src/bin/proxmox-backup-proxy.rs

   1 use std::sync::{Mutex, Arc};
   2 use std::path::{Path, PathBuf};
   3 use std::os::unix::io::AsRawFd;
   4 use std::future::Future;
   5 use std::pin::Pin;
   6
   7 use anyhow::{bail, format_err, Error};
   8 use futures::*;
   9 use http::request::Parts;
  10 use http::Response;
  11 use hyper::{Body, StatusCode};
  12 use hyper::header;
  13 use url::form_urlencoded;
  14
  15 use openssl::ssl::{SslMethod, SslAcceptor, SslFiletype};
  16 use tokio_stream::wrappers::ReceiverStream;
  17 use serde_json::{json, Value};
  18 use http::{Method, HeaderMap};
  19
  20 use proxmox_sys::linux::socket::set_tcp_keepalive;
  21 use proxmox_sys::fs::CreateOptions;
  22 use proxmox_lang::try_block;
  23 use proxmox_router::{RpcEnvironment, RpcEnvironmentType, UserInformation};
  24 use proxmox_http::client::{RateLimitedStream, ShareableRateLimit};
  25 use proxmox_sys::{task_log, task_warn};
  26 use proxmox_sys::logrotate::LogRotate;
  27
  28 use pbs_datastore::DataStore;
  29
  30 use proxmox_rest_server::{
  31     rotate_task_log_archive, extract_cookie , AuthError, ApiConfig, RestServer, RestEnvironment,
  32     ServerAdapter, WorkerTask, cleanup_old_tasks,
  33 };
  34
  35 use proxmox_backup::rrd_cache::{
  36     initialize_rrd_cache, rrd_update_gauge, rrd_update_derive, rrd_sync_journal,
  37 };
  38 use proxmox_backup::{
  39     TRAFFIC_CONTROL_CACHE,
  40     server::{
  41         auth::check_pbs_auth,
  42         jobstate::{
  43             self,
  44             Job,
  45         },
  46     },
  47 };
  48
  49 use pbs_buildcfg::configdir;
  50 use proxmox_time::CalendarEvent;
  51
  52 use pbs_api_types::{
  53     Authid, TapeBackupJobConfig, VerificationJobConfig, SyncJobConfig, DataStoreConfig,
  54     PruneOptions,
  55 };
  56
  57 use proxmox_rest_server::daemon;
  58
  59 use proxmox_backup::server;
  60 use proxmox_backup::auth_helpers::*;
  61 use proxmox_backup::tools::{
  62     PROXMOX_BACKUP_TCP_KEEPALIVE_TIME,
  63     disks::{
  64         DiskManage,
  65         zfs_dataset_stats,
  66     },
  67 };
  68
  69
  70 use proxmox_backup::api2::pull::do_sync_job;
  71 use proxmox_backup::api2::tape::backup::do_tape_backup_job;
  72 use proxmox_backup::server::do_verification_job;
  73 use proxmox_backup::server::do_prune_job;
  74
  75 fn main() -> Result<(), Error> {
  76     proxmox_backup::tools::setup_safe_path_env();
  77
  78     let backup_uid = pbs_config::backup_user()?.uid;
  79     let backup_gid = pbs_config::backup_group()?.gid;
  80     let running_uid = nix::unistd::Uid::effective();
  81     let running_gid = nix::unistd::Gid::effective();
  82
  83     if running_uid != backup_uid || running_gid != backup_gid {
  84         bail!("proxy not running as backup user or group (got uid {} gid {})", running_uid, running_gid);
  85     }
  86
  87     proxmox_async::runtime::main(run())
  88 }
  89
  90
  91 struct ProxmoxBackupProxyAdapter;
  92
  93 impl ServerAdapter for ProxmoxBackupProxyAdapter {
  94
  95     fn get_index(
  96         &self,
  97         env: RestEnvironment,
  98         parts: Parts,
  99     ) -> Pin<Box<dyn Future<Output = Response<Body>> + Send>> {
 100         Box::pin(get_index_future(env, parts))
 101     }
 102
 103     fn check_auth<'a>(
 104         &'a self,
 105         headers: &'a HeaderMap,
 106         method: &'a Method,
 107     ) -> Pin<Box<dyn Future<Output = Result<(String, Box<dyn UserInformation + Sync + Send>), AuthError>> + Send + 'a>> {
 108         Box::pin(async move {
 109             check_pbs_auth(headers, method).await
 110         })
 111     }
 112 }
 113
 114 fn extract_lang_header(headers: &http::HeaderMap) -> Option<String> {
 115     if let Some(Ok(cookie)) = headers.get("COOKIE").map(|v| v.to_str()) {
 116         return extract_cookie(cookie, "PBSLangCookie");
 117     }
 118     None
 119 }
 120
 121 async fn get_index_future(
 122     env: RestEnvironment,
 123     parts: Parts,
 124 ) -> Response<Body> {
 125
 126     let auth_id = env.get_auth_id();
 127     let api = env.api_config();
 128     let language = extract_lang_header(&parts.headers);
 129
 130      // fixme: make all IO async
 131
 132     let (userid, csrf_token) = match auth_id {
 133         Some(auth_id) => {
 134             let auth_id = auth_id.parse::<Authid>();
 135             match auth_id {
 136                 Ok(auth_id) if !auth_id.is_token() => {
 137                     let userid = auth_id.user().clone();
 138                     let new_csrf_token = assemble_csrf_prevention_token(csrf_secret(), &userid);
 139                     (Some(userid), Some(new_csrf_token))
 140                 }
 141                 _ => (None, None)
 142             }
 143         }
 144         None => (None, None),
 145     };
 146
 147     let nodename = proxmox_sys::nodename();
 148     let user = userid.as_ref().map(|u| u.as_str()).unwrap_or("");
 149
 150     let csrf_token = csrf_token.unwrap_or_else(|| String::from(""));
 151
 152     let mut debug = false;
 153     let mut template_file = "index";
 154
 155     if let Some(query_str) = parts.uri.query() {
 156         for (k, v) in form_urlencoded::parse(query_str.as_bytes()).into_owned() {
 157             if k == "debug" && v != "0" && v != "false" {
 158                 debug = true;
 159             } else if k == "console" {
 160                 template_file = "console";
 161             }
 162         }
 163     }
 164
 165     let mut lang = String::from("");
 166     if let Some(language) = language {
 167         if Path::new(&format!("/usr/share/pbs-i18n/pbs-lang-{}.js", language)).exists() {
 168             lang = language;
 169         }
 170     }
 171
 172     let data = json!({
 173         "NodeName": nodename,
 174         "UserName": user,
 175         "CSRFPreventionToken": csrf_token,
 176         "language": lang,
 177         "debug": debug,
 178     });
 179
 180     let (ct, index) = match api.render_template(template_file, &data) {
 181         Ok(index) => ("text/html", index),
 182         Err(err) => ("text/plain", format!("Error rendering template: {}", err)),
 183     };
 184
 185     let mut resp = Response::builder()
 186         .status(StatusCode::OK)
 187         .header(header::CONTENT_TYPE, ct)
 188         .body(index.into())
 189         .unwrap();
 190
 191     if let Some(userid) = userid {
 192         resp.extensions_mut().insert(Authid::from((userid, None)));
 193     }
 194
 195     resp
 196 }
 197
 198 async fn run() -> Result<(), Error> {
 199     if let Err(err) = syslog::init(
 200         syslog::Facility::LOG_DAEMON,
 201         log::LevelFilter::Info,
 202         Some("proxmox-backup-proxy")) {
 203         bail!("unable to inititialize syslog - {}", err);
 204     }
 205
 206     // Note: To debug early connection error use
 207     // PROXMOX_DEBUG=1 ./target/release/proxmox-backup-proxy
 208     let debug = std::env::var("PROXMOX_DEBUG").is_ok();
 209
 210     let _ = public_auth_key(); // load with lazy_static
 211     let _ = csrf_secret(); // load with lazy_static
 212
 213     let rrd_cache = initialize_rrd_cache()?;
 214     rrd_cache.apply_journal()?;
 215
 216     let mut config = ApiConfig::new(
 217         pbs_buildcfg::JS_DIR,
 218         &proxmox_backup::api2::ROUTER,
 219         RpcEnvironmentType::PUBLIC,
 220         ProxmoxBackupProxyAdapter,
 221     )?;
 222
 223     config.add_alias("novnc", "/usr/share/novnc-pve");
 224     config.add_alias("extjs", "/usr/share/javascript/extjs");
 225     config.add_alias("qrcodejs", "/usr/share/javascript/qrcodejs");
 226     config.add_alias("fontawesome", "/usr/share/fonts-font-awesome");
 227     config.add_alias("xtermjs", "/usr/share/pve-xtermjs");
 228     config.add_alias("locale", "/usr/share/pbs-i18n");
 229     config.add_alias("widgettoolkit", "/usr/share/javascript/proxmox-widget-toolkit");
 230     config.add_alias("docs", "/usr/share/doc/proxmox-backup/html");
 231
 232     let mut indexpath = PathBuf::from(pbs_buildcfg::JS_DIR);
 233     indexpath.push("index.hbs");
 234     config.register_template("index", &indexpath)?;
 235     config.register_template("console", "/usr/share/pve-xtermjs/index.html.hbs")?;
 236
 237     let backup_user = pbs_config::backup_user()?;
 238     let mut commando_sock = proxmox_rest_server::CommandSocket::new(proxmox_rest_server::our_ctrl_sock(), backup_user.gid);
 239
 240     let dir_opts = CreateOptions::new().owner(backup_user.uid).group(backup_user.gid);
 241     let file_opts = CreateOptions::new().owner(backup_user.uid).group(backup_user.gid);
 242
 243     config.enable_access_log(
 244         pbs_buildcfg::API_ACCESS_LOG_FN,
 245         Some(dir_opts.clone()),
 246         Some(file_opts.clone()),
 247         &mut commando_sock,
 248     )?;
 249
 250     config.enable_auth_log(
 251         pbs_buildcfg::API_AUTH_LOG_FN,
 252         Some(dir_opts.clone()),
 253         Some(file_opts.clone()),
 254         &mut commando_sock,
 255     )?;
 256
 257     let rest_server = RestServer::new(config);
 258     proxmox_rest_server::init_worker_tasks(pbs_buildcfg::PROXMOX_BACKUP_LOG_DIR_M!().into(), file_opts.clone())?;
 259
 260     //openssl req -x509 -newkey rsa:4096 -keyout /etc/proxmox-backup/proxy.key -out /etc/proxmox-backup/proxy.pem -nodes
 261
 262     // we build the initial acceptor here as we cannot start if this fails
 263     let acceptor = make_tls_acceptor()?;
 264     let acceptor = Arc::new(Mutex::new(acceptor));
 265
 266     // to renew the acceptor we just add a command-socket handler
 267     commando_sock.register_command(
 268         "reload-certificate".to_string(),
 269         {
 270             let acceptor = Arc::clone(&acceptor);
 271             move |_value| -> Result<_, Error> {
 272                 log::info!("reloading certificate");
 273                 match make_tls_acceptor() {
 274                     Err(err) => log::error!("error reloading certificate: {}", err),
 275                     Ok(new_acceptor) => {
 276                         let mut guard = acceptor.lock().unwrap();
 277                         *guard = new_acceptor;
 278                     }
 279                 }
 280                 Ok(Value::Null)
 281             }
 282         },
 283     )?;
 284
 285     // to remove references for not configured datastores
 286     commando_sock.register_command(
 287         "datastore-removed".to_string(),
 288         |_value| {
 289             if let Err(err) = DataStore::remove_unused_datastores() {
 290                 log::error!("could not refresh datastores: {}", err);
 291             }
 292             Ok(Value::Null)
 293         }
 294     )?;
 295
 296     let server = daemon::create_daemon(
 297         ([0,0,0,0,0,0,0,0], 8007).into(),
 298         move |listener| {
 299
 300             let connections = accept_connections(listener, acceptor, debug);
 301             let connections = hyper::server::accept::from_stream(ReceiverStream::new(connections));
 302
 303             Ok(async {
 304                 daemon::systemd_notify(daemon::SystemdNotify::Ready)?;
 305
 306                 hyper::Server::builder(connections)
 307                     .serve(rest_server)
 308                     .with_graceful_shutdown(proxmox_rest_server::shutdown_future())
 309                     .map_err(Error::from)
 310                     .await
 311             })
 312         },
 313     );
 314
 315     proxmox_rest_server::write_pid(pbs_buildcfg::PROXMOX_BACKUP_PROXY_PID_FN)?;
 316
 317     let init_result: Result<(), Error> = try_block!({
 318         proxmox_rest_server::register_task_control_commands(&mut commando_sock)?;
 319         commando_sock.spawn()?;
 320         proxmox_rest_server::catch_shutdown_signal()?;
 321         proxmox_rest_server::catch_reload_signal()?;
 322         Ok(())
 323     });
 324
 325     if let Err(err) = init_result {
 326         bail!("unable to start daemon - {}", err);
 327     }
 328
 329     start_task_scheduler();
 330     start_stat_generator();
 331     start_traffic_control_updater();
 332
 333     server.await?;
 334     log::info!("server shutting down, waiting for active workers to complete");
 335     proxmox_rest_server::last_worker_future().await?;
 336     log::info!("done - exit server");
 337
 338     Ok(())
 339 }
 340
 341 fn make_tls_acceptor() -> Result<SslAcceptor, Error> {
 342     let key_path = configdir!("/proxy.key");
 343     let cert_path = configdir!("/proxy.pem");
 344
 345     let (config, _) = proxmox_backup::config::node::config()?;
 346     let ciphers_tls_1_3 = config.ciphers_tls_1_3;
 347     let ciphers_tls_1_2 = config.ciphers_tls_1_2;
 348
 349     let mut acceptor = SslAcceptor::mozilla_intermediate_v5(SslMethod::tls()).unwrap();
 350     if let Some(ciphers) = ciphers_tls_1_3.as_deref() {
 351         acceptor.set_ciphersuites(ciphers)?;
 352     }
 353     if let Some(ciphers) = ciphers_tls_1_2.as_deref() {
 354         acceptor.set_cipher_list(ciphers)?;
 355     }
 356     acceptor.set_private_key_file(key_path, SslFiletype::PEM)
 357         .map_err(|err| format_err!("unable to read proxy key {} - {}", key_path, err))?;
 358     acceptor.set_certificate_chain_file(cert_path)
 359         .map_err(|err| format_err!("unable to read proxy cert {} - {}", cert_path, err))?;
 360     acceptor.set_options(openssl::ssl::SslOptions::NO_RENEGOTIATION);
 361     acceptor.check_private_key().unwrap();
 362
 363     Ok(acceptor.build())
 364 }
 365
 366 type ClientStreamResult =
 367     Result<std::pin::Pin<Box<tokio_openssl::SslStream<RateLimitedStream<tokio::net::TcpStream>>>>, Error>;
 368 const MAX_PENDING_ACCEPTS: usize = 1024;
 369
 370 fn accept_connections(
 371     listener: tokio::net::TcpListener,
 372     acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>,
 373     debug: bool,
 374 ) -> tokio::sync::mpsc::Receiver<ClientStreamResult> {
 375
 376     let (sender, receiver) = tokio::sync::mpsc::channel(MAX_PENDING_ACCEPTS);
 377
 378     tokio::spawn(accept_connection(listener, acceptor, debug, sender));
 379
 380     receiver
 381 }
 382
 383 async fn accept_connection(
 384     listener: tokio::net::TcpListener,
 385     acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>,
 386     debug: bool,
 387     sender: tokio::sync::mpsc::Sender<ClientStreamResult>,
 388 ) {
 389     let accept_counter = Arc::new(());
 390
 391     loop {
 392         let (sock, _addr) = match listener.accept().await {
 393             Ok(conn) => conn,
 394             Err(err) =>  {
 395                 eprintln!("error accepting tcp connection: {}", err);
 396                 continue;
 397             }
 398         };
 399
 400         sock.set_nodelay(true).unwrap();
 401         let _ = set_tcp_keepalive(sock.as_raw_fd(), PROXMOX_BACKUP_TCP_KEEPALIVE_TIME);
 402
 403         let peer = sock.peer_addr().ok();
 404         let sock = RateLimitedStream::with_limiter_update_cb(sock, move || lookup_rate_limiter(peer));
 405
 406         let ssl = { // limit acceptor_guard scope
 407             // Acceptor can be reloaded using the command socket "reload-certificate" command
 408             let acceptor_guard = acceptor.lock().unwrap();
 409
 410             match openssl::ssl::Ssl::new(acceptor_guard.context()) {
 411                 Ok(ssl) => ssl,
 412                 Err(err) => {
 413                     eprintln!("failed to create Ssl object from Acceptor context - {}", err);
 414                     continue;
 415                 },
 416             }
 417         };
 418
 419         let stream = match tokio_openssl::SslStream::new(ssl, sock) {
 420             Ok(stream) => stream,
 421             Err(err) => {
 422                 eprintln!("failed to create SslStream using ssl and connection socket - {}", err);
 423                 continue;
 424             },
 425         };
 426
 427         let mut stream = Box::pin(stream);
 428         let sender = sender.clone();
 429
 430         if Arc::strong_count(&accept_counter) > MAX_PENDING_ACCEPTS {
 431             eprintln!("connection rejected - to many open connections");
 432             continue;
 433         }
 434
 435         let accept_counter = Arc::clone(&accept_counter);
 436         tokio::spawn(async move {
 437             let accept_future = tokio::time::timeout(
 438                 Duration::new(10, 0), stream.as_mut().accept());
 439
 440             let result = accept_future.await;
 441
 442             match result {
 443                 Ok(Ok(())) => {
 444                     if sender.send(Ok(stream)).await.is_err() && debug {
 445                         eprintln!("detect closed connection channel");
 446                     }
 447                 }
 448                 Ok(Err(err)) => {
 449                     if debug {
 450                         eprintln!("https handshake failed - {}", err);
 451                     }
 452                 }
 453                 Err(_) => {
 454                     if debug {
 455                         eprintln!("https handshake timeout");
 456                     }
 457                 }
 458             }
 459
 460             drop(accept_counter); // decrease reference count
 461         });
 462     }
 463 }
 464
 465 fn start_stat_generator() {
 466     let abort_future = proxmox_rest_server::shutdown_future();
 467     let future = Box::pin(run_stat_generator());
 468     let task = futures::future::select(future, abort_future);
 469     tokio::spawn(task.map(|_| ()));
 470 }
 471
 472 fn start_task_scheduler() {
 473     let abort_future = proxmox_rest_server::shutdown_future();
 474     let future = Box::pin(run_task_scheduler());
 475     let task = futures::future::select(future, abort_future);
 476     tokio::spawn(task.map(|_| ()));
 477 }
 478
 479 fn start_traffic_control_updater() {
 480     let abort_future = proxmox_rest_server::shutdown_future();
 481     let future = Box::pin(run_traffic_control_updater());
 482     let task = futures::future::select(future, abort_future);
 483     tokio::spawn(task.map(|_| ()));
 484 }
 485
 486 use std::time::{SystemTime, Instant, Duration, UNIX_EPOCH};
 487
 488 fn next_minute() -> Result<Instant, Error> {
 489     let now = SystemTime::now();
 490     let epoch_now = now.duration_since(UNIX_EPOCH)?;
 491     let epoch_next = Duration::from_secs((epoch_now.as_secs()/60  + 1)*60);
 492     Ok(Instant::now() + epoch_next - epoch_now)
 493 }
 494
 495 async fn run_task_scheduler() {
 496
 497     let mut count: usize = 0;
 498
 499     loop {
 500         count += 1;
 501
 502         let delay_target = match next_minute() {  // try to run very minute
 503             Ok(d) => d,
 504             Err(err) => {
 505                 eprintln!("task scheduler: compute next minute failed - {}", err);
 506                 tokio::time::sleep_until(tokio::time::Instant::from_std(Instant::now() + Duration::from_secs(60))).await;
 507                 continue;
 508             }
 509         };
 510
 511         if count > 2 { // wait 1..2 minutes before starting
 512             match schedule_tasks().catch_unwind().await {
 513                 Err(panic) => {
 514                     match panic.downcast::<&str>() {
 515                         Ok(msg) => {
 516                             eprintln!("task scheduler panic: {}", msg);
 517                         }
 518                         Err(_) => {
 519                             eprintln!("task scheduler panic - unknown type");
 520                         }
 521                     }
 522                 }
 523                 Ok(Err(err)) => {
 524                     eprintln!("task scheduler failed - {:?}", err);
 525                 }
 526                 Ok(Ok(_)) => {}
 527             }
 528         }
 529
 530         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
 531     }
 532 }
 533
 534 async fn schedule_tasks() -> Result<(), Error> {
 535
 536     schedule_datastore_garbage_collection().await;
 537     schedule_datastore_prune().await;
 538     schedule_datastore_sync_jobs().await;
 539     schedule_datastore_verify_jobs().await;
 540     schedule_tape_backup_jobs().await;
 541     schedule_task_log_rotate().await;
 542
 543     Ok(())
 544 }
 545
 546 async fn schedule_datastore_garbage_collection() {
 547
 548     let config = match pbs_config::datastore::config() {
 549         Err(err) => {
 550             eprintln!("unable to read datastore config - {}", err);
 551             return;
 552         }
 553         Ok((config, _digest)) => config,
 554     };
 555
 556     for (store, (_, store_config)) in config.sections {
 557         let datastore = match DataStore::lookup_datastore(&store) {
 558             Ok(datastore) => datastore,
 559             Err(err) => {
 560                 eprintln!("lookup_datastore failed - {}", err);
 561                 continue;
 562             }
 563         };
 564
 565         let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
 566             Ok(c) => c,
 567             Err(err) => {
 568                 eprintln!("datastore config from_value failed - {}", err);
 569                 continue;
 570             }
 571         };
 572
 573         let event_str = match store_config.gc_schedule {
 574             Some(event_str) => event_str,
 575             None => continue,
 576         };
 577
 578         let event: CalendarEvent = match event_str.parse() {
 579             Ok(event) => event,
 580             Err(err) => {
 581                 eprintln!("unable to parse schedule '{}' - {}", event_str, err);
 582                 continue;
 583             }
 584         };
 585
 586         if datastore.garbage_collection_running() { continue; }
 587
 588         let worker_type = "garbage_collection";
 589
 590         let last = match jobstate::last_run_time(worker_type, &store) {
 591             Ok(time) => time,
 592             Err(err) => {
 593                 eprintln!("could not get last run time of {} {}: {}", worker_type, store, err);
 594                 continue;
 595             }
 596         };
 597
 598         let next = match event.compute_next_event(last) {
 599             Ok(Some(next)) => next,
 600             Ok(None) => continue,
 601             Err(err) => {
 602                 eprintln!("compute_next_event for '{}' failed - {}", event_str, err);
 603                 continue;
 604             }
 605         };
 606
 607         let now = proxmox_time::epoch_i64();
 608
 609         if next > now  { continue; }
 610
 611         let job = match Job::new(worker_type, &store) {
 612             Ok(job) => job,
 613             Err(_) => continue, // could not get lock
 614         };
 615
 616         let auth_id = Authid::root_auth_id();
 617
 618         if let Err(err) = crate::server::do_garbage_collection_job(job, datastore, auth_id, Some(event_str), false) {
 619             eprintln!("unable to start garbage collection job on datastore {} - {}", store, err);
 620         }
 621     }
 622 }
 623
 624 async fn schedule_datastore_prune() {
 625
 626     let config = match pbs_config::datastore::config() {
 627         Err(err) => {
 628             eprintln!("unable to read datastore config - {}", err);
 629             return;
 630         }
 631         Ok((config, _digest)) => config,
 632     };
 633
 634     for (store, (_, store_config)) in config.sections {
 635
 636         let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
 637             Ok(c) => c,
 638             Err(err) => {
 639                 eprintln!("datastore '{}' config from_value failed - {}", store, err);
 640                 continue;
 641             }
 642         };
 643
 644         let event_str = match store_config.prune_schedule {
 645             Some(event_str) => event_str,
 646             None => continue,
 647         };
 648
 649         let prune_options = PruneOptions {
 650             keep_last: store_config.keep_last,
 651             keep_hourly: store_config.keep_hourly,
 652             keep_daily:  store_config.keep_daily,
 653             keep_weekly: store_config.keep_weekly,
 654             keep_monthly: store_config.keep_monthly,
 655             keep_yearly: store_config.keep_yearly,
 656         };
 657
 658         if !pbs_datastore::prune::keeps_something(&prune_options) { // no prune settings - keep all
 659             continue;
 660         }
 661
 662         let worker_type = "prune";
 663         if check_schedule(worker_type, &event_str, &store) {
 664             let job = match Job::new(worker_type, &store) {
 665                 Ok(job) => job,
 666                 Err(_) => continue, // could not get lock
 667             };
 668
 669             let auth_id = Authid::root_auth_id().clone();
 670             if let Err(err) = do_prune_job(job, prune_options, store.clone(), &auth_id, Some(event_str)) {
 671                 eprintln!("unable to start datastore prune job {} - {}", &store, err);
 672             }
 673         };
 674     }
 675 }
 676
 677 async fn schedule_datastore_sync_jobs() {
 678
 679
 680     let config = match pbs_config::sync::config() {
 681         Err(err) => {
 682             eprintln!("unable to read sync job config - {}", err);
 683             return;
 684         }
 685         Ok((config, _digest)) => config,
 686     };
 687
 688     for (job_id, (_, job_config)) in config.sections {
 689         let job_config: SyncJobConfig = match serde_json::from_value(job_config) {
 690             Ok(c) => c,
 691             Err(err) => {
 692                 eprintln!("sync job config from_value failed - {}", err);
 693                 continue;
 694             }
 695         };
 696
 697         let event_str = match job_config.schedule {
 698             Some(ref event_str) => event_str.clone(),
 699             None => continue,
 700         };
 701
 702         let worker_type = "syncjob";
 703         if check_schedule(worker_type, &event_str, &job_id) {
 704             let job = match Job::new(worker_type, &job_id) {
 705                 Ok(job) => job,
 706                 Err(_) => continue, // could not get lock
 707             };
 708
 709             let auth_id = Authid::root_auth_id().clone();
 710             if let Err(err) = do_sync_job(job, job_config, &auth_id, Some(event_str), false) {
 711                 eprintln!("unable to start datastore sync job {} - {}", &job_id, err);
 712             }
 713         };
 714     }
 715 }
 716
 717 async fn schedule_datastore_verify_jobs() {
 718
 719     let config = match pbs_config::verify::config() {
 720         Err(err) => {
 721             eprintln!("unable to read verification job config - {}", err);
 722             return;
 723         }
 724         Ok((config, _digest)) => config,
 725     };
 726     for (job_id, (_, job_config)) in config.sections {
 727         let job_config: VerificationJobConfig = match serde_json::from_value(job_config) {
 728             Ok(c) => c,
 729             Err(err) => {
 730                 eprintln!("verification job config from_value failed - {}", err);
 731                 continue;
 732             }
 733         };
 734         let event_str = match job_config.schedule {
 735             Some(ref event_str) => event_str.clone(),
 736             None => continue,
 737         };
 738
 739         let worker_type = "verificationjob";
 740         let auth_id = Authid::root_auth_id().clone();
 741         if check_schedule(worker_type, &event_str, &job_id) {
 742             let job = match Job::new(worker_type, &job_id) {
 743                 Ok(job) => job,
 744                 Err(_) => continue, // could not get lock
 745             };
 746             if let Err(err) = do_verification_job(job, job_config, &auth_id, Some(event_str), false) {
 747                 eprintln!("unable to start datastore verification job {} - {}", &job_id, err);
 748             }
 749         };
 750     }
 751 }
 752
 753 async fn schedule_tape_backup_jobs() {
 754
 755     let config = match pbs_config::tape_job::config() {
 756         Err(err) => {
 757             eprintln!("unable to read tape job config - {}", err);
 758             return;
 759         }
 760         Ok((config, _digest)) => config,
 761     };
 762     for (job_id, (_, job_config)) in config.sections {
 763         let job_config: TapeBackupJobConfig = match serde_json::from_value(job_config) {
 764             Ok(c) => c,
 765             Err(err) => {
 766                 eprintln!("tape backup job config from_value failed - {}", err);
 767                 continue;
 768             }
 769         };
 770         let event_str = match job_config.schedule {
 771             Some(ref event_str) => event_str.clone(),
 772             None => continue,
 773         };
 774
 775         let worker_type = "tape-backup-job";
 776         let auth_id = Authid::root_auth_id().clone();
 777         if check_schedule(worker_type, &event_str, &job_id) {
 778             let job = match Job::new(worker_type, &job_id) {
 779                 Ok(job) => job,
 780                 Err(_) => continue, // could not get lock
 781             };
 782             if let Err(err) = do_tape_backup_job(job, job_config.setup, &auth_id, Some(event_str), false) {
 783                 eprintln!("unable to start tape backup job {} - {}", &job_id, err);
 784             }
 785         };
 786     }
 787 }
 788
 789
 790 async fn schedule_task_log_rotate() {
 791
 792     let worker_type = "logrotate";
 793     let job_id = "access-log_and_task-archive";
 794
 795     // schedule daily at 00:00 like normal logrotate
 796     let schedule = "00:00";
 797
 798     if !check_schedule(worker_type, schedule, job_id) {
 799         // if we never ran the rotation, schedule instantly
 800         match jobstate::JobState::load(worker_type, job_id) {
 801             Ok(state) => match state {
 802                 jobstate::JobState::Created { .. } => {},
 803                 _ => return,
 804             },
 805             _ => return,
 806         }
 807     }
 808
 809     let mut job = match Job::new(worker_type, job_id) {
 810         Ok(job) => job,
 811         Err(_) => return, // could not get lock
 812     };
 813
 814     if let Err(err) = WorkerTask::new_thread(
 815         worker_type,
 816         None,
 817         Authid::root_auth_id().to_string(),
 818         false,
 819         move |worker| {
 820             job.start(&worker.upid().to_string())?;
 821             task_log!(worker, "starting task log rotation");
 822
 823             let result = try_block!({
 824                 let max_size = 512 * 1024 - 1; // an entry has ~ 100b, so > 5000 entries/file
 825                 let max_files = 20; // times twenty files gives > 100000 task entries
 826
 827                 let user = pbs_config::backup_user()?;
 828                 let options = proxmox_sys::fs::CreateOptions::new()
 829                     .owner(user.uid)
 830                     .group(user.gid);
 831
 832                 let has_rotated = rotate_task_log_archive(
 833                     max_size,
 834                     true,
 835                     Some(max_files),
 836                     Some(options.clone()),
 837                 )?;
 838
 839                 if has_rotated {
 840                     task_log!(worker, "task log archive was rotated");
 841                 } else {
 842                     task_log!(worker, "task log archive was not rotated");
 843                 }
 844
 845                 let max_size = 32 * 1024 * 1024 - 1;
 846                 let max_files = 14;
 847
 848
 849                 let mut logrotate = LogRotate::new(
 850                     pbs_buildcfg::API_ACCESS_LOG_FN,
 851                     true,
 852                     Some(max_files),
 853                     Some(options.clone()),
 854                 )?;
 855
 856                 if logrotate.rotate(max_size)? {
 857                     println!("rotated access log, telling daemons to re-open log file");
 858                     proxmox_async::runtime::block_on(command_reopen_access_logfiles())?;
 859                     task_log!(worker, "API access log was rotated");
 860                 } else {
 861                     task_log!(worker, "API access log was not rotated");
 862                 }
 863
 864                 let mut logrotate = LogRotate::new(
 865                     pbs_buildcfg::API_AUTH_LOG_FN,
 866                     true,
 867                     Some(max_files),
 868                     Some(options),
 869                 )?;
 870
 871                 if logrotate.rotate(max_size)? {
 872                     println!("rotated auth log, telling daemons to re-open log file");
 873                     proxmox_async::runtime::block_on(command_reopen_auth_logfiles())?;
 874                     task_log!(worker, "API authentication log was rotated");
 875                 } else {
 876                     task_log!(worker, "API authentication log was not rotated");
 877                 }
 878
 879                 if has_rotated {
 880                     task_log!(worker, "cleaning up old task logs");
 881                     if let Err(err) = cleanup_old_tasks(true) {
 882                         task_warn!(worker, "could not completely cleanup old tasks: {}", err);
 883                     }
 884                 }
 885
 886                 Ok(())
 887             });
 888
 889             let status = worker.create_state(&result);
 890
 891             if let Err(err) = job.finish(status) {
 892                 eprintln!("could not finish job state for {}: {}", worker_type, err);
 893             }
 894
 895             result
 896         },
 897     ) {
 898         eprintln!("unable to start task log rotation: {}", err);
 899     }
 900
 901 }
 902
 903 async fn command_reopen_access_logfiles() -> Result<(), Error> {
 904     // only care about the most recent daemon instance for each, proxy & api, as other older ones
 905     // should not respond to new requests anyway, but only finish their current one and then exit.
 906     let sock = proxmox_rest_server::our_ctrl_sock();
 907     let f1 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-access-log-reopen\"}\n");
 908
 909     let pid = proxmox_rest_server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?;
 910     let sock = proxmox_rest_server::ctrl_sock_from_pid(pid);
 911     let f2 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-access-log-reopen\"}\n");
 912
 913     match futures::join!(f1, f2) {
 914         (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)),
 915         (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)),
 916         (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)),
 917         _ => Ok(()),
 918     }
 919 }
 920
 921 async fn command_reopen_auth_logfiles() -> Result<(), Error> {
 922     // only care about the most recent daemon instance for each, proxy & api, as other older ones
 923     // should not respond to new requests anyway, but only finish their current one and then exit.
 924     let sock = proxmox_rest_server::our_ctrl_sock();
 925     let f1 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n");
 926
 927     let pid = proxmox_rest_server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?;
 928     let sock = proxmox_rest_server::ctrl_sock_from_pid(pid);
 929     let f2 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n");
 930
 931     match futures::join!(f1, f2) {
 932         (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)),
 933         (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)),
 934         (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)),
 935         _ => Ok(()),
 936     }
 937 }
 938
 939 async fn run_stat_generator() {
 940
 941     loop {
 942         let delay_target = Instant::now() +  Duration::from_secs(10);
 943
 944         generate_host_stats().await;
 945
 946         rrd_sync_journal();
 947
 948         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
 949
 950      }
 951
 952 }
 953
 954 async fn generate_host_stats() {
 955     match tokio::task::spawn_blocking(generate_host_stats_sync).await {
 956         Ok(()) => (),
 957         Err(err) => log::error!("generate_host_stats paniced: {}", err),
 958     }
 959 }
 960
 961 fn generate_host_stats_sync() {
 962     use proxmox_sys::linux::procfs::{
 963         read_meminfo, read_proc_stat, read_proc_net_dev, read_loadavg};
 964
 965     match read_proc_stat() {
 966         Ok(stat) => {
 967             rrd_update_gauge("host/cpu", stat.cpu);
 968             rrd_update_gauge("host/iowait", stat.iowait_percent);
 969         }
 970         Err(err) => {
 971             eprintln!("read_proc_stat failed - {}", err);
 972         }
 973     }
 974
 975     match read_meminfo() {
 976         Ok(meminfo) => {
 977             rrd_update_gauge("host/memtotal", meminfo.memtotal as f64);
 978             rrd_update_gauge("host/memused", meminfo.memused as f64);
 979             rrd_update_gauge("host/swaptotal", meminfo.swaptotal as f64);
 980             rrd_update_gauge("host/swapused", meminfo.swapused as f64);
 981         }
 982         Err(err) => {
 983             eprintln!("read_meminfo failed - {}", err);
 984         }
 985     }
 986
 987     match read_proc_net_dev() {
 988         Ok(netdev) => {
 989             use pbs_config::network::is_physical_nic;
 990             let mut netin = 0;
 991             let mut netout = 0;
 992             for item in netdev {
 993                 if !is_physical_nic(&item.device) { continue; }
 994                 netin += item.receive;
 995                 netout += item.send;
 996             }
 997             rrd_update_derive("host/netin", netin as f64);
 998             rrd_update_derive("host/netout", netout as f64);
 999         }
1000         Err(err) => {
1001             eprintln!("read_prox_net_dev failed - {}", err);
1002         }
1003     }
1004
1005     match read_loadavg() {
1006         Ok(loadavg) => {
1007             rrd_update_gauge("host/loadavg", loadavg.0 as f64);
1008         }
1009         Err(err) => {
1010             eprintln!("read_loadavg failed - {}", err);
1011         }
1012     }
1013
1014     let disk_manager = DiskManage::new();
1015
1016     gather_disk_stats(disk_manager.clone(), Path::new("/"), "host");
1017
1018     match pbs_config::datastore::config() {
1019         Ok((config, _)) => {
1020             let datastore_list: Vec<DataStoreConfig> =
1021                 config.convert_to_typed_array("datastore").unwrap_or_default();
1022
1023             for config in datastore_list {
1024
1025                 let rrd_prefix = format!("datastore/{}", config.name);
1026                 let path = std::path::Path::new(&config.path);
1027                 gather_disk_stats(disk_manager.clone(), path, &rrd_prefix);
1028             }
1029         }
1030         Err(err) => {
1031             eprintln!("read datastore config failed - {}", err);
1032         }
1033     }
1034 }
1035
1036 fn check_schedule(worker_type: &str, event_str: &str, id: &str) -> bool {
1037     let event: CalendarEvent = match event_str.parse() {
1038         Ok(event) => event,
1039         Err(err) => {
1040             eprintln!("unable to parse schedule '{}' - {}", event_str, err);
1041             return false;
1042         }
1043     };
1044
1045     let last = match jobstate::last_run_time(worker_type, id) {
1046         Ok(time) => time,
1047         Err(err) => {
1048             eprintln!("could not get last run time of {} {}: {}", worker_type, id, err);
1049             return false;
1050         }
1051     };
1052
1053     let next = match event.compute_next_event(last) {
1054         Ok(Some(next)) => next,
1055         Ok(None) => return false,
1056         Err(err) => {
1057             eprintln!("compute_next_event for '{}' failed - {}", event_str, err);
1058             return false;
1059         }
1060     };
1061
1062     let now = proxmox_time::epoch_i64();
1063     next <= now
1064 }
1065
1066 fn gather_disk_stats(disk_manager: Arc<DiskManage>, path: &Path, rrd_prefix: &str) {
1067
1068     match proxmox_backup::tools::disks::disk_usage(path) {
1069         Ok(status) => {
1070             let rrd_key = format!("{}/total", rrd_prefix);
1071             rrd_update_gauge(&rrd_key, status.total as f64);
1072             let rrd_key = format!("{}/used", rrd_prefix);
1073             rrd_update_gauge(&rrd_key, status.used as f64);
1074         }
1075         Err(err) => {
1076             eprintln!("read disk_usage on {:?} failed - {}", path, err);
1077         }
1078     }
1079
1080     match disk_manager.find_mounted_device(path) {
1081         Ok(None) => {},
1082         Ok(Some((fs_type, device, source))) => {
1083             let mut device_stat = None;
1084             match (fs_type.as_str(), source) {
1085                 ("zfs", Some(source)) => match source.into_string() {
1086                     Ok(dataset) => match zfs_dataset_stats(&dataset) {
1087                         Ok(stat) => device_stat = Some(stat),
1088                         Err(err) => eprintln!("zfs_dataset_stats({:?}) failed - {}", dataset, err),
1089                     },
1090                     Err(source) => {
1091                         eprintln!("zfs_pool_stats({:?}) failed - invalid characters", source)
1092                     }
1093                 },
1094                 _ => {
1095                     if let Ok(disk) = disk_manager.clone().disk_by_dev_num(device.into_dev_t()) {
1096                         match disk.read_stat() {
1097                             Ok(stat) => device_stat = stat,
1098                             Err(err) => eprintln!("disk.read_stat {:?} failed - {}", path, err),
1099                         }
1100                     }
1101                 }
1102             }
1103             if let Some(stat) = device_stat {
1104                 let rrd_key = format!("{}/read_ios", rrd_prefix);
1105                 rrd_update_derive(&rrd_key, stat.read_ios as f64);
1106                 let rrd_key = format!("{}/read_bytes", rrd_prefix);
1107                 rrd_update_derive(&rrd_key, (stat.read_sectors*512) as f64);
1108
1109                 let rrd_key = format!("{}/write_ios", rrd_prefix);
1110                 rrd_update_derive(&rrd_key, stat.write_ios as f64);
1111                 let rrd_key = format!("{}/write_bytes", rrd_prefix);
1112                 rrd_update_derive(&rrd_key, (stat.write_sectors*512) as f64);
1113
1114                 let rrd_key = format!("{}/io_ticks", rrd_prefix);
1115                 rrd_update_derive(&rrd_key, (stat.io_ticks as f64)/1000.0);
1116             }
1117         }
1118         Err(err) => {
1119             eprintln!("find_mounted_device failed - {}", err);
1120         }
1121     }
1122 }
1123
1124 // Rate Limiter lookup
1125
1126 // Test WITH
1127 // proxmox-backup-client restore vm/201/2021-10-22T09:55:56Z drive-scsi0.img img1.img --repository localhost:store2
1128
1129 async fn run_traffic_control_updater() {
1130
1131      loop {
1132         let delay_target = Instant::now() +  Duration::from_secs(1);
1133
1134         {
1135             let mut cache = TRAFFIC_CONTROL_CACHE.lock().unwrap();
1136             cache.compute_current_rates();
1137         }
1138
1139          tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
1140      }
1141
1142 }
1143
1144 fn lookup_rate_limiter(
1145     peer: Option<std::net::SocketAddr>,
1146 ) -> (Option<Arc<dyn ShareableRateLimit>>, Option<Arc<dyn ShareableRateLimit>>) {
1147     let mut cache = TRAFFIC_CONTROL_CACHE.lock().unwrap();
1148
1149     let now = proxmox_time::epoch_i64();
1150
1151     cache.reload(now);
1152
1153     let (_rule_name, read_limiter, write_limiter) = cache.lookup_rate_limiter(peer, now);
1154
1155     (read_limiter, write_limiter)
1156 }