1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773
|
// Copyright 2018 The gVisor Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// Package compressio provides parallel compression and decompression, as well
// as optional SHA-256 hashing. It also provides another storage variant
// (nocompressio) that does not compress data but tracks its integrity.
//
// The stream format is defined as follows.
//
// /------------------------------------------------------\
// | chunk size (4-bytes) |
// +------------------------------------------------------+
// | (optional) hash (32-bytes) |
// +------------------------------------------------------+
// | compressed data size (4-bytes) |
// +------------------------------------------------------+
// | compressed data |
// +------------------------------------------------------+
// | (optional) hash (32-bytes) |
// +------------------------------------------------------+
// | compressed data size (4-bytes) |
// +------------------------------------------------------+
// | ...... |
// \------------------------------------------------------/
//
// where each subsequent hash is calculated from the following items in order
//
// compressed data
// compressed data size
// previous hash
//
// so the stream integrity cannot be compromised by switching and mixing
// compressed chunks.
package compressio
import (
"bytes"
"compress/flate"
"crypto/hmac"
"crypto/sha256"
"encoding/binary"
"errors"
"hash"
"io"
"runtime"
"gvisor.dev/gvisor/pkg/sync"
)
var bufPool = sync.Pool{
New: func() any {
return bytes.NewBuffer(nil)
},
}
var chunkPool = sync.Pool{
New: func() any {
return new(chunk)
},
}
// chunk is a unit of work.
type chunk struct {
// compressed is compressed data.
//
// This will always be returned to the bufPool directly when work has
// finished (in schedule) and therefore must be allocated.
compressed *bytes.Buffer
// uncompressed is the uncompressed data.
//
// This is not returned to the bufPool automatically, since it may
// correspond to a inline slice (provided directly to Read or Write).
uncompressed *bytes.Buffer
// The current hash object. Only used in compress mode.
h hash.Hash
// The hash from previous chunks. Only used in uncompress mode.
lastSum []byte
// The expected hash after current chunk. Only used in uncompress mode.
sum []byte
}
// newChunk allocates a new chunk object (or pulls one from the pool). Buffers
// will be allocated if nil is provided for compressed or uncompressed.
func newChunk(lastSum []byte, sum []byte, compressed *bytes.Buffer, uncompressed *bytes.Buffer) *chunk {
c := chunkPool.Get().(*chunk)
c.lastSum = lastSum
c.sum = sum
if compressed != nil {
c.compressed = compressed
} else {
c.compressed = bufPool.Get().(*bytes.Buffer)
}
if uncompressed != nil {
c.uncompressed = uncompressed
} else {
c.uncompressed = bufPool.Get().(*bytes.Buffer)
}
return c
}
// result is the result of some work; it includes the original chunk.
type result struct {
*chunk
err error
}
// worker is a compression/decompression worker.
//
// The associated worker goroutine reads in uncompressed buffers from input and
// writes compressed buffers to its output. Alternatively, the worker reads
// compressed buffers from input and writes uncompressed buffers to its output.
//
// The goroutine will exit when input is closed, and the goroutine will close
// output.
type worker struct {
hashPool *hashPool
input chan *chunk
output chan result
// scratch is a temporary buffer used for marshalling. This is declared
// unfront here to avoid reallocation.
scratch [4]byte
}
// work is the main work routine; see worker.
func (w *worker) work(compress bool, level int) {
defer close(w.output)
var h hash.Hash
for c := range w.input {
if h == nil && w.hashPool != nil {
h = w.hashPool.getHash()
}
if compress {
mw := io.Writer(c.compressed)
if h != nil {
mw = io.MultiWriter(mw, h)
}
// Encode this slice.
fw, err := flate.NewWriter(mw, level)
if err != nil {
w.output <- result{c, err}
continue
}
// Encode the input.
if _, err := io.CopyN(fw, c.uncompressed, int64(c.uncompressed.Len())); err != nil {
w.output <- result{c, err}
continue
}
if err := fw.Close(); err != nil {
w.output <- result{c, err}
continue
}
// Write the hash, if enabled.
if h != nil {
binary.BigEndian.PutUint32(w.scratch[:], uint32(c.compressed.Len()))
h.Write(w.scratch[:4])
c.h = h
h = nil
}
} else {
// Check the hash of the compressed contents.
if h != nil {
h.Write(c.compressed.Bytes())
binary.BigEndian.PutUint32(w.scratch[:], uint32(c.compressed.Len()))
h.Write(w.scratch[:4])
io.CopyN(h, bytes.NewReader(c.lastSum), int64(len(c.lastSum)))
sum := h.Sum(nil)
h.Reset()
if !hmac.Equal(c.sum, sum) {
w.output <- result{c, ErrHashMismatch}
continue
}
}
// Decode this slice.
fr := flate.NewReader(c.compressed)
// Decode the input.
if _, err := io.Copy(c.uncompressed, fr); err != nil {
w.output <- result{c, err}
continue
}
}
// Send the output.
w.output <- result{c, nil}
}
}
type hashPool struct {
// mu protects the hash list.
mu sync.Mutex
// key is the key used to create hash objects.
key []byte
// hashes is the hash object free list. Note that this cannot be
// globally shared across readers or writers, as it is key-specific.
hashes []hash.Hash
}
// getHash gets a hash object for the pool. It should only be called when the
// pool key is non-nil.
func (p *hashPool) getHash() hash.Hash {
p.mu.Lock()
defer p.mu.Unlock()
if len(p.hashes) == 0 {
return hmac.New(sha256.New, p.key)
}
h := p.hashes[len(p.hashes)-1]
p.hashes = p.hashes[:len(p.hashes)-1]
return h
}
func (p *hashPool) putHash(h hash.Hash) {
h.Reset()
p.mu.Lock()
defer p.mu.Unlock()
p.hashes = append(p.hashes, h)
}
// pool is common functionality for reader/writers.
type pool struct {
// workers are the compression/decompression workers.
workers []worker
// chunkSize is the chunk size. This is the first four bytes in the
// stream and is shared across both the reader and writer.
chunkSize uint32
// mu protects below; it is generally the responsibility of users to
// acquire this mutex before calling any methods on the pool.
mu sync.Mutex
// nextInput is the next worker for input (scheduling).
nextInput int
// nextOutput is the next worker for output (result).
nextOutput int
// buf is the current active buffer; the exact semantics of this buffer
// depending on whether this is a reader or a writer.
buf *bytes.Buffer
// lasSum records the hash of the last chunk processed.
lastSum []byte
// hashPool is the hash object pool. It cannot be embedded into pool
// itself as worker refers to it and that would stop pool from being
// GCed.
hashPool *hashPool
}
// init initializes the worker pool.
//
// This should only be called once.
func (p *pool) init(key []byte, workers int, compress bool, level int) {
if key != nil {
p.hashPool = &hashPool{key: key}
}
p.workers = make([]worker, workers)
for i := 0; i < len(p.workers); i++ {
p.workers[i] = worker{
hashPool: p.hashPool,
input: make(chan *chunk, 1),
output: make(chan result, 1),
}
go p.workers[i].work(compress, level) // S/R-SAFE: In save path only.
}
runtime.SetFinalizer(p, (*pool).stop)
}
// stop stops all workers.
func (p *pool) stop() {
for i := 0; i < len(p.workers); i++ {
close(p.workers[i].input)
}
p.workers = nil
p.hashPool = nil
}
// handleResult calls the callback.
func handleResult(r result, callback func(*chunk) error) error {
defer func() {
r.chunk.compressed.Reset()
bufPool.Put(r.chunk.compressed)
chunkPool.Put(r.chunk)
}()
if r.err != nil {
return r.err
}
return callback(r.chunk)
}
// schedule schedules the given buffers.
//
// If c is non-nil, then it will return as soon as the chunk is scheduled. If c
// is nil, then it will return only when no more work is left to do.
//
// If no callback function is provided, then the output channel will be
// ignored. You must be sure that the input is schedulable in this case.
func (p *pool) schedule(c *chunk, callback func(*chunk) error) error {
for {
var (
inputChan chan *chunk
outputChan chan result
)
if c != nil && len(p.workers) != 0 {
inputChan = p.workers[(p.nextInput+1)%len(p.workers)].input
}
if callback != nil && p.nextOutput != p.nextInput && len(p.workers) != 0 {
outputChan = p.workers[(p.nextOutput+1)%len(p.workers)].output
}
if inputChan == nil && outputChan == nil {
return nil
}
select {
case inputChan <- c:
p.nextInput++
return nil
case r := <-outputChan:
p.nextOutput++
if err := handleResult(r, callback); err != nil {
return err
}
}
}
}
// Reader is a compressed reader.
type Reader struct {
pool
// in is the source.
in io.Reader
// scratch is a temporary buffer used for marshalling. This is declared
// unfront here to avoid reallocation.
scratch [4]byte
}
var _ io.Reader = (*Reader)(nil)
// NewReader returns a new compressed reader. If key is non-nil, the data stream
// is assumed to contain expected hash values, which will be compared against
// hash values computed from the compressed bytes. See package comments for
// details.
func NewReader(in io.Reader, key []byte) (*Reader, error) {
r := &Reader{
in: in,
}
// Use double buffering for read.
r.init(key, 2*runtime.GOMAXPROCS(0), false, 0)
if _, err := io.ReadFull(in, r.scratch[:4]); err != nil {
return nil, err
}
r.chunkSize = binary.BigEndian.Uint32(r.scratch[:4])
if r.hashPool != nil {
h := r.hashPool.getHash()
binary.BigEndian.PutUint32(r.scratch[:], r.chunkSize)
h.Write(r.scratch[:4])
r.lastSum = h.Sum(nil)
r.hashPool.putHash(h)
sum := make([]byte, len(r.lastSum))
if _, err := io.ReadFull(r.in, sum); err != nil {
return nil, err
}
if !hmac.Equal(r.lastSum, sum) {
return nil, ErrHashMismatch
}
}
return r, nil
}
// errNewBuffer is returned when a new buffer is completed.
var errNewBuffer = errors.New("buffer ready")
// ErrHashMismatch is returned if the hash does not match.
var ErrHashMismatch = errors.New("hash mismatch")
// Read implements io.Reader.Read.
func (r *Reader) Read(p []byte) (int, error) {
r.mu.Lock()
defer r.mu.Unlock()
// Total bytes completed; this is declared up front because it must be
// adjustable by the callback below.
done := 0
// Total bytes pending in the asynchronous workers for buffers. This is
// used to process the proper regions of the input as inline buffers.
var (
pendingPre = r.nextInput - r.nextOutput
pendingInline = 0
)
// Define our callback for completed work.
callback := func(c *chunk) error {
// Check for an inline buffer.
if pendingPre == 0 && pendingInline > 0 {
pendingInline--
done += c.uncompressed.Len()
return nil
}
// Copy the resulting buffer to our intermediate one, and
// return errNewBuffer to ensure that we aren't called a second
// time. This error code is handled specially below.
//
// c.buf will be freed and return to the pool when it is done.
if pendingPre > 0 {
pendingPre--
}
r.buf = c.uncompressed
return errNewBuffer
}
for done < len(p) {
// Do we have buffered data available?
if r.buf != nil {
n, err := r.buf.Read(p[done:])
done += n
if err == io.EOF {
// This is the uncompressed buffer, it can be
// returned to the pool at this point.
r.buf.Reset()
bufPool.Put(r.buf)
r.buf = nil
} else if err != nil {
// Should never happen.
defer r.stop()
return done, err
}
continue
}
// Read the length of the next chunk and reset the
// reader. The length is used to limit the reader.
//
// See writer.flush.
if _, err := io.ReadFull(r.in, r.scratch[:4]); err != nil {
// This is generally okay as long as there
// are still buffers outstanding. We actually
// just wait for completion of those buffers here
// and continue our loop.
if err := r.schedule(nil, callback); err == nil {
// We've actually finished all buffers; this is
// the normal EOF exit path.
defer r.stop()
return done, io.EOF
} else if err == errNewBuffer {
// A new buffer is now available.
continue
} else {
// Some other error occurred; we cannot
// process any further.
defer r.stop()
return done, err
}
}
l := binary.BigEndian.Uint32(r.scratch[:4])
// Read this chunk and schedule decompression.
compressed := bufPool.Get().(*bytes.Buffer)
if _, err := io.CopyN(compressed, r.in, int64(l)); err != nil {
// Some other error occurred; see above.
if err == io.EOF {
err = io.ErrUnexpectedEOF
}
return done, err
}
var sum []byte
if r.hashPool != nil {
sum = make([]byte, len(r.lastSum))
if _, err := io.ReadFull(r.in, sum); err != nil {
if err == io.EOF {
err = io.ErrUnexpectedEOF
}
return done, err
}
}
// Are we doing inline decoding?
//
// Note that we need to check the length here against
// bytes.MinRead, since the bytes library will choose to grow
// the slice if the available capacity is not at least
// bytes.MinRead. This limits inline decoding to chunkSizes
// that are at least bytes.MinRead (which is not unreasonable).
var c *chunk
start := done + ((pendingPre + pendingInline) * int(r.chunkSize))
if len(p) >= start+int(r.chunkSize) && len(p) >= start+bytes.MinRead {
c = newChunk(r.lastSum, sum, compressed, bytes.NewBuffer(p[start:start]))
pendingInline++
} else {
c = newChunk(r.lastSum, sum, compressed, nil)
}
r.lastSum = sum
if err := r.schedule(c, callback); err == errNewBuffer {
// A new buffer was completed while we were reading.
// That's great, but we need to force schedule the
// current buffer so that it does not get lost.
//
// It is safe to pass nil as an output function here,
// because we know that we just freed up a slot above.
r.schedule(c, nil)
} else if err != nil {
// Some other error occurred; see above.
defer r.stop()
return done, err
}
}
// Make sure that everything has been decoded successfully, otherwise
// parts of p may not actually have completed.
for pendingInline > 0 {
if err := r.schedule(nil, func(c *chunk) error {
if err := callback(c); err != nil {
return err
}
// The nil case means that an inline buffer has
// completed. The callback will have already removed
// the inline buffer from the map, so we just return an
// error to check the top of the loop again.
return errNewBuffer
}); err != errNewBuffer {
// Some other error occurred; see above.
return done, err
}
}
// Need to return done here, since it may have been adjusted by the
// callback to compensation for partial reads on some inline buffer.
return done, nil
}
// Writer is a compressed writer.
type Writer struct {
pool
// out is the underlying writer.
out io.Writer
// closed indicates whether the file has been closed.
closed bool
// scratch is a temporary buffer used for marshalling. This is declared
// unfront here to avoid reallocation.
scratch [4]byte
}
var _ io.Writer = (*Writer)(nil)
// NewWriter returns a new compressed writer. If key is non-nil, hash values are
// generated and written out for compressed bytes. See package comments for
// details.
//
// The recommended chunkSize is on the order of 1M. Extra memory may be
// buffered (in the form of read-ahead, or buffered writes), and is limited to
// O(chunkSize * [1+GOMAXPROCS]).
func NewWriter(out io.Writer, key []byte, chunkSize uint32, level int) (*Writer, error) {
w := &Writer{
pool: pool{
chunkSize: chunkSize,
buf: bufPool.Get().(*bytes.Buffer),
},
out: out,
}
w.init(key, 1+runtime.GOMAXPROCS(0), true, level)
binary.BigEndian.PutUint32(w.scratch[:], chunkSize)
if _, err := w.out.Write(w.scratch[:4]); err != nil {
return nil, err
}
if w.hashPool != nil {
h := w.hashPool.getHash()
binary.BigEndian.PutUint32(w.scratch[:], chunkSize)
h.Write(w.scratch[:4])
w.lastSum = h.Sum(nil)
w.hashPool.putHash(h)
if _, err := io.CopyN(w.out, bytes.NewReader(w.lastSum), int64(len(w.lastSum))); err != nil {
return nil, err
}
}
return w, nil
}
// flush writes a single buffer.
func (w *Writer) flush(c *chunk) error {
// Prefix each chunk with a length; this allows the reader to safely
// limit reads while buffering.
l := uint32(c.compressed.Len())
binary.BigEndian.PutUint32(w.scratch[:], l)
if _, err := w.out.Write(w.scratch[:4]); err != nil {
return err
}
// Write out to the stream.
if _, err := io.CopyN(w.out, c.compressed, int64(c.compressed.Len())); err != nil {
return err
}
if w.hashPool != nil {
io.CopyN(c.h, bytes.NewReader(w.lastSum), int64(len(w.lastSum)))
sum := c.h.Sum(nil)
w.hashPool.putHash(c.h)
c.h = nil
if _, err := io.CopyN(w.out, bytes.NewReader(sum), int64(len(sum))); err != nil {
return err
}
w.lastSum = sum
}
return nil
}
// Write implements io.Writer.Write.
func (w *Writer) Write(p []byte) (int, error) {
w.mu.Lock()
defer w.mu.Unlock()
// Did we close already?
if w.closed {
return 0, io.ErrUnexpectedEOF
}
// See above; we need to track in the same way.
var (
pendingPre = w.nextInput - w.nextOutput
pendingInline = 0
)
callback := func(c *chunk) error {
if pendingPre > 0 {
pendingPre--
err := w.flush(c)
c.uncompressed.Reset()
bufPool.Put(c.uncompressed)
return err
}
if pendingInline > 0 {
pendingInline--
return w.flush(c)
}
panic("both pendingPre and pendingInline exhausted")
}
for done := 0; done < len(p); {
// Construct an inline buffer if we're doing an inline
// encoding; see above regarding the bytes.MinRead constraint.
inline := false
if w.buf.Len() == 0 && len(p) >= done+int(w.chunkSize) && len(p) >= done+bytes.MinRead {
bufPool.Put(w.buf) // Return to the pool; never scheduled.
w.buf = bytes.NewBuffer(p[done : done+int(w.chunkSize)])
done += int(w.chunkSize)
pendingInline++
inline = true
}
// Do we need to flush w.buf? Note that this case should be hit
// immediately following the inline case above.
left := int(w.chunkSize) - w.buf.Len()
if left == 0 {
if err := w.schedule(newChunk(nil, nil, nil, w.buf), callback); err != nil {
return done, err
}
if !inline {
pendingPre++
}
// Reset the buffer, since this has now been scheduled
// for compression. Note that this may be trampled
// immediately by the bufPool.Put(w.buf) above if the
// next buffer happens to be inline, but that's okay.
w.buf = bufPool.Get().(*bytes.Buffer)
continue
}
// Read from p into w.buf.
toWrite := len(p) - done
if toWrite > left {
toWrite = left
}
n, err := w.buf.Write(p[done : done+toWrite])
done += n
if err != nil {
return done, err
}
}
// Make sure that everything has been flushed, we can't return until
// all the contents from p have been used.
for pendingInline > 0 {
if err := w.schedule(nil, func(c *chunk) error {
if err := callback(c); err != nil {
return err
}
// The flush was successful, return errNewBuffer here
// to break from the loop and check the condition
// again.
return errNewBuffer
}); err != errNewBuffer {
return len(p), err
}
}
return len(p), nil
}
// Close implements io.Closer.Close.
func (w *Writer) Close() error {
w.mu.Lock()
defer w.mu.Unlock()
// Did we already close? After the call to Close, we always mark as
// closed, regardless of whether the flush is successful.
if w.closed {
return io.ErrUnexpectedEOF
}
w.closed = true
defer w.stop()
// Schedule any remaining partial buffer; we pass w.flush directly here
// because the final buffer is guaranteed to not be an inline buffer.
if w.buf.Len() > 0 {
if err := w.schedule(newChunk(nil, nil, nil, w.buf), w.flush); err != nil {
return err
}
}
// Flush all scheduled buffers; see above.
if err := w.schedule(nil, w.flush); err != nil {
return err
}
// Close the underlying writer (if necessary).
if closer, ok := w.out.(io.Closer); ok {
return closer.Close()
}
return nil
}
|