]>
Commit | Line | Data |
---|---|---|
1 | use std::sync::{Mutex, Arc}; | |
2 | use std::path::{Path, PathBuf}; | |
3 | use std::os::unix::io::AsRawFd; | |
4 | ||
5 | use anyhow::{bail, format_err, Error}; | |
6 | use futures::*; | |
7 | use http::request::Parts; | |
8 | use http::Response; | |
9 | use hyper::{Body, StatusCode}; | |
10 | use hyper::header; | |
11 | use url::form_urlencoded; | |
12 | ||
13 | use openssl::ssl::{SslMethod, SslAcceptor, SslFiletype}; | |
14 | use tokio_stream::wrappers::ReceiverStream; | |
15 | use serde_json::{json, Value}; | |
16 | ||
17 | use proxmox::try_block; | |
18 | use proxmox::api::RpcEnvironmentType; | |
19 | use proxmox::sys::linux::socket::set_tcp_keepalive; | |
20 | use proxmox::tools::fs::CreateOptions; | |
21 | ||
22 | use proxmox_rest_server::{ApiConfig, RestServer}; | |
23 | ||
24 | use proxmox_backup::{ | |
25 | backup::DataStore, | |
26 | server::{ | |
27 | auth::default_api_auth, | |
28 | WorkerTask, | |
29 | jobstate::{ | |
30 | self, | |
31 | Job, | |
32 | }, | |
33 | rotate_task_log_archive, | |
34 | }, | |
35 | }; | |
36 | ||
37 | use pbs_buildcfg::configdir; | |
38 | use proxmox_systemd::time::{compute_next_event, parse_calendar_event}; | |
39 | use pbs_tools::logrotate::LogRotate; | |
40 | ||
41 | use pbs_api_types::{ | |
42 | Authid, TapeBackupJobConfig, VerificationJobConfig, SyncJobConfig, DataStoreConfig, | |
43 | PruneOptions, | |
44 | }; | |
45 | ||
46 | use proxmox_rest_server::daemon; | |
47 | ||
48 | use proxmox_backup::server; | |
49 | use proxmox_backup::auth_helpers::*; | |
50 | use proxmox_backup::tools::{ | |
51 | PROXMOX_BACKUP_TCP_KEEPALIVE_TIME, | |
52 | disks::{ | |
53 | DiskManage, | |
54 | zfs_pool_stats, | |
55 | get_pool_from_dataset, | |
56 | }, | |
57 | }; | |
58 | ||
59 | ||
60 | use proxmox_backup::api2::pull::do_sync_job; | |
61 | use proxmox_backup::api2::tape::backup::do_tape_backup_job; | |
62 | use proxmox_backup::server::do_verification_job; | |
63 | use proxmox_backup::server::do_prune_job; | |
64 | ||
65 | fn main() -> Result<(), Error> { | |
66 | proxmox_backup::tools::setup_safe_path_env(); | |
67 | ||
68 | let backup_uid = pbs_config::backup_user()?.uid; | |
69 | let backup_gid = pbs_config::backup_group()?.gid; | |
70 | let running_uid = nix::unistd::Uid::effective(); | |
71 | let running_gid = nix::unistd::Gid::effective(); | |
72 | ||
73 | if running_uid != backup_uid || running_gid != backup_gid { | |
74 | bail!("proxy not running as backup user or group (got uid {} gid {})", running_uid, running_gid); | |
75 | } | |
76 | ||
77 | pbs_runtime::main(run()) | |
78 | } | |
79 | ||
80 | fn get_index( | |
81 | auth_id: Option<String>, | |
82 | language: Option<String>, | |
83 | api: &ApiConfig, | |
84 | parts: Parts, | |
85 | ) -> Response<Body> { | |
86 | ||
87 | let (userid, csrf_token) = match auth_id { | |
88 | Some(auth_id) => { | |
89 | let auth_id = auth_id.parse::<Authid>(); | |
90 | match auth_id { | |
91 | Ok(auth_id) if !auth_id.is_token() => { | |
92 | let userid = auth_id.user().clone(); | |
93 | let new_csrf_token = assemble_csrf_prevention_token(csrf_secret(), &userid); | |
94 | (Some(userid), Some(new_csrf_token)) | |
95 | } | |
96 | _ => (None, None) | |
97 | } | |
98 | } | |
99 | None => (None, None), | |
100 | }; | |
101 | ||
102 | let nodename = proxmox::tools::nodename(); | |
103 | let user = userid.as_ref().map(|u| u.as_str()).unwrap_or(""); | |
104 | ||
105 | let csrf_token = csrf_token.unwrap_or_else(|| String::from("")); | |
106 | ||
107 | let mut debug = false; | |
108 | let mut template_file = "index"; | |
109 | ||
110 | if let Some(query_str) = parts.uri.query() { | |
111 | for (k, v) in form_urlencoded::parse(query_str.as_bytes()).into_owned() { | |
112 | if k == "debug" && v != "0" && v != "false" { | |
113 | debug = true; | |
114 | } else if k == "console" { | |
115 | template_file = "console"; | |
116 | } | |
117 | } | |
118 | } | |
119 | ||
120 | let mut lang = String::from(""); | |
121 | if let Some(language) = language { | |
122 | if Path::new(&format!("/usr/share/pbs-i18n/pbs-lang-{}.js", language)).exists() { | |
123 | lang = language; | |
124 | } | |
125 | } | |
126 | ||
127 | let data = json!({ | |
128 | "NodeName": nodename, | |
129 | "UserName": user, | |
130 | "CSRFPreventionToken": csrf_token, | |
131 | "language": lang, | |
132 | "debug": debug, | |
133 | }); | |
134 | ||
135 | let (ct, index) = match api.render_template(template_file, &data) { | |
136 | Ok(index) => ("text/html", index), | |
137 | Err(err) => ("text/plain", format!("Error rendering template: {}", err)), | |
138 | }; | |
139 | ||
140 | let mut resp = Response::builder() | |
141 | .status(StatusCode::OK) | |
142 | .header(header::CONTENT_TYPE, ct) | |
143 | .body(index.into()) | |
144 | .unwrap(); | |
145 | ||
146 | if let Some(userid) = userid { | |
147 | resp.extensions_mut().insert(Authid::from((userid, None))); | |
148 | } | |
149 | ||
150 | resp | |
151 | } | |
152 | ||
153 | async fn run() -> Result<(), Error> { | |
154 | if let Err(err) = syslog::init( | |
155 | syslog::Facility::LOG_DAEMON, | |
156 | log::LevelFilter::Info, | |
157 | Some("proxmox-backup-proxy")) { | |
158 | bail!("unable to inititialize syslog - {}", err); | |
159 | } | |
160 | ||
161 | // Note: To debug early connection error use | |
162 | // PROXMOX_DEBUG=1 ./target/release/proxmox-backup-proxy | |
163 | let debug = std::env::var("PROXMOX_DEBUG").is_ok(); | |
164 | ||
165 | let _ = public_auth_key(); // load with lazy_static | |
166 | let _ = csrf_secret(); // load with lazy_static | |
167 | ||
168 | let mut config = ApiConfig::new( | |
169 | pbs_buildcfg::JS_DIR, | |
170 | &proxmox_backup::api2::ROUTER, | |
171 | RpcEnvironmentType::PUBLIC, | |
172 | default_api_auth(), | |
173 | get_index, | |
174 | )?; | |
175 | ||
176 | config.add_alias("novnc", "/usr/share/novnc-pve"); | |
177 | config.add_alias("extjs", "/usr/share/javascript/extjs"); | |
178 | config.add_alias("qrcodejs", "/usr/share/javascript/qrcodejs"); | |
179 | config.add_alias("fontawesome", "/usr/share/fonts-font-awesome"); | |
180 | config.add_alias("xtermjs", "/usr/share/pve-xtermjs"); | |
181 | config.add_alias("locale", "/usr/share/pbs-i18n"); | |
182 | config.add_alias("widgettoolkit", "/usr/share/javascript/proxmox-widget-toolkit"); | |
183 | config.add_alias("docs", "/usr/share/doc/proxmox-backup/html"); | |
184 | ||
185 | let mut indexpath = PathBuf::from(pbs_buildcfg::JS_DIR); | |
186 | indexpath.push("index.hbs"); | |
187 | config.register_template("index", &indexpath)?; | |
188 | config.register_template("console", "/usr/share/pve-xtermjs/index.html.hbs")?; | |
189 | ||
190 | let backup_user = pbs_config::backup_user()?; | |
191 | let mut commando_sock = proxmox_rest_server::CommandoSocket::new(crate::server::our_ctrl_sock(), backup_user.gid); | |
192 | ||
193 | let dir_opts = CreateOptions::new().owner(backup_user.uid).group(backup_user.gid); | |
194 | let file_opts = CreateOptions::new().owner(backup_user.uid).group(backup_user.gid); | |
195 | ||
196 | config.enable_file_log( | |
197 | pbs_buildcfg::API_ACCESS_LOG_FN, | |
198 | Some(dir_opts.clone()), | |
199 | Some(file_opts.clone()), | |
200 | &mut commando_sock, | |
201 | )?; | |
202 | ||
203 | config.enable_auth_log( | |
204 | pbs_buildcfg::API_AUTH_LOG_FN, | |
205 | Some(dir_opts), | |
206 | Some(file_opts), | |
207 | &mut commando_sock, | |
208 | )?; | |
209 | ||
210 | let rest_server = RestServer::new(config); | |
211 | ||
212 | //openssl req -x509 -newkey rsa:4096 -keyout /etc/proxmox-backup/proxy.key -out /etc/proxmox-backup/proxy.pem -nodes | |
213 | ||
214 | // we build the initial acceptor here as we cannot start if this fails | |
215 | let acceptor = make_tls_acceptor()?; | |
216 | let acceptor = Arc::new(Mutex::new(acceptor)); | |
217 | ||
218 | // to renew the acceptor we just add a command-socket handler | |
219 | commando_sock.register_command( | |
220 | "reload-certificate".to_string(), | |
221 | { | |
222 | let acceptor = Arc::clone(&acceptor); | |
223 | move |_value| -> Result<_, Error> { | |
224 | log::info!("reloading certificate"); | |
225 | match make_tls_acceptor() { | |
226 | Err(err) => log::error!("error reloading certificate: {}", err), | |
227 | Ok(new_acceptor) => { | |
228 | let mut guard = acceptor.lock().unwrap(); | |
229 | *guard = new_acceptor; | |
230 | } | |
231 | } | |
232 | Ok(Value::Null) | |
233 | } | |
234 | }, | |
235 | )?; | |
236 | ||
237 | // to remove references for not configured datastores | |
238 | commando_sock.register_command( | |
239 | "datastore-removed".to_string(), | |
240 | |_value| { | |
241 | if let Err(err) = proxmox_backup::backup::DataStore::remove_unused_datastores() { | |
242 | log::error!("could not refresh datastores: {}", err); | |
243 | } | |
244 | Ok(Value::Null) | |
245 | } | |
246 | )?; | |
247 | ||
248 | let server = daemon::create_daemon( | |
249 | ([0,0,0,0,0,0,0,0], 8007).into(), | |
250 | move |listener, ready| { | |
251 | ||
252 | let connections = accept_connections(listener, acceptor, debug); | |
253 | let connections = hyper::server::accept::from_stream(ReceiverStream::new(connections)); | |
254 | ||
255 | Ok(ready | |
256 | .and_then(|_| hyper::Server::builder(connections) | |
257 | .serve(rest_server) | |
258 | .with_graceful_shutdown(proxmox_rest_server::shutdown_future()) | |
259 | .map_err(Error::from) | |
260 | ) | |
261 | .map_err(|err| eprintln!("server error: {}", err)) | |
262 | .map(|_| ()) | |
263 | ) | |
264 | }, | |
265 | "proxmox-backup-proxy.service", | |
266 | ); | |
267 | ||
268 | server::write_pid(pbs_buildcfg::PROXMOX_BACKUP_PROXY_PID_FN)?; | |
269 | daemon::systemd_notify(daemon::SystemdNotify::Ready)?; | |
270 | ||
271 | let init_result: Result<(), Error> = try_block!({ | |
272 | server::register_task_control_commands(&mut commando_sock)?; | |
273 | commando_sock.spawn()?; | |
274 | proxmox_rest_server::server_state_init()?; | |
275 | Ok(()) | |
276 | }); | |
277 | ||
278 | if let Err(err) = init_result { | |
279 | bail!("unable to start daemon - {}", err); | |
280 | } | |
281 | ||
282 | start_task_scheduler(); | |
283 | start_stat_generator(); | |
284 | ||
285 | server.await?; | |
286 | log::info!("server shutting down, waiting for active workers to complete"); | |
287 | proxmox_rest_server::last_worker_future().await?; | |
288 | log::info!("done - exit server"); | |
289 | ||
290 | Ok(()) | |
291 | } | |
292 | ||
293 | fn make_tls_acceptor() -> Result<SslAcceptor, Error> { | |
294 | let key_path = configdir!("/proxy.key"); | |
295 | let cert_path = configdir!("/proxy.pem"); | |
296 | ||
297 | let mut acceptor = SslAcceptor::mozilla_intermediate_v5(SslMethod::tls()).unwrap(); | |
298 | acceptor.set_private_key_file(key_path, SslFiletype::PEM) | |
299 | .map_err(|err| format_err!("unable to read proxy key {} - {}", key_path, err))?; | |
300 | acceptor.set_certificate_chain_file(cert_path) | |
301 | .map_err(|err| format_err!("unable to read proxy cert {} - {}", cert_path, err))?; | |
302 | acceptor.check_private_key().unwrap(); | |
303 | ||
304 | Ok(acceptor.build()) | |
305 | } | |
306 | ||
307 | type ClientStreamResult = | |
308 | Result<std::pin::Pin<Box<tokio_openssl::SslStream<tokio::net::TcpStream>>>, Error>; | |
309 | const MAX_PENDING_ACCEPTS: usize = 1024; | |
310 | ||
311 | fn accept_connections( | |
312 | listener: tokio::net::TcpListener, | |
313 | acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>, | |
314 | debug: bool, | |
315 | ) -> tokio::sync::mpsc::Receiver<ClientStreamResult> { | |
316 | ||
317 | let (sender, receiver) = tokio::sync::mpsc::channel(MAX_PENDING_ACCEPTS); | |
318 | ||
319 | tokio::spawn(accept_connection(listener, acceptor, debug, sender)); | |
320 | ||
321 | receiver | |
322 | } | |
323 | ||
324 | async fn accept_connection( | |
325 | listener: tokio::net::TcpListener, | |
326 | acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>, | |
327 | debug: bool, | |
328 | sender: tokio::sync::mpsc::Sender<ClientStreamResult>, | |
329 | ) { | |
330 | let accept_counter = Arc::new(()); | |
331 | ||
332 | loop { | |
333 | let (sock, _addr) = match listener.accept().await { | |
334 | Ok(conn) => conn, | |
335 | Err(err) => { | |
336 | eprintln!("error accepting tcp connection: {}", err); | |
337 | continue; | |
338 | } | |
339 | }; | |
340 | ||
341 | sock.set_nodelay(true).unwrap(); | |
342 | let _ = set_tcp_keepalive(sock.as_raw_fd(), PROXMOX_BACKUP_TCP_KEEPALIVE_TIME); | |
343 | ||
344 | let ssl = { // limit acceptor_guard scope | |
345 | // Acceptor can be reloaded using the command socket "reload-certificate" command | |
346 | let acceptor_guard = acceptor.lock().unwrap(); | |
347 | ||
348 | match openssl::ssl::Ssl::new(acceptor_guard.context()) { | |
349 | Ok(ssl) => ssl, | |
350 | Err(err) => { | |
351 | eprintln!("failed to create Ssl object from Acceptor context - {}", err); | |
352 | continue; | |
353 | }, | |
354 | } | |
355 | }; | |
356 | ||
357 | let stream = match tokio_openssl::SslStream::new(ssl, sock) { | |
358 | Ok(stream) => stream, | |
359 | Err(err) => { | |
360 | eprintln!("failed to create SslStream using ssl and connection socket - {}", err); | |
361 | continue; | |
362 | }, | |
363 | }; | |
364 | ||
365 | let mut stream = Box::pin(stream); | |
366 | let sender = sender.clone(); | |
367 | ||
368 | if Arc::strong_count(&accept_counter) > MAX_PENDING_ACCEPTS { | |
369 | eprintln!("connection rejected - to many open connections"); | |
370 | continue; | |
371 | } | |
372 | ||
373 | let accept_counter = Arc::clone(&accept_counter); | |
374 | tokio::spawn(async move { | |
375 | let accept_future = tokio::time::timeout( | |
376 | Duration::new(10, 0), stream.as_mut().accept()); | |
377 | ||
378 | let result = accept_future.await; | |
379 | ||
380 | match result { | |
381 | Ok(Ok(())) => { | |
382 | if sender.send(Ok(stream)).await.is_err() && debug { | |
383 | eprintln!("detect closed connection channel"); | |
384 | } | |
385 | } | |
386 | Ok(Err(err)) => { | |
387 | if debug { | |
388 | eprintln!("https handshake failed - {}", err); | |
389 | } | |
390 | } | |
391 | Err(_) => { | |
392 | if debug { | |
393 | eprintln!("https handshake timeout"); | |
394 | } | |
395 | } | |
396 | } | |
397 | ||
398 | drop(accept_counter); // decrease reference count | |
399 | }); | |
400 | } | |
401 | } | |
402 | ||
403 | fn start_stat_generator() { | |
404 | let abort_future = proxmox_rest_server::shutdown_future(); | |
405 | let future = Box::pin(run_stat_generator()); | |
406 | let task = futures::future::select(future, abort_future); | |
407 | tokio::spawn(task.map(|_| ())); | |
408 | } | |
409 | ||
410 | fn start_task_scheduler() { | |
411 | let abort_future = proxmox_rest_server::shutdown_future(); | |
412 | let future = Box::pin(run_task_scheduler()); | |
413 | let task = futures::future::select(future, abort_future); | |
414 | tokio::spawn(task.map(|_| ())); | |
415 | } | |
416 | ||
417 | use std::time::{SystemTime, Instant, Duration, UNIX_EPOCH}; | |
418 | ||
419 | fn next_minute() -> Result<Instant, Error> { | |
420 | let now = SystemTime::now(); | |
421 | let epoch_now = now.duration_since(UNIX_EPOCH)?; | |
422 | let epoch_next = Duration::from_secs((epoch_now.as_secs()/60 + 1)*60); | |
423 | Ok(Instant::now() + epoch_next - epoch_now) | |
424 | } | |
425 | ||
426 | async fn run_task_scheduler() { | |
427 | ||
428 | let mut count: usize = 0; | |
429 | ||
430 | loop { | |
431 | count += 1; | |
432 | ||
433 | let delay_target = match next_minute() { // try to run very minute | |
434 | Ok(d) => d, | |
435 | Err(err) => { | |
436 | eprintln!("task scheduler: compute next minute failed - {}", err); | |
437 | tokio::time::sleep_until(tokio::time::Instant::from_std(Instant::now() + Duration::from_secs(60))).await; | |
438 | continue; | |
439 | } | |
440 | }; | |
441 | ||
442 | if count > 2 { // wait 1..2 minutes before starting | |
443 | match schedule_tasks().catch_unwind().await { | |
444 | Err(panic) => { | |
445 | match panic.downcast::<&str>() { | |
446 | Ok(msg) => { | |
447 | eprintln!("task scheduler panic: {}", msg); | |
448 | } | |
449 | Err(_) => { | |
450 | eprintln!("task scheduler panic - unknown type"); | |
451 | } | |
452 | } | |
453 | } | |
454 | Ok(Err(err)) => { | |
455 | eprintln!("task scheduler failed - {:?}", err); | |
456 | } | |
457 | Ok(Ok(_)) => {} | |
458 | } | |
459 | } | |
460 | ||
461 | tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await; | |
462 | } | |
463 | } | |
464 | ||
465 | async fn schedule_tasks() -> Result<(), Error> { | |
466 | ||
467 | schedule_datastore_garbage_collection().await; | |
468 | schedule_datastore_prune().await; | |
469 | schedule_datastore_sync_jobs().await; | |
470 | schedule_datastore_verify_jobs().await; | |
471 | schedule_tape_backup_jobs().await; | |
472 | schedule_task_log_rotate().await; | |
473 | ||
474 | Ok(()) | |
475 | } | |
476 | ||
477 | async fn schedule_datastore_garbage_collection() { | |
478 | ||
479 | let config = match pbs_config::datastore::config() { | |
480 | Err(err) => { | |
481 | eprintln!("unable to read datastore config - {}", err); | |
482 | return; | |
483 | } | |
484 | Ok((config, _digest)) => config, | |
485 | }; | |
486 | ||
487 | for (store, (_, store_config)) in config.sections { | |
488 | let datastore = match DataStore::lookup_datastore(&store) { | |
489 | Ok(datastore) => datastore, | |
490 | Err(err) => { | |
491 | eprintln!("lookup_datastore failed - {}", err); | |
492 | continue; | |
493 | } | |
494 | }; | |
495 | ||
496 | let store_config: DataStoreConfig = match serde_json::from_value(store_config) { | |
497 | Ok(c) => c, | |
498 | Err(err) => { | |
499 | eprintln!("datastore config from_value failed - {}", err); | |
500 | continue; | |
501 | } | |
502 | }; | |
503 | ||
504 | let event_str = match store_config.gc_schedule { | |
505 | Some(event_str) => event_str, | |
506 | None => continue, | |
507 | }; | |
508 | ||
509 | let event = match parse_calendar_event(&event_str) { | |
510 | Ok(event) => event, | |
511 | Err(err) => { | |
512 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
513 | continue; | |
514 | } | |
515 | }; | |
516 | ||
517 | if datastore.garbage_collection_running() { continue; } | |
518 | ||
519 | let worker_type = "garbage_collection"; | |
520 | ||
521 | let last = match jobstate::last_run_time(worker_type, &store) { | |
522 | Ok(time) => time, | |
523 | Err(err) => { | |
524 | eprintln!("could not get last run time of {} {}: {}", worker_type, store, err); | |
525 | continue; | |
526 | } | |
527 | }; | |
528 | ||
529 | let next = match compute_next_event(&event, last, false) { | |
530 | Ok(Some(next)) => next, | |
531 | Ok(None) => continue, | |
532 | Err(err) => { | |
533 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
534 | continue; | |
535 | } | |
536 | }; | |
537 | ||
538 | let now = proxmox::tools::time::epoch_i64(); | |
539 | ||
540 | if next > now { continue; } | |
541 | ||
542 | let job = match Job::new(worker_type, &store) { | |
543 | Ok(job) => job, | |
544 | Err(_) => continue, // could not get lock | |
545 | }; | |
546 | ||
547 | let auth_id = Authid::root_auth_id(); | |
548 | ||
549 | if let Err(err) = crate::server::do_garbage_collection_job(job, datastore, auth_id, Some(event_str), false) { | |
550 | eprintln!("unable to start garbage collection job on datastore {} - {}", store, err); | |
551 | } | |
552 | } | |
553 | } | |
554 | ||
555 | async fn schedule_datastore_prune() { | |
556 | ||
557 | let config = match pbs_config::datastore::config() { | |
558 | Err(err) => { | |
559 | eprintln!("unable to read datastore config - {}", err); | |
560 | return; | |
561 | } | |
562 | Ok((config, _digest)) => config, | |
563 | }; | |
564 | ||
565 | for (store, (_, store_config)) in config.sections { | |
566 | ||
567 | let store_config: DataStoreConfig = match serde_json::from_value(store_config) { | |
568 | Ok(c) => c, | |
569 | Err(err) => { | |
570 | eprintln!("datastore '{}' config from_value failed - {}", store, err); | |
571 | continue; | |
572 | } | |
573 | }; | |
574 | ||
575 | let event_str = match store_config.prune_schedule { | |
576 | Some(event_str) => event_str, | |
577 | None => continue, | |
578 | }; | |
579 | ||
580 | let prune_options = PruneOptions { | |
581 | keep_last: store_config.keep_last, | |
582 | keep_hourly: store_config.keep_hourly, | |
583 | keep_daily: store_config.keep_daily, | |
584 | keep_weekly: store_config.keep_weekly, | |
585 | keep_monthly: store_config.keep_monthly, | |
586 | keep_yearly: store_config.keep_yearly, | |
587 | }; | |
588 | ||
589 | if !pbs_datastore::prune::keeps_something(&prune_options) { // no prune settings - keep all | |
590 | continue; | |
591 | } | |
592 | ||
593 | let worker_type = "prune"; | |
594 | if check_schedule(worker_type, &event_str, &store) { | |
595 | let job = match Job::new(worker_type, &store) { | |
596 | Ok(job) => job, | |
597 | Err(_) => continue, // could not get lock | |
598 | }; | |
599 | ||
600 | let auth_id = Authid::root_auth_id().clone(); | |
601 | if let Err(err) = do_prune_job(job, prune_options, store.clone(), &auth_id, Some(event_str)) { | |
602 | eprintln!("unable to start datastore prune job {} - {}", &store, err); | |
603 | } | |
604 | }; | |
605 | } | |
606 | } | |
607 | ||
608 | async fn schedule_datastore_sync_jobs() { | |
609 | ||
610 | ||
611 | let config = match pbs_config::sync::config() { | |
612 | Err(err) => { | |
613 | eprintln!("unable to read sync job config - {}", err); | |
614 | return; | |
615 | } | |
616 | Ok((config, _digest)) => config, | |
617 | }; | |
618 | ||
619 | for (job_id, (_, job_config)) in config.sections { | |
620 | let job_config: SyncJobConfig = match serde_json::from_value(job_config) { | |
621 | Ok(c) => c, | |
622 | Err(err) => { | |
623 | eprintln!("sync job config from_value failed - {}", err); | |
624 | continue; | |
625 | } | |
626 | }; | |
627 | ||
628 | let event_str = match job_config.schedule { | |
629 | Some(ref event_str) => event_str.clone(), | |
630 | None => continue, | |
631 | }; | |
632 | ||
633 | let worker_type = "syncjob"; | |
634 | if check_schedule(worker_type, &event_str, &job_id) { | |
635 | let job = match Job::new(worker_type, &job_id) { | |
636 | Ok(job) => job, | |
637 | Err(_) => continue, // could not get lock | |
638 | }; | |
639 | ||
640 | let auth_id = Authid::root_auth_id().clone(); | |
641 | if let Err(err) = do_sync_job(job, job_config, &auth_id, Some(event_str), false) { | |
642 | eprintln!("unable to start datastore sync job {} - {}", &job_id, err); | |
643 | } | |
644 | }; | |
645 | } | |
646 | } | |
647 | ||
648 | async fn schedule_datastore_verify_jobs() { | |
649 | ||
650 | let config = match pbs_config::verify::config() { | |
651 | Err(err) => { | |
652 | eprintln!("unable to read verification job config - {}", err); | |
653 | return; | |
654 | } | |
655 | Ok((config, _digest)) => config, | |
656 | }; | |
657 | for (job_id, (_, job_config)) in config.sections { | |
658 | let job_config: VerificationJobConfig = match serde_json::from_value(job_config) { | |
659 | Ok(c) => c, | |
660 | Err(err) => { | |
661 | eprintln!("verification job config from_value failed - {}", err); | |
662 | continue; | |
663 | } | |
664 | }; | |
665 | let event_str = match job_config.schedule { | |
666 | Some(ref event_str) => event_str.clone(), | |
667 | None => continue, | |
668 | }; | |
669 | ||
670 | let worker_type = "verificationjob"; | |
671 | let auth_id = Authid::root_auth_id().clone(); | |
672 | if check_schedule(worker_type, &event_str, &job_id) { | |
673 | let job = match Job::new(&worker_type, &job_id) { | |
674 | Ok(job) => job, | |
675 | Err(_) => continue, // could not get lock | |
676 | }; | |
677 | if let Err(err) = do_verification_job(job, job_config, &auth_id, Some(event_str), false) { | |
678 | eprintln!("unable to start datastore verification job {} - {}", &job_id, err); | |
679 | } | |
680 | }; | |
681 | } | |
682 | } | |
683 | ||
684 | async fn schedule_tape_backup_jobs() { | |
685 | ||
686 | let config = match pbs_config::tape_job::config() { | |
687 | Err(err) => { | |
688 | eprintln!("unable to read tape job config - {}", err); | |
689 | return; | |
690 | } | |
691 | Ok((config, _digest)) => config, | |
692 | }; | |
693 | for (job_id, (_, job_config)) in config.sections { | |
694 | let job_config: TapeBackupJobConfig = match serde_json::from_value(job_config) { | |
695 | Ok(c) => c, | |
696 | Err(err) => { | |
697 | eprintln!("tape backup job config from_value failed - {}", err); | |
698 | continue; | |
699 | } | |
700 | }; | |
701 | let event_str = match job_config.schedule { | |
702 | Some(ref event_str) => event_str.clone(), | |
703 | None => continue, | |
704 | }; | |
705 | ||
706 | let worker_type = "tape-backup-job"; | |
707 | let auth_id = Authid::root_auth_id().clone(); | |
708 | if check_schedule(worker_type, &event_str, &job_id) { | |
709 | let job = match Job::new(&worker_type, &job_id) { | |
710 | Ok(job) => job, | |
711 | Err(_) => continue, // could not get lock | |
712 | }; | |
713 | if let Err(err) = do_tape_backup_job(job, job_config.setup, &auth_id, Some(event_str), false) { | |
714 | eprintln!("unable to start tape backup job {} - {}", &job_id, err); | |
715 | } | |
716 | }; | |
717 | } | |
718 | } | |
719 | ||
720 | ||
721 | async fn schedule_task_log_rotate() { | |
722 | ||
723 | let worker_type = "logrotate"; | |
724 | let job_id = "access-log_and_task-archive"; | |
725 | ||
726 | // schedule daily at 00:00 like normal logrotate | |
727 | let schedule = "00:00"; | |
728 | ||
729 | if !check_schedule(worker_type, schedule, job_id) { | |
730 | // if we never ran the rotation, schedule instantly | |
731 | match jobstate::JobState::load(worker_type, job_id) { | |
732 | Ok(state) => match state { | |
733 | jobstate::JobState::Created { .. } => {}, | |
734 | _ => return, | |
735 | }, | |
736 | _ => return, | |
737 | } | |
738 | } | |
739 | ||
740 | let mut job = match Job::new(worker_type, job_id) { | |
741 | Ok(job) => job, | |
742 | Err(_) => return, // could not get lock | |
743 | }; | |
744 | ||
745 | if let Err(err) = WorkerTask::new_thread( | |
746 | worker_type, | |
747 | None, | |
748 | Authid::root_auth_id().clone(), | |
749 | false, | |
750 | move |worker| { | |
751 | job.start(&worker.upid().to_string())?; | |
752 | worker.log("starting task log rotation".to_string()); | |
753 | ||
754 | let result = try_block!({ | |
755 | let max_size = 512 * 1024 - 1; // an entry has ~ 100b, so > 5000 entries/file | |
756 | let max_files = 20; // times twenty files gives > 100000 task entries | |
757 | let has_rotated = rotate_task_log_archive(max_size, true, Some(max_files))?; | |
758 | if has_rotated { | |
759 | worker.log("task log archive was rotated".to_string()); | |
760 | } else { | |
761 | worker.log("task log archive was not rotated".to_string()); | |
762 | } | |
763 | ||
764 | let max_size = 32 * 1024 * 1024 - 1; | |
765 | let max_files = 14; | |
766 | let mut logrotate = LogRotate::new(pbs_buildcfg::API_ACCESS_LOG_FN, true) | |
767 | .ok_or_else(|| format_err!("could not get API access log file names"))?; | |
768 | ||
769 | if logrotate.rotate(max_size, None, Some(max_files))? { | |
770 | println!("rotated access log, telling daemons to re-open log file"); | |
771 | pbs_runtime::block_on(command_reopen_access_logfiles())?; | |
772 | worker.log("API access log was rotated".to_string()); | |
773 | } else { | |
774 | worker.log("API access log was not rotated".to_string()); | |
775 | } | |
776 | ||
777 | let mut logrotate = LogRotate::new(pbs_buildcfg::API_AUTH_LOG_FN, true) | |
778 | .ok_or_else(|| format_err!("could not get API auth log file names"))?; | |
779 | ||
780 | if logrotate.rotate(max_size, None, Some(max_files))? { | |
781 | println!("rotated auth log, telling daemons to re-open log file"); | |
782 | pbs_runtime::block_on(command_reopen_auth_logfiles())?; | |
783 | worker.log("API authentication log was rotated".to_string()); | |
784 | } else { | |
785 | worker.log("API authentication log was not rotated".to_string()); | |
786 | } | |
787 | ||
788 | Ok(()) | |
789 | }); | |
790 | ||
791 | let status = worker.create_state(&result); | |
792 | ||
793 | if let Err(err) = job.finish(status) { | |
794 | eprintln!("could not finish job state for {}: {}", worker_type, err); | |
795 | } | |
796 | ||
797 | result | |
798 | }, | |
799 | ) { | |
800 | eprintln!("unable to start task log rotation: {}", err); | |
801 | } | |
802 | ||
803 | } | |
804 | ||
805 | async fn command_reopen_access_logfiles() -> Result<(), Error> { | |
806 | // only care about the most recent daemon instance for each, proxy & api, as other older ones | |
807 | // should not respond to new requests anyway, but only finish their current one and then exit. | |
808 | let sock = crate::server::our_ctrl_sock(); | |
809 | let f1 = proxmox_rest_server::send_command(sock, "{\"command\":\"api-access-log-reopen\"}\n"); | |
810 | ||
811 | let pid = crate::server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?; | |
812 | let sock = crate::server::ctrl_sock_from_pid(pid); | |
813 | let f2 = proxmox_rest_server::send_command(sock, "{\"command\":\"api-access-log-reopen\"}\n"); | |
814 | ||
815 | match futures::join!(f1, f2) { | |
816 | (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)), | |
817 | (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)), | |
818 | (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)), | |
819 | _ => Ok(()), | |
820 | } | |
821 | } | |
822 | ||
823 | async fn command_reopen_auth_logfiles() -> Result<(), Error> { | |
824 | // only care about the most recent daemon instance for each, proxy & api, as other older ones | |
825 | // should not respond to new requests anyway, but only finish their current one and then exit. | |
826 | let sock = crate::server::our_ctrl_sock(); | |
827 | let f1 = proxmox_rest_server::send_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n"); | |
828 | ||
829 | let pid = crate::server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?; | |
830 | let sock = crate::server::ctrl_sock_from_pid(pid); | |
831 | let f2 = proxmox_rest_server::send_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n"); | |
832 | ||
833 | match futures::join!(f1, f2) { | |
834 | (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)), | |
835 | (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)), | |
836 | (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)), | |
837 | _ => Ok(()), | |
838 | } | |
839 | } | |
840 | ||
841 | async fn run_stat_generator() { | |
842 | ||
843 | let mut count = 0; | |
844 | loop { | |
845 | count += 1; | |
846 | let save = if count >= 6 { count = 0; true } else { false }; | |
847 | ||
848 | let delay_target = Instant::now() + Duration::from_secs(10); | |
849 | ||
850 | generate_host_stats(save).await; | |
851 | ||
852 | tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await; | |
853 | ||
854 | } | |
855 | ||
856 | } | |
857 | ||
858 | fn rrd_update_gauge(name: &str, value: f64, save: bool) { | |
859 | use proxmox_backup::rrd; | |
860 | if let Err(err) = rrd::update_value(name, value, rrd::DST::Gauge, save) { | |
861 | eprintln!("rrd::update_value '{}' failed - {}", name, err); | |
862 | } | |
863 | } | |
864 | ||
865 | fn rrd_update_derive(name: &str, value: f64, save: bool) { | |
866 | use proxmox_backup::rrd; | |
867 | if let Err(err) = rrd::update_value(name, value, rrd::DST::Derive, save) { | |
868 | eprintln!("rrd::update_value '{}' failed - {}", name, err); | |
869 | } | |
870 | } | |
871 | ||
872 | async fn generate_host_stats(save: bool) { | |
873 | use proxmox::sys::linux::procfs::{ | |
874 | read_meminfo, read_proc_stat, read_proc_net_dev, read_loadavg}; | |
875 | ||
876 | pbs_runtime::block_in_place(move || { | |
877 | ||
878 | match read_proc_stat() { | |
879 | Ok(stat) => { | |
880 | rrd_update_gauge("host/cpu", stat.cpu, save); | |
881 | rrd_update_gauge("host/iowait", stat.iowait_percent, save); | |
882 | } | |
883 | Err(err) => { | |
884 | eprintln!("read_proc_stat failed - {}", err); | |
885 | } | |
886 | } | |
887 | ||
888 | match read_meminfo() { | |
889 | Ok(meminfo) => { | |
890 | rrd_update_gauge("host/memtotal", meminfo.memtotal as f64, save); | |
891 | rrd_update_gauge("host/memused", meminfo.memused as f64, save); | |
892 | rrd_update_gauge("host/swaptotal", meminfo.swaptotal as f64, save); | |
893 | rrd_update_gauge("host/swapused", meminfo.swapused as f64, save); | |
894 | } | |
895 | Err(err) => { | |
896 | eprintln!("read_meminfo failed - {}", err); | |
897 | } | |
898 | } | |
899 | ||
900 | match read_proc_net_dev() { | |
901 | Ok(netdev) => { | |
902 | use pbs_config::network::is_physical_nic; | |
903 | let mut netin = 0; | |
904 | let mut netout = 0; | |
905 | for item in netdev { | |
906 | if !is_physical_nic(&item.device) { continue; } | |
907 | netin += item.receive; | |
908 | netout += item.send; | |
909 | } | |
910 | rrd_update_derive("host/netin", netin as f64, save); | |
911 | rrd_update_derive("host/netout", netout as f64, save); | |
912 | } | |
913 | Err(err) => { | |
914 | eprintln!("read_prox_net_dev failed - {}", err); | |
915 | } | |
916 | } | |
917 | ||
918 | match read_loadavg() { | |
919 | Ok(loadavg) => { | |
920 | rrd_update_gauge("host/loadavg", loadavg.0 as f64, save); | |
921 | } | |
922 | Err(err) => { | |
923 | eprintln!("read_loadavg failed - {}", err); | |
924 | } | |
925 | } | |
926 | ||
927 | let disk_manager = DiskManage::new(); | |
928 | ||
929 | gather_disk_stats(disk_manager.clone(), Path::new("/"), "host", save); | |
930 | ||
931 | match pbs_config::datastore::config() { | |
932 | Ok((config, _)) => { | |
933 | let datastore_list: Vec<DataStoreConfig> = | |
934 | config.convert_to_typed_array("datastore").unwrap_or_default(); | |
935 | ||
936 | for config in datastore_list { | |
937 | ||
938 | let rrd_prefix = format!("datastore/{}", config.name); | |
939 | let path = std::path::Path::new(&config.path); | |
940 | gather_disk_stats(disk_manager.clone(), path, &rrd_prefix, save); | |
941 | } | |
942 | } | |
943 | Err(err) => { | |
944 | eprintln!("read datastore config failed - {}", err); | |
945 | } | |
946 | } | |
947 | ||
948 | }); | |
949 | } | |
950 | ||
951 | fn check_schedule(worker_type: &str, event_str: &str, id: &str) -> bool { | |
952 | let event = match parse_calendar_event(event_str) { | |
953 | Ok(event) => event, | |
954 | Err(err) => { | |
955 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
956 | return false; | |
957 | } | |
958 | }; | |
959 | ||
960 | let last = match jobstate::last_run_time(worker_type, &id) { | |
961 | Ok(time) => time, | |
962 | Err(err) => { | |
963 | eprintln!("could not get last run time of {} {}: {}", worker_type, id, err); | |
964 | return false; | |
965 | } | |
966 | }; | |
967 | ||
968 | let next = match compute_next_event(&event, last, false) { | |
969 | Ok(Some(next)) => next, | |
970 | Ok(None) => return false, | |
971 | Err(err) => { | |
972 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
973 | return false; | |
974 | } | |
975 | }; | |
976 | ||
977 | let now = proxmox::tools::time::epoch_i64(); | |
978 | next <= now | |
979 | } | |
980 | ||
981 | fn gather_disk_stats(disk_manager: Arc<DiskManage>, path: &Path, rrd_prefix: &str, save: bool) { | |
982 | ||
983 | match proxmox_backup::tools::disks::disk_usage(path) { | |
984 | Ok(status) => { | |
985 | let rrd_key = format!("{}/total", rrd_prefix); | |
986 | rrd_update_gauge(&rrd_key, status.total as f64, save); | |
987 | let rrd_key = format!("{}/used", rrd_prefix); | |
988 | rrd_update_gauge(&rrd_key, status.used as f64, save); | |
989 | } | |
990 | Err(err) => { | |
991 | eprintln!("read disk_usage on {:?} failed - {}", path, err); | |
992 | } | |
993 | } | |
994 | ||
995 | match disk_manager.find_mounted_device(path) { | |
996 | Ok(None) => {}, | |
997 | Ok(Some((fs_type, device, source))) => { | |
998 | let mut device_stat = None; | |
999 | match fs_type.as_str() { | |
1000 | "zfs" => { | |
1001 | if let Some(source) = source { | |
1002 | let pool = get_pool_from_dataset(&source).unwrap_or(&source); | |
1003 | match zfs_pool_stats(pool) { | |
1004 | Ok(stat) => device_stat = stat, | |
1005 | Err(err) => eprintln!("zfs_pool_stats({:?}) failed - {}", pool, err), | |
1006 | } | |
1007 | } | |
1008 | } | |
1009 | _ => { | |
1010 | if let Ok(disk) = disk_manager.clone().disk_by_dev_num(device.into_dev_t()) { | |
1011 | match disk.read_stat() { | |
1012 | Ok(stat) => device_stat = stat, | |
1013 | Err(err) => eprintln!("disk.read_stat {:?} failed - {}", path, err), | |
1014 | } | |
1015 | } | |
1016 | } | |
1017 | } | |
1018 | if let Some(stat) = device_stat { | |
1019 | let rrd_key = format!("{}/read_ios", rrd_prefix); | |
1020 | rrd_update_derive(&rrd_key, stat.read_ios as f64, save); | |
1021 | let rrd_key = format!("{}/read_bytes", rrd_prefix); | |
1022 | rrd_update_derive(&rrd_key, (stat.read_sectors*512) as f64, save); | |
1023 | ||
1024 | let rrd_key = format!("{}/write_ios", rrd_prefix); | |
1025 | rrd_update_derive(&rrd_key, stat.write_ios as f64, save); | |
1026 | let rrd_key = format!("{}/write_bytes", rrd_prefix); | |
1027 | rrd_update_derive(&rrd_key, (stat.write_sectors*512) as f64, save); | |
1028 | ||
1029 | let rrd_key = format!("{}/io_ticks", rrd_prefix); | |
1030 | rrd_update_derive(&rrd_key, (stat.io_ticks as f64)/1000.0, save); | |
1031 | } | |
1032 | } | |
1033 | Err(err) => { | |
1034 | eprintln!("find_mounted_device failed - {}", err); | |
1035 | } | |
1036 | } | |
1037 | } |