Adding upstream version 18.2.2.upstream/18.2.2

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

29 files changed, 6195 insertions, 0 deletions

diff --git a/src/rocksdb/port/README b/src/rocksdb/port/README
new file mode 100644
index 000000000..422563e25
--- /dev/null
+++ b/src/rocksdb/port/README
@@ -0,0 +1,10 @@
+This directory contains interfaces and implementations that isolate the
+rest of the package from platform details.
+
+Code in the rest of the package includes "port.h" from this directory.
+"port.h" in turn includes a platform specific "port_<platform>.h" file
+that provides the platform specific implementation.
+
+See port_posix.h for an example of what must be provided in a platform
+specific header file.
+
diff --git a/src/rocksdb/port/jemalloc_helper.h b/src/rocksdb/port/jemalloc_helper.h
new file mode 100644
index 000000000..f085f6226
--- /dev/null
+++ b/src/rocksdb/port/jemalloc_helper.h
@@ -0,0 +1,107 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#if defined(__clang__) && defined(__GLIBC__)
+// glibc's `posix_memalign()` declaration specifies `throw()` while clang's
+// declaration does not. There is a hack in clang to make its re-declaration
+// compatible with glibc's if they are declared consecutively. That hack breaks
+// if yet another `posix_memalign()` declaration comes between glibc's and
+// clang's declarations. Include "mm_malloc.h" here ensures glibc's and clang's
+// declarations both come before "jemalloc.h"'s `posix_memalign()` declaration.
+//
+// This problem could also be avoided if "jemalloc.h"'s `posix_memalign()`
+// declaration did not specify `throw()` when built with clang.
+#include <mm_malloc.h>
+#endif
+
+#ifdef ROCKSDB_JEMALLOC
+#ifdef __FreeBSD__
+#include <malloc_np.h>
+#define JEMALLOC_USABLE_SIZE_CONST const
+#else
+#define JEMALLOC_MANGLE
+#include <jemalloc/jemalloc.h>
+#endif
+
+#ifndef JEMALLOC_CXX_THROW
+#define JEMALLOC_CXX_THROW
+#endif
+
+#if defined(OS_WIN) && defined(_MSC_VER)
+
+// MSVC does not have weak symbol support. As long as ROCKSDB_JEMALLOC is
+// defined, Jemalloc memory allocator is used.
+static inline bool HasJemalloc() { return true; }
+
+#else
+
+// definitions for compatibility with older versions of jemalloc
+#if !defined(JEMALLOC_ALLOCATOR)
+#define JEMALLOC_ALLOCATOR
+#endif
+#if !defined(JEMALLOC_RESTRICT_RETURN)
+#define JEMALLOC_RESTRICT_RETURN
+#endif
+#if !defined(JEMALLOC_NOTHROW)
+#define JEMALLOC_NOTHROW JEMALLOC_ATTR(nothrow)
+#endif
+#if !defined(JEMALLOC_ALLOC_SIZE)
+#ifdef JEMALLOC_HAVE_ATTR_ALLOC_SIZE
+#define JEMALLOC_ALLOC_SIZE(s) JEMALLOC_ATTR(alloc_size(s))
+#else
+#define JEMALLOC_ALLOC_SIZE(s)
+#endif
+#endif
+
+// Declare non-standard jemalloc APIs as weak symbols. We can null-check these
+// symbols to detect whether jemalloc is linked with the binary.
+extern "C" JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN void JEMALLOC_NOTHROW *
+mallocx(size_t, int) JEMALLOC_ATTR(malloc) JEMALLOC_ALLOC_SIZE(1)
+    __attribute__((__weak__));
+extern "C" JEMALLOC_ALLOCATOR JEMALLOC_RESTRICT_RETURN void JEMALLOC_NOTHROW *
+rallocx(void *, size_t, int) JEMALLOC_ALLOC_SIZE(2) __attribute__((__weak__));
+extern "C" size_t JEMALLOC_NOTHROW xallocx(void *, size_t, size_t, int)
+    __attribute__((__weak__));
+extern "C" size_t JEMALLOC_NOTHROW sallocx(const void *, int)
+    JEMALLOC_ATTR(pure) __attribute__((__weak__));
+extern "C" void JEMALLOC_NOTHROW dallocx(void *, int) __attribute__((__weak__));
+extern "C" void JEMALLOC_NOTHROW sdallocx(void *, size_t, int)
+    __attribute__((__weak__));
+extern "C" size_t JEMALLOC_NOTHROW nallocx(size_t, int) JEMALLOC_ATTR(pure)
+    __attribute__((__weak__));
+extern "C" int JEMALLOC_NOTHROW mallctl(const char *, void *, size_t *, void *,
+                                        size_t) __attribute__((__weak__));
+extern "C" int JEMALLOC_NOTHROW mallctlnametomib(const char *, size_t *,
+                                                 size_t *)
+    __attribute__((__weak__));
+extern "C" int JEMALLOC_NOTHROW mallctlbymib(const size_t *, size_t, void *,
+                                             size_t *, void *, size_t)
+    __attribute__((__weak__));
+extern "C" void JEMALLOC_NOTHROW
+malloc_stats_print(void (*)(void *, const char *), void *, const char *)
+    __attribute__((__weak__));
+extern "C" size_t JEMALLOC_NOTHROW
+malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void *) JEMALLOC_CXX_THROW
+    __attribute__((__weak__));
+
+// Check if Jemalloc is linked with the binary. Note the main program might be
+// using a different memory allocator even this method return true.
+// It is loosely based on folly::usingJEMalloc(), minus the check that actually
+// allocate memory and see if it is through jemalloc, to handle the dlopen()
+// case:
+// https://github.com/facebook/folly/blob/76cf8b5841fb33137cfbf8b224f0226437c855bc/folly/memory/Malloc.h#L147
+static inline bool HasJemalloc() {
+  return mallocx != nullptr && rallocx != nullptr && xallocx != nullptr &&
+         sallocx != nullptr && dallocx != nullptr && sdallocx != nullptr &&
+         nallocx != nullptr && mallctl != nullptr &&
+         mallctlnametomib != nullptr && mallctlbymib != nullptr &&
+         malloc_stats_print != nullptr && malloc_usable_size != nullptr;
+}
+
+#endif
+
+#endif  // ROCKSDB_JEMALLOC
diff --git a/src/rocksdb/port/lang.h b/src/rocksdb/port/lang.h
new file mode 100644
index 000000000..52c597acd
--- /dev/null
+++ b/src/rocksdb/port/lang.h
@@ -0,0 +1,70 @@
+// Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#pragma once
+
+#ifndef FALLTHROUGH_INTENDED
+#if defined(__clang__)
+#define FALLTHROUGH_INTENDED [[clang::fallthrough]]
+#elif defined(__GNUC__) && __GNUC__ >= 7
+#define FALLTHROUGH_INTENDED [[gnu::fallthrough]]
+#else
+#define FALLTHROUGH_INTENDED \
+  do {                       \
+  } while (0)
+#endif
+#endif
+
+#define DECLARE_DEFAULT_MOVES(Name) \
+  Name(Name&&) noexcept = default;  \
+  Name& operator=(Name&&) = default
+
+// ASAN (Address sanitizer)
+
+#if defined(__clang__)
+#if defined(__has_feature)
+#if __has_feature(address_sanitizer)
+#define MUST_FREE_HEAP_ALLOCATIONS 1
+#endif  // __has_feature(address_sanitizer)
+#endif  // defined(__has_feature)
+#else   // __clang__
+#ifdef __SANITIZE_ADDRESS__
+#define MUST_FREE_HEAP_ALLOCATIONS 1
+#endif  // __SANITIZE_ADDRESS__
+#endif  // __clang__
+
+#ifdef ROCKSDB_VALGRIND_RUN
+#define MUST_FREE_HEAP_ALLOCATIONS 1
+#endif  // ROCKSDB_VALGRIND_RUN
+
+// Coding guidelines say to avoid static objects with non-trivial destructors,
+// because it's easy to cause trouble (UB) in static destruction. This
+// macro makes it easier to define static objects that are normally never
+// destructed, except are destructed when running under ASAN. This should
+// avoid unexpected, unnecessary destruction behavior in production.
+// Note that constructor arguments can be provided as in
+//   STATIC_AVOID_DESTRUCTION(Foo, foo)(arg1, arg2);
+#ifdef MUST_FREE_HEAP_ALLOCATIONS
+#define STATIC_AVOID_DESTRUCTION(Type, name) static Type name
+constexpr bool kMustFreeHeapAllocations = true;
+#else
+#define STATIC_AVOID_DESTRUCTION(Type, name) static Type& name = *new Type
+constexpr bool kMustFreeHeapAllocations = false;
+#endif
+
+// TSAN (Thread sanitizer)
+
+// For simplicity, standardize on the GCC define
+#if defined(__clang__)
+#if defined(__has_feature) && __has_feature(thread_sanitizer)
+#define __SANITIZE_THREAD__ 1
+#endif  // __has_feature(thread_sanitizer)
+#endif  // __clang__
+
+#ifdef __SANITIZE_THREAD__
+#define TSAN_SUPPRESSION __attribute__((no_sanitize("thread")))
+#else
+#define TSAN_SUPPRESSION
+#endif  // TSAN_SUPPRESSION
diff --git a/src/rocksdb/port/likely.h b/src/rocksdb/port/likely.h
new file mode 100644
index 000000000..0bd90d701
--- /dev/null
+++ b/src/rocksdb/port/likely.h
@@ -0,0 +1,18 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#pragma once
+
+#if defined(__GNUC__) && __GNUC__ >= 4
+#define LIKELY(x) (__builtin_expect((x), 1))
+#define UNLIKELY(x) (__builtin_expect((x), 0))
+#else
+#define LIKELY(x) (x)
+#define UNLIKELY(x) (x)
+#endif
diff --git a/src/rocksdb/port/malloc.h b/src/rocksdb/port/malloc.h
new file mode 100644
index 000000000..f973263e2
--- /dev/null
+++ b/src/rocksdb/port/malloc.h
@@ -0,0 +1,17 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+#pragma once
+
+#ifdef ROCKSDB_MALLOC_USABLE_SIZE
+#ifdef OS_FREEBSD
+#include <malloc_np.h>
+#else
+#include <malloc.h>
+#endif  // OS_FREEBSD
+#endif  // ROCKSDB_MALLOC_USABLE_SIZE
diff --git a/src/rocksdb/port/port.h b/src/rocksdb/port/port.h
new file mode 100644
index 000000000..13aa56d47
--- /dev/null
+++ b/src/rocksdb/port/port.h
@@ -0,0 +1,21 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#pragma once
+
+#include <string>
+
+// Include the appropriate platform specific file below.  If you are
+// porting to a new platform, see "port_example.h" for documentation
+// of what the new port_<platform>.h file must provide.
+#if defined(ROCKSDB_PLATFORM_POSIX)
+#include "port/port_posix.h"
+#elif defined(OS_WIN)
+#include "port/win/port_win.h"
+#endif
diff --git a/src/rocksdb/port/port_dirent.h b/src/rocksdb/port/port_dirent.h
new file mode 100644
index 000000000..2b23e2f07
--- /dev/null
+++ b/src/rocksdb/port/port_dirent.h
@@ -0,0 +1,44 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// See port_example.h for documentation for the following types/functions.
+
+#pragma once
+
+#ifdef ROCKSDB_PLATFORM_POSIX
+#include <dirent.h>
+#include <sys/types.h>
+#elif defined(OS_WIN)
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+struct dirent {
+  char d_name[_MAX_PATH]; /* filename */
+};
+
+struct DIR;
+
+DIR* opendir(const char* name);
+
+dirent* readdir(DIR* dirp);
+
+int closedir(DIR* dirp);
+
+}  // namespace port
+
+using port::closedir;
+using port::DIR;
+using port::dirent;
+using port::opendir;
+using port::readdir;
+
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif  // OS_WIN
diff --git a/src/rocksdb/port/port_example.h b/src/rocksdb/port/port_example.h
new file mode 100644
index 000000000..794149a69
--- /dev/null
+++ b/src/rocksdb/port/port_example.h
@@ -0,0 +1,101 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// This file contains the specification, but not the implementations,
+// of the types/operations/etc. that should be defined by a platform
+// specific port_<platform>.h file.  Use this file as a reference for
+// how to port this package to a new platform.
+
+#pragma once
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+// TODO(jorlow): Many of these belong more in the environment class rather than
+//               here. We should try moving them and see if it affects perf.
+
+// The following boolean constant must be true on a little-endian machine
+// and false otherwise.
+static const bool kLittleEndian = true /* or some other expression */;
+
+// ------------------ Threading -------------------
+
+// A Mutex represents an exclusive lock.
+class Mutex {
+ public:
+  Mutex();
+  ~Mutex();
+
+  // Lock the mutex.  Waits until other lockers have exited.
+  // Will deadlock if the mutex is already locked by this thread.
+  void Lock();
+
+  // Unlock the mutex.
+  // REQUIRES: This mutex was locked by this thread.
+  void Unlock();
+
+  // Optionally crash if this thread does not hold this mutex.
+  // The implementation must be fast, especially if NDEBUG is
+  // defined.  The implementation is allowed to skip all checks.
+  void AssertHeld();
+};
+
+class CondVar {
+ public:
+  explicit CondVar(Mutex* mu);
+  ~CondVar();
+
+  // Atomically release *mu and block on this condition variable until
+  // either a call to SignalAll(), or a call to Signal() that picks
+  // this thread to wakeup.
+  // REQUIRES: this thread holds *mu
+  void Wait();
+
+  // If there are some threads waiting, wake up at least one of them.
+  void Signal();
+
+  // Wake up all waiting threads.
+  void SignallAll();
+};
+
+// Thread-safe initialization.
+// Used as follows:
+//      static port::OnceType init_control = LEVELDB_ONCE_INIT;
+//      static void Initializer() { ... do something ...; }
+//      ...
+//      port::InitOnce(&init_control, &Initializer);
+using OnceType = intptr_t;
+#define LEVELDB_ONCE_INIT 0
+extern void InitOnce(port::OnceType*, void (*initializer)());
+
+// ------------------ Compression -------------------
+
+// Store the snappy compression of "input[0,input_length-1]" in *output.
+// Returns false if snappy is not supported by this port.
+extern bool Snappy_Compress(const char* input, size_t input_length,
+                            std::string* output);
+
+// If input[0,input_length-1] looks like a valid snappy compressed
+// buffer, store the size of the uncompressed data in *result and
+// return true.  Else return false.
+extern bool Snappy_GetUncompressedLength(const char* input, size_t length,
+                                         size_t* result);
+
+// Attempt to snappy uncompress input[0,input_length-1] into *output.
+// Returns true if successful, false if the input is invalid lightweight
+// compressed data.
+//
+// REQUIRES: at least the first "n" bytes of output[] must be writable
+// where "n" is the result of a successful call to
+// Snappy_GetUncompressedLength.
+extern bool Snappy_Uncompress(const char* input_data, size_t input_length,
+                              char* output);
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/port_posix.cc b/src/rocksdb/port/port_posix.cc
new file mode 100644
index 000000000..3872293b8
--- /dev/null
+++ b/src/rocksdb/port/port_posix.cc
@@ -0,0 +1,300 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if !defined(OS_WIN)
+
+#include "port/port_posix.h"
+
+#include <assert.h>
+#if defined(__i386__) || defined(__x86_64__)
+#include <cpuid.h>
+#endif
+#include <errno.h>
+#include <sched.h>
+#include <signal.h>
+#include <stdio.h>
+#include <string.h>
+#include <sys/resource.h>
+#include <sys/time.h>
+#include <unistd.h>
+
+#include <cstdlib>
+#include <fstream>
+#include <string>
+
+#include "util/string_util.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+// We want to give users opportunity to default all the mutexes to adaptive if
+// not specified otherwise. This enables a quick way to conduct various
+// performance related experiements.
+//
+// NB! Support for adaptive mutexes is turned on by definining
+// ROCKSDB_PTHREAD_ADAPTIVE_MUTEX during the compilation. If you use RocksDB
+// build environment then this happens automatically; otherwise it's up to the
+// consumer to define the identifier.
+#ifdef ROCKSDB_DEFAULT_TO_ADAPTIVE_MUTEX
+extern const bool kDefaultToAdaptiveMutex = true;
+#else
+extern const bool kDefaultToAdaptiveMutex = false;
+#endif
+
+namespace port {
+
+static int PthreadCall(const char* label, int result) {
+  if (result != 0 && result != ETIMEDOUT && result != EBUSY) {
+    fprintf(stderr, "pthread %s: %s\n", label, errnoStr(result).c_str());
+    abort();
+  }
+  return result;
+}
+
+Mutex::Mutex(bool adaptive) {
+  (void)adaptive;
+#ifdef ROCKSDB_PTHREAD_ADAPTIVE_MUTEX
+  if (!adaptive) {
+    PthreadCall("init mutex", pthread_mutex_init(&mu_, nullptr));
+  } else {
+    pthread_mutexattr_t mutex_attr;
+    PthreadCall("init mutex attr", pthread_mutexattr_init(&mutex_attr));
+    PthreadCall("set mutex attr", pthread_mutexattr_settype(
+                                      &mutex_attr, PTHREAD_MUTEX_ADAPTIVE_NP));
+    PthreadCall("init mutex", pthread_mutex_init(&mu_, &mutex_attr));
+    PthreadCall("destroy mutex attr", pthread_mutexattr_destroy(&mutex_attr));
+  }
+#else
+  PthreadCall("init mutex", pthread_mutex_init(&mu_, nullptr));
+#endif  // ROCKSDB_PTHREAD_ADAPTIVE_MUTEX
+}
+
+Mutex::~Mutex() { PthreadCall("destroy mutex", pthread_mutex_destroy(&mu_)); }
+
+void Mutex::Lock() {
+  PthreadCall("lock", pthread_mutex_lock(&mu_));
+#ifndef NDEBUG
+  locked_ = true;
+#endif
+}
+
+void Mutex::Unlock() {
+#ifndef NDEBUG
+  locked_ = false;
+#endif
+  PthreadCall("unlock", pthread_mutex_unlock(&mu_));
+}
+
+bool Mutex::TryLock() {
+  bool ret = PthreadCall("trylock", pthread_mutex_trylock(&mu_)) == 0;
+#ifndef NDEBUG
+  if (ret) {
+    locked_ = true;
+  }
+#endif
+  return ret;
+}
+
+void Mutex::AssertHeld() {
+#ifndef NDEBUG
+  assert(locked_);
+#endif
+}
+
+CondVar::CondVar(Mutex* mu) : mu_(mu) {
+  PthreadCall("init cv", pthread_cond_init(&cv_, nullptr));
+}
+
+CondVar::~CondVar() { PthreadCall("destroy cv", pthread_cond_destroy(&cv_)); }
+
+void CondVar::Wait() {
+#ifndef NDEBUG
+  mu_->locked_ = false;
+#endif
+  PthreadCall("wait", pthread_cond_wait(&cv_, &mu_->mu_));
+#ifndef NDEBUG
+  mu_->locked_ = true;
+#endif
+}
+
+bool CondVar::TimedWait(uint64_t abs_time_us) {
+  struct timespec ts;
+  ts.tv_sec = static_cast<time_t>(abs_time_us / 1000000);
+  ts.tv_nsec = static_cast<suseconds_t>((abs_time_us % 1000000) * 1000);
+
+#ifndef NDEBUG
+  mu_->locked_ = false;
+#endif
+  int err = pthread_cond_timedwait(&cv_, &mu_->mu_, &ts);
+#ifndef NDEBUG
+  mu_->locked_ = true;
+#endif
+  if (err == ETIMEDOUT) {
+    return true;
+  }
+  if (err != 0) {
+    PthreadCall("timedwait", err);
+  }
+  return false;
+}
+
+void CondVar::Signal() { PthreadCall("signal", pthread_cond_signal(&cv_)); }
+
+void CondVar::SignalAll() {
+  PthreadCall("broadcast", pthread_cond_broadcast(&cv_));
+}
+
+RWMutex::RWMutex() {
+  PthreadCall("init mutex", pthread_rwlock_init(&mu_, nullptr));
+}
+
+RWMutex::~RWMutex() {
+  PthreadCall("destroy mutex", pthread_rwlock_destroy(&mu_));
+}
+
+void RWMutex::ReadLock() {
+  PthreadCall("read lock", pthread_rwlock_rdlock(&mu_));
+}
+
+void RWMutex::WriteLock() {
+  PthreadCall("write lock", pthread_rwlock_wrlock(&mu_));
+}
+
+void RWMutex::ReadUnlock() {
+  PthreadCall("read unlock", pthread_rwlock_unlock(&mu_));
+}
+
+void RWMutex::WriteUnlock() {
+  PthreadCall("write unlock", pthread_rwlock_unlock(&mu_));
+}
+
+int PhysicalCoreID() {
+#if defined(ROCKSDB_SCHED_GETCPU_PRESENT) && defined(__x86_64__) && \
+    (__GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 22))
+  // sched_getcpu uses VDSO getcpu() syscall since 2.22. I believe Linux offers
+  // VDSO support only on x86_64. This is the fastest/preferred method if
+  // available.
+  int cpuno = sched_getcpu();
+  if (cpuno < 0) {
+    return -1;
+  }
+  return cpuno;
+#elif defined(__x86_64__) || defined(__i386__)
+  // clang/gcc both provide cpuid.h, which defines __get_cpuid(), for x86_64 and
+  // i386.
+  unsigned eax, ebx = 0, ecx, edx;
+  if (!__get_cpuid(1, &eax, &ebx, &ecx, &edx)) {
+    return -1;
+  }
+  return ebx >> 24;
+#else
+  // give up, the caller can generate a random number or something.
+  return -1;
+#endif
+}
+
+void InitOnce(OnceType* once, void (*initializer)()) {
+  PthreadCall("once", pthread_once(once, initializer));
+}
+
+void Crash(const std::string& srcfile, int srcline) {
+  fprintf(stdout, "Crashing at %s:%d\n", srcfile.c_str(), srcline);
+  fflush(stdout);
+  kill(getpid(), SIGTERM);
+}
+
+int GetMaxOpenFiles() {
+#if defined(RLIMIT_NOFILE)
+  struct rlimit no_files_limit;
+  if (getrlimit(RLIMIT_NOFILE, &no_files_limit) != 0) {
+    return -1;
+  }
+  // protect against overflow
+  if (static_cast<uintmax_t>(no_files_limit.rlim_cur) >=
+      static_cast<uintmax_t>(std::numeric_limits<int>::max())) {
+    return std::numeric_limits<int>::max();
+  }
+  return static_cast<int>(no_files_limit.rlim_cur);
+#endif
+  return -1;
+}
+
+void* cacheline_aligned_alloc(size_t size) {
+#if __GNUC__ < 5 && defined(__SANITIZE_ADDRESS__)
+  return malloc(size);
+#elif (_POSIX_C_SOURCE >= 200112L || _XOPEN_SOURCE >= 600 || defined(__APPLE__))
+  void* m;
+  errno = posix_memalign(&m, CACHE_LINE_SIZE, size);
+  return errno ? nullptr : m;
+#else
+  return malloc(size);
+#endif
+}
+
+void cacheline_aligned_free(void* memblock) { free(memblock); }
+
+static size_t GetPageSize() {
+#if defined(OS_LINUX) || defined(_SC_PAGESIZE)
+  long v = sysconf(_SC_PAGESIZE);
+  if (v >= 1024) {
+    return static_cast<size_t>(v);
+  }
+#endif
+  // Default assume 4KB
+  return 4U * 1024U;
+}
+
+const size_t kPageSize = GetPageSize();
+
+void SetCpuPriority(ThreadId id, CpuPriority priority) {
+#ifdef OS_LINUX
+  sched_param param;
+  param.sched_priority = 0;
+  switch (priority) {
+    case CpuPriority::kHigh:
+      sched_setscheduler(id, SCHED_OTHER, &param);
+      setpriority(PRIO_PROCESS, id, -20);
+      break;
+    case CpuPriority::kNormal:
+      sched_setscheduler(id, SCHED_OTHER, &param);
+      setpriority(PRIO_PROCESS, id, 0);
+      break;
+    case CpuPriority::kLow:
+      sched_setscheduler(id, SCHED_OTHER, &param);
+      setpriority(PRIO_PROCESS, id, 19);
+      break;
+    case CpuPriority::kIdle:
+      sched_setscheduler(id, SCHED_IDLE, &param);
+      break;
+    default:
+      assert(false);
+  }
+#else
+  (void)id;
+  (void)priority;
+#endif
+}
+
+int64_t GetProcessID() { return getpid(); }
+
+bool GenerateRfcUuid(std::string* output) {
+  output->clear();
+  std::ifstream f("/proc/sys/kernel/random/uuid");
+  std::getline(f, /*&*/ *output);
+  if (output->size() == 36) {
+    return true;
+  } else {
+    output->clear();
+    return false;
+  }
+}
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/port_posix.h b/src/rocksdb/port/port_posix.h
new file mode 100644
index 000000000..ec6aa281d
--- /dev/null
+++ b/src/rocksdb/port/port_posix.h
@@ -0,0 +1,241 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// See port_example.h for documentation for the following types/functions.
+
+#pragma once
+
+#include <thread>
+
+#include "rocksdb/options.h"
+#include "rocksdb/rocksdb_namespace.h"
+
+// size_t printf formatting named in the manner of C99 standard formatting
+// strings such as PRIu64
+// in fact, we could use that one
+#define ROCKSDB_PRIszt "zu"
+
+#define __declspec(S)
+
+#undef PLATFORM_IS_LITTLE_ENDIAN
+#if defined(OS_MACOSX)
+#include <machine/endian.h>
+#if defined(__DARWIN_LITTLE_ENDIAN) && defined(__DARWIN_BYTE_ORDER)
+#define PLATFORM_IS_LITTLE_ENDIAN \
+  (__DARWIN_BYTE_ORDER == __DARWIN_LITTLE_ENDIAN)
+#endif
+#elif defined(OS_SOLARIS)
+#include <sys/isa_defs.h>
+#ifdef _LITTLE_ENDIAN
+#define PLATFORM_IS_LITTLE_ENDIAN true
+#else
+#define PLATFORM_IS_LITTLE_ENDIAN false
+#endif
+#include <alloca.h>
+#elif defined(OS_AIX)
+#include <arpa/nameser_compat.h>
+#include <sys/types.h>
+#define PLATFORM_IS_LITTLE_ENDIAN (BYTE_ORDER == LITTLE_ENDIAN)
+#include <alloca.h>
+#elif defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || \
+    defined(OS_DRAGONFLYBSD) || defined(OS_ANDROID)
+#include <sys/endian.h>
+#include <sys/types.h>
+#define PLATFORM_IS_LITTLE_ENDIAN (_BYTE_ORDER == _LITTLE_ENDIAN)
+#else
+#include <endian.h>
+#endif
+#include <pthread.h>
+#include <stdint.h>
+#include <string.h>
+
+#include <limits>
+#include <string>
+
+#ifndef PLATFORM_IS_LITTLE_ENDIAN
+#define PLATFORM_IS_LITTLE_ENDIAN (__BYTE_ORDER == __LITTLE_ENDIAN)
+#endif
+
+#if defined(OS_MACOSX) || defined(OS_SOLARIS) || defined(OS_FREEBSD) ||      \
+    defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLYBSD) || \
+    defined(OS_ANDROID) || defined(CYGWIN) || defined(OS_AIX)
+// Use fread/fwrite/fflush on platforms without _unlocked variants
+#define fread_unlocked fread
+#define fwrite_unlocked fwrite
+#define fflush_unlocked fflush
+#endif
+
+#if defined(OS_MACOSX) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || \
+    defined(OS_DRAGONFLYBSD)
+// Use fsync() on platforms without fdatasync()
+#define fdatasync fsync
+#endif
+
+#if defined(OS_ANDROID) && __ANDROID_API__ < 9
+// fdatasync() was only introduced in API level 9 on Android. Use fsync()
+// when targeting older platforms.
+#define fdatasync fsync
+#endif
+
+namespace ROCKSDB_NAMESPACE {
+
+extern const bool kDefaultToAdaptiveMutex;
+
+namespace port {
+constexpr bool kLittleEndian = PLATFORM_IS_LITTLE_ENDIAN;
+#undef PLATFORM_IS_LITTLE_ENDIAN
+
+class CondVar;
+
+class Mutex {
+ public:
+  static const char* kName() { return "pthread_mutex_t"; }
+
+  explicit Mutex(bool adaptive = kDefaultToAdaptiveMutex);
+  // No copying
+  Mutex(const Mutex&) = delete;
+  void operator=(const Mutex&) = delete;
+
+  ~Mutex();
+
+  void Lock();
+  void Unlock();
+
+  bool TryLock();
+
+  // this will assert if the mutex is not locked
+  // it does NOT verify that mutex is held by a calling thread
+  void AssertHeld();
+
+  // Also implement std Lockable
+  inline void lock() { Lock(); }
+  inline void unlock() { Unlock(); }
+  inline bool try_lock() { return TryLock(); }
+
+ private:
+  friend class CondVar;
+  pthread_mutex_t mu_;
+#ifndef NDEBUG
+  bool locked_ = false;
+#endif
+};
+
+class RWMutex {
+ public:
+  RWMutex();
+  // No copying allowed
+  RWMutex(const RWMutex&) = delete;
+  void operator=(const RWMutex&) = delete;
+
+  ~RWMutex();
+
+  void ReadLock();
+  void WriteLock();
+  void ReadUnlock();
+  void WriteUnlock();
+  void AssertHeld() {}
+
+ private:
+  pthread_rwlock_t mu_;  // the underlying platform mutex
+};
+
+class CondVar {
+ public:
+  explicit CondVar(Mutex* mu);
+  ~CondVar();
+  void Wait();
+  // Timed condition wait.  Returns true if timeout occurred.
+  bool TimedWait(uint64_t abs_time_us);
+  void Signal();
+  void SignalAll();
+
+ private:
+  pthread_cond_t cv_;
+  Mutex* mu_;
+};
+
+using Thread = std::thread;
+
+static inline void AsmVolatilePause() {
+#if defined(__i386__) || defined(__x86_64__)
+  asm volatile("pause");
+#elif defined(__aarch64__)
+  asm volatile("isb");
+#elif defined(__powerpc64__)
+  asm volatile("or 27,27,27");
+#endif
+  // it's okay for other platforms to be no-ops
+}
+
+// Returns -1 if not available on this platform
+extern int PhysicalCoreID();
+
+using OnceType = pthread_once_t;
+#define LEVELDB_ONCE_INIT PTHREAD_ONCE_INIT
+extern void InitOnce(OnceType* once, void (*initializer)());
+
+#ifndef CACHE_LINE_SIZE
+// To test behavior with non-native cache line size, e.g. for
+// Bloom filters, set TEST_CACHE_LINE_SIZE to the desired test size.
+// This disables ALIGN_AS to keep it from failing compilation.
+#ifdef TEST_CACHE_LINE_SIZE
+#define CACHE_LINE_SIZE TEST_CACHE_LINE_SIZE
+#define ALIGN_AS(n) /*empty*/
+#else
+#if defined(__s390__)
+#if defined(__GNUC__) && __GNUC__ < 7
+#define CACHE_LINE_SIZE 64U
+#else
+#define CACHE_LINE_SIZE 256U
+#endif
+#elif defined(__powerpc__) || defined(__aarch64__)
+#define CACHE_LINE_SIZE 128U
+#else
+#define CACHE_LINE_SIZE 64U
+#endif
+#define ALIGN_AS(n) alignas(n)
+#endif
+#endif
+
+static_assert((CACHE_LINE_SIZE & (CACHE_LINE_SIZE - 1)) == 0,
+              "Cache line size must be a power of 2 number of bytes");
+
+extern void* cacheline_aligned_alloc(size_t size);
+
+extern void cacheline_aligned_free(void* memblock);
+
+#if defined(__aarch64__)
+//  __builtin_prefetch(..., 1) turns into a prefetch into prfm pldl3keep. On
+// arm64 we want this as close to the core as possible to turn it into a
+// L1 prefetech unless locality == 0 in which case it will be turned into a
+// non-temporal prefetch
+#define PREFETCH(addr, rw, locality) \
+  __builtin_prefetch(addr, rw, locality >= 1 ? 3 : locality)
+#else
+#define PREFETCH(addr, rw, locality) __builtin_prefetch(addr, rw, locality)
+#endif
+
+extern void Crash(const std::string& srcfile, int srcline);
+
+extern int GetMaxOpenFiles();
+
+extern const size_t kPageSize;
+
+using ThreadId = pid_t;
+
+extern void SetCpuPriority(ThreadId id, CpuPriority priority);
+
+int64_t GetProcessID();
+
+// Uses platform APIs to generate a 36-character RFC-4122 UUID. Returns
+// true on success or false on failure.
+bool GenerateRfcUuid(std::string* output);
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/stack_trace.cc b/src/rocksdb/port/stack_trace.cc
new file mode 100644
index 000000000..ef7144947
--- /dev/null
+++ b/src/rocksdb/port/stack_trace.cc
@@ -0,0 +1,202 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+#include "port/stack_trace.h"
+
+#if defined(ROCKSDB_LITE) ||                                                  \
+    !(defined(ROCKSDB_BACKTRACE) || defined(OS_MACOSX)) || defined(CYGWIN) || \
+    defined(OS_SOLARIS) || defined(OS_WIN)
+
+// noop
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+void InstallStackTraceHandler() {}
+void PrintStack(int /*first_frames_to_skip*/) {}
+void PrintAndFreeStack(void* /*callstack*/, int /*num_frames*/) {}
+void* SaveStack(int* /*num_frames*/, int /*first_frames_to_skip*/) {
+  return nullptr;
+}
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#else
+
+#include <cxxabi.h>
+#include <execinfo.h>
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#if defined(OS_FREEBSD)
+#include <sys/sysctl.h>
+#endif
+#ifdef OS_LINUX
+#include <sys/prctl.h>
+#endif
+
+#include "port/lang.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+namespace {
+
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_GNU_KFREEBSD)
+const char* GetExecutableName() {
+  static char name[1024];
+
+#if !defined(OS_FREEBSD)
+  char link[1024];
+  snprintf(link, sizeof(link), "/proc/%d/exe", getpid());
+  auto read = readlink(link, name, sizeof(name) - 1);
+  if (-1 == read) {
+    return nullptr;
+  } else {
+    name[read] = 0;
+    return name;
+  }
+#else
+  int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
+  size_t namesz = sizeof(name);
+
+  auto ret = sysctl(mib, 4, name, &namesz, nullptr, 0);
+  if (-1 == ret) {
+    return nullptr;
+  } else {
+    return name;
+  }
+#endif
+}
+
+void PrintStackTraceLine(const char* symbol, void* frame) {
+  static const char* executable = GetExecutableName();
+  if (symbol) {
+    fprintf(stderr, "%s ", symbol);
+  }
+  if (executable) {
+    // out source to addr2line, for the address translation
+    const int kLineMax = 256;
+    char cmd[kLineMax];
+    snprintf(cmd, kLineMax, "addr2line %p -e %s -f -C 2>&1", frame, executable);
+    auto f = popen(cmd, "r");
+    if (f) {
+      char line[kLineMax];
+      while (fgets(line, sizeof(line), f)) {
+        line[strlen(line) - 1] = 0;  // remove newline
+        fprintf(stderr, "%s\t", line);
+      }
+      pclose(f);
+    }
+  } else {
+    fprintf(stderr, " %p", frame);
+  }
+
+  fprintf(stderr, "\n");
+}
+#elif defined(OS_MACOSX)
+
+void PrintStackTraceLine(const char* symbol, void* frame) {
+  static int pid = getpid();
+  // out source to atos, for the address translation
+  const int kLineMax = 256;
+  char cmd[kLineMax];
+  snprintf(cmd, kLineMax, "xcrun atos %p -p %d  2>&1", frame, pid);
+  auto f = popen(cmd, "r");
+  if (f) {
+    char line[kLineMax];
+    while (fgets(line, sizeof(line), f)) {
+      line[strlen(line) - 1] = 0;  // remove newline
+      fprintf(stderr, "%s\t", line);
+    }
+    pclose(f);
+  } else if (symbol) {
+    fprintf(stderr, "%s ", symbol);
+  }
+
+  fprintf(stderr, "\n");
+}
+
+#endif
+
+}  // namespace
+
+void PrintStack(void* frames[], int num_frames) {
+  auto symbols = backtrace_symbols(frames, num_frames);
+
+  for (int i = 0; i < num_frames; ++i) {
+    fprintf(stderr, "#%-2d  ", i);
+    PrintStackTraceLine((symbols != nullptr) ? symbols[i] : nullptr, frames[i]);
+  }
+  free(symbols);
+}
+
+void PrintStack(int first_frames_to_skip) {
+  const int kMaxFrames = 100;
+  void* frames[kMaxFrames];
+
+  auto num_frames = backtrace(frames, kMaxFrames);
+  PrintStack(&frames[first_frames_to_skip], num_frames - first_frames_to_skip);
+}
+
+void PrintAndFreeStack(void* callstack, int num_frames) {
+  PrintStack(static_cast<void**>(callstack), num_frames);
+  free(callstack);
+}
+
+void* SaveStack(int* num_frames, int first_frames_to_skip) {
+  const int kMaxFrames = 100;
+  void* frames[kMaxFrames];
+
+  auto count = backtrace(frames, kMaxFrames);
+  *num_frames = count - first_frames_to_skip;
+  void* callstack = malloc(sizeof(void*) * *num_frames);
+  memcpy(callstack, &frames[first_frames_to_skip], sizeof(void*) * *num_frames);
+  return callstack;
+}
+
+static void StackTraceHandler(int sig) {
+  // reset to default handler
+  signal(sig, SIG_DFL);
+  fprintf(stderr, "Received signal %d (%s)\n", sig, strsignal(sig));
+  // skip the top three signal handler related frames
+  PrintStack(3);
+
+  // Efforts to fix or suppress TSAN warnings "signal-unsafe call inside of
+  // a signal" have failed, so just warn the user about them.
+#ifdef __SANITIZE_THREAD__
+  fprintf(stderr,
+          "==> NOTE: any above warnings about \"signal-unsafe call\" are\n"
+          "==> ignorable, as they are expected when generating a stack\n"
+          "==> trace because of a signal under TSAN. Consider why the\n"
+          "==> signal was generated to begin with, and the stack trace\n"
+          "==> in the TSAN warning can be useful for that. (The stack\n"
+          "==> trace printed by the signal handler is likely obscured\n"
+          "==> by TSAN output.)\n");
+#endif
+
+  // re-signal to default handler (so we still get core dump if needed...)
+  raise(sig);
+}
+
+void InstallStackTraceHandler() {
+  // just use the plain old signal as it's simple and sufficient
+  // for this use case
+  signal(SIGILL, StackTraceHandler);
+  signal(SIGSEGV, StackTraceHandler);
+  signal(SIGBUS, StackTraceHandler);
+  signal(SIGABRT, StackTraceHandler);
+  // Allow ouside debugger to attach, even with Yama security restrictions
+#ifdef PR_SET_PTRACER_ANY
+  (void)prctl(PR_SET_PTRACER, PR_SET_PTRACER_ANY, 0, 0, 0);
+#endif
+}
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/stack_trace.h b/src/rocksdb/port/stack_trace.h
new file mode 100644
index 000000000..5b3bf9320
--- /dev/null
+++ b/src/rocksdb/port/stack_trace.h
@@ -0,0 +1,31 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+//
+#pragma once
+
+#include "rocksdb/rocksdb_namespace.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+// Install a signal handler to print callstack on the following signals:
+// SIGILL SIGSEGV SIGBUS SIGABRT
+// And also (Linux ony for now) overrides security settings to allow outside
+// processes to attach to this one as a debugger. ONLY USE FOR NON-SECURITY
+// CRITICAL PROCESSES such as unit tests or benchmarking tools.
+// Currently supports only some POSIX implementations. No-op otherwise.
+void InstallStackTraceHandler();
+
+// Prints stack, skips skip_first_frames frames
+void PrintStack(int first_frames_to_skip = 0);
+
+// Prints the given callstack
+void PrintAndFreeStack(void* callstack, int num_frames);
+
+// Save the current callstack
+void* SaveStack(int* num_frame, int first_frames_to_skip = 0);
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/sys_time.h b/src/rocksdb/port/sys_time.h
new file mode 100644
index 000000000..f2137526b
--- /dev/null
+++ b/src/rocksdb/port/sys_time.h
@@ -0,0 +1,63 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+// This file is a portable substitute for sys/time.h which does not exist on
+// Windows
+
+#pragma once
+
+#include "rocksdb/rocksdb_namespace.h"
+
+#if defined(OS_WIN) && (defined(_MSC_VER) || defined(__MINGW32__))
+
+#include <time.h>
+
+namespace ROCKSDB_NAMESPACE {
+
+namespace port {
+
+struct TimeVal {
+  long tv_sec;
+  long tv_usec;
+};
+
+void GetTimeOfDay(TimeVal* tv, struct timezone* tz);
+
+inline struct tm* LocalTimeR(const time_t* timep, struct tm* result) {
+  errno_t ret = localtime_s(result, timep);
+  return (ret == 0) ? result : NULL;
+}
+
+}  // namespace port
+
+}  // namespace ROCKSDB_NAMESPACE
+
+#else
+#include <sys/time.h>
+#include <time.h>
+
+namespace ROCKSDB_NAMESPACE {
+
+namespace port {
+
+using TimeVal = struct timeval;
+
+inline void GetTimeOfDay(TimeVal* tv, struct timezone* tz) {
+  gettimeofday(tv, tz);
+}
+
+inline struct tm* LocalTimeR(const time_t* timep, struct tm* result) {
+  return localtime_r(timep, result);
+}
+
+}  // namespace port
+
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/util_logger.h b/src/rocksdb/port/util_logger.h
new file mode 100644
index 000000000..ce7e3a941
--- /dev/null
+++ b/src/rocksdb/port/util_logger.h
@@ -0,0 +1,18 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#pragma once
+
+// Include the appropriate platform specific file below.  If you are
+// porting to a new platform, see "port_example.h" for documentation
+// of what the new port_<platform>.h file must provide.
+
+#if defined(OS_WIN)
+#include "port/win/win_logger.h"
+#endif
diff --git a/src/rocksdb/port/win/env_default.cc b/src/rocksdb/port/win/env_default.cc
new file mode 100644
index 000000000..48853f26e
--- /dev/null
+++ b/src/rocksdb/port/win/env_default.cc
@@ -0,0 +1,45 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+
+#include <mutex>
+
+#include "port/win/env_win.h"
+#include "rocksdb/env.h"
+#include "test_util/sync_point.h"
+#include "util/compression_context_cache.h"
+#include "util/thread_local.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+// We choose not to destroy the env because joining the threads from the
+// system loader
+//    which destroys the statics (same as from DLLMain) creates a system loader
+//    dead-lock.
+//    in this manner any remaining threads are terminated OK.
+namespace {
+std::once_flag winenv_once_flag;
+Env* envptr;
+};  // namespace
+}  // namespace port
+
+Env* Env::Default() {
+  ThreadLocalPtr::InitSingletons();
+  CompressionContextCache::InitSingleton();
+  INIT_SYNC_POINT_SINGLETONS();
+  std::call_once(port::winenv_once_flag,
+                 []() { port::envptr = new port::WinEnv(); });
+  return port::envptr;
+}
+
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/win/env_win.cc b/src/rocksdb/port/win/env_win.cc
new file mode 100644
index 000000000..2262eb59c
--- /dev/null
+++ b/src/rocksdb/port/win/env_win.cc
@@ -0,0 +1,1437 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+
+#include "port/win/env_win.h"
+
+#include <direct.h>  // _rmdir, _mkdir, _getcwd
+#include <errno.h>
+#include <io.h>   // _access
+#include <rpc.h>  // for uuid generation
+#include <shlwapi.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <windows.h>
+#include <winioctl.h>
+
+#include <algorithm>
+#include <ctime>
+#include <thread>
+
+#include "monitoring/iostats_context_imp.h"
+#include "monitoring/thread_status_updater.h"
+#include "monitoring/thread_status_util.h"
+#include "port/lang.h"
+#include "port/port.h"
+#include "port/port_dirent.h"
+#include "port/win/io_win.h"
+#include "port/win/win_logger.h"
+#include "rocksdb/env.h"
+#include "rocksdb/slice.h"
+#include "strsafe.h"
+#include "util/string_util.h"
+
+// Undefine the functions  windows might use (again)...
+#undef GetCurrentTime
+#undef DeleteFile
+#undef LoadLibrary
+
+namespace ROCKSDB_NAMESPACE {
+
+ThreadStatusUpdater* CreateThreadStatusUpdater() {
+  return new ThreadStatusUpdater();
+}
+
+namespace {
+
+// Sector size used when physical sector size cannot be obtained from device.
+static const size_t kSectorSize = 512;
+
+// RAII helpers for HANDLEs
+const auto CloseHandleFunc = [](HANDLE h) { ::CloseHandle(h); };
+using UniqueCloseHandlePtr = std::unique_ptr<void, decltype(CloseHandleFunc)>;
+
+const auto FindCloseFunc = [](HANDLE h) { ::FindClose(h); };
+using UniqueFindClosePtr = std::unique_ptr<void, decltype(FindCloseFunc)>;
+
+void WinthreadCall(const char* label, std::error_code result) {
+  if (0 != result.value()) {
+    fprintf(stderr, "Winthread %s: %s\n", label,
+            errnoStr(result.value()).c_str());
+    abort();
+  }
+}
+
+}  // namespace
+
+namespace port {
+WinClock::WinClock()
+    : perf_counter_frequency_(0),
+      nano_seconds_per_period_(0),
+      GetSystemTimePreciseAsFileTime_(NULL) {
+  {
+    LARGE_INTEGER qpf;
+    BOOL ret __attribute__((__unused__));
+    ret = QueryPerformanceFrequency(&qpf);
+    assert(ret == TRUE);
+    perf_counter_frequency_ = qpf.QuadPart;
+
+    if (std::nano::den % perf_counter_frequency_ == 0) {
+      nano_seconds_per_period_ = std::nano::den / perf_counter_frequency_;
+    }
+  }
+
+  HMODULE module = GetModuleHandle("kernel32.dll");
+  if (module != NULL) {
+    GetSystemTimePreciseAsFileTime_ = (FnGetSystemTimePreciseAsFileTime)(
+        void*)GetProcAddress(module, "GetSystemTimePreciseAsFileTime");
+  }
+}
+
+void WinClock::SleepForMicroseconds(int micros) {
+  std::this_thread::sleep_for(std::chrono::microseconds(micros));
+}
+
+std::string WinClock::TimeToString(uint64_t secondsSince1970) {
+  std::string result;
+
+  const time_t seconds = secondsSince1970;
+  const int maxsize = 64;
+
+  struct tm t;
+  errno_t ret = localtime_s(&t, &seconds);
+
+  if (ret) {
+    result = std::to_string(seconds);
+  } else {
+    result.resize(maxsize);
+    char* p = &result[0];
+
+    int len =
+        snprintf(p, maxsize, "%04d/%02d/%02d-%02d:%02d:%02d ", t.tm_year + 1900,
+                 t.tm_mon + 1, t.tm_mday, t.tm_hour, t.tm_min, t.tm_sec);
+    assert(len > 0);
+
+    result.resize(len);
+  }
+
+  return result;
+}
+
+uint64_t WinClock::NowMicros() {
+  if (GetSystemTimePreciseAsFileTime_ != NULL) {
+    // all std::chrono clocks on windows proved to return
+    // values that may repeat that is not good enough for some uses.
+    const int64_t c_UnixEpochStartTicks = 116444736000000000LL;
+    const int64_t c_FtToMicroSec = 10;
+
+    // This interface needs to return system time and not
+    // just any microseconds because it is often used as an argument
+    // to TimedWait() on condition variable
+    FILETIME ftSystemTime;
+    GetSystemTimePreciseAsFileTime_(&ftSystemTime);
+
+    LARGE_INTEGER li;
+    li.LowPart = ftSystemTime.dwLowDateTime;
+    li.HighPart = ftSystemTime.dwHighDateTime;
+    // Subtract unix epoch start
+    li.QuadPart -= c_UnixEpochStartTicks;
+    // Convert to microsecs
+    li.QuadPart /= c_FtToMicroSec;
+    return li.QuadPart;
+  }
+  return std::chrono::duration_cast<std::chrono::microseconds>(
+             std::chrono::system_clock::now().time_since_epoch())
+      .count();
+}
+
+uint64_t WinClock::NowNanos() {
+  if (nano_seconds_per_period_ != 0) {
+    // all std::chrono clocks on windows have the same resolution that is only
+    // good enough for microseconds but not nanoseconds
+    // On Windows 8 and Windows 2012 Server
+    // GetSystemTimePreciseAsFileTime(&current_time) can be used
+    LARGE_INTEGER li;
+    QueryPerformanceCounter(&li);
+    // Convert performance counter to nanoseconds by precomputed ratio.
+    // Directly multiply nano::den with li.QuadPart causes overflow.
+    // Only do this when nano::den is divisible by perf_counter_frequency_,
+    // which most likely is the case in reality. If it's not, fall back to
+    // high_resolution_clock, which may be less precise under old compilers.
+    li.QuadPart *= nano_seconds_per_period_;
+    return li.QuadPart;
+  }
+  return std::chrono::duration_cast<std::chrono::nanoseconds>(
+             std::chrono::high_resolution_clock::now().time_since_epoch())
+      .count();
+}
+
+Status WinClock::GetCurrentTime(int64_t* unix_time) {
+  time_t time = std::time(nullptr);
+  if (time == (time_t)(-1)) {
+    return Status::NotSupported("Failed to get time");
+  }
+
+  *unix_time = time;
+  return Status::OK();
+}
+
+WinFileSystem::WinFileSystem(const std::shared_ptr<SystemClock>& clock)
+    : clock_(clock), page_size_(4 * 1024), allocation_granularity_(page_size_) {
+  SYSTEM_INFO sinfo;
+  GetSystemInfo(&sinfo);
+
+  page_size_ = sinfo.dwPageSize;
+  allocation_granularity_ = sinfo.dwAllocationGranularity;
+}
+
+const std::shared_ptr<WinFileSystem>& WinFileSystem::Default() {
+  STATIC_AVOID_DESTRUCTION(std::shared_ptr<WinFileSystem>, fs)
+  (std::make_shared<WinFileSystem>(WinClock::Default()));
+  return fs;
+}
+
+WinEnvIO::WinEnvIO(Env* hosted_env) : hosted_env_(hosted_env) {}
+
+WinEnvIO::~WinEnvIO() {}
+
+IOStatus WinFileSystem::DeleteFile(const std::string& fname,
+                                   const IOOptions& /*options*/,
+                                   IODebugContext* /*dbg*/) {
+  IOStatus result;
+
+  BOOL ret = RX_DeleteFile(RX_FN(fname).c_str());
+
+  if (!ret) {
+    auto lastError = GetLastError();
+    result = IOErrorFromWindowsError("Failed to delete: " + fname, lastError);
+  }
+
+  return result;
+}
+
+IOStatus WinFileSystem::Truncate(const std::string& fname, size_t size,
+                                 const IOOptions& /*options*/,
+                                 IODebugContext* /*dbg*/) {
+  IOStatus s;
+  int result = ROCKSDB_NAMESPACE::port::Truncate(fname, size);
+  if (result != 0) {
+    s = IOError("Failed to truncate: " + fname, errno);
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::NewSequentialFile(
+    const std::string& fname, const FileOptions& options,
+    std::unique_ptr<FSSequentialFile>* result, IODebugContext* /*dbg*/) {
+  IOStatus s;
+
+  result->reset();
+
+  // Corruption test needs to rename and delete files of these kind
+  // while they are still open with another handle. For that reason we
+  // allow share_write and delete(allows rename).
+  HANDLE hFile = INVALID_HANDLE_VALUE;
+
+  DWORD fileFlags = FILE_ATTRIBUTE_READONLY;
+
+  if (options.use_direct_reads && !options.use_mmap_reads) {
+    fileFlags |= FILE_FLAG_NO_BUFFERING;
+  }
+
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile = RX_CreateFile(
+        RX_FN(fname).c_str(), GENERIC_READ,
+        FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL,
+        OPEN_EXISTING,  // Original fopen mode is "rb"
+        fileFlags, NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("Failed to open NewSequentialFile" + fname,
+                                lastError);
+  } else {
+    result->reset(new WinSequentialFile(fname, hFile, options));
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::NewRandomAccessFile(
+    const std::string& fname, const FileOptions& options,
+    std::unique_ptr<FSRandomAccessFile>* result, IODebugContext* dbg) {
+  result->reset();
+  IOStatus s;
+
+  // Open the file for read-only random access
+  // Random access is to disable read-ahead as the system reads too much data
+  DWORD fileFlags = FILE_ATTRIBUTE_READONLY;
+
+  if (options.use_direct_reads && !options.use_mmap_reads) {
+    fileFlags |= FILE_FLAG_NO_BUFFERING;
+  } else {
+    fileFlags |= FILE_FLAG_RANDOM_ACCESS;
+  }
+
+  /// Shared access is necessary for corruption test to pass
+  // almost all tests would work with a possible exception of fault_injection
+  HANDLE hFile = 0;
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile =
+        RX_CreateFile(RX_FN(fname).c_str(), GENERIC_READ,
+                      FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
+                      NULL, OPEN_EXISTING, fileFlags, NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError(
+        "NewRandomAccessFile failed to Create/Open: " + fname, lastError);
+  }
+
+  UniqueCloseHandlePtr fileGuard(hFile, CloseHandleFunc);
+
+  // CAUTION! This will map the entire file into the process address space.
+  // Not recommended for 32-bit platforms.
+  if (options.use_mmap_reads) {
+    uint64_t fileSize;
+
+    s = GetFileSize(fname, IOOptions(), &fileSize, dbg);
+
+    if (s.ok()) {
+      // Will not map empty files
+      if (fileSize == 0) {
+        return IOError("NewRandomAccessFile failed to map empty file: " + fname,
+                       EINVAL);
+      }
+
+      HANDLE hMap = RX_CreateFileMapping(hFile, NULL, PAGE_READONLY,
+                                         0,  // At its present length
+                                         0,
+                                         NULL);  // Mapping name
+
+      if (!hMap) {
+        auto lastError = GetLastError();
+        return IOErrorFromWindowsError(
+            "Failed to create file mapping for NewRandomAccessFile: " + fname,
+            lastError);
+      }
+
+      UniqueCloseHandlePtr mapGuard(hMap, CloseHandleFunc);
+
+      const void* mapped_region =
+          MapViewOfFileEx(hMap, FILE_MAP_READ,
+                          0,  // High DWORD of access start
+                          0,  // Low DWORD
+                          static_cast<SIZE_T>(fileSize),
+                          NULL);  // Let the OS choose the mapping
+
+      if (!mapped_region) {
+        auto lastError = GetLastError();
+        return IOErrorFromWindowsError(
+            "Failed to MapViewOfFile for NewRandomAccessFile: " + fname,
+            lastError);
+      }
+
+      result->reset(new WinMmapReadableFile(fname, hFile, hMap, mapped_region,
+                                            static_cast<size_t>(fileSize)));
+
+      mapGuard.release();
+      fileGuard.release();
+    }
+  } else {
+    result->reset(new WinRandomAccessFile(fname, hFile, page_size_, options));
+    fileGuard.release();
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::OpenWritableFile(
+    const std::string& fname, const FileOptions& options,
+    std::unique_ptr<FSWritableFile>* result, bool reopen) {
+  const size_t c_BufferCapacity = 64 * 1024;
+
+  EnvOptions local_options(options);
+
+  result->reset();
+  IOStatus s;
+
+  DWORD fileFlags = FILE_ATTRIBUTE_NORMAL;
+
+  if (local_options.use_direct_writes && !local_options.use_mmap_writes) {
+    fileFlags = FILE_FLAG_NO_BUFFERING | FILE_FLAG_WRITE_THROUGH;
+  }
+
+  // Desired access. We are want to write only here but if we want to memory
+  // map
+  // the file then there is no write only mode so we have to create it
+  // Read/Write
+  // However, MapViewOfFile specifies only Write only
+  DWORD desired_access = GENERIC_WRITE;
+  DWORD shared_mode = FILE_SHARE_READ;
+
+  if (local_options.use_mmap_writes) {
+    desired_access |= GENERIC_READ;
+  } else {
+    // Adding this solely for tests to pass (fault_injection_test,
+    // wal_manager_test).
+    shared_mode |= (FILE_SHARE_WRITE | FILE_SHARE_DELETE);
+  }
+
+  // This will always truncate the file
+  DWORD creation_disposition = CREATE_ALWAYS;
+  if (reopen) {
+    creation_disposition = OPEN_ALWAYS;
+  }
+
+  HANDLE hFile = 0;
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile = RX_CreateFile(
+        RX_FN(fname).c_str(),
+        desired_access,  // Access desired
+        shared_mode,
+        NULL,  // Security attributes
+        // Posix env says (reopen) ? (O_CREATE | O_APPEND) : O_CREAT | O_TRUNC
+        creation_disposition,
+        fileFlags,  // Flags
+        NULL);      // Template File
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError(
+        "Failed to create a NewWritableFile: " + fname, lastError);
+  }
+
+  // We will start writing at the end, appending
+  if (reopen) {
+    LARGE_INTEGER zero_move;
+    zero_move.QuadPart = 0;
+    BOOL ret = SetFilePointerEx(hFile, zero_move, NULL, FILE_END);
+    if (!ret) {
+      auto lastError = GetLastError();
+      return IOErrorFromWindowsError(
+          "Failed to create a ReopenWritableFile move to the end: " + fname,
+          lastError);
+    }
+  }
+
+  if (options.use_mmap_writes) {
+    // We usually do not use mmmapping on SSD and thus we pass memory
+    // page_size
+    result->reset(new WinMmapFile(fname, hFile, page_size_,
+                                  allocation_granularity_, local_options));
+  } else {
+    // Here we want the buffer allocation to be aligned by the SSD page size
+    // and to be a multiple of it
+    result->reset(new WinWritableFile(fname, hFile, GetPageSize(),
+                                      c_BufferCapacity, local_options));
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::NewWritableFile(const std::string& fname,
+                                        const FileOptions& options,
+                                        std::unique_ptr<FSWritableFile>* result,
+                                        IODebugContext* /*dbg*/) {
+  return OpenWritableFile(fname, options, result, false);
+}
+
+IOStatus WinFileSystem::ReopenWritableFile(
+    const std::string& fname, const FileOptions& options,
+    std::unique_ptr<FSWritableFile>* result, IODebugContext* /*dbg*/) {
+  return OpenWritableFile(fname, options, result, true);
+}
+
+IOStatus WinFileSystem::NewRandomRWFile(const std::string& fname,
+                                        const FileOptions& options,
+                                        std::unique_ptr<FSRandomRWFile>* result,
+                                        IODebugContext* /*dbg*/) {
+  IOStatus s;
+
+  // Open the file for read-only random access
+  // Random access is to disable read-ahead as the system reads too much data
+  DWORD desired_access = GENERIC_READ | GENERIC_WRITE;
+  DWORD shared_mode = FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE;
+  DWORD creation_disposition = OPEN_EXISTING;  // Fail if file does not exist
+  DWORD file_flags = FILE_FLAG_RANDOM_ACCESS;
+
+  if (options.use_direct_reads && options.use_direct_writes) {
+    file_flags |= FILE_FLAG_NO_BUFFERING;
+  }
+
+  /// Shared access is necessary for corruption test to pass
+  // almost all tests would work with a possible exception of fault_injection
+  HANDLE hFile = 0;
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile = RX_CreateFile(RX_FN(fname).c_str(), desired_access, shared_mode,
+                          NULL,  // Security attributes
+                          creation_disposition, file_flags, NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError(
+        "NewRandomRWFile failed to Create/Open: " + fname, lastError);
+  }
+
+  UniqueCloseHandlePtr fileGuard(hFile, CloseHandleFunc);
+  result->reset(new WinRandomRWFile(fname, hFile, GetPageSize(), options));
+  fileGuard.release();
+
+  return s;
+}
+
+IOStatus WinFileSystem::NewMemoryMappedFileBuffer(
+    const std::string& fname, std::unique_ptr<MemoryMappedFileBuffer>* result) {
+  IOStatus s;
+  result->reset();
+
+  DWORD fileFlags = FILE_ATTRIBUTE_READONLY;
+
+  HANDLE hFile = INVALID_HANDLE_VALUE;
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile = RX_CreateFile(
+        RX_FN(fname).c_str(), GENERIC_READ | GENERIC_WRITE,
+        FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL,
+        OPEN_EXISTING,  // Open only if it exists
+        fileFlags, NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError(
+        "Failed to open NewMemoryMappedFileBuffer: " + fname, lastError);
+    return s;
+  }
+  UniqueCloseHandlePtr fileGuard(hFile, CloseHandleFunc);
+
+  uint64_t fileSize = 0;
+  s = GetFileSize(fname, IOOptions(), &fileSize, nullptr);
+  if (!s.ok()) {
+    return s;
+  }
+  // Will not map empty files
+  if (fileSize == 0) {
+    return IOStatus::NotSupported(
+        "NewMemoryMappedFileBuffer can not map zero length files: " + fname);
+  }
+
+  // size_t is 32-bit with 32-bit builds
+  if (fileSize > std::numeric_limits<size_t>::max()) {
+    return IOStatus::NotSupported(
+        "The specified file size does not fit into 32-bit memory addressing: " +
+        fname);
+  }
+
+  HANDLE hMap = RX_CreateFileMapping(hFile, NULL, PAGE_READWRITE,
+                                     0,  // Whole file at its present length
+                                     0,
+                                     NULL);  // Mapping name
+
+  if (!hMap) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError(
+        "Failed to create file mapping for: " + fname, lastError);
+  }
+  UniqueCloseHandlePtr mapGuard(hMap, CloseHandleFunc);
+
+  void* base = MapViewOfFileEx(hMap, FILE_MAP_WRITE,
+                               0,  // High DWORD of access start
+                               0,  // Low DWORD
+                               static_cast<SIZE_T>(fileSize),
+                               NULL);  // Let the OS choose the mapping
+
+  if (!base) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError(
+        "Failed to MapViewOfFile for NewMemoryMappedFileBuffer: " + fname,
+        lastError);
+  }
+
+  result->reset(new WinMemoryMappedBuffer(hFile, hMap, base,
+                                          static_cast<size_t>(fileSize)));
+
+  mapGuard.release();
+  fileGuard.release();
+
+  return s;
+}
+
+IOStatus WinFileSystem::NewDirectory(const std::string& name,
+                                     const IOOptions& /*options*/,
+                                     std::unique_ptr<FSDirectory>* result,
+                                     IODebugContext* /*dbg*/) {
+  IOStatus s;
+  // Must be nullptr on failure
+  result->reset();
+
+  if (!DirExists(name)) {
+    s = IOErrorFromWindowsError("open folder: " + name, ERROR_DIRECTORY);
+    return s;
+  }
+
+  HANDLE handle = INVALID_HANDLE_VALUE;
+  // 0 - for access means read metadata
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    handle = RX_CreateFile(
+        RX_FN(name).c_str(), 0,
+        FILE_SHARE_DELETE | FILE_SHARE_READ | FILE_SHARE_WRITE, NULL,
+        OPEN_EXISTING,
+        FILE_FLAG_BACKUP_SEMANTICS,  // make opening folders possible
+        NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == handle) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("open folder: " + name, lastError);
+    return s;
+  }
+
+  result->reset(new WinDirectory(name, handle));
+
+  return s;
+}
+
+IOStatus WinFileSystem::FileExists(const std::string& fname,
+                                   const IOOptions& /*opts*/,
+                                   IODebugContext* /*dbg*/) {
+  IOStatus s;
+  // TODO: This does not follow symbolic links at this point
+  // which is consistent with _access() impl on windows
+  // but can be added
+  WIN32_FILE_ATTRIBUTE_DATA attrs;
+  if (FALSE == RX_GetFileAttributesEx(RX_FN(fname).c_str(),
+                                      GetFileExInfoStandard, &attrs)) {
+    auto lastError = GetLastError();
+    switch (lastError) {
+      case ERROR_ACCESS_DENIED:
+      case ERROR_NOT_FOUND:
+      case ERROR_FILE_NOT_FOUND:
+      case ERROR_PATH_NOT_FOUND:
+        s = IOStatus::NotFound();
+        break;
+      default:
+        s = IOErrorFromWindowsError("Unexpected error for: " + fname,
+                                    lastError);
+        break;
+    }
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::GetChildren(const std::string& dir,
+                                    const IOOptions& /*opts*/,
+                                    std::vector<std::string>* result,
+                                    IODebugContext* /*dbg*/) {
+  IOStatus status;
+  result->clear();
+
+  RX_WIN32_FIND_DATA data;
+  memset(&data, 0, sizeof(data));
+  std::string pattern(dir);
+  pattern.append("\\").append("*");
+
+  HANDLE handle =
+      RX_FindFirstFileEx(RX_FN(pattern).c_str(),
+                         // Do not want alternative name
+                         FindExInfoBasic, &data, FindExSearchNameMatch,
+                         NULL,  // lpSearchFilter
+                         0);
+
+  if (handle == INVALID_HANDLE_VALUE) {
+    auto lastError = GetLastError();
+    switch (lastError) {
+      case ERROR_NOT_FOUND:
+      case ERROR_ACCESS_DENIED:
+      case ERROR_FILE_NOT_FOUND:
+      case ERROR_PATH_NOT_FOUND:
+        status = IOStatus::NotFound();
+        break;
+      default:
+        status = IOErrorFromWindowsError("Failed to GetChhildren for: " + dir,
+                                         lastError);
+    }
+    return status;
+  }
+
+  UniqueFindClosePtr fc(handle, FindCloseFunc);
+
+  // For safety
+  data.cFileName[MAX_PATH - 1] = 0;
+
+  while (true) {
+    // filter out '.' and '..' directory entries
+    // which appear only on some platforms
+    const bool ignore =
+        ((data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) != 0) &&
+        (RX_FNCMP(data.cFileName, ".") == 0 ||
+         RX_FNCMP(data.cFileName, "..") == 0);
+    if (!ignore) {
+      auto x = RX_FILESTRING(data.cFileName, RX_FNLEN(data.cFileName));
+      result->push_back(FN_TO_RX(x));
+    }
+
+    BOOL ret = -RX_FindNextFile(handle, &data);
+    // If the function fails the return value is zero
+    // and non-zero otherwise. Not TRUE or FALSE.
+    if (ret == FALSE) {
+      // Posix does not care why we stopped
+      break;
+    }
+    data.cFileName[MAX_PATH - 1] = 0;
+  }
+  return status;
+}
+
+IOStatus WinFileSystem::CreateDir(const std::string& name,
+                                  const IOOptions& /*opts*/,
+                                  IODebugContext* /*dbg*/) {
+  IOStatus result;
+  BOOL ret = RX_CreateDirectory(RX_FN(name).c_str(), NULL);
+  if (!ret) {
+    auto lastError = GetLastError();
+    result = IOErrorFromWindowsError("Failed to create a directory: " + name,
+                                     lastError);
+  }
+
+  return result;
+}
+
+IOStatus WinFileSystem::CreateDirIfMissing(const std::string& name,
+                                           const IOOptions& /*opts*/,
+                                           IODebugContext* /*dbg*/) {
+  IOStatus result;
+
+  if (DirExists(name)) {
+    return result;
+  }
+
+  BOOL ret = RX_CreateDirectory(RX_FN(name).c_str(), NULL);
+  if (!ret) {
+    auto lastError = GetLastError();
+    if (lastError != ERROR_ALREADY_EXISTS) {
+      result = IOErrorFromWindowsError("Failed to create a directory: " + name,
+                                       lastError);
+    } else {
+      result = IOStatus::IOError(name + ": exists but is not a directory");
+    }
+  }
+  return result;
+}
+
+IOStatus WinFileSystem::DeleteDir(const std::string& name,
+                                  const IOOptions& /*options*/,
+                                  IODebugContext* /*dbg*/) {
+  IOStatus result;
+  BOOL ret = RX_RemoveDirectory(RX_FN(name).c_str());
+  if (!ret) {
+    auto lastError = GetLastError();
+    result =
+        IOErrorFromWindowsError("Failed to remove dir: " + name, lastError);
+  }
+  return result;
+}
+
+IOStatus WinFileSystem::GetFileSize(const std::string& fname,
+                                    const IOOptions& /*opts*/, uint64_t* size,
+                                    IODebugContext* /*dbg*/) {
+  IOStatus s;
+
+  WIN32_FILE_ATTRIBUTE_DATA attrs;
+  if (RX_GetFileAttributesEx(RX_FN(fname).c_str(), GetFileExInfoStandard,
+                             &attrs)) {
+    ULARGE_INTEGER file_size;
+    file_size.HighPart = attrs.nFileSizeHigh;
+    file_size.LowPart = attrs.nFileSizeLow;
+    *size = file_size.QuadPart;
+  } else {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("Can not get size for: " + fname, lastError);
+  }
+  return s;
+}
+
+uint64_t WinFileSystem::FileTimeToUnixTime(const FILETIME& ftTime) {
+  const uint64_t c_FileTimePerSecond = 10000000U;
+  // UNIX epoch starts on 1970-01-01T00:00:00Z
+  // Windows FILETIME starts on 1601-01-01T00:00:00Z
+  // Therefore, we need to subtract the below number of seconds from
+  // the seconds that we obtain from FILETIME with an obvious loss of
+  // precision
+  const uint64_t c_SecondBeforeUnixEpoch = 11644473600U;
+
+  ULARGE_INTEGER li;
+  li.HighPart = ftTime.dwHighDateTime;
+  li.LowPart = ftTime.dwLowDateTime;
+
+  uint64_t result =
+      (li.QuadPart / c_FileTimePerSecond) - c_SecondBeforeUnixEpoch;
+  return result;
+}
+
+IOStatus WinFileSystem::GetFileModificationTime(const std::string& fname,
+                                                const IOOptions& /*opts*/,
+                                                uint64_t* file_mtime,
+                                                IODebugContext* /*dbg*/) {
+  IOStatus s;
+
+  WIN32_FILE_ATTRIBUTE_DATA attrs;
+  if (RX_GetFileAttributesEx(RX_FN(fname).c_str(), GetFileExInfoStandard,
+                             &attrs)) {
+    *file_mtime = FileTimeToUnixTime(attrs.ftLastWriteTime);
+  } else {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError(
+        "Can not get file modification time for: " + fname, lastError);
+    *file_mtime = 0;
+  }
+
+  return s;
+}
+
+IOStatus WinFileSystem::RenameFile(const std::string& src,
+                                   const std::string& target,
+                                   const IOOptions& /*opts*/,
+                                   IODebugContext* /*dbg*/) {
+  IOStatus result;
+
+  // rename() is not capable of replacing the existing file as on Linux
+  // so use OS API directly
+  if (!RX_MoveFileEx(RX_FN(src).c_str(), RX_FN(target).c_str(),
+                     MOVEFILE_REPLACE_EXISTING)) {
+    DWORD lastError = GetLastError();
+
+    std::string text("Failed to rename: ");
+    text.append(src).append(" to: ").append(target);
+
+    result = IOErrorFromWindowsError(text, lastError);
+  }
+
+  return result;
+}
+
+IOStatus WinFileSystem::LinkFile(const std::string& src,
+                                 const std::string& target,
+                                 const IOOptions& /*opts*/,
+                                 IODebugContext* /*dbg*/) {
+  IOStatus result;
+
+  if (!RX_CreateHardLink(RX_FN(target).c_str(), RX_FN(src).c_str(), NULL)) {
+    DWORD lastError = GetLastError();
+    if (lastError == ERROR_NOT_SAME_DEVICE) {
+      return IOStatus::NotSupported("No cross FS links allowed");
+    }
+
+    std::string text("Failed to link: ");
+    text.append(src).append(" to: ").append(target);
+
+    result = IOErrorFromWindowsError(text, lastError);
+  }
+
+  return result;
+}
+
+IOStatus WinFileSystem::NumFileLinks(const std::string& fname,
+                                     const IOOptions& /*opts*/, uint64_t* count,
+                                     IODebugContext* /*dbg*/) {
+  IOStatus s;
+  HANDLE handle =
+      RX_CreateFile(RX_FN(fname).c_str(), 0,
+                    FILE_SHARE_DELETE | FILE_SHARE_READ | FILE_SHARE_WRITE,
+                    NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
+
+  if (INVALID_HANDLE_VALUE == handle) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("NumFileLinks: " + fname, lastError);
+    return s;
+  }
+  UniqueCloseHandlePtr handle_guard(handle, CloseHandleFunc);
+  FILE_STANDARD_INFO standard_info;
+  if (0 != GetFileInformationByHandleEx(handle, FileStandardInfo,
+                                        &standard_info,
+                                        sizeof(standard_info))) {
+    *count = standard_info.NumberOfLinks;
+  } else {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("GetFileInformationByHandleEx: " + fname,
+                                lastError);
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::AreFilesSame(const std::string& first,
+                                     const std::string& second,
+                                     const IOOptions& /*opts*/, bool* res,
+                                     IODebugContext* /*dbg*/) {
+// For MinGW builds
+#if (_WIN32_WINNT == _WIN32_WINNT_VISTA)
+  IOStatus s = IOStatus::NotSupported();
+#else
+  assert(res != nullptr);
+  IOStatus s;
+  if (res == nullptr) {
+    s = IOStatus::InvalidArgument("res");
+    return s;
+  }
+
+  // 0 - for access means read metadata
+  HANDLE file_1 = RX_CreateFile(
+      RX_FN(first).c_str(), 0,
+      FILE_SHARE_DELETE | FILE_SHARE_READ | FILE_SHARE_WRITE, NULL,
+      OPEN_EXISTING,
+      FILE_FLAG_BACKUP_SEMANTICS,  // make opening folders possible
+      NULL);
+
+  if (INVALID_HANDLE_VALUE == file_1) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("open file: " + first, lastError);
+    return s;
+  }
+  UniqueCloseHandlePtr g_1(file_1, CloseHandleFunc);
+
+  HANDLE file_2 = RX_CreateFile(
+      RX_FN(second).c_str(), 0,
+      FILE_SHARE_DELETE | FILE_SHARE_READ | FILE_SHARE_WRITE, NULL,
+      OPEN_EXISTING,
+      FILE_FLAG_BACKUP_SEMANTICS,  // make opening folders possible
+      NULL);
+
+  if (INVALID_HANDLE_VALUE == file_2) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("open file: " + second, lastError);
+    return s;
+  }
+  UniqueCloseHandlePtr g_2(file_2, CloseHandleFunc);
+
+  FILE_ID_INFO FileInfo_1;
+  BOOL result = GetFileInformationByHandleEx(file_1, FileIdInfo, &FileInfo_1,
+                                             sizeof(FileInfo_1));
+
+  if (!result) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("stat file: " + first, lastError);
+    return s;
+  }
+
+  FILE_ID_INFO FileInfo_2;
+  result = GetFileInformationByHandleEx(file_2, FileIdInfo, &FileInfo_2,
+                                        sizeof(FileInfo_2));
+
+  if (!result) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("stat file: " + second, lastError);
+    return s;
+  }
+
+  if (FileInfo_1.VolumeSerialNumber == FileInfo_2.VolumeSerialNumber) {
+    *res =
+        (0 == memcmp(FileInfo_1.FileId.Identifier, FileInfo_2.FileId.Identifier,
+                     sizeof(FileInfo_1.FileId.Identifier)));
+  } else {
+    *res = false;
+  }
+#endif
+  return s;
+}
+
+IOStatus WinFileSystem::LockFile(const std::string& lockFname,
+                                 const IOOptions& /*opts*/, FileLock** lock,
+                                 IODebugContext* /*dbg*/) {
+  assert(lock != nullptr);
+
+  *lock = NULL;
+  IOStatus result;
+
+  // No-sharing, this is a LOCK file
+  const DWORD ExclusiveAccessON = 0;
+
+  // Obtain exclusive access to the LOCK file
+  // Previously, instead of NORMAL attr we set DELETE on close and that worked
+  // well except with fault_injection test that insists on deleting it.
+  HANDLE hFile = 0;
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile = RX_CreateFile(RX_FN(lockFname).c_str(),
+                          (GENERIC_READ | GENERIC_WRITE), ExclusiveAccessON,
+                          NULL, CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    result = IOErrorFromWindowsError("Failed to create lock file: " + lockFname,
+                                     lastError);
+  } else {
+    *lock = new WinFileLock(hFile);
+  }
+
+  return result;
+}
+
+IOStatus WinFileSystem::UnlockFile(FileLock* lock, const IOOptions& /*opts*/,
+                                   IODebugContext* /*dbg*/) {
+  IOStatus result;
+
+  assert(lock != nullptr);
+
+  delete lock;
+
+  return result;
+}
+
+IOStatus WinFileSystem::GetTestDirectory(const IOOptions& opts,
+                                         std::string* result,
+                                         IODebugContext* dbg) {
+  std::string output;
+
+  const char* env = getenv("TEST_TMPDIR");
+  if (env && env[0] != '\0') {
+    output = env;
+  } else {
+    env = getenv("TMP");
+
+    if (env && env[0] != '\0') {
+      output = env;
+    } else {
+      output = "c:\\tmp";
+    }
+  }
+  CreateDir(output, opts, dbg);
+
+  output.append("\\testrocksdb-");
+  output.append(std::to_string(GetCurrentProcessId()));
+
+  CreateDir(output, opts, dbg);
+
+  output.swap(*result);
+
+  return IOStatus::OK();
+}
+
+IOStatus WinFileSystem::NewLogger(const std::string& fname,
+                                  const IOOptions& /*opts*/,
+                                  std::shared_ptr<Logger>* result,
+                                  IODebugContext* /*dbg*/) {
+  IOStatus s;
+
+  result->reset();
+
+  HANDLE hFile = 0;
+  {
+    IOSTATS_TIMER_GUARD(open_nanos);
+    hFile = RX_CreateFile(
+        RX_FN(fname).c_str(), GENERIC_WRITE,
+        FILE_SHARE_READ | FILE_SHARE_DELETE,  // In RocksDb log files are
+        // renamed and deleted before
+        // they are closed. This enables
+        // doing so.
+        NULL,
+        CREATE_ALWAYS,  // Original fopen mode is "w"
+        FILE_ATTRIBUTE_NORMAL, NULL);
+  }
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("Failed to open LogFile" + fname, lastError);
+  } else {
+    {
+      // With log files we want to set the true creation time as of now
+      // because the system
+      // for some reason caches the attributes of the previous file that just
+      // been renamed from
+      // this name so auto_roll_logger_test fails
+      FILETIME ft;
+      GetSystemTimeAsFileTime(&ft);
+      // Set creation, last access and last write time to the same value
+      SetFileTime(hFile, &ft, &ft, &ft);
+    }
+    result->reset(new WinLogger(&WinEnvThreads::gettid, clock_.get(), hFile));
+  }
+  return s;
+}
+
+IOStatus WinFileSystem::IsDirectory(const std::string& path,
+                                    const IOOptions& /*opts*/, bool* is_dir,
+                                    IODebugContext* /*dbg*/) {
+  BOOL ret = RX_PathIsDirectory(RX_FN(path).c_str());
+  if (is_dir) {
+    *is_dir = ret ? true : false;
+  }
+  return IOStatus::OK();
+}
+
+Status WinEnvIO::GetHostName(char* name, uint64_t len) {
+  Status s;
+  DWORD nSize = static_cast<DWORD>(
+      std::min<uint64_t>(len, std::numeric_limits<DWORD>::max()));
+
+  if (!::GetComputerNameA(name, &nSize)) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("GetHostName", lastError);
+  } else {
+    name[nSize] = 0;
+  }
+
+  return s;
+}
+
+IOStatus WinFileSystem::GetAbsolutePath(const std::string& db_path,
+                                        const IOOptions& /*options*/,
+                                        std::string* output_path,
+                                        IODebugContext* dbg) {
+  // Check if we already have an absolute path
+  // For test compatibility we will consider starting slash as an
+  // absolute path
+  if ((!db_path.empty() && (db_path[0] == '\\' || db_path[0] == '/')) ||
+      !RX_PathIsRelative(RX_FN(db_path).c_str())) {
+    *output_path = db_path;
+    return IOStatus::OK();
+  }
+
+  RX_FILESTRING result;
+  result.resize(MAX_PATH);
+
+  // Hopefully no changes the current directory while we do this
+  // however _getcwd also suffers from the same limitation
+  DWORD len = RX_GetCurrentDirectory(MAX_PATH, &result[0]);
+  if (len == 0) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError("Failed to get current working directory",
+                                   lastError);
+  }
+
+  result.resize(len);
+  std::string res = FN_TO_RX(result);
+
+  res.swap(*output_path);
+  return IOStatus::OK();
+}
+
+IOStatus WinFileSystem::GetFreeSpace(const std::string& path,
+                                     const IOOptions& /*options*/,
+                                     uint64_t* diskfree,
+                                     IODebugContext* /*dbg*/) {
+  assert(diskfree != nullptr);
+  ULARGE_INTEGER freeBytes;
+  BOOL f = RX_GetDiskFreeSpaceEx(RX_FN(path).c_str(), &freeBytes, NULL, NULL);
+  if (f) {
+    *diskfree = freeBytes.QuadPart;
+    return IOStatus::OK();
+  } else {
+    DWORD lastError = GetLastError();
+    return IOErrorFromWindowsError("Failed to get free space: " + path,
+                                   lastError);
+  }
+}
+
+FileOptions WinFileSystem::OptimizeForLogWrite(
+    const FileOptions& file_options, const DBOptions& db_options) const {
+  FileOptions optimized(file_options);
+  // These two the same as default optimizations
+  optimized.bytes_per_sync = db_options.wal_bytes_per_sync;
+  optimized.writable_file_max_buffer_size =
+      db_options.writable_file_max_buffer_size;
+
+  // This adversely affects %999 on windows
+  optimized.use_mmap_writes = false;
+  // Direct writes will produce a huge perf impact on
+  // Windows. Pre-allocate space for WAL.
+  optimized.use_direct_writes = false;
+  return optimized;
+}
+
+FileOptions WinFileSystem::OptimizeForManifestWrite(
+    const FileOptions& options) const {
+  FileOptions optimized(options);
+  optimized.use_mmap_writes = false;
+  optimized.use_direct_reads = false;
+  return optimized;
+}
+
+FileOptions WinFileSystem::OptimizeForManifestRead(
+    const FileOptions& file_options) const {
+  FileOptions optimized(file_options);
+  optimized.use_mmap_writes = false;
+  optimized.use_direct_reads = false;
+  return optimized;
+}
+
+// Returns true iff the named directory exists and is a directory.
+bool WinFileSystem::DirExists(const std::string& dname) {
+  WIN32_FILE_ATTRIBUTE_DATA attrs;
+  if (RX_GetFileAttributesEx(RX_FN(dname).c_str(), GetFileExInfoStandard,
+                             &attrs)) {
+    return 0 != (attrs.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY);
+  }
+  return false;
+}
+
+size_t WinFileSystem::GetSectorSize(const std::string& fname) {
+  size_t sector_size = kSectorSize;
+
+  // obtain device handle
+  char devicename[7] = "\\\\.\\";
+  int erresult = 0;
+  if (RX_PathIsRelative(RX_FN(fname).c_str())) {
+    RX_FILESTRING rx_current_dir;
+    rx_current_dir.resize(MAX_PATH);
+    DWORD len = RX_GetCurrentDirectory(MAX_PATH, &rx_current_dir[0]);
+    if (len == 0) {
+      return sector_size;
+    }
+    rx_current_dir.resize(len);
+    std::string current_dir = FN_TO_RX(rx_current_dir);
+    erresult =
+        strncat_s(devicename, sizeof(devicename), current_dir.c_str(), 2);
+  } else {
+    erresult = strncat_s(devicename, sizeof(devicename), fname.c_str(), 2);
+  }
+
+  if (erresult) {
+    assert(false);
+    return sector_size;
+  }
+
+  HANDLE hDevice = CreateFile(devicename, 0, 0, nullptr, OPEN_EXISTING,
+                              FILE_ATTRIBUTE_NORMAL, nullptr);
+
+  if (hDevice == INVALID_HANDLE_VALUE) {
+    return sector_size;
+  }
+
+  STORAGE_PROPERTY_QUERY spropertyquery;
+  spropertyquery.PropertyId = StorageAccessAlignmentProperty;
+  spropertyquery.QueryType = PropertyStandardQuery;
+
+  BYTE output_buffer[sizeof(STORAGE_ACCESS_ALIGNMENT_DESCRIPTOR)];
+  DWORD output_bytes = 0;
+
+  BOOL ret = DeviceIoControl(
+      hDevice, IOCTL_STORAGE_QUERY_PROPERTY, &spropertyquery,
+      sizeof(spropertyquery), output_buffer,
+      sizeof(STORAGE_ACCESS_ALIGNMENT_DESCRIPTOR), &output_bytes, nullptr);
+
+  if (ret) {
+    sector_size = ((STORAGE_ACCESS_ALIGNMENT_DESCRIPTOR*)output_buffer)
+                      ->BytesPerLogicalSector;
+  } else {
+    // many devices do not support StorageProcessAlignmentProperty. Any failure
+    // here and we fall back to logical alignment
+
+    DISK_GEOMETRY_EX geometry = {0};
+    ret = DeviceIoControl(hDevice, IOCTL_DISK_GET_DRIVE_GEOMETRY, nullptr, 0,
+                          &geometry, sizeof(geometry), &output_bytes, nullptr);
+    if (ret) {
+      sector_size = geometry.Geometry.BytesPerSector;
+    }
+  }
+
+  if (hDevice != INVALID_HANDLE_VALUE) {
+    CloseHandle(hDevice);
+  }
+
+  return sector_size;
+}
+
+////////////////////////////////////////////////////////////////////////
+// WinEnvThreads
+
+WinEnvThreads::WinEnvThreads(Env* hosted_env)
+    : hosted_env_(hosted_env), thread_pools_(Env::Priority::TOTAL) {
+  for (int pool_id = 0; pool_id < Env::Priority::TOTAL; ++pool_id) {
+    thread_pools_[pool_id].SetThreadPriority(
+        static_cast<Env::Priority>(pool_id));
+    // This allows later initializing the thread-local-env of each thread.
+    thread_pools_[pool_id].SetHostEnv(hosted_env);
+  }
+}
+
+WinEnvThreads::~WinEnvThreads() {
+  WaitForJoin();
+
+  for (auto& thpool : thread_pools_) {
+    thpool.JoinAllThreads();
+  }
+}
+
+void WinEnvThreads::Schedule(void (*function)(void*), void* arg,
+                             Env::Priority pri, void* tag,
+                             void (*unschedFunction)(void* arg)) {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  thread_pools_[pri].Schedule(function, arg, tag, unschedFunction);
+}
+
+int WinEnvThreads::UnSchedule(void* arg, Env::Priority pri) {
+  return thread_pools_[pri].UnSchedule(arg);
+}
+
+namespace {
+
+struct StartThreadState {
+  void (*user_function)(void*);
+  void* arg;
+};
+
+void* StartThreadWrapper(void* arg) {
+  std::unique_ptr<StartThreadState> state(
+      reinterpret_cast<StartThreadState*>(arg));
+  state->user_function(state->arg);
+  return nullptr;
+}
+
+}  // namespace
+
+void WinEnvThreads::StartThread(void (*function)(void* arg), void* arg) {
+  std::unique_ptr<StartThreadState> state(new StartThreadState);
+  state->user_function = function;
+  state->arg = arg;
+  try {
+    Thread th(&StartThreadWrapper, state.get());
+    state.release();
+
+    std::lock_guard<std::mutex> lg(mu_);
+    threads_to_join_.push_back(std::move(th));
+
+  } catch (const std::system_error& ex) {
+    WinthreadCall("start thread", ex.code());
+  }
+}
+
+void WinEnvThreads::WaitForJoin() {
+  for (auto& th : threads_to_join_) {
+    th.join();
+  }
+  threads_to_join_.clear();
+}
+
+unsigned int WinEnvThreads::GetThreadPoolQueueLen(Env::Priority pri) const {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  return thread_pools_[pri].GetQueueLen();
+}
+
+int WinEnvThreads::ReserveThreads(int threads_to_reserved, Env::Priority pri) {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  return thread_pools_[pri].ReserveThreads(threads_to_reserved);
+}
+
+int WinEnvThreads::ReleaseThreads(int threads_to_released, Env::Priority pri) {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  return thread_pools_[pri].ReleaseThreads(threads_to_released);
+}
+
+uint64_t WinEnvThreads::gettid() {
+  uint64_t thread_id = GetCurrentThreadId();
+  return thread_id;
+}
+
+uint64_t WinEnvThreads::GetThreadID() const { return gettid(); }
+
+void WinEnvThreads::SetBackgroundThreads(int num, Env::Priority pri) {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  thread_pools_[pri].SetBackgroundThreads(num);
+}
+
+int WinEnvThreads::GetBackgroundThreads(Env::Priority pri) {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  return thread_pools_[pri].GetBackgroundThreads();
+}
+
+void WinEnvThreads::IncBackgroundThreadsIfNeeded(int num, Env::Priority pri) {
+  assert(pri >= Env::Priority::BOTTOM && pri <= Env::Priority::HIGH);
+  thread_pools_[pri].IncBackgroundThreadsIfNeeded(num);
+}
+
+/////////////////////////////////////////////////////////////////////////
+// WinEnv
+
+WinEnv::WinEnv()
+    : CompositeEnv(WinFileSystem::Default(), WinClock::Default()),
+      winenv_io_(this),
+      winenv_threads_(this) {
+  // Protected member of the base class
+  thread_status_updater_ = CreateThreadStatusUpdater();
+}
+
+WinEnv::~WinEnv() {
+  // All threads must be joined before the deletion of
+  // thread_status_updater_.
+  delete thread_status_updater_;
+}
+
+Status WinEnv::GetThreadList(std::vector<ThreadStatus>* thread_list) {
+  assert(thread_status_updater_);
+  return thread_status_updater_->GetThreadList(thread_list);
+}
+
+Status WinEnv::GetHostName(char* name, uint64_t len) {
+  return winenv_io_.GetHostName(name, len);
+}
+
+void WinEnv::Schedule(void (*function)(void*), void* arg, Env::Priority pri,
+                      void* tag, void (*unschedFunction)(void* arg)) {
+  return winenv_threads_.Schedule(function, arg, pri, tag, unschedFunction);
+}
+
+int WinEnv::UnSchedule(void* arg, Env::Priority pri) {
+  return winenv_threads_.UnSchedule(arg, pri);
+}
+
+void WinEnv::StartThread(void (*function)(void* arg), void* arg) {
+  return winenv_threads_.StartThread(function, arg);
+}
+
+void WinEnv::WaitForJoin() { return winenv_threads_.WaitForJoin(); }
+
+unsigned int WinEnv::GetThreadPoolQueueLen(Env::Priority pri) const {
+  return winenv_threads_.GetThreadPoolQueueLen(pri);
+}
+int WinEnv::ReserveThreads(int threads_to_reserved, Env::Priority pri) {
+  return winenv_threads_.ReserveThreads(threads_to_reserved, pri);
+}
+
+int WinEnv::ReleaseThreads(int threads_to_released, Env::Priority pri) {
+  return winenv_threads_.ReleaseThreads(threads_to_released, pri);
+}
+
+uint64_t WinEnv::GetThreadID() const { return winenv_threads_.GetThreadID(); }
+
+// Allow increasing the number of worker threads.
+void WinEnv::SetBackgroundThreads(int num, Env::Priority pri) {
+  return winenv_threads_.SetBackgroundThreads(num, pri);
+}
+
+int WinEnv::GetBackgroundThreads(Env::Priority pri) {
+  return winenv_threads_.GetBackgroundThreads(pri);
+}
+
+void WinEnv::IncBackgroundThreadsIfNeeded(int num, Env::Priority pri) {
+  return winenv_threads_.IncBackgroundThreadsIfNeeded(num, pri);
+}
+
+}  // namespace port
+
+std::shared_ptr<FileSystem> FileSystem::Default() {
+  return port::WinFileSystem::Default();
+}
+
+const std::shared_ptr<SystemClock>& SystemClock::Default() {
+  STATIC_AVOID_DESTRUCTION(std::shared_ptr<SystemClock>, clock)
+  (std::make_shared<port::WinClock>());
+  return clock;
+}
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/win/env_win.h b/src/rocksdb/port/win/env_win.h
new file mode 100644
index 000000000..8fbfb8246
--- /dev/null
+++ b/src/rocksdb/port/win/env_win.h
@@ -0,0 +1,304 @@
+// Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// An Env is an interface used by the rocksdb implementation to access
+// operating system functionality like the filesystem etc.  Callers
+// may wish to provide a custom Env object when opening a database to
+// get fine gain control; e.g., to rate limit file system operations.
+//
+// All Env implementations are safe for concurrent access from
+// multiple threads without any external synchronization.
+
+#pragma once
+#include <stdint.h>
+#include <windows.h>
+
+#include <mutex>
+#include <string>
+#include <vector>
+
+#include "env/composite_env_wrapper.h"
+#include "port/port.h"
+#include "rocksdb/env.h"
+#include "rocksdb/file_system.h"
+#include "rocksdb/system_clock.h"
+#include "util/threadpool_imp.h"
+
+#undef GetCurrentTime
+#undef DeleteFile
+#undef LoadLibrary
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+// Currently not designed for inheritance but rather a replacement
+class WinEnvThreads {
+ public:
+  explicit WinEnvThreads(Env* hosted_env);
+
+  ~WinEnvThreads();
+
+  WinEnvThreads(const WinEnvThreads&) = delete;
+  WinEnvThreads& operator=(const WinEnvThreads&) = delete;
+
+  void Schedule(void (*function)(void*), void* arg, Env::Priority pri,
+                void* tag, void (*unschedFunction)(void* arg));
+
+  int UnSchedule(void* arg, Env::Priority pri);
+
+  void StartThread(void (*function)(void* arg), void* arg);
+
+  void WaitForJoin();
+
+  unsigned int GetThreadPoolQueueLen(Env::Priority pri) const;
+
+  int ReserveThreads(int threads_to_be_reserved, Env::Priority pri);
+
+  int ReleaseThreads(int threads_to_be_released, Env::Priority pri);
+
+  static uint64_t gettid();
+
+  uint64_t GetThreadID() const;
+
+  // Allow increasing the number of worker threads.
+  void SetBackgroundThreads(int num, Env::Priority pri);
+  int GetBackgroundThreads(Env::Priority pri);
+
+  void IncBackgroundThreadsIfNeeded(int num, Env::Priority pri);
+
+ private:
+  Env* hosted_env_;
+  mutable std::mutex mu_;
+  std::vector<ThreadPoolImpl> thread_pools_;
+  std::vector<Thread> threads_to_join_;
+};
+
+class WinClock : public SystemClock {
+ public:
+  WinClock();
+  virtual ~WinClock() {}
+
+  static const char* kClassName() { return "WindowsClock"; }
+  const char* Name() const override { return kDefaultName(); }
+  const char* NickName() const override { return kClassName(); }
+
+  uint64_t NowMicros() override;
+
+  uint64_t NowNanos() override;
+
+  // 0 indicates not supported
+  uint64_t CPUMicros() override { return 0; }
+  void SleepForMicroseconds(int micros) override;
+
+  Status GetCurrentTime(int64_t* unix_time) override;
+  // Converts seconds-since-Jan-01-1970 to a printable string
+  virtual std::string TimeToString(uint64_t time);
+
+  uint64_t GetPerfCounterFrequency() const { return perf_counter_frequency_; }
+
+ private:
+  using FnGetSystemTimePreciseAsFileTime = VOID(WINAPI*)(LPFILETIME);
+
+  uint64_t perf_counter_frequency_;
+  uint64_t nano_seconds_per_period_;
+  FnGetSystemTimePreciseAsFileTime GetSystemTimePreciseAsFileTime_;
+};
+
+class WinFileSystem : public FileSystem {
+ public:
+  static const std::shared_ptr<WinFileSystem>& Default();
+  WinFileSystem(const std::shared_ptr<SystemClock>& clock);
+  ~WinFileSystem() {}
+  static const char* kClassName() { return "WinFS"; }
+  const char* Name() const override { return kClassName(); }
+  const char* NickName() const { return kDefaultName(); }
+
+  static size_t GetSectorSize(const std::string& fname);
+  size_t GetPageSize() const { return page_size_; }
+  size_t GetAllocationGranularity() const { return allocation_granularity_; }
+
+  IOStatus DeleteFile(const std::string& fname, const IOOptions& options,
+                      IODebugContext* dbg) override;
+
+  // Truncate the named file to the specified size.
+  IOStatus Truncate(const std::string& /*fname*/, size_t /*size*/,
+                    const IOOptions& /*options*/,
+                    IODebugContext* /*dbg*/) override;
+  IOStatus NewSequentialFile(const std::string& fname,
+                             const FileOptions& file_opts,
+                             std::unique_ptr<FSSequentialFile>* result,
+                             IODebugContext* dbg) override;
+
+  IOStatus NewRandomAccessFile(const std::string& fname,
+                               const FileOptions& options,
+                               std::unique_ptr<FSRandomAccessFile>* result,
+                               IODebugContext* /*dbg*/) override;
+  IOStatus NewWritableFile(const std::string& f, const FileOptions& file_opts,
+                           std::unique_ptr<FSWritableFile>* r,
+                           IODebugContext* dbg) override;
+  IOStatus ReopenWritableFile(const std::string& fname,
+                              const FileOptions& options,
+                              std::unique_ptr<FSWritableFile>* result,
+                              IODebugContext* dbg) override;
+
+  IOStatus NewRandomRWFile(const std::string& fname,
+                           const FileOptions& file_opts,
+                           std::unique_ptr<FSRandomRWFile>* result,
+                           IODebugContext* dbg) override;
+  IOStatus NewMemoryMappedFileBuffer(
+      const std::string& fname,
+      std::unique_ptr<MemoryMappedFileBuffer>* result) override;
+
+  IOStatus NewDirectory(const std::string& name, const IOOptions& io_opts,
+                        std::unique_ptr<FSDirectory>* result,
+                        IODebugContext* dbg) override;
+  IOStatus FileExists(const std::string& f, const IOOptions& io_opts,
+                      IODebugContext* dbg) override;
+  IOStatus GetChildren(const std::string& dir, const IOOptions& io_opts,
+                       std::vector<std::string>* r,
+                       IODebugContext* dbg) override;
+  IOStatus CreateDir(const std::string& dirname, const IOOptions& options,
+                     IODebugContext* dbg) override;
+
+  // Creates directory if missing. Return Ok if it exists, or successful in
+  // Creating.
+  IOStatus CreateDirIfMissing(const std::string& dirname,
+                              const IOOptions& options,
+                              IODebugContext* dbg) override;
+
+  // Delete the specified directory.
+  IOStatus DeleteDir(const std::string& dirname, const IOOptions& options,
+                     IODebugContext* dbg) override;
+  // Store the size of fname in *file_size.
+  IOStatus GetFileSize(const std::string& fname, const IOOptions& options,
+                       uint64_t* file_size, IODebugContext* dbg) override;
+  // Store the last modification time of fname in *file_mtime.
+  IOStatus GetFileModificationTime(const std::string& fname,
+                                   const IOOptions& options,
+                                   uint64_t* file_mtime,
+                                   IODebugContext* dbg) override;
+  // Rename file src to target.
+  IOStatus RenameFile(const std::string& src, const std::string& target,
+                      const IOOptions& options, IODebugContext* dbg) override;
+
+  // Hard Link file src to target.
+  IOStatus LinkFile(const std::string& /*src*/, const std::string& /*target*/,
+                    const IOOptions& /*options*/,
+                    IODebugContext* /*dbg*/) override;
+  IOStatus NumFileLinks(const std::string& /*fname*/,
+                        const IOOptions& /*options*/, uint64_t* /*count*/,
+                        IODebugContext* /*dbg*/) override;
+  IOStatus AreFilesSame(const std::string& /*first*/,
+                        const std::string& /*second*/,
+                        const IOOptions& /*options*/, bool* /*res*/,
+                        IODebugContext* /*dbg*/) override;
+  IOStatus LockFile(const std::string& fname, const IOOptions& options,
+                    FileLock** lock, IODebugContext* dbg) override;
+  IOStatus UnlockFile(FileLock* lock, const IOOptions& options,
+                      IODebugContext* dbg) override;
+  IOStatus GetTestDirectory(const IOOptions& options, std::string* path,
+                            IODebugContext* dbg) override;
+
+  // Create and returns a default logger (an instance of EnvLogger) for storing
+  // informational messages. Derived classes can override to provide custom
+  // logger.
+  IOStatus NewLogger(const std::string& fname, const IOOptions& io_opts,
+                     std::shared_ptr<Logger>* result,
+                     IODebugContext* dbg) override;
+  // Get full directory name for this db.
+  IOStatus GetAbsolutePath(const std::string& db_path, const IOOptions& options,
+                           std::string* output_path,
+                           IODebugContext* dbg) override;
+  IOStatus IsDirectory(const std::string& /*path*/, const IOOptions& options,
+                       bool* is_dir, IODebugContext* /*dgb*/) override;
+  // This seems to clash with a macro on Windows, so #undef it here
+#undef GetFreeSpace
+  IOStatus GetFreeSpace(const std::string& /*path*/,
+                        const IOOptions& /*options*/, uint64_t* /*diskfree*/,
+                        IODebugContext* /*dbg*/) override;
+  FileOptions OptimizeForLogWrite(const FileOptions& file_options,
+                                  const DBOptions& db_options) const override;
+  FileOptions OptimizeForManifestRead(
+      const FileOptions& file_options) const override;
+  FileOptions OptimizeForManifestWrite(
+      const FileOptions& file_options) const override;
+
+ protected:
+  static uint64_t FileTimeToUnixTime(const FILETIME& ftTime);
+  // Returns true iff the named directory exists and is a directory.
+
+  virtual bool DirExists(const std::string& dname);
+  // Helper for NewWritable and ReopenWritableFile
+  virtual IOStatus OpenWritableFile(const std::string& fname,
+                                    const FileOptions& options,
+                                    std::unique_ptr<FSWritableFile>* result,
+                                    bool reopen);
+
+ private:
+  std::shared_ptr<SystemClock> clock_;
+  size_t page_size_;
+  size_t allocation_granularity_;
+};
+
+// Designed for inheritance so can be re-used
+// but certain parts replaced
+class WinEnvIO {
+ public:
+  explicit WinEnvIO(Env* hosted_env);
+
+  virtual ~WinEnvIO();
+
+  virtual Status GetHostName(char* name, uint64_t len);
+
+ private:
+  Env* hosted_env_;
+};
+
+class WinEnv : public CompositeEnv {
+ public:
+  WinEnv();
+
+  ~WinEnv();
+  static const char* kClassName() { return "WinEnv"; }
+  const char* Name() const override { return kClassName(); }
+  const char* NickName() const override { return kDefaultName(); }
+
+  Status GetHostName(char* name, uint64_t len) override;
+
+  Status GetThreadList(std::vector<ThreadStatus>* thread_list) override;
+
+  void Schedule(void (*function)(void*), void* arg, Env::Priority pri,
+                void* tag, void (*unschedFunction)(void* arg)) override;
+
+  int UnSchedule(void* arg, Env::Priority pri) override;
+
+  void StartThread(void (*function)(void* arg), void* arg) override;
+
+  void WaitForJoin() override;
+
+  unsigned int GetThreadPoolQueueLen(Env::Priority pri) const override;
+
+  int ReserveThreads(int threads_to_be_reserved, Env::Priority pri) override;
+
+  int ReleaseThreads(int threads_to_be_released, Env::Priority pri) override;
+
+  uint64_t GetThreadID() const override;
+
+  // Allow increasing the number of worker threads.
+  void SetBackgroundThreads(int num, Env::Priority pri) override;
+  int GetBackgroundThreads(Env::Priority pri) override;
+
+  void IncBackgroundThreadsIfNeeded(int num, Env::Priority pri) override;
+
+ private:
+  WinEnvIO winenv_io_;
+  WinEnvThreads winenv_threads_;
+};
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/win/io_win.cc b/src/rocksdb/port/win/io_win.cc
new file mode 100644
index 000000000..4fa735518
--- /dev/null
+++ b/src/rocksdb/port/win/io_win.cc
@@ -0,0 +1,1101 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+
+#include "port/win/io_win.h"
+
+#include "env_win.h"
+#include "monitoring/iostats_context_imp.h"
+#include "test_util/sync_point.h"
+#include "util/aligned_buffer.h"
+#include "util/coding.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+/*
+ * DirectIOHelper
+ */
+namespace {
+
+const size_t kSectorSize = 512;
+
+inline bool IsPowerOfTwo(const size_t alignment) {
+  return ((alignment) & (alignment - 1)) == 0;
+}
+
+inline bool IsAligned(size_t alignment, const void* ptr) {
+  return ((uintptr_t(ptr)) & (alignment - 1)) == 0;
+}
+}  // namespace
+
+std::string GetWindowsErrSz(DWORD err) {
+  std::string Err;
+  LPSTR lpMsgBuf = nullptr;
+  FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM |
+                     FORMAT_MESSAGE_IGNORE_INSERTS,
+                 NULL, err,
+                 0,  // Default language
+                 reinterpret_cast<LPSTR>(&lpMsgBuf), 0, NULL);
+
+  if (lpMsgBuf) {
+    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.
+IOStatus pwrite(const WinFileData* file_data, const Slice& data,
+                uint64_t offset, size_t& bytes_written) {
+  IOStatus s;
+  bytes_written = 0;
+
+  size_t num_bytes = data.size();
+  if (num_bytes > std::numeric_limits<DWORD>::max()) {
+    // May happen in 64-bit builds where size_t is 64-bits but
+    // long is still 32-bit, but that's the API here at the moment
+    return IOStatus::InvalidArgument(
+        "num_bytes is too large for a single write: " + file_data->GetName());
+  }
+
+  OVERLAPPED overlapped = {0};
+  ULARGE_INTEGER offsetUnion;
+  offsetUnion.QuadPart = offset;
+
+  overlapped.Offset = offsetUnion.LowPart;
+  overlapped.OffsetHigh = offsetUnion.HighPart;
+
+  DWORD bytesWritten = 0;
+
+  if (FALSE == WriteFile(file_data->GetFileHandle(), data.data(),
+                         static_cast<DWORD>(num_bytes), &bytesWritten,
+                         &overlapped)) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError("WriteFile failed: " + file_data->GetName(),
+                                lastError);
+  } else {
+    bytes_written = bytesWritten;
+  }
+
+  return s;
+}
+
+// See comments for pwrite above
+IOStatus pread(const WinFileData* file_data, char* src, size_t num_bytes,
+               uint64_t offset, size_t& bytes_read) {
+  IOStatus s;
+  bytes_read = 0;
+
+  if (num_bytes > std::numeric_limits<DWORD>::max()) {
+    return IOStatus::InvalidArgument(
+        "num_bytes is too large for a single read: " + file_data->GetName());
+  }
+
+  OVERLAPPED overlapped = {0};
+  ULARGE_INTEGER offsetUnion;
+  offsetUnion.QuadPart = offset;
+
+  overlapped.Offset = offsetUnion.LowPart;
+  overlapped.OffsetHigh = offsetUnion.HighPart;
+
+  DWORD bytesRead = 0;
+
+  if (FALSE == ReadFile(file_data->GetFileHandle(), src,
+                        static_cast<DWORD>(num_bytes), &bytesRead,
+                        &overlapped)) {
+    auto lastError = GetLastError();
+    // EOF is OK with zero bytes read
+    if (lastError != ERROR_HANDLE_EOF) {
+      s = IOErrorFromWindowsError("ReadFile failed: " + file_data->GetName(),
+                                  lastError);
+    }
+  } else {
+    bytes_read = bytesRead;
+  }
+
+  return s;
+}
+
+// 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.
+IOStatus fallocate(const std::string& filename, HANDLE hFile,
+                   uint64_t to_size) {
+  IOStatus 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;
+}
+
+IOStatus ftruncate(const std::string& filename, HANDLE hFile, uint64_t toSize) {
+  IOStatus 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*/) {
+  // Returning 0 is safe as it causes the table reader to generate a unique ID.
+  // This is suboptimal for performance as it prevents multiple table readers
+  // for the same file from sharing cached blocks. For example, if users have
+  // a low value for `max_open_files`, there can be many table readers opened
+  // for the same file.
+  //
+  // TODO: this is a temporarily solution as it is safe but not optimal for
+  // performance. For more details see discussion in
+  // https://github.com/facebook/rocksdb/pull/5844.
+  return 0;
+}
+
+WinFileData::WinFileData(const std::string& filename, HANDLE hFile,
+                         bool direct_io)
+    : filename_(filename),
+      hFile_(hFile),
+      use_direct_io_(direct_io),
+      sector_size_(WinFileSystem::GetSectorSize(filename)) {}
+
+bool WinFileData::IsSectorAligned(const size_t off) const {
+  return (off & (sector_size_ - 1)) == 0;
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+// WinMmapReadableFile
+
+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) {}
+
+WinMmapReadableFile::~WinMmapReadableFile() {
+  BOOL ret __attribute__((__unused__));
+  ret = ::UnmapViewOfFile(mapped_region_);
+  assert(ret);
+
+  ret = ::CloseHandle(hMap_);
+  assert(ret);
+}
+
+IOStatus WinMmapReadableFile::Read(uint64_t offset, size_t n,
+                                   const IOOptions& /*options*/, Slice* result,
+                                   char* scratch,
+                                   IODebugContext* /*dbg*/) const {
+  IOStatus s;
+
+  if (offset > length_) {
+    *result = Slice();
+    return IOError(filename_, EINVAL);
+  } else if (offset + n > length_) {
+    n = length_ - static_cast<size_t>(offset);
+  }
+  *result = Slice(reinterpret_cast<const char*>(mapped_region_) + offset, n);
+  return s;
+}
+
+IOStatus WinMmapReadableFile::InvalidateCache(size_t offset, size_t length) {
+  return IOStatus::OK();
+}
+
+size_t WinMmapReadableFile::GetUniqueId(char* id, size_t max_size) const {
+  return GetUniqueIdFromFile(hFile_, id, max_size);
+}
+
+///////////////////////////////////////////////////////////////////////////////
+/// WinMmapFile
+
+// Can only truncate or reserve to a sector size aligned if
+// used on files that are opened with Unbuffered I/O
+IOStatus WinMmapFile::TruncateFile(uint64_t toSize) {
+  return ftruncate(filename_, hFile_, toSize);
+}
+
+IOStatus WinMmapFile::UnmapCurrentRegion() {
+  IOStatus 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;
+}
+
+IOStatus WinMmapFile::MapNewRegion(const IOOptions& options,
+                                   IODebugContext* dbg) {
+  IOStatus status;
+
+  assert(mapped_begin_ == nullptr);
+
+  size_t minDiskSize = static_cast<size_t>(file_offset_) + view_size_;
+
+  if (minDiskSize > reserved_size_) {
+    status = Allocate(file_offset_, view_size_, options, dbg);
+    if (!status.ok()) {
+      return status;
+    }
+  }
+
+  // Need to remap
+  if (hMap_ == NULL || reserved_size_ > mapping_size_) {
+    if (hMap_ != NULL) {
+      // Unmap the previous one
+      BOOL ret __attribute__((__unused__));
+      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;
+}
+
+IOStatus 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 FileOptions& options)
+    : WinFileData(fname, hFile, false),
+      FSWritableFile(options),
+      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(IOOptions(), nullptr);
+  }
+}
+
+IOStatus WinMmapFile::Append(const Slice& data, const IOOptions& options,
+                             IODebugContext* dbg) {
+  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) {
+      IOStatus s = UnmapCurrentRegion();
+      if (s.ok()) {
+        s = MapNewRegion(options, dbg);
+      }
+
+      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 IOStatus::OK();
+}
+
+// Means Close() will properly take care of truncate
+// and it does not need any additional information
+IOStatus WinMmapFile::Truncate(uint64_t size, const IOOptions& /*options*/,
+                               IODebugContext* /*dbg*/) {
+  return IOStatus::OK();
+}
+
+IOStatus WinMmapFile::Close(const IOOptions& options, IODebugContext* dbg) {
+  IOStatus 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(options, dbg);
+
+  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(options, dbg);
+    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;
+}
+
+IOStatus WinMmapFile::Flush(const IOOptions& /*options*/,
+                            IODebugContext* /*dbg*/) {
+  return IOStatus::OK();
+}
+
+// Flush only data
+IOStatus WinMmapFile::Sync(const IOOptions& /*options*/,
+                           IODebugContext* /*dbg*/) {
+  IOStatus 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.
+ */
+IOStatus WinMmapFile::Fsync(const IOOptions& options, IODebugContext* dbg) {
+  IOStatus s = Sync(options, dbg);
+
+  // 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(const IOOptions& /*options*/,
+                                  IODebugContext* /*dbg*/) {
+  size_t used = dst_ - mapped_begin_;
+  return file_offset_ + used;
+}
+
+IOStatus WinMmapFile::InvalidateCache(size_t offset, size_t length) {
+  return IOStatus::OK();
+}
+
+IOStatus WinMmapFile::Allocate(uint64_t offset, uint64_t len,
+                               const IOOptions& /*options*/,
+                               IODebugContext* /*dbg*/) {
+  IOStatus status;
+  TEST_KILL_RANDOM("WinMmapFile::Allocate");
+
+  // 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(static_cast<size_t>(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);
+}
+
+//////////////////////////////////////////////////////////////////////////////////
+// WinSequentialFile
+
+WinSequentialFile::WinSequentialFile(const std::string& fname, HANDLE f,
+                                     const FileOptions& options)
+    : WinFileData(fname, f, options.use_direct_reads) {}
+
+WinSequentialFile::~WinSequentialFile() {
+  assert(hFile_ != INVALID_HANDLE_VALUE);
+}
+
+IOStatus WinSequentialFile::Read(size_t n, const IOOptions& /*opts*/,
+                                 Slice* result, char* scratch,
+                                 IODebugContext* /*dbg*/) {
+  IOStatus s;
+  size_t r = 0;
+
+  assert(result != nullptr);
+  if (WinFileData::use_direct_io()) {
+    return IOStatus::NotSupported("Read() does not support direct_io");
+  }
+
+  // Windows ReadFile API accepts a DWORD.
+  // While it is possible to read in a loop if n is too big
+  // it is an unlikely case.
+  if (n > std::numeric_limits<DWORD>::max()) {
+    return IOStatus::InvalidArgument("n is too big for a single ReadFile: " +
+                                     filename_);
+  }
+
+  DWORD bytesToRead =
+      static_cast<DWORD>(n);  // cast is safe due to the check above
+  DWORD bytesRead = 0;
+  BOOL ret = ReadFile(hFile_, scratch, bytesToRead, &bytesRead, NULL);
+  if (ret != FALSE) {
+    r = bytesRead;
+  } else {
+    auto lastError = GetLastError();
+    if (lastError != ERROR_HANDLE_EOF) {
+      s = IOErrorFromWindowsError("ReadFile failed: " + filename_, lastError);
+    }
+  }
+
+  *result = Slice(scratch, r);
+  return s;
+}
+
+IOStatus WinSequentialFile::PositionedReadInternal(char* src, size_t numBytes,
+                                                   uint64_t offset,
+                                                   size_t& bytes_read) const {
+  return pread(this, src, numBytes, offset, bytes_read);
+}
+
+IOStatus WinSequentialFile::PositionedRead(uint64_t offset, size_t n,
+                                           const IOOptions& /*opts*/,
+                                           Slice* result, char* scratch,
+                                           IODebugContext* /*dbg*/) {
+  if (!WinFileData::use_direct_io()) {
+    return IOStatus::NotSupported("This function is only used for direct_io");
+  }
+
+  assert(IsSectorAligned(static_cast<size_t>(offset)));
+  assert(IsSectorAligned(static_cast<size_t>(n)));
+
+  size_t bytes_read = 0;  // out param
+  IOStatus s = PositionedReadInternal(scratch, static_cast<size_t>(n), offset,
+                                      bytes_read);
+  *result = Slice(scratch, bytes_read);
+  return s;
+}
+
+IOStatus 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 > static_cast<uint64_t>(std::numeric_limits<LONGLONG>::max())) {
+    return IOStatus::InvalidArgument(
+        "n is too large for a single SetFilePointerEx() call" + filename_);
+  }
+
+  LARGE_INTEGER li;
+  li.QuadPart = static_cast<LONGLONG>(n);  // cast is safe due to the check
+                                           // above
+  BOOL ret = SetFilePointerEx(hFile_, li, NULL, FILE_CURRENT);
+  if (ret == FALSE) {
+    auto lastError = GetLastError();
+    return IOErrorFromWindowsError("Skip SetFilePointerEx():" + filename_,
+                                   lastError);
+  }
+  return IOStatus::OK();
+}
+
+IOStatus WinSequentialFile::InvalidateCache(size_t offset, size_t length) {
+  return IOStatus::OK();
+}
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+/// WinRandomAccessBase
+
+inline IOStatus WinRandomAccessImpl::PositionedReadInternal(
+    char* src, size_t numBytes, uint64_t offset, size_t& bytes_read) const {
+  return pread(file_base_, src, numBytes, offset, bytes_read);
+}
+
+inline WinRandomAccessImpl::WinRandomAccessImpl(WinFileData* file_base,
+                                                size_t alignment,
+                                                const FileOptions& options)
+    : file_base_(file_base),
+      alignment_(std::max(alignment, file_base->GetSectorSize())) {
+  assert(!options.use_mmap_reads);
+}
+
+inline IOStatus WinRandomAccessImpl::ReadImpl(uint64_t offset, size_t n,
+                                              Slice* result,
+                                              char* scratch) const {
+  // Check buffer alignment
+  if (file_base_->use_direct_io()) {
+    assert(file_base_->IsSectorAligned(static_cast<size_t>(offset)));
+    assert(IsAligned(alignment_, scratch));
+  }
+
+  if (n == 0) {
+    *result = Slice(scratch, 0);
+    return IOStatus::OK();
+  }
+
+  size_t bytes_read = 0;
+  IOStatus s = PositionedReadInternal(scratch, n, offset, bytes_read);
+  *result = Slice(scratch, bytes_read);
+  return s;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////
+/// WinRandomAccessFile
+
+WinRandomAccessFile::WinRandomAccessFile(const std::string& fname, HANDLE hFile,
+                                         size_t alignment,
+                                         const FileOptions& options)
+    : WinFileData(fname, hFile, options.use_direct_reads),
+      WinRandomAccessImpl(this, alignment, options) {}
+
+WinRandomAccessFile::~WinRandomAccessFile() {}
+
+IOStatus WinRandomAccessFile::Read(uint64_t offset, size_t n,
+                                   const IOOptions& /*options*/, Slice* result,
+                                   char* scratch,
+                                   IODebugContext* /*dbg*/) const {
+  return ReadImpl(offset, n, result, scratch);
+}
+
+IOStatus WinRandomAccessFile::InvalidateCache(size_t offset, size_t length) {
+  return IOStatus::OK();
+}
+
+size_t WinRandomAccessFile::GetUniqueId(char* id, size_t max_size) const {
+  return GetUniqueIdFromFile(GetFileHandle(), id, max_size);
+}
+
+size_t WinRandomAccessFile::GetRequiredBufferAlignment() const {
+  return GetAlignment();
+}
+
+/////////////////////////////////////////////////////////////////////////////
+// WinWritableImpl
+//
+
+inline IOStatus WinWritableImpl::PreallocateInternal(uint64_t spaceToReserve) {
+  return fallocate(file_data_->GetName(), file_data_->GetFileHandle(),
+                   spaceToReserve);
+}
+
+inline WinWritableImpl::WinWritableImpl(WinFileData* file_data,
+                                        size_t alignment)
+    : file_data_(file_data),
+      alignment_(std::max(alignment, file_data->GetSectorSize())),
+      next_write_offset_(0),
+      reservedsize_(0) {
+  // Query current position in case ReopenWritableFile is called
+  // This position is only important for buffered writes
+  // for unbuffered writes we explicitely specify the position.
+  LARGE_INTEGER zero_move;
+  zero_move.QuadPart = 0;  // Do not move
+  LARGE_INTEGER pos;
+  pos.QuadPart = 0;
+  BOOL ret = SetFilePointerEx(file_data_->GetFileHandle(), zero_move, &pos,
+                              FILE_CURRENT);
+  // Querying no supped to fail
+  if (ret != 0) {
+    next_write_offset_ = pos.QuadPart;
+  } else {
+    assert(false);
+  }
+}
+
+inline IOStatus WinWritableImpl::AppendImpl(const Slice& data) {
+  IOStatus s;
+
+  if (data.size() > std::numeric_limits<DWORD>::max()) {
+    return IOStatus::InvalidArgument("data is too long for a single write" +
+                                     file_data_->GetName());
+  }
+
+  size_t bytes_written = 0;  // out param
+
+  if (file_data_->use_direct_io()) {
+    // With no offset specified we are appending
+    // to the end of the file
+    assert(file_data_->IsSectorAligned(next_write_offset_));
+    assert(file_data_->IsSectorAligned(data.size()));
+    assert(IsAligned(static_cast<size_t>(GetAlignment()), data.data()));
+    s = pwrite(file_data_, data, next_write_offset_, bytes_written);
+  } 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 {
+      bytes_written = bytesWritten;
+    }
+  }
+
+  if (s.ok()) {
+    if (bytes_written == data.size()) {
+      // This matters for direct_io cases where
+      // we rely on the fact that next_write_offset_
+      // is sector aligned
+      next_write_offset_ += bytes_written;
+    } else {
+      s = IOStatus::IOError("Failed to write all bytes: " +
+                            file_data_->GetName());
+    }
+  }
+
+  return s;
+}
+
+inline IOStatus WinWritableImpl::PositionedAppendImpl(const Slice& data,
+                                                      uint64_t offset) {
+  if (file_data_->use_direct_io()) {
+    assert(file_data_->IsSectorAligned(static_cast<size_t>(offset)));
+    assert(file_data_->IsSectorAligned(data.size()));
+    assert(IsAligned(static_cast<size_t>(GetAlignment()), data.data()));
+  }
+
+  size_t bytes_written = 0;
+  IOStatus s = pwrite(file_data_, data, offset, bytes_written);
+
+  if (s.ok()) {
+    if (bytes_written == data.size()) {
+      // For sequential write this would be simple
+      // size extension by data.size()
+      uint64_t write_end = offset + bytes_written;
+      if (write_end >= next_write_offset_) {
+        next_write_offset_ = write_end;
+      }
+    } else {
+      s = IOStatus::IOError("Failed to write all of the requested data: " +
+                            file_data_->GetName());
+    }
+  }
+  return s;
+}
+
+inline IOStatus WinWritableImpl::TruncateImpl(uint64_t size) {
+  // It is tempting to check for the size for sector alignment
+  // but truncation may come at the end and there is not a requirement
+  // for this to be sector aligned so long as we do not attempt to write
+  // after that. The interface docs state that the behavior is undefined
+  // in that case.
+  IOStatus s =
+      ftruncate(file_data_->GetName(), file_data_->GetFileHandle(), size);
+
+  if (s.ok()) {
+    next_write_offset_ = size;
+  }
+  return s;
+}
+
+inline IOStatus WinWritableImpl::CloseImpl() {
+  IOStatus s;
+
+  auto hFile = file_data_->GetFileHandle();
+  assert(INVALID_HANDLE_VALUE != hFile);
+
+  if (!::FlushFileBuffers(hFile)) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError(
+        "FlushFileBuffers failed at Close() for: " + file_data_->GetName(),
+        lastError);
+  }
+
+  if (!file_data_->CloseFile() && s.ok()) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError(
+        "CloseHandle failed for: " + file_data_->GetName(), lastError);
+  }
+  return s;
+}
+
+inline IOStatus WinWritableImpl::SyncImpl(const IOOptions& /*options*/,
+                                          IODebugContext* /*dbg*/) {
+  IOStatus s;
+  if (!::FlushFileBuffers(file_data_->GetFileHandle())) {
+    auto lastError = GetLastError();
+    s = IOErrorFromWindowsError(
+        "FlushFileBuffers failed at Sync() for: " + file_data_->GetName(),
+        lastError);
+  }
+  return s;
+}
+
+inline IOStatus WinWritableImpl::AllocateImpl(uint64_t offset, uint64_t len) {
+  IOStatus status;
+  TEST_KILL_RANDOM("WinWritableFile::Allocate");
+
+  // 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(static_cast<size_t>(offset + len),
+                                  static_cast<size_t>(alignment_));
+  // Nothing to do
+  if (spaceToReserve <= reservedsize_) {
+    return status;
+  }
+
+  IOSTATS_TIMER_GUARD(allocate_nanos);
+  status = PreallocateInternal(spaceToReserve);
+  if (status.ok()) {
+    reservedsize_ = spaceToReserve;
+  }
+  return status;
+}
+
+////////////////////////////////////////////////////////////////////////////////
+/// WinWritableFile
+
+WinWritableFile::WinWritableFile(const std::string& fname, HANDLE hFile,
+                                 size_t alignment, size_t /* capacity */,
+                                 const FileOptions& options)
+    : WinFileData(fname, hFile, options.use_direct_writes),
+      WinWritableImpl(this, alignment),
+      FSWritableFile(options) {
+  assert(!options.use_mmap_writes);
+}
+
+WinWritableFile::~WinWritableFile() {}
+
+// Indicates if the class makes use of direct I/O
+bool WinWritableFile::use_direct_io() const {
+  return WinFileData::use_direct_io();
+}
+
+size_t WinWritableFile::GetRequiredBufferAlignment() const {
+  return static_cast<size_t>(GetAlignment());
+}
+
+IOStatus WinWritableFile::Append(const Slice& data,
+                                 const IOOptions& /*options*/,
+                                 IODebugContext* /*dbg*/) {
+  return AppendImpl(data);
+}
+
+IOStatus WinWritableFile::PositionedAppend(const Slice& data, uint64_t offset,
+                                           const IOOptions& /*options*/,
+                                           IODebugContext* /*dbg*/) {
+  return PositionedAppendImpl(data, offset);
+}
+
+// Need to implement this so the file is truncated correctly
+// when buffered and unbuffered mode
+IOStatus WinWritableFile::Truncate(uint64_t size, const IOOptions& /*options*/,
+                                   IODebugContext* /*dbg*/) {
+  return TruncateImpl(size);
+}
+
+IOStatus WinWritableFile::Close(const IOOptions& /*options*/,
+                                IODebugContext* /*dbg*/) {
+  return CloseImpl();
+}
+
+// write out the cached data to the OS cache
+// This is now taken care of the WritableFileWriter
+IOStatus WinWritableFile::Flush(const IOOptions& /*options*/,
+                                IODebugContext* /*dbg*/) {
+  return IOStatus::OK();
+}
+
+IOStatus WinWritableFile::Sync(const IOOptions& options, IODebugContext* dbg) {
+  return SyncImpl(options, dbg);
+}
+
+IOStatus WinWritableFile::Fsync(const IOOptions& options, IODebugContext* dbg) {
+  return SyncImpl(options, dbg);
+}
+
+bool WinWritableFile::IsSyncThreadSafe() const { return true; }
+
+uint64_t WinWritableFile::GetFileSize(const IOOptions& /*options*/,
+                                      IODebugContext* /*dbg*/) {
+  return GetFileNextWriteOffset();
+}
+
+IOStatus WinWritableFile::Allocate(uint64_t offset, uint64_t len,
+                                   const IOOptions& /*options*/,
+                                   IODebugContext* /*dbg*/) {
+  return AllocateImpl(offset, len);
+}
+
+size_t WinWritableFile::GetUniqueId(char* id, size_t max_size) const {
+  return GetUniqueIdFromFile(GetFileHandle(), id, max_size);
+}
+
+/////////////////////////////////////////////////////////////////////////
+/// WinRandomRWFile
+
+WinRandomRWFile::WinRandomRWFile(const std::string& fname, HANDLE hFile,
+                                 size_t alignment, const FileOptions& options)
+    : WinFileData(fname, hFile,
+                  options.use_direct_reads && options.use_direct_writes),
+      WinRandomAccessImpl(this, alignment, options),
+      WinWritableImpl(this, alignment) {}
+
+bool WinRandomRWFile::use_direct_io() const {
+  return WinFileData::use_direct_io();
+}
+
+size_t WinRandomRWFile::GetRequiredBufferAlignment() const {
+  assert(WinRandomAccessImpl::GetAlignment() ==
+         WinWritableImpl::GetAlignment());
+  return static_cast<size_t>(WinRandomAccessImpl::GetAlignment());
+}
+
+IOStatus WinRandomRWFile::Write(uint64_t offset, const Slice& data,
+                                const IOOptions& /*options*/,
+                                IODebugContext* /*dbg*/) {
+  return PositionedAppendImpl(data, offset);
+}
+
+IOStatus WinRandomRWFile::Read(uint64_t offset, size_t n,
+                               const IOOptions& /*options*/, Slice* result,
+                               char* scratch, IODebugContext* /*dbg*/) const {
+  return ReadImpl(offset, n, result, scratch);
+}
+
+IOStatus WinRandomRWFile::Flush(const IOOptions& /*options*/,
+                                IODebugContext* /*dbg*/) {
+  return IOStatus::OK();
+}
+
+IOStatus WinRandomRWFile::Sync(const IOOptions& options, IODebugContext* dbg) {
+  return SyncImpl(options, dbg);
+}
+
+IOStatus WinRandomRWFile::Close(const IOOptions& /*options*/,
+                                IODebugContext* /*dbg*/) {
+  return CloseImpl();
+}
+
+//////////////////////////////////////////////////////////////////////////
+/// WinMemoryMappedBufer
+WinMemoryMappedBuffer::~WinMemoryMappedBuffer() {
+  BOOL ret
+#if defined(_MSC_VER)
+      = FALSE;
+#else
+      __attribute__((__unused__));
+#endif
+  if (base_ != nullptr) {
+    ret = ::UnmapViewOfFile(base_);
+    assert(ret);
+    base_ = nullptr;
+  }
+  if (map_handle_ != NULL && map_handle_ != INVALID_HANDLE_VALUE) {
+    ret = ::CloseHandle(map_handle_);
+    assert(ret);
+    map_handle_ = NULL;
+  }
+  if (file_handle_ != NULL && file_handle_ != INVALID_HANDLE_VALUE) {
+    ret = ::CloseHandle(file_handle_);
+    assert(ret);
+    file_handle_ = NULL;
+  }
+}
+
+//////////////////////////////////////////////////////////////////////////
+/// WinDirectory
+
+IOStatus WinDirectory::Fsync(const IOOptions& /*options*/,
+                             IODebugContext* /*dbg*/) {
+  return IOStatus::OK();
+}
+
+IOStatus WinDirectory::Close(const IOOptions& /*options*/,
+                             IODebugContext* /*dbg*/) {
+  IOStatus s = IOStatus::OK();
+  BOOL ret __attribute__((__unused__));
+  if (handle_ != INVALID_HANDLE_VALUE) {
+    ret = ::CloseHandle(handle_);
+    if (!ret) {
+      auto lastError = GetLastError();
+      s = IOErrorFromWindowsError("Directory closes failed for : " + GetName(),
+                                  lastError);
+    }
+    handle_ = NULL;
+  }
+  return s;
+}
+
+size_t WinDirectory::GetUniqueId(char* id, size_t max_size) const {
+  return GetUniqueIdFromFile(handle_, id, max_size);
+}
+//////////////////////////////////////////////////////////////////////////
+/// WinFileLock
+
+WinFileLock::~WinFileLock() {
+  BOOL ret __attribute__((__unused__));
+  ret = ::CloseHandle(hFile_);
+  assert(ret);
+}
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/win/io_win.h b/src/rocksdb/port/win/io_win.h
new file mode 100644
index 000000000..a4fee8346
--- /dev/null
+++ b/src/rocksdb/port/win/io_win.h
@@ -0,0 +1,508 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+#pragma once
+
+#include <stdint.h>
+#include <windows.h>
+
+#include <mutex>
+#include <string>
+
+#include "rocksdb/file_system.h"
+#include "rocksdb/status.h"
+#include "util/aligned_buffer.h"
+#include "util/string_util.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+std::string GetWindowsErrSz(DWORD err);
+
+inline IOStatus IOErrorFromWindowsError(const std::string& context, DWORD err) {
+  return ((err == ERROR_HANDLE_DISK_FULL) || (err == ERROR_DISK_FULL))
+             ? IOStatus::NoSpace(context, GetWindowsErrSz(err))
+         : ((err == ERROR_FILE_NOT_FOUND) || (err == ERROR_PATH_NOT_FOUND))
+             ? IOStatus::PathNotFound(context, GetWindowsErrSz(err))
+             : IOStatus::IOError(context, GetWindowsErrSz(err));
+}
+
+inline IOStatus IOErrorFromLastWindowsError(const std::string& context) {
+  return IOErrorFromWindowsError(context, GetLastError());
+}
+
+inline IOStatus IOError(const std::string& context, int err_number) {
+  return (err_number == ENOSPC)
+             ? IOStatus::NoSpace(context, errnoStr(err_number).c_str())
+         : (err_number == ENOENT)
+             ? IOStatus::PathNotFound(context, errnoStr(err_number).c_str())
+             : IOStatus::IOError(context, errnoStr(err_number).c_str());
+}
+
+class WinFileData;
+
+IOStatus pwrite(const WinFileData* file_data, const Slice& data,
+                uint64_t offset, size_t& bytes_written);
+
+IOStatus pread(const WinFileData* file_data, char* src, size_t num_bytes,
+               uint64_t offset, size_t& bytes_read);
+
+IOStatus fallocate(const std::string& filename, HANDLE hFile, uint64_t to_size);
+
+IOStatus ftruncate(const std::string& filename, HANDLE hFile, uint64_t toSize);
+
+size_t GetUniqueIdFromFile(HANDLE hFile, char* id, size_t max_size);
+
+class WinFileData {
+ protected:
+  const std::string filename_;
+  HANDLE hFile_;
+  // If true, the I/O issued would be direct I/O which the buffer
+  // will need to be aligned (not sure there is a guarantee that the buffer
+  // passed in is aligned).
+  const bool use_direct_io_;
+  const size_t sector_size_;
+
+ public:
+  // We want this class be usable both for inheritance (prive
+  // or protected) and for containment so __ctor and __dtor public
+  WinFileData(const std::string& filename, HANDLE hFile, bool direct_io);
+
+  virtual ~WinFileData() { this->CloseFile(); }
+
+  bool CloseFile() {
+    bool result = true;
+
+    if (hFile_ != NULL && hFile_ != INVALID_HANDLE_VALUE) {
+      result = ::CloseHandle(hFile_);
+      assert(result);
+      hFile_ = NULL;
+    }
+    return result;
+  }
+
+  const std::string& GetName() const { return filename_; }
+
+  HANDLE GetFileHandle() const { return hFile_; }
+
+  bool use_direct_io() const { return use_direct_io_; }
+
+  size_t GetSectorSize() const { return sector_size_; }
+
+  bool IsSectorAligned(const size_t off) const;
+
+  WinFileData(const WinFileData&) = delete;
+  WinFileData& operator=(const WinFileData&) = delete;
+};
+
+class WinSequentialFile : protected WinFileData, public FSSequentialFile {
+  // Override for behavior change when creating a custom env
+  virtual IOStatus PositionedReadInternal(char* src, size_t numBytes,
+                                          uint64_t offset,
+                                          size_t& bytes_read) const;
+
+ public:
+  WinSequentialFile(const std::string& fname, HANDLE f,
+                    const FileOptions& options);
+
+  ~WinSequentialFile();
+
+  WinSequentialFile(const WinSequentialFile&) = delete;
+  WinSequentialFile& operator=(const WinSequentialFile&) = delete;
+
+  IOStatus Read(size_t n, const IOOptions& options, Slice* result,
+                char* scratch, IODebugContext* dbg) override;
+  IOStatus PositionedRead(uint64_t offset, size_t n, const IOOptions& options,
+                          Slice* result, char* scratch,
+                          IODebugContext* dbg) override;
+
+  IOStatus Skip(uint64_t n) override;
+
+  IOStatus InvalidateCache(size_t offset, size_t length) override;
+
+  virtual bool use_direct_io() const override {
+    return WinFileData::use_direct_io();
+  }
+};
+
+// mmap() based random-access
+class WinMmapReadableFile : private WinFileData, public FSRandomAccessFile {
+  HANDLE hMap_;
+
+  const void* mapped_region_;
+  const size_t length_;
+
+ public:
+  // mapped_region_[0,length-1] contains the mmapped contents of the file.
+  WinMmapReadableFile(const std::string& fileName, HANDLE hFile, HANDLE hMap,
+                      const void* mapped_region, size_t length);
+
+  ~WinMmapReadableFile();
+
+  WinMmapReadableFile(const WinMmapReadableFile&) = delete;
+  WinMmapReadableFile& operator=(const WinMmapReadableFile&) = delete;
+
+  IOStatus Read(uint64_t offset, size_t n, const IOOptions& options,
+                Slice* result, char* scratch,
+                IODebugContext* dbg) const override;
+
+  virtual IOStatus InvalidateCache(size_t offset, size_t length) override;
+
+  virtual size_t GetUniqueId(char* id, size_t max_size) const override;
+};
+
+// We preallocate and use memcpy to append new
+// data to the file.  This is safe since we either properly close the
+// file before reading from it, or for log files, the reading code
+// knows enough to skip zero suffixes.
+class WinMmapFile : private WinFileData, public FSWritableFile {
+ private:
+  HANDLE hMap_;
+
+  const size_t page_size_;  // We flush the mapping view in page_size
+  // increments. We may decide if this is a memory
+  // page size or SSD page size
+  const size_t
+      allocation_granularity_;  // View must start at such a granularity
+
+  size_t reserved_size_;  // Preallocated size
+
+  size_t mapping_size_;  // The max size of the mapping object
+  // we want to guess the final file size to minimize the remapping
+  size_t view_size_;  // How much memory to map into a view at a time
+
+  char* mapped_begin_;  // Must begin at the file offset that is aligned with
+  // allocation_granularity_
+  char* mapped_end_;
+  char* dst_;  // Where to write next  (in range [mapped_begin_,mapped_end_])
+  char* last_sync_;  // Where have we synced up to
+
+  uint64_t file_offset_;  // Offset of mapped_begin_ in file
+
+  // Do we have unsynced writes?
+  bool pending_sync_;
+
+  // Can only truncate or reserve to a sector size aligned if
+  // used on files that are opened with Unbuffered I/O
+  IOStatus TruncateFile(uint64_t toSize);
+
+  IOStatus UnmapCurrentRegion();
+
+  IOStatus MapNewRegion(const IOOptions& options, IODebugContext* dbg);
+
+  virtual IOStatus PreallocateInternal(uint64_t spaceToReserve);
+
+ public:
+  WinMmapFile(const std::string& fname, HANDLE hFile, size_t page_size,
+              size_t allocation_granularity, const FileOptions& options);
+
+  ~WinMmapFile();
+
+  WinMmapFile(const WinMmapFile&) = delete;
+  WinMmapFile& operator=(const WinMmapFile&) = delete;
+
+  IOStatus Append(const Slice& data, const IOOptions& options,
+                  IODebugContext* dbg) override;
+  IOStatus Append(const Slice& data, const IOOptions& opts,
+                  const DataVerificationInfo& /* verification_info */,
+                  IODebugContext* dbg) override {
+    return Append(data, opts, dbg);
+  }
+
+  // Means Close() will properly take care of truncate
+  // and it does not need any additional information
+  IOStatus Truncate(uint64_t size, const IOOptions& options,
+                    IODebugContext* dbg) override;
+
+  IOStatus Close(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus Flush(const IOOptions& options, IODebugContext* dbg) override;
+
+  // Flush only data
+  IOStatus Sync(const IOOptions& options, IODebugContext* dbg) override;
+
+  /**
+   * Flush data as well as metadata to stable storage.
+   */
+  IOStatus Fsync(const IOOptions& options, IODebugContext* dbg) override;
+
+  /**
+   * 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 GetFileSize(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus InvalidateCache(size_t offset, size_t length) override;
+
+  IOStatus Allocate(uint64_t offset, uint64_t len, const IOOptions& options,
+                    IODebugContext* dbg) override;
+
+  virtual size_t GetUniqueId(char* id, size_t max_size) const override;
+};
+
+class WinRandomAccessImpl {
+ protected:
+  WinFileData* file_base_;
+  size_t alignment_;
+
+  // Override for behavior change when creating a custom env
+  virtual IOStatus PositionedReadInternal(char* src, size_t numBytes,
+                                          uint64_t offset,
+                                          size_t& bytes_read) const;
+
+  WinRandomAccessImpl(WinFileData* file_base, size_t alignment,
+                      const FileOptions& options);
+
+  virtual ~WinRandomAccessImpl() {}
+
+  IOStatus ReadImpl(uint64_t offset, size_t n, Slice* result,
+                    char* scratch) const;
+
+  size_t GetAlignment() const { return alignment_; }
+
+ public:
+  WinRandomAccessImpl(const WinRandomAccessImpl&) = delete;
+  WinRandomAccessImpl& operator=(const WinRandomAccessImpl&) = delete;
+};
+
+// pread() based random-access
+class WinRandomAccessFile
+    : private WinFileData,
+      protected WinRandomAccessImpl,  // Want to be able to override
+                                      // PositionedReadInternal
+      public FSRandomAccessFile {
+ public:
+  WinRandomAccessFile(const std::string& fname, HANDLE hFile, size_t alignment,
+                      const FileOptions& options);
+
+  ~WinRandomAccessFile();
+
+  IOStatus Read(uint64_t offset, size_t n, const IOOptions& options,
+                Slice* result, char* scratch,
+                IODebugContext* dbg) const override;
+
+  virtual size_t GetUniqueId(char* id, size_t max_size) const override;
+
+  virtual bool use_direct_io() const override {
+    return WinFileData::use_direct_io();
+  }
+
+  IOStatus InvalidateCache(size_t offset, size_t length) override;
+
+  virtual size_t GetRequiredBufferAlignment() const override;
+};
+
+// This is a sequential write class. It has been mimicked (as others) after
+// the original Posix class. We add support for unbuffered I/O on windows as
+// well
+// we utilize the original buffer as an alignment buffer to write directly to
+// file with no buffering.
+// No buffering requires that the provided buffer is aligned to the physical
+// sector size (SSD page size) and
+// that all SetFilePointer() operations to occur with such an alignment.
+// We thus always write in sector/page size increments to the drive and leave
+// the tail for the next write OR for Close() at which point we pad with zeros.
+// No padding is required for
+// buffered access.
+class WinWritableImpl {
+ protected:
+  WinFileData* file_data_;
+  const uint64_t alignment_;
+  uint64_t
+      next_write_offset_;  // Needed because Windows does not support O_APPEND
+  uint64_t reservedsize_;  // how far we have reserved space
+
+  virtual IOStatus PreallocateInternal(uint64_t spaceToReserve);
+
+  WinWritableImpl(WinFileData* file_data, size_t alignment);
+
+  ~WinWritableImpl() {}
+
+  uint64_t GetAlignment() const { return alignment_; }
+
+  IOStatus AppendImpl(const Slice& data);
+
+  // Requires that the data is aligned as specified by
+  // GetRequiredBufferAlignment()
+  IOStatus PositionedAppendImpl(const Slice& data, uint64_t offset);
+
+  IOStatus TruncateImpl(uint64_t size);
+
+  IOStatus CloseImpl();
+
+  IOStatus SyncImpl(const IOOptions& options, IODebugContext* dbg);
+
+  uint64_t GetFileNextWriteOffset() {
+    // Double accounting now here with WritableFileWriter
+    // and this size will be wrong when unbuffered access is used
+    // but tests implement their own writable files and do not use
+    // WritableFileWrapper
+    // so we need to squeeze a square peg through
+    // a round hole here.
+    return next_write_offset_;
+  }
+
+  IOStatus AllocateImpl(uint64_t offset, uint64_t len);
+
+ public:
+  WinWritableImpl(const WinWritableImpl&) = delete;
+  WinWritableImpl& operator=(const WinWritableImpl&) = delete;
+};
+
+class WinWritableFile : private WinFileData,
+                        protected WinWritableImpl,
+                        public FSWritableFile {
+ public:
+  WinWritableFile(const std::string& fname, HANDLE hFile, size_t alignment,
+                  size_t capacity, const FileOptions& options);
+
+  ~WinWritableFile();
+
+  IOStatus Append(const Slice& data, const IOOptions& options,
+                  IODebugContext* dbg) override;
+  IOStatus Append(const Slice& data, const IOOptions& opts,
+                  const DataVerificationInfo& /* verification_info */,
+                  IODebugContext* dbg) override {
+    return Append(data, opts, dbg);
+  }
+
+  // Requires that the data is aligned as specified by
+  // GetRequiredBufferAlignment()
+  IOStatus PositionedAppend(const Slice& data, uint64_t offset,
+                            const IOOptions& options,
+                            IODebugContext* dbg) override;
+  IOStatus PositionedAppend(const Slice& data, uint64_t offset,
+                            const IOOptions& opts,
+                            const DataVerificationInfo& /* verification_info */,
+                            IODebugContext* dbg) override {
+    return PositionedAppend(data, offset, opts, dbg);
+  }
+
+  // Need to implement this so the file is truncated correctly
+  // when buffered and unbuffered mode
+  IOStatus Truncate(uint64_t size, const IOOptions& options,
+                    IODebugContext* dbg) override;
+
+  IOStatus Close(const IOOptions& options, IODebugContext* dbg) override;
+
+  // write out the cached data to the OS cache
+  // This is now taken care of the WritableFileWriter
+  IOStatus Flush(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus Sync(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus Fsync(const IOOptions& options, IODebugContext* dbg) override;
+
+  virtual bool IsSyncThreadSafe() const override;
+
+  // Indicates if the class makes use of direct I/O
+  // Use PositionedAppend
+  virtual bool use_direct_io() const override;
+
+  virtual size_t GetRequiredBufferAlignment() const override;
+
+  uint64_t GetFileSize(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus Allocate(uint64_t offset, uint64_t len, const IOOptions& options,
+                    IODebugContext* dbg) override;
+
+  virtual size_t GetUniqueId(char* id, size_t max_size) const override;
+};
+
+class WinRandomRWFile : private WinFileData,
+                        protected WinRandomAccessImpl,
+                        protected WinWritableImpl,
+                        public FSRandomRWFile {
+ public:
+  WinRandomRWFile(const std::string& fname, HANDLE hFile, size_t alignment,
+                  const FileOptions& options);
+
+  ~WinRandomRWFile() {}
+
+  // Indicates if the class makes use of direct I/O
+  // If false you must pass aligned buffer to Write()
+  virtual bool use_direct_io() const override;
+
+  // Use the returned alignment value to allocate aligned
+  // buffer for Write() when use_direct_io() returns true
+  virtual size_t GetRequiredBufferAlignment() const override;
+
+  // Write bytes in `data` at  offset `offset`, Returns Status::OK() on success.
+  // Pass aligned buffer when use_direct_io() returns true.
+  IOStatus Write(uint64_t offset, const Slice& data, const IOOptions& options,
+                 IODebugContext* dbg) override;
+
+  // Read up to `n` bytes starting from offset `offset` and store them in
+  // result, provided `scratch` size should be at least `n`.
+  // Returns Status::OK() on success.
+  IOStatus Read(uint64_t offset, size_t n, const IOOptions& options,
+                Slice* result, char* scratch,
+                IODebugContext* dbg) const override;
+
+  IOStatus Flush(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus Sync(const IOOptions& options, IODebugContext* dbg) override;
+
+  IOStatus Fsync(const IOOptions& options, IODebugContext* dbg) override {
+    return Sync(options, dbg);
+  }
+
+  IOStatus Close(const IOOptions& options, IODebugContext* dbg) override;
+};
+
+class WinMemoryMappedBuffer : public MemoryMappedFileBuffer {
+ private:
+  HANDLE file_handle_;
+  HANDLE map_handle_;
+
+ public:
+  WinMemoryMappedBuffer(HANDLE file_handle, HANDLE map_handle, void* base,
+                        size_t size)
+      : MemoryMappedFileBuffer(base, size),
+        file_handle_(file_handle),
+        map_handle_(map_handle) {}
+  ~WinMemoryMappedBuffer() override;
+};
+
+class WinDirectory : public FSDirectory {
+  const std::string filename_;
+  HANDLE handle_;
+
+ public:
+  explicit WinDirectory(const std::string& filename, HANDLE h) noexcept
+      : filename_(filename), handle_(h) {
+    assert(handle_ != INVALID_HANDLE_VALUE);
+  }
+  ~WinDirectory() {
+    if (handle_ != NULL) {
+      IOStatus s = WinDirectory::Close(IOOptions(), nullptr);
+      s.PermitUncheckedError();
+    }
+  }
+  const std::string& GetName() const { return filename_; }
+  IOStatus Fsync(const IOOptions& options, IODebugContext* dbg) override;
+  IOStatus Close(const IOOptions& options, IODebugContext* dbg) override;
+
+  size_t GetUniqueId(char* id, size_t max_size) const override;
+};
+
+class WinFileLock : public FileLock {
+ public:
+  explicit WinFileLock(HANDLE hFile) : hFile_(hFile) {
+    assert(hFile != NULL);
+    assert(hFile != INVALID_HANDLE_VALUE);
+  }
+
+  ~WinFileLock();
+
+ private:
+  HANDLE hFile_;
+};
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/win/port_win.cc b/src/rocksdb/port/win/port_win.cc
new file mode 100644
index 000000000..37e8f655c
--- /dev/null
+++ b/src/rocksdb/port/win/port_win.cc
@@ -0,0 +1,303 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+
+#include "port/win/port_win.h"
+
+#include <assert.h>
+#include <io.h>
+#include <rpc.h>
+#include <stdio.h>
+#include <string.h>
+
+#include <chrono>
+#include <cstdlib>
+#include <exception>
+#include <memory>
+
+#include "port/port_dirent.h"
+#include "port/sys_time.h"
+
+#ifdef ROCKSDB_WINDOWS_UTF8_FILENAMES
+// utf8 <-> utf16
+#include <codecvt>
+#include <locale>
+#include <string>
+#endif
+
+#include "logging/logging.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+extern const bool kDefaultToAdaptiveMutex = false;
+
+namespace port {
+
+#ifdef ROCKSDB_WINDOWS_UTF8_FILENAMES
+std::string utf16_to_utf8(const std::wstring& utf16) {
+  std::wstring_convert<std::codecvt_utf8_utf16<wchar_t>, wchar_t> convert;
+  return convert.to_bytes(utf16);
+}
+
+std::wstring utf8_to_utf16(const std::string& utf8) {
+  std::wstring_convert<std::codecvt_utf8_utf16<wchar_t>> converter;
+  return converter.from_bytes(utf8);
+}
+#endif
+
+void GetTimeOfDay(TimeVal* tv, struct timezone* /* tz */) {
+  std::chrono::microseconds usNow(
+      std::chrono::duration_cast<std::chrono::microseconds>(
+          std::chrono::system_clock::now().time_since_epoch()));
+
+  std::chrono::seconds secNow(
+      std::chrono::duration_cast<std::chrono::seconds>(usNow));
+
+  tv->tv_sec = static_cast<long>(secNow.count());
+  tv->tv_usec = static_cast<long>(
+      usNow.count() -
+      std::chrono::duration_cast<std::chrono::microseconds>(secNow).count());
+}
+
+Mutex::~Mutex() {}
+
+CondVar::~CondVar() {}
+
+void CondVar::Wait() {
+  // Caller must ensure that mutex is held prior to calling this method
+  std::unique_lock<std::mutex> lk(mu_->getLock(), std::adopt_lock);
+#ifndef NDEBUG
+  mu_->locked_ = false;
+#endif
+  cv_.wait(lk);
+#ifndef NDEBUG
+  mu_->locked_ = true;
+#endif
+  // Release ownership of the lock as we don't want it to be unlocked when
+  // it goes out of scope (as we adopted the lock and didn't lock it ourselves)
+  lk.release();
+}
+
+bool CondVar::TimedWait(uint64_t abs_time_us) {
+  // MSVC++ library implements wait_until in terms of wait_for so
+  // we need to convert absolute wait into relative wait.
+  std::chrono::microseconds usAbsTime(abs_time_us);
+
+  std::chrono::microseconds usNow(
+      std::chrono::duration_cast<std::chrono::microseconds>(
+          std::chrono::system_clock::now().time_since_epoch()));
+  std::chrono::microseconds relTimeUs = (usAbsTime > usNow)
+                                            ? (usAbsTime - usNow)
+                                            : std::chrono::microseconds::zero();
+
+  // Caller must ensure that mutex is held prior to calling this method
+  std::unique_lock<std::mutex> lk(mu_->getLock(), std::adopt_lock);
+
+  // Work around https://github.com/microsoft/STL/issues/369
+#if defined(_MSC_VER) && \
+    (!defined(_MSVC_STL_UPDATE) || _MSVC_STL_UPDATE < 202008L)
+  if (relTimeUs == std::chrono::microseconds::zero()) {
+    lk.unlock();
+    lk.lock();
+  }
+#endif
+#ifndef NDEBUG
+  mu_->locked_ = false;
+#endif
+  std::cv_status cvStatus = cv_.wait_for(lk, relTimeUs);
+#ifndef NDEBUG
+  mu_->locked_ = true;
+#endif
+  // Release ownership of the lock as we don't want it to be unlocked when
+  // it goes out of scope (as we adopted the lock and didn't lock it ourselves)
+  lk.release();
+
+  if (cvStatus == std::cv_status::timeout) {
+    return true;
+  }
+
+  return false;
+}
+
+void CondVar::Signal() { cv_.notify_one(); }
+
+void CondVar::SignalAll() { cv_.notify_all(); }
+
+int PhysicalCoreID() { return GetCurrentProcessorNumber(); }
+
+void InitOnce(OnceType* once, void (*initializer)()) {
+  std::call_once(once->flag_, initializer);
+}
+
+// Private structure, exposed only by pointer
+struct DIR {
+  HANDLE handle_;
+  bool firstread_;
+  RX_WIN32_FIND_DATA data_;
+  dirent entry_;
+
+  DIR() : handle_(INVALID_HANDLE_VALUE), firstread_(true) {}
+
+  DIR(const DIR&) = delete;
+  DIR& operator=(const DIR&) = delete;
+
+  ~DIR() {
+    if (INVALID_HANDLE_VALUE != handle_) {
+      ::FindClose(handle_);
+    }
+  }
+};
+
+DIR* opendir(const char* name) {
+  if (!name || *name == 0) {
+    errno = ENOENT;
+    return nullptr;
+  }
+
+  std::string pattern(name);
+  pattern.append("\\").append("*");
+
+  std::unique_ptr<DIR> dir(new DIR);
+
+  dir->handle_ =
+      RX_FindFirstFileEx(RX_FN(pattern).c_str(),
+                         FindExInfoBasic,  // Do not want alternative name
+                         &dir->data_, FindExSearchNameMatch,
+                         NULL,  // lpSearchFilter
+                         0);
+
+  if (dir->handle_ == INVALID_HANDLE_VALUE) {
+    return nullptr;
+  }
+
+  RX_FILESTRING x(dir->data_.cFileName, RX_FNLEN(dir->data_.cFileName));
+  strcpy_s(dir->entry_.d_name, sizeof(dir->entry_.d_name), FN_TO_RX(x).c_str());
+
+  return dir.release();
+}
+
+struct dirent* readdir(DIR* dirp) {
+  if (!dirp || dirp->handle_ == INVALID_HANDLE_VALUE) {
+    errno = EBADF;
+    return nullptr;
+  }
+
+  if (dirp->firstread_) {
+    dirp->firstread_ = false;
+    return &dirp->entry_;
+  }
+
+  auto ret = RX_FindNextFile(dirp->handle_, &dirp->data_);
+
+  if (ret == 0) {
+    return nullptr;
+  }
+
+  RX_FILESTRING x(dirp->data_.cFileName, RX_FNLEN(dirp->data_.cFileName));
+  strcpy_s(dirp->entry_.d_name, sizeof(dirp->entry_.d_name),
+           FN_TO_RX(x).c_str());
+
+  return &dirp->entry_;
+}
+
+int closedir(DIR* dirp) {
+  delete dirp;
+  return 0;
+}
+
+int truncate(const char* path, int64_t length) {
+  if (path == nullptr) {
+    errno = EFAULT;
+    return -1;
+  }
+  return ROCKSDB_NAMESPACE::port::Truncate(path, length);
+}
+
+int Truncate(std::string path, int64_t len) {
+  if (len < 0) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  HANDLE hFile =
+      RX_CreateFile(RX_FN(path).c_str(), GENERIC_READ | GENERIC_WRITE,
+                    FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
+                    NULL,           // Security attrs
+                    OPEN_EXISTING,  // Truncate existing file only
+                    FILE_ATTRIBUTE_NORMAL, NULL);
+
+  if (INVALID_HANDLE_VALUE == hFile) {
+    auto lastError = GetLastError();
+    if (lastError == ERROR_FILE_NOT_FOUND) {
+      errno = ENOENT;
+    } else if (lastError == ERROR_ACCESS_DENIED) {
+      errno = EACCES;
+    } else {
+      errno = EIO;
+    }
+    return -1;
+  }
+
+  int result = 0;
+  FILE_END_OF_FILE_INFO end_of_file;
+  end_of_file.EndOfFile.QuadPart = len;
+
+  if (!SetFileInformationByHandle(hFile, FileEndOfFileInfo, &end_of_file,
+                                  sizeof(FILE_END_OF_FILE_INFO))) {
+    errno = EIO;
+    result = -1;
+  }
+
+  CloseHandle(hFile);
+  return result;
+}
+
+void Crash(const std::string& srcfile, int srcline) {
+  fprintf(stdout, "Crashing at %s:%d\n", srcfile.c_str(), srcline);
+  fflush(stdout);
+  abort();
+}
+
+int GetMaxOpenFiles() { return -1; }
+
+// Assume 4KB page size
+const size_t kPageSize = 4U * 1024U;
+
+void SetCpuPriority(ThreadId id, CpuPriority priority) {
+  // Not supported
+  (void)id;
+  (void)priority;
+}
+
+int64_t GetProcessID() { return GetCurrentProcessId(); }
+
+bool GenerateRfcUuid(std::string* output) {
+  UUID uuid;
+  UuidCreateSequential(&uuid);
+
+  RPC_CSTR rpc_str;
+  auto status = UuidToStringA(&uuid, &rpc_str);
+  if (status != RPC_S_OK) {
+    return false;
+  }
+
+  // rpc_str is nul-terminated
+  *output = reinterpret_cast<char*>(rpc_str);
+
+  status = RpcStringFreeA(&rpc_str);
+  assert(status == RPC_S_OK);
+
+  return true;
+}
+
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/win/port_win.h b/src/rocksdb/port/win/port_win.h
new file mode 100644
index 000000000..989b5620b
--- /dev/null
+++ b/src/rocksdb/port/win/port_win.h
@@ -0,0 +1,378 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// See port_example.h for documentation for the following types/functions.
+
+#pragma once
+
+// Always want minimum headers
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif
+
+#include <windows.h>
+#include <string>
+#include <thread>
+#include <string.h>
+#include <mutex>
+#include <limits>
+#include <condition_variable>
+#include <malloc.h>
+#include <intrin.h>
+#include <process.h>
+
+#include <stdint.h>
+
+#include "port/win/win_thread.h"
+
+#include "rocksdb/options.h"
+
+#undef min
+#undef max
+#undef DeleteFile
+#undef GetCurrentTime
+
+#ifndef strcasecmp
+#define strcasecmp _stricmp
+#endif
+
+#undef GetCurrentTime
+#undef DeleteFile
+
+#ifndef _SSIZE_T_DEFINED
+using ssize_t = SSIZE_T;
+#endif
+
+// size_t printf formatting named in the manner of C99 standard formatting
+// strings such as PRIu64
+// in fact, we could use that one
+#ifndef ROCKSDB_PRIszt
+#define ROCKSDB_PRIszt "Iu"
+#endif
+
+#ifdef _MSC_VER
+#define __attribute__(A)
+
+#endif
+
+namespace ROCKSDB_NAMESPACE {
+
+#define PREFETCH(addr, rw, locality)
+
+extern const bool kDefaultToAdaptiveMutex;
+
+namespace port {
+
+// "Windows is designed to run on little-endian computer architectures."
+// https://docs.microsoft.com/en-us/windows/win32/sysinfo/registry-value-types
+constexpr bool kLittleEndian = true;
+#undef PLATFORM_IS_LITTLE_ENDIAN
+
+class CondVar;
+
+class Mutex {
+ public:
+  static const char* kName() { return "std::mutex"; }
+
+  explicit Mutex(bool IGNORED_adaptive = kDefaultToAdaptiveMutex)
+#ifndef NDEBUG
+      : locked_(false)
+#endif
+  {
+    (void)IGNORED_adaptive;
+  }
+
+  ~Mutex();
+
+  void Lock() {
+    mutex_.lock();
+#ifndef NDEBUG
+    locked_ = true;
+#endif
+  }
+
+  void Unlock() {
+#ifndef NDEBUG
+    locked_ = false;
+#endif
+    mutex_.unlock();
+  }
+
+  bool TryLock() {
+    bool ret = mutex_.try_lock();
+#ifndef NDEBUG
+    if (ret) {
+      locked_ = true;
+    }
+#endif
+    return ret;
+  }
+
+  // this will assert if the mutex is not locked
+  // it does NOT verify that mutex is held by a calling thread
+  void AssertHeld() {
+#ifndef NDEBUG
+    assert(locked_);
+#endif
+  }
+
+  // Also implement std Lockable
+  inline void lock() { Lock(); }
+  inline void unlock() { Unlock(); }
+  inline bool try_lock() { return TryLock(); }
+
+  // Mutex is move only with lock ownership transfer
+  Mutex(const Mutex&) = delete;
+  void operator=(const Mutex&) = delete;
+
+ private:
+  friend class CondVar;
+
+  std::mutex& getLock() { return mutex_; }
+
+  std::mutex mutex_;
+#ifndef NDEBUG
+  bool locked_;
+#endif
+};
+
+class RWMutex {
+ public:
+  RWMutex() { InitializeSRWLock(&srwLock_); }
+  // No copying allowed
+  RWMutex(const RWMutex&) = delete;
+  void operator=(const RWMutex&) = delete;
+
+  void ReadLock() { AcquireSRWLockShared(&srwLock_); }
+
+  void WriteLock() { AcquireSRWLockExclusive(&srwLock_); }
+
+  void ReadUnlock() { ReleaseSRWLockShared(&srwLock_); }
+
+  void WriteUnlock() { ReleaseSRWLockExclusive(&srwLock_); }
+
+  // Empty as in POSIX
+  void AssertHeld() {}
+
+ private:
+  SRWLOCK srwLock_;
+};
+
+class CondVar {
+ public:
+  explicit CondVar(Mutex* mu) : mu_(mu) {}
+
+  ~CondVar();
+  void Wait();
+  bool TimedWait(uint64_t expiration_time);
+  void Signal();
+  void SignalAll();
+
+  // Condition var is not copy/move constructible
+  CondVar(const CondVar&) = delete;
+  CondVar& operator=(const CondVar&) = delete;
+
+  CondVar(CondVar&&) = delete;
+  CondVar& operator=(CondVar&&) = delete;
+
+ private:
+  std::condition_variable cv_;
+  Mutex* mu_;
+};
+
+#ifdef _POSIX_THREADS
+using Thread = std::thread;
+#else
+// Wrapper around the platform efficient
+// or otherwise preferrable implementation
+using Thread = WindowsThread;
+#endif
+
+// OnceInit type helps emulate
+// Posix semantics with initialization
+// adopted in the project
+struct OnceType {
+  struct Init {};
+
+  OnceType() {}
+  OnceType(const Init&) {}
+  OnceType(const OnceType&) = delete;
+  OnceType& operator=(const OnceType&) = delete;
+
+  std::once_flag flag_;
+};
+
+#define LEVELDB_ONCE_INIT port::OnceType::Init()
+extern void InitOnce(OnceType* once, void (*initializer)());
+
+#ifndef CACHE_LINE_SIZE
+#define CACHE_LINE_SIZE 64U
+#endif
+
+#ifdef ROCKSDB_JEMALLOC
+// Separate inlines so they can be replaced if needed
+void* jemalloc_aligned_alloc(size_t size, size_t alignment) noexcept;
+void jemalloc_aligned_free(void* p) noexcept;
+#endif
+
+inline void* cacheline_aligned_alloc(size_t size) {
+#ifdef ROCKSDB_JEMALLOC
+  return jemalloc_aligned_alloc(size, CACHE_LINE_SIZE);
+#else
+  return _aligned_malloc(size, CACHE_LINE_SIZE);
+#endif
+}
+
+inline void cacheline_aligned_free(void* memblock) {
+#ifdef ROCKSDB_JEMALLOC
+  jemalloc_aligned_free(memblock);
+#else
+  _aligned_free(memblock);
+#endif
+}
+
+extern const size_t kPageSize;
+
+// Part of C++11
+#define ALIGN_AS(n) alignas(n)
+
+static inline void AsmVolatilePause() {
+#if defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) || defined(_M_ARM)
+  YieldProcessor();
+#endif
+  // it would be nice to get "wfe" on ARM here
+}
+
+extern int PhysicalCoreID();
+
+// For Thread Local Storage abstraction
+using pthread_key_t = DWORD;
+
+inline int pthread_key_create(pthread_key_t* key, void (*destructor)(void*)) {
+  // Not used
+  (void)destructor;
+
+  pthread_key_t k = TlsAlloc();
+  if (TLS_OUT_OF_INDEXES == k) {
+    return ENOMEM;
+  }
+
+  *key = k;
+  return 0;
+}
+
+inline int pthread_key_delete(pthread_key_t key) {
+  if (!TlsFree(key)) {
+    return EINVAL;
+  }
+  return 0;
+}
+
+inline int pthread_setspecific(pthread_key_t key, const void* value) {
+  if (!TlsSetValue(key, const_cast<void*>(value))) {
+    return ENOMEM;
+  }
+  return 0;
+}
+
+inline void* pthread_getspecific(pthread_key_t key) {
+  void* result = TlsGetValue(key);
+  if (!result) {
+    if (GetLastError() != ERROR_SUCCESS) {
+      errno = EINVAL;
+    } else {
+      errno = NOERROR;
+    }
+  }
+  return result;
+}
+
+// UNIX equiv although errno numbers will be off
+// using C-runtime to implement. Note, this does not
+// feel space with zeros in case the file is extended.
+int truncate(const char* path, int64_t length);
+int Truncate(std::string path, int64_t length);
+void Crash(const std::string& srcfile, int srcline);
+extern int GetMaxOpenFiles();
+std::string utf16_to_utf8(const std::wstring& utf16);
+std::wstring utf8_to_utf16(const std::string& utf8);
+
+using ThreadId = int;
+
+extern void SetCpuPriority(ThreadId id, CpuPriority priority);
+
+int64_t GetProcessID();
+
+// Uses platform APIs to generate a 36-character RFC-4122 UUID. Returns
+// true on success or false on failure.
+bool GenerateRfcUuid(std::string* output);
+
+}  // namespace port
+
+#ifdef ROCKSDB_WINDOWS_UTF8_FILENAMES
+
+#define RX_FILESTRING std::wstring
+#define RX_FN(a) ROCKSDB_NAMESPACE::port::utf8_to_utf16(a)
+#define FN_TO_RX(a) ROCKSDB_NAMESPACE::port::utf16_to_utf8(a)
+#define RX_FNCMP(a, b) ::wcscmp(a, RX_FN(b).c_str())
+#define RX_FNLEN(a) ::wcslen(a)
+
+#define RX_DeleteFile DeleteFileW
+#define RX_CreateFile CreateFileW
+#define RX_CreateFileMapping CreateFileMappingW
+#define RX_GetFileAttributesEx GetFileAttributesExW
+#define RX_FindFirstFileEx FindFirstFileExW
+#define RX_FindNextFile FindNextFileW
+#define RX_WIN32_FIND_DATA WIN32_FIND_DATAW
+#define RX_CreateDirectory CreateDirectoryW
+#define RX_RemoveDirectory RemoveDirectoryW
+#define RX_GetFileAttributesEx GetFileAttributesExW
+#define RX_MoveFileEx MoveFileExW
+#define RX_CreateHardLink CreateHardLinkW
+#define RX_PathIsRelative PathIsRelativeW
+#define RX_GetCurrentDirectory GetCurrentDirectoryW
+#define RX_GetDiskFreeSpaceEx GetDiskFreeSpaceExW
+#define RX_PathIsDirectory PathIsDirectoryW
+
+#else
+
+#define RX_FILESTRING std::string
+#define RX_FN(a) a
+#define FN_TO_RX(a) a
+#define RX_FNCMP(a, b) strcmp(a, b)
+#define RX_FNLEN(a) strlen(a)
+
+#define RX_DeleteFile DeleteFileA
+#define RX_CreateFile CreateFileA
+#define RX_CreateFileMapping CreateFileMappingA
+#define RX_GetFileAttributesEx GetFileAttributesExA
+#define RX_FindFirstFileEx FindFirstFileExA
+#define RX_CreateDirectory CreateDirectoryA
+#define RX_FindNextFile FindNextFileA
+#define RX_WIN32_FIND_DATA WIN32_FIND_DATAA
+#define RX_CreateDirectory CreateDirectoryA
+#define RX_RemoveDirectory RemoveDirectoryA
+#define RX_GetFileAttributesEx GetFileAttributesExA
+#define RX_MoveFileEx MoveFileExA
+#define RX_CreateHardLink CreateHardLinkA
+#define RX_PathIsRelative PathIsRelativeA
+#define RX_GetCurrentDirectory GetCurrentDirectoryA
+#define RX_GetDiskFreeSpaceEx GetDiskFreeSpaceExA
+#define RX_PathIsDirectory PathIsDirectoryA
+
+#endif
+
+using port::pthread_getspecific;
+using port::pthread_key_create;
+using port::pthread_key_delete;
+using port::pthread_key_t;
+using port::pthread_setspecific;
+using port::truncate;
+
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/win/win_jemalloc.cc b/src/rocksdb/port/win/win_jemalloc.cc
new file mode 100644
index 000000000..cf38f55b7
--- /dev/null
+++ b/src/rocksdb/port/win/win_jemalloc.cc
@@ -0,0 +1,80 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+
+#ifndef ROCKSDB_JEMALLOC
+#error This file can only be part of jemalloc aware build
+#endif
+
+#include <stdexcept>
+
+#include "jemalloc/jemalloc.h"
+#include "port/win/port_win.h"
+
+#if defined(ZSTD) && defined(ZSTD_STATIC_LINKING_ONLY)
+#include <zstd.h>
+#if (ZSTD_VERSION_NUMBER >= 500)
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+void* JemallocAllocateForZSTD(void* /* opaque */, size_t size) {
+  return je_malloc(size);
+}
+void JemallocDeallocateForZSTD(void* /* opaque */, void* address) {
+  je_free(address);
+}
+ZSTD_customMem GetJeZstdAllocationOverrides() {
+  return {JemallocAllocateForZSTD, JemallocDeallocateForZSTD, nullptr};
+}
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+#endif  // (ZSTD_VERSION_NUMBER >= 500)
+#endif  // defined(ZSTD) defined(ZSTD_STATIC_LINKING_ONLY)
+
+// Global operators to be replaced by a linker when this file is
+// a part of the build
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+void* jemalloc_aligned_alloc(size_t size, size_t alignment) noexcept {
+  return je_aligned_alloc(alignment, size);
+}
+void jemalloc_aligned_free(void* p) noexcept { je_free(p); }
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+void* operator new(size_t size) {
+  void* p = je_malloc(size);
+  if (!p) {
+    throw std::bad_alloc();
+  }
+  return p;
+}
+
+void* operator new[](size_t size) {
+  void* p = je_malloc(size);
+  if (!p) {
+    throw std::bad_alloc();
+  }
+  return p;
+}
+
+void operator delete(void* p) {
+  if (p) {
+    je_free(p);
+  }
+}
+
+void operator delete[](void* p) {
+  if (p) {
+    je_free(p);
+  }
+}
+
+#endif
diff --git a/src/rocksdb/port/win/win_logger.cc b/src/rocksdb/port/win/win_logger.cc
new file mode 100644
index 000000000..072ea419a
--- /dev/null
+++ b/src/rocksdb/port/win/win_logger.cc
@@ -0,0 +1,192 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// Logger implementation that can be shared by all environments
+// where enough posix functionality is available.
+
+#if defined(OS_WIN)
+
+#include "port/win/win_logger.h"
+
+#include <fcntl.h>
+#include <stdio.h>
+#include <time.h>
+
+#include <algorithm>
+#include <atomic>
+
+#include "monitoring/iostats_context_imp.h"
+#include "port/sys_time.h"
+#include "port/win/env_win.h"
+#include "port/win/io_win.h"
+#include "rocksdb/env.h"
+#include "rocksdb/system_clock.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+namespace port {
+
+WinLogger::WinLogger(uint64_t (*gettid)(), SystemClock* clock, HANDLE file,
+                     const InfoLogLevel log_level)
+    : Logger(log_level),
+      file_(file),
+      gettid_(gettid),
+      log_size_(0),
+      last_flush_micros_(0),
+      clock_(clock),
+      flush_pending_(false) {
+  assert(file_ != NULL);
+  assert(file_ != INVALID_HANDLE_VALUE);
+}
+
+void WinLogger::DebugWriter(const char* str, int len) {
+  assert(file_ != INVALID_HANDLE_VALUE);
+  DWORD bytesWritten = 0;
+  BOOL ret = WriteFile(file_, str, len, &bytesWritten, NULL);
+  if (ret == FALSE) {
+    std::string errSz = GetWindowsErrSz(GetLastError());
+    fprintf(stderr, "%s", errSz.c_str());
+  }
+}
+
+WinLogger::~WinLogger() { CloseInternal().PermitUncheckedError(); }
+
+Status WinLogger::CloseImpl() { return CloseInternal(); }
+
+Status WinLogger::CloseInternal() {
+  Status s;
+  if (INVALID_HANDLE_VALUE != file_) {
+    BOOL ret = FlushFileBuffers(file_);
+    if (ret == 0) {
+      auto lastError = GetLastError();
+      s = IOErrorFromWindowsError("Failed to flush LOG on Close() ", lastError);
+    }
+    ret = CloseHandle(file_);
+    // On error the return value is zero
+    if (ret == 0 && s.ok()) {
+      auto lastError = GetLastError();
+      s = IOErrorFromWindowsError("Failed to flush LOG on Close() ", lastError);
+    }
+    file_ = INVALID_HANDLE_VALUE;
+    closed_ = true;
+  }
+  return s;
+}
+
+void WinLogger::Flush() {
+  assert(file_ != INVALID_HANDLE_VALUE);
+  if (flush_pending_) {
+    flush_pending_ = false;
+    // With Windows API writes go to OS buffers directly so no fflush needed
+    // unlike with C runtime API. We don't flush all the way to disk
+    // for perf reasons.
+  }
+
+  last_flush_micros_ = clock_->NowMicros();
+}
+
+void WinLogger::Logv(const char* format, va_list ap) {
+  IOSTATS_TIMER_GUARD(logger_nanos);
+  assert(file_ != INVALID_HANDLE_VALUE);
+
+  const uint64_t thread_id = (*gettid_)();
+
+  // We try twice: the first time with a fixed-size stack allocated buffer,
+  // and the second time with a much larger dynamically allocated buffer.
+  char buffer[500];
+  std::unique_ptr<char[]> largeBuffer;
+  for (int iter = 0; iter < 2; ++iter) {
+    char* base;
+    int bufsize;
+    if (iter == 0) {
+      bufsize = sizeof(buffer);
+      base = buffer;
+    } else {
+      bufsize = 30000;
+      largeBuffer.reset(new char[bufsize]);
+      base = largeBuffer.get();
+    }
+
+    char* p = base;
+    char* limit = base + bufsize;
+
+    port::TimeVal now_tv;
+    port::GetTimeOfDay(&now_tv, nullptr);
+    const time_t seconds = now_tv.tv_sec;
+    struct tm t;
+    localtime_s(&t, &seconds);
+    p += snprintf(p, limit - p, "%04d/%02d/%02d-%02d:%02d:%02d.%06d %llx ",
+                  t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, t.tm_hour,
+                  t.tm_min, t.tm_sec, static_cast<int>(now_tv.tv_usec),
+                  static_cast<long long unsigned int>(thread_id));
+
+    // Print the message
+    if (p < limit) {
+      va_list backup_ap;
+      va_copy(backup_ap, ap);
+      int done = vsnprintf(p, limit - p, format, backup_ap);
+      if (done > 0) {
+        p += done;
+      } else {
+        continue;
+      }
+      va_end(backup_ap);
+    }
+
+    // Truncate to available space if necessary
+    if (p >= limit) {
+      if (iter == 0) {
+        continue;  // Try again with larger buffer
+      } else {
+        p = limit - 1;
+      }
+    }
+
+    // Add newline if necessary
+    if (p == base || p[-1] != '\n') {
+      *p++ = '\n';
+    }
+
+    assert(p <= limit);
+    const size_t write_size = p - base;
+
+    DWORD bytesWritten = 0;
+    BOOL ret = WriteFile(file_, base, static_cast<DWORD>(write_size),
+                         &bytesWritten, NULL);
+    if (ret == FALSE) {
+      std::string errSz = GetWindowsErrSz(GetLastError());
+      fprintf(stderr, "%s", errSz.c_str());
+    }
+
+    flush_pending_ = true;
+    assert((bytesWritten == write_size) || (ret == FALSE));
+    if (bytesWritten > 0) {
+      log_size_ += write_size;
+    }
+
+    uint64_t now_micros =
+        static_cast<uint64_t>(now_tv.tv_sec) * 1000000 + now_tv.tv_usec;
+    if (now_micros - last_flush_micros_ >= flush_every_seconds_ * 1000000) {
+      flush_pending_ = false;
+      // With Windows API writes go to OS buffers directly so no fflush needed
+      // unlike with C runtime API. We don't flush all the way to disk
+      // for perf reasons.
+      last_flush_micros_ = now_micros;
+    }
+    break;
+  }
+}
+
+size_t WinLogger::GetLogFileSize() const { return log_size_; }
+
+}  // namespace port
+
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
diff --git a/src/rocksdb/port/win/win_logger.h b/src/rocksdb/port/win/win_logger.h
new file mode 100644
index 000000000..1ca4610e9
--- /dev/null
+++ b/src/rocksdb/port/win/win_logger.h
@@ -0,0 +1,64 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+//
+// Logger implementation that can be shared by all environments
+// where enough posix functionality is available.
+
+#pragma once
+
+#include <stdint.h>
+#include <windows.h>
+
+#include <atomic>
+#include <memory>
+
+#include "rocksdb/env.h"
+
+namespace ROCKSDB_NAMESPACE {
+class SystemClock;
+
+namespace port {
+class WinLogger : public ROCKSDB_NAMESPACE::Logger {
+ public:
+  WinLogger(uint64_t (*gettid)(), SystemClock* clock, HANDLE file,
+            const InfoLogLevel log_level = InfoLogLevel::ERROR_LEVEL);
+
+  virtual ~WinLogger();
+
+  WinLogger(const WinLogger&) = delete;
+
+  WinLogger& operator=(const WinLogger&) = delete;
+
+  void Flush() override;
+
+  using ROCKSDB_NAMESPACE::Logger::Logv;
+  void Logv(const char* format, va_list ap) override;
+
+  size_t GetLogFileSize() const override;
+
+  void DebugWriter(const char* str, int len);
+
+ protected:
+  Status CloseImpl() override;
+
+ private:
+  HANDLE file_;
+  uint64_t (*gettid_)();  // Return the thread id for the current thread
+  std::atomic_size_t log_size_;
+  std::atomic_uint_fast64_t last_flush_micros_;
+  SystemClock* clock_;
+  bool flush_pending_;
+
+  Status CloseInternal();
+
+  const static uint64_t flush_every_seconds_ = 5;
+};
+}  // namespace port
+
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/win/win_thread.cc b/src/rocksdb/port/win/win_thread.cc
new file mode 100644
index 000000000..3c82e736e
--- /dev/null
+++ b/src/rocksdb/port/win/win_thread.cc
@@ -0,0 +1,170 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+// Most Mingw builds support std::thread only when using posix threads.
+// In that case, some of these functions will be unavailable.
+// Note that we're using either WindowsThread or std::thread, depending on
+// which one is available.
+#ifndef _POSIX_THREADS
+
+#include "port/win/win_thread.h"
+
+#include <assert.h>
+#include <process.h>  // __beginthreadex
+#include <windows.h>
+
+#include <stdexcept>
+#include <system_error>
+#include <thread>
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+struct WindowsThread::Data {
+  std::function<void()> func_;
+  uintptr_t handle_;
+
+  Data(std::function<void()>&& func) : func_(std::move(func)), handle_(0) {}
+
+  Data(const Data&) = delete;
+  Data& operator=(const Data&) = delete;
+
+  static unsigned int __stdcall ThreadProc(void* arg);
+};
+
+void WindowsThread::Init(std::function<void()>&& func) {
+  data_ = std::make_shared<Data>(std::move(func));
+  // We create another instance of std::shared_ptr to get an additional ref
+  // since we may detach and destroy this instance before the threadproc
+  // may start to run. We choose to allocate this additional ref on the heap
+  // so we do not need to synchronize and allow this thread to proceed
+  std::unique_ptr<std::shared_ptr<Data>> th_data(
+      new std::shared_ptr<Data>(data_));
+
+  data_->handle_ = _beginthreadex(NULL,
+                                  0,  // stack size
+                                  &Data::ThreadProc, th_data.get(),
+                                  0,  // init flag
+                                  &th_id_);
+
+  if (data_->handle_ == 0) {
+    throw std::system_error(
+        std::make_error_code(std::errc::resource_unavailable_try_again),
+        "Unable to create a thread");
+  }
+  th_data.release();
+}
+
+WindowsThread::WindowsThread() : data_(nullptr), th_id_(0) {}
+
+WindowsThread::~WindowsThread() {
+  // Must be joined or detached
+  // before destruction.
+  // This is the same as std::thread
+  if (data_) {
+    if (joinable()) {
+      assert(false);
+      std::terminate();
+    }
+    data_.reset();
+  }
+}
+
+WindowsThread::WindowsThread(WindowsThread&& o) noexcept : WindowsThread() {
+  *this = std::move(o);
+}
+
+WindowsThread& WindowsThread::operator=(WindowsThread&& o) noexcept {
+  if (joinable()) {
+    assert(false);
+    std::terminate();
+  }
+
+  data_ = std::move(o.data_);
+
+  // Per spec both instances will have the same id
+  th_id_ = o.th_id_;
+
+  return *this;
+}
+
+bool WindowsThread::joinable() const { return (data_ && data_->handle_ != 0); }
+
+WindowsThread::native_handle_type WindowsThread::native_handle() const {
+  return reinterpret_cast<native_handle_type>(data_->handle_);
+}
+
+unsigned WindowsThread::hardware_concurrency() {
+  return std::thread::hardware_concurrency();
+}
+
+void WindowsThread::join() {
+  if (!joinable()) {
+    assert(false);
+    throw std::system_error(std::make_error_code(std::errc::invalid_argument),
+                            "Thread is no longer joinable");
+  }
+
+  if (GetThreadId(GetCurrentThread()) == th_id_) {
+    assert(false);
+    throw std::system_error(
+        std::make_error_code(std::errc::resource_deadlock_would_occur),
+        "Can not join itself");
+  }
+
+  auto ret =
+      WaitForSingleObject(reinterpret_cast<HANDLE>(data_->handle_), INFINITE);
+  if (ret != WAIT_OBJECT_0) {
+    auto lastError = GetLastError();
+    assert(false);
+    throw std::system_error(static_cast<int>(lastError), std::system_category(),
+                            "WaitForSingleObjectFailed: thread join");
+  }
+
+  BOOL rc
+#if defined(_MSC_VER)
+      = FALSE;
+#else
+      __attribute__((__unused__));
+#endif
+  rc = CloseHandle(reinterpret_cast<HANDLE>(data_->handle_));
+  assert(rc != 0);
+  data_->handle_ = 0;
+}
+
+bool WindowsThread::detach() {
+  if (!joinable()) {
+    assert(false);
+    throw std::system_error(std::make_error_code(std::errc::invalid_argument),
+                            "Thread is no longer available");
+  }
+
+  BOOL ret = CloseHandle(reinterpret_cast<HANDLE>(data_->handle_));
+  data_->handle_ = 0;
+
+  return (ret != 0);
+}
+
+void WindowsThread::swap(WindowsThread& o) {
+  data_.swap(o.data_);
+  std::swap(th_id_, o.th_id_);
+}
+
+unsigned int __stdcall WindowsThread::Data::ThreadProc(void* arg) {
+  auto ptr = reinterpret_cast<std::shared_ptr<Data>*>(arg);
+  std::unique_ptr<std::shared_ptr<Data>> data(ptr);
+  (*data)->func_();
+  return 0;
+}
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif  // !_POSIX_THREADS
+#endif  // OS_WIN
diff --git a/src/rocksdb/port/win/win_thread.h b/src/rocksdb/port/win/win_thread.h
new file mode 100644
index 000000000..916033b77
--- /dev/null
+++ b/src/rocksdb/port/win/win_thread.h
@@ -0,0 +1,117 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#pragma once
+
+#ifndef _POSIX_THREADS
+
+#include <functional>
+#include <memory>
+#include <type_traits>
+
+#include "rocksdb/rocksdb_namespace.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+
+// This class is a replacement for std::thread
+// 2 reasons we do not like std::thread:
+//  -- is that it dynamically allocates its internals that are automatically
+//     freed when  the thread terminates and not on the destruction of the
+//     object. This makes it difficult to control the source of memory
+//     allocation
+//  -  This implements Pimpl so we can easily replace the guts of the
+//      object in our private version if necessary.
+class WindowsThread {
+  struct Data;
+
+  std::shared_ptr<Data> data_;
+  unsigned int th_id_;
+
+  void Init(std::function<void()>&&);
+
+ public:
+  using native_handle_type = void*;
+
+  // Construct with no thread
+  WindowsThread();
+
+  // Template constructor
+  //
+  // This templated constructor accomplishes several things
+  //
+  // - Allows the class as whole to be not a template
+  //
+  // - take "universal" references to support both _lvalues and _rvalues
+  //
+  // -  because this constructor is a catchall case in many respects it
+  //    may prevent us from using both the default __ctor, the move __ctor.
+  //    Also it may circumvent copy __ctor deletion. To work around this
+  //    we make sure this one has at least one argument and eliminate
+  //    it from the overload  selection when WindowsThread is the first
+  //    argument.
+  //
+  // - construct with Fx(Ax...) with a variable number of types/arguments.
+  //
+  // - Gathers together the callable object with its arguments and constructs
+  //   a single callable entity
+  //
+  // - Makes use of std::function to convert it to a specification-template
+  //   dependent type that both checks the signature conformance to ensure
+  //   that all of the necessary arguments are provided and allows pimpl
+  //   implementation.
+  template <class Fn, class... Args,
+            class = typename std::enable_if<!std::is_same<
+                typename std::decay<Fn>::type, WindowsThread>::value>::type>
+  explicit WindowsThread(Fn&& fx, Args&&... ax) : WindowsThread() {
+    // Use binder to create a single callable entity
+    auto binder = std::bind(std::forward<Fn>(fx), std::forward<Args>(ax)...);
+    // Use std::function to take advantage of the type erasure
+    // so we can still hide implementation within pimpl
+    // This also makes sure that the binder signature is compliant
+    std::function<void()> target = binder;
+
+    Init(std::move(target));
+  }
+
+  ~WindowsThread();
+
+  WindowsThread(const WindowsThread&) = delete;
+
+  WindowsThread& operator=(const WindowsThread&) = delete;
+
+  WindowsThread(WindowsThread&&) noexcept;
+
+  WindowsThread& operator=(WindowsThread&&) noexcept;
+
+  bool joinable() const;
+
+  unsigned int get_id() const { return th_id_; }
+
+  native_handle_type native_handle() const;
+
+  static unsigned hardware_concurrency();
+
+  void join();
+
+  bool detach();
+
+  void swap(WindowsThread&);
+};
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+namespace std {
+inline void swap(ROCKSDB_NAMESPACE::port::WindowsThread& th1,
+                 ROCKSDB_NAMESPACE::port::WindowsThread& th2) {
+  th1.swap(th2);
+}
+}  // namespace std
+
+#endif  // !_POSIX_THREADS
diff --git a/src/rocksdb/port/win/xpress_win.cc b/src/rocksdb/port/win/xpress_win.cc
new file mode 100644
index 000000000..21904d502
--- /dev/null
+++ b/src/rocksdb/port/win/xpress_win.cc
@@ -0,0 +1,210 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#if defined(OS_WIN)
+
+#include "port/win/xpress_win.h"
+
+#include <windows.h>
+
+#include <cassert>
+#include <iostream>
+#include <limits>
+#include <memory>
+
+#ifdef XPRESS
+
+// Put this under ifdef so windows systems w/o this
+// can still build
+#include <compressapi.h>
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+namespace xpress {
+
+// Helpers
+namespace {
+
+auto CloseCompressorFun = [](void* h) {
+  if (NULL != h) {
+    ::CloseCompressor(reinterpret_cast<COMPRESSOR_HANDLE>(h));
+  }
+};
+
+auto CloseDecompressorFun = [](void* h) {
+  if (NULL != h) {
+    ::CloseDecompressor(reinterpret_cast<DECOMPRESSOR_HANDLE>(h));
+  }
+};
+}  // namespace
+
+bool Compress(const char* input, size_t length, std::string* output) {
+  assert(input != nullptr);
+  assert(output != nullptr);
+
+  if (length == 0) {
+    output->clear();
+    return true;
+  }
+
+  COMPRESS_ALLOCATION_ROUTINES* allocRoutinesPtr = nullptr;
+
+  COMPRESSOR_HANDLE compressor = NULL;
+
+  BOOL success =
+      CreateCompressor(COMPRESS_ALGORITHM_XPRESS,  //  Compression Algorithm
+                       allocRoutinesPtr,  //  Optional allocation routine
+                       &compressor);      //  Handle
+
+  if (!success) {
+#ifdef _DEBUG
+    std::cerr << "XPRESS: Failed to create Compressor LastError: "
+              << GetLastError() << std::endl;
+#endif
+    return false;
+  }
+
+  std::unique_ptr<void, decltype(CloseCompressorFun)> compressorGuard(
+      compressor, CloseCompressorFun);
+
+  SIZE_T compressedBufferSize = 0;
+
+  //  Query compressed buffer size.
+  success = ::Compress(compressor,                //  Compressor Handle
+                       const_cast<char*>(input),  //  Input buffer
+                       length,                    //  Uncompressed data size
+                       NULL,                      //  Compressed Buffer
+                       0,                         //  Compressed Buffer size
+                       &compressedBufferSize);    //  Compressed Data size
+
+  if (!success) {
+    auto lastError = GetLastError();
+
+    if (lastError != ERROR_INSUFFICIENT_BUFFER) {
+#ifdef _DEBUG
+      std::cerr
+          << "XPRESS: Failed to estimate compressed buffer size LastError "
+          << lastError << std::endl;
+#endif
+      return false;
+    }
+  }
+
+  assert(compressedBufferSize > 0);
+
+  std::string result;
+  result.resize(compressedBufferSize);
+
+  SIZE_T compressedDataSize = 0;
+
+  //  Compress
+  success = ::Compress(compressor,                //  Compressor Handle
+                       const_cast<char*>(input),  //  Input buffer
+                       length,                    //  Uncompressed data size
+                       &result[0],                //  Compressed Buffer
+                       compressedBufferSize,      //  Compressed Buffer size
+                       &compressedDataSize);      //  Compressed Data size
+
+  if (!success) {
+#ifdef _DEBUG
+    std::cerr << "XPRESS: Failed to compress LastError " << GetLastError()
+              << std::endl;
+#endif
+    return false;
+  }
+
+  result.resize(compressedDataSize);
+  output->swap(result);
+
+  return true;
+}
+
+char* Decompress(const char* input_data, size_t input_length,
+                 size_t* uncompressed_size) {
+  assert(input_data != nullptr);
+  assert(uncompressed_size != nullptr);
+
+  if (input_length == 0) {
+    return nullptr;
+  }
+
+  COMPRESS_ALLOCATION_ROUTINES* allocRoutinesPtr = nullptr;
+
+  DECOMPRESSOR_HANDLE decompressor = NULL;
+
+  BOOL success =
+      CreateDecompressor(COMPRESS_ALGORITHM_XPRESS,  //  Compression Algorithm
+                         allocRoutinesPtr,  //  Optional allocation routine
+                         &decompressor);    //  Handle
+
+  if (!success) {
+#ifdef _DEBUG
+    std::cerr << "XPRESS: Failed to create Decompressor LastError "
+              << GetLastError() << std::endl;
+#endif
+    return nullptr;
+  }
+
+  std::unique_ptr<void, decltype(CloseDecompressorFun)> compressorGuard(
+      decompressor, CloseDecompressorFun);
+
+  SIZE_T decompressedBufferSize = 0;
+
+  success = ::Decompress(decompressor,                   //  Compressor Handle
+                         const_cast<char*>(input_data),  //  Compressed data
+                         input_length,              //  Compressed data size
+                         NULL,                      //  Buffer set to NULL
+                         0,                         //  Buffer size set to 0
+                         &decompressedBufferSize);  //  Decompressed Data size
+
+  if (!success) {
+    auto lastError = GetLastError();
+
+    if (lastError != ERROR_INSUFFICIENT_BUFFER) {
+#ifdef _DEBUG
+      std::cerr
+          << "XPRESS: Failed to estimate decompressed buffer size LastError "
+          << lastError << std::endl;
+#endif
+      return nullptr;
+    }
+  }
+
+  assert(decompressedBufferSize > 0);
+
+  // The callers are deallocating using delete[]
+  // thus we must allocate with new[]
+  std::unique_ptr<char[]> outputBuffer(new char[decompressedBufferSize]);
+
+  SIZE_T decompressedDataSize = 0;
+
+  success = ::Decompress(decompressor, const_cast<char*>(input_data),
+                         input_length, outputBuffer.get(),
+                         decompressedBufferSize, &decompressedDataSize);
+
+  if (!success) {
+#ifdef _DEBUG
+    std::cerr << "XPRESS: Failed to decompress LastError " << GetLastError()
+              << std::endl;
+#endif
+    return nullptr;
+  }
+
+  *uncompressed_size = decompressedDataSize;
+
+  // Return the raw buffer to the caller supporting the tradition
+  return outputBuffer.release();
+}
+}  // namespace xpress
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
+
+#endif
+
+#endif
diff --git a/src/rocksdb/port/win/xpress_win.h b/src/rocksdb/port/win/xpress_win.h
new file mode 100644
index 000000000..187adffa6
--- /dev/null
+++ b/src/rocksdb/port/win/xpress_win.h
@@ -0,0 +1,26 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#pragma once
+
+#include <string>
+
+#include "rocksdb/rocksdb_namespace.h"
+
+namespace ROCKSDB_NAMESPACE {
+namespace port {
+namespace xpress {
+
+bool Compress(const char* input, size_t length, std::string* output);
+
+char* Decompress(const char* input_data, size_t input_length,
+                 size_t* uncompressed_size);
+}  // namespace xpress
+}  // namespace port
+}  // namespace ROCKSDB_NAMESPACE
diff --git a/src/rocksdb/port/xpress.h b/src/rocksdb/port/xpress.h
new file mode 100644
index 000000000..457025f66
--- /dev/null
+++ b/src/rocksdb/port/xpress.h
@@ -0,0 +1,17 @@
+//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root 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.
+
+#pragma once
+
+// Xpress on Windows is implemeted using Win API
+#if defined(ROCKSDB_PLATFORM_POSIX)
+#error "Xpress compression not implemented"
+#elif defined(OS_WIN)
+#include "port/win/xpress_win.h"
+#endif