src/backup/datastore.rs

   1 use std::collections::{HashSet, HashMap};
   2 use std::io::{self, Write};
   3 use std::path::{Path, PathBuf};
   4 use std::sync::{Arc, Mutex};
   5 use std::convert::TryFrom;
   6 use std::str::FromStr;
   7 use std::time::Duration;
   8
   9 use anyhow::{bail, format_err, Error};
  10 use lazy_static::lazy_static;
  11
  12 use proxmox::tools::fs::{replace_file, file_read_optional_string, CreateOptions};
  13
  14 use pbs_api_types::{UPID, DataStoreConfig, Authid, GarbageCollectionStatus};
  15 use pbs_datastore::{task_log, task_warn};
  16 use pbs_datastore::DataBlob;
  17 use pbs_datastore::backup_info::{BackupGroup, BackupDir};
  18 use pbs_datastore::chunk_store::ChunkStore;
  19 use pbs_datastore::dynamic_index::{DynamicIndexReader, DynamicIndexWriter};
  20 use pbs_datastore::fixed_index::{FixedIndexReader, FixedIndexWriter};
  21 use pbs_datastore::index::IndexFile;
  22 use pbs_datastore::manifest::{
  23     MANIFEST_BLOB_NAME, MANIFEST_LOCK_NAME, CLIENT_LOG_BLOB_NAME,
  24     ArchiveType, BackupManifest,
  25     archive_type,
  26 };
  27 use pbs_datastore::task::TaskState;
  28 use pbs_tools::format::HumanByte;
  29 use pbs_tools::fs::{lock_dir_noblock, DirLockGuard};
  30 use pbs_tools::process_locker::ProcessLockSharedGuard;
  31 use pbs_config::{open_backup_lockfile, BackupLockGuard};
  32
  33 use crate::tools::fail_on_shutdown;
  34
  35 lazy_static! {
  36     static ref DATASTORE_MAP: Mutex<HashMap<String, Arc<DataStore>>> = Mutex::new(HashMap::new());
  37 }
  38
  39 /// checks if auth_id is owner, or, if owner is a token, if
  40 /// auth_id is the user of the token
  41 pub fn check_backup_owner(
  42     owner: &Authid,
  43     auth_id: &Authid,
  44 ) -> Result<(), Error> {
  45     let correct_owner = owner == auth_id
  46         || (owner.is_token() && &Authid::from(owner.user().clone()) == auth_id);
  47     if !correct_owner {
  48         bail!("backup owner check failed ({} != {})", auth_id, owner);
  49     }
  50     Ok(())
  51 }
  52
  53 /// Datastore Management
  54 ///
  55 /// A Datastore can store severals backups, and provides the
  56 /// management interface for backup.
  57 pub struct DataStore {
  58     chunk_store: Arc<ChunkStore>,
  59     gc_mutex: Mutex<()>,
  60     last_gc_status: Mutex<GarbageCollectionStatus>,
  61     verify_new: bool,
  62 }
  63
  64 impl DataStore {
  65
  66     pub fn lookup_datastore(name: &str) -> Result<Arc<DataStore>, Error> {
  67
  68         let (config, _digest) = pbs_config::datastore::config()?;
  69         let config: DataStoreConfig = config.lookup("datastore", name)?;
  70         let path = PathBuf::from(&config.path);
  71
  72         let mut map = DATASTORE_MAP.lock().unwrap();
  73
  74         if let Some(datastore) = map.get(name) {
  75             // Compare Config - if changed, create new Datastore object!
  76             if datastore.chunk_store.base() == path &&
  77                 datastore.verify_new == config.verify_new.unwrap_or(false)
  78             {
  79                 return Ok(datastore.clone());
  80             }
  81         }
  82
  83         let datastore = DataStore::open_with_path(name, &path, config)?;
  84
  85         let datastore = Arc::new(datastore);
  86         map.insert(name.to_string(), datastore.clone());
  87
  88         Ok(datastore)
  89     }
  90
  91     /// removes all datastores that are not configured anymore
  92     pub fn remove_unused_datastores() -> Result<(), Error>{
  93         let (config, _digest) = pbs_config::datastore::config()?;
  94
  95         let mut map = DATASTORE_MAP.lock().unwrap();
  96         // removes all elements that are not in the config
  97         map.retain(|key, _| {
  98             config.sections.contains_key(key)
  99         });
 100         Ok(())
 101     }
 102
 103     fn open_with_path(store_name: &str, path: &Path, config: DataStoreConfig) -> Result<Self, Error> {
 104         let chunk_store = ChunkStore::open(store_name, path)?;
 105
 106         let mut gc_status_path = chunk_store.base_path();
 107         gc_status_path.push(".gc-status");
 108
 109         let gc_status = if let Some(state) = file_read_optional_string(gc_status_path)? {
 110             match serde_json::from_str(&state) {
 111                 Ok(state) => state,
 112                 Err(err) => {
 113                     eprintln!("error reading gc-status: {}", err);
 114                     GarbageCollectionStatus::default()
 115                 }
 116             }
 117         } else {
 118             GarbageCollectionStatus::default()
 119         };
 120
 121         Ok(Self {
 122             chunk_store: Arc::new(chunk_store),
 123             gc_mutex: Mutex::new(()),
 124             last_gc_status: Mutex::new(gc_status),
 125             verify_new: config.verify_new.unwrap_or(false),
 126         })
 127     }
 128
 129     pub fn get_chunk_iterator(
 130         &self,
 131     ) -> Result<
 132         impl Iterator<Item = (Result<pbs_tools::fs::ReadDirEntry, Error>, usize, bool)>,
 133         Error
 134     > {
 135         self.chunk_store.get_chunk_iterator()
 136     }
 137
 138     pub fn create_fixed_writer<P: AsRef<Path>>(&self, filename: P, size: usize, chunk_size: usize) -> Result<FixedIndexWriter, Error> {
 139
 140         let index = FixedIndexWriter::create(self.chunk_store.clone(), filename.as_ref(), size, chunk_size)?;
 141
 142         Ok(index)
 143     }
 144
 145     pub fn open_fixed_reader<P: AsRef<Path>>(&self, filename: P) -> Result<FixedIndexReader, Error> {
 146
 147         let full_path =  self.chunk_store.relative_path(filename.as_ref());
 148
 149         let index = FixedIndexReader::open(&full_path)?;
 150
 151         Ok(index)
 152     }
 153
 154     pub fn create_dynamic_writer<P: AsRef<Path>>(
 155         &self, filename: P,
 156     ) -> Result<DynamicIndexWriter, Error> {
 157
 158         let index = DynamicIndexWriter::create(
 159             self.chunk_store.clone(), filename.as_ref())?;
 160
 161         Ok(index)
 162     }
 163
 164     pub fn open_dynamic_reader<P: AsRef<Path>>(&self, filename: P) -> Result<DynamicIndexReader, Error> {
 165
 166         let full_path =  self.chunk_store.relative_path(filename.as_ref());
 167
 168         let index = DynamicIndexReader::open(&full_path)?;
 169
 170         Ok(index)
 171     }
 172
 173     pub fn open_index<P>(&self, filename: P) -> Result<Box<dyn IndexFile + Send>, Error>
 174     where
 175         P: AsRef<Path>,
 176     {
 177         let filename = filename.as_ref();
 178         let out: Box<dyn IndexFile + Send> =
 179             match archive_type(filename)? {
 180                 ArchiveType::DynamicIndex => Box::new(self.open_dynamic_reader(filename)?),
 181                 ArchiveType::FixedIndex => Box::new(self.open_fixed_reader(filename)?),
 182                 _ => bail!("cannot open index file of unknown type: {:?}", filename),
 183             };
 184         Ok(out)
 185     }
 186
 187     /// Fast index verification - only check if chunks exists
 188     pub fn fast_index_verification(
 189         &self,
 190         index: &dyn IndexFile,
 191         checked: &mut HashSet<[u8;32]>,
 192     ) -> Result<(), Error> {
 193
 194         for pos in 0..index.index_count() {
 195             let info = index.chunk_info(pos).unwrap();
 196             if checked.contains(&info.digest) {
 197                 continue;
 198             }
 199
 200             self.stat_chunk(&info.digest).
 201                 map_err(|err| {
 202                     format_err!(
 203                         "fast_index_verification error, stat_chunk {} failed - {}",
 204                         proxmox::tools::digest_to_hex(&info.digest),
 205                         err,
 206                     )
 207                 })?;
 208
 209             checked.insert(info.digest);
 210         }
 211
 212         Ok(())
 213     }
 214
 215     pub fn name(&self) -> &str {
 216         self.chunk_store.name()
 217     }
 218
 219     pub fn base_path(&self) -> PathBuf {
 220         self.chunk_store.base_path()
 221     }
 222
 223     /// Cleanup a backup directory
 224     ///
 225     /// Removes all files not mentioned in the manifest.
 226     pub fn cleanup_backup_dir(&self, backup_dir: &BackupDir, manifest: &BackupManifest
 227     ) ->  Result<(), Error> {
 228
 229         let mut full_path = self.base_path();
 230         full_path.push(backup_dir.relative_path());
 231
 232         let mut wanted_files = HashSet::new();
 233         wanted_files.insert(MANIFEST_BLOB_NAME.to_string());
 234         wanted_files.insert(CLIENT_LOG_BLOB_NAME.to_string());
 235         manifest.files().iter().for_each(|item| { wanted_files.insert(item.filename.clone()); });
 236
 237         for item in pbs_tools::fs::read_subdir(libc::AT_FDCWD, &full_path)? {
 238             if let Ok(item) = item {
 239                 if let Some(file_type) = item.file_type() {
 240                     if file_type != nix::dir::Type::File { continue; }
 241                 }
 242                 let file_name = item.file_name().to_bytes();
 243                 if file_name == b"." || file_name == b".." { continue; };
 244
 245                 if let Ok(name) = std::str::from_utf8(file_name) {
 246                     if wanted_files.contains(name) { continue; }
 247                 }
 248                 println!("remove unused file {:?}", item.file_name());
 249                 let dirfd = item.parent_fd();
 250                 let _res = unsafe { libc::unlinkat(dirfd, item.file_name().as_ptr(), 0) };
 251             }
 252         }
 253
 254         Ok(())
 255     }
 256
 257     /// Returns the absolute path for a backup_group
 258     pub fn group_path(&self, backup_group: &BackupGroup) -> PathBuf {
 259         let mut full_path = self.base_path();
 260         full_path.push(backup_group.group_path());
 261         full_path
 262     }
 263
 264     /// Returns the absolute path for backup_dir
 265     pub fn snapshot_path(&self, backup_dir: &BackupDir) -> PathBuf {
 266         let mut full_path = self.base_path();
 267         full_path.push(backup_dir.relative_path());
 268         full_path
 269     }
 270
 271     /// Remove a complete backup group including all snapshots
 272     pub fn remove_backup_group(&self, backup_group: &BackupGroup) ->  Result<(), Error> {
 273
 274         let full_path = self.group_path(backup_group);
 275
 276         let _guard = pbs_tools::fs::lock_dir_noblock(&full_path, "backup group", "possible running backup")?;
 277
 278         log::info!("removing backup group {:?}", full_path);
 279
 280         // remove all individual backup dirs first to ensure nothing is using them
 281         for snap in backup_group.list_backups(&self.base_path())? {
 282             self.remove_backup_dir(&snap.backup_dir, false)?;
 283         }
 284
 285         // no snapshots left, we can now safely remove the empty folder
 286         std::fs::remove_dir_all(&full_path)
 287             .map_err(|err| {
 288                 format_err!(
 289                     "removing backup group directory {:?} failed - {}",
 290                     full_path,
 291                     err,
 292                 )
 293             })?;
 294
 295         Ok(())
 296     }
 297
 298     /// Remove a backup directory including all content
 299     pub fn remove_backup_dir(&self, backup_dir: &BackupDir, force: bool) ->  Result<(), Error> {
 300
 301         let full_path = self.snapshot_path(backup_dir);
 302
 303         let (_guard, _manifest_guard);
 304         if !force {
 305             _guard = lock_dir_noblock(&full_path, "snapshot", "possibly running or in use")?;
 306             _manifest_guard = self.lock_manifest(backup_dir)?;
 307         }
 308
 309         log::info!("removing backup snapshot {:?}", full_path);
 310         std::fs::remove_dir_all(&full_path)
 311             .map_err(|err| {
 312                 format_err!(
 313                     "removing backup snapshot {:?} failed - {}",
 314                     full_path,
 315                     err,
 316                 )
 317             })?;
 318
 319         // the manifest does not exists anymore, we do not need to keep the lock
 320         if let Ok(path) = self.manifest_lock_path(backup_dir) {
 321             // ignore errors
 322             let _ = std::fs::remove_file(path);
 323         }
 324
 325         Ok(())
 326     }
 327
 328     /// Returns the time of the last successful backup
 329     ///
 330     /// Or None if there is no backup in the group (or the group dir does not exist).
 331     pub fn last_successful_backup(&self, backup_group: &BackupGroup) -> Result<Option<i64>, Error> {
 332         let base_path = self.base_path();
 333         let mut group_path = base_path.clone();
 334         group_path.push(backup_group.group_path());
 335
 336         if group_path.exists() {
 337             backup_group.last_successful_backup(&base_path)
 338         } else {
 339             Ok(None)
 340         }
 341     }
 342
 343     /// Returns the backup owner.
 344     ///
 345     /// The backup owner is the entity who first created the backup group.
 346     pub fn get_owner(&self, backup_group: &BackupGroup) -> Result<Authid, Error> {
 347         let mut full_path = self.base_path();
 348         full_path.push(backup_group.group_path());
 349         full_path.push("owner");
 350         let owner = proxmox::tools::fs::file_read_firstline(full_path)?;
 351         Ok(owner.trim_end().parse()?) // remove trailing newline
 352     }
 353
 354     pub fn owns_backup(&self, backup_group: &BackupGroup, auth_id: &Authid) -> Result<bool, Error> {
 355         let owner = self.get_owner(backup_group)?;
 356
 357         Ok(check_backup_owner(&owner, auth_id).is_ok())
 358     }
 359
 360     /// Set the backup owner.
 361     pub fn set_owner(
 362         &self,
 363         backup_group: &BackupGroup,
 364         auth_id: &Authid,
 365         force: bool,
 366     ) -> Result<(), Error> {
 367         let mut path = self.base_path();
 368         path.push(backup_group.group_path());
 369         path.push("owner");
 370
 371         let mut open_options = std::fs::OpenOptions::new();
 372         open_options.write(true);
 373         open_options.truncate(true);
 374
 375         if force {
 376             open_options.create(true);
 377         } else {
 378             open_options.create_new(true);
 379         }
 380
 381         let mut file = open_options.open(&path)
 382             .map_err(|err| format_err!("unable to create owner file {:?} - {}", path, err))?;
 383
 384         writeln!(file, "{}", auth_id)
 385             .map_err(|err| format_err!("unable to write owner file  {:?} - {}", path, err))?;
 386
 387         Ok(())
 388     }
 389
 390     /// Create (if it does not already exists) and lock a backup group
 391     ///
 392     /// And set the owner to 'userid'. If the group already exists, it returns the
 393     /// current owner (instead of setting the owner).
 394     ///
 395     /// This also acquires an exclusive lock on the directory and returns the lock guard.
 396     pub fn create_locked_backup_group(
 397         &self,
 398         backup_group: &BackupGroup,
 399         auth_id: &Authid,
 400     ) -> Result<(Authid, DirLockGuard), Error> {
 401         // create intermediate path first:
 402         let mut full_path = self.base_path();
 403         full_path.push(backup_group.backup_type());
 404         std::fs::create_dir_all(&full_path)?;
 405
 406         full_path.push(backup_group.backup_id());
 407
 408         // create the last component now
 409         match std::fs::create_dir(&full_path) {
 410             Ok(_) => {
 411                 let guard = lock_dir_noblock(&full_path, "backup group", "another backup is already running")?;
 412                 self.set_owner(backup_group, auth_id, false)?;
 413                 let owner = self.get_owner(backup_group)?; // just to be sure
 414                 Ok((owner, guard))
 415             }
 416             Err(ref err) if err.kind() == io::ErrorKind::AlreadyExists => {
 417                 let guard = lock_dir_noblock(&full_path, "backup group", "another backup is already running")?;
 418                 let owner = self.get_owner(backup_group)?; // just to be sure
 419                 Ok((owner, guard))
 420             }
 421             Err(err) => bail!("unable to create backup group {:?} - {}", full_path, err),
 422         }
 423     }
 424
 425     /// Creates a new backup snapshot inside a BackupGroup
 426     ///
 427     /// The BackupGroup directory needs to exist.
 428     pub fn create_locked_backup_dir(&self, backup_dir: &BackupDir)
 429         -> Result<(PathBuf, bool, DirLockGuard), Error>
 430     {
 431         let relative_path = backup_dir.relative_path();
 432         let mut full_path = self.base_path();
 433         full_path.push(&relative_path);
 434
 435         let lock = ||
 436             lock_dir_noblock(&full_path, "snapshot", "internal error - tried creating snapshot that's already in use");
 437
 438         match std::fs::create_dir(&full_path) {
 439             Ok(_) => Ok((relative_path, true, lock()?)),
 440             Err(ref e) if e.kind() == io::ErrorKind::AlreadyExists => Ok((relative_path, false, lock()?)),
 441             Err(e) => Err(e.into())
 442         }
 443     }
 444
 445     pub fn list_images(&self) -> Result<Vec<PathBuf>, Error> {
 446         let base = self.base_path();
 447
 448         let mut list = vec![];
 449
 450         use walkdir::WalkDir;
 451
 452         let walker = WalkDir::new(&base).into_iter();
 453
 454         // make sure we skip .chunks (and other hidden files to keep it simple)
 455         fn is_hidden(entry: &walkdir::DirEntry) -> bool {
 456             entry.file_name()
 457                 .to_str()
 458                 .map(|s| s.starts_with('.'))
 459                 .unwrap_or(false)
 460         }
 461         let handle_entry_err = |err: walkdir::Error| {
 462             if let Some(inner) = err.io_error() {
 463                 if let Some(path) = err.path() {
 464                     if inner.kind() == io::ErrorKind::PermissionDenied {
 465                         // only allow to skip ext4 fsck directory, avoid GC if, for example,
 466                         // a user got file permissions wrong on datastore rsync to new server
 467                         if err.depth() > 1 || !path.ends_with("lost+found") {
 468                             bail!("cannot continue garbage-collection safely, permission denied on: {:?}", path)
 469                         }
 470                     } else {
 471                         bail!("unexpected error on datastore traversal: {} - {:?}", inner, path)
 472                     }
 473                 } else {
 474                     bail!("unexpected error on datastore traversal: {}", inner)
 475                 }
 476             }
 477             Ok(())
 478         };
 479         for entry in walker.filter_entry(|e| !is_hidden(e)) {
 480             let path = match entry {
 481                 Ok(entry) => entry.into_path(),
 482                 Err(err) => {
 483                     handle_entry_err(err)?;
 484                     continue
 485                 },
 486             };
 487             if let Ok(archive_type) = archive_type(&path) {
 488                 if archive_type == ArchiveType::FixedIndex || archive_type == ArchiveType::DynamicIndex {
 489                     list.push(path);
 490                 }
 491             }
 492         }
 493
 494         Ok(list)
 495     }
 496
 497     // mark chunks  used by ``index`` as used
 498     fn index_mark_used_chunks<I: IndexFile>(
 499         &self,
 500         index: I,
 501         file_name: &Path, // only used for error reporting
 502         status: &mut GarbageCollectionStatus,
 503         worker: &dyn TaskState,
 504     ) -> Result<(), Error> {
 505
 506         status.index_file_count += 1;
 507         status.index_data_bytes += index.index_bytes();
 508
 509         for pos in 0..index.index_count() {
 510             worker.check_abort()?;
 511             fail_on_shutdown()?;
 512             let digest = index.index_digest(pos).unwrap();
 513             if !self.chunk_store.cond_touch_chunk(digest, false)? {
 514                 task_warn!(
 515                     worker,
 516                     "warning: unable to access non-existent chunk {}, required by {:?}",
 517                     proxmox::tools::digest_to_hex(digest),
 518                     file_name,
 519                 );
 520
 521                 // touch any corresponding .bad files to keep them around, meaning if a chunk is
 522                 // rewritten correctly they will be removed automatically, as well as if no index
 523                 // file requires the chunk anymore (won't get to this loop then)
 524                 for i in 0..=9 {
 525                     let bad_ext = format!("{}.bad", i);
 526                     let mut bad_path = PathBuf::new();
 527                     bad_path.push(self.chunk_path(digest).0);
 528                     bad_path.set_extension(bad_ext);
 529                     self.chunk_store.cond_touch_path(&bad_path, false)?;
 530                 }
 531             }
 532         }
 533         Ok(())
 534     }
 535
 536     fn mark_used_chunks(
 537         &self,
 538         status: &mut GarbageCollectionStatus,
 539         worker: &dyn TaskState,
 540     ) -> Result<(), Error> {
 541
 542         let image_list = self.list_images()?;
 543         let image_count = image_list.len();
 544
 545         let mut last_percentage: usize = 0;
 546
 547         let mut strange_paths_count: u64 = 0;
 548
 549         for (i, img) in image_list.into_iter().enumerate() {
 550
 551             worker.check_abort()?;
 552             fail_on_shutdown()?;
 553
 554             if let Some(backup_dir_path) = img.parent() {
 555                 let backup_dir_path = backup_dir_path.strip_prefix(self.base_path())?;
 556                 if let Some(backup_dir_str) = backup_dir_path.to_str() {
 557                     if BackupDir::from_str(backup_dir_str).is_err() {
 558                         strange_paths_count += 1;
 559                     }
 560                 }
 561             }
 562
 563             match std::fs::File::open(&img) {
 564                 Ok(file) => {
 565                     if let Ok(archive_type) = archive_type(&img) {
 566                         if archive_type == ArchiveType::FixedIndex {
 567                             let index = FixedIndexReader::new(file).map_err(|e| {
 568                                 format_err!("can't read index '{}' - {}", img.to_string_lossy(), e)
 569                             })?;
 570                             self.index_mark_used_chunks(index, &img, status, worker)?;
 571                         } else if archive_type == ArchiveType::DynamicIndex {
 572                             let index = DynamicIndexReader::new(file).map_err(|e| {
 573                                 format_err!("can't read index '{}' - {}", img.to_string_lossy(), e)
 574                             })?;
 575                             self.index_mark_used_chunks(index, &img, status, worker)?;
 576                         }
 577                     }
 578                 }
 579                 Err(err) if err.kind() == io::ErrorKind::NotFound => (), // ignore vanished files
 580                 Err(err) => bail!("can't open index {} - {}", img.to_string_lossy(), err),
 581             }
 582
 583             let percentage = (i + 1) * 100 / image_count;
 584             if percentage > last_percentage {
 585                 task_log!(
 586                     worker,
 587                     "marked {}% ({} of {} index files)",
 588                     percentage,
 589                     i + 1,
 590                     image_count,
 591                 );
 592                 last_percentage = percentage;
 593             }
 594         }
 595
 596         if strange_paths_count > 0 {
 597             task_log!(
 598                 worker,
 599                 "found (and marked) {} index files outside of expected directory scheme",
 600                 strange_paths_count,
 601             );
 602         }
 603
 604
 605         Ok(())
 606     }
 607
 608     pub fn last_gc_status(&self) -> GarbageCollectionStatus {
 609         self.last_gc_status.lock().unwrap().clone()
 610     }
 611
 612     pub fn garbage_collection_running(&self) -> bool {
 613         !matches!(self.gc_mutex.try_lock(), Ok(_))
 614     }
 615
 616     pub fn garbage_collection(&self, worker: &dyn TaskState, upid: &UPID) -> Result<(), Error> {
 617
 618         if let Ok(ref mut _mutex) = self.gc_mutex.try_lock() {
 619
 620             // avoids that we run GC if an old daemon process has still a
 621             // running backup writer, which is not save as we have no "oldest
 622             // writer" information and thus no safe atime cutoff
 623             let _exclusive_lock =  self.chunk_store.try_exclusive_lock()?;
 624
 625             let phase1_start_time = proxmox::tools::time::epoch_i64();
 626             let oldest_writer = self.chunk_store.oldest_writer().unwrap_or(phase1_start_time);
 627
 628             let mut gc_status = GarbageCollectionStatus::default();
 629             gc_status.upid = Some(upid.to_string());
 630
 631             task_log!(worker, "Start GC phase1 (mark used chunks)");
 632
 633             self.mark_used_chunks(&mut gc_status, worker)?;
 634
 635             task_log!(worker, "Start GC phase2 (sweep unused chunks)");
 636             self.chunk_store.sweep_unused_chunks(
 637                 oldest_writer,
 638                 phase1_start_time,
 639                 &mut gc_status,
 640                 worker,
 641                 fail_on_shutdown,
 642             )?;
 643
 644             task_log!(
 645                 worker,
 646                 "Removed garbage: {}",
 647                 HumanByte::from(gc_status.removed_bytes),
 648             );
 649             task_log!(worker, "Removed chunks: {}", gc_status.removed_chunks);
 650             if gc_status.pending_bytes > 0 {
 651                 task_log!(
 652                     worker,
 653                     "Pending removals: {} (in {} chunks)",
 654                     HumanByte::from(gc_status.pending_bytes),
 655                     gc_status.pending_chunks,
 656                 );
 657             }
 658             if gc_status.removed_bad > 0 {
 659                 task_log!(worker, "Removed bad chunks: {}", gc_status.removed_bad);
 660             }
 661
 662             if gc_status.still_bad > 0 {
 663                 task_log!(worker, "Leftover bad chunks: {}", gc_status.still_bad);
 664             }
 665
 666             task_log!(
 667                 worker,
 668                 "Original data usage: {}",
 669                 HumanByte::from(gc_status.index_data_bytes),
 670             );
 671
 672             if gc_status.index_data_bytes > 0 {
 673                 let comp_per = (gc_status.disk_bytes as f64 * 100.)/gc_status.index_data_bytes as f64;
 674                 task_log!(
 675                     worker,
 676                     "On-Disk usage: {} ({:.2}%)",
 677                     HumanByte::from(gc_status.disk_bytes),
 678                     comp_per,
 679                 );
 680             }
 681
 682             task_log!(worker, "On-Disk chunks: {}", gc_status.disk_chunks);
 683
 684             let deduplication_factor = if gc_status.disk_bytes > 0 {
 685                 (gc_status.index_data_bytes as f64)/(gc_status.disk_bytes as f64)
 686             } else {
 687                 1.0
 688             };
 689
 690             task_log!(worker, "Deduplication factor: {:.2}", deduplication_factor);
 691
 692             if gc_status.disk_chunks > 0 {
 693                 let avg_chunk = gc_status.disk_bytes/(gc_status.disk_chunks as u64);
 694                 task_log!(worker, "Average chunk size: {}", HumanByte::from(avg_chunk));
 695             }
 696
 697             if let Ok(serialized) = serde_json::to_string(&gc_status) {
 698                 let mut path = self.base_path();
 699                 path.push(".gc-status");
 700
 701                 let backup_user = pbs_config::backup_user()?;
 702                 let mode = nix::sys::stat::Mode::from_bits_truncate(0o0644);
 703                 // set the correct owner/group/permissions while saving file
 704                 // owner(rw) = backup, group(r)= backup
 705                 let options = CreateOptions::new()
 706                     .perm(mode)
 707                     .owner(backup_user.uid)
 708                     .group(backup_user.gid);
 709
 710                 // ignore errors
 711                 let _ = replace_file(path, serialized.as_bytes(), options);
 712             }
 713
 714             *self.last_gc_status.lock().unwrap() = gc_status;
 715
 716         } else {
 717             bail!("Start GC failed - (already running/locked)");
 718         }
 719
 720         Ok(())
 721     }
 722
 723     pub fn try_shared_chunk_store_lock(&self) -> Result<ProcessLockSharedGuard, Error> {
 724         self.chunk_store.try_shared_lock()
 725     }
 726
 727     pub fn chunk_path(&self, digest:&[u8; 32]) -> (PathBuf, String) {
 728         self.chunk_store.chunk_path(digest)
 729     }
 730
 731     pub fn cond_touch_chunk(&self, digest: &[u8; 32], fail_if_not_exist: bool) -> Result<bool, Error> {
 732         self.chunk_store.cond_touch_chunk(digest, fail_if_not_exist)
 733     }
 734
 735     pub fn insert_chunk(
 736         &self,
 737         chunk: &DataBlob,
 738         digest: &[u8; 32],
 739     ) -> Result<(bool, u64), Error> {
 740         self.chunk_store.insert_chunk(chunk, digest)
 741     }
 742
 743     pub fn load_blob(&self, backup_dir: &BackupDir, filename: &str) -> Result<DataBlob, Error> {
 744         let mut path = self.base_path();
 745         path.push(backup_dir.relative_path());
 746         path.push(filename);
 747
 748         proxmox::try_block!({
 749             let mut file = std::fs::File::open(&path)?;
 750             DataBlob::load_from_reader(&mut file)
 751         }).map_err(|err| format_err!("unable to load blob '{:?}' - {}", path, err))
 752     }
 753
 754
 755     pub fn stat_chunk(&self, digest: &[u8; 32]) -> Result<std::fs::Metadata, Error> {
 756         let (chunk_path, _digest_str) = self.chunk_store.chunk_path(digest);
 757         std::fs::metadata(chunk_path).map_err(Error::from)
 758     }
 759
 760     pub fn load_chunk(&self, digest: &[u8; 32]) -> Result<DataBlob, Error> {
 761
 762         let (chunk_path, digest_str) = self.chunk_store.chunk_path(digest);
 763
 764         proxmox::try_block!({
 765             let mut file = std::fs::File::open(&chunk_path)?;
 766             DataBlob::load_from_reader(&mut file)
 767         }).map_err(|err| format_err!(
 768             "store '{}', unable to load chunk '{}' - {}",
 769             self.name(),
 770             digest_str,
 771             err,
 772         ))
 773     }
 774
 775     /// Returns the filename to lock a manifest
 776     ///
 777     /// Also creates the basedir. The lockfile is located in
 778     /// '/run/proxmox-backup/locks/{datastore}/{type}/{id}/{timestamp}.index.json.lck'
 779     fn manifest_lock_path(
 780         &self,
 781         backup_dir: &BackupDir,
 782     ) -> Result<String, Error> {
 783         let mut path = format!(
 784             "/run/proxmox-backup/locks/{}/{}/{}",
 785             self.name(),
 786             backup_dir.group().backup_type(),
 787             backup_dir.group().backup_id(),
 788         );
 789         std::fs::create_dir_all(&path)?;
 790         use std::fmt::Write;
 791         write!(path, "/{}{}", backup_dir.backup_time_string(), &MANIFEST_LOCK_NAME)?;
 792
 793         Ok(path)
 794     }
 795
 796     fn lock_manifest(
 797         &self,
 798         backup_dir: &BackupDir,
 799     ) -> Result<BackupLockGuard, Error> {
 800         let path = self.manifest_lock_path(backup_dir)?;
 801
 802         // update_manifest should never take a long time, so if someone else has
 803         // the lock we can simply block a bit and should get it soon
 804         open_backup_lockfile(&path, Some(Duration::from_secs(5)), true)
 805             .map_err(|err| {
 806                 format_err!(
 807                     "unable to acquire manifest lock {:?} - {}", &path, err
 808                 )
 809             })
 810     }
 811
 812     /// Load the manifest without a lock. Must not be written back.
 813     pub fn load_manifest(
 814         &self,
 815         backup_dir: &BackupDir,
 816     ) -> Result<(BackupManifest, u64), Error> {
 817         let blob = self.load_blob(backup_dir, MANIFEST_BLOB_NAME)?;
 818         let raw_size = blob.raw_size();
 819         let manifest = BackupManifest::try_from(blob)?;
 820         Ok((manifest, raw_size))
 821     }
 822
 823     /// Update the manifest of the specified snapshot. Never write a manifest directly,
 824     /// only use this method - anything else may break locking guarantees.
 825     pub fn update_manifest(
 826         &self,
 827         backup_dir: &BackupDir,
 828         update_fn: impl FnOnce(&mut BackupManifest),
 829     ) -> Result<(), Error> {
 830
 831         let _guard = self.lock_manifest(backup_dir)?;
 832         let (mut manifest, _) = self.load_manifest(&backup_dir)?;
 833
 834         update_fn(&mut manifest);
 835
 836         let manifest = serde_json::to_value(manifest)?;
 837         let manifest = serde_json::to_string_pretty(&manifest)?;
 838         let blob = DataBlob::encode(manifest.as_bytes(), None, true)?;
 839         let raw_data = blob.raw_data();
 840
 841         let mut path = self.base_path();
 842         path.push(backup_dir.relative_path());
 843         path.push(MANIFEST_BLOB_NAME);
 844
 845         // atomic replace invalidates flock - no other writes past this point!
 846         replace_file(&path, raw_data, CreateOptions::new())?;
 847
 848         Ok(())
 849     }
 850
 851     pub fn verify_new(&self) -> bool {
 852         self.verify_new
 853     }
 854
 855     /// returns a list of chunks sorted by their inode number on disk
 856     /// chunks that could not be stat'ed are at the end of the list
 857     pub fn get_chunks_in_order<F, A>(
 858         &self,
 859         index: &Box<dyn IndexFile + Send>,
 860         skip_chunk: F,
 861         check_abort: A,
 862     ) -> Result<Vec<(usize, u64)>, Error>
 863     where
 864         F: Fn(&[u8; 32]) -> bool,
 865         A: Fn(usize) -> Result<(), Error>,
 866     {
 867         let index_count = index.index_count();
 868         let mut chunk_list = Vec::with_capacity(index_count);
 869         use std::os::unix::fs::MetadataExt;
 870         for pos in 0..index_count {
 871             check_abort(pos)?;
 872
 873             let info = index.chunk_info(pos).unwrap();
 874
 875             if skip_chunk(&info.digest) {
 876                 continue;
 877             }
 878
 879             let ino = match self.stat_chunk(&info.digest) {
 880                 Err(_) => u64::MAX, // could not stat, move to end of list
 881                 Ok(metadata) => metadata.ino(),
 882             };
 883
 884             chunk_list.push((pos, ino));
 885         }
 886
 887         // sorting by inode improves data locality, which makes it lots faster on spinners
 888         chunk_list.sort_unstable_by(|(_, ino_a), (_, ino_b)| ino_a.cmp(&ino_b));
 889
 890         Ok(chunk_list)
 891     }
 892 }