src/server/worker_task.rs

   1 use std::collections::{HashMap, VecDeque};
   2 use std::fs::File;
   3 use std::path::Path;
   4 use std::io::{Read, Write, BufRead, BufReader};
   5 use std::panic::UnwindSafe;
   6 use std::sync::atomic::{AtomicBool, Ordering};
   7 use std::sync::{Arc, Mutex};
   8
   9 use anyhow::{bail, format_err, Error};
  10 use futures::*;
  11 use lazy_static::lazy_static;
  12 use nix::unistd::Pid;
  13 use serde_json::{json, Value};
  14 use serde::{Serialize, Deserialize};
  15 use tokio::sync::oneshot;
  16
  17 use proxmox::sys::linux::procfs;
  18 use proxmox::try_block;
  19 use proxmox::tools::fs::{create_path, open_file_locked, replace_file, CreateOptions};
  20
  21 use super::UPID;
  22
  23 use crate::tools::logrotate::{LogRotate, LogRotateFiles};
  24 use crate::tools::FileLogger;
  25 use crate::api2::types::Userid;
  26
  27 macro_rules! PROXMOX_BACKUP_VAR_RUN_DIR_M { () => ("/run/proxmox-backup") }
  28 macro_rules! PROXMOX_BACKUP_LOG_DIR_M { () => ("/var/log/proxmox-backup") }
  29 macro_rules! PROXMOX_BACKUP_TASK_DIR_M { () => (concat!( PROXMOX_BACKUP_LOG_DIR_M!(), "/tasks")) }
  30
  31 pub const PROXMOX_BACKUP_VAR_RUN_DIR: &str = PROXMOX_BACKUP_VAR_RUN_DIR_M!();
  32 pub const PROXMOX_BACKUP_LOG_DIR: &str = PROXMOX_BACKUP_LOG_DIR_M!();
  33 pub const PROXMOX_BACKUP_TASK_DIR: &str = PROXMOX_BACKUP_TASK_DIR_M!();
  34 pub const PROXMOX_BACKUP_TASK_LOCK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/.active.lock");
  35 pub const PROXMOX_BACKUP_ACTIVE_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/active");
  36 pub const PROXMOX_BACKUP_INDEX_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/index");
  37 pub const PROXMOX_BACKUP_ARCHIVE_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/archive");
  38
  39 const MAX_INDEX_TASKS: usize = 1000;
  40
  41 lazy_static! {
  42     static ref WORKER_TASK_LIST: Mutex<HashMap<usize, Arc<WorkerTask>>> = Mutex::new(HashMap::new());
  43
  44     static ref MY_PID: i32 = unsafe { libc::getpid() };
  45     static ref MY_PID_PSTART: u64 = procfs::PidStat::read_from_pid(Pid::from_raw(*MY_PID))
  46         .unwrap()
  47         .starttime;
  48 }
  49
  50 /// Test if the task is still running
  51 pub async fn worker_is_active(upid: &UPID) -> Result<bool, Error> {
  52     if (upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART) {
  53         return Ok(WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id));
  54     }
  55
  56     if !procfs::check_process_running_pstart(upid.pid, upid.pstart).is_some() {
  57         return Ok(false);
  58     }
  59
  60     let socketname = format!(
  61         "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, upid.pid);
  62
  63     let cmd = json!({
  64         "command": "status",
  65         "upid": upid.to_string(),
  66     });
  67
  68     let status = super::send_command(socketname, cmd).await?;
  69
  70     if let Some(active) = status.as_bool() {
  71         Ok(active)
  72     } else {
  73         bail!("got unexpected result {:?} (expected bool)", status);
  74     }
  75 }
  76
  77 /// Test if the task is still running (fast but inaccurate implementation)
  78 ///
  79 /// If the task is spanned from a different process, we simply return if
  80 /// that process is still running. This information is good enough to detect
  81 /// stale tasks...
  82 pub fn worker_is_active_local(upid: &UPID) -> bool {
  83     if (upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART) {
  84         WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id)
  85     } else {
  86         procfs::check_process_running_pstart(upid.pid, upid.pstart).is_some()
  87     }
  88 }
  89
  90 pub fn create_task_control_socket() -> Result<(), Error> {
  91
  92     let socketname = format!(
  93         "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, *MY_PID);
  94
  95     let control_future = super::create_control_socket(socketname, |param| {
  96         let param = param.as_object()
  97             .ok_or_else(|| format_err!("unable to parse parameters (expected json object)"))?;
  98         if param.keys().count() != 2 { bail!("wrong number of parameters"); }
  99
 100         let command = param["command"].as_str()
 101             .ok_or_else(|| format_err!("unable to parse parameters (missing command)"))?;
 102
 103         // we have only two commands for now
 104         if !(command == "abort-task" || command == "status") { bail!("got unknown command '{}'", command); }
 105
 106         let upid_str = param["upid"].as_str()
 107             .ok_or_else(|| format_err!("unable to parse parameters (missing upid)"))?;
 108
 109         let upid = upid_str.parse::<UPID>()?;
 110
 111         if !((upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART)) {
 112             bail!("upid does not belong to this process");
 113         }
 114
 115         let hash = WORKER_TASK_LIST.lock().unwrap();
 116
 117         match command {
 118             "abort-task" => {
 119                 if let Some(ref worker) = hash.get(&upid.task_id) {
 120                     worker.request_abort();
 121                 } else {
 122                     // assume task is already stopped
 123                 }
 124                 Ok(Value::Null)
 125             }
 126             "status" => {
 127                 let active = hash.contains_key(&upid.task_id);
 128                 Ok(active.into())
 129             }
 130             _ => {
 131                 bail!("got unknown command '{}'", command);
 132             }
 133         }
 134     })?;
 135
 136     tokio::spawn(control_future);
 137
 138     Ok(())
 139 }
 140
 141 pub fn abort_worker_async(upid: UPID) {
 142     tokio::spawn(async move {
 143         if let Err(err) = abort_worker(upid).await {
 144             eprintln!("abort worker failed - {}", err);
 145         }
 146     });
 147 }
 148
 149 pub async fn abort_worker(upid: UPID) -> Result<(), Error> {
 150
 151     let target_pid = upid.pid;
 152
 153     let socketname = format!(
 154         "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, target_pid);
 155
 156     let cmd = json!({
 157         "command": "abort-task",
 158         "upid": upid.to_string(),
 159     });
 160
 161     super::send_command(socketname, cmd).map_ok(|_| ()).await
 162 }
 163
 164 fn parse_worker_status_line(line: &str) -> Result<(String, UPID, Option<TaskState>), Error> {
 165
 166     let data = line.splitn(3, ' ').collect::<Vec<&str>>();
 167
 168     let len = data.len();
 169
 170     match len {
 171         1 => Ok((data[0].to_owned(), data[0].parse::<UPID>()?, None)),
 172         3 => {
 173             let endtime = i64::from_str_radix(data[1], 16)?;
 174             let state = TaskState::from_endtime_and_message(endtime, data[2])?;
 175             Ok((data[0].to_owned(), data[0].parse::<UPID>()?, Some(state)))
 176         }
 177         _ => bail!("wrong number of components"),
 178     }
 179 }
 180
 181 /// Create task log directory with correct permissions
 182 pub fn create_task_log_dirs() -> Result<(), Error> {
 183
 184     try_block!({
 185         let backup_user = crate::backup::backup_user()?;
 186         let opts = CreateOptions::new()
 187             .owner(backup_user.uid)
 188             .group(backup_user.gid);
 189
 190         create_path(PROXMOX_BACKUP_LOG_DIR, None, Some(opts.clone()))?;
 191         create_path(PROXMOX_BACKUP_TASK_DIR, None, Some(opts.clone()))?;
 192         create_path(PROXMOX_BACKUP_VAR_RUN_DIR, None, Some(opts))?;
 193         Ok(())
 194     }).map_err(|err: Error| format_err!("unable to create task log dir - {}", err))?;
 195
 196     Ok(())
 197 }
 198
 199 /// Read endtime (time of last log line) and exitstatus from task log file
 200 /// If there is not a single line with at valid datetime, we assume the
 201 /// starttime to be the endtime
 202 pub fn upid_read_status(upid: &UPID) -> Result<TaskState, Error> {
 203
 204     let mut status = TaskState::Unknown { endtime: upid.starttime };
 205
 206     let path = upid.log_path();
 207
 208     let mut file = File::open(path)?;
 209
 210     /// speedup - only read tail
 211     use std::io::Seek;
 212     use std::io::SeekFrom;
 213     let _ = file.seek(SeekFrom::End(-8192)); // ignore errors
 214
 215     let mut data = Vec::with_capacity(8192);
 216     file.read_to_end(&mut data)?;
 217
 218     // task logs should end with newline, we do not want it here
 219     if data.len() > 0 && data[data.len()-1] == b'\n' {
 220         data.pop();
 221     }
 222
 223     let last_line = {
 224         let mut start = 0;
 225         for pos in (0..data.len()).rev() {
 226             if data[pos] == b'\n' {
 227                 start = data.len().min(pos + 1);
 228                 break;
 229             }
 230         }
 231         &data[start..]
 232     };
 233
 234     let last_line = std::str::from_utf8(last_line)
 235         .map_err(|err| format_err!("upid_read_status: utf8 parse failed: {}", err))?;
 236
 237     let mut iter = last_line.splitn(2, ": ");
 238     if let Some(time_str) = iter.next() {
 239         if let Ok(endtime) = proxmox::tools::time::parse_rfc3339(time_str) {
 240             if let Some(rest) = iter.next().and_then(|rest| rest.strip_prefix("TASK ")) {
 241                 if let Ok(state) = TaskState::from_endtime_and_message(endtime, rest) {
 242                     status = state;
 243                 }
 244             }
 245         }
 246     }
 247
 248     Ok(status)
 249 }
 250
 251 /// Task State
 252 #[derive(Debug, PartialEq, Eq, Serialize, Deserialize)]
 253 pub enum TaskState {
 254     /// The Task ended with an undefined state
 255     Unknown { endtime: i64 },
 256     /// The Task ended and there were no errors or warnings
 257     OK { endtime: i64 },
 258     /// The Task had 'count' amount of warnings and no errors
 259     Warning { count: u64, endtime: i64 },
 260     /// The Task ended with the error described in 'message'
 261     Error { message: String, endtime: i64 },
 262 }
 263
 264 impl TaskState {
 265     pub fn endtime(&self) -> i64 {
 266         match *self {
 267             TaskState::Unknown { endtime } => endtime,
 268             TaskState::OK { endtime } => endtime,
 269             TaskState::Warning { endtime, .. } => endtime,
 270             TaskState::Error { endtime, .. } => endtime,
 271         }
 272     }
 273
 274     fn result_text(&self) -> String {
 275         match self {
 276             TaskState::Error { message, .. } => format!("TASK ERROR: {}", message),
 277             other => format!("TASK {}", other),
 278         }
 279     }
 280
 281     fn from_endtime_and_message(endtime: i64, s: &str) -> Result<Self, Error> {
 282         if s == "unknown" {
 283             Ok(TaskState::Unknown { endtime })
 284         } else if s == "OK" {
 285             Ok(TaskState::OK { endtime })
 286         } else if s.starts_with("WARNINGS: ") {
 287             let count: u64 = s[10..].parse()?;
 288             Ok(TaskState::Warning{ count, endtime })
 289         } else if s.len() > 0 {
 290             let message = if s.starts_with("ERROR: ") { &s[7..] } else { s }.to_string();
 291             Ok(TaskState::Error{ message, endtime })
 292         } else {
 293             bail!("unable to parse Task Status '{}'", s);
 294         }
 295     }
 296 }
 297
 298 impl std::cmp::PartialOrd for TaskState {
 299     fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
 300         Some(self.endtime().cmp(&other.endtime()))
 301     }
 302 }
 303
 304 impl std::cmp::Ord for TaskState {
 305     fn cmp(&self, other: &Self) -> std::cmp::Ordering {
 306         self.endtime().cmp(&other.endtime())
 307     }
 308 }
 309
 310 impl std::fmt::Display for TaskState {
 311     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
 312         match self {
 313             TaskState::Unknown { .. } => write!(f, "unknown"),
 314             TaskState::OK { .. }=> write!(f, "OK"),
 315             TaskState::Warning { count, .. } => write!(f, "WARNINGS: {}", count),
 316             TaskState::Error { message, .. } => write!(f, "{}", message),
 317         }
 318     }
 319 }
 320
 321 /// Task details including parsed UPID
 322 ///
 323 /// If there is no `state`, the task is still running.
 324 #[derive(Debug)]
 325 pub struct TaskListInfo {
 326     /// The parsed UPID
 327     pub upid: UPID,
 328     /// UPID string representation
 329     pub upid_str: String,
 330     /// Task `(endtime, status)` if already finished
 331     pub state: Option<TaskState>, // endtime, status
 332 }
 333
 334 fn lock_task_list_files(exclusive: bool) -> Result<std::fs::File, Error> {
 335     let backup_user = crate::backup::backup_user()?;
 336
 337     let lock = open_file_locked(PROXMOX_BACKUP_TASK_LOCK_FN, std::time::Duration::new(10, 0), exclusive)?;
 338     nix::unistd::chown(PROXMOX_BACKUP_TASK_LOCK_FN, Some(backup_user.uid), Some(backup_user.gid))?;
 339
 340     Ok(lock)
 341 }
 342
 343 /// checks if the Task Archive is bigger that 'size_threshold' bytes, and
 344 /// rotates it if it is
 345 pub fn rotate_task_log_archive(size_threshold: u64, compress: bool, max_files: Option<usize>) -> Result<bool, Error> {
 346     let _lock = lock_task_list_files(true)?;
 347     let path = Path::new(PROXMOX_BACKUP_ARCHIVE_TASK_FN);
 348     let metadata = match path.metadata() {
 349         Ok(metadata) => metadata,
 350         Err(err) if err.kind() == std::io::ErrorKind::NotFound => return Ok(false),
 351         Err(err) =>  bail!("unable to open task archive - {}", err),
 352     };
 353
 354     if metadata.len() > size_threshold {
 355         let mut logrotate = LogRotate::new(PROXMOX_BACKUP_ARCHIVE_TASK_FN, compress).ok_or_else(|| format_err!("could not get archive file names"))?;
 356         let backup_user = crate::backup::backup_user()?;
 357         logrotate.rotate(
 358             CreateOptions::new()
 359                 .owner(backup_user.uid)
 360                 .group(backup_user.gid),
 361             max_files,
 362         )?;
 363         Ok(true)
 364     } else {
 365         Ok(false)
 366     }
 367 }
 368
 369 // atomically read/update the task list, update status of finished tasks
 370 // new_upid is added to the list when specified.
 371 fn update_active_workers(new_upid: Option<&UPID>) -> Result<(), Error> {
 372
 373     let backup_user = crate::backup::backup_user()?;
 374
 375     let lock = lock_task_list_files(true)?;
 376
 377     let mut finish_list: Vec<TaskListInfo> = read_task_file_from_path(PROXMOX_BACKUP_INDEX_TASK_FN)?;
 378     let mut active_list: Vec<TaskListInfo> = read_task_file_from_path(PROXMOX_BACKUP_ACTIVE_TASK_FN)?
 379         .into_iter()
 380         .filter_map(|info| {
 381             if info.state.is_some() {
 382                 // this can happen when the active file still includes finished tasks
 383                 finish_list.push(info);
 384                 return None;
 385             }
 386
 387             if !worker_is_active_local(&info.upid) {
 388                 println!("Detected stopped UPID {}", &info.upid_str);
 389                 let now = proxmox::tools::time::epoch_i64();
 390                 let status = upid_read_status(&info.upid)
 391                     .unwrap_or_else(|_| TaskState::Unknown { endtime: now });
 392                 finish_list.push(TaskListInfo {
 393                     upid: info.upid,
 394                     upid_str: info.upid_str,
 395                     state: Some(status)
 396                 });
 397                 return None;
 398             }
 399
 400             Some(info)
 401         }).collect();
 402
 403     if let Some(upid) = new_upid {
 404         active_list.push(TaskListInfo { upid: upid.clone(), upid_str: upid.to_string(), state: None });
 405     }
 406
 407     let active_raw = render_task_list(&active_list);
 408
 409     replace_file(
 410         PROXMOX_BACKUP_ACTIVE_TASK_FN,
 411         active_raw.as_bytes(),
 412         CreateOptions::new()
 413             .owner(backup_user.uid)
 414             .group(backup_user.gid),
 415     )?;
 416
 417     finish_list.sort_unstable_by(|a, b| {
 418         match (&a.state, &b.state) {
 419             (Some(s1), Some(s2)) => s1.cmp(&s2),
 420             (Some(_), None) => std::cmp::Ordering::Less,
 421             (None, Some(_)) => std::cmp::Ordering::Greater,
 422             _ => a.upid.starttime.cmp(&b.upid.starttime),
 423         }
 424     });
 425
 426
 427     let start = if finish_list.len() > MAX_INDEX_TASKS {
 428         finish_list.len() - MAX_INDEX_TASKS
 429     } else {
 430         0
 431     };
 432
 433     let end = (start+MAX_INDEX_TASKS).min(finish_list.len());
 434
 435     let index_raw = if end > start {
 436         render_task_list(&finish_list[start..end])
 437     } else {
 438         "".to_string()
 439     };
 440
 441     replace_file(
 442         PROXMOX_BACKUP_INDEX_TASK_FN,
 443         index_raw.as_bytes(),
 444         CreateOptions::new()
 445             .owner(backup_user.uid)
 446             .group(backup_user.gid),
 447     )?;
 448
 449     if !finish_list.is_empty() && start > 0 {
 450         match std::fs::OpenOptions::new().append(true).create(true).open(PROXMOX_BACKUP_ARCHIVE_TASK_FN) {
 451             Ok(mut writer) => {
 452                 for info in &finish_list[0..start] {
 453                     writer.write_all(render_task_line(&info).as_bytes())?;
 454                 }
 455             },
 456             Err(err) => bail!("could not write task archive - {}", err),
 457         }
 458
 459         nix::unistd::chown(PROXMOX_BACKUP_ARCHIVE_TASK_FN, Some(backup_user.uid), Some(backup_user.gid))?;
 460     }
 461
 462     drop(lock);
 463
 464     Ok(())
 465 }
 466
 467 fn render_task_line(info: &TaskListInfo) -> String {
 468     let mut raw = String::new();
 469     if let Some(status) = &info.state {
 470         raw.push_str(&format!("{} {:08X} {}\n", info.upid_str, status.endtime(), status));
 471     } else {
 472         raw.push_str(&info.upid_str);
 473         raw.push('\n');
 474     }
 475
 476     raw
 477 }
 478
 479 fn render_task_list(list: &[TaskListInfo]) -> String {
 480     let mut raw = String::new();
 481     for info in list {
 482         raw.push_str(&render_task_line(&info));
 483     }
 484     raw
 485 }
 486
 487 // note this is not locked, caller has to make sure it is
 488 // this will skip (and log) lines that are not valid status lines
 489 fn read_task_file<R: Read>(reader: R) -> Result<Vec<TaskListInfo>, Error>
 490 {
 491     let reader = BufReader::new(reader);
 492     let mut list = Vec::new();
 493     for line in reader.lines() {
 494         let line = line?;
 495         match parse_worker_status_line(&line) {
 496             Ok((upid_str, upid, state)) => list.push(TaskListInfo {
 497                 upid_str,
 498                 upid,
 499                 state
 500             }),
 501             Err(err) => {
 502                 eprintln!("unable to parse worker status '{}' - {}", line, err);
 503                 continue;
 504             }
 505         };
 506     }
 507
 508     Ok(list)
 509 }
 510
 511 // note this is not locked, caller has to make sure it is
 512 fn read_task_file_from_path<P>(path: P) -> Result<Vec<TaskListInfo>, Error>
 513 where
 514     P: AsRef<std::path::Path> + std::fmt::Debug,
 515 {
 516     let file = match File::open(&path) {
 517         Ok(f) => f,
 518         Err(err) if err.kind() == std::io::ErrorKind::NotFound => return Ok(Vec::new()),
 519         Err(err) => bail!("unable to open task list {:?} - {}", path, err),
 520     };
 521
 522     read_task_file(file)
 523 }
 524
 525 enum TaskFile {
 526     Active,
 527     Index,
 528     Archive,
 529     End,
 530 }
 531
 532 pub struct TaskListInfoIterator {
 533     list: VecDeque<TaskListInfo>,
 534     file: TaskFile,
 535     archive: Option<LogRotateFiles>,
 536     lock: Option<File>,
 537 }
 538
 539 impl TaskListInfoIterator {
 540     pub fn new(active_only: bool) -> Result<Self, Error> {
 541         let (read_lock, active_list) = {
 542             let lock = lock_task_list_files(false)?;
 543             let active_list = read_task_file_from_path(PROXMOX_BACKUP_ACTIVE_TASK_FN)?;
 544
 545             let needs_update = active_list
 546                 .iter()
 547                 .any(|info| info.state.is_some() || !worker_is_active_local(&info.upid));
 548
 549             if needs_update {
 550                 drop(lock);
 551                 update_active_workers(None)?;
 552                 let lock = lock_task_list_files(false)?;
 553                 let active_list = read_task_file_from_path(PROXMOX_BACKUP_ACTIVE_TASK_FN)?;
 554                 (lock, active_list)
 555             } else {
 556                 (lock, active_list)
 557             }
 558         };
 559
 560         let archive = if active_only {
 561             None
 562         } else {
 563             let logrotate = LogRotate::new(PROXMOX_BACKUP_ARCHIVE_TASK_FN, true).ok_or_else(|| format_err!("could not get archive file names"))?;
 564             Some(logrotate.files())
 565         };
 566
 567         let file = if active_only { TaskFile::End } else { TaskFile::Active };
 568         let lock = if active_only { None } else { Some(read_lock) };
 569
 570         Ok(Self {
 571             list: active_list.into(),
 572             file,
 573             archive,
 574             lock,
 575         })
 576     }
 577 }
 578
 579 impl Iterator for TaskListInfoIterator {
 580     type Item = Result<TaskListInfo, Error>;
 581
 582     fn next(&mut self) -> Option<Self::Item> {
 583         loop {
 584             if let Some(element) = self.list.pop_back() {
 585                 return Some(Ok(element));
 586             } else {
 587                 match self.file {
 588                     TaskFile::Active => {
 589                         let index = match read_task_file_from_path(PROXMOX_BACKUP_INDEX_TASK_FN) {
 590                             Ok(index) => index,
 591                             Err(err) => return Some(Err(err)),
 592                         };
 593                         self.list.append(&mut index.into());
 594                         self.file = TaskFile::Index;
 595                     },
 596                     TaskFile::Index | TaskFile::Archive => {
 597                         if let Some(mut archive) = self.archive.take() {
 598                             if let Some(file) = archive.next() {
 599                                 let list = match read_task_file(file) {
 600                                     Ok(list) => list,
 601                                     Err(err) => return Some(Err(err)),
 602                                 };
 603                                 self.list.append(&mut list.into());
 604                                 self.archive = Some(archive);
 605                                 self.file = TaskFile::Archive;
 606                                 continue;
 607                             }
 608                         }
 609                         self.file = TaskFile::End;
 610                         self.lock.take();
 611                         return None;
 612                     }
 613                     TaskFile::End => return None,
 614                 }
 615             }
 616         }
 617     }
 618 }
 619
 620 /// Launch long running worker tasks.
 621 ///
 622 /// A worker task can either be a whole thread, or a simply tokio
 623 /// task/future. Each task can `log()` messages, which are stored
 624 /// persistently to files. Task should poll the `abort_requested`
 625 /// flag, and stop execution when requested.
 626 #[derive(Debug)]
 627 pub struct WorkerTask {
 628     upid: UPID,
 629     data: Mutex<WorkerTaskData>,
 630     abort_requested: AtomicBool,
 631 }
 632
 633 impl std::fmt::Display for WorkerTask {
 634
 635     fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
 636         self.upid.fmt(f)
 637     }
 638 }
 639
 640 #[derive(Debug)]
 641 struct WorkerTaskData {
 642     logger: FileLogger,
 643     progress: f64, // 0..1
 644     warn_count: u64,
 645     pub abort_listeners: Vec<oneshot::Sender<()>>,
 646 }
 647
 648 impl Drop for WorkerTask {
 649
 650     fn drop(&mut self) {
 651         println!("unregister worker");
 652     }
 653 }
 654
 655 impl WorkerTask {
 656
 657     pub fn new(worker_type: &str, worker_id: Option<String>, userid: Userid, to_stdout: bool) -> Result<Arc<Self>, Error> {
 658         println!("register worker");
 659
 660         let upid = UPID::new(worker_type, worker_id, userid)?;
 661         let task_id = upid.task_id;
 662
 663         let mut path = std::path::PathBuf::from(PROXMOX_BACKUP_TASK_DIR);
 664
 665         path.push(format!("{:02X}", upid.pstart % 256));
 666
 667         let backup_user = crate::backup::backup_user()?;
 668
 669         create_path(&path, None, Some(CreateOptions::new().owner(backup_user.uid).group(backup_user.gid)))?;
 670
 671         path.push(upid.to_string());
 672
 673         println!("FILE: {:?}", path);
 674
 675         let logger = FileLogger::new(&path, to_stdout)?;
 676         nix::unistd::chown(&path, Some(backup_user.uid), Some(backup_user.gid))?;
 677
 678         let worker = Arc::new(Self {
 679             upid: upid.clone(),
 680             abort_requested: AtomicBool::new(false),
 681             data: Mutex::new(WorkerTaskData {
 682                 logger,
 683                 progress: 0.0,
 684                 warn_count: 0,
 685                 abort_listeners: vec![],
 686             }),
 687         });
 688
 689         // scope to drop the lock again after inserting
 690         {
 691             let mut hash = WORKER_TASK_LIST.lock().unwrap();
 692             hash.insert(task_id, worker.clone());
 693             super::set_worker_count(hash.len());
 694         }
 695
 696         update_active_workers(Some(&upid))?;
 697
 698         Ok(worker)
 699     }
 700
 701     /// Spawn a new tokio task/future.
 702     pub fn spawn<F, T>(
 703         worker_type: &str,
 704         worker_id: Option<String>,
 705         userid: Userid,
 706         to_stdout: bool,
 707         f: F,
 708     ) -> Result<String, Error>
 709         where F: Send + 'static + FnOnce(Arc<WorkerTask>) -> T,
 710               T: Send + 'static + Future<Output = Result<(), Error>>,
 711     {
 712         let worker = WorkerTask::new(worker_type, worker_id, userid, to_stdout)?;
 713         let upid_str = worker.upid.to_string();
 714         let f = f(worker.clone());
 715         tokio::spawn(async move {
 716             let result = f.await;
 717             worker.log_result(&result);
 718         });
 719
 720         Ok(upid_str)
 721     }
 722
 723     /// Create a new worker thread.
 724     pub fn new_thread<F>(
 725         worker_type: &str,
 726         worker_id: Option<String>,
 727         userid: Userid,
 728         to_stdout: bool,
 729         f: F,
 730     ) -> Result<String, Error>
 731         where F: Send + UnwindSafe + 'static + FnOnce(Arc<WorkerTask>) -> Result<(), Error>
 732     {
 733         println!("register worker thread");
 734
 735         let worker = WorkerTask::new(worker_type, worker_id, userid, to_stdout)?;
 736         let upid_str = worker.upid.to_string();
 737
 738         let _child = std::thread::Builder::new().name(upid_str.clone()).spawn(move || {
 739             let worker1 = worker.clone();
 740             let result = match std::panic::catch_unwind(move || f(worker1)) {
 741                 Ok(r) => r,
 742                 Err(panic) => {
 743                     match panic.downcast::<&str>() {
 744                         Ok(panic_msg) => {
 745                             Err(format_err!("worker panicked: {}", panic_msg))
 746                         }
 747                         Err(_) => {
 748                             Err(format_err!("worker panicked: unknown type."))
 749                         }
 750                     }
 751                 }
 752             };
 753
 754             worker.log_result(&result);
 755         });
 756
 757         Ok(upid_str)
 758     }
 759
 760     /// create state from self and a result
 761     pub fn create_state(&self, result: &Result<(), Error>) -> TaskState {
 762         let warn_count = self.data.lock().unwrap().warn_count;
 763
 764         let endtime = proxmox::tools::time::epoch_i64();
 765
 766         if let Err(err) = result {
 767             TaskState::Error { message: err.to_string(), endtime }
 768         } else if warn_count > 0 {
 769             TaskState::Warning { count: warn_count, endtime }
 770         } else {
 771             TaskState::OK { endtime }
 772         }
 773     }
 774
 775     /// Log task result, remove task from running list
 776     pub fn log_result(&self, result: &Result<(), Error>) {
 777         let state = self.create_state(result);
 778         self.log(state.result_text());
 779
 780         WORKER_TASK_LIST.lock().unwrap().remove(&self.upid.task_id);
 781         let _ = update_active_workers(None);
 782         super::set_worker_count(WORKER_TASK_LIST.lock().unwrap().len());
 783     }
 784
 785     /// Log a message.
 786     pub fn log<S: AsRef<str>>(&self, msg: S) {
 787         let mut data = self.data.lock().unwrap();
 788         data.logger.log(msg);
 789     }
 790
 791     /// Log a message as warning.
 792     pub fn warn<S: AsRef<str>>(&self, msg: S) {
 793         let mut data = self.data.lock().unwrap();
 794         data.logger.log(format!("WARN: {}", msg.as_ref()));
 795         data.warn_count += 1;
 796     }
 797
 798     /// Set progress indicator
 799     pub fn progress(&self, progress: f64) {
 800         if progress >= 0.0 && progress <= 1.0 {
 801             let mut data = self.data.lock().unwrap();
 802             data.progress = progress;
 803         } else {
 804            // fixme:  log!("task '{}': ignoring strange value for progress '{}'", self.upid, progress);
 805         }
 806     }
 807
 808     /// Request abort
 809     pub fn request_abort(&self) {
 810         eprintln!("set abort flag for worker {}", self.upid);
 811         self.abort_requested.store(true, Ordering::SeqCst);
 812         // noitify listeners
 813         let mut data = self.data.lock().unwrap();
 814         loop {
 815             match data.abort_listeners.pop() {
 816                 None => { break; },
 817                 Some(ch) => {
 818                     let _ = ch.send(()); // ignore erros here
 819                 },
 820             }
 821         }
 822     }
 823
 824     /// Test if abort was requested.
 825     pub fn abort_requested(&self) -> bool {
 826         self.abort_requested.load(Ordering::SeqCst)
 827     }
 828
 829     /// Fail if abort was requested.
 830     pub fn fail_on_abort(&self) -> Result<(), Error> {
 831         if self.abort_requested() {
 832             bail!("abort requested - aborting task");
 833         }
 834         Ok(())
 835     }
 836
 837     /// Get a future which resolves on task abort
 838     pub fn abort_future(&self) ->  oneshot::Receiver<()> {
 839         let (tx, rx) = oneshot::channel::<()>();
 840
 841         let mut data = self.data.lock().unwrap();
 842         if self.abort_requested() {
 843             let _ = tx.send(());
 844         } else {
 845             data.abort_listeners.push(tx);
 846         }
 847         rx
 848     }
 849
 850     pub fn upid(&self) -> &UPID {
 851         &self.upid
 852     }
 853 }
 854
 855 impl crate::task::TaskState for WorkerTask {
 856     fn check_abort(&self) -> Result<(), Error> {
 857         self.fail_on_abort()
 858     }
 859
 860     fn log(&self, level: log::Level, message: &std::fmt::Arguments) {
 861         match level {
 862             log::Level::Error => self.warn(&message.to_string()),
 863             log::Level::Warn => self.warn(&message.to_string()),
 864             log::Level::Info => self.log(&message.to_string()),
 865             log::Level::Debug => self.log(&format!("DEBUG: {}", message)),
 866             log::Level::Trace => self.log(&format!("TRACE: {}", message)),
 867         }
 868     }
 869 }