khenaidoo | 106c61a | 2021-08-11 18:05:46 -0400 | [diff] [blame] | 1 | // Copyright 2019+ Klaus Post. All rights reserved. |
| 2 | // License information can be found in the LICENSE file. |
| 3 | // Based on work by Yann Collet, released under BSD License. |
| 4 | |
| 5 | package zstd |
| 6 | |
| 7 | import ( |
| 8 | "errors" |
| 9 | "io" |
| 10 | "sync" |
| 11 | ) |
| 12 | |
| 13 | // Decoder provides decoding of zstandard streams. |
| 14 | // The decoder has been designed to operate without allocations after a warmup. |
| 15 | // This means that you should store the decoder for best performance. |
| 16 | // To re-use a stream decoder, use the Reset(r io.Reader) error to switch to another stream. |
| 17 | // A decoder can safely be re-used even if the previous stream failed. |
| 18 | // To release the resources, you must call the Close() function on a decoder. |
| 19 | type Decoder struct { |
| 20 | o decoderOptions |
| 21 | |
| 22 | // Unreferenced decoders, ready for use. |
| 23 | decoders chan *blockDec |
| 24 | |
| 25 | // Streams ready to be decoded. |
| 26 | stream chan decodeStream |
| 27 | |
| 28 | // Current read position used for Reader functionality. |
| 29 | current decoderState |
| 30 | |
| 31 | // Custom dictionaries. |
| 32 | // Always uses copies. |
| 33 | dicts map[uint32]dict |
| 34 | |
| 35 | // streamWg is the waitgroup for all streams |
| 36 | streamWg sync.WaitGroup |
| 37 | } |
| 38 | |
| 39 | // decoderState is used for maintaining state when the decoder |
| 40 | // is used for streaming. |
| 41 | type decoderState struct { |
| 42 | // current block being written to stream. |
| 43 | decodeOutput |
| 44 | |
| 45 | // output in order to be written to stream. |
| 46 | output chan decodeOutput |
| 47 | |
| 48 | // cancel remaining output. |
| 49 | cancel chan struct{} |
| 50 | |
| 51 | flushed bool |
| 52 | } |
| 53 | |
| 54 | var ( |
| 55 | // Check the interfaces we want to support. |
| 56 | _ = io.WriterTo(&Decoder{}) |
| 57 | _ = io.Reader(&Decoder{}) |
| 58 | ) |
| 59 | |
| 60 | // NewReader creates a new decoder. |
| 61 | // A nil Reader can be provided in which case Reset can be used to start a decode. |
| 62 | // |
| 63 | // A Decoder can be used in two modes: |
| 64 | // |
| 65 | // 1) As a stream, or |
| 66 | // 2) For stateless decoding using DecodeAll. |
| 67 | // |
| 68 | // Only a single stream can be decoded concurrently, but the same decoder |
| 69 | // can run multiple concurrent stateless decodes. It is even possible to |
| 70 | // use stateless decodes while a stream is being decoded. |
| 71 | // |
| 72 | // The Reset function can be used to initiate a new stream, which is will considerably |
| 73 | // reduce the allocations normally caused by NewReader. |
| 74 | func NewReader(r io.Reader, opts ...DOption) (*Decoder, error) { |
| 75 | initPredefined() |
| 76 | var d Decoder |
| 77 | d.o.setDefault() |
| 78 | for _, o := range opts { |
| 79 | err := o(&d.o) |
| 80 | if err != nil { |
| 81 | return nil, err |
| 82 | } |
| 83 | } |
| 84 | d.current.output = make(chan decodeOutput, d.o.concurrent) |
| 85 | d.current.flushed = true |
| 86 | |
| 87 | if r == nil { |
| 88 | d.current.err = ErrDecoderNilInput |
| 89 | } |
| 90 | |
| 91 | // Transfer option dicts. |
| 92 | d.dicts = make(map[uint32]dict, len(d.o.dicts)) |
| 93 | for _, dc := range d.o.dicts { |
| 94 | d.dicts[dc.id] = dc |
| 95 | } |
| 96 | d.o.dicts = nil |
| 97 | |
| 98 | // Create decoders |
| 99 | d.decoders = make(chan *blockDec, d.o.concurrent) |
| 100 | for i := 0; i < d.o.concurrent; i++ { |
| 101 | dec := newBlockDec(d.o.lowMem) |
| 102 | dec.localFrame = newFrameDec(d.o) |
| 103 | d.decoders <- dec |
| 104 | } |
| 105 | |
| 106 | if r == nil { |
| 107 | return &d, nil |
| 108 | } |
| 109 | return &d, d.Reset(r) |
| 110 | } |
| 111 | |
| 112 | // Read bytes from the decompressed stream into p. |
| 113 | // Returns the number of bytes written and any error that occurred. |
| 114 | // When the stream is done, io.EOF will be returned. |
| 115 | func (d *Decoder) Read(p []byte) (int, error) { |
| 116 | if d.stream == nil { |
| 117 | return 0, ErrDecoderNilInput |
| 118 | } |
| 119 | var n int |
| 120 | for { |
| 121 | if len(d.current.b) > 0 { |
| 122 | filled := copy(p, d.current.b) |
| 123 | p = p[filled:] |
| 124 | d.current.b = d.current.b[filled:] |
| 125 | n += filled |
| 126 | } |
| 127 | if len(p) == 0 { |
| 128 | break |
| 129 | } |
| 130 | if len(d.current.b) == 0 { |
| 131 | // We have an error and no more data |
| 132 | if d.current.err != nil { |
| 133 | break |
| 134 | } |
| 135 | if !d.nextBlock(n == 0) { |
| 136 | return n, nil |
| 137 | } |
| 138 | } |
| 139 | } |
| 140 | if len(d.current.b) > 0 { |
| 141 | if debug { |
| 142 | println("returning", n, "still bytes left:", len(d.current.b)) |
| 143 | } |
| 144 | // Only return error at end of block |
| 145 | return n, nil |
| 146 | } |
| 147 | if d.current.err != nil { |
| 148 | d.drainOutput() |
| 149 | } |
| 150 | if debug { |
| 151 | println("returning", n, d.current.err, len(d.decoders)) |
| 152 | } |
| 153 | return n, d.current.err |
| 154 | } |
| 155 | |
| 156 | // Reset will reset the decoder the supplied stream after the current has finished processing. |
| 157 | // Note that this functionality cannot be used after Close has been called. |
| 158 | // Reset can be called with a nil reader to release references to the previous reader. |
| 159 | // After being called with a nil reader, no other operations than Reset or DecodeAll or Close |
| 160 | // should be used. |
| 161 | func (d *Decoder) Reset(r io.Reader) error { |
| 162 | if d.current.err == ErrDecoderClosed { |
| 163 | return d.current.err |
| 164 | } |
| 165 | |
| 166 | d.drainOutput() |
| 167 | |
| 168 | if r == nil { |
| 169 | d.current.err = ErrDecoderNilInput |
| 170 | d.current.flushed = true |
| 171 | return nil |
| 172 | } |
| 173 | |
| 174 | if d.stream == nil { |
| 175 | d.stream = make(chan decodeStream, 1) |
| 176 | d.streamWg.Add(1) |
| 177 | go d.startStreamDecoder(d.stream) |
| 178 | } |
| 179 | |
| 180 | // If bytes buffer and < 1MB, do sync decoding anyway. |
| 181 | if bb, ok := r.(byter); ok && bb.Len() < 1<<20 { |
| 182 | bb2 := bb |
| 183 | if debug { |
| 184 | println("*bytes.Buffer detected, doing sync decode, len:", bb.Len()) |
| 185 | } |
| 186 | b := bb2.Bytes() |
| 187 | var dst []byte |
| 188 | if cap(d.current.b) > 0 { |
| 189 | dst = d.current.b |
| 190 | } |
| 191 | |
| 192 | dst, err := d.DecodeAll(b, dst[:0]) |
| 193 | if err == nil { |
| 194 | err = io.EOF |
| 195 | } |
| 196 | d.current.b = dst |
| 197 | d.current.err = err |
| 198 | d.current.flushed = true |
| 199 | if debug { |
| 200 | println("sync decode to", len(dst), "bytes, err:", err) |
| 201 | } |
| 202 | return nil |
| 203 | } |
| 204 | |
| 205 | // Remove current block. |
| 206 | d.current.decodeOutput = decodeOutput{} |
| 207 | d.current.err = nil |
| 208 | d.current.cancel = make(chan struct{}) |
| 209 | d.current.flushed = false |
| 210 | d.current.d = nil |
| 211 | |
| 212 | d.stream <- decodeStream{ |
| 213 | r: r, |
| 214 | output: d.current.output, |
| 215 | cancel: d.current.cancel, |
| 216 | } |
| 217 | return nil |
| 218 | } |
| 219 | |
| 220 | // drainOutput will drain the output until errEndOfStream is sent. |
| 221 | func (d *Decoder) drainOutput() { |
| 222 | if d.current.cancel != nil { |
| 223 | println("cancelling current") |
| 224 | close(d.current.cancel) |
| 225 | d.current.cancel = nil |
| 226 | } |
| 227 | if d.current.d != nil { |
| 228 | if debug { |
| 229 | printf("re-adding current decoder %p, decoders: %d", d.current.d, len(d.decoders)) |
| 230 | } |
| 231 | d.decoders <- d.current.d |
| 232 | d.current.d = nil |
| 233 | d.current.b = nil |
| 234 | } |
| 235 | if d.current.output == nil || d.current.flushed { |
| 236 | println("current already flushed") |
| 237 | return |
| 238 | } |
| 239 | for v := range d.current.output { |
| 240 | if v.d != nil { |
| 241 | if debug { |
| 242 | printf("re-adding decoder %p", v.d) |
| 243 | } |
| 244 | d.decoders <- v.d |
| 245 | } |
| 246 | if v.err == errEndOfStream { |
| 247 | println("current flushed") |
| 248 | d.current.flushed = true |
| 249 | return |
| 250 | } |
| 251 | } |
| 252 | } |
| 253 | |
| 254 | // WriteTo writes data to w until there's no more data to write or when an error occurs. |
| 255 | // The return value n is the number of bytes written. |
| 256 | // Any error encountered during the write is also returned. |
| 257 | func (d *Decoder) WriteTo(w io.Writer) (int64, error) { |
| 258 | if d.stream == nil { |
| 259 | return 0, ErrDecoderNilInput |
| 260 | } |
| 261 | var n int64 |
| 262 | for { |
| 263 | if len(d.current.b) > 0 { |
| 264 | n2, err2 := w.Write(d.current.b) |
| 265 | n += int64(n2) |
| 266 | if err2 != nil && d.current.err == nil { |
| 267 | d.current.err = err2 |
| 268 | break |
| 269 | } |
| 270 | } |
| 271 | if d.current.err != nil { |
| 272 | break |
| 273 | } |
| 274 | d.nextBlock(true) |
| 275 | } |
| 276 | err := d.current.err |
| 277 | if err != nil { |
| 278 | d.drainOutput() |
| 279 | } |
| 280 | if err == io.EOF { |
| 281 | err = nil |
| 282 | } |
| 283 | return n, err |
| 284 | } |
| 285 | |
| 286 | // DecodeAll allows stateless decoding of a blob of bytes. |
| 287 | // Output will be appended to dst, so if the destination size is known |
| 288 | // you can pre-allocate the destination slice to avoid allocations. |
| 289 | // DecodeAll can be used concurrently. |
| 290 | // The Decoder concurrency limits will be respected. |
| 291 | func (d *Decoder) DecodeAll(input, dst []byte) ([]byte, error) { |
| 292 | if d.current.err == ErrDecoderClosed { |
| 293 | return dst, ErrDecoderClosed |
| 294 | } |
| 295 | |
| 296 | // Grab a block decoder and frame decoder. |
| 297 | block := <-d.decoders |
| 298 | frame := block.localFrame |
| 299 | defer func() { |
| 300 | if debug { |
| 301 | printf("re-adding decoder: %p", block) |
| 302 | } |
| 303 | frame.rawInput = nil |
| 304 | frame.bBuf = nil |
| 305 | d.decoders <- block |
| 306 | }() |
| 307 | frame.bBuf = input |
| 308 | |
| 309 | for { |
| 310 | frame.history.reset() |
| 311 | err := frame.reset(&frame.bBuf) |
| 312 | if err == io.EOF { |
| 313 | if debug { |
| 314 | println("frame reset return EOF") |
| 315 | } |
| 316 | return dst, nil |
| 317 | } |
| 318 | if frame.DictionaryID != nil { |
| 319 | dict, ok := d.dicts[*frame.DictionaryID] |
| 320 | if !ok { |
| 321 | return nil, ErrUnknownDictionary |
| 322 | } |
| 323 | frame.history.setDict(&dict) |
| 324 | } |
| 325 | if err != nil { |
| 326 | return dst, err |
| 327 | } |
| 328 | if frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) { |
| 329 | return dst, ErrDecoderSizeExceeded |
| 330 | } |
| 331 | if frame.FrameContentSize > 0 && frame.FrameContentSize < 1<<30 { |
| 332 | // Never preallocate moe than 1 GB up front. |
| 333 | if cap(dst)-len(dst) < int(frame.FrameContentSize) { |
| 334 | dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize)) |
| 335 | copy(dst2, dst) |
| 336 | dst = dst2 |
| 337 | } |
| 338 | } |
| 339 | if cap(dst) == 0 { |
| 340 | // Allocate len(input) * 2 by default if nothing is provided |
| 341 | // and we didn't get frame content size. |
| 342 | size := len(input) * 2 |
| 343 | // Cap to 1 MB. |
| 344 | if size > 1<<20 { |
| 345 | size = 1 << 20 |
| 346 | } |
| 347 | if uint64(size) > d.o.maxDecodedSize { |
| 348 | size = int(d.o.maxDecodedSize) |
| 349 | } |
| 350 | dst = make([]byte, 0, size) |
| 351 | } |
| 352 | |
| 353 | dst, err = frame.runDecoder(dst, block) |
| 354 | if err != nil { |
| 355 | return dst, err |
| 356 | } |
| 357 | if len(frame.bBuf) == 0 { |
| 358 | if debug { |
| 359 | println("frame dbuf empty") |
| 360 | } |
| 361 | break |
| 362 | } |
| 363 | } |
| 364 | return dst, nil |
| 365 | } |
| 366 | |
| 367 | // nextBlock returns the next block. |
| 368 | // If an error occurs d.err will be set. |
| 369 | // Optionally the function can block for new output. |
| 370 | // If non-blocking mode is used the returned boolean will be false |
| 371 | // if no data was available without blocking. |
| 372 | func (d *Decoder) nextBlock(blocking bool) (ok bool) { |
| 373 | if d.current.d != nil { |
| 374 | if debug { |
| 375 | printf("re-adding current decoder %p", d.current.d) |
| 376 | } |
| 377 | d.decoders <- d.current.d |
| 378 | d.current.d = nil |
| 379 | } |
| 380 | if d.current.err != nil { |
| 381 | // Keep error state. |
| 382 | return blocking |
| 383 | } |
| 384 | |
| 385 | if blocking { |
| 386 | d.current.decodeOutput = <-d.current.output |
| 387 | } else { |
| 388 | select { |
| 389 | case d.current.decodeOutput = <-d.current.output: |
| 390 | default: |
| 391 | return false |
| 392 | } |
| 393 | } |
| 394 | if debug { |
| 395 | println("got", len(d.current.b), "bytes, error:", d.current.err) |
| 396 | } |
| 397 | return true |
| 398 | } |
| 399 | |
| 400 | // Close will release all resources. |
| 401 | // It is NOT possible to reuse the decoder after this. |
| 402 | func (d *Decoder) Close() { |
| 403 | if d.current.err == ErrDecoderClosed { |
| 404 | return |
| 405 | } |
| 406 | d.drainOutput() |
| 407 | if d.stream != nil { |
| 408 | close(d.stream) |
| 409 | d.streamWg.Wait() |
| 410 | d.stream = nil |
| 411 | } |
| 412 | if d.decoders != nil { |
| 413 | close(d.decoders) |
| 414 | for dec := range d.decoders { |
| 415 | dec.Close() |
| 416 | } |
| 417 | d.decoders = nil |
| 418 | } |
| 419 | if d.current.d != nil { |
| 420 | d.current.d.Close() |
| 421 | d.current.d = nil |
| 422 | } |
| 423 | d.current.err = ErrDecoderClosed |
| 424 | } |
| 425 | |
| 426 | // IOReadCloser returns the decoder as an io.ReadCloser for convenience. |
| 427 | // Any changes to the decoder will be reflected, so the returned ReadCloser |
| 428 | // can be reused along with the decoder. |
| 429 | // io.WriterTo is also supported by the returned ReadCloser. |
| 430 | func (d *Decoder) IOReadCloser() io.ReadCloser { |
| 431 | return closeWrapper{d: d} |
| 432 | } |
| 433 | |
| 434 | // closeWrapper wraps a function call as a closer. |
| 435 | type closeWrapper struct { |
| 436 | d *Decoder |
| 437 | } |
| 438 | |
| 439 | // WriteTo forwards WriteTo calls to the decoder. |
| 440 | func (c closeWrapper) WriteTo(w io.Writer) (n int64, err error) { |
| 441 | return c.d.WriteTo(w) |
| 442 | } |
| 443 | |
| 444 | // Read forwards read calls to the decoder. |
| 445 | func (c closeWrapper) Read(p []byte) (n int, err error) { |
| 446 | return c.d.Read(p) |
| 447 | } |
| 448 | |
| 449 | // Close closes the decoder. |
| 450 | func (c closeWrapper) Close() error { |
| 451 | c.d.Close() |
| 452 | return nil |
| 453 | } |
| 454 | |
| 455 | type decodeOutput struct { |
| 456 | d *blockDec |
| 457 | b []byte |
| 458 | err error |
| 459 | } |
| 460 | |
| 461 | type decodeStream struct { |
| 462 | r io.Reader |
| 463 | |
| 464 | // Blocks ready to be written to output. |
| 465 | output chan decodeOutput |
| 466 | |
| 467 | // cancel reading from the input |
| 468 | cancel chan struct{} |
| 469 | } |
| 470 | |
| 471 | // errEndOfStream indicates that everything from the stream was read. |
| 472 | var errEndOfStream = errors.New("end-of-stream") |
| 473 | |
| 474 | // Create Decoder: |
| 475 | // Spawn n block decoders. These accept tasks to decode a block. |
| 476 | // Create goroutine that handles stream processing, this will send history to decoders as they are available. |
| 477 | // Decoders update the history as they decode. |
| 478 | // When a block is returned: |
| 479 | // a) history is sent to the next decoder, |
| 480 | // b) content written to CRC. |
| 481 | // c) return data to WRITER. |
| 482 | // d) wait for next block to return data. |
| 483 | // Once WRITTEN, the decoders reused by the writer frame decoder for re-use. |
| 484 | func (d *Decoder) startStreamDecoder(inStream chan decodeStream) { |
| 485 | defer d.streamWg.Done() |
| 486 | frame := newFrameDec(d.o) |
| 487 | for stream := range inStream { |
| 488 | if debug { |
| 489 | println("got new stream") |
| 490 | } |
| 491 | br := readerWrapper{r: stream.r} |
| 492 | decodeStream: |
| 493 | for { |
| 494 | frame.history.reset() |
| 495 | err := frame.reset(&br) |
| 496 | if debug && err != nil { |
| 497 | println("Frame decoder returned", err) |
| 498 | } |
| 499 | if err == nil && frame.DictionaryID != nil { |
| 500 | dict, ok := d.dicts[*frame.DictionaryID] |
| 501 | if !ok { |
| 502 | err = ErrUnknownDictionary |
| 503 | } else { |
| 504 | frame.history.setDict(&dict) |
| 505 | } |
| 506 | } |
| 507 | if err != nil { |
| 508 | stream.output <- decodeOutput{ |
| 509 | err: err, |
| 510 | } |
| 511 | break |
| 512 | } |
| 513 | if debug { |
| 514 | println("starting frame decoder") |
| 515 | } |
| 516 | |
| 517 | // This goroutine will forward history between frames. |
| 518 | frame.frameDone.Add(1) |
| 519 | frame.initAsync() |
| 520 | |
| 521 | go frame.startDecoder(stream.output) |
| 522 | decodeFrame: |
| 523 | // Go through all blocks of the frame. |
| 524 | for { |
| 525 | dec := <-d.decoders |
| 526 | select { |
| 527 | case <-stream.cancel: |
| 528 | if !frame.sendErr(dec, io.EOF) { |
| 529 | // To not let the decoder dangle, send it back. |
| 530 | stream.output <- decodeOutput{d: dec} |
| 531 | } |
| 532 | break decodeStream |
| 533 | default: |
| 534 | } |
| 535 | err := frame.next(dec) |
| 536 | switch err { |
| 537 | case io.EOF: |
| 538 | // End of current frame, no error |
| 539 | println("EOF on next block") |
| 540 | break decodeFrame |
| 541 | case nil: |
| 542 | continue |
| 543 | default: |
| 544 | println("block decoder returned", err) |
| 545 | break decodeStream |
| 546 | } |
| 547 | } |
| 548 | // All blocks have started decoding, check if there are more frames. |
| 549 | println("waiting for done") |
| 550 | frame.frameDone.Wait() |
| 551 | println("done waiting...") |
| 552 | } |
| 553 | frame.frameDone.Wait() |
| 554 | println("Sending EOS") |
| 555 | stream.output <- decodeOutput{err: errEndOfStream} |
| 556 | } |
| 557 | } |