reboot_count: u32, custom_timeout: Option<i32>) { 1118 let list_boots_cmd = "sudo last | grep -c reboot"; 1119 let boot_count = self 1120 .ssh_command(list_boots_cmd) 1121 .unwrap() 1122 .trim() 1123 .parse::<u32>() 1124 .unwrap_or_default(); 1125 1126 assert_eq!(boot_count, current_reboot_count + 1); 1127 self.ssh_command("sudo reboot").unwrap(); 1128 1129 self.wait_vm_boot(custom_timeout).unwrap(); 1130 let boot_count = self 1131 .ssh_command(list_boots_cmd) 1132 .unwrap() 1133 .trim() 1134 .parse::<u32>() 1135 .unwrap_or_default(); 1136 assert_eq!(boot_count, current_reboot_count + 2); 1137 } 1138 1139 pub fn enable_memory_hotplug(&self) { 1140 self.ssh_command("echo online | sudo tee /sys/devices/system/memory/auto_online_blocks") 1141 .unwrap(); 1142 } 1143 1144 pub fn check_devices_common( 1145 &self, 1146 socket: Option<&String>, 1147 console_text: Option<&String>, 1148 pmem_path: Option<&String>, 1149 ) { 1150 // Check block devices are readable 1151 self.ssh_command("sudo dd if=/dev/vda of=/dev/null bs=1M iflag=direct count=1024") 1152 .unwrap(); 1153 self.ssh_command("sudo dd if=/dev/vdb of=/dev/null bs=1M iflag=direct count=8") 1154 .unwrap(); 1155 // Check if the rng device is readable 1156 self.ssh_command("sudo head -c 1000 /dev/hwrng > /dev/null") 1157 .unwrap(); 1158 // Check vsock 1159 if let Some(socket) = socket { 1160 self.check_vsock(socket.as_str()); 1161 } 1162 // Check if the console is usable 1163 if let Some(console_text) = console_text { 1164 let console_cmd = format!("echo {console_text} | sudo tee /dev/hvc0"); 1165 self.ssh_command(&console_cmd).unwrap(); 1166 } 1167 // The net device is 'automatically' exercised through the above 'ssh' commands 1168 1169 // Check if the pmem device is usable 1170 if let Some(pmem_path) = pmem_path { 1171 assert_eq!( 1172 self.ssh_command(&format!("ls {pmem_path}")).unwrap().trim(), 1173 pmem_path 1174 ); 1175 assert_eq!( 1176 self.ssh_command(&format!("sudo mount {pmem_path} /mnt")) 1177 .unwrap(), 1178 "" 1179 ); 1180 assert_eq!(self.ssh_command("ls /mnt").unwrap(), "lost+found\n"); 1181 self.ssh_command("echo test123 | sudo tee /mnt/test") 1182 .unwrap(); 1183 assert_eq!(self.ssh_command("sudo umount /mnt").unwrap(), ""); 1184 assert_eq!(self.ssh_command("ls /mnt").unwrap(), ""); 1185 1186 assert_eq!( 1187 self.ssh_command(&format!("sudo mount {pmem_path} /mnt")) 1188 .unwrap(), 1189 "" 1190 ); 1191 assert_eq!( 1192 self.ssh_command("sudo cat /mnt/test || true") 1193 .unwrap() 1194 .trim(), 1195 "test123" 1196 ); 1197 self.ssh_command("sudo rm /mnt/test").unwrap(); 1198 assert_eq!(self.ssh_command("sudo umount /mnt").unwrap(), ""); 1199 } 1200 } 1201 } 1202 1203 pub enum VerbosityLevel { 1204 Warn, 1205 Info, 1206 Debug, 1207 } 1208 1209 impl Default for VerbosityLevel { 1210 fn default() -> Self { 1211 Self::Warn 1212 } 1213 } 1214 1215 impl Display for VerbosityLevel { 1216 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { 1217 use VerbosityLevel::*; 1218 match self { 1219 Warn => (), 1220 Info => write!(f, "-v")?, 1221 Debug => write!(f, "-vv")?, 1222 } 1223 Ok(()) 1224 } 1225 } 1226 1227 pub struct GuestCommand<'a> { 1228 command: Command, 1229 guest: &'a Guest, 1230 capture_output: bool, 1231 print_cmd: bool, 1232 verbosity: VerbosityLevel, 1233 } 1234 1235 impl<'a> GuestCommand<'a> { 1236 pub fn new(guest: &'a Guest) -> Self { 1237 Self::new_with_binary_path(guest, &clh_command("cloud-hypervisor")) 1238 } 1239 1240 pub fn new_with_binary_path(guest: &'a Guest, binary_path: &str) -> Self { 1241 Self { 1242 command: Command::new(binary_path), 1243 guest, 1244 capture_output: false, 1245 print_cmd: true, 1246 verbosity: VerbosityLevel::Info, 1247 } 1248 } 1249 1250 pub fn verbosity(&mut self, verbosity: VerbosityLevel) -> &mut Self { 1251 self.verbosity = verbosity; 1252 self 1253 } 1254 1255 pub fn capture_output(&mut self) -> &mut Self { 1256 self.capture_output = true; 1257 self 1258 } 1259 1260 pub fn set_print_cmd(&mut self, print_cmd: bool) -> &mut Self { 1261 self.print_cmd = print_cmd; 1262 self 1263 } 1264 1265 pub fn spawn(&mut self) -> io::Result<Child> { 1266 use VerbosityLevel::*; 1267 match &self.verbosity { 1268 Warn => {} 1269 Info => { 1270 self.command.arg("-v"); 1271 } 1272 Debug => { 1273 self.command.args(["-vv"]); 1274 } 1275 }; 1276 1277 if self.print_cmd { 1278 println!( 1279 "\n\n==== Start cloud-hypervisor command-line ====\n\n\ 1280 {:?}\n\ 1281 \n==== End cloud-hypervisor command-line ====\n\n", 1282 self.command 1283 ); 1284 } 1285 1286 if self.capture_output { 1287 let child = self 1288 .command 1289 .stderr(Stdio::piped()) 1290 .stdout(Stdio::piped()) 1291 .spawn() 1292 .unwrap(); 1293 1294 let fd = child.stdout.as_ref().unwrap().as_raw_fd(); 1295 let pipesize = unsafe { libc::fcntl(fd, libc::F_SETPIPE_SZ, PIPE_SIZE) }; 1296 if pipesize == -1 { 1297 return Err(io::Error::last_os_error()); 1298 } 1299 let fd = child.stderr.as_ref().unwrap().as_raw_fd(); 1300 let pipesize1 = unsafe { libc::fcntl(fd, libc::F_SETPIPE_SZ, PIPE_SIZE) }; 1301 if pipesize1 == -1 { 1302 return Err(io::Error::last_os_error()); 1303 } 1304 1305 if pipesize >= PIPE_SIZE && pipesize1 >= PIPE_SIZE { 1306 Ok(child) 1307 } else { 1308 Err(std::io::Error::new( 1309 std::io::ErrorKind::Other, 1310 format!( 1311 "resizing pipe w/ 'fnctl' failed: stdout pipesize {pipesize}, stderr pipesize {pipesize1}" 1312 ), 1313 )) 1314 } 1315 } else { 1316 self.command.spawn() 1317 } 1318 } 1319 1320 pub fn args<I, S>(&mut self, args: I) -> &mut Self 1321 where 1322 I: IntoIterator<Item = S>, 1323 S: AsRef<OsStr>, 1324 { 1325 self.command.args(args); 1326 self 1327 } 1328 1329 pub fn default_disks(&mut self) -> &mut Self { 1330 if self.guest.disk_config.disk(DiskType::CloudInit).is_some() { 1331 self.args([ 1332 "--disk", 1333 format!( 1334 "path={}", 1335 self.guest 1336 .disk_config 1337 .disk(DiskType::OperatingSystem) 1338 .unwrap() 1339 ) 1340 .as_str(), 1341 format!( 1342 "path={}", 1343 self.guest.disk_config.disk(DiskType::CloudInit).unwrap() 1344 ) 1345 .as_str(), 1346 ]) 1347 } else { 1348 self.args([ 1349 "--disk", 1350 format!( 1351 "path={}", 1352 self.guest 1353 .disk_config 1354 .disk(DiskType::OperatingSystem) 1355 .unwrap() 1356 ) 1357 .as_str(), 1358 ]) 1359 } 1360 } 1361 1362 pub fn default_net(&mut self) -> &mut Self { 1363 self.args(["--net", self.guest.default_net_string().as_str()]) 1364 } 1365 } 1366 1367 pub fn clh_command(cmd: &str) -> String { 1368 env::var("BUILD_TARGET").map_or( 1369 format!("target/x86_64-unknown-linux-gnu/release/{cmd}"), 1370 |target| format!("target/{target}/release/{cmd}"), 1371 ) 1372 } 1373 1374 pub fn parse_iperf3_output(output: &[u8], sender: bool, bandwidth: bool) -> Result<f64, Error> { 1375 std::panic::catch_unwind(|| { 1376 let s = String::from_utf8_lossy(output); 1377 let v: Value = serde_json::from_str(&s).expect("'iperf3' parse error: invalid json output"); 1378 1379 if bandwidth { 1380 if sender { 1381 v["end"]["sum_sent"]["bits_per_second"] 1382 .as_f64() 1383 .expect("'iperf3' parse error: missing entry 'end.sum_sent.bits_per_second'") 1384 } else { 1385 v["end"]["sum_received"]["bits_per_second"].as_f64().expect( 1386 "'iperf3' parse error: missing entry 'end.sum_received.bits_per_second'", 1387 ) 1388 } 1389 } else { 1390 // iperf does not distinguish sent vs received in this case. 1391 1392 let lost_packets = v["end"]["sum"]["lost_packets"] 1393 .as_f64() 1394 .expect("'iperf3' parse error: missing entry 'end.sum.lost_packets'"); 1395 let packets = v["end"]["sum"]["packets"] 1396 .as_f64() 1397 .expect("'iperf3' parse error: missing entry 'end.sum.packets'"); 1398 let seconds = v["end"]["sum"]["seconds"] 1399 .as_f64() 1400 .expect("'iperf3' parse error: missing entry 'end.sum.seconds'"); 1401 1402 (packets - lost_packets) / seconds 1403 } 1404 }) 1405 .map_err(|_| { 1406 eprintln!( 1407 "==== Start iperf3 output ===\n\n{}\n\n=== End iperf3 output ===\n\n", 1408 String::from_utf8_lossy(output) 1409 ); 1410 Error::Iperf3Parse 1411 }) 1412 } 1413 1414 #[derive(Clone)] 1415 pub enum FioOps { 1416 Read, 1417 RandomRead, 1418 Write, 1419 RandomWrite, 1420 ReadWrite, 1421 RandRW, 1422 } 1423 1424 impl fmt::Display for FioOps { 1425 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { 1426 match self { 1427 FioOps::Read => write!(f, "read"), 1428 FioOps::RandomRead => write!(f, "randread"), 1429 FioOps::Write => write!(f, "write"), 1430 FioOps::RandomWrite => write!(f, "randwrite"), 1431 FioOps::ReadWrite => write!(f, "rw"), 1432 FioOps::RandRW => write!(f, "randrw"), 1433 } 1434 } 1435 } 1436 1437 pub fn parse_fio_output(output: &str, fio_ops: &FioOps, num_jobs: u32) -> Result<f64, Error> { 1438 std::panic::catch_unwind(|| { 1439 let v: Value = 1440 serde_json::from_str(output).expect("'fio' parse error: invalid json output"); 1441 let jobs = v["jobs"] 1442 .as_array() 1443 .expect("'fio' parse error: missing entry 'jobs'"); 1444 assert_eq!( 1445 jobs.len(), 1446 num_jobs as usize, 1447 "'fio' parse error: Unexpected number of 'fio' jobs." 1448 ); 1449 1450 let (read, write) = match fio_ops { 1451 FioOps::Read | FioOps::RandomRead => (true, false), 1452 FioOps::Write | FioOps::RandomWrite => (false, true), 1453 FioOps::ReadWrite | FioOps::RandRW => (true, true), 1454 }; 1455 1456 let mut total_bps = 0_f64; 1457 for j in jobs { 1458 if read { 1459 let bytes = j["read"]["io_bytes"] 1460 .as_u64() 1461 .expect("'fio' parse error: missing entry 'read.io_bytes'"); 1462 let runtime = j["read"]["runtime"] 1463 .as_u64() 1464 .expect("'fio' parse error: missing entry 'read.runtime'") 1465 as f64 1466 / 1000_f64; 1467 total_bps += bytes as f64 / runtime; 1468 } 1469 if write { 1470 let bytes = j["write"]["io_bytes"] 1471 .as_u64() 1472 .expect("'fio' parse error: missing entry 'write.io_bytes'"); 1473 let runtime = j["write"]["runtime"] 1474 .as_u64() 1475 .expect("'fio' parse error: missing entry 'write.runtime'") 1476 as f64 1477 / 1000_f64; 1478 total_bps += bytes as f64 / runtime; 1479 } 1480 } 1481 1482 total_bps 1483 }) 1484 .map_err(|_| { 1485 eprintln!("=== Start Fio output ===\n\n{output}\n\n=== End Fio output ===\n\n"); 1486 Error::FioOutputParse 1487 }) 1488 } 1489 1490 pub fn parse_fio_output_iops(output: &str, fio_ops: &FioOps, num_jobs: u32) -> Result<f64, Error> { 1491 std::panic::catch_unwind(|| { 1492 let v: Value = 1493 serde_json::from_str(output).expect("'fio' parse error: invalid json output"); 1494 let jobs = v["jobs"] 1495 .as_array() 1496 .expect("'fio' parse error: missing entry 'jobs'"); 1497 assert_eq!( 1498 jobs.len(), 1499 num_jobs as usize, 1500 "'fio' parse error: Unexpected number of 'fio' jobs." 1501 ); 1502 1503 let (read, write) = match fio_ops { 1504 FioOps::Read | FioOps::RandomRead => (true, false), 1505 FioOps::Write | FioOps::RandomWrite => (false, true), 1506 FioOps::ReadWrite | FioOps::RandRW => (true, true), 1507 }; 1508 1509 let mut total_iops = 0_f64; 1510 for j in jobs { 1511 if read { 1512 let ios = j["read"]["total_ios"] 1513 .as_u64() 1514 .expect("'fio' parse error: missing entry 'read.total_ios'"); 1515 let runtime = j["read"]["runtime"] 1516 .as_u64() 1517 .expect("'fio' parse error: missing entry 'read.runtime'") 1518 as f64 1519 / 1000_f64; 1520 total_iops += ios as f64 / runtime; 1521 } 1522 if write { 1523 let ios = j["write"]["total_ios"] 1524 .as_u64() 1525 .expect("'fio' parse error: missing entry 'write.total_ios'"); 1526 let runtime = j["write"]["runtime"] 1527 .as_u64() 1528 .expect("'fio' parse error: missing entry 'write.runtime'") 1529 as f64 1530 / 1000_f64; 1531 total_iops += ios as f64 / runtime; 1532 } 1533 } 1534 1535 total_iops 1536 }) 1537 .map_err(|_| { 1538 eprintln!("=== Start Fio output ===\n\n{output}\n\n=== End Fio output ===\n\n"); 1539 Error::FioOutputParse 1540 }) 1541 } 1542 1543 // Wait the child process for a given timeout 1544 fn child_wait_timeout(child: &mut Child, timeout: u64) -> Result<(), WaitTimeoutError> { 1545 match child.wait_timeout(Duration::from_secs(timeout)) { 1546 Err(e) => { 1547 return Err(WaitTimeoutError::General(e)); 1548 } 1549 Ok(s) => match s { 1550 None => { 1551 return Err(WaitTimeoutError::Timedout); 1552 } 1553 Some(s) => { 1554 if !s.success() { 1555 return Err(WaitTimeoutError::ExitStatus); 1556 } 1557 } 1558 }, 1559 } 1560 1561 Ok(()) 1562 } 1563 1564 pub fn measure_virtio_net_throughput( 1565 test_timeout: u32, 1566 queue_pairs: u32, 1567 guest: &Guest, 1568 receive: bool, 1569 bandwidth: bool, 1570 ) -> Result<f64, Error> { 1571 let default_port = 5201; 1572 1573 // 1. start the iperf3 server on the guest 1574 for n in 0..queue_pairs { 1575 guest.ssh_command(&format!("iperf3 -s -p {} -D", default_port + n))?; 1576 } 1577 1578 thread::sleep(Duration::new(1, 0)); 1579 1580 // 2. start the iperf3 client on host to measure RX through-put 1581 let mut clients = Vec::new(); 1582 for n in 0..queue_pairs { 1583 let mut cmd = Command::new("iperf3"); 1584 cmd.args([ 1585 "-J", // Output in JSON format 1586 "-c", 1587 &guest.network.guest_ip, 1588 "-p", 1589 &format!("{}", default_port + n), 1590 "-t", 1591 &format!("{test_timeout}"), 1592 "-i", 1593 "0", 1594 ]); 1595 // For measuring the guest transmit throughput (as a sender), 1596 // use reverse mode of the iperf3 client on the host 1597 if !receive { 1598 cmd.args(["-R"]); 1599 } 1600 // Use UDP stream to measure packets per second. The bitrate is set to 1601 // 1T to make sure it saturates the link. 1602 if !bandwidth { 1603 cmd.args(["-u", "-b", "1T"]); 1604 } 1605 let client = cmd 1606 .stderr(Stdio::piped()) 1607 .stdout(Stdio::piped()) 1608 .spawn() 1609 .map_err(Error::Spawn)?; 1610 1611 clients.push(client); 1612 } 1613 1614 let mut err: Option<Error> = None; 1615 let mut results = Vec::new(); 1616 let mut failed = false; 1617 for c in clients { 1618 let mut c = c; 1619 if let Err(e) = child_wait_timeout(&mut c, test_timeout as u64 + 5) { 1620 err = Some(Error::WaitTimeout(e)); 1621 failed = true; 1622 } 1623 1624 if !failed { 1625 // Safe to unwrap as we know the child has terminated succesffully 1626 let output = c.wait_with_output().unwrap(); 1627 results.push(parse_iperf3_output(&output.stdout, receive, bandwidth)?); 1628 } else { 1629 let _ = c.kill(); 1630 let output = c.wait_with_output().unwrap(); 1631 println!( 1632 "=============== Client output [Error] ===============\n\n{}\n\n===========end============\n\n", 1633 String::from_utf8_lossy(&output.stdout) 1634 ); 1635 } 1636 } 1637 1638 if let Some(e) = err { 1639 Err(e) 1640 } else { 1641 Ok(results.iter().sum()) 1642 } 1643 } 1644 1645 pub fn parse_ethr_latency_output(output: &[u8]) -> Result<Vec<f64>, Error> { 1646 std::panic::catch_unwind(|| { 1647 let s = String::from_utf8_lossy(output); 1648 let mut latency = Vec::new(); 1649 for l in s.lines() { 1650 let v: Value = serde_json::from_str(l).expect("'ethr' parse error: invalid json line"); 1651 // Skip header/summary lines 1652 if let Some(avg) = v["Avg"].as_str() { 1653 // Assume the latency unit is always "us" 1654 latency.push( 1655 avg.split("us").collect::<Vec<&str>>()[0] 1656 .parse::<f64>() 1657 .expect("'ethr' parse error: invalid 'Avg' entry"), 1658 ); 1659 } 1660 } 1661 1662 assert!( 1663 !latency.is_empty(), 1664 "'ethr' parse error: no valid latency data found" 1665 ); 1666 1667 latency 1668 }) 1669 .map_err(|_| { 1670 eprintln!( 1671 "=== Start ethr output ===\n\n{}\n\n=== End ethr output ===\n\n", 1672 String::from_utf8_lossy(output) 1673 ); 1674 Error::EthrLogParse 1675 }) 1676 } 1677 1678 pub fn measure_virtio_net_latency(guest: &Guest, test_timeout: u32) -> Result<Vec<f64>, Error> { 1679 // copy the 'ethr' tool to the guest image 1680 let ethr_path = "/usr/local/bin/ethr"; 1681 let ethr_remote_path = "/tmp/ethr"; 1682 scp_to_guest( 1683 Path::new(ethr_path), 1684 Path::new(ethr_remote_path), 1685 &guest.network.guest_ip, 1686 //DEFAULT_SSH_RETRIES, 1687 1, 1688 DEFAULT_SSH_TIMEOUT, 1689 )?; 1690 1691 // Start the ethr server on the guest 1692 guest.ssh_command(&format!("{ethr_remote_path} -s &> /dev/null &"))?; 1693 1694 thread::sleep(Duration::new(10, 0)); 1695 1696 // Start the ethr client on the host 1697 let log_file = guest 1698 .tmp_dir 1699 .as_path() 1700 .join("ethr.client.log") 1701 .to_str() 1702 .unwrap() 1703 .to_string(); 1704 let mut c = Command::new(ethr_path) 1705 .args([ 1706 "-c", 1707 &guest.network.guest_ip, 1708 "-t", 1709 "l", 1710 "-o", 1711 &log_file, // file output is JSON format 1712 "-d", 1713 &format!("{test_timeout}s"), 1714 ]) 1715 .stderr(Stdio::piped()) 1716 .stdout(Stdio::piped()) 1717 .spawn() 1718 .map_err(Error::Spawn)?; 1719 1720 if let Err(e) = child_wait_timeout(&mut c, test_timeout as u64 + 5).map_err(Error::WaitTimeout) 1721 { 1722 let _ = c.kill(); 1723 return Err(e); 1724 } 1725 1726 // Parse the ethr latency test output 1727 let content = fs::read(log_file).map_err(Error::EthrLogFile)?; 1728 parse_ethr_latency_output(&content) 1729 } 1730