]> git.proxmox.com Git - proxmox-backup.git/blob - src/server/worker_task.rs
replace file_set_contents with replace_file
[proxmox-backup.git] / src / server / worker_task.rs
1 use failure::*;
2 use lazy_static::lazy_static;
3 use chrono::Local;
4
5 use tokio::sync::oneshot;
6 use futures::*;
7 use std::sync::{Arc, Mutex};
8 use std::collections::HashMap;
9 use std::sync::atomic::{AtomicBool, Ordering};
10 use std::io::{BufRead, BufReader};
11 use std::fs::File;
12 use std::panic::UnwindSafe;
13
14 use serde_json::{json, Value};
15
16 use proxmox::tools::{
17 try_block,
18 fs::{create_path, replace_file, CreateOptions},
19 };
20
21 use super::UPID;
22
23 use crate::tools::FileLogger;
24
25 macro_rules! PROXMOX_BACKUP_VAR_RUN_DIR_M { () => ("/run/proxmox-backup") }
26 macro_rules! PROXMOX_BACKUP_LOG_DIR_M { () => ("/var/log/proxmox-backup") }
27 macro_rules! PROXMOX_BACKUP_TASK_DIR_M { () => (concat!( PROXMOX_BACKUP_LOG_DIR_M!(), "/tasks")) }
28
29 pub const PROXMOX_BACKUP_VAR_RUN_DIR: &str = PROXMOX_BACKUP_VAR_RUN_DIR_M!();
30 pub const PROXMOX_BACKUP_LOG_DIR: &str = PROXMOX_BACKUP_LOG_DIR_M!();
31 pub const PROXMOX_BACKUP_TASK_DIR: &str = PROXMOX_BACKUP_TASK_DIR_M!();
32 pub const PROXMOX_BACKUP_TASK_LOCK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/.active.lock");
33 pub const PROXMOX_BACKUP_ACTIVE_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/active");
34
35 lazy_static! {
36 static ref WORKER_TASK_LIST: Mutex<HashMap<usize, Arc<WorkerTask>>> = Mutex::new(HashMap::new());
37
38 static ref MY_PID: i32 = unsafe { libc::getpid() };
39 static ref MY_PID_PSTART: u64 = proxmox::sys::linux::procfs::read_proc_pid_stat(*MY_PID).unwrap().starttime;
40 }
41
42 /// Test if the task is still running
43 pub fn worker_is_active(upid: &UPID) -> bool {
44
45 if (upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART) {
46 WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id)
47 } else {
48 use proxmox::sys::linux::procfs;
49 procfs::check_process_running_pstart(upid.pid, upid.pstart).is_some()
50 }
51 }
52
53 pub fn create_task_control_socket() -> Result<(), Error> {
54
55 let socketname = format!(
56 "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, *MY_PID);
57
58 let control_future = super::create_control_socket(socketname, |param| {
59 let param = param.as_object()
60 .ok_or_else(|| format_err!("unable to parse parameters (expected json object)"))?;
61 if param.keys().count() != 2 { bail!("wrong number of parameters"); }
62
63 let command = param.get("command")
64 .ok_or_else(|| format_err!("unable to parse parameters (missing command)"))?;
65
66 // this is the only command for now
67 if command != "abort-task" { bail!("got unknown command '{}'", command); }
68
69 let upid_str = param["upid"].as_str()
70 .ok_or_else(|| format_err!("unable to parse parameters (missing upid)"))?;
71
72 let upid = upid_str.parse::<UPID>()?;
73
74 if !((upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART)) {
75 bail!("upid does not belong to this process");
76 }
77
78 let hash = WORKER_TASK_LIST.lock().unwrap();
79 if let Some(ref worker) = hash.get(&upid.task_id) {
80 worker.request_abort();
81 } else {
82 // assume task is already stopped
83 }
84 Ok(Value::Null)
85 })?;
86
87 tokio::spawn(control_future);
88
89 Ok(())
90 }
91
92 pub fn abort_worker_async(upid: UPID) {
93 tokio::spawn(async move {
94 if let Err(err) = abort_worker(upid).await {
95 eprintln!("abort worker failed - {}", err);
96 }
97 });
98 }
99
100 pub fn abort_worker(upid: UPID) -> impl Future<Output = Result<(), Error>> {
101
102 let target_pid = upid.pid;
103
104 let socketname = format!(
105 "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, target_pid);
106
107 let cmd = json!({
108 "command": "abort-task",
109 "upid": upid.to_string(),
110 });
111
112 super::send_command(socketname, cmd).map_ok(|_| ())
113 }
114
115 fn parse_worker_status_line(line: &str) -> Result<(String, UPID, Option<(i64, String)>), Error> {
116
117 let data = line.splitn(3, ' ').collect::<Vec<&str>>();
118
119 let len = data.len();
120
121 match len {
122 1 => Ok((data[0].to_owned(), data[0].parse::<UPID>()?, None)),
123 3 => {
124 let endtime = i64::from_str_radix(data[1], 16)?;
125 Ok((data[0].to_owned(), data[0].parse::<UPID>()?, Some((endtime, data[2].to_owned()))))
126 }
127 _ => bail!("wrong number of components"),
128 }
129 }
130
131 /// Create task log directory with correct permissions
132 pub fn create_task_log_dirs() -> Result<(), Error> {
133
134 try_block!({
135 let (backup_uid, backup_gid) = crate::tools::getpwnam_ugid("backup")?;
136 let opts = CreateOptions::new()
137 .owner(nix::unistd::Uid::from_raw(backup_uid))
138 .group(nix::unistd::Gid::from_raw(backup_gid));
139
140 create_path(PROXMOX_BACKUP_LOG_DIR, None, Some(opts.clone()))?;
141 create_path(PROXMOX_BACKUP_TASK_DIR, None, Some(opts.clone()))?;
142 create_path(PROXMOX_BACKUP_VAR_RUN_DIR, None, Some(opts))?;
143 Ok(())
144 }).map_err(|err: Error| format_err!("unable to create task log dir - {}", err))?;
145
146 Ok(())
147 }
148
149 /// Read exits status from task log file
150 pub fn upid_read_status(upid: &UPID) -> Result<String, Error> {
151 let mut status = String::from("unknown");
152
153 let path = upid.log_path();
154
155 let mut file = File::open(path)?;
156
157 /// speedup - only read tail
158 use std::io::Seek;
159 use std::io::SeekFrom;
160 let _ = file.seek(SeekFrom::End(-8192)); // ignore errors
161
162 let reader = BufReader::new(file);
163
164 for line in reader.lines() {
165 let line = line?;
166
167 let mut iter = line.splitn(2, ": TASK ");
168 if iter.next() == None { continue; }
169 match iter.next() {
170 None => continue,
171 Some(rest) => {
172 if rest == "OK" {
173 status = String::from(rest);
174 } else if rest.starts_with("ERROR: ") {
175 status = String::from(&rest[7..]);
176 }
177 }
178 }
179 }
180
181 Ok(status)
182 }
183
184 /// Task details including parsed UPID
185 ///
186 /// If there is no `state`, the task is still running.
187 #[derive(Debug)]
188 pub struct TaskListInfo {
189 /// The parsed UPID
190 pub upid: UPID,
191 /// UPID string representation
192 pub upid_str: String,
193 /// Task `(endtime, status)` if already finished
194 ///
195 /// The `status` ise iether `unknown`, `OK`, or `ERROR: ...`
196 pub state: Option<(i64, String)>, // endtime, status
197 }
198
199 // atomically read/update the task list, update status of finished tasks
200 // new_upid is added to the list when specified.
201 // Returns a sorted list of known tasks,
202 fn update_active_workers(new_upid: Option<&UPID>) -> Result<Vec<TaskListInfo>, Error> {
203
204 let (backup_uid, backup_gid) = crate::tools::getpwnam_ugid("backup")?;
205 let uid = nix::unistd::Uid::from_raw(backup_uid);
206 let gid = nix::unistd::Gid::from_raw(backup_gid);
207
208 let lock = crate::tools::open_file_locked(PROXMOX_BACKUP_TASK_LOCK_FN, std::time::Duration::new(10, 0))?;
209 nix::unistd::chown(PROXMOX_BACKUP_TASK_LOCK_FN, Some(uid), Some(gid))?;
210
211 let reader = match File::open(PROXMOX_BACKUP_ACTIVE_TASK_FN) {
212 Ok(f) => Some(BufReader::new(f)),
213 Err(err) => {
214 if err.kind() == std::io::ErrorKind::NotFound {
215 None
216 } else {
217 bail!("unable to open active worker {:?} - {}", PROXMOX_BACKUP_ACTIVE_TASK_FN, err);
218 }
219 }
220 };
221
222 let mut active_list = vec![];
223 let mut finish_list = vec![];
224
225 if let Some(lines) = reader.map(|r| r.lines()) {
226
227 for line in lines {
228 let line = line?;
229 match parse_worker_status_line(&line) {
230 Err(err) => bail!("unable to parse active worker status '{}' - {}", line, err),
231 Ok((upid_str, upid, state)) => {
232
233 let running = worker_is_active(&upid);
234
235 if running {
236 active_list.push(TaskListInfo { upid, upid_str, state: None });
237 } else {
238 match state {
239 None => {
240 println!("Detected stoped UPID {}", upid_str);
241 let status = upid_read_status(&upid)
242 .unwrap_or_else(|_| String::from("unknown"));
243 finish_list.push(TaskListInfo {
244 upid, upid_str, state: Some((Local::now().timestamp(), status))
245 });
246 }
247 Some((endtime, status)) => {
248 finish_list.push(TaskListInfo {
249 upid, upid_str, state: Some((endtime, status))
250 })
251 }
252 }
253 }
254 }
255 }
256 }
257 }
258
259 if let Some(upid) = new_upid {
260 active_list.push(TaskListInfo { upid: upid.clone(), upid_str: upid.to_string(), state: None });
261 }
262
263 // assemble list without duplicates
264 // we include all active tasks,
265 // and fill up to 1000 entries with finished tasks
266
267 let max = 1000;
268
269 let mut task_hash = HashMap::new();
270
271 for info in active_list {
272 task_hash.insert(info.upid_str.clone(), info);
273 }
274
275 for info in finish_list {
276 if task_hash.len() > max { break; }
277 if !task_hash.contains_key(&info.upid_str) {
278 task_hash.insert(info.upid_str.clone(), info);
279 }
280 }
281
282 let mut task_list: Vec<TaskListInfo> = vec![];
283 for (_, info) in task_hash { task_list.push(info); }
284
285 task_list.sort_unstable_by(|b, a| { // lastest on top
286 match (&a.state, &b.state) {
287 (Some(s1), Some(s2)) => s1.0.cmp(&s2.0),
288 (Some(_), None) => std::cmp::Ordering::Less,
289 (None, Some(_)) => std::cmp::Ordering::Greater,
290 _ => a.upid.starttime.cmp(&b.upid.starttime),
291 }
292 });
293
294 let mut raw = String::new();
295 for info in &task_list {
296 if let Some((endtime, status)) = &info.state {
297 raw.push_str(&format!("{} {:08X} {}\n", info.upid_str, endtime, status));
298 } else {
299 raw.push_str(&info.upid_str);
300 raw.push('\n');
301 }
302 }
303
304 replace_file(
305 PROXMOX_BACKUP_ACTIVE_TASK_FN,
306 raw.as_bytes(),
307 CreateOptions::new()
308 .owner(uid)
309 .group(gid),
310 )?;
311
312 drop(lock);
313
314 Ok(task_list)
315 }
316
317 /// Returns a sorted list of known tasks
318 ///
319 /// The list is sorted by `(starttime, endtime)` in ascending order
320 pub fn read_task_list() -> Result<Vec<TaskListInfo>, Error> {
321 update_active_workers(None)
322 }
323
324 /// Launch long running worker tasks.
325 ///
326 /// A worker task can either be a whole thread, or a simply tokio
327 /// task/future. Each task can `log()` messages, which are stored
328 /// persistently to files. Task should poll the `abort_requested`
329 /// flag, and stop execution when requested.
330 #[derive(Debug)]
331 pub struct WorkerTask {
332 upid: UPID,
333 data: Mutex<WorkerTaskData>,
334 abort_requested: AtomicBool,
335 }
336
337 impl std::fmt::Display for WorkerTask {
338
339 fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
340 self.upid.fmt(f)
341 }
342 }
343
344 #[derive(Debug)]
345 struct WorkerTaskData {
346 logger: FileLogger,
347 progress: f64, // 0..1
348 pub abort_listeners: Vec<oneshot::Sender<()>>,
349 }
350
351 impl Drop for WorkerTask {
352
353 fn drop(&mut self) {
354 println!("unregister worker");
355 }
356 }
357
358 impl WorkerTask {
359
360 pub fn new(worker_type: &str, worker_id: Option<String>, username: &str, to_stdout: bool) -> Result<Arc<Self>, Error> {
361 println!("register worker");
362
363 let upid = UPID::new(worker_type, worker_id, username)?;
364 let task_id = upid.task_id;
365
366 let mut path = std::path::PathBuf::from(PROXMOX_BACKUP_TASK_DIR);
367
368 path.push(format!("{:02X}", upid.pstart % 256));
369
370 let (backup_uid, backup_gid) = crate::tools::getpwnam_ugid("backup")?;
371 let uid = nix::unistd::Uid::from_raw(backup_uid);
372 let gid = nix::unistd::Gid::from_raw(backup_gid);
373
374 create_path(&path, None, Some(CreateOptions::new().owner(uid).group(gid)))?;
375
376 path.push(upid.to_string());
377
378 println!("FILE: {:?}", path);
379
380 let logger = FileLogger::new(&path, to_stdout)?;
381 nix::unistd::chown(&path, Some(uid), Some(gid))?;
382
383 update_active_workers(Some(&upid))?;
384
385 let worker = Arc::new(Self {
386 upid,
387 abort_requested: AtomicBool::new(false),
388 data: Mutex::new(WorkerTaskData {
389 logger,
390 progress: 0.0,
391 abort_listeners: vec![],
392 }),
393 });
394
395 let mut hash = WORKER_TASK_LIST.lock().unwrap();
396
397 hash.insert(task_id, worker.clone());
398 super::set_worker_count(hash.len());
399
400 Ok(worker)
401 }
402
403 /// Spawn a new tokio task/future.
404 pub fn spawn<F, T>(
405 worker_type: &str,
406 worker_id: Option<String>,
407 username: &str,
408 to_stdout: bool,
409 f: F,
410 ) -> Result<String, Error>
411 where F: Send + 'static + FnOnce(Arc<WorkerTask>) -> T,
412 T: Send + 'static + Future<Output = Result<(), Error>>,
413 {
414 let worker = WorkerTask::new(worker_type, worker_id, username, to_stdout)?;
415 let upid_str = worker.upid.to_string();
416 let f = f(worker.clone());
417 tokio::spawn(async move {
418 let result = f.await;
419 worker.log_result(&result);
420 });
421
422 Ok(upid_str)
423 }
424
425 /// Create a new worker thread.
426 pub fn new_thread<F>(
427 worker_type: &str,
428 worker_id: Option<String>,
429 username: &str,
430 to_stdout: bool,
431 f: F,
432 ) -> Result<String, Error>
433 where F: Send + UnwindSafe + 'static + FnOnce(Arc<WorkerTask>) -> Result<(), Error>
434 {
435 println!("register worker thread");
436
437 let (p, c) = oneshot::channel::<()>();
438
439 let worker = WorkerTask::new(worker_type, worker_id, username, to_stdout)?;
440 let upid_str = worker.upid.to_string();
441
442 let _child = std::thread::spawn(move || {
443 let worker1 = worker.clone();
444 let result = match std::panic::catch_unwind(move || f(worker1)) {
445 Ok(r) => r,
446 Err(panic) => {
447 match panic.downcast::<&str>() {
448 Ok(panic_msg) => {
449 Err(format_err!("worker panicked: {}", panic_msg))
450 }
451 Err(_) => {
452 Err(format_err!("worker panicked: unknown type."))
453 }
454 }
455 }
456 };
457
458 worker.log_result(&result);
459 p.send(()).unwrap();
460 });
461
462 tokio::spawn(c.map(|_| ()));
463
464 Ok(upid_str)
465 }
466
467 /// Log task result, remove task from running list
468 pub fn log_result(&self, result: &Result<(), Error>) {
469
470 if let Err(err) = result {
471 self.log(&format!("TASK ERROR: {}", err));
472 } else {
473 self.log("TASK OK");
474 }
475
476 WORKER_TASK_LIST.lock().unwrap().remove(&self.upid.task_id);
477 let _ = update_active_workers(None);
478 super::set_worker_count(WORKER_TASK_LIST.lock().unwrap().len());
479 }
480
481 /// Log a message.
482 pub fn log<S: AsRef<str>>(&self, msg: S) {
483 let mut data = self.data.lock().unwrap();
484 data.logger.log(msg);
485 }
486
487 /// Set progress indicator
488 pub fn progress(&self, progress: f64) {
489 if progress >= 0.0 && progress <= 1.0 {
490 let mut data = self.data.lock().unwrap();
491 data.progress = progress;
492 } else {
493 // fixme: log!("task '{}': ignoring strange value for progress '{}'", self.upid, progress);
494 }
495 }
496
497 /// Request abort
498 pub fn request_abort(&self) {
499 eprintln!("set abort flag for worker {}", self.upid);
500 self.abort_requested.store(true, Ordering::SeqCst);
501 // noitify listeners
502 let mut data = self.data.lock().unwrap();
503 loop {
504 match data.abort_listeners.pop() {
505 None => { break; },
506 Some(ch) => {
507 let _ = ch.send(()); // ignore erros here
508 },
509 }
510 }
511 }
512
513 /// Test if abort was requested.
514 pub fn abort_requested(&self) -> bool {
515 self.abort_requested.load(Ordering::SeqCst)
516 }
517
518 /// Fail if abort was requested.
519 pub fn fail_on_abort(&self) -> Result<(), Error> {
520 if self.abort_requested() {
521 bail!("task '{}': abort requested - aborting task", self.upid);
522 }
523 Ok(())
524 }
525
526 /// Get a future which resolves on task abort
527 pub fn abort_future(&self) -> oneshot::Receiver<()> {
528 let (tx, rx) = oneshot::channel::<()>();
529
530 let mut data = self.data.lock().unwrap();
531 if self.abort_requested() {
532 let _ = tx.send(());
533 } else {
534 data.abort_listeners.push(tx);
535 }
536 rx
537 }
538 }