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