khenaidoo | ffe076b | 2019-01-15 16:08:08 -0500 | [diff] [blame^] | 1 | package bolt |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "fmt" |
| 6 | "hash/fnv" |
| 7 | "log" |
| 8 | "os" |
| 9 | "runtime" |
| 10 | "runtime/debug" |
| 11 | "strings" |
| 12 | "sync" |
| 13 | "time" |
| 14 | "unsafe" |
| 15 | ) |
| 16 | |
| 17 | // The largest step that can be taken when remapping the mmap. |
| 18 | const maxMmapStep = 1 << 30 // 1GB |
| 19 | |
| 20 | // The data file format version. |
| 21 | const version = 2 |
| 22 | |
| 23 | // Represents a marker value to indicate that a file is a Bolt DB. |
| 24 | const magic uint32 = 0xED0CDAED |
| 25 | |
| 26 | // IgnoreNoSync specifies whether the NoSync field of a DB is ignored when |
| 27 | // syncing changes to a file. This is required as some operating systems, |
| 28 | // such as OpenBSD, do not have a unified buffer cache (UBC) and writes |
| 29 | // must be synchronized using the msync(2) syscall. |
| 30 | const IgnoreNoSync = runtime.GOOS == "openbsd" |
| 31 | |
| 32 | // Default values if not set in a DB instance. |
| 33 | const ( |
| 34 | DefaultMaxBatchSize int = 1000 |
| 35 | DefaultMaxBatchDelay = 10 * time.Millisecond |
| 36 | DefaultAllocSize = 16 * 1024 * 1024 |
| 37 | ) |
| 38 | |
| 39 | // default page size for db is set to the OS page size. |
| 40 | var defaultPageSize = os.Getpagesize() |
| 41 | |
| 42 | // DB represents a collection of buckets persisted to a file on disk. |
| 43 | // All data access is performed through transactions which can be obtained through the DB. |
| 44 | // All the functions on DB will return a ErrDatabaseNotOpen if accessed before Open() is called. |
| 45 | type DB struct { |
| 46 | // When enabled, the database will perform a Check() after every commit. |
| 47 | // A panic is issued if the database is in an inconsistent state. This |
| 48 | // flag has a large performance impact so it should only be used for |
| 49 | // debugging purposes. |
| 50 | StrictMode bool |
| 51 | |
| 52 | // Setting the NoSync flag will cause the database to skip fsync() |
| 53 | // calls after each commit. This can be useful when bulk loading data |
| 54 | // into a database and you can restart the bulk load in the event of |
| 55 | // a system failure or database corruption. Do not set this flag for |
| 56 | // normal use. |
| 57 | // |
| 58 | // If the package global IgnoreNoSync constant is true, this value is |
| 59 | // ignored. See the comment on that constant for more details. |
| 60 | // |
| 61 | // THIS IS UNSAFE. PLEASE USE WITH CAUTION. |
| 62 | NoSync bool |
| 63 | |
| 64 | // When true, skips the truncate call when growing the database. |
| 65 | // Setting this to true is only safe on non-ext3/ext4 systems. |
| 66 | // Skipping truncation avoids preallocation of hard drive space and |
| 67 | // bypasses a truncate() and fsync() syscall on remapping. |
| 68 | // |
| 69 | // https://github.com/boltdb/bolt/issues/284 |
| 70 | NoGrowSync bool |
| 71 | |
| 72 | // If you want to read the entire database fast, you can set MmapFlag to |
| 73 | // syscall.MAP_POPULATE on Linux 2.6.23+ for sequential read-ahead. |
| 74 | MmapFlags int |
| 75 | |
| 76 | // MaxBatchSize is the maximum size of a batch. Default value is |
| 77 | // copied from DefaultMaxBatchSize in Open. |
| 78 | // |
| 79 | // If <=0, disables batching. |
| 80 | // |
| 81 | // Do not change concurrently with calls to Batch. |
| 82 | MaxBatchSize int |
| 83 | |
| 84 | // MaxBatchDelay is the maximum delay before a batch starts. |
| 85 | // Default value is copied from DefaultMaxBatchDelay in Open. |
| 86 | // |
| 87 | // If <=0, effectively disables batching. |
| 88 | // |
| 89 | // Do not change concurrently with calls to Batch. |
| 90 | MaxBatchDelay time.Duration |
| 91 | |
| 92 | // AllocSize is the amount of space allocated when the database |
| 93 | // needs to create new pages. This is done to amortize the cost |
| 94 | // of truncate() and fsync() when growing the data file. |
| 95 | AllocSize int |
| 96 | |
| 97 | path string |
| 98 | file *os.File |
| 99 | lockfile *os.File // windows only |
| 100 | dataref []byte // mmap'ed readonly, write throws SEGV |
| 101 | data *[maxMapSize]byte |
| 102 | datasz int |
| 103 | filesz int // current on disk file size |
| 104 | meta0 *meta |
| 105 | meta1 *meta |
| 106 | pageSize int |
| 107 | opened bool |
| 108 | rwtx *Tx |
| 109 | txs []*Tx |
| 110 | freelist *freelist |
| 111 | stats Stats |
| 112 | |
| 113 | pagePool sync.Pool |
| 114 | |
| 115 | batchMu sync.Mutex |
| 116 | batch *batch |
| 117 | |
| 118 | rwlock sync.Mutex // Allows only one writer at a time. |
| 119 | metalock sync.Mutex // Protects meta page access. |
| 120 | mmaplock sync.RWMutex // Protects mmap access during remapping. |
| 121 | statlock sync.RWMutex // Protects stats access. |
| 122 | |
| 123 | ops struct { |
| 124 | writeAt func(b []byte, off int64) (n int, err error) |
| 125 | } |
| 126 | |
| 127 | // Read only mode. |
| 128 | // When true, Update() and Begin(true) return ErrDatabaseReadOnly immediately. |
| 129 | readOnly bool |
| 130 | } |
| 131 | |
| 132 | // Path returns the path to currently open database file. |
| 133 | func (db *DB) Path() string { |
| 134 | return db.path |
| 135 | } |
| 136 | |
| 137 | // GoString returns the Go string representation of the database. |
| 138 | func (db *DB) GoString() string { |
| 139 | return fmt.Sprintf("bolt.DB{path:%q}", db.path) |
| 140 | } |
| 141 | |
| 142 | // String returns the string representation of the database. |
| 143 | func (db *DB) String() string { |
| 144 | return fmt.Sprintf("DB<%q>", db.path) |
| 145 | } |
| 146 | |
| 147 | // Open creates and opens a database at the given path. |
| 148 | // If the file does not exist then it will be created automatically. |
| 149 | // Passing in nil options will cause Bolt to open the database with the default options. |
| 150 | func Open(path string, mode os.FileMode, options *Options) (*DB, error) { |
| 151 | var db = &DB{opened: true} |
| 152 | |
| 153 | // Set default options if no options are provided. |
| 154 | if options == nil { |
| 155 | options = DefaultOptions |
| 156 | } |
| 157 | db.NoGrowSync = options.NoGrowSync |
| 158 | db.MmapFlags = options.MmapFlags |
| 159 | |
| 160 | // Set default values for later DB operations. |
| 161 | db.MaxBatchSize = DefaultMaxBatchSize |
| 162 | db.MaxBatchDelay = DefaultMaxBatchDelay |
| 163 | db.AllocSize = DefaultAllocSize |
| 164 | |
| 165 | flag := os.O_RDWR |
| 166 | if options.ReadOnly { |
| 167 | flag = os.O_RDONLY |
| 168 | db.readOnly = true |
| 169 | } |
| 170 | |
| 171 | // Open data file and separate sync handler for metadata writes. |
| 172 | db.path = path |
| 173 | var err error |
| 174 | if db.file, err = os.OpenFile(db.path, flag|os.O_CREATE, mode); err != nil { |
| 175 | _ = db.close() |
| 176 | return nil, err |
| 177 | } |
| 178 | |
| 179 | // Lock file so that other processes using Bolt in read-write mode cannot |
| 180 | // use the database at the same time. This would cause corruption since |
| 181 | // the two processes would write meta pages and free pages separately. |
| 182 | // The database file is locked exclusively (only one process can grab the lock) |
| 183 | // if !options.ReadOnly. |
| 184 | // The database file is locked using the shared lock (more than one process may |
| 185 | // hold a lock at the same time) otherwise (options.ReadOnly is set). |
| 186 | if err := flock(db, mode, !db.readOnly, options.Timeout); err != nil { |
| 187 | _ = db.close() |
| 188 | return nil, err |
| 189 | } |
| 190 | |
| 191 | // Default values for test hooks |
| 192 | db.ops.writeAt = db.file.WriteAt |
| 193 | |
| 194 | // Initialize the database if it doesn't exist. |
| 195 | if info, err := db.file.Stat(); err != nil { |
| 196 | return nil, err |
| 197 | } else if info.Size() == 0 { |
| 198 | // Initialize new files with meta pages. |
| 199 | if err := db.init(); err != nil { |
| 200 | return nil, err |
| 201 | } |
| 202 | } else { |
| 203 | // Read the first meta page to determine the page size. |
| 204 | var buf [0x1000]byte |
| 205 | if _, err := db.file.ReadAt(buf[:], 0); err == nil { |
| 206 | m := db.pageInBuffer(buf[:], 0).meta() |
| 207 | if err := m.validate(); err != nil { |
| 208 | // If we can't read the page size, we can assume it's the same |
| 209 | // as the OS -- since that's how the page size was chosen in the |
| 210 | // first place. |
| 211 | // |
| 212 | // If the first page is invalid and this OS uses a different |
| 213 | // page size than what the database was created with then we |
| 214 | // are out of luck and cannot access the database. |
| 215 | db.pageSize = os.Getpagesize() |
| 216 | } else { |
| 217 | db.pageSize = int(m.pageSize) |
| 218 | } |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | // Initialize page pool. |
| 223 | db.pagePool = sync.Pool{ |
| 224 | New: func() interface{} { |
| 225 | return make([]byte, db.pageSize) |
| 226 | }, |
| 227 | } |
| 228 | |
| 229 | // Memory map the data file. |
| 230 | if err := db.mmap(options.InitialMmapSize); err != nil { |
| 231 | _ = db.close() |
| 232 | return nil, err |
| 233 | } |
| 234 | |
| 235 | // Read in the freelist. |
| 236 | db.freelist = newFreelist() |
| 237 | db.freelist.read(db.page(db.meta().freelist)) |
| 238 | |
| 239 | // Mark the database as opened and return. |
| 240 | return db, nil |
| 241 | } |
| 242 | |
| 243 | // mmap opens the underlying memory-mapped file and initializes the meta references. |
| 244 | // minsz is the minimum size that the new mmap can be. |
| 245 | func (db *DB) mmap(minsz int) error { |
| 246 | db.mmaplock.Lock() |
| 247 | defer db.mmaplock.Unlock() |
| 248 | |
| 249 | info, err := db.file.Stat() |
| 250 | if err != nil { |
| 251 | return fmt.Errorf("mmap stat error: %s", err) |
| 252 | } else if int(info.Size()) < db.pageSize*2 { |
| 253 | return fmt.Errorf("file size too small") |
| 254 | } |
| 255 | |
| 256 | // Ensure the size is at least the minimum size. |
| 257 | var size = int(info.Size()) |
| 258 | if size < minsz { |
| 259 | size = minsz |
| 260 | } |
| 261 | size, err = db.mmapSize(size) |
| 262 | if err != nil { |
| 263 | return err |
| 264 | } |
| 265 | |
| 266 | // Dereference all mmap references before unmapping. |
| 267 | if db.rwtx != nil { |
| 268 | db.rwtx.root.dereference() |
| 269 | } |
| 270 | |
| 271 | // Unmap existing data before continuing. |
| 272 | if err := db.munmap(); err != nil { |
| 273 | return err |
| 274 | } |
| 275 | |
| 276 | // Memory-map the data file as a byte slice. |
| 277 | if err := mmap(db, size); err != nil { |
| 278 | return err |
| 279 | } |
| 280 | |
| 281 | // Save references to the meta pages. |
| 282 | db.meta0 = db.page(0).meta() |
| 283 | db.meta1 = db.page(1).meta() |
| 284 | |
| 285 | // Validate the meta pages. We only return an error if both meta pages fail |
| 286 | // validation, since meta0 failing validation means that it wasn't saved |
| 287 | // properly -- but we can recover using meta1. And vice-versa. |
| 288 | err0 := db.meta0.validate() |
| 289 | err1 := db.meta1.validate() |
| 290 | if err0 != nil && err1 != nil { |
| 291 | return err0 |
| 292 | } |
| 293 | |
| 294 | return nil |
| 295 | } |
| 296 | |
| 297 | // munmap unmaps the data file from memory. |
| 298 | func (db *DB) munmap() error { |
| 299 | if err := munmap(db); err != nil { |
| 300 | return fmt.Errorf("unmap error: " + err.Error()) |
| 301 | } |
| 302 | return nil |
| 303 | } |
| 304 | |
| 305 | // mmapSize determines the appropriate size for the mmap given the current size |
| 306 | // of the database. The minimum size is 32KB and doubles until it reaches 1GB. |
| 307 | // Returns an error if the new mmap size is greater than the max allowed. |
| 308 | func (db *DB) mmapSize(size int) (int, error) { |
| 309 | // Double the size from 32KB until 1GB. |
| 310 | for i := uint(15); i <= 30; i++ { |
| 311 | if size <= 1<<i { |
| 312 | return 1 << i, nil |
| 313 | } |
| 314 | } |
| 315 | |
| 316 | // Verify the requested size is not above the maximum allowed. |
| 317 | if size > maxMapSize { |
| 318 | return 0, fmt.Errorf("mmap too large") |
| 319 | } |
| 320 | |
| 321 | // If larger than 1GB then grow by 1GB at a time. |
| 322 | sz := int64(size) |
| 323 | if remainder := sz % int64(maxMmapStep); remainder > 0 { |
| 324 | sz += int64(maxMmapStep) - remainder |
| 325 | } |
| 326 | |
| 327 | // Ensure that the mmap size is a multiple of the page size. |
| 328 | // This should always be true since we're incrementing in MBs. |
| 329 | pageSize := int64(db.pageSize) |
| 330 | if (sz % pageSize) != 0 { |
| 331 | sz = ((sz / pageSize) + 1) * pageSize |
| 332 | } |
| 333 | |
| 334 | // If we've exceeded the max size then only grow up to the max size. |
| 335 | if sz > maxMapSize { |
| 336 | sz = maxMapSize |
| 337 | } |
| 338 | |
| 339 | return int(sz), nil |
| 340 | } |
| 341 | |
| 342 | // init creates a new database file and initializes its meta pages. |
| 343 | func (db *DB) init() error { |
| 344 | // Set the page size to the OS page size. |
| 345 | db.pageSize = os.Getpagesize() |
| 346 | |
| 347 | // Create two meta pages on a buffer. |
| 348 | buf := make([]byte, db.pageSize*4) |
| 349 | for i := 0; i < 2; i++ { |
| 350 | p := db.pageInBuffer(buf[:], pgid(i)) |
| 351 | p.id = pgid(i) |
| 352 | p.flags = metaPageFlag |
| 353 | |
| 354 | // Initialize the meta page. |
| 355 | m := p.meta() |
| 356 | m.magic = magic |
| 357 | m.version = version |
| 358 | m.pageSize = uint32(db.pageSize) |
| 359 | m.freelist = 2 |
| 360 | m.root = bucket{root: 3} |
| 361 | m.pgid = 4 |
| 362 | m.txid = txid(i) |
| 363 | m.checksum = m.sum64() |
| 364 | } |
| 365 | |
| 366 | // Write an empty freelist at page 3. |
| 367 | p := db.pageInBuffer(buf[:], pgid(2)) |
| 368 | p.id = pgid(2) |
| 369 | p.flags = freelistPageFlag |
| 370 | p.count = 0 |
| 371 | |
| 372 | // Write an empty leaf page at page 4. |
| 373 | p = db.pageInBuffer(buf[:], pgid(3)) |
| 374 | p.id = pgid(3) |
| 375 | p.flags = leafPageFlag |
| 376 | p.count = 0 |
| 377 | |
| 378 | // Write the buffer to our data file. |
| 379 | if _, err := db.ops.writeAt(buf, 0); err != nil { |
| 380 | return err |
| 381 | } |
| 382 | if err := fdatasync(db); err != nil { |
| 383 | return err |
| 384 | } |
| 385 | |
| 386 | return nil |
| 387 | } |
| 388 | |
| 389 | // Close releases all database resources. |
| 390 | // All transactions must be closed before closing the database. |
| 391 | func (db *DB) Close() error { |
| 392 | db.rwlock.Lock() |
| 393 | defer db.rwlock.Unlock() |
| 394 | |
| 395 | db.metalock.Lock() |
| 396 | defer db.metalock.Unlock() |
| 397 | |
| 398 | db.mmaplock.RLock() |
| 399 | defer db.mmaplock.RUnlock() |
| 400 | |
| 401 | return db.close() |
| 402 | } |
| 403 | |
| 404 | func (db *DB) close() error { |
| 405 | if !db.opened { |
| 406 | return nil |
| 407 | } |
| 408 | |
| 409 | db.opened = false |
| 410 | |
| 411 | db.freelist = nil |
| 412 | |
| 413 | // Clear ops. |
| 414 | db.ops.writeAt = nil |
| 415 | |
| 416 | // Close the mmap. |
| 417 | if err := db.munmap(); err != nil { |
| 418 | return err |
| 419 | } |
| 420 | |
| 421 | // Close file handles. |
| 422 | if db.file != nil { |
| 423 | // No need to unlock read-only file. |
| 424 | if !db.readOnly { |
| 425 | // Unlock the file. |
| 426 | if err := funlock(db); err != nil { |
| 427 | log.Printf("bolt.Close(): funlock error: %s", err) |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | // Close the file descriptor. |
| 432 | if err := db.file.Close(); err != nil { |
| 433 | return fmt.Errorf("db file close: %s", err) |
| 434 | } |
| 435 | db.file = nil |
| 436 | } |
| 437 | |
| 438 | db.path = "" |
| 439 | return nil |
| 440 | } |
| 441 | |
| 442 | // Begin starts a new transaction. |
| 443 | // Multiple read-only transactions can be used concurrently but only one |
| 444 | // write transaction can be used at a time. Starting multiple write transactions |
| 445 | // will cause the calls to block and be serialized until the current write |
| 446 | // transaction finishes. |
| 447 | // |
| 448 | // Transactions should not be dependent on one another. Opening a read |
| 449 | // transaction and a write transaction in the same goroutine can cause the |
| 450 | // writer to deadlock because the database periodically needs to re-mmap itself |
| 451 | // as it grows and it cannot do that while a read transaction is open. |
| 452 | // |
| 453 | // If a long running read transaction (for example, a snapshot transaction) is |
| 454 | // needed, you might want to set DB.InitialMmapSize to a large enough value |
| 455 | // to avoid potential blocking of write transaction. |
| 456 | // |
| 457 | // IMPORTANT: You must close read-only transactions after you are finished or |
| 458 | // else the database will not reclaim old pages. |
| 459 | func (db *DB) Begin(writable bool) (*Tx, error) { |
| 460 | if writable { |
| 461 | return db.beginRWTx() |
| 462 | } |
| 463 | return db.beginTx() |
| 464 | } |
| 465 | |
| 466 | func (db *DB) beginTx() (*Tx, error) { |
| 467 | // Lock the meta pages while we initialize the transaction. We obtain |
| 468 | // the meta lock before the mmap lock because that's the order that the |
| 469 | // write transaction will obtain them. |
| 470 | db.metalock.Lock() |
| 471 | |
| 472 | // Obtain a read-only lock on the mmap. When the mmap is remapped it will |
| 473 | // obtain a write lock so all transactions must finish before it can be |
| 474 | // remapped. |
| 475 | db.mmaplock.RLock() |
| 476 | |
| 477 | // Exit if the database is not open yet. |
| 478 | if !db.opened { |
| 479 | db.mmaplock.RUnlock() |
| 480 | db.metalock.Unlock() |
| 481 | return nil, ErrDatabaseNotOpen |
| 482 | } |
| 483 | |
| 484 | // Create a transaction associated with the database. |
| 485 | t := &Tx{} |
| 486 | t.init(db) |
| 487 | |
| 488 | // Keep track of transaction until it closes. |
| 489 | db.txs = append(db.txs, t) |
| 490 | n := len(db.txs) |
| 491 | |
| 492 | // Unlock the meta pages. |
| 493 | db.metalock.Unlock() |
| 494 | |
| 495 | // Update the transaction stats. |
| 496 | db.statlock.Lock() |
| 497 | db.stats.TxN++ |
| 498 | db.stats.OpenTxN = n |
| 499 | db.statlock.Unlock() |
| 500 | |
| 501 | return t, nil |
| 502 | } |
| 503 | |
| 504 | func (db *DB) beginRWTx() (*Tx, error) { |
| 505 | // If the database was opened with Options.ReadOnly, return an error. |
| 506 | if db.readOnly { |
| 507 | return nil, ErrDatabaseReadOnly |
| 508 | } |
| 509 | |
| 510 | // Obtain writer lock. This is released by the transaction when it closes. |
| 511 | // This enforces only one writer transaction at a time. |
| 512 | db.rwlock.Lock() |
| 513 | |
| 514 | // Once we have the writer lock then we can lock the meta pages so that |
| 515 | // we can set up the transaction. |
| 516 | db.metalock.Lock() |
| 517 | defer db.metalock.Unlock() |
| 518 | |
| 519 | // Exit if the database is not open yet. |
| 520 | if !db.opened { |
| 521 | db.rwlock.Unlock() |
| 522 | return nil, ErrDatabaseNotOpen |
| 523 | } |
| 524 | |
| 525 | // Create a transaction associated with the database. |
| 526 | t := &Tx{writable: true} |
| 527 | t.init(db) |
| 528 | db.rwtx = t |
| 529 | |
| 530 | // Free any pages associated with closed read-only transactions. |
| 531 | var minid txid = 0xFFFFFFFFFFFFFFFF |
| 532 | for _, t := range db.txs { |
| 533 | if t.meta.txid < minid { |
| 534 | minid = t.meta.txid |
| 535 | } |
| 536 | } |
| 537 | if minid > 0 { |
| 538 | db.freelist.release(minid - 1) |
| 539 | } |
| 540 | |
| 541 | return t, nil |
| 542 | } |
| 543 | |
| 544 | // removeTx removes a transaction from the database. |
| 545 | func (db *DB) removeTx(tx *Tx) { |
| 546 | // Release the read lock on the mmap. |
| 547 | db.mmaplock.RUnlock() |
| 548 | |
| 549 | // Use the meta lock to restrict access to the DB object. |
| 550 | db.metalock.Lock() |
| 551 | |
| 552 | // Remove the transaction. |
| 553 | for i, t := range db.txs { |
| 554 | if t == tx { |
| 555 | db.txs = append(db.txs[:i], db.txs[i+1:]...) |
| 556 | break |
| 557 | } |
| 558 | } |
| 559 | n := len(db.txs) |
| 560 | |
| 561 | // Unlock the meta pages. |
| 562 | db.metalock.Unlock() |
| 563 | |
| 564 | // Merge statistics. |
| 565 | db.statlock.Lock() |
| 566 | db.stats.OpenTxN = n |
| 567 | db.stats.TxStats.add(&tx.stats) |
| 568 | db.statlock.Unlock() |
| 569 | } |
| 570 | |
| 571 | // Update executes a function within the context of a read-write managed transaction. |
| 572 | // If no error is returned from the function then the transaction is committed. |
| 573 | // If an error is returned then the entire transaction is rolled back. |
| 574 | // Any error that is returned from the function or returned from the commit is |
| 575 | // returned from the Update() method. |
| 576 | // |
| 577 | // Attempting to manually commit or rollback within the function will cause a panic. |
| 578 | func (db *DB) Update(fn func(*Tx) error) error { |
| 579 | t, err := db.Begin(true) |
| 580 | if err != nil { |
| 581 | return err |
| 582 | } |
| 583 | |
| 584 | // Make sure the transaction rolls back in the event of a panic. |
| 585 | defer func() { |
| 586 | if t.db != nil { |
| 587 | t.rollback() |
| 588 | } |
| 589 | }() |
| 590 | |
| 591 | // Mark as a managed tx so that the inner function cannot manually commit. |
| 592 | t.managed = true |
| 593 | |
| 594 | // If an error is returned from the function then rollback and return error. |
| 595 | err = fn(t) |
| 596 | t.managed = false |
| 597 | if err != nil { |
| 598 | _ = t.Rollback() |
| 599 | return err |
| 600 | } |
| 601 | |
| 602 | return t.Commit() |
| 603 | } |
| 604 | |
| 605 | // View executes a function within the context of a managed read-only transaction. |
| 606 | // Any error that is returned from the function is returned from the View() method. |
| 607 | // |
| 608 | // Attempting to manually rollback within the function will cause a panic. |
| 609 | func (db *DB) View(fn func(*Tx) error) error { |
| 610 | t, err := db.Begin(false) |
| 611 | if err != nil { |
| 612 | return err |
| 613 | } |
| 614 | |
| 615 | // Make sure the transaction rolls back in the event of a panic. |
| 616 | defer func() { |
| 617 | if t.db != nil { |
| 618 | t.rollback() |
| 619 | } |
| 620 | }() |
| 621 | |
| 622 | // Mark as a managed tx so that the inner function cannot manually rollback. |
| 623 | t.managed = true |
| 624 | |
| 625 | // If an error is returned from the function then pass it through. |
| 626 | err = fn(t) |
| 627 | t.managed = false |
| 628 | if err != nil { |
| 629 | _ = t.Rollback() |
| 630 | return err |
| 631 | } |
| 632 | |
| 633 | if err := t.Rollback(); err != nil { |
| 634 | return err |
| 635 | } |
| 636 | |
| 637 | return nil |
| 638 | } |
| 639 | |
| 640 | // Batch calls fn as part of a batch. It behaves similar to Update, |
| 641 | // except: |
| 642 | // |
| 643 | // 1. concurrent Batch calls can be combined into a single Bolt |
| 644 | // transaction. |
| 645 | // |
| 646 | // 2. the function passed to Batch may be called multiple times, |
| 647 | // regardless of whether it returns error or not. |
| 648 | // |
| 649 | // This means that Batch function side effects must be idempotent and |
| 650 | // take permanent effect only after a successful return is seen in |
| 651 | // caller. |
| 652 | // |
| 653 | // The maximum batch size and delay can be adjusted with DB.MaxBatchSize |
| 654 | // and DB.MaxBatchDelay, respectively. |
| 655 | // |
| 656 | // Batch is only useful when there are multiple goroutines calling it. |
| 657 | func (db *DB) Batch(fn func(*Tx) error) error { |
| 658 | errCh := make(chan error, 1) |
| 659 | |
| 660 | db.batchMu.Lock() |
| 661 | if (db.batch == nil) || (db.batch != nil && len(db.batch.calls) >= db.MaxBatchSize) { |
| 662 | // There is no existing batch, or the existing batch is full; start a new one. |
| 663 | db.batch = &batch{ |
| 664 | db: db, |
| 665 | } |
| 666 | db.batch.timer = time.AfterFunc(db.MaxBatchDelay, db.batch.trigger) |
| 667 | } |
| 668 | db.batch.calls = append(db.batch.calls, call{fn: fn, err: errCh}) |
| 669 | if len(db.batch.calls) >= db.MaxBatchSize { |
| 670 | // wake up batch, it's ready to run |
| 671 | go db.batch.trigger() |
| 672 | } |
| 673 | db.batchMu.Unlock() |
| 674 | |
| 675 | err := <-errCh |
| 676 | if err == trySolo { |
| 677 | err = db.Update(fn) |
| 678 | } |
| 679 | return err |
| 680 | } |
| 681 | |
| 682 | type call struct { |
| 683 | fn func(*Tx) error |
| 684 | err chan<- error |
| 685 | } |
| 686 | |
| 687 | type batch struct { |
| 688 | db *DB |
| 689 | timer *time.Timer |
| 690 | start sync.Once |
| 691 | calls []call |
| 692 | } |
| 693 | |
| 694 | // trigger runs the batch if it hasn't already been run. |
| 695 | func (b *batch) trigger() { |
| 696 | b.start.Do(b.run) |
| 697 | } |
| 698 | |
| 699 | // run performs the transactions in the batch and communicates results |
| 700 | // back to DB.Batch. |
| 701 | func (b *batch) run() { |
| 702 | b.db.batchMu.Lock() |
| 703 | b.timer.Stop() |
| 704 | // Make sure no new work is added to this batch, but don't break |
| 705 | // other batches. |
| 706 | if b.db.batch == b { |
| 707 | b.db.batch = nil |
| 708 | } |
| 709 | b.db.batchMu.Unlock() |
| 710 | |
| 711 | retry: |
| 712 | for len(b.calls) > 0 { |
| 713 | var failIdx = -1 |
| 714 | err := b.db.Update(func(tx *Tx) error { |
| 715 | for i, c := range b.calls { |
| 716 | if err := safelyCall(c.fn, tx); err != nil { |
| 717 | failIdx = i |
| 718 | return err |
| 719 | } |
| 720 | } |
| 721 | return nil |
| 722 | }) |
| 723 | |
| 724 | if failIdx >= 0 { |
| 725 | // take the failing transaction out of the batch. it's |
| 726 | // safe to shorten b.calls here because db.batch no longer |
| 727 | // points to us, and we hold the mutex anyway. |
| 728 | c := b.calls[failIdx] |
| 729 | b.calls[failIdx], b.calls = b.calls[len(b.calls)-1], b.calls[:len(b.calls)-1] |
| 730 | // tell the submitter re-run it solo, continue with the rest of the batch |
| 731 | c.err <- trySolo |
| 732 | continue retry |
| 733 | } |
| 734 | |
| 735 | // pass success, or bolt internal errors, to all callers |
| 736 | for _, c := range b.calls { |
| 737 | if c.err != nil { |
| 738 | c.err <- err |
| 739 | } |
| 740 | } |
| 741 | break retry |
| 742 | } |
| 743 | } |
| 744 | |
| 745 | // trySolo is a special sentinel error value used for signaling that a |
| 746 | // transaction function should be re-run. It should never be seen by |
| 747 | // callers. |
| 748 | var trySolo = errors.New("batch function returned an error and should be re-run solo") |
| 749 | |
| 750 | type panicked struct { |
| 751 | reason interface{} |
| 752 | } |
| 753 | |
| 754 | func (p panicked) Error() string { |
| 755 | if err, ok := p.reason.(error); ok { |
| 756 | return err.Error() |
| 757 | } |
| 758 | return fmt.Sprintf("panic: %v", p.reason) |
| 759 | } |
| 760 | |
| 761 | func safelyCall(fn func(*Tx) error, tx *Tx) (err error) { |
| 762 | defer func() { |
| 763 | if p := recover(); p != nil { |
| 764 | err = panicked{p} |
| 765 | } |
| 766 | }() |
| 767 | return fn(tx) |
| 768 | } |
| 769 | |
| 770 | // Sync executes fdatasync() against the database file handle. |
| 771 | // |
| 772 | // This is not necessary under normal operation, however, if you use NoSync |
| 773 | // then it allows you to force the database file to sync against the disk. |
| 774 | func (db *DB) Sync() error { return fdatasync(db) } |
| 775 | |
| 776 | // Stats retrieves ongoing performance stats for the database. |
| 777 | // This is only updated when a transaction closes. |
| 778 | func (db *DB) Stats() Stats { |
| 779 | db.statlock.RLock() |
| 780 | defer db.statlock.RUnlock() |
| 781 | return db.stats |
| 782 | } |
| 783 | |
| 784 | // This is for internal access to the raw data bytes from the C cursor, use |
| 785 | // carefully, or not at all. |
| 786 | func (db *DB) Info() *Info { |
| 787 | return &Info{uintptr(unsafe.Pointer(&db.data[0])), db.pageSize} |
| 788 | } |
| 789 | |
| 790 | // page retrieves a page reference from the mmap based on the current page size. |
| 791 | func (db *DB) page(id pgid) *page { |
| 792 | pos := id * pgid(db.pageSize) |
| 793 | return (*page)(unsafe.Pointer(&db.data[pos])) |
| 794 | } |
| 795 | |
| 796 | // pageInBuffer retrieves a page reference from a given byte array based on the current page size. |
| 797 | func (db *DB) pageInBuffer(b []byte, id pgid) *page { |
| 798 | return (*page)(unsafe.Pointer(&b[id*pgid(db.pageSize)])) |
| 799 | } |
| 800 | |
| 801 | // meta retrieves the current meta page reference. |
| 802 | func (db *DB) meta() *meta { |
| 803 | // We have to return the meta with the highest txid which doesn't fail |
| 804 | // validation. Otherwise, we can cause errors when in fact the database is |
| 805 | // in a consistent state. metaA is the one with the higher txid. |
| 806 | metaA := db.meta0 |
| 807 | metaB := db.meta1 |
| 808 | if db.meta1.txid > db.meta0.txid { |
| 809 | metaA = db.meta1 |
| 810 | metaB = db.meta0 |
| 811 | } |
| 812 | |
| 813 | // Use higher meta page if valid. Otherwise fallback to previous, if valid. |
| 814 | if err := metaA.validate(); err == nil { |
| 815 | return metaA |
| 816 | } else if err := metaB.validate(); err == nil { |
| 817 | return metaB |
| 818 | } |
| 819 | |
| 820 | // This should never be reached, because both meta1 and meta0 were validated |
| 821 | // on mmap() and we do fsync() on every write. |
| 822 | panic("bolt.DB.meta(): invalid meta pages") |
| 823 | } |
| 824 | |
| 825 | // allocate returns a contiguous block of memory starting at a given page. |
| 826 | func (db *DB) allocate(count int) (*page, error) { |
| 827 | // Allocate a temporary buffer for the page. |
| 828 | var buf []byte |
| 829 | if count == 1 { |
| 830 | buf = db.pagePool.Get().([]byte) |
| 831 | } else { |
| 832 | buf = make([]byte, count*db.pageSize) |
| 833 | } |
| 834 | p := (*page)(unsafe.Pointer(&buf[0])) |
| 835 | p.overflow = uint32(count - 1) |
| 836 | |
| 837 | // Use pages from the freelist if they are available. |
| 838 | if p.id = db.freelist.allocate(count); p.id != 0 { |
| 839 | return p, nil |
| 840 | } |
| 841 | |
| 842 | // Resize mmap() if we're at the end. |
| 843 | p.id = db.rwtx.meta.pgid |
| 844 | var minsz = int((p.id+pgid(count))+1) * db.pageSize |
| 845 | if minsz >= db.datasz { |
| 846 | if err := db.mmap(minsz); err != nil { |
| 847 | return nil, fmt.Errorf("mmap allocate error: %s", err) |
| 848 | } |
| 849 | } |
| 850 | |
| 851 | // Move the page id high water mark. |
| 852 | db.rwtx.meta.pgid += pgid(count) |
| 853 | |
| 854 | return p, nil |
| 855 | } |
| 856 | |
| 857 | // grow grows the size of the database to the given sz. |
| 858 | func (db *DB) grow(sz int) error { |
| 859 | // Ignore if the new size is less than available file size. |
| 860 | if sz <= db.filesz { |
| 861 | return nil |
| 862 | } |
| 863 | |
| 864 | // If the data is smaller than the alloc size then only allocate what's needed. |
| 865 | // Once it goes over the allocation size then allocate in chunks. |
| 866 | if db.datasz < db.AllocSize { |
| 867 | sz = db.datasz |
| 868 | } else { |
| 869 | sz += db.AllocSize |
| 870 | } |
| 871 | |
| 872 | // Truncate and fsync to ensure file size metadata is flushed. |
| 873 | // https://github.com/boltdb/bolt/issues/284 |
| 874 | if !db.NoGrowSync && !db.readOnly { |
| 875 | if runtime.GOOS != "windows" { |
| 876 | if err := db.file.Truncate(int64(sz)); err != nil { |
| 877 | return fmt.Errorf("file resize error: %s", err) |
| 878 | } |
| 879 | } |
| 880 | if err := db.file.Sync(); err != nil { |
| 881 | return fmt.Errorf("file sync error: %s", err) |
| 882 | } |
| 883 | } |
| 884 | |
| 885 | db.filesz = sz |
| 886 | return nil |
| 887 | } |
| 888 | |
| 889 | func (db *DB) IsReadOnly() bool { |
| 890 | return db.readOnly |
| 891 | } |
| 892 | |
| 893 | // Options represents the options that can be set when opening a database. |
| 894 | type Options struct { |
| 895 | // Timeout is the amount of time to wait to obtain a file lock. |
| 896 | // When set to zero it will wait indefinitely. This option is only |
| 897 | // available on Darwin and Linux. |
| 898 | Timeout time.Duration |
| 899 | |
| 900 | // Sets the DB.NoGrowSync flag before memory mapping the file. |
| 901 | NoGrowSync bool |
| 902 | |
| 903 | // Open database in read-only mode. Uses flock(..., LOCK_SH |LOCK_NB) to |
| 904 | // grab a shared lock (UNIX). |
| 905 | ReadOnly bool |
| 906 | |
| 907 | // Sets the DB.MmapFlags flag before memory mapping the file. |
| 908 | MmapFlags int |
| 909 | |
| 910 | // InitialMmapSize is the initial mmap size of the database |
| 911 | // in bytes. Read transactions won't block write transaction |
| 912 | // if the InitialMmapSize is large enough to hold database mmap |
| 913 | // size. (See DB.Begin for more information) |
| 914 | // |
| 915 | // If <=0, the initial map size is 0. |
| 916 | // If initialMmapSize is smaller than the previous database size, |
| 917 | // it takes no effect. |
| 918 | InitialMmapSize int |
| 919 | } |
| 920 | |
| 921 | // DefaultOptions represent the options used if nil options are passed into Open(). |
| 922 | // No timeout is used which will cause Bolt to wait indefinitely for a lock. |
| 923 | var DefaultOptions = &Options{ |
| 924 | Timeout: 0, |
| 925 | NoGrowSync: false, |
| 926 | } |
| 927 | |
| 928 | // Stats represents statistics about the database. |
| 929 | type Stats struct { |
| 930 | // Freelist stats |
| 931 | FreePageN int // total number of free pages on the freelist |
| 932 | PendingPageN int // total number of pending pages on the freelist |
| 933 | FreeAlloc int // total bytes allocated in free pages |
| 934 | FreelistInuse int // total bytes used by the freelist |
| 935 | |
| 936 | // Transaction stats |
| 937 | TxN int // total number of started read transactions |
| 938 | OpenTxN int // number of currently open read transactions |
| 939 | |
| 940 | TxStats TxStats // global, ongoing stats. |
| 941 | } |
| 942 | |
| 943 | // Sub calculates and returns the difference between two sets of database stats. |
| 944 | // This is useful when obtaining stats at two different points and time and |
| 945 | // you need the performance counters that occurred within that time span. |
| 946 | func (s *Stats) Sub(other *Stats) Stats { |
| 947 | if other == nil { |
| 948 | return *s |
| 949 | } |
| 950 | var diff Stats |
| 951 | diff.FreePageN = s.FreePageN |
| 952 | diff.PendingPageN = s.PendingPageN |
| 953 | diff.FreeAlloc = s.FreeAlloc |
| 954 | diff.FreelistInuse = s.FreelistInuse |
| 955 | diff.TxN = other.TxN - s.TxN |
| 956 | diff.TxStats = s.TxStats.Sub(&other.TxStats) |
| 957 | return diff |
| 958 | } |
| 959 | |
| 960 | func (s *Stats) add(other *Stats) { |
| 961 | s.TxStats.add(&other.TxStats) |
| 962 | } |
| 963 | |
| 964 | type Info struct { |
| 965 | Data uintptr |
| 966 | PageSize int |
| 967 | } |
| 968 | |
| 969 | type meta struct { |
| 970 | magic uint32 |
| 971 | version uint32 |
| 972 | pageSize uint32 |
| 973 | flags uint32 |
| 974 | root bucket |
| 975 | freelist pgid |
| 976 | pgid pgid |
| 977 | txid txid |
| 978 | checksum uint64 |
| 979 | } |
| 980 | |
| 981 | // validate checks the marker bytes and version of the meta page to ensure it matches this binary. |
| 982 | func (m *meta) validate() error { |
| 983 | if m.magic != magic { |
| 984 | return ErrInvalid |
| 985 | } else if m.version != version { |
| 986 | return ErrVersionMismatch |
| 987 | } else if m.checksum != 0 && m.checksum != m.sum64() { |
| 988 | return ErrChecksum |
| 989 | } |
| 990 | return nil |
| 991 | } |
| 992 | |
| 993 | // copy copies one meta object to another. |
| 994 | func (m *meta) copy(dest *meta) { |
| 995 | *dest = *m |
| 996 | } |
| 997 | |
| 998 | // write writes the meta onto a page. |
| 999 | func (m *meta) write(p *page) { |
| 1000 | if m.root.root >= m.pgid { |
| 1001 | panic(fmt.Sprintf("root bucket pgid (%d) above high water mark (%d)", m.root.root, m.pgid)) |
| 1002 | } else if m.freelist >= m.pgid { |
| 1003 | panic(fmt.Sprintf("freelist pgid (%d) above high water mark (%d)", m.freelist, m.pgid)) |
| 1004 | } |
| 1005 | |
| 1006 | // Page id is either going to be 0 or 1 which we can determine by the transaction ID. |
| 1007 | p.id = pgid(m.txid % 2) |
| 1008 | p.flags |= metaPageFlag |
| 1009 | |
| 1010 | // Calculate the checksum. |
| 1011 | m.checksum = m.sum64() |
| 1012 | |
| 1013 | m.copy(p.meta()) |
| 1014 | } |
| 1015 | |
| 1016 | // generates the checksum for the meta. |
| 1017 | func (m *meta) sum64() uint64 { |
| 1018 | var h = fnv.New64a() |
| 1019 | _, _ = h.Write((*[unsafe.Offsetof(meta{}.checksum)]byte)(unsafe.Pointer(m))[:]) |
| 1020 | return h.Sum64() |
| 1021 | } |
| 1022 | |
| 1023 | // _assert will panic with a given formatted message if the given condition is false. |
| 1024 | func _assert(condition bool, msg string, v ...interface{}) { |
| 1025 | if !condition { |
| 1026 | panic(fmt.Sprintf("assertion failed: "+msg, v...)) |
| 1027 | } |
| 1028 | } |
| 1029 | |
| 1030 | func warn(v ...interface{}) { fmt.Fprintln(os.Stderr, v...) } |
| 1031 | func warnf(msg string, v ...interface{}) { fmt.Fprintf(os.Stderr, msg+"\n", v...) } |
| 1032 | |
| 1033 | func printstack() { |
| 1034 | stack := strings.Join(strings.Split(string(debug.Stack()), "\n")[2:], "\n") |
| 1035 | fmt.Fprintln(os.Stderr, stack) |
| 1036 | } |