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