src/bin/proxmox-backup-proxy.rs

   1 use std::future::Future;
   2 use std::os::unix::io::AsRawFd;
   3 use std::path::{Path, PathBuf};
   4 use std::pin::Pin;
   5 use std::sync::{Arc, Mutex};
   6
   7 use anyhow::{bail, format_err, Error};
   8 use futures::*;
   9 use http::request::Parts;
  10 use http::Response;
  11 use hyper::header;
  12 use hyper::{Body, StatusCode};
  13 use url::form_urlencoded;
  14
  15 use http::{HeaderMap, Method};
  16 use openssl::ssl::{SslAcceptor, SslFiletype, SslMethod};
  17 use serde_json::{json, Value};
  18 use tokio_stream::wrappers::ReceiverStream;
  19
  20 use proxmox_http::client::{RateLimitedStream, ShareableRateLimit};
  21 use proxmox_lang::try_block;
  22 use proxmox_router::{RpcEnvironment, RpcEnvironmentType, UserInformation};
  23 use proxmox_sys::fs::CreateOptions;
  24 use proxmox_sys::linux::socket::set_tcp_keepalive;
  25 use proxmox_sys::logrotate::LogRotate;
  26 use proxmox_sys::{task_log, task_warn};
  27
  28 use pbs_datastore::DataStore;
  29
  30 use proxmox_rest_server::{
  31     cleanup_old_tasks, cookie_from_header, rotate_task_log_archive, ApiConfig, AuthError,
  32     RestEnvironment, RestServer, ServerAdapter, WorkerTask,
  33 };
  34
  35 use proxmox_backup::rrd_cache::{
  36     initialize_rrd_cache, rrd_sync_journal, rrd_update_derive, rrd_update_gauge,
  37 };
  38 use proxmox_backup::{
  39     server::{
  40         auth::check_pbs_auth,
  41         jobstate::{self, Job},
  42     },
  43     traffic_control_cache::TRAFFIC_CONTROL_CACHE,
  44 };
  45
  46 use pbs_buildcfg::configdir;
  47 use proxmox_time::CalendarEvent;
  48
  49 use pbs_api_types::{
  50     Authid, DataStoreConfig, PruneOptions, SyncJobConfig, TapeBackupJobConfig,
  51     VerificationJobConfig, Operation
  52 };
  53
  54 use proxmox_rest_server::daemon;
  55
  56 use proxmox_backup::auth_helpers::*;
  57 use proxmox_backup::server;
  58 use proxmox_backup::tools::{
  59     disks::{zfs_dataset_stats, DiskManage},
  60     PROXMOX_BACKUP_TCP_KEEPALIVE_TIME,
  61 };
  62
  63 use proxmox_backup::api2::pull::do_sync_job;
  64 use proxmox_backup::api2::tape::backup::do_tape_backup_job;
  65 use proxmox_backup::server::do_prune_job;
  66 use proxmox_backup::server::do_verification_job;
  67
  68 fn main() -> Result<(), Error> {
  69     pbs_tools::setup_libc_malloc_opts();
  70
  71     proxmox_backup::tools::setup_safe_path_env();
  72
  73     let backup_uid = pbs_config::backup_user()?.uid;
  74     let backup_gid = pbs_config::backup_group()?.gid;
  75     let running_uid = nix::unistd::Uid::effective();
  76     let running_gid = nix::unistd::Gid::effective();
  77
  78     if running_uid != backup_uid || running_gid != backup_gid {
  79         bail!(
  80             "proxy not running as backup user or group (got uid {} gid {})",
  81             running_uid,
  82             running_gid
  83         );
  84     }
  85
  86     proxmox_async::runtime::main(run())
  87 }
  88
  89 struct ProxmoxBackupProxyAdapter;
  90
  91 impl ServerAdapter for ProxmoxBackupProxyAdapter {
  92     fn get_index(
  93         &self,
  94         env: RestEnvironment,
  95         parts: Parts,
  96     ) -> Pin<Box<dyn Future<Output = Response<Body>> + Send>> {
  97         Box::pin(get_index_future(env, parts))
  98     }
  99
 100     fn check_auth<'a>(
 101         &'a self,
 102         headers: &'a HeaderMap,
 103         method: &'a Method,
 104     ) -> Pin<Box<dyn Future<Output = Result<(String, Box<dyn UserInformation + Sync + Send>), AuthError>> + Send + 'a>> {
 105         Box::pin(async move {
 106             check_pbs_auth(headers, method).await
 107         })
 108     }
 109 }
 110
 111 /// check for a cookie with the user-preferred language, fallback to the config one if not set or
 112 /// not existing
 113 fn get_language(headers: &http::HeaderMap) -> String {
 114     let exists = |l: &str| Path::new(&format!("/usr/share/pbs-i18n/pbs-lang-{}.js", l)).exists();
 115
 116     match cookie_from_header(headers, "PBSLangCookie") {
 117         Some(cookie_lang) if exists(&cookie_lang) => cookie_lang,
 118         _ => match proxmox_backup::config::node::config().map(|(cfg, _)| cfg.default_lang) {
 119             Ok(Some(default_lang)) if exists(&default_lang) => default_lang,
 120             _ => String::from(""),
 121         },
 122     }
 123 }
 124
 125 async fn get_index_future(env: RestEnvironment, parts: Parts) -> Response<Body> {
 126     let auth_id = env.get_auth_id();
 127     let api = env.api_config();
 128
 129     // fixme: make all IO async
 130
 131     let (userid, csrf_token) = match auth_id {
 132         Some(auth_id) => {
 133             let auth_id = auth_id.parse::<Authid>();
 134             match auth_id {
 135                 Ok(auth_id) if !auth_id.is_token() => {
 136                     let userid = auth_id.user().clone();
 137                     let new_csrf_token = assemble_csrf_prevention_token(csrf_secret(), &userid);
 138                     (Some(userid), Some(new_csrf_token))
 139                 }
 140                 _ => (None, None),
 141             }
 142         }
 143         None => (None, None),
 144     };
 145
 146     let nodename = proxmox_sys::nodename();
 147     let user = userid.as_ref().map(|u| u.as_str()).unwrap_or("");
 148
 149     let csrf_token = csrf_token.unwrap_or_else(|| String::from(""));
 150
 151     let mut debug = false;
 152     let mut template_file = "index";
 153
 154     if let Some(query_str) = parts.uri.query() {
 155         for (k, v) in form_urlencoded::parse(query_str.as_bytes()).into_owned() {
 156             if k == "debug" && v != "0" && v != "false" {
 157                 debug = true;
 158             } else if k == "console" {
 159                 template_file = "console";
 160             }
 161         }
 162     }
 163
 164     let data = json!({
 165         "NodeName": nodename,
 166         "UserName": user,
 167         "CSRFPreventionToken": csrf_token,
 168         "language": get_language(&parts.headers),
 169         "debug": debug,
 170     });
 171
 172     let (ct, index) = match api.render_template(template_file, &data) {
 173         Ok(index) => ("text/html", index),
 174         Err(err) => ("text/plain", format!("Error rendering template: {}", err)),
 175     };
 176
 177     let mut resp = Response::builder()
 178         .status(StatusCode::OK)
 179         .header(header::CONTENT_TYPE, ct)
 180         .body(index.into())
 181         .unwrap();
 182
 183     if let Some(userid) = userid {
 184         resp.extensions_mut().insert(Authid::from((userid, None)));
 185     }
 186
 187     resp
 188 }
 189
 190 async fn run() -> Result<(), Error> {
 191     // Note: To debug early connection error use
 192     // PROXMOX_DEBUG=1 ./target/release/proxmox-backup-proxy
 193     let debug = std::env::var("PROXMOX_DEBUG").is_ok();
 194
 195     if let Err(err) = syslog::init(
 196         syslog::Facility::LOG_DAEMON,
 197         if debug { log::LevelFilter::Debug } else { log::LevelFilter::Info },
 198         Some("proxmox-backup-proxy"),
 199     ) {
 200         bail!("unable to inititialize syslog - {}", err);
 201     }
 202
 203     let _ = public_auth_key(); // load with lazy_static
 204     let _ = csrf_secret(); // load with lazy_static
 205
 206     let rrd_cache = initialize_rrd_cache()?;
 207     rrd_cache.apply_journal()?;
 208
 209     let mut config = ApiConfig::new(
 210         pbs_buildcfg::JS_DIR,
 211         &proxmox_backup::api2::ROUTER,
 212         RpcEnvironmentType::PUBLIC,
 213         ProxmoxBackupProxyAdapter,
 214     )?;
 215
 216     config.add_alias("novnc", "/usr/share/novnc-pve");
 217     config.add_alias("extjs", "/usr/share/javascript/extjs");
 218     config.add_alias("qrcodejs", "/usr/share/javascript/qrcodejs");
 219     config.add_alias("fontawesome", "/usr/share/fonts-font-awesome");
 220     config.add_alias("xtermjs", "/usr/share/pve-xtermjs");
 221     config.add_alias("locale", "/usr/share/pbs-i18n");
 222     config.add_alias(
 223         "widgettoolkit",
 224         "/usr/share/javascript/proxmox-widget-toolkit",
 225     );
 226     config.add_alias("docs", "/usr/share/doc/proxmox-backup/html");
 227
 228     let mut indexpath = PathBuf::from(pbs_buildcfg::JS_DIR);
 229     indexpath.push("index.hbs");
 230     config.register_template("index", &indexpath)?;
 231     config.register_template("console", "/usr/share/pve-xtermjs/index.html.hbs")?;
 232
 233     let backup_user = pbs_config::backup_user()?;
 234     let mut commando_sock = proxmox_rest_server::CommandSocket::new(
 235         proxmox_rest_server::our_ctrl_sock(),
 236         backup_user.gid,
 237     );
 238
 239     let dir_opts = CreateOptions::new()
 240         .owner(backup_user.uid)
 241         .group(backup_user.gid);
 242     let file_opts = CreateOptions::new()
 243         .owner(backup_user.uid)
 244         .group(backup_user.gid);
 245
 246     config.enable_access_log(
 247         pbs_buildcfg::API_ACCESS_LOG_FN,
 248         Some(dir_opts.clone()),
 249         Some(file_opts.clone()),
 250         &mut commando_sock,
 251     )?;
 252
 253     config.enable_auth_log(
 254         pbs_buildcfg::API_AUTH_LOG_FN,
 255         Some(dir_opts.clone()),
 256         Some(file_opts.clone()),
 257         &mut commando_sock,
 258     )?;
 259
 260     let rest_server = RestServer::new(config);
 261     proxmox_rest_server::init_worker_tasks(
 262         pbs_buildcfg::PROXMOX_BACKUP_LOG_DIR_M!().into(),
 263         file_opts.clone(),
 264     )?;
 265
 266     //openssl req -x509 -newkey rsa:4096 -keyout /etc/proxmox-backup/proxy.key -out /etc/proxmox-backup/proxy.pem -nodes
 267
 268     // we build the initial acceptor here as we cannot start if this fails
 269     let acceptor = make_tls_acceptor()?;
 270     let acceptor = Arc::new(Mutex::new(acceptor));
 271
 272     // to renew the acceptor we just add a command-socket handler
 273     commando_sock.register_command("reload-certificate".to_string(), {
 274         let acceptor = Arc::clone(&acceptor);
 275         move |_value| -> Result<_, Error> {
 276             log::info!("reloading certificate");
 277             match make_tls_acceptor() {
 278                 Err(err) => log::error!("error reloading certificate: {}", err),
 279                 Ok(new_acceptor) => {
 280                     let mut guard = acceptor.lock().unwrap();
 281                     *guard = new_acceptor;
 282                 }
 283             }
 284             Ok(Value::Null)
 285         }
 286     })?;
 287
 288     // to remove references for not configured datastores
 289     commando_sock.register_command("datastore-removed".to_string(), |_value| {
 290         if let Err(err) = DataStore::remove_unused_datastores() {
 291             log::error!("could not refresh datastores: {}", err);
 292         }
 293         Ok(Value::Null)
 294     })?;
 295
 296     let server = daemon::create_daemon(([0, 0, 0, 0, 0, 0, 0, 0], 8007).into(), move |listener| {
 297         let connections = accept_connections(listener, acceptor, debug);
 298         let connections = hyper::server::accept::from_stream(ReceiverStream::new(connections));
 299
 300         Ok(async {
 301             daemon::systemd_notify(daemon::SystemdNotify::Ready)?;
 302
 303             hyper::Server::builder(connections)
 304                 .serve(rest_server)
 305                 .with_graceful_shutdown(proxmox_rest_server::shutdown_future())
 306                 .map_err(Error::from)
 307                 .await
 308         })
 309     });
 310
 311     proxmox_rest_server::write_pid(pbs_buildcfg::PROXMOX_BACKUP_PROXY_PID_FN)?;
 312
 313     let init_result: Result<(), Error> = try_block!({
 314         proxmox_rest_server::register_task_control_commands(&mut commando_sock)?;
 315         commando_sock.spawn()?;
 316         proxmox_rest_server::catch_shutdown_signal()?;
 317         proxmox_rest_server::catch_reload_signal()?;
 318         Ok(())
 319     });
 320
 321     if let Err(err) = init_result {
 322         bail!("unable to start daemon - {}", err);
 323     }
 324
 325     start_task_scheduler();
 326     start_stat_generator();
 327     start_traffic_control_updater();
 328
 329     server.await?;
 330     log::info!("server shutting down, waiting for active workers to complete");
 331     proxmox_rest_server::last_worker_future().await?;
 332     log::info!("done - exit server");
 333
 334     Ok(())
 335 }
 336
 337 fn make_tls_acceptor() -> Result<SslAcceptor, Error> {
 338     let key_path = configdir!("/proxy.key");
 339     let cert_path = configdir!("/proxy.pem");
 340
 341     let (config, _) = proxmox_backup::config::node::config()?;
 342     let ciphers_tls_1_3 = config.ciphers_tls_1_3;
 343     let ciphers_tls_1_2 = config.ciphers_tls_1_2;
 344
 345     let mut acceptor = SslAcceptor::mozilla_intermediate_v5(SslMethod::tls()).unwrap();
 346     if let Some(ciphers) = ciphers_tls_1_3.as_deref() {
 347         acceptor.set_ciphersuites(ciphers)?;
 348     }
 349     if let Some(ciphers) = ciphers_tls_1_2.as_deref() {
 350         acceptor.set_cipher_list(ciphers)?;
 351     }
 352     acceptor
 353         .set_private_key_file(key_path, SslFiletype::PEM)
 354         .map_err(|err| format_err!("unable to read proxy key {} - {}", key_path, err))?;
 355     acceptor
 356         .set_certificate_chain_file(cert_path)
 357         .map_err(|err| format_err!("unable to read proxy cert {} - {}", cert_path, err))?;
 358     acceptor.set_options(openssl::ssl::SslOptions::NO_RENEGOTIATION);
 359     acceptor.check_private_key().unwrap();
 360
 361     Ok(acceptor.build())
 362 }
 363
 364 type ClientStreamResult = Result<
 365     std::pin::Pin<Box<tokio_openssl::SslStream<RateLimitedStream<tokio::net::TcpStream>>>>,
 366     Error,
 367 >;
 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     let (sender, receiver) = tokio::sync::mpsc::channel(MAX_PENDING_ACCEPTS);
 376
 377     tokio::spawn(accept_connection(listener, acceptor, debug, sender));
 378
 379     receiver
 380 }
 381
 382 async fn accept_connection(
 383     listener: tokio::net::TcpListener,
 384     acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>,
 385     debug: bool,
 386     sender: tokio::sync::mpsc::Sender<ClientStreamResult>,
 387 ) {
 388     let accept_counter = Arc::new(());
 389
 390     loop {
 391         let (sock, peer) = match listener.accept().await {
 392             Ok(conn) => conn,
 393             Err(err) => {
 394                 eprintln!("error accepting tcp connection: {}", err);
 395                 continue;
 396             }
 397         };
 398
 399         sock.set_nodelay(true).unwrap();
 400         let _ = set_tcp_keepalive(sock.as_raw_fd(), PROXMOX_BACKUP_TCP_KEEPALIVE_TIME);
 401
 402         let sock =
 403             RateLimitedStream::with_limiter_update_cb(sock, move || lookup_rate_limiter(peer));
 404
 405         let ssl = {
 406             // 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!(
 414                         "failed to create Ssl object from Acceptor context - {}",
 415                         err
 416                     );
 417                     continue;
 418                 }
 419             }
 420         };
 421
 422         let stream = match tokio_openssl::SslStream::new(ssl, sock) {
 423             Ok(stream) => stream,
 424             Err(err) => {
 425                 eprintln!(
 426                     "failed to create SslStream using ssl and connection socket - {}",
 427                     err
 428                 );
 429                 continue;
 430             }
 431         };
 432
 433         let mut stream = Box::pin(stream);
 434         let sender = sender.clone();
 435
 436         if Arc::strong_count(&accept_counter) > MAX_PENDING_ACCEPTS {
 437             eprintln!("connection rejected - to many open connections");
 438             continue;
 439         }
 440
 441         let accept_counter = Arc::clone(&accept_counter);
 442         tokio::spawn(async move {
 443             let accept_future =
 444                 tokio::time::timeout(Duration::new(10, 0), stream.as_mut().accept());
 445
 446             let result = accept_future.await;
 447
 448             match result {
 449                 Ok(Ok(())) => {
 450                     if sender.send(Ok(stream)).await.is_err() && debug {
 451                         eprintln!("detect closed connection channel");
 452                     }
 453                 }
 454                 Ok(Err(err)) => {
 455                     if debug {
 456                         eprintln!("https handshake failed - {}", err);
 457                     }
 458                 }
 459                 Err(_) => {
 460                     if debug {
 461                         eprintln!("https handshake timeout");
 462                     }
 463                 }
 464             }
 465
 466             drop(accept_counter); // decrease reference count
 467         });
 468     }
 469 }
 470
 471 fn start_stat_generator() {
 472     let abort_future = proxmox_rest_server::shutdown_future();
 473     let future = Box::pin(run_stat_generator());
 474     let task = futures::future::select(future, abort_future);
 475     tokio::spawn(task.map(|_| ()));
 476 }
 477
 478 fn start_task_scheduler() {
 479     let abort_future = proxmox_rest_server::shutdown_future();
 480     let future = Box::pin(run_task_scheduler());
 481     let task = futures::future::select(future, abort_future);
 482     tokio::spawn(task.map(|_| ()));
 483 }
 484
 485 fn start_traffic_control_updater() {
 486     let abort_future = proxmox_rest_server::shutdown_future();
 487     let future = Box::pin(run_traffic_control_updater());
 488     let task = futures::future::select(future, abort_future);
 489     tokio::spawn(task.map(|_| ()));
 490 }
 491
 492 use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH};
 493
 494 fn next_minute() -> Result<Instant, Error> {
 495     let now = SystemTime::now();
 496     let epoch_now = now.duration_since(UNIX_EPOCH)?;
 497     let epoch_next = Duration::from_secs((epoch_now.as_secs() / 60 + 1) * 60);
 498     Ok(Instant::now() + epoch_next - epoch_now)
 499 }
 500
 501 async fn run_task_scheduler() {
 502     let mut count: usize = 0;
 503
 504     loop {
 505         count += 1;
 506
 507         let delay_target = match next_minute() {
 508             // try to run very minute
 509             Ok(d) => d,
 510             Err(err) => {
 511                 eprintln!("task scheduler: compute next minute failed - {}", err);
 512                 tokio::time::sleep_until(tokio::time::Instant::from_std(
 513                     Instant::now() + Duration::from_secs(60),
 514                 ))
 515                 .await;
 516                 continue;
 517             }
 518         };
 519
 520         if count > 2 {
 521             // wait 1..2 minutes before starting
 522             match schedule_tasks().catch_unwind().await {
 523                 Err(panic) => match panic.downcast::<&str>() {
 524                     Ok(msg) => {
 525                         eprintln!("task scheduler panic: {}", msg);
 526                     }
 527                     Err(_) => {
 528                         eprintln!("task scheduler panic - unknown type");
 529                     }
 530                 },
 531                 Ok(Err(err)) => {
 532                     eprintln!("task scheduler failed - {:?}", err);
 533                 }
 534                 Ok(Ok(_)) => {}
 535             }
 536         }
 537
 538         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
 539     }
 540 }
 541
 542 async fn schedule_tasks() -> Result<(), Error> {
 543     schedule_datastore_garbage_collection().await;
 544     schedule_datastore_prune().await;
 545     schedule_datastore_sync_jobs().await;
 546     schedule_datastore_verify_jobs().await;
 547     schedule_tape_backup_jobs().await;
 548     schedule_task_log_rotate().await;
 549
 550     Ok(())
 551 }
 552
 553 async fn schedule_datastore_garbage_collection() {
 554     let config = match pbs_config::datastore::config() {
 555         Err(err) => {
 556             eprintln!("unable to read datastore config - {}", err);
 557             return;
 558         }
 559         Ok((config, _digest)) => config,
 560     };
 561
 562     for (store, (_, store_config)) in config.sections {
 563         let datastore = match DataStore::lookup_datastore(&store, Some(Operation::Write)) {
 564             Ok(datastore) => datastore,
 565             Err(err) => {
 566                 eprintln!("lookup_datastore failed - {}", err);
 567                 continue;
 568             }
 569         };
 570
 571         let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
 572             Ok(c) => c,
 573             Err(err) => {
 574                 eprintln!("datastore config from_value failed - {}", err);
 575                 continue;
 576             }
 577         };
 578
 579         let event_str = match store_config.gc_schedule {
 580             Some(event_str) => event_str,
 581             None => continue,
 582         };
 583
 584         let event: CalendarEvent = match event_str.parse() {
 585             Ok(event) => event,
 586             Err(err) => {
 587                 eprintln!("unable to parse schedule '{}' - {}", event_str, err);
 588                 continue;
 589             }
 590         };
 591
 592         if datastore.garbage_collection_running() {
 593             continue;
 594         }
 595
 596         let worker_type = "garbage_collection";
 597
 598         let last = match jobstate::last_run_time(worker_type, &store) {
 599             Ok(time) => time,
 600             Err(err) => {
 601                 eprintln!(
 602                     "could not get last run time of {} {}: {}",
 603                     worker_type, store, err
 604                 );
 605                 continue;
 606             }
 607         };
 608
 609         let next = match event.compute_next_event(last) {
 610             Ok(Some(next)) => next,
 611             Ok(None) => continue,
 612             Err(err) => {
 613                 eprintln!("compute_next_event for '{}' failed - {}", event_str, err);
 614                 continue;
 615             }
 616         };
 617
 618         let now = proxmox_time::epoch_i64();
 619
 620         if next > now {
 621             continue;
 622         }
 623
 624         let job = match Job::new(worker_type, &store) {
 625             Ok(job) => job,
 626             Err(_) => continue, // could not get lock
 627         };
 628
 629         let auth_id = Authid::root_auth_id();
 630
 631         if let Err(err) = crate::server::do_garbage_collection_job(
 632             job,
 633             datastore,
 634             auth_id,
 635             Some(event_str),
 636             false,
 637         ) {
 638             eprintln!(
 639                 "unable to start garbage collection job on datastore {} - {}",
 640                 store, err
 641             );
 642         }
 643     }
 644 }
 645
 646 async fn schedule_datastore_prune() {
 647     let config = match pbs_config::datastore::config() {
 648         Err(err) => {
 649             eprintln!("unable to read datastore config - {}", err);
 650             return;
 651         }
 652         Ok((config, _digest)) => config,
 653     };
 654
 655     for (store, (_, store_config)) in config.sections {
 656         let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
 657             Ok(c) => c,
 658             Err(err) => {
 659                 eprintln!("datastore '{}' config from_value failed - {}", store, err);
 660                 continue;
 661             }
 662         };
 663
 664         let event_str = match store_config.prune_schedule {
 665             Some(event_str) => event_str,
 666             None => continue,
 667         };
 668
 669         let prune_options = PruneOptions {
 670             keep_last: store_config.keep_last,
 671             keep_hourly: store_config.keep_hourly,
 672             keep_daily: store_config.keep_daily,
 673             keep_weekly: store_config.keep_weekly,
 674             keep_monthly: store_config.keep_monthly,
 675             keep_yearly: store_config.keep_yearly,
 676         };
 677
 678         if !pbs_datastore::prune::keeps_something(&prune_options) {
 679             // no prune settings - keep all
 680             continue;
 681         }
 682
 683         let worker_type = "prune";
 684         if check_schedule(worker_type, &event_str, &store) {
 685             let job = match Job::new(worker_type, &store) {
 686                 Ok(job) => job,
 687                 Err(_) => continue, // could not get lock
 688             };
 689
 690             let auth_id = Authid::root_auth_id().clone();
 691             if let Err(err) =
 692                 do_prune_job(job, prune_options, store.clone(), &auth_id, Some(event_str))
 693             {
 694                 eprintln!("unable to start datastore prune job {} - {}", &store, err);
 695             }
 696         };
 697     }
 698 }
 699
 700 async fn schedule_datastore_sync_jobs() {
 701     let config = match pbs_config::sync::config() {
 702         Err(err) => {
 703             eprintln!("unable to read sync job config - {}", err);
 704             return;
 705         }
 706         Ok((config, _digest)) => config,
 707     };
 708
 709     for (job_id, (_, job_config)) in config.sections {
 710         let job_config: SyncJobConfig = match serde_json::from_value(job_config) {
 711             Ok(c) => c,
 712             Err(err) => {
 713                 eprintln!("sync job config from_value failed - {}", err);
 714                 continue;
 715             }
 716         };
 717
 718         let event_str = match job_config.schedule {
 719             Some(ref event_str) => event_str.clone(),
 720             None => continue,
 721         };
 722
 723         let worker_type = "syncjob";
 724         if check_schedule(worker_type, &event_str, &job_id) {
 725             let job = match Job::new(worker_type, &job_id) {
 726                 Ok(job) => job,
 727                 Err(_) => continue, // could not get lock
 728             };
 729
 730             let auth_id = Authid::root_auth_id().clone();
 731             if let Err(err) = do_sync_job(job, job_config, &auth_id, Some(event_str), false) {
 732                 eprintln!("unable to start datastore sync job {} - {}", &job_id, err);
 733             }
 734         };
 735     }
 736 }
 737
 738 async fn schedule_datastore_verify_jobs() {
 739     let config = match pbs_config::verify::config() {
 740         Err(err) => {
 741             eprintln!("unable to read verification job config - {}", err);
 742             return;
 743         }
 744         Ok((config, _digest)) => config,
 745     };
 746     for (job_id, (_, job_config)) in config.sections {
 747         let job_config: VerificationJobConfig = match serde_json::from_value(job_config) {
 748             Ok(c) => c,
 749             Err(err) => {
 750                 eprintln!("verification job config from_value failed - {}", err);
 751                 continue;
 752             }
 753         };
 754         let event_str = match job_config.schedule {
 755             Some(ref event_str) => event_str.clone(),
 756             None => continue,
 757         };
 758
 759         let worker_type = "verificationjob";
 760         let auth_id = Authid::root_auth_id().clone();
 761         if check_schedule(worker_type, &event_str, &job_id) {
 762             let job = match Job::new(worker_type, &job_id) {
 763                 Ok(job) => job,
 764                 Err(_) => continue, // could not get lock
 765             };
 766             if let Err(err) = do_verification_job(job, job_config, &auth_id, Some(event_str), false)
 767             {
 768                 eprintln!(
 769                     "unable to start datastore verification job {} - {}",
 770                     &job_id, err
 771                 );
 772             }
 773         };
 774     }
 775 }
 776
 777 async fn schedule_tape_backup_jobs() {
 778     let config = match pbs_config::tape_job::config() {
 779         Err(err) => {
 780             eprintln!("unable to read tape job config - {}", err);
 781             return;
 782         }
 783         Ok((config, _digest)) => config,
 784     };
 785     for (job_id, (_, job_config)) in config.sections {
 786         let job_config: TapeBackupJobConfig = match serde_json::from_value(job_config) {
 787             Ok(c) => c,
 788             Err(err) => {
 789                 eprintln!("tape backup job config from_value failed - {}", err);
 790                 continue;
 791             }
 792         };
 793         let event_str = match job_config.schedule {
 794             Some(ref event_str) => event_str.clone(),
 795             None => continue,
 796         };
 797
 798         let worker_type = "tape-backup-job";
 799         let auth_id = Authid::root_auth_id().clone();
 800         if check_schedule(worker_type, &event_str, &job_id) {
 801             let job = match Job::new(worker_type, &job_id) {
 802                 Ok(job) => job,
 803                 Err(_) => continue, // could not get lock
 804             };
 805             if let Err(err) =
 806                 do_tape_backup_job(job, job_config.setup, &auth_id, Some(event_str), false)
 807             {
 808                 eprintln!("unable to start tape backup job {} - {}", &job_id, err);
 809             }
 810         };
 811     }
 812 }
 813
 814 async fn schedule_task_log_rotate() {
 815     let worker_type = "logrotate";
 816     let job_id = "access-log_and_task-archive";
 817
 818     // schedule daily at 00:00 like normal logrotate
 819     let schedule = "00:00";
 820
 821     if !check_schedule(worker_type, schedule, job_id) {
 822         // if we never ran the rotation, schedule instantly
 823         match jobstate::JobState::load(worker_type, job_id) {
 824             Ok(state) => match state {
 825                 jobstate::JobState::Created { .. } => {}
 826                 _ => return,
 827             },
 828             _ => return,
 829         }
 830     }
 831
 832     let mut job = match Job::new(worker_type, job_id) {
 833         Ok(job) => job,
 834         Err(_) => return, // could not get lock
 835     };
 836
 837     if let Err(err) = WorkerTask::new_thread(
 838         worker_type,
 839         None,
 840         Authid::root_auth_id().to_string(),
 841         false,
 842         move |worker| {
 843             job.start(&worker.upid().to_string())?;
 844             task_log!(worker, "starting task log rotation");
 845
 846             let result = try_block!({
 847                 let max_size = 512 * 1024 - 1; // an entry has ~ 100b, so > 5000 entries/file
 848                 let max_files = 20; // times twenty files gives > 100000 task entries
 849
 850                 let max_days = proxmox_backup::config::node::config()
 851                     .map(|(cfg, _)| cfg.task_log_max_days)
 852                     .ok()
 853                     .flatten();
 854
 855                 let user = pbs_config::backup_user()?;
 856                 let options = proxmox_sys::fs::CreateOptions::new()
 857                     .owner(user.uid)
 858                     .group(user.gid);
 859
 860                 let has_rotated = rotate_task_log_archive(
 861                     max_size,
 862                     true,
 863                     Some(max_files),
 864                     max_days,
 865                     Some(options.clone()),
 866                 )?;
 867
 868                 if has_rotated {
 869                     task_log!(worker, "task log archive was rotated");
 870                 } else {
 871                     task_log!(worker, "task log archive was not rotated");
 872                 }
 873
 874                 let max_size = 32 * 1024 * 1024 - 1;
 875                 let max_files = 14;
 876
 877                 let mut logrotate = LogRotate::new(
 878                     pbs_buildcfg::API_ACCESS_LOG_FN,
 879                     true,
 880                     Some(max_files),
 881                     Some(options.clone()),
 882                 )?;
 883
 884                 if logrotate.rotate(max_size)? {
 885                     println!("rotated access log, telling daemons to re-open log file");
 886                     proxmox_async::runtime::block_on(command_reopen_access_logfiles())?;
 887                     task_log!(worker, "API access log was rotated");
 888                 } else {
 889                     task_log!(worker, "API access log was not rotated");
 890                 }
 891
 892                 let mut logrotate = LogRotate::new(
 893                     pbs_buildcfg::API_AUTH_LOG_FN,
 894                     true,
 895                     Some(max_files),
 896                     Some(options),
 897                 )?;
 898
 899                 if logrotate.rotate(max_size)? {
 900                     println!("rotated auth log, telling daemons to re-open log file");
 901                     proxmox_async::runtime::block_on(command_reopen_auth_logfiles())?;
 902                     task_log!(worker, "API authentication log was rotated");
 903                 } else {
 904                     task_log!(worker, "API authentication log was not rotated");
 905                 }
 906
 907                 if has_rotated {
 908                     task_log!(worker, "cleaning up old task logs");
 909                     if let Err(err) = cleanup_old_tasks(&worker, true) {
 910                         task_warn!(worker, "could not completely cleanup old tasks: {}", err);
 911                     }
 912                 }
 913
 914                 Ok(())
 915             });
 916
 917             let status = worker.create_state(&result);
 918
 919             if let Err(err) = job.finish(status) {
 920                 eprintln!("could not finish job state for {}: {}", worker_type, err);
 921             }
 922
 923             result
 924         },
 925     ) {
 926         eprintln!("unable to start task log rotation: {}", err);
 927     }
 928 }
 929
 930 async fn command_reopen_access_logfiles() -> Result<(), Error> {
 931     // only care about the most recent daemon instance for each, proxy & api, as other older ones
 932     // should not respond to new requests anyway, but only finish their current one and then exit.
 933     let sock = proxmox_rest_server::our_ctrl_sock();
 934     let f1 =
 935         proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-access-log-reopen\"}\n");
 936
 937     let pid = proxmox_rest_server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?;
 938     let sock = proxmox_rest_server::ctrl_sock_from_pid(pid);
 939     let f2 =
 940         proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-access-log-reopen\"}\n");
 941
 942     match futures::join!(f1, f2) {
 943         (Err(e1), Err(e2)) => Err(format_err!(
 944             "reopen commands failed, proxy: {}; api: {}",
 945             e1,
 946             e2
 947         )),
 948         (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)),
 949         (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)),
 950         _ => Ok(()),
 951     }
 952 }
 953
 954 async fn command_reopen_auth_logfiles() -> Result<(), Error> {
 955     // only care about the most recent daemon instance for each, proxy & api, as other older ones
 956     // should not respond to new requests anyway, but only finish their current one and then exit.
 957     let sock = proxmox_rest_server::our_ctrl_sock();
 958     let f1 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n");
 959
 960     let pid = proxmox_rest_server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?;
 961     let sock = proxmox_rest_server::ctrl_sock_from_pid(pid);
 962     let f2 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n");
 963
 964     match futures::join!(f1, f2) {
 965         (Err(e1), Err(e2)) => Err(format_err!(
 966             "reopen commands failed, proxy: {}; api: {}",
 967             e1,
 968             e2
 969         )),
 970         (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)),
 971         (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)),
 972         _ => Ok(()),
 973     }
 974 }
 975
 976 async fn run_stat_generator() {
 977     loop {
 978         let delay_target = Instant::now() + Duration::from_secs(10);
 979
 980         generate_host_stats().await;
 981
 982         rrd_sync_journal();
 983
 984         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
 985     }
 986 }
 987
 988 async fn generate_host_stats() {
 989     match tokio::task::spawn_blocking(generate_host_stats_sync).await {
 990         Ok(()) => (),
 991         Err(err) => log::error!("generate_host_stats paniced: {}", err),
 992     }
 993 }
 994
 995 fn generate_host_stats_sync() {
 996     use proxmox_sys::linux::procfs::{
 997         read_loadavg, read_meminfo, read_proc_net_dev, read_proc_stat,
 998     };
 999
1000     match read_proc_stat() {
1001         Ok(stat) => {
1002             rrd_update_gauge("host/cpu", stat.cpu);
1003             rrd_update_gauge("host/iowait", stat.iowait_percent);
1004         }
1005         Err(err) => {
1006             eprintln!("read_proc_stat failed - {}", err);
1007         }
1008     }
1009
1010     match read_meminfo() {
1011         Ok(meminfo) => {
1012             rrd_update_gauge("host/memtotal", meminfo.memtotal as f64);
1013             rrd_update_gauge("host/memused", meminfo.memused as f64);
1014             rrd_update_gauge("host/swaptotal", meminfo.swaptotal as f64);
1015             rrd_update_gauge("host/swapused", meminfo.swapused as f64);
1016         }
1017         Err(err) => {
1018             eprintln!("read_meminfo failed - {}", err);
1019         }
1020     }
1021
1022     match read_proc_net_dev() {
1023         Ok(netdev) => {
1024             use pbs_config::network::is_physical_nic;
1025             let mut netin = 0;
1026             let mut netout = 0;
1027             for item in netdev {
1028                 if !is_physical_nic(&item.device) {
1029                     continue;
1030                 }
1031                 netin += item.receive;
1032                 netout += item.send;
1033             }
1034             rrd_update_derive("host/netin", netin as f64);
1035             rrd_update_derive("host/netout", netout as f64);
1036         }
1037         Err(err) => {
1038             eprintln!("read_prox_net_dev failed - {}", err);
1039         }
1040     }
1041
1042     match read_loadavg() {
1043         Ok(loadavg) => {
1044             rrd_update_gauge("host/loadavg", loadavg.0 as f64);
1045         }
1046         Err(err) => {
1047             eprintln!("read_loadavg failed - {}", err);
1048         }
1049     }
1050
1051     let disk_manager = DiskManage::new();
1052
1053     gather_disk_stats(disk_manager.clone(), Path::new("/"), "host");
1054
1055     match pbs_config::datastore::config() {
1056         Ok((config, _)) => {
1057             let datastore_list: Vec<DataStoreConfig> = config
1058                 .convert_to_typed_array("datastore")
1059                 .unwrap_or_default();
1060
1061             for config in datastore_list {
1062                 let rrd_prefix = format!("datastore/{}", config.name);
1063                 let path = std::path::Path::new(&config.path);
1064                 gather_disk_stats(disk_manager.clone(), path, &rrd_prefix);
1065             }
1066         }
1067         Err(err) => {
1068             eprintln!("read datastore config failed - {}", err);
1069         }
1070     }
1071 }
1072
1073 fn check_schedule(worker_type: &str, event_str: &str, id: &str) -> bool {
1074     let event: CalendarEvent = match event_str.parse() {
1075         Ok(event) => event,
1076         Err(err) => {
1077             eprintln!("unable to parse schedule '{}' - {}", event_str, err);
1078             return false;
1079         }
1080     };
1081
1082     let last = match jobstate::last_run_time(worker_type, id) {
1083         Ok(time) => time,
1084         Err(err) => {
1085             eprintln!(
1086                 "could not get last run time of {} {}: {}",
1087                 worker_type, id, err
1088             );
1089             return false;
1090         }
1091     };
1092
1093     let next = match event.compute_next_event(last) {
1094         Ok(Some(next)) => next,
1095         Ok(None) => return false,
1096         Err(err) => {
1097             eprintln!("compute_next_event for '{}' failed - {}", event_str, err);
1098             return false;
1099         }
1100     };
1101
1102     let now = proxmox_time::epoch_i64();
1103     next <= now
1104 }
1105
1106 fn gather_disk_stats(disk_manager: Arc<DiskManage>, path: &Path, rrd_prefix: &str) {
1107     match proxmox_backup::tools::disks::disk_usage(path) {
1108         Ok(status) => {
1109             let rrd_key = format!("{}/total", rrd_prefix);
1110             rrd_update_gauge(&rrd_key, status.total as f64);
1111             let rrd_key = format!("{}/used", rrd_prefix);
1112             rrd_update_gauge(&rrd_key, status.used as f64);
1113         }
1114         Err(err) => {
1115             eprintln!("read disk_usage on {:?} failed - {}", path, err);
1116         }
1117     }
1118
1119     match disk_manager.find_mounted_device(path) {
1120         Ok(None) => {}
1121         Ok(Some((fs_type, device, source))) => {
1122             let mut device_stat = None;
1123             match (fs_type.as_str(), source) {
1124                 ("zfs", Some(source)) => match source.into_string() {
1125                     Ok(dataset) => match zfs_dataset_stats(&dataset) {
1126                         Ok(stat) => device_stat = Some(stat),
1127                         Err(err) => eprintln!("zfs_dataset_stats({:?}) failed - {}", dataset, err),
1128                     },
1129                     Err(source) => {
1130                         eprintln!("zfs_pool_stats({:?}) failed - invalid characters", source)
1131                     }
1132                 },
1133                 _ => {
1134                     if let Ok(disk) = disk_manager.clone().disk_by_dev_num(device.into_dev_t()) {
1135                         match disk.read_stat() {
1136                             Ok(stat) => device_stat = stat,
1137                             Err(err) => eprintln!("disk.read_stat {:?} failed - {}", path, err),
1138                         }
1139                     }
1140                 }
1141             }
1142             if let Some(stat) = device_stat {
1143                 let rrd_key = format!("{}/read_ios", rrd_prefix);
1144                 rrd_update_derive(&rrd_key, stat.read_ios as f64);
1145                 let rrd_key = format!("{}/read_bytes", rrd_prefix);
1146                 rrd_update_derive(&rrd_key, (stat.read_sectors * 512) as f64);
1147
1148                 let rrd_key = format!("{}/write_ios", rrd_prefix);
1149                 rrd_update_derive(&rrd_key, stat.write_ios as f64);
1150                 let rrd_key = format!("{}/write_bytes", rrd_prefix);
1151                 rrd_update_derive(&rrd_key, (stat.write_sectors * 512) as f64);
1152
1153                 let rrd_key = format!("{}/io_ticks", rrd_prefix);
1154                 rrd_update_derive(&rrd_key, (stat.io_ticks as f64) / 1000.0);
1155             }
1156         }
1157         Err(err) => {
1158             eprintln!("find_mounted_device failed - {}", err);
1159         }
1160     }
1161 }
1162
1163 // Rate Limiter lookup
1164
1165 // Test WITH
1166 // proxmox-backup-client restore vm/201/2021-10-22T09:55:56Z drive-scsi0.img img1.img --repository localhost:store2
1167
1168 async fn run_traffic_control_updater() {
1169     loop {
1170         let delay_target = Instant::now() + Duration::from_secs(1);
1171
1172         {
1173             let mut cache = TRAFFIC_CONTROL_CACHE.lock().unwrap();
1174             cache.compute_current_rates();
1175         }
1176
1177         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
1178     }
1179 }
1180
1181 fn lookup_rate_limiter(
1182     peer: std::net::SocketAddr,
1183 ) -> (
1184     Option<Arc<dyn ShareableRateLimit>>,
1185     Option<Arc<dyn ShareableRateLimit>>,
1186 ) {
1187     let mut cache = TRAFFIC_CONTROL_CACHE.lock().unwrap();
1188
1189     let now = proxmox_time::epoch_i64();
1190
1191     cache.reload(now);
1192
1193     let (_rule_name, read_limiter, write_limiter) = cache.lookup_rate_limiter(peer, now);
1194
1195     (read_limiter, write_limiter)
1196 }