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