src/tape/pool_writer.rs

   1 use std::collections::HashSet;
   2 use std::path::Path;
   3 use std::time::SystemTime;
   4
   5 use anyhow::{bail, Error};
   6
   7 use proxmox::tools::Uuid;
   8
   9 use crate::{
  10     task_log,
  11     backup::{
  12         DataStore,
  13     },
  14     server::WorkerTask,
  15     tape::{
  16         TAPE_STATUS_DIR,
  17         MAX_CHUNK_ARCHIVE_SIZE,
  18         COMMIT_BLOCK_SIZE,
  19         TapeWrite,
  20         SnapshotReader,
  21         SnapshotChunkIterator,
  22         MediaPool,
  23         MediaId,
  24         MediaCatalog,
  25         MediaSetCatalog,
  26         file_formats::{
  27             MediaSetLabel,
  28             ChunkArchiveWriter,
  29             tape_write_snapshot_archive,
  30         },
  31         drive::{
  32             TapeDriver,
  33             request_and_load_media,
  34             tape_alert_flags_critical,
  35             media_changer,
  36         },
  37     },
  38     config::tape_encryption_keys::load_key_configs,
  39 };
  40
  41
  42 struct PoolWriterState {
  43     drive: Box<dyn TapeDriver>,
  44     catalog: MediaCatalog,
  45     // tell if we already moved to EOM
  46     at_eom: bool,
  47     // bytes written after the last tape fush/sync
  48     bytes_written: usize,
  49 }
  50
  51 impl PoolWriterState {
  52
  53     fn commit(&mut self) -> Result<(), Error> {
  54         self.drive.sync()?; // sync all data to the tape
  55         self.catalog.commit()?; // then commit the catalog
  56         self.bytes_written = 0;
  57         Ok(())
  58     }
  59 }
  60
  61 /// Helper to manage a backup job, writing several tapes of a pool
  62 pub struct PoolWriter {
  63     pool: MediaPool,
  64     drive_name: String,
  65     status: Option<PoolWriterState>,
  66     media_set_catalog: MediaSetCatalog,
  67 }
  68
  69 impl PoolWriter {
  70
  71     pub fn new(mut pool: MediaPool, drive_name: &str) -> Result<Self, Error> {
  72
  73         let current_time = proxmox::tools::time::epoch_i64();
  74
  75         pool.start_write_session(current_time)?;
  76
  77         let mut media_set_catalog = MediaSetCatalog::new();
  78
  79         // load all catalogs read-only at start
  80         for media_uuid in pool.current_media_list()? {
  81             let media_catalog = MediaCatalog::open(
  82                 Path::new(TAPE_STATUS_DIR),
  83                 &media_uuid,
  84                 false,
  85                 false,
  86             )?;
  87             media_set_catalog.append_catalog(media_catalog)?;
  88         }
  89
  90         Ok(Self {
  91             pool,
  92             drive_name: drive_name.to_string(),
  93             status: None,
  94             media_set_catalog,
  95          })
  96     }
  97
  98     pub fn pool(&mut self) -> &mut MediaPool {
  99         &mut self.pool
 100     }
 101
 102     /// Set media status to FULL (persistent - stores pool status)
 103     pub fn set_media_status_full(&mut self, uuid: &Uuid) -> Result<(), Error> {
 104         self.pool.set_media_status_full(&uuid)?;
 105         Ok(())
 106     }
 107
 108     pub fn contains_snapshot(&self, snapshot: &str) -> bool {
 109         if let Some(PoolWriterState { ref catalog, .. }) = self.status {
 110             if catalog.contains_snapshot(snapshot) {
 111                 return true;
 112             }
 113         }
 114         self.media_set_catalog.contains_snapshot(snapshot)
 115     }
 116
 117     /// Eject media and drop PoolWriterState (close drive)
 118     pub fn eject_media(&mut self, worker: &WorkerTask) -> Result<(), Error> {
 119         let mut status = match self.status.take() {
 120             Some(status) => status,
 121             None => return Ok(()), // no media loaded
 122         };
 123
 124         let (drive_config, _digest) = crate::config::drive::config()?;
 125
 126         if let Some((mut changer, _)) = media_changer(&drive_config, &self.drive_name)? {
 127             worker.log("eject media");
 128             status.drive.eject_media()?; // rewind and eject early, so that unload_media is faster
 129             drop(status); // close drive
 130             worker.log("unload media");
 131             changer.unload_media(None)?; //eject and unload
 132         } else {
 133             worker.log("standalone drive - ejecting media");
 134             status.drive.eject_media()?;
 135         }
 136
 137         Ok(())
 138     }
 139
 140     /// Export current media set and drop PoolWriterState (close drive)
 141     pub fn export_media_set(&mut self, worker: &WorkerTask) -> Result<(), Error> {
 142         let mut status = self.status.take();
 143
 144         let (drive_config, _digest) = crate::config::drive::config()?;
 145
 146         if let Some((mut changer, _)) = media_changer(&drive_config, &self.drive_name)? {
 147
 148             if let Some(ref mut status) = status {
 149                 worker.log("eject media");
 150                 status.drive.eject_media()?; // rewind and eject early, so that unload_media is faster
 151             }
 152             drop(status); // close drive
 153
 154             worker.log("unload media");
 155             changer.unload_media(None)?;
 156
 157             for media_uuid in self.pool.current_media_list()? {
 158                 let media = self.pool.lookup_media(media_uuid)?;
 159                 let label_text = media.label_text();
 160                 if let Some(slot) = changer.export_media(label_text)? {
 161                     worker.log(format!("exported media '{}' to import/export slot {}", label_text, slot));
 162                 } else {
 163                     worker.warn(format!("export failed - media '{}' is not online", label_text));
 164                 }
 165             }
 166
 167         } else if let Some(mut status) = status {
 168             worker.log("standalone drive - ejecting media instead of export");
 169             status.drive.eject_media()?;
 170         }
 171
 172         Ok(())
 173     }
 174
 175     /// commit changes to tape and catalog
 176     ///
 177     /// This is done automatically during a backupsession, but needs to
 178     /// be called explicitly before dropping the PoolWriter
 179     pub fn commit(&mut self) -> Result<(), Error> {
 180         if let Some(ref mut status) = self.status {
 181             status.commit()?;
 182         }
 183         Ok(())
 184     }
 185
 186     /// Load a writable media into the drive
 187     pub fn load_writable_media(&mut self, worker: &WorkerTask) -> Result<Uuid, Error> {
 188         let last_media_uuid = match self.status {
 189             Some(PoolWriterState { ref catalog, .. }) => Some(catalog.uuid().clone()),
 190             None => None,
 191         };
 192
 193         let current_time = proxmox::tools::time::epoch_i64();
 194         let media_uuid = self.pool.alloc_writable_media(current_time)?;
 195
 196         let media = self.pool.lookup_media(&media_uuid).unwrap();
 197
 198         let media_changed = match last_media_uuid {
 199             Some(ref last_media_uuid) => last_media_uuid != &media_uuid,
 200             None => true,
 201         };
 202
 203         if !media_changed {
 204             return Ok(media_uuid);
 205         }
 206
 207         task_log!(worker, "allocated new writable media '{}'", media.label_text());
 208
 209         // remove read-only catalog (we store a writable version in status)
 210         self.media_set_catalog.remove_catalog(&media_uuid);
 211
 212         if let Some(PoolWriterState {mut drive, catalog, .. }) = self.status.take() {
 213             self.media_set_catalog.append_catalog(catalog)?;
 214             task_log!(worker, "eject current media");
 215             drive.eject_media()?;
 216         }
 217
 218         let (drive_config, _digest) = crate::config::drive::config()?;
 219
 220         let (mut drive, old_media_id) =
 221             request_and_load_media(worker, &drive_config, &self.drive_name, media.label())?;
 222
 223         // test for critical tape alert flags
 224         if let Ok(alert_flags) = drive.tape_alert_flags() {
 225             if !alert_flags.is_empty() {
 226                 worker.log(format!("TapeAlertFlags: {:?}", alert_flags));
 227                 if tape_alert_flags_critical(alert_flags) {
 228                     self.pool.set_media_status_damaged(&media_uuid)?;
 229                     bail!("aborting due to critical tape alert flags: {:?}", alert_flags);
 230                 }
 231             }
 232         }
 233
 234         let catalog = update_media_set_label(
 235             worker,
 236             drive.as_mut(),
 237             old_media_id.media_set_label,
 238             media.id(),
 239         )?;
 240
 241         let media_set = media.media_set_label().clone().unwrap();
 242
 243         let encrypt_fingerprint = media_set
 244             .encryption_key_fingerprint
 245             .clone()
 246             .map(|fp| (fp, media_set.uuid.clone()));
 247
 248         drive.set_encryption(encrypt_fingerprint)?;
 249
 250         self.status = Some(PoolWriterState { drive, catalog, at_eom: false, bytes_written: 0 });
 251
 252         Ok(media_uuid)
 253     }
 254
 255     /// uuid of currently loaded BackupMedia
 256     pub fn current_media_uuid(&self) -> Result<&Uuid, Error> {
 257         match self.status {
 258             Some(PoolWriterState { ref catalog, ..}) => Ok(catalog.uuid()),
 259             None => bail!("PoolWriter - no media loaded"),
 260         }
 261     }
 262
 263     /// Move to EOM (if not aleady there), then creates a new snapshot
 264     /// archive writing specified files (as .pxar) into it. On
 265     /// success, this return 'Ok(true)' and the media catalog gets
 266     /// updated.
 267
 268     /// Please note that this may fail when there is not enough space
 269     /// on the media (return value 'Ok(false, _)'). In that case, the
 270     /// archive is marked incomplete, and we do not use it. The caller
 271     /// should mark the media as full and try again using another
 272     /// media.
 273     pub fn append_snapshot_archive(
 274         &mut self,
 275         worker: &WorkerTask,
 276         snapshot_reader: &SnapshotReader,
 277     ) -> Result<(bool, usize), Error> {
 278
 279         let status = match self.status {
 280             Some(ref mut status) => status,
 281             None => bail!("PoolWriter - no media loaded"),
 282         };
 283
 284         if !status.at_eom {
 285             worker.log(String::from("moving to end of media"));
 286             status.drive.move_to_eom()?;
 287             status.at_eom = true;
 288         }
 289
 290         let current_file_number = status.drive.current_file_number()?;
 291         if current_file_number < 2 {
 292             bail!("got strange file position number from drive ({})", current_file_number);
 293         }
 294
 295         let (done, bytes_written) = {
 296             let mut writer: Box<dyn TapeWrite> = status.drive.write_file()?;
 297
 298             match tape_write_snapshot_archive(writer.as_mut(), snapshot_reader)? {
 299                 Some(content_uuid) => {
 300                     status.catalog.register_snapshot(
 301                         content_uuid,
 302                         current_file_number,
 303                         &snapshot_reader.snapshot().to_string(),
 304                     )?;
 305                     (true, writer.bytes_written())
 306                 }
 307                 None => (false, writer.bytes_written()),
 308             }
 309         };
 310
 311         status.bytes_written += bytes_written;
 312
 313         let request_sync = status.bytes_written >= COMMIT_BLOCK_SIZE;
 314
 315         if !done || request_sync {
 316             status.commit()?;
 317         }
 318
 319         Ok((done, bytes_written))
 320     }
 321
 322     /// Move to EOM (if not aleady there), then creates a new chunk
 323     /// archive and writes chunks from 'chunk_iter'. This stops when
 324     /// it detect LEOM or when we reach max archive size
 325     /// (4GB). Written chunks are registered in the media catalog.
 326     pub fn append_chunk_archive(
 327         &mut self,
 328         worker: &WorkerTask,
 329         datastore: &DataStore,
 330         chunk_iter: &mut std::iter::Peekable<SnapshotChunkIterator>,
 331     ) -> Result<(bool, usize), Error> {
 332
 333         let status = match self.status {
 334             Some(ref mut status) => status,
 335             None => bail!("PoolWriter - no media loaded"),
 336         };
 337
 338         if !status.at_eom {
 339             worker.log(String::from("moving to end of media"));
 340             status.drive.move_to_eom()?;
 341             status.at_eom = true;
 342         }
 343
 344         let current_file_number = status.drive.current_file_number()?;
 345         if current_file_number < 2 {
 346             bail!("got strange file position number from drive ({})", current_file_number);
 347         }
 348         let writer = status.drive.write_file()?;
 349
 350         let start_time = SystemTime::now();
 351
 352         let (saved_chunks, content_uuid, leom, bytes_written) = write_chunk_archive(
 353             worker,
 354             writer,
 355             datastore,
 356             chunk_iter,
 357             &self.media_set_catalog,
 358             &status.catalog,
 359             MAX_CHUNK_ARCHIVE_SIZE,
 360         )?;
 361
 362         status.bytes_written += bytes_written;
 363
 364         let elapsed =  start_time.elapsed()?.as_secs_f64();
 365         worker.log(format!(
 366             "wrote {:.2} MB ({} MB/s)",
 367             bytes_written as f64 / (1024.0*1024.0),
 368             (bytes_written as f64)/(1024.0*1024.0*elapsed),
 369         ));
 370
 371         let request_sync = status.bytes_written >= COMMIT_BLOCK_SIZE;
 372
 373         // register chunks in media_catalog
 374         status.catalog.start_chunk_archive(content_uuid, current_file_number)?;
 375         for digest in saved_chunks {
 376             status.catalog.register_chunk(&digest)?;
 377         }
 378         status.catalog.end_chunk_archive()?;
 379
 380         if leom || request_sync {
 381             status.commit()?;
 382         }
 383
 384         Ok((leom, bytes_written))
 385     }
 386 }
 387
 388 /// write up to <max_size> of chunks
 389 fn write_chunk_archive<'a>(
 390     worker: &WorkerTask,
 391     writer: Box<dyn 'a + TapeWrite>,
 392     datastore: &DataStore,
 393     chunk_iter: &mut std::iter::Peekable<SnapshotChunkIterator>,
 394     media_set_catalog: &MediaSetCatalog,
 395     media_catalog: &MediaCatalog,
 396     max_size: usize,
 397 ) -> Result<(Vec<[u8;32]>, Uuid, bool, usize), Error> {
 398
 399     let (mut writer, content_uuid) = ChunkArchiveWriter::new(writer, true)?;
 400
 401     let mut chunk_index: HashSet<[u8;32]> = HashSet::new();
 402
 403     // we want to get the chunk list in correct order
 404     let mut chunk_list: Vec<[u8;32]> = Vec::new();
 405
 406     let mut leom = false;
 407
 408     loop {
 409         let digest = match chunk_iter.next() {
 410             None => break,
 411             Some(digest) => digest?,
 412         };
 413         if media_catalog.contains_chunk(&digest)
 414             || chunk_index.contains(&digest)
 415             || media_set_catalog.contains_chunk(&digest)
 416         {
 417             continue;
 418         }
 419
 420         let blob = datastore.load_chunk(&digest)?;
 421         //println!("CHUNK {} size {}", proxmox::tools::digest_to_hex(&digest), blob.raw_size());
 422
 423         match writer.try_write_chunk(&digest, &blob) {
 424              Ok(true) => {
 425                 chunk_index.insert(digest);
 426                 chunk_list.push(digest);
 427             }
 428             Ok(false) => {
 429                 leom = true;
 430                 break;
 431             }
 432             Err(err) => bail!("write chunk failed - {}", err),
 433         }
 434
 435         if writer.bytes_written() > max_size {
 436             worker.log("Chunk Archive max size reached, closing archive".to_string());
 437             break;
 438         }
 439     }
 440
 441     writer.finish()?;
 442
 443     Ok((chunk_list, content_uuid, leom, writer.bytes_written()))
 444 }
 445
 446 // Compare the media set label. If the media is empty, or the existing
 447 // set label does not match the expected media set, overwrite the
 448 // media set label.
 449 fn update_media_set_label(
 450     worker: &WorkerTask,
 451     drive: &mut dyn TapeDriver,
 452     old_set: Option<MediaSetLabel>,
 453     media_id: &MediaId,
 454 ) -> Result<MediaCatalog, Error> {
 455
 456     let media_catalog;
 457
 458     let new_set = match media_id.media_set_label {
 459         None => bail!("got media without media set - internal error"),
 460         Some(ref set) => set,
 461     };
 462
 463     let key_config = if let Some(ref fingerprint) = new_set.encryption_key_fingerprint {
 464         let (config_map, _digest) = load_key_configs()?;
 465         match config_map.get(fingerprint) {
 466             Some(key_config) => Some(key_config.clone()),
 467             None => {
 468                 bail!("unable to find tape encryption key config '{}'", fingerprint);
 469             }
 470         }
 471     } else {
 472         None
 473     };
 474
 475     let status_path = Path::new(TAPE_STATUS_DIR);
 476
 477     match old_set {
 478         None => {
 479             worker.log("wrinting new media set label".to_string());
 480             drive.write_media_set_label(new_set, key_config.as_ref())?;
 481             media_catalog = MediaCatalog::overwrite(status_path, media_id, false)?;
 482         }
 483         Some(media_set_label) => {
 484             if new_set.uuid == media_set_label.uuid {
 485                 if new_set.seq_nr != media_set_label.seq_nr {
 486                     bail!("got media with wrong media sequence number ({} != {}",
 487                           new_set.seq_nr,media_set_label.seq_nr);
 488                 }
 489                 if new_set.encryption_key_fingerprint != media_set_label.encryption_key_fingerprint {
 490                     bail!("detected changed encryption fingerprint - internal error");
 491                 }
 492                 media_catalog = MediaCatalog::open(status_path, &media_id.label.uuid, true, false)?;
 493             } else {
 494                 worker.log(
 495                     format!("wrinting new media set label (overwrite '{}/{}')",
 496                             media_set_label.uuid.to_string(), media_set_label.seq_nr)
 497                 );
 498
 499                 drive.write_media_set_label(new_set, key_config.as_ref())?;
 500                 media_catalog = MediaCatalog::overwrite(status_path, media_id, false)?;
 501             }
 502         }
 503     }
 504
 505     // todo: verify last content/media_catalog somehow?
 506     drive.move_to_eom()?; // just to be sure
 507
 508     Ok(media_catalog)
 509 }