X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=engines%2Fwindowsaio.c;h=13d7f19402a1759fa8c28f7e11b6e13686098e93;hp=f66b9a2ec3465dee7c2d6b5b7f4c0e23cb9f9828;hb=ce4d13ca162df4127ec3b5911553802c53396705;hpb=9b8365618309572d8fd2579c8ea3132db89f843f

diff --git a/engines/windowsaio.c b/engines/windowsaio.c
index f66b9a2e..13d7f194 100644
--- a/engines/windowsaio.c
+++ b/engines/windowsaio.c
@@ -1,6 +1,7 @@
 /*
- * Native Windows async IO engine
- * Copyright (C) 2010 Bruce Cran <bruce@cran.org.uk>
+ * windowsaio engine
+ *
+ * IO engine using Windows IO Completion Ports.
  */
 
 #include <stdio.h>
@@ -8,27 +9,25 @@
 #include <unistd.h>
 #include <signal.h>
 #include <errno.h>
-#include <windows.h>
 
 #include "../fio.h"
 
-typedef struct {
- OVERLAPPED o;
- struct io_u *io_u;
-} FIO_OVERLAPPED;
+typedef BOOL (WINAPI *CANCELIOEX)(HANDLE hFile, LPOVERLAPPED lpOverlapped);
 
-struct windowsaio_data {
-	HANDLE *io_handles;
-	unsigned int io_index;
-	FIO_OVERLAPPED *ovls;
+int geterrno_from_win_error (DWORD code, int deferrno);
 
-	HANDLE iothread;
-	HANDLE iothread_stopped;
-	BOOL iothread_running;
+struct fio_overlapped {
+	OVERLAPPED o;
+	struct io_u *io_u;
+	BOOL io_complete;
+};
 
+struct windowsaio_data {
 	struct io_u **aio_events;
+	HANDLE iocp;
+	HANDLE iothread;
 	HANDLE iocomplete_event;
-	BOOL have_cancelioex;
+	BOOL iothread_running;
 };
 
 struct thread_ctx {
@@ -36,192 +35,86 @@ struct thread_ctx {
 	struct windowsaio_data *wd;
 };
 
-static void PrintError(LPCSTR lpszFunction);
-static int fio_windowsaio_cancel(struct thread_data *td,
-			       struct io_u *io_u);
-static BOOL TimedOut(DWORD start_count, DWORD end_count);
-static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
-				    unsigned int max, struct timespec *t);
-static struct io_u *fio_windowsaio_event(struct thread_data *td, int event);
-static int fio_windowsaio_queue(struct thread_data *td,
-			      struct io_u *io_u);
-static void fio_windowsaio_cleanup(struct thread_data *td);
 static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter);
-static int fio_windowsaio_init(struct thread_data *td);
-static int fio_windowsaio_open_file(struct thread_data *td, struct fio_file *f);
-static int fio_windowsaio_close_file(struct thread_data fio_unused *td, struct fio_file *f);
 
-/* CancelIoEx isn't in Cygwin's w32api */
-BOOL WINAPI CancelIoEx(
-  HANDLE hFile,
-  LPOVERLAPPED lpOverlapped
-);
-
-
-
-int sync_file_range(int fd, off64_t offset, off64_t nbytes,
-			   unsigned int flags)
-{
-	errno = ENOSYS;
-	return -1;
-}
-
-static void PrintError(LPCSTR lpszFunction)
-{
-	// Retrieve the system error message for the last-error code
-
-	LPSTR lpMsgBuf;
-	DWORD dw = GetLastError();
-
-	FormatMessage(
-		FORMAT_MESSAGE_ALLOCATE_BUFFER |
-		FORMAT_MESSAGE_FROM_SYSTEM |
-		FORMAT_MESSAGE_IGNORE_INSERTS,
-		NULL,
-		dw,
-		MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
-		(LPTSTR)&lpMsgBuf,
-		0, NULL );
-
-	log_err("%s - %s", lpszFunction, lpMsgBuf);
-	LocalFree(lpMsgBuf);
-}
-
-static int fio_windowsaio_cancel(struct thread_data *td,
-			       struct io_u *io_u)
+static int fio_windowsaio_init(struct thread_data *td)
 {
+	struct windowsaio_data *wd;
 	int rc = 0;
 
-	struct windowsaio_data *wd = td->io_ops->data;
-
-	/* If we're running on Vista, we can cancel individual IO requests */
-	if (wd->have_cancelioex) {
-		FIO_OVERLAPPED *ovl = io_u->engine_data;
-		if (!CancelIoEx(io_u->file->hFile, &ovl->o))
-			rc = 1;
-	} else
+	wd = calloc(1, sizeof(struct windowsaio_data));
+	if (wd == NULL) {
+		 log_err("windowsaio: failed to allocate memory for engine data\n");
 		rc = 1;
+	}
 
-	return rc;
-}
-
-static BOOL TimedOut(DWORD start_count, DWORD end_count)
-{
-	BOOL expired = FALSE;
-	DWORD current_time;
-
-	current_time = GetTickCount();
-
-	if ((end_count > start_count) && current_time >= end_count)
-		expired = TRUE;
-	else if (current_time < start_count && current_time > end_count)
-		expired = TRUE;
+	if (!rc) {
+		wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*));
+		if (wd->aio_events == NULL) {
+			log_err("windowsaio: failed to allocate memory for aio events list\n");
+			rc = 1;
+		}
+	}
 
-	return expired;
-}
+	if (!rc) {
+		/* Create an auto-reset event */
+		wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL);
+		if (wd->iocomplete_event == NULL) {
+			log_err("windowsaio: failed to create io complete event handle\n");
+			rc = 1;
+		}
+	}
 
-static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
-				    unsigned int max, struct timespec *t)
-{
-	struct windowsaio_data *wd = td->io_ops->data;
-	struct flist_head *entry;
-	unsigned int dequeued = 0;
-	struct io_u *io_u;
-	DWORD start_count = 0, end_count = 0;
-	BOOL timedout = FALSE;
-	unsigned int mswait = 100;
+	if (rc) {
+		if (wd != NULL) {
+			if (wd->aio_events != NULL)
+				free(wd->aio_events);
 
-	if (t != NULL) {
-		mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
-		start_count = GetTickCount();
-		end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
+			free(wd);
+		}
 	}
 
-	while (dequeued < min && !timedout) {
-		flist_for_each(entry, &td->io_u_busylist) {
-			io_u = flist_entry(entry, struct io_u, list);
-
-			if (io_u->seen == 1) {
-				io_u->seen = 2;
-				wd->aio_events[dequeued] = io_u;
-				dequeued++;
-			}
+	td->io_ops_data = wd;
 
-			if (dequeued == max)
-				break;
-		}
+	if (!rc) {
+		struct thread_ctx *ctx;
+		struct windowsaio_data *wd;
+		HANDLE hFile;
 
-		if (dequeued < min) {
-			WaitForSingleObject(wd->iocomplete_event, mswait);
+		hFile = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 0);
+		if (hFile == INVALID_HANDLE_VALUE) {
+			log_err("windowsaio: failed to create io completion port\n");
+			rc = 1;
 		}
 
-		if (t != NULL && TimedOut(start_count, end_count))
-			timedout = TRUE;
-	}
+		wd = td->io_ops_data;
+		wd->iothread_running = TRUE;
+		wd->iocp = hFile;
 
-	return dequeued;
-}
+		if (!rc)
+			ctx = malloc(sizeof(struct thread_ctx));
 
-static struct io_u *fio_windowsaio_event(struct thread_data *td, int event)
-{
-	struct windowsaio_data *wd = td->io_ops->data;
-	return wd->aio_events[event];
-}
+		if (!rc && ctx == NULL) {
+			log_err("windowsaio: failed to allocate memory for thread context structure\n");
+			CloseHandle(hFile);
+			rc = 1;
+		}
 
-static int fio_windowsaio_queue(struct thread_data *td,
-			      struct io_u *io_u)
-{
-	struct windowsaio_data *wd;
-	DWORD iobytes;
-	BOOL success = TRUE;
-	int ind;
-	int rc;
+		if (!rc) {
+			DWORD threadid;
 
-	fio_ro_check(td, io_u);
+			ctx->iocp = hFile;
+			ctx->wd = wd;
+			wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, &threadid);
 
-	wd = td->io_ops->data;
-	ind = wd->io_index;
-
-	ResetEvent(wd->io_handles[ind]);
-	wd->ovls[ind].o.Internal = 0;
-	wd->ovls[ind].o.InternalHigh = 0;
-	wd->ovls[ind].o.Offset = io_u->offset & 0xFFFFFFFF;
-	wd->ovls[ind].o.OffsetHigh = io_u->offset >> 32;
-	wd->ovls[ind].o.hEvent = wd->io_handles[ind];
-	wd->ovls[ind].io_u = io_u;
-
-	io_u->engine_data = &wd->ovls[ind];
-	io_u->seen = 0;
-
-	if (io_u->ddir == DDIR_WRITE) {
-		success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, &wd->ovls[ind].o);
-	} else if (io_u->ddir == DDIR_READ) {
-		success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, &wd->ovls[ind].o);
-	} else if (io_u->ddir == DDIR_SYNC     ||
-			 io_u->ddir == DDIR_DATASYNC ||
-			 io_u->ddir == DDIR_SYNC_FILE_RANGE)
-	{
-		FlushFileBuffers(io_u->file->hFile);
-		return FIO_Q_COMPLETED;
-	} else if (io_u->ddir == DDIR_TRIM) {
-		log_info("explicit TRIM isn't supported on Windows");
-		return FIO_Q_COMPLETED;
-	} else
-		assert(0);
+			if (wd->iothread != NULL)
+				fio_setaffinity(threadid, td->o.cpumask);
+			else
+				log_err("windowsaio: failed to create io completion thread\n");
+		}
 
-	if (success) {
-		io_u->seen = 1;
-		io_u->resid = io_u->xfer_buflen - iobytes;
-		io_u->error = 0;
-		rc = FIO_Q_COMPLETED;
-	} else if (!success && GetLastError() == ERROR_IO_PENDING) {
-		wd->io_index = (wd->io_index + 1) % (2 * td->o.iodepth);
-		rc = FIO_Q_QUEUED;
-	} else {
-		PrintError(__func__);
-		io_u->error = GetLastError();
-		io_u->resid = io_u->xfer_buflen;
-		rc = FIO_Q_COMPLETED;
+		if (rc || wd->iothread == NULL)
+			rc = 1;
 	}
 
 	return rc;
@@ -229,158 +122,79 @@ static int fio_windowsaio_queue(struct thread_data *td,
 
 static void fio_windowsaio_cleanup(struct thread_data *td)
 {
-	int i;
 	struct windowsaio_data *wd;
 
-	wd = td->io_ops->data;
-
-	WaitForSingleObject(wd->iothread_stopped, INFINITE);
+	wd = td->io_ops_data;
 
 	if (wd != NULL) {
+		wd->iothread_running = FALSE;
+		WaitForSingleObject(wd->iothread, INFINITE);
+
 		CloseHandle(wd->iothread);
-		CloseHandle(wd->iothread_stopped);
 		CloseHandle(wd->iocomplete_event);
 
-		for (i = 0; i < 2 * td->o.iodepth; i++) {
-			CloseHandle(wd->io_handles[i]);
-		}
-
 		free(wd->aio_events);
-		free(wd->io_handles);
-		free(wd->ovls);
 		free(wd);
 
-		td->io_ops->data = NULL;
-	}
-}
-
-/* Runs as a thread and waits for queued IO to complete */
-static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter)
-{
-	OVERLAPPED *ovl;
-	FIO_OVERLAPPED *fov;
-	struct io_u *io_u;
-	struct windowsaio_data *wd;
-	struct thread_ctx *ctx;
-	ULONG_PTR ulKey = 0;
-	DWORD bytes;
-
-	ctx = (struct thread_ctx*)lpParameter;
-	wd = ctx->wd;
-
-	while (ctx->wd->iothread_running) {
-		if (!GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey, &ovl, 250))
-			continue;
-
-		fov = CONTAINING_RECORD(ovl, FIO_OVERLAPPED, o);
-		io_u = fov->io_u;
-
-		if (io_u->seen != 0)
-			continue;
-
-		if (ovl->Internal == ERROR_SUCCESS) {
-			io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
-			io_u->error = 0;
-		} else {
-			io_u->resid = io_u->xfer_buflen;
-			io_u->error = ovl->Internal;
-		}
-
-		io_u->seen = 1;
-		SetEvent(wd->iocomplete_event);
+		td->io_ops_data = NULL;
 	}
-
-	CloseHandle(ctx->iocp);
-	SetEvent(ctx->wd->iothread_stopped);
-	free(ctx);
-
-	return 0;
 }
 
-static int fio_windowsaio_init(struct thread_data *td)
+static int windowsaio_invalidate_cache(struct fio_file *f)
 {
-	struct windowsaio_data *wd;
-	OSVERSIONINFO osInfo;
+	DWORD error;
+	DWORD isharemode = (FILE_SHARE_DELETE | FILE_SHARE_READ |
+				FILE_SHARE_WRITE);
+	HANDLE ihFile;
 	int rc = 0;
 
-	wd = malloc(sizeof(struct windowsaio_data));
-	if (wd != NULL)
-		ZeroMemory(wd, sizeof(struct windowsaio_data));
-	else
-		rc = 1;
-
-	if (!rc) {
-		wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*));
-		if (wd->aio_events == NULL)
+	/*
+	 * Encourage Windows to drop cached parts of a file by temporarily
+	 * opening it for non-buffered access. Note: this will only work when
+	 * the following is the only thing with the file open on the whole
+	 * system.
+	 */
+	dprint(FD_IO, "windowaio: attempt invalidate cache for %s\n",
+			f->file_name);
+	ihFile = CreateFile(f->file_name, 0, isharemode, NULL, OPEN_EXISTING,
+			FILE_FLAG_NO_BUFFERING, NULL);
+
+	if (ihFile != INVALID_HANDLE_VALUE) {
+		if (!CloseHandle(ihFile)) {
+			error = GetLastError();
+			log_info("windowsaio: invalidation fd close %s "
+				 "failed: error %d\n", f->file_name, error);
 			rc = 1;
-	}
-
-	if (!rc) {
-		wd->io_handles = malloc(2 * td->o.iodepth * sizeof(HANDLE));
-		if (wd->io_handles == NULL)
-			rc = 1;
-	}
-
-	if (!rc) {
-		wd->ovls = malloc(2 * td->o.iodepth * sizeof(FIO_OVERLAPPED));
-		if (wd->ovls == NULL)
-			rc = 1;
-	}
-
-	if (!rc) {
-		/* Create an auto-reset event */
-		wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL);
-		if (wd->iocomplete_event == NULL)
+		}
+	} else {
+		error = GetLastError();
+		if (error != ERROR_FILE_NOT_FOUND) {
+			log_info("windowsaio: cache invalidation of %s failed: "
+					"error %d\n", f->file_name, error);
 			rc = 1;
-	}
-
-	if (!rc) {
-		osInfo.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
-		GetVersionEx(&osInfo);
-
-		if (osInfo.dwMajorVersion >= 6)
-			wd->have_cancelioex = TRUE;
-		else
-			wd->have_cancelioex = FALSE;
-	}
-
-	if (rc) {
-		PrintError(__func__);
-		if (wd != NULL) {
-			if (wd->ovls != NULL)
-				free(wd->ovls);
-			if (wd->io_handles != NULL)
-				free(wd->io_handles);
-			if (wd->aio_events != NULL)
-				free(wd->aio_events);
-
-			free(wd);
 		}
 	}
 
-	td->io_ops->data = wd;
-	return 0;
+	return rc;
 }
 
 static int fio_windowsaio_open_file(struct thread_data *td, struct fio_file *f)
 {
 	int rc = 0;
-	HANDLE hFile;
-	DWORD flags = FILE_ATTRIBUTE_TEMPORARY | FILE_FLAG_POSIX_SEMANTICS | FILE_FLAG_OVERLAPPED;
+	DWORD flags = FILE_FLAG_POSIX_SEMANTICS | FILE_FLAG_OVERLAPPED;
 	DWORD sharemode = FILE_SHARE_READ | FILE_SHARE_WRITE;
 	DWORD openmode = OPEN_ALWAYS;
 	DWORD access;
-	int i;
 
 	dprint(FD_FILE, "fd open %s\n", f->file_name);
 
 	if (f->filetype == FIO_TYPE_PIPE) {
-		log_err("fio: windowsaio doesn't support pipes\n");
+		log_err("windowsaio: pipes are not supported\n");
 		return 1;
 	}
 
 	if (!strcmp(f->file_name, "-")) {
-		log_err("fio: can't read/write to stdin/out\n");
+		log_err("windowsaio: can't read/write to stdin/out\n");
 		return 1;
 	}
 
@@ -389,100 +203,272 @@ static int fio_windowsaio_open_file(struct thread_data *td, struct fio_file *f)
 	if (td->o.sync_io)
 		flags |= FILE_FLAG_WRITE_THROUGH;
 
-
-	if (td->o.td_ddir == TD_DDIR_READ  ||
-		td->o.td_ddir == TD_DDIR_WRITE ||
-		td->o.td_ddir == TD_DDIR_RANDRW)
-	{
-		flags |= FILE_FLAG_SEQUENTIAL_SCAN;
-	}
-	else
-	{
+	/*
+	 * Inform Windows whether we're going to be doing sequential or
+	 * random IO so it can tune the Cache Manager
+	 */
+	switch (td->o.fadvise_hint) {
+	case F_ADV_TYPE:
+		if (td_random(td))
+			flags |= FILE_FLAG_RANDOM_ACCESS;
+		else
+			flags |= FILE_FLAG_SEQUENTIAL_SCAN;
+		break;
+	case F_ADV_RANDOM:
 		flags |= FILE_FLAG_RANDOM_ACCESS;
+		break;
+	case F_ADV_SEQUENTIAL:
+		flags |= FILE_FLAG_SEQUENTIAL_SCAN;
+		break;
+	case F_ADV_NONE:
+		break;
+	default:
+		log_err("fio: unknown fadvise type %d\n", td->o.fadvise_hint);
 	}
 
-	if (td_read(td) || read_only)
+	if (!td_write(td) || read_only)
 		access = GENERIC_READ;
 	else
 		access = (GENERIC_READ | GENERIC_WRITE);
 
-	if (td->o.create_on_open > 0)
+	if (td->o.create_on_open)
 		openmode = OPEN_ALWAYS;
 	else
 		openmode = OPEN_EXISTING;
 
+	/* If we're going to use direct I/O, Windows will try and invalidate
+	 * its cache at that point so there's no need to do it here */
+	if (td->o.invalidate_cache && !td->o.odirect)
+		windowsaio_invalidate_cache(f);
+
 	f->hFile = CreateFile(f->file_name, access, sharemode,
 		NULL, openmode, flags, NULL);
 
 	if (f->hFile == INVALID_HANDLE_VALUE) {
-		PrintError(__func__);
+		log_err("windowsaio: failed to open file \"%s\"\n", f->file_name);
 		rc = 1;
 	}
 
-	/* Only set up the competion port and thread if we're not just
+	/* Only set up the completion port and thread if we're not just
 	 * querying the device size */
-	if (!rc && td->io_ops->data != NULL) {
+	if (!rc && td->io_ops_data != NULL) {
 		struct windowsaio_data *wd;
-		struct thread_ctx *ctx;
-		hFile = CreateIoCompletionPort(f->hFile, NULL, 0, 0);
 
-		wd = td->io_ops->data;
+		wd = td->io_ops_data;
 
-		wd->io_index = 0;
-		wd->iothread_running = TRUE;
-		/* Create a manual-reset event */
-		wd->iothread_stopped = CreateEvent(NULL, TRUE, FALSE, NULL);
+		if (CreateIoCompletionPort(f->hFile, wd->iocp, 0, 0) == NULL) {
+			log_err("windowsaio: failed to create io completion port\n");
+			rc = 1;
+		}
+	}
+
+	return rc;
+}
+
+static int fio_windowsaio_close_file(struct thread_data fio_unused *td, struct fio_file *f)
+{
+	int rc = 0;
+
+	dprint(FD_FILE, "fd close %s\n", f->file_name);
 
-		if (wd->iothread_stopped == NULL)
+	if (f->hFile != INVALID_HANDLE_VALUE) {
+		if (!CloseHandle(f->hFile)) {
+			log_info("windowsaio: failed to close file handle for \"%s\"\n", f->file_name);
 			rc = 1;
+		}
+	}
 
-		if (!rc) {
-			for (i = 0; i < 2 * td->o.iodepth; i++) {
-				/* Create a manual-reset event for putting in OVERLAPPED */
-				wd->io_handles[i] = CreateEvent(NULL, TRUE, FALSE, NULL);
-				if (wd->io_handles[i] == NULL) {
-					PrintError(__func__);
-					rc = 1;
-					break;
-				}
+	f->hFile = INVALID_HANDLE_VALUE;
+	return rc;
+}
+
+static BOOL timeout_expired(DWORD start_count, DWORD end_count)
+{
+	BOOL expired = FALSE;
+	DWORD current_time;
+
+	current_time = GetTickCount();
+
+	if ((end_count > start_count) && current_time >= end_count)
+		expired = TRUE;
+	else if (current_time < start_count && current_time > end_count)
+		expired = TRUE;
+
+	return expired;
+}
+
+static struct io_u* fio_windowsaio_event(struct thread_data *td, int event)
+{
+	struct windowsaio_data *wd = td->io_ops_data;
+	return wd->aio_events[event];
+}
+
+static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
+				    unsigned int max,
+				    const struct timespec *t)
+{
+	struct windowsaio_data *wd = td->io_ops_data;
+	unsigned int dequeued = 0;
+	struct io_u *io_u;
+	int i;
+	struct fio_overlapped *fov;
+	DWORD start_count = 0;
+	DWORD end_count = 0;
+	DWORD status;
+	DWORD mswait = 250;
+
+	if (t != NULL) {
+		mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
+		start_count = GetTickCount();
+		end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
+	}
+
+	do {
+		io_u_qiter(&td->io_u_all, io_u, i) {
+			if (!(io_u->flags & IO_U_F_FLIGHT))
+				continue;
+
+			fov = (struct fio_overlapped*)io_u->engine_data;
+
+			if (fov->io_complete) {
+				fov->io_complete = FALSE;
+				wd->aio_events[dequeued] = io_u;
+				dequeued++;
 			}
-		}
 
-		if (!rc) {
-			ctx = malloc(sizeof(struct thread_ctx));
-			ctx->iocp = hFile;
-			ctx->wd = wd;
+		}
+		if (dequeued >= min)
+			break;
 
-			wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, NULL);
+		if (dequeued < min) {
+			status = WaitForSingleObject(wd->iocomplete_event, mswait);
+			if (status != WAIT_OBJECT_0 && dequeued >= min)
+				break;
 		}
 
-		if (rc || wd->iothread == NULL) {
-			PrintError(__func__);
-			rc = 1;
+		if (dequeued >= min ||
+		    (t != NULL && timeout_expired(start_count, end_count)))
+			break;
+	} while (1);
+
+	return dequeued;
+}
+
+static enum fio_q_status fio_windowsaio_queue(struct thread_data *td,
+					      struct io_u *io_u)
+{
+	struct fio_overlapped *o = io_u->engine_data;
+	LPOVERLAPPED lpOvl = &o->o;
+	BOOL success = FALSE;
+	int rc = FIO_Q_COMPLETED;
+
+	fio_ro_check(td, io_u);
+
+	lpOvl->Internal = 0;
+	lpOvl->InternalHigh = 0;
+	lpOvl->Offset = io_u->offset & 0xFFFFFFFF;
+	lpOvl->OffsetHigh = io_u->offset >> 32;
+
+	switch (io_u->ddir) {
+	case DDIR_WRITE:
+		success = WriteFile(io_u->file->hFile, io_u->xfer_buf,
+					io_u->xfer_buflen, NULL, lpOvl);
+		break;
+	case DDIR_READ:
+		success = ReadFile(io_u->file->hFile, io_u->xfer_buf,
+					io_u->xfer_buflen, NULL, lpOvl);
+		break;
+	case DDIR_SYNC:
+	case DDIR_DATASYNC:
+	case DDIR_SYNC_FILE_RANGE:
+		success = FlushFileBuffers(io_u->file->hFile);
+		if (!success) {
+			log_err("windowsaio: failed to flush file buffers\n");
+			io_u->error = win_to_posix_error(GetLastError());
 		}
+
+		return FIO_Q_COMPLETED;
+	case DDIR_TRIM:
+		log_err("windowsaio: manual TRIM isn't supported on Windows\n");
+		io_u->error = 1;
+		io_u->resid = io_u->xfer_buflen;
+		return FIO_Q_COMPLETED;
+	default:
+		assert(0);
+		break;
+	}
+
+	if (success || GetLastError() == ERROR_IO_PENDING)
+		rc = FIO_Q_QUEUED;
+	else {
+		io_u->error = win_to_posix_error(GetLastError());
+		io_u->resid = io_u->xfer_buflen;
 	}
 
 	return rc;
 }
 
-static int fio_windowsaio_close_file(struct thread_data fio_unused *td, struct fio_file *f)
+/* Runs as a thread and waits for queued IO to complete */
+static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter)
 {
+	OVERLAPPED *ovl;
+	struct fio_overlapped *fov;
+	struct io_u *io_u;
 	struct windowsaio_data *wd;
+	struct thread_ctx *ctx;
+	ULONG_PTR ulKey = 0;
+	DWORD bytes;
 
-	dprint(FD_FILE, "fd close %s\n", f->file_name);
+	ctx = (struct thread_ctx*)lpParameter;
+	wd = ctx->wd;
 
-	if (td->io_ops->data != NULL) {
-		wd = td->io_ops->data;
-		wd->iothread_running = FALSE;
-		WaitForSingleObject(wd->iothread_stopped, INFINITE);
-	}
+	do {
+		BOOL ret;
 
-	if (f->hFile != INVALID_HANDLE_VALUE) {
-		if (!CloseHandle(f->hFile))
-			PrintError(__func__);
+		ret = GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey,
+						&ovl, 250);
+		if (!ret && ovl == NULL)
+			continue;
+
+		fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o);
+		io_u = fov->io_u;
+
+		if (ovl->Internal == ERROR_SUCCESS) {
+			io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
+			io_u->error = 0;
+		} else {
+			io_u->resid = io_u->xfer_buflen;
+			io_u->error = win_to_posix_error(GetLastError());
+		}
+
+		fov->io_complete = TRUE;
+		SetEvent(wd->iocomplete_event);
+	} while (ctx->wd->iothread_running);
+
+	CloseHandle(ctx->iocp);
+	free(ctx);
+	return 0;
+}
+
+static void fio_windowsaio_io_u_free(struct thread_data *td, struct io_u *io_u)
+{
+	struct fio_overlapped *o = io_u->engine_data;
+
+	if (o) {
+		io_u->engine_data = NULL;
+		free(o);
 	}
+}
 
-	f->hFile = INVALID_HANDLE_VALUE;
+static int fio_windowsaio_io_u_init(struct thread_data *td, struct io_u *io_u)
+{
+	struct fio_overlapped *o;
+
+	o = malloc(sizeof(*o));
+	o->io_complete = FALSE;
+	o->io_u = io_u;
+	o->o.hEvent = NULL;
+	io_u->engine_data = o;
 	return 0;
 }
 
@@ -491,21 +477,22 @@ static struct ioengine_ops ioengine = {
 	.version	= FIO_IOOPS_VERSION,
 	.init		= fio_windowsaio_init,
 	.queue		= fio_windowsaio_queue,
-	.cancel		= fio_windowsaio_cancel,
 	.getevents	= fio_windowsaio_getevents,
 	.event		= fio_windowsaio_event,
 	.cleanup	= fio_windowsaio_cleanup,
 	.open_file	= fio_windowsaio_open_file,
 	.close_file	= fio_windowsaio_close_file,
-	.get_file_size	= generic_get_file_size
+	.get_file_size	= generic_get_file_size,
+	.io_u_init	= fio_windowsaio_io_u_init,
+	.io_u_free	= fio_windowsaio_io_u_free,
 };
 
-static void fio_init fio_posixaio_register(void)
+static void fio_init fio_windowsaio_register(void)
 {
 	register_ioengine(&ioengine);
 }
 
-static void fio_exit fio_posixaio_unregister(void)
+static void fio_exit fio_windowsaio_unregister(void)
 {
 	unregister_ioengine(&ioengine);
 }