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