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
   7 use anyhow::{bail, format_err, Error};
   8 use lazy_static::lazy_static;
   9
  10 use proxmox::tools::fs::{replace_file, CreateOptions};
  11
  12 use super::backup_info::{BackupGroup, BackupDir};
  13 use super::chunk_store::ChunkStore;
  14 use super::dynamic_index::{DynamicIndexReader, DynamicIndexWriter};
  15 use super::fixed_index::{FixedIndexReader, FixedIndexWriter};
  16 use super::manifest::{MANIFEST_BLOB_NAME, CLIENT_LOG_BLOB_NAME, BackupManifest};
  17 use super::index::*;
  18 use super::{DataBlob, ArchiveType, archive_type};
  19 use crate::config::datastore;
  20 use crate::task::TaskState;
  21 use crate::tools;
  22 use crate::tools::format::HumanByte;
  23 use crate::tools::fs::{lock_dir_noblock, DirLockGuard};
  24 use crate::api2::types::{GarbageCollectionStatus, Userid};
  25 use crate::server::UPID;
  26
  27 lazy_static! {
  28     static ref DATASTORE_MAP: Mutex<HashMap<String, Arc<DataStore>>> = Mutex::new(HashMap::new());
  29 }
  30
  31 /// Datastore Management
  32 ///
  33 /// A Datastore can store severals backups, and provides the
  34 /// management interface for backup.
  35 pub struct DataStore {
  36     chunk_store: Arc<ChunkStore>,
  37     gc_mutex: Mutex<bool>,
  38     last_gc_status: Mutex<GarbageCollectionStatus>,
  39 }
  40
  41 impl DataStore {
  42
  43     pub fn lookup_datastore(name: &str) -> Result<Arc<DataStore>, Error> {
  44
  45         let (config, _digest) = datastore::config()?;
  46         let config: datastore::DataStoreConfig = config.lookup("datastore", name)?;
  47
  48         let mut map = DATASTORE_MAP.lock().unwrap();
  49
  50         if let Some(datastore) = map.get(name) {
  51             // Compare Config - if changed, create new Datastore object!
  52             if datastore.chunk_store.base == PathBuf::from(&config.path) {
  53                 return Ok(datastore.clone());
  54             }
  55         }
  56
  57         let datastore = DataStore::open(name)?;
  58
  59         let datastore = Arc::new(datastore);
  60         map.insert(name.to_string(), datastore.clone());
  61
  62         Ok(datastore)
  63     }
  64
  65     pub fn open(store_name: &str) -> Result<Self, Error> {
  66
  67         let (config, _digest) = datastore::config()?;
  68         let (_, store_config) = config.sections.get(store_name)
  69             .ok_or(format_err!("no such datastore '{}'", store_name))?;
  70
  71         let path = store_config["path"].as_str().unwrap();
  72
  73         Self::open_with_path(store_name, Path::new(path))
  74     }
  75
  76     pub fn open_with_path(store_name: &str, path: &Path) -> Result<Self, Error> {
  77         let chunk_store = ChunkStore::open(store_name, path)?;
  78
  79         let gc_status = GarbageCollectionStatus::default();
  80
  81         Ok(Self {
  82             chunk_store: Arc::new(chunk_store),
  83             gc_mutex: Mutex::new(false),
  84             last_gc_status: Mutex::new(gc_status),
  85         })
  86     }
  87
  88     pub fn get_chunk_iterator(
  89         &self,
  90     ) -> Result<
  91         impl Iterator<Item = (Result<tools::fs::ReadDirEntry, Error>, usize, bool)>,
  92         Error
  93     > {
  94         self.chunk_store.get_chunk_iterator()
  95     }
  96
  97     pub fn create_fixed_writer<P: AsRef<Path>>(&self, filename: P, size: usize, chunk_size: usize) -> Result<FixedIndexWriter, Error> {
  98
  99         let index = FixedIndexWriter::create(self.chunk_store.clone(), filename.as_ref(), size, chunk_size)?;
 100
 101         Ok(index)
 102     }
 103
 104     pub fn open_fixed_reader<P: AsRef<Path>>(&self, filename: P) -> Result<FixedIndexReader, Error> {
 105
 106         let full_path =  self.chunk_store.relative_path(filename.as_ref());
 107
 108         let index = FixedIndexReader::open(&full_path)?;
 109
 110         Ok(index)
 111     }
 112
 113     pub fn create_dynamic_writer<P: AsRef<Path>>(
 114         &self, filename: P,
 115     ) -> Result<DynamicIndexWriter, Error> {
 116
 117         let index = DynamicIndexWriter::create(
 118             self.chunk_store.clone(), filename.as_ref())?;
 119
 120         Ok(index)
 121     }
 122
 123     pub fn open_dynamic_reader<P: AsRef<Path>>(&self, filename: P) -> Result<DynamicIndexReader, Error> {
 124
 125         let full_path =  self.chunk_store.relative_path(filename.as_ref());
 126
 127         let index = DynamicIndexReader::open(&full_path)?;
 128
 129         Ok(index)
 130     }
 131
 132     pub fn open_index<P>(&self, filename: P) -> Result<Box<dyn IndexFile + Send>, Error>
 133     where
 134         P: AsRef<Path>,
 135     {
 136         let filename = filename.as_ref();
 137         let out: Box<dyn IndexFile + Send> =
 138             match archive_type(filename)? {
 139                 ArchiveType::DynamicIndex => Box::new(self.open_dynamic_reader(filename)?),
 140                 ArchiveType::FixedIndex => Box::new(self.open_fixed_reader(filename)?),
 141                 _ => bail!("cannot open index file of unknown type: {:?}", filename),
 142             };
 143         Ok(out)
 144     }
 145
 146     pub fn name(&self) -> &str {
 147         self.chunk_store.name()
 148     }
 149
 150     pub fn base_path(&self) -> PathBuf {
 151         self.chunk_store.base_path()
 152     }
 153
 154     /// Cleanup a backup directory
 155     ///
 156     /// Removes all files not mentioned in the manifest.
 157     pub fn cleanup_backup_dir(&self, backup_dir: &BackupDir, manifest: &BackupManifest
 158     ) ->  Result<(), Error> {
 159
 160         let mut full_path = self.base_path();
 161         full_path.push(backup_dir.relative_path());
 162
 163         let mut wanted_files = HashSet::new();
 164         wanted_files.insert(MANIFEST_BLOB_NAME.to_string());
 165         wanted_files.insert(CLIENT_LOG_BLOB_NAME.to_string());
 166         manifest.files().iter().for_each(|item| { wanted_files.insert(item.filename.clone()); });
 167
 168         for item in tools::fs::read_subdir(libc::AT_FDCWD, &full_path)? {
 169             if let Ok(item) = item {
 170                 if let Some(file_type) = item.file_type() {
 171                     if file_type != nix::dir::Type::File { continue; }
 172                 }
 173                 let file_name = item.file_name().to_bytes();
 174                 if file_name == b"." || file_name == b".." { continue; };
 175
 176                 if let Ok(name) = std::str::from_utf8(file_name) {
 177                     if wanted_files.contains(name) { continue; }
 178                 }
 179                 println!("remove unused file {:?}", item.file_name());
 180                 let dirfd = item.parent_fd();
 181                 let _res = unsafe { libc::unlinkat(dirfd, item.file_name().as_ptr(), 0) };
 182             }
 183         }
 184
 185         Ok(())
 186     }
 187
 188     /// Returns the absolute path for a backup_group
 189     pub fn group_path(&self, backup_group: &BackupGroup) -> PathBuf {
 190         let mut full_path = self.base_path();
 191         full_path.push(backup_group.group_path());
 192         full_path
 193     }
 194
 195     /// Returns the absolute path for backup_dir
 196     pub fn snapshot_path(&self, backup_dir: &BackupDir) -> PathBuf {
 197         let mut full_path = self.base_path();
 198         full_path.push(backup_dir.relative_path());
 199         full_path
 200     }
 201
 202     /// Remove a complete backup group including all snapshots
 203     pub fn remove_backup_group(&self, backup_group: &BackupGroup) ->  Result<(), Error> {
 204
 205         let full_path = self.group_path(backup_group);
 206
 207         let _guard = tools::fs::lock_dir_noblock(&full_path, "backup group", "possible running backup")?;
 208
 209         log::info!("removing backup group {:?}", full_path);
 210
 211         // remove all individual backup dirs first to ensure nothing is using them
 212         for snap in backup_group.list_backups(&self.base_path())? {
 213             self.remove_backup_dir(&snap.backup_dir, false)?;
 214         }
 215
 216         // no snapshots left, we can now safely remove the empty folder
 217         std::fs::remove_dir_all(&full_path)
 218             .map_err(|err| {
 219                 format_err!(
 220                     "removing backup group directory {:?} failed - {}",
 221                     full_path,
 222                     err,
 223                 )
 224             })?;
 225
 226         Ok(())
 227     }
 228
 229     /// Remove a backup directory including all content
 230     pub fn remove_backup_dir(&self, backup_dir: &BackupDir, force: bool) ->  Result<(), Error> {
 231
 232         let full_path = self.snapshot_path(backup_dir);
 233
 234         let _guard;
 235         if !force {
 236             _guard = lock_dir_noblock(&full_path, "snapshot", "possibly running or in use")?;
 237         }
 238
 239         log::info!("removing backup snapshot {:?}", full_path);
 240         std::fs::remove_dir_all(&full_path)
 241             .map_err(|err| {
 242                 format_err!(
 243                     "removing backup snapshot {:?} failed - {}",
 244                     full_path,
 245                     err,
 246                 )
 247             })?;
 248
 249         Ok(())
 250     }
 251
 252     /// Returns the time of the last successful backup
 253     ///
 254     /// Or None if there is no backup in the group (or the group dir does not exist).
 255     pub fn last_successful_backup(&self, backup_group: &BackupGroup) -> Result<Option<i64>, Error> {
 256         let base_path = self.base_path();
 257         let mut group_path = base_path.clone();
 258         group_path.push(backup_group.group_path());
 259
 260         if group_path.exists() {
 261             backup_group.last_successful_backup(&base_path)
 262         } else {
 263             Ok(None)
 264         }
 265     }
 266
 267     /// Returns the backup owner.
 268     ///
 269     /// The backup owner is the user who first created the backup group.
 270     pub fn get_owner(&self, backup_group: &BackupGroup) -> Result<Userid, Error> {
 271         let mut full_path = self.base_path();
 272         full_path.push(backup_group.group_path());
 273         full_path.push("owner");
 274         let owner = proxmox::tools::fs::file_read_firstline(full_path)?;
 275         Ok(owner.trim_end().parse()?) // remove trailing newline
 276     }
 277
 278     /// Set the backup owner.
 279     pub fn set_owner(
 280         &self,
 281         backup_group: &BackupGroup,
 282         userid: &Userid,
 283         force: bool,
 284     ) -> Result<(), Error> {
 285         let mut path = self.base_path();
 286         path.push(backup_group.group_path());
 287         path.push("owner");
 288
 289         let mut open_options = std::fs::OpenOptions::new();
 290         open_options.write(true);
 291         open_options.truncate(true);
 292
 293         if force {
 294             open_options.create(true);
 295         } else {
 296             open_options.create_new(true);
 297         }
 298
 299         let mut file = open_options.open(&path)
 300             .map_err(|err| format_err!("unable to create owner file {:?} - {}", path, err))?;
 301
 302         writeln!(file, "{}", userid)
 303             .map_err(|err| format_err!("unable to write owner file  {:?} - {}", path, err))?;
 304
 305         Ok(())
 306     }
 307
 308     /// Create (if it does not already exists) and lock a backup group
 309     ///
 310     /// And set the owner to 'userid'. If the group already exists, it returns the
 311     /// current owner (instead of setting the owner).
 312     ///
 313     /// This also acquires an exclusive lock on the directory and returns the lock guard.
 314     pub fn create_locked_backup_group(
 315         &self,
 316         backup_group: &BackupGroup,
 317         userid: &Userid,
 318     ) -> Result<(Userid, DirLockGuard), Error> {
 319         // create intermediate path first:
 320         let base_path = self.base_path();
 321
 322         let mut full_path = base_path.clone();
 323         full_path.push(backup_group.backup_type());
 324         std::fs::create_dir_all(&full_path)?;
 325
 326         full_path.push(backup_group.backup_id());
 327
 328         // create the last component now
 329         match std::fs::create_dir(&full_path) {
 330             Ok(_) => {
 331                 let guard = lock_dir_noblock(&full_path, "backup group", "another backup is already running")?;
 332                 self.set_owner(backup_group, userid, false)?;
 333                 let owner = self.get_owner(backup_group)?; // just to be sure
 334                 Ok((owner, guard))
 335             }
 336             Err(ref err) if err.kind() == io::ErrorKind::AlreadyExists => {
 337                 let guard = lock_dir_noblock(&full_path, "backup group", "another backup is already running")?;
 338                 let owner = self.get_owner(backup_group)?; // just to be sure
 339                 Ok((owner, guard))
 340             }
 341             Err(err) => bail!("unable to create backup group {:?} - {}", full_path, err),
 342         }
 343     }
 344
 345     /// Creates a new backup snapshot inside a BackupGroup
 346     ///
 347     /// The BackupGroup directory needs to exist.
 348     pub fn create_locked_backup_dir(&self, backup_dir: &BackupDir)
 349         -> Result<(PathBuf, bool, DirLockGuard), Error>
 350     {
 351         let relative_path = backup_dir.relative_path();
 352         let mut full_path = self.base_path();
 353         full_path.push(&relative_path);
 354
 355         let lock = ||
 356             lock_dir_noblock(&full_path, "snapshot", "internal error - tried creating snapshot that's already in use");
 357
 358         match std::fs::create_dir(&full_path) {
 359             Ok(_) => Ok((relative_path, true, lock()?)),
 360             Err(ref e) if e.kind() == io::ErrorKind::AlreadyExists => Ok((relative_path, false, lock()?)),
 361             Err(e) => Err(e.into())
 362         }
 363     }
 364
 365     pub fn list_images(&self) -> Result<Vec<PathBuf>, Error> {
 366         let base = self.base_path();
 367
 368         let mut list = vec![];
 369
 370         use walkdir::WalkDir;
 371
 372         let walker = WalkDir::new(&base).same_file_system(true).into_iter();
 373
 374         // make sure we skip .chunks (and other hidden files to keep it simple)
 375         fn is_hidden(entry: &walkdir::DirEntry) -> bool {
 376             entry.file_name()
 377                 .to_str()
 378                 .map(|s| s.starts_with("."))
 379                 .unwrap_or(false)
 380         }
 381         let handle_entry_err = |err: walkdir::Error| {
 382             if let Some(inner) = err.io_error() {
 383                 let path = err.path().unwrap_or(Path::new(""));
 384                 match inner.kind() {
 385                     io::ErrorKind::PermissionDenied => {
 386                         // only allow to skip ext4 fsck directory, avoid GC if, for example,
 387                         // a user got file permissions wrong on datastore rsync to new server
 388                         if err.depth() > 1 || !path.ends_with("lost+found") {
 389                             bail!("cannot continue garbage-collection safely, permission denied on: {}", path.display())
 390                         }
 391                     },
 392                     _ => bail!("unexpected error on datastore traversal: {} - {}", inner, path.display()),
 393                 }
 394             }
 395             Ok(())
 396         };
 397         for entry in walker.filter_entry(|e| !is_hidden(e)) {
 398             let path = match entry {
 399                 Ok(entry) => entry.into_path(),
 400                 Err(err) => {
 401                     handle_entry_err(err)?;
 402                     continue
 403                 },
 404             };
 405             if let Ok(archive_type) = archive_type(&path) {
 406                 if archive_type == ArchiveType::FixedIndex || archive_type == ArchiveType::DynamicIndex {
 407                     list.push(path);
 408                 }
 409             }
 410         }
 411
 412         Ok(list)
 413     }
 414
 415     // mark chunks  used by ``index`` as used
 416     fn index_mark_used_chunks<I: IndexFile>(
 417         &self,
 418         index: I,
 419         file_name: &Path, // only used for error reporting
 420         status: &mut GarbageCollectionStatus,
 421         worker: &dyn TaskState,
 422     ) -> Result<(), Error> {
 423
 424         status.index_file_count += 1;
 425         status.index_data_bytes += index.index_bytes();
 426
 427         for pos in 0..index.index_count() {
 428             worker.check_abort()?;
 429             tools::fail_on_shutdown()?;
 430             let digest = index.index_digest(pos).unwrap();
 431             if let Err(err) = self.chunk_store.touch_chunk(digest) {
 432                 crate::task_warn!(
 433                     worker,
 434                     "warning: unable to access chunk {}, required by {:?} - {}",
 435                     proxmox::tools::digest_to_hex(digest),
 436                     file_name,
 437                     err,
 438                 );
 439             }
 440         }
 441         Ok(())
 442     }
 443
 444     fn mark_used_chunks(
 445         &self,
 446         status: &mut GarbageCollectionStatus,
 447         worker: &dyn TaskState,
 448     ) -> Result<(), Error> {
 449
 450         let image_list = self.list_images()?;
 451
 452         let image_count = image_list.len();
 453
 454         let mut done = 0;
 455
 456         let mut last_percentage: usize = 0;
 457
 458         for path in image_list {
 459
 460             worker.check_abort()?;
 461             tools::fail_on_shutdown()?;
 462
 463             if let Ok(archive_type) = archive_type(&path) {
 464                 if archive_type == ArchiveType::FixedIndex {
 465                     let index = self.open_fixed_reader(&path)?;
 466                     self.index_mark_used_chunks(index, &path, status, worker)?;
 467                 } else if archive_type == ArchiveType::DynamicIndex {
 468                     let index = self.open_dynamic_reader(&path)?;
 469                     self.index_mark_used_chunks(index, &path, status, worker)?;
 470                 }
 471             }
 472             done += 1;
 473
 474             let percentage = done*100/image_count;
 475             if percentage > last_percentage {
 476                 crate::task_log!(
 477                     worker,
 478                     "percentage done: phase1 {}% ({} of {} index files)",
 479                     percentage,
 480                     done,
 481                     image_count,
 482                 );
 483                 last_percentage = percentage;
 484             }
 485         }
 486
 487         Ok(())
 488     }
 489
 490     pub fn last_gc_status(&self) -> GarbageCollectionStatus {
 491         self.last_gc_status.lock().unwrap().clone()
 492     }
 493
 494     pub fn garbage_collection_running(&self) -> bool {
 495         if let Ok(_) = self.gc_mutex.try_lock() { false } else { true }
 496     }
 497
 498     pub fn garbage_collection(&self, worker: &dyn TaskState, upid: &UPID) -> Result<(), Error> {
 499
 500         if let Ok(ref mut _mutex) = self.gc_mutex.try_lock() {
 501
 502             // avoids that we run GC if an old daemon process has still a
 503             // running backup writer, which is not save as we have no "oldest
 504             // writer" information and thus no safe atime cutoff
 505             let _exclusive_lock =  self.chunk_store.try_exclusive_lock()?;
 506
 507             let phase1_start_time = proxmox::tools::time::epoch_i64();
 508             let oldest_writer = self.chunk_store.oldest_writer().unwrap_or(phase1_start_time);
 509
 510             let mut gc_status = GarbageCollectionStatus::default();
 511             gc_status.upid = Some(upid.to_string());
 512
 513             crate::task_log!(worker, "Start GC phase1 (mark used chunks)");
 514
 515             self.mark_used_chunks(&mut gc_status, worker)?;
 516
 517             crate::task_log!(worker, "Start GC phase2 (sweep unused chunks)");
 518             self.chunk_store.sweep_unused_chunks(
 519                 oldest_writer,
 520                 phase1_start_time,
 521                 &mut gc_status,
 522                 worker,
 523             )?;
 524
 525             crate::task_log!(
 526                 worker,
 527                 "Removed garbage: {}",
 528                 HumanByte::from(gc_status.removed_bytes),
 529             );
 530             crate::task_log!(worker, "Removed chunks: {}", gc_status.removed_chunks);
 531             if gc_status.pending_bytes > 0 {
 532                 crate::task_log!(
 533                     worker,
 534                     "Pending removals: {} (in {} chunks)",
 535                     HumanByte::from(gc_status.pending_bytes),
 536                     gc_status.pending_chunks,
 537                 );
 538             }
 539             if gc_status.removed_bad > 0 {
 540                 crate::task_log!(worker, "Removed bad files: {}", gc_status.removed_bad);
 541             }
 542
 543             crate::task_log!(
 544                 worker,
 545                 "Original data usage: {}",
 546                 HumanByte::from(gc_status.index_data_bytes),
 547             );
 548
 549             if gc_status.index_data_bytes > 0 {
 550                 let comp_per = (gc_status.disk_bytes as f64 * 100.)/gc_status.index_data_bytes as f64;
 551                 crate::task_log!(
 552                     worker,
 553                     "On-Disk usage: {} ({:.2}%)",
 554                     HumanByte::from(gc_status.disk_bytes),
 555                     comp_per,
 556                 );
 557             }
 558
 559             crate::task_log!(worker, "On-Disk chunks: {}", gc_status.disk_chunks);
 560
 561             if gc_status.disk_chunks > 0 {
 562                 let avg_chunk = gc_status.disk_bytes/(gc_status.disk_chunks as u64);
 563                 crate::task_log!(worker, "Average chunk size: {}", HumanByte::from(avg_chunk));
 564             }
 565
 566             *self.last_gc_status.lock().unwrap() = gc_status;
 567
 568         } else {
 569             bail!("Start GC failed - (already running/locked)");
 570         }
 571
 572         Ok(())
 573     }
 574
 575     pub fn try_shared_chunk_store_lock(&self) -> Result<tools::ProcessLockSharedGuard, Error> {
 576         self.chunk_store.try_shared_lock()
 577     }
 578
 579     pub fn chunk_path(&self, digest:&[u8; 32]) -> (PathBuf, String) {
 580         self.chunk_store.chunk_path(digest)
 581     }
 582
 583     pub fn cond_touch_chunk(&self, digest: &[u8; 32], fail_if_not_exist: bool) -> Result<bool, Error> {
 584         self.chunk_store.cond_touch_chunk(digest, fail_if_not_exist)
 585     }
 586
 587     pub fn insert_chunk(
 588         &self,
 589         chunk: &DataBlob,
 590         digest: &[u8; 32],
 591     ) -> Result<(bool, u64), Error> {
 592         self.chunk_store.insert_chunk(chunk, digest)
 593     }
 594
 595     pub fn load_blob(&self, backup_dir: &BackupDir, filename: &str) -> Result<DataBlob, Error> {
 596         let mut path = self.base_path();
 597         path.push(backup_dir.relative_path());
 598         path.push(filename);
 599
 600         proxmox::try_block!({
 601             let mut file = std::fs::File::open(&path)?;
 602             DataBlob::load_from_reader(&mut file)
 603         }).map_err(|err| format_err!("unable to load blob '{:?}' - {}", path, err))
 604     }
 605
 606
 607     pub fn load_chunk(&self, digest: &[u8; 32]) -> Result<DataBlob, Error> {
 608
 609         let (chunk_path, digest_str) = self.chunk_store.chunk_path(digest);
 610
 611         proxmox::try_block!({
 612             let mut file = std::fs::File::open(&chunk_path)?;
 613             DataBlob::load_from_reader(&mut file)
 614         }).map_err(|err| format_err!(
 615             "store '{}', unable to load chunk '{}' - {}",
 616             self.name(),
 617             digest_str,
 618             err,
 619         ))
 620      }
 621
 622     pub fn load_manifest(
 623         &self,
 624         backup_dir: &BackupDir,
 625     ) -> Result<(BackupManifest, u64), Error> {
 626         let blob = self.load_blob(backup_dir, MANIFEST_BLOB_NAME)?;
 627         let raw_size = blob.raw_size();
 628         let manifest = BackupManifest::try_from(blob)?;
 629         Ok((manifest, raw_size))
 630     }
 631
 632     pub fn store_manifest(
 633         &self,
 634         backup_dir: &BackupDir,
 635         manifest: BackupManifest,
 636     ) -> Result<(), Error> {
 637         let manifest = serde_json::to_value(manifest)?;
 638         let manifest = serde_json::to_string_pretty(&manifest)?;
 639         let blob = DataBlob::encode(manifest.as_bytes(), None, true)?;
 640         let raw_data = blob.raw_data();
 641
 642         let mut path = self.base_path();
 643         path.push(backup_dir.relative_path());
 644         path.push(MANIFEST_BLOB_NAME);
 645
 646         replace_file(&path, raw_data, CreateOptions::new())?;
 647
 648         Ok(())
 649     }
 650 }