io_win.cc 32.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
//  This source code is licensed under the BSD-style license found in the
//  LICENSE file in the root directory of this source tree. An additional grant
//  of patent rights can be found in the PATENTS file in the same directory.
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.

#include "port/win/io_win.h"

#include "util/sync_point.h"
#include "util/coding.h"
#include "util/iostats_context_imp.h"
#include "util/aligned_buffer.h"


namespace rocksdb {
namespace port {

D
Dmitri Smirnov 已提交
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
/*
* DirectIOHelper
*/
namespace {

const size_t kSectorSize = 512;

inline
bool IsPowerOfTwo(const size_t alignment) {
  return ((alignment) & (alignment - 1)) == 0;
}

inline
bool IsSectorAligned(const size_t off) { 
  return (off & (kSectorSize - 1)) == 0;
}

inline
bool IsAligned(size_t alignment, const void* ptr) {
  return ((uintptr_t(ptr)) & (alignment - 1)) == 0;
}
}


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
std::string GetWindowsErrSz(DWORD err) {
  LPSTR lpMsgBuf;
  FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM |
    FORMAT_MESSAGE_IGNORE_INSERTS,
    NULL, err,
    0,  // Default language
    reinterpret_cast<LPSTR>(&lpMsgBuf), 0, NULL);

  std::string Err = lpMsgBuf;
  LocalFree(lpMsgBuf);
  return Err;
}

// We preserve the original name of this interface to denote the original idea
// behind it.
// All reads happen by a specified offset and pwrite interface does not change
// the position of the file pointer. Judging from the man page and errno it does
// execute
// lseek atomically to return the position of the file back where it was.
// WriteFile() does not
// have this capability. Therefore, for both pread and pwrite the pointer is
// advanced to the next position
// which is fine for writes because they are (should be) sequential.
// Because all the reads/writes happen by the specified offset, the caller in
// theory should not
// rely on the current file offset.
SSIZE_T pwrite(HANDLE hFile, const char* src, size_t numBytes,
  uint64_t offset) {
  assert(numBytes <= std::numeric_limits<DWORD>::max());
  OVERLAPPED overlapped = { 0 };
  ULARGE_INTEGER offsetUnion;
  offsetUnion.QuadPart = offset;

  overlapped.Offset = offsetUnion.LowPart;
  overlapped.OffsetHigh = offsetUnion.HighPart;

  SSIZE_T result = 0;

  unsigned long bytesWritten = 0;

  if (FALSE == WriteFile(hFile, src, static_cast<DWORD>(numBytes), &bytesWritten,
    &overlapped)) {
    result = -1;
  } else {
    result = bytesWritten;
  }

  return result;
}

// See comments for pwrite above
SSIZE_T pread(HANDLE hFile, char* src, size_t numBytes, uint64_t offset) {
  assert(numBytes <= std::numeric_limits<DWORD>::max());
  OVERLAPPED overlapped = { 0 };
  ULARGE_INTEGER offsetUnion;
  offsetUnion.QuadPart = offset;

  overlapped.Offset = offsetUnion.LowPart;
  overlapped.OffsetHigh = offsetUnion.HighPart;

  SSIZE_T result = 0;

  unsigned long bytesRead = 0;

  if (FALSE == ReadFile(hFile, src, static_cast<DWORD>(numBytes), &bytesRead,
    &overlapped)) {
    return -1;
  } else {
    result = bytesRead;
  }

  return result;
}

// SetFileInformationByHandle() is capable of fast pre-allocates.
// However, this does not change the file end position unless the file is
// truncated and the pre-allocated space is not considered filled with zeros.
Status fallocate(const std::string& filename, HANDLE hFile,
  uint64_t to_size) {
  Status status;

  FILE_ALLOCATION_INFO alloc_info;
  alloc_info.AllocationSize.QuadPart = to_size;

  if (!SetFileInformationByHandle(hFile, FileAllocationInfo, &alloc_info,
    sizeof(FILE_ALLOCATION_INFO))) {
    auto lastError = GetLastError();
    status = IOErrorFromWindowsError(
      "Failed to pre-allocate space: " + filename, lastError);
  }

  return status;
}

Status ftruncate(const std::string& filename, HANDLE hFile,
  uint64_t toSize) {
  Status status;

  FILE_END_OF_FILE_INFO end_of_file;
  end_of_file.EndOfFile.QuadPart = toSize;

  if (!SetFileInformationByHandle(hFile, FileEndOfFileInfo, &end_of_file,
    sizeof(FILE_END_OF_FILE_INFO))) {
    auto lastError = GetLastError();
    status = IOErrorFromWindowsError("Failed to Set end of file: " + filename,
      lastError);
  }

  return status;
}

size_t GetUniqueIdFromFile(HANDLE hFile, char* id, size_t max_size) {

  if (max_size < kMaxVarint64Length * 3) {
    return 0;
  }

D
Dmitri Smirnov 已提交
162 163
  // This function has to be re-worked for cases when
  // ReFS file system introduced on Windows Server 2012 is used
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
  BY_HANDLE_FILE_INFORMATION FileInfo;

  BOOL result = GetFileInformationByHandle(hFile, &FileInfo);

  TEST_SYNC_POINT_CALLBACK("GetUniqueIdFromFile:FS_IOC_GETVERSION", &result);

  if (!result) {
    return 0;
  }

  char* rid = id;
  rid = EncodeVarint64(rid, uint64_t(FileInfo.dwVolumeSerialNumber));
  rid = EncodeVarint64(rid, uint64_t(FileInfo.nFileIndexHigh));
  rid = EncodeVarint64(rid, uint64_t(FileInfo.nFileIndexLow));

  assert(rid >= id);
  return static_cast<size_t>(rid - id);
}

183 184 185
////////////////////////////////////////////////////////////////////////////////////////////////////
// WinMmapReadableFile

A
Aaron Gao 已提交
186 187 188 189 190 191 192 193
WinMmapReadableFile::WinMmapReadableFile(const std::string& fileName,
                                         HANDLE hFile, HANDLE hMap,
                                         const void* mapped_region,
                                         size_t length)
    : WinFileData(fileName, hFile, false /* use_direct_io */),
      hMap_(hMap),
      mapped_region_(mapped_region),
      length_(length) {}
194 195 196 197 198 199 200 201 202 203 204 205 206 207 208

WinMmapReadableFile::~WinMmapReadableFile() {
  BOOL ret = ::UnmapViewOfFile(mapped_region_);
  assert(ret);

  ret = ::CloseHandle(hMap_);
  assert(ret);
}

Status WinMmapReadableFile::Read(uint64_t offset, size_t n, Slice* result,
  char* scratch) const {
  Status s;

  if (offset > length_) {
    *result = Slice();
209
    return IOError(filename_, EINVAL);
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
  } else if (offset + n > length_) {
    n = length_ - offset;
  }
  *result =
    Slice(reinterpret_cast<const char*>(mapped_region_)+offset, n);
  return s;
}

Status WinMmapReadableFile::InvalidateCache(size_t offset, size_t length) {
  return Status::OK();
}

size_t WinMmapReadableFile::GetUniqueId(char* id, size_t max_size) const {
  return GetUniqueIdFromFile(hFile_, id, max_size);
}

226 227 228 229
///////////////////////////////////////////////////////////////////////////////
/// WinMmapFile


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
// Can only truncate or reserve to a sector size aligned if
// used on files that are opened with Unbuffered I/O
Status WinMmapFile::TruncateFile(uint64_t toSize) {
  return ftruncate(filename_, hFile_, toSize);
}

Status WinMmapFile::UnmapCurrentRegion() {
  Status status;

  if (mapped_begin_ != nullptr) {
    if (!::UnmapViewOfFile(mapped_begin_)) {
      status = IOErrorFromWindowsError(
        "Failed to unmap file view: " + filename_, GetLastError());
    }

    // Move on to the next portion of the file
    file_offset_ += view_size_;

    // UnmapView automatically sends data to disk but not the metadata
    // which is good and provides some equivalent of fdatasync() on Linux
    // therefore, we donot need separate flag for metadata
    mapped_begin_ = nullptr;
    mapped_end_ = nullptr;
    dst_ = nullptr;

    last_sync_ = nullptr;
    pending_sync_ = false;
  }

  return status;
}

Status WinMmapFile::MapNewRegion() {

  Status status;

  assert(mapped_begin_ == nullptr);

  size_t minDiskSize = file_offset_ + view_size_;

  if (minDiskSize > reserved_size_) {
    status = Allocate(file_offset_, view_size_);
    if (!status.ok()) {
      return status;
    }
  }

  // Need to remap
  if (hMap_ == NULL || reserved_size_ > mapping_size_) {

    if (hMap_ != NULL) {
      // Unmap the previous one
      BOOL ret = ::CloseHandle(hMap_);
      assert(ret);
      hMap_ = NULL;
    }

    ULARGE_INTEGER mappingSize;
    mappingSize.QuadPart = reserved_size_;

    hMap_ = CreateFileMappingA(
      hFile_,
      NULL,                  // Security attributes
      PAGE_READWRITE,        // There is not a write only mode for mapping
      mappingSize.HighPart,  // Enable mapping the whole file but the actual
      // amount mapped is determined by MapViewOfFile
      mappingSize.LowPart,
      NULL);  // Mapping name

    if (NULL == hMap_) {
      return IOErrorFromWindowsError(
        "WindowsMmapFile failed to create file mapping for: " + filename_,
        GetLastError());
    }

    mapping_size_ = reserved_size_;
  }

  ULARGE_INTEGER offset;
  offset.QuadPart = file_offset_;

  // View must begin at the granularity aligned offset
  mapped_begin_ = reinterpret_cast<char*>(
    MapViewOfFileEx(hMap_, FILE_MAP_WRITE, offset.HighPart, offset.LowPart,
    view_size_, NULL));

  if (!mapped_begin_) {
    status = IOErrorFromWindowsError(
      "WindowsMmapFile failed to map file view: " + filename_,
      GetLastError());
  } else {
    mapped_end_ = mapped_begin_ + view_size_;
    dst_ = mapped_begin_;
    last_sync_ = mapped_begin_;
    pending_sync_ = false;
  }
  return status;
}

Status WinMmapFile::PreallocateInternal(uint64_t spaceToReserve) {
  return fallocate(filename_, hFile_, spaceToReserve);
}

WinMmapFile::WinMmapFile(const std::string& fname, HANDLE hFile, size_t page_size,
  size_t allocation_granularity, const EnvOptions& options)
335
  : WinFileData(fname, hFile, false),
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
  hMap_(NULL),
  page_size_(page_size),
  allocation_granularity_(allocation_granularity),
  reserved_size_(0),
  mapping_size_(0),
  view_size_(0),
  mapped_begin_(nullptr),
  mapped_end_(nullptr),
  dst_(nullptr),
  last_sync_(nullptr),
  file_offset_(0),
  pending_sync_(false) {
  // Allocation granularity must be obtained from GetSystemInfo() and must be
  // a power of two.
  assert(allocation_granularity > 0);
  assert((allocation_granularity & (allocation_granularity - 1)) == 0);

  assert(page_size > 0);
  assert((page_size & (page_size - 1)) == 0);

  // Only for memory mapped writes
  assert(options.use_mmap_writes);

  // View size must be both the multiple of allocation_granularity AND the
  // page size and the granularity is usually a multiple of a page size.
  const size_t viewSize = 32 * 1024; // 32Kb similar to the Windows File Cache in buffered mode
  view_size_ = Roundup(viewSize, allocation_granularity_);
}

WinMmapFile::~WinMmapFile() {
  if (hFile_) {
    this->Close();
  }
}

Status WinMmapFile::Append(const Slice& data) {
  const char* src = data.data();
  size_t left = data.size();

  while (left > 0) {
    assert(mapped_begin_ <= dst_);
    size_t avail = mapped_end_ - dst_;

    if (avail == 0) {
      Status s = UnmapCurrentRegion();
      if (s.ok()) {
        s = MapNewRegion();
      }

      if (!s.ok()) {
        return s;
      }
    } else {
      size_t n = std::min(left, avail);
      memcpy(dst_, src, n);
      dst_ += n;
      src += n;
      left -= n;
      pending_sync_ = true;
    }
  }

  // Now make sure that the last partial page is padded with zeros if needed
  size_t bytesToPad = Roundup(size_t(dst_), page_size_) - size_t(dst_);
  if (bytesToPad > 0) {
    memset(dst_, 0, bytesToPad);
  }

  return Status::OK();
}

// Means Close() will properly take care of truncate
// and it does not need any additional information
Status WinMmapFile::Truncate(uint64_t size) {
  return Status::OK();
}

Status WinMmapFile::Close() {
  Status s;

  assert(NULL != hFile_);

  // We truncate to the precise size so no
  // uninitialized data at the end. SetEndOfFile
  // which we use does not write zeros and it is good.
  uint64_t targetSize = GetFileSize();

  if (mapped_begin_ != nullptr) {
    // Sync before unmapping to make sure everything
    // is on disk and there is not a lazy writing
    // so we are deterministic with the tests
    Sync();
    s = UnmapCurrentRegion();
  }

  if (NULL != hMap_) {
    BOOL ret = ::CloseHandle(hMap_);
    if (!ret && s.ok()) {
      auto lastError = GetLastError();
      s = IOErrorFromWindowsError(
        "Failed to Close mapping for file: " + filename_, lastError);
    }

    hMap_ = NULL;
  }

  if (hFile_ != NULL) {

    TruncateFile(targetSize);

    BOOL ret = ::CloseHandle(hFile_);
    hFile_ = NULL;

    if (!ret && s.ok()) {
      auto lastError = GetLastError();
      s = IOErrorFromWindowsError(
        "Failed to close file map handle: " + filename_, lastError);
    }
  }

  return s;
}

Status WinMmapFile::Flush() { return Status::OK(); }

// Flush only data
Status WinMmapFile::Sync() {
  Status s;

  // Some writes occurred since last sync
  if (dst_ > last_sync_) {
    assert(mapped_begin_);
    assert(dst_);
    assert(dst_ > mapped_begin_);
    assert(dst_ < mapped_end_);

    size_t page_begin =
      TruncateToPageBoundary(page_size_, last_sync_ - mapped_begin_);
    size_t page_end =
      TruncateToPageBoundary(page_size_, dst_ - mapped_begin_ - 1);

    // Flush only the amount of that is a multiple of pages
    if (!::FlushViewOfFile(mapped_begin_ + page_begin,
      (page_end - page_begin) + page_size_)) {
      s = IOErrorFromWindowsError("Failed to FlushViewOfFile: " + filename_,
        GetLastError());
    } else {
      last_sync_ = dst_;
    }
  }

  return s;
}

/**
* Flush data as well as metadata to stable storage.
*/
Status WinMmapFile::Fsync() {
  Status s = Sync();

  // Flush metadata
  if (s.ok() && pending_sync_) {
    if (!::FlushFileBuffers(hFile_)) {
      s = IOErrorFromWindowsError("Failed to FlushFileBuffers: " + filename_,
        GetLastError());
    }
    pending_sync_ = false;
  }

  return s;
}

/**
* Get the size of valid data in the file. This will not match the
* size that is returned from the filesystem because we use mmap
* to extend file by map_size every time.
*/
uint64_t WinMmapFile::GetFileSize() {
  size_t used = dst_ - mapped_begin_;
  return file_offset_ + used;
}

Status WinMmapFile::InvalidateCache(size_t offset, size_t length) {
  return Status::OK();
}

Status WinMmapFile::Allocate(uint64_t offset, uint64_t len) {
  Status status;
  TEST_KILL_RANDOM("WinMmapFile::Allocate", rocksdb_kill_odds);

  // Make sure that we reserve an aligned amount of space
  // since the reservation block size is driven outside so we want
  // to check if we are ok with reservation here
  size_t spaceToReserve = Roundup(offset + len, view_size_);
  // Nothing to do
  if (spaceToReserve <= reserved_size_) {
    return status;
  }

  IOSTATS_TIMER_GUARD(allocate_nanos);
  status = PreallocateInternal(spaceToReserve);
  if (status.ok()) {
    reserved_size_ = spaceToReserve;
  }
  return status;
}

size_t WinMmapFile::GetUniqueId(char* id, size_t max_size) const {
  return GetUniqueIdFromFile(hFile_, id, max_size);
}

547 548 549
//////////////////////////////////////////////////////////////////////////////////
// WinSequentialFile

550
WinSequentialFile::WinSequentialFile(const std::string& fname, HANDLE f,
A
Aaron Gao 已提交
551 552
                                     const EnvOptions& options)
    : WinFileData(fname, f, options.use_direct_reads) {}
553 554

WinSequentialFile::~WinSequentialFile() {
555
  assert(hFile_ != INVALID_HANDLE_VALUE);
556 557 558
}

Status WinSequentialFile::Read(size_t n, Slice* result, char* scratch) {
D
Dmitri Smirnov 已提交
559
  assert(result != nullptr && !WinFileData::use_direct_io());
560 561 562 563 564 565 566 567 568 569 570 571
  Status s;
  size_t r = 0;

  // Windows ReadFile API accepts a DWORD.
  // While it is possible to read in a loop if n is > UINT_MAX
  // it is a highly unlikely case.
  if (n > UINT_MAX) {
    return IOErrorFromWindowsError(filename_, ERROR_INVALID_PARAMETER);
  }

  DWORD bytesToRead = static_cast<DWORD>(n); //cast is safe due to the check above
  DWORD bytesRead = 0;
572
  BOOL ret = ReadFile(hFile_, scratch, bytesToRead, &bytesRead, NULL);
573 574 575 576 577 578 579 580 581 582 583
  if (ret == TRUE) {
    r = bytesRead;
  } else {
    return IOErrorFromWindowsError(filename_, GetLastError());
  }

  *result = Slice(scratch, r);

  return s;
}

D
Dmitri Smirnov 已提交
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
SSIZE_T WinSequentialFile::PositionedReadInternal(char* src, size_t numBytes,
  uint64_t offset) const {
  return pread(GetFileHandle(), src, numBytes, offset);
}

Status WinSequentialFile::PositionedRead(uint64_t offset, size_t n, Slice* result,
  char* scratch) {

  Status s;

  assert(WinFileData::use_direct_io());

  // Windows ReadFile API accepts a DWORD.
  // While it is possible to read in a loop if n is > UINT_MAX
  // it is a highly unlikely case.
  if (n > UINT_MAX) {
    return IOErrorFromWindowsError(GetName(), ERROR_INVALID_PARAMETER);
  }

  auto r = PositionedReadInternal(scratch, n, offset);

  if (r < 0) {
    auto lastError = GetLastError();
    // Posix impl wants to treat reads from beyond
    // of the file as OK.
    if (lastError != ERROR_HANDLE_EOF) {
      s = IOErrorFromWindowsError(GetName(), lastError);
    }
  }

  *result = Slice(scratch, (r < 0) ? 0 : size_t(r));
  return s;
}


619 620 621 622 623 624 625 626 627
Status WinSequentialFile::Skip(uint64_t n) {
  // Can't handle more than signed max as SetFilePointerEx accepts a signed 64-bit
  // integer. As such it is a highly unlikley case to have n so large.
  if (n > _I64_MAX) {
    return IOErrorFromWindowsError(filename_, ERROR_INVALID_PARAMETER);
  }

  LARGE_INTEGER li;
  li.QuadPart = static_cast<int64_t>(n); //cast is safe due to the check above
628
  BOOL ret = SetFilePointerEx(hFile_, li, NULL, FILE_CURRENT);
629 630 631 632 633 634 635 636 637 638
  if (ret == FALSE) {
    return IOErrorFromWindowsError(filename_, GetLastError());
  }
  return Status::OK();
}

Status WinSequentialFile::InvalidateCache(size_t offset, size_t length) {
  return Status::OK();
}

639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655
//////////////////////////////////////////////////////////////////////////////////////////////////
/// WinRandomAccessBase

// Helper
void CalculateReadParameters(size_t alignment, uint64_t offset,
  size_t bytes_requested,
  size_t& actual_bytes_toread,
  uint64_t& first_page_start) {

  first_page_start = TruncateToPageBoundary(alignment, offset);
  const uint64_t last_page_start =
    TruncateToPageBoundary(alignment, offset + bytes_requested - 1);
  actual_bytes_toread = (last_page_start - first_page_start) + alignment;
}

SSIZE_T WinRandomAccessImpl::ReadIntoBuffer(uint64_t user_offset,
  uint64_t first_page_start,
656 657 658 659 660 661
  size_t bytes_to_read, size_t& left,
  AlignedBuffer& buffer, char* dest) const {
  assert(buffer.CurrentSize() == 0);
  assert(buffer.Capacity() >= bytes_to_read);

  SSIZE_T read =
662 663
    PositionedReadInternal(buffer.Destination(), bytes_to_read,
      first_page_start);
664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680

  if (read > 0) {
    buffer.Size(read);

    // Let's figure out how much we read from the users standpoint
    if ((first_page_start + buffer.CurrentSize()) > user_offset) {
      assert(first_page_start <= user_offset);
      size_t buffer_offset = user_offset - first_page_start;
      read = buffer.Read(dest, buffer_offset, left);
    } else {
      read = 0;
    }
    left -= read;
  }
  return read;
}

681 682
SSIZE_T WinRandomAccessImpl::ReadIntoOneShotBuffer(uint64_t user_offset,
  uint64_t first_page_start,
683 684 685 686 687 688 689 690 691 692
  size_t bytes_to_read, size_t& left,
  char* dest) const {
  AlignedBuffer bigBuffer;
  bigBuffer.Alignment(buffer_.Alignment());
  bigBuffer.AllocateNewBuffer(bytes_to_read);

  return ReadIntoBuffer(user_offset, first_page_start, bytes_to_read, left,
    bigBuffer, dest);
}

693
SSIZE_T WinRandomAccessImpl::ReadIntoInstanceBuffer(uint64_t user_offset,
694 695 696 697 698 699 700 701 702 703 704 705 706
  uint64_t first_page_start,
  size_t bytes_to_read, size_t& left,
  char* dest) const {
  SSIZE_T read = ReadIntoBuffer(user_offset, first_page_start, bytes_to_read,
    left, buffer_, dest);

  if (read > 0) {
    buffered_start_ = first_page_start;
  }

  return read;
}

707 708
SSIZE_T WinRandomAccessImpl::PositionedReadInternal(char* src,
  size_t numBytes,
709
  uint64_t offset) const {
710
  return pread(file_base_->GetFileHandle(), src, numBytes, offset);
711 712
}

713 714 715 716 717
inline
WinRandomAccessImpl::WinRandomAccessImpl(WinFileData* file_base,
  size_t alignment,
  const EnvOptions& options) :
    file_base_(file_base),
718 719 720 721 722
    read_ahead_(false),
    compaction_readahead_size_(options.compaction_readahead_size),
    random_access_max_buffer_size_(options.random_access_max_buffer_size),
    buffer_(),
    buffered_start_(0) {
723

724 725
  assert(!options.use_mmap_reads);

D
Dmitri Smirnov 已提交
726 727 728
  // Do not allocate the buffer either until the first request or
  // until there is a call to allocate a read-ahead buffer
  buffer_.Alignment(alignment);
729 730
}

731 732
inline
Status WinRandomAccessImpl::ReadImpl(uint64_t offset, size_t n, Slice* result,
733 734 735 736 737 738 739 740 741 742 743 744
  char* scratch) const {

  Status s;
  SSIZE_T r = -1;
  size_t left = n;
  char* dest = scratch;

  if (n == 0) {
    *result = Slice(scratch, 0);
    return s;
  }

A
Aaron Gao 已提交
745
  // When in direct I/O mode we need to do the following changes:
746 747
  // - use our own aligned buffer
  // - always read at the offset of that is a multiple of alignment
748
  if (file_base_->use_direct_io()) {
749 750 751 752 753
    uint64_t first_page_start = 0;
    size_t actual_bytes_toread = 0;
    size_t bytes_requested = left;

    if (!read_ahead_ && random_access_max_buffer_size_ == 0) {
754 755
      CalculateReadParameters(buffer_.Alignment(), offset, bytes_requested,
        actual_bytes_toread,
756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787
        first_page_start);

      assert(actual_bytes_toread > 0);

      r = ReadIntoOneShotBuffer(offset, first_page_start,
        actual_bytes_toread, left, dest);
    } else {

      std::unique_lock<std::mutex> lock(buffer_mut_);

      // Let's see if at least some of the requested data is already
      // in the buffer
      if (offset >= buffered_start_ &&
        offset < (buffered_start_ + buffer_.CurrentSize())) {
        size_t buffer_offset = offset - buffered_start_;
        r = buffer_.Read(dest, buffer_offset, left);
        assert(r >= 0);

        left -= size_t(r);
        offset += r;
        dest += r;
      }

      // Still some left or none was buffered
      if (left > 0) {
        // Figure out the start/end offset for reading and amount to read
        bytes_requested = left;

        if (read_ahead_ && bytes_requested < compaction_readahead_size_) {
          bytes_requested = compaction_readahead_size_;
        }

788 789
        CalculateReadParameters(buffer_.Alignment(), offset, bytes_requested,
          actual_bytes_toread,
790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823
          first_page_start);

        assert(actual_bytes_toread > 0);

        if (buffer_.Capacity() < actual_bytes_toread) {
          // If we are in read-ahead mode or the requested size
          // exceeds max buffer size then use one-shot
          // big buffer otherwise reallocate main buffer
          if (read_ahead_ ||
            (actual_bytes_toread > random_access_max_buffer_size_)) {
            // Unlock the mutex since we are not using instance buffer
            lock.unlock();
            r = ReadIntoOneShotBuffer(offset, first_page_start,
              actual_bytes_toread, left, dest);
          } else {
            buffer_.AllocateNewBuffer(actual_bytes_toread);
            r = ReadIntoInstanceBuffer(offset, first_page_start,
              actual_bytes_toread, left, dest);
          }
        } else {
          buffer_.Clear();
          r = ReadIntoInstanceBuffer(offset, first_page_start,
            actual_bytes_toread, left, dest);
        }
      }
    }
  } else {
    r = PositionedReadInternal(scratch, left, offset);
    if (r > 0) {
      left -= r;
    }
  }

  if (r < 0) {
824 825 826 827 828 829
    auto lastError = GetLastError();
    // Posix impl wants to treat reads from beyond
    // of the file as OK.
    if(lastError != ERROR_HANDLE_EOF) {
      s = IOErrorFromWindowsError(file_base_->GetName(), lastError);
    }
830
  }
831 832 833

  *result = Slice(scratch, (r < 0) ? 0 : n - left);

834 835 836
  return s;
}

837 838
inline
void WinRandomAccessImpl::HintImpl(RandomAccessFile::AccessPattern pattern) {
839
  if (pattern == RandomAccessFile::SEQUENTIAL && file_base_->use_direct_io() &&
A
Aaron Gao 已提交
840
      compaction_readahead_size_ > 0) {
841 842 843 844 845 846 847 848 849 850 851 852 853
    std::lock_guard<std::mutex> lg(buffer_mut_);
    if (!read_ahead_) {
      read_ahead_ = true;
      // This would allocate read-ahead size + 2 alignments
      // - one for memory alignment which added implicitly by AlignedBuffer
      // - We add one more alignment because we will read one alignment more
      // from disk
      buffer_.AllocateNewBuffer(compaction_readahead_size_ +
        buffer_.Alignment());
    }
  }
}

854 855 856
///////////////////////////////////////////////////////////////////////////////////////////////////
/// WinRandomAccessFile

A
Aaron Gao 已提交
857 858 859 860 861
WinRandomAccessFile::WinRandomAccessFile(const std::string& fname, HANDLE hFile,
                                         size_t alignment,
                                         const EnvOptions& options)
    : WinFileData(fname, hFile, options.use_direct_reads),
      WinRandomAccessImpl(this, alignment, options) {}
862

863
WinRandomAccessFile::~WinRandomAccessFile() {
864 865
}

866 867 868
Status WinRandomAccessFile::Read(uint64_t offset, size_t n, Slice* result,
  char* scratch) const {
  return ReadImpl(offset, n, result, scratch);
869 870
}

871 872
void WinRandomAccessFile::EnableReadAhead() {
  HintImpl(SEQUENTIAL);
873 874
}

875 876
bool WinRandomAccessFile::ShouldForwardRawRequest() const {
  return true;
877 878
}

879 880
void WinRandomAccessFile::Hint(AccessPattern pattern) {
  HintImpl(pattern);
881 882
}

883 884
Status WinRandomAccessFile::InvalidateCache(size_t offset, size_t length) {
  return Status::OK();
885 886
}

887 888 889 890
size_t WinRandomAccessFile::GetUniqueId(char* id, size_t max_size) const {
  return GetUniqueIdFromFile(GetFileHandle(), id, max_size);
}

D
Dmitri Smirnov 已提交
891 892 893 894
size_t WinRandomAccessFile::GetRequiredBufferAlignment() const {
  return GetAlignment();
}

895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
/////////////////////////////////////////////////////////////////////////////
// WinWritableImpl
//

inline
Status WinWritableImpl::PreallocateInternal(uint64_t spaceToReserve) {
  return fallocate(file_data_->GetName(), file_data_->GetFileHandle(), spaceToReserve);
}

WinWritableImpl::WinWritableImpl(WinFileData* file_data, size_t alignment)
  : file_data_(file_data),
  alignment_(alignment),
  filesize_(0),
  reservedsize_(0) {
}

Status WinWritableImpl::AppendImpl(const Slice& data) {
912

D
Dmitri Smirnov 已提交
913 914
  Status s;

915 916
  assert(data.size() < std::numeric_limits<DWORD>::max());

D
Dmitri Smirnov 已提交
917
  uint64_t written = 0;
918

919
  if (file_data_->use_direct_io()) {
D
Dmitri Smirnov 已提交
920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952

    // With no offset specified we are appending
    // to the end of the file

    assert(IsSectorAligned(filesize_));
    assert(IsSectorAligned(data.size()));
    assert(IsAligned(GetAlignement(), data.data()));

    SSIZE_T ret = pwrite(file_data_->GetFileHandle(), data.data(),
     data.size(), filesize_);

    if (ret < 0) {
      auto lastError = GetLastError();
      s = IOErrorFromWindowsError(
        "Failed to pwrite for: " + file_data_->GetName(), lastError);
    }
    else {
      written = ret;
    }

  } else {

    DWORD bytesWritten = 0;
    if (!WriteFile(file_data_->GetFileHandle(), data.data(),
      static_cast<DWORD>(data.size()), &bytesWritten, NULL)) {
      auto lastError = GetLastError();
      s = IOErrorFromWindowsError(
        "Failed to WriteFile: " + file_data_->GetName(),
        lastError);
    }
    else {
      written = bytesWritten;
    }
953
  }
D
Dmitri Smirnov 已提交
954 955 956

  if(s.ok()) {
    assert(written == data.size());
957 958 959 960 961 962
    filesize_ += data.size();
  }

  return s;
}

963
Status WinWritableImpl::PositionedAppendImpl(const Slice& data, uint64_t offset) {
D
Dmitri Smirnov 已提交
964

965
  if(file_data_->use_direct_io()) {
D
Dmitri Smirnov 已提交
966 967 968 969 970
    assert(IsSectorAligned(offset));
    assert(IsSectorAligned(data.size()));
    assert(IsAligned(GetAlignement(), data.data()));
  }

971 972
  Status s;

973
  SSIZE_T ret = pwrite(file_data_->GetFileHandle(), data.data(), data.size(), offset);
974 975 976 977 978

  // Error break
  if (ret < 0) {
    auto lastError = GetLastError();
    s = IOErrorFromWindowsError(
979 980 981
      "Failed to pwrite for: " + file_data_->GetName(), lastError);
  }
  else {
982
    assert(size_t(ret) == data.size());
A
Aaron Gao 已提交
983
    // For sequential write this would be simple
984 985 986 987 988
    // size extension by data.size()
    uint64_t write_end = offset + data.size();
    if (write_end >= filesize_) {
      filesize_ = write_end;
    }
989 990 991 992
  }
  return s;
}

993 994 995 996 997 998
// Need to implement this so the file is truncated correctly
// when buffered and unbuffered mode
inline
Status WinWritableImpl::TruncateImpl(uint64_t size) {
  Status s = ftruncate(file_data_->GetName(), file_data_->GetFileHandle(),
    size);
999 1000 1001 1002 1003 1004
  if (s.ok()) {
    filesize_ = size;
  }
  return s;
}

1005
Status WinWritableImpl::CloseImpl() {
1006 1007 1008

  Status s;

1009 1010
  auto hFile = file_data_->GetFileHandle();
  assert(INVALID_HANDLE_VALUE != hFile);
1011

1012
  if (fsync(hFile) < 0) {
1013
    auto lastError = GetLastError();
1014 1015
    s = IOErrorFromWindowsError("fsync failed at Close() for: " +
      file_data_->GetName(),
1016 1017 1018
      lastError);
  }

1019
  if(!file_data_->CloseFile()) {
1020
    auto lastError = GetLastError();
1021
    s = IOErrorFromWindowsError("CloseHandle failed for: " + file_data_->GetName(),
1022 1023 1024 1025 1026
      lastError);
  }
  return s;
}

1027
Status WinWritableImpl::SyncImpl() {
1028 1029
  Status s;
  // Calls flush buffers
1030
  if (fsync(file_data_->GetFileHandle()) < 0) {
1031
    auto lastError = GetLastError();
A
Aaron Gao 已提交
1032 1033
    s = IOErrorFromWindowsError(
        "fsync failed at Sync() for: " + file_data_->GetName(), lastError);
1034 1035 1036 1037 1038
  }
  return s;
}


1039
Status WinWritableImpl::AllocateImpl(uint64_t offset, uint64_t len) {
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
  Status status;
  TEST_KILL_RANDOM("WinWritableFile::Allocate", rocksdb_kill_odds);

  // Make sure that we reserve an aligned amount of space
  // since the reservation block size is driven outside so we want
  // to check if we are ok with reservation here
  size_t spaceToReserve = Roundup(offset + len, alignment_);
  // Nothing to do
  if (spaceToReserve <= reservedsize_) {
    return status;
  }

  IOSTATS_TIMER_GUARD(allocate_nanos);
  status = PreallocateInternal(spaceToReserve);
  if (status.ok()) {
    reservedsize_ = spaceToReserve;
  }
  return status;
}

1060 1061 1062 1063

////////////////////////////////////////////////////////////////////////////////
/// WinWritableFile

A
Aaron Gao 已提交
1064 1065 1066 1067 1068
WinWritableFile::WinWritableFile(const std::string& fname, HANDLE hFile,
                                 size_t alignment, size_t /* capacity */,
                                 const EnvOptions& options)
    : WinFileData(fname, hFile, options.use_direct_writes),
      WinWritableImpl(this, alignment) {
1069 1070 1071 1072 1073 1074
  assert(!options.use_mmap_writes);
}

WinWritableFile::~WinWritableFile() {
}

A
Aaron Gao 已提交
1075
// Indicates if the class makes use of direct I/O
1076
bool WinWritableFile::use_direct_io() const { return WinFileData::use_direct_io(); }
1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109

size_t WinWritableFile::GetRequiredBufferAlignment() const {
  return GetAlignement();
}

Status WinWritableFile::Append(const Slice& data) {
  return AppendImpl(data);
}

Status WinWritableFile::PositionedAppend(const Slice& data, uint64_t offset) {
  return PositionedAppendImpl(data, offset);
}

// Need to implement this so the file is truncated correctly
// when buffered and unbuffered mode
Status WinWritableFile::Truncate(uint64_t size) {
  return TruncateImpl(size);
}

Status WinWritableFile::Close() {
  return CloseImpl();
}

  // write out the cached data to the OS cache
  // This is now taken care of the WritableFileWriter
Status WinWritableFile::Flush() {
  return Status::OK();
}

Status WinWritableFile::Sync() {
  return SyncImpl();
}

A
Aaron Gao 已提交
1110
Status WinWritableFile::Fsync() { return SyncImpl(); }
1111 1112 1113 1114 1115 1116 1117 1118 1119

uint64_t WinWritableFile::GetFileSize() {
  return GetFileSizeImpl();
}

Status WinWritableFile::Allocate(uint64_t offset, uint64_t len) {
  return AllocateImpl(offset, len);
}

1120
size_t WinWritableFile::GetUniqueId(char* id, size_t max_size) const {
1121 1122 1123 1124 1125 1126
  return GetUniqueIdFromFile(GetFileHandle(), id, max_size);
}

/////////////////////////////////////////////////////////////////////////
/// WinRandomRWFile

A
Aaron Gao 已提交
1127 1128 1129 1130 1131 1132
WinRandomRWFile::WinRandomRWFile(const std::string& fname, HANDLE hFile,
                                 size_t alignment, const EnvOptions& options)
    : WinFileData(fname, hFile,
                  options.use_direct_reads && options.use_direct_writes),
      WinRandomAccessImpl(this, alignment, options),
      WinWritableImpl(this, alignment) {}
1133

1134
bool WinRandomRWFile::use_direct_io() const { return WinFileData::use_direct_io(); }
1135

1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
size_t WinRandomRWFile::GetRequiredBufferAlignment() const {
  return GetAlignement();
}

bool WinRandomRWFile::ShouldForwardRawRequest() const {
  return true;
}

void WinRandomRWFile::EnableReadAhead() {
  HintImpl(RandomAccessFile::SEQUENTIAL);
}

Status WinRandomRWFile::Write(uint64_t offset, const Slice & data) {
  return PositionedAppendImpl(data, offset);
}

A
Aaron Gao 已提交
1152 1153
Status WinRandomRWFile::Read(uint64_t offset, size_t n, Slice* result,
                             char* scratch) const {
1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
  return ReadImpl(offset, n, result, scratch);
}

Status WinRandomRWFile::Flush() {
  return Status::OK();
}

Status WinRandomRWFile::Sync() {
  return SyncImpl();
}

Status WinRandomRWFile::Close() {
  return CloseImpl();
}

//////////////////////////////////////////////////////////////////////////
/// WinDirectory

1172 1173
Status WinDirectory::Fsync() { return Status::OK(); }

1174 1175 1176
//////////////////////////////////////////////////////////////////////////
/// WinFileLock

1177 1178 1179 1180 1181 1182 1183
WinFileLock::~WinFileLock() {
  BOOL ret = ::CloseHandle(hFile_);
  assert(ret);
}

}
}