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