[fio.git] / engines / windowsaio.c

/*
 * windowsaio engine
 *
 * IO engine using Windows IO Completion Ports.
 */

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <signal.h>
#include <errno.h>
#include <windows.h>

#include "../fio.h"

typedef BOOL (WINAPI *CANCELIOEX)(HANDLE hFile, LPOVERLAPPED lpOverlapped);

int geterrno_from_win_error (DWORD code, int deferrno);

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;
	CANCELIOEX pCancelIoEx;
	BOOL iothread_running;
};

struct thread_ctx {
	HANDLE iocp;
	struct windowsaio_data *wd;
};

static int fio_windowsaio_cancel(struct thread_data *td,
				   struct io_u *io_u);
static BOOL timeout_expired(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);

static int fio_windowsaio_init(struct thread_data *td)
{
	struct windowsaio_data *wd;
	HANDLE hKernel32Dll;
	int rc = 0;

	wd = calloc(1, sizeof(struct windowsaio_data));
	if (wd == NULL) {
		 log_err("windowsaio: failed to allocate memory for engine data\n");
		rc = 1;
	}

	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;
		}
	}

	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;
		}
	}

	if (rc) {
		if (wd != NULL) {
			if (wd->aio_events != NULL)
				free(wd->aio_events);

			free(wd);
		}
	}

	hKernel32Dll = GetModuleHandle("kernel32.dll");
	wd->pCancelIoEx = (CANCELIOEX)GetProcAddress(hKernel32Dll, "CancelIoEx");
	td->io_ops->data = wd;

	if (!rc) {
		struct thread_ctx *ctx;
		struct windowsaio_data *wd;
		HANDLE hFile;

		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;
		}

		wd = td->io_ops->data;
		wd->iothread_running = TRUE;
		wd->iocp = hFile;

		if (!rc)
			ctx = malloc(sizeof(struct thread_ctx));

		if (!rc && ctx == NULL)
		{
			log_err("windowsaio: failed to allocate memory for thread context structure\n");
			CloseHandle(hFile);
			rc = 1;
		}

		if (!rc)
		{
			ctx->iocp = hFile;
			ctx->wd = wd;
			wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, NULL);
			if (wd->iothread == NULL)
				log_err("windowsaio: failed to create io completion thread\n");
		}

		if (rc || wd->iothread == NULL)
			rc = 1;
	}

	return rc;
}

static void fio_windowsaio_cleanup(struct thread_data *td)
{
	struct windowsaio_data *wd;

	wd = td->io_ops->data;

	if (wd != NULL) {
		wd->iothread_running = FALSE;
		WaitForSingleObject(wd->iothread, INFINITE);

		CloseHandle(wd->iothread);
		CloseHandle(wd->iocomplete_event);

		free(wd->aio_events);
		free(wd);

		td->io_ops->data = NULL;
	}
}


static int fio_windowsaio_open_file(struct thread_data *td, struct fio_file *f)
{
	int rc = 0;
	DWORD flags = FILE_FLAG_POSIX_SEMANTICS | FILE_FLAG_OVERLAPPED;
	DWORD sharemode = FILE_SHARE_READ | FILE_SHARE_WRITE;
	DWORD openmode = OPEN_ALWAYS;
	DWORD access;

	dprint(FD_FILE, "fd open %s\n", f->file_name);

	if (f->filetype == FIO_TYPE_PIPE) {
		log_err("windowsaio: pipes are not supported\n");
		return 1;
	}

	if (!strcmp(f->file_name, "-")) {
		log_err("windowsaio: can't read/write to stdin/out\n");
		return 1;
	}

	if (td->o.odirect)
		flags |= FILE_FLAG_NO_BUFFERING;
	if (td->o.sync_io)
		flags |= FILE_FLAG_WRITE_THROUGH;

	/*
	 * Inform Windows whether we're going to be doing sequential or
	 * random io so it can tune the Cache Manager
	 */
	if (td->o.td_ddir == TD_DDIR_READ  ||
		td->o.td_ddir == TD_DDIR_WRITE)
		flags |= FILE_FLAG_SEQUENTIAL_SCAN;
	else
		flags |= FILE_FLAG_RANDOM_ACCESS;

	if (!td_write(td) || read_only)
		access = GENERIC_READ;
	else
		access = (GENERIC_READ | GENERIC_WRITE);

	if (td->o.create_on_open)
		openmode = OPEN_ALWAYS;
	else
		openmode = OPEN_EXISTING;

	f->hFile = CreateFile(f->file_name, access, sharemode,
		NULL, openmode, flags, NULL);

	if (f->hFile == INVALID_HANDLE_VALUE) {
		log_err("windowsaio: failed to open file \"%s\"\n", f->file_name);
		rc = 1;
	}

	/* Only set up the completion port and thread if we're not just
	 * querying the device size */
	if (!rc && td->io_ops->data != NULL) {
		struct windowsaio_data *wd;

		wd = td->io_ops->data;

		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 (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;
		}
	}

	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, struct timespec *t)
{
	struct windowsaio_data *wd = td->io_ops->data;
	struct flist_head *entry;
	unsigned int dequeued = 0;
	struct io_u *io_u;
	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 {
		flist_for_each(entry, &td->io_u_busylist) {
			io_u = flist_entry(entry, struct io_u, list);
			fov = (struct fio_overlapped*)io_u->engine_data;

			if (fov->io_complete) {
				fov->io_complete = FALSE;
				ResetEvent(fov->o.hEvent);
				wd->aio_events[dequeued] = io_u;
				dequeued++;
			}

			if (dequeued >= min)
				break;
		}

		if (dequeued < min) {
			status = WaitForSingleObject(wd->iocomplete_event, mswait);
			if (status != WAIT_OBJECT_0 && dequeued >= min)
				break;
		}

		if (dequeued >= min || (t != NULL && timeout_expired(start_count, end_count)))
			break;
	} while (1);

	return dequeued;
}

static int fio_windowsaio_queue(struct thread_data *td, struct io_u *io_u)
{
	struct fio_overlapped *o = io_u->engine_data;
	LPOVERLAPPED lpOvl = &o->o;
	DWORD iobytes;
	BOOL success = FALSE;
	int rc = FIO_Q_COMPLETED;

	fio_ro_check(td, io_u);

	lpOvl->Internal = STATUS_PENDING;
	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, &iobytes, lpOvl);
		break;
	case DDIR_READ:
		success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, 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;
		break;
	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;
		break;
	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;
}

/* 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;

	ctx = (struct thread_ctx*)lpParameter;
	wd = ctx->wd;

	do {
		if (!GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey, &ovl, 250) && 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 int fio_windowsaio_cancel(struct thread_data *td,
				   struct io_u *io_u)
{
	int rc = 0;

	struct windowsaio_data *wd = td->io_ops->data;

	/* If we're running on Vista or newer, we can cancel individual IO requests */
	if (wd->pCancelIoEx != NULL) {
		struct fio_overlapped *ovl = io_u->engine_data;

		if (!wd->pCancelIoEx(io_u->file->hFile, &ovl->o)) {
			log_err("windowsaio: failed to cancel io\n");
			rc = 1;
		}
	} else
		rc = 1;

	return rc;
}

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) {
		CloseHandle(o->o.hEvent);
		io_u->engine_data = NULL;
		free(o);
	}
}

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 = CreateEvent(NULL, TRUE, FALSE, NULL);
	if (o->o.hEvent == NULL) {
		log_err("windowsaio: failed to create event handle\n");
		free(o);
		return 1;
	}

	io_u->engine_data = o;
	return 0;
}

static struct ioengine_ops ioengine = {
	.name		= "windowsaio",
	.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,
	.io_u_init	= fio_windowsaio_io_u_init,
	.io_u_free	= fio_windowsaio_io_u_free,
};

static void fio_init fio_windowsaio_register(void)
{
	register_ioengine(&ioengine);
}

static void fio_exit fio_windowsaio_unregister(void)
{
	unregister_ioengine(&ioengine);
}
Commit	Line	Data
	1	/*
	2	* windowsaio engine
	3	*
	4	* IO engine using Windows IO Completion Ports.
	5	*/
	6
	7	#include <stdio.h>
	8	#include <stdlib.h>
	9	#include <unistd.h>
	10	#include <signal.h>
	11	#include <errno.h>
	12	#include <windows.h>
	13
	14	#include "../fio.h"
	15
	16	typedef BOOL (WINAPI *CANCELIOEX)(HANDLE hFile, LPOVERLAPPED lpOverlapped);
	17
	18	int geterrno_from_win_error (DWORD code, int deferrno);
	19
	20	struct fio_overlapped {
	21	OVERLAPPED o;
	22	struct io_u *io_u;
	23	BOOL io_complete;
	24	};
	25
	26	struct windowsaio_data {
	27	struct io_u **aio_events;
	28	HANDLE iocp;
	29	HANDLE iothread;
	30	HANDLE iocomplete_event;
	31	CANCELIOEX pCancelIoEx;
	32	BOOL iothread_running;
	33	};
	34
	35	struct thread_ctx {
	36	HANDLE iocp;
	37	struct windowsaio_data *wd;
	38	};
	39
	40	static int fio_windowsaio_cancel(struct thread_data *td,
	41	struct io_u *io_u);
	42	static BOOL timeout_expired(DWORD start_count, DWORD end_count);
	43	static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
	44	unsigned int max, struct timespec *t);
	45	static struct io_u fio_windowsaio_event(struct thread_data td, int event);
	46	static int fio_windowsaio_queue(struct thread_data *td,
	47	struct io_u *io_u);
	48	static void fio_windowsaio_cleanup(struct thread_data *td);
	49	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter);
	50	static int fio_windowsaio_init(struct thread_data *td);
	51	static int fio_windowsaio_open_file(struct thread_data td, struct fio_file f);
	52	static int fio_windowsaio_close_file(struct thread_data fio_unused td, struct fio_file f);
	53
	54	static int fio_windowsaio_init(struct thread_data *td)
	55	{
	56	struct windowsaio_data *wd;
	57	HANDLE hKernel32Dll;
	58	int rc = 0;
	59
	60	wd = calloc(1, sizeof(struct windowsaio_data));
	61	if (wd == NULL) {
	62	log_err("windowsaio: failed to allocate memory for engine data\n");
	63	rc = 1;
	64	}
	65
	66	if (!rc) {
	67	wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*));
	68	if (wd->aio_events == NULL) {
	69	log_err("windowsaio: failed to allocate memory for aio events list\n");
	70	rc = 1;
	71	}
	72	}
	73
	74	if (!rc) {
	75	/* Create an auto-reset event */
	76	wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL);
	77	if (wd->iocomplete_event == NULL) {
	78	log_err("windowsaio: failed to create io complete event handle\n");
	79	rc = 1;
	80	}
	81	}
	82
	83	if (rc) {
	84	if (wd != NULL) {
	85	if (wd->aio_events != NULL)
	86	free(wd->aio_events);
	87
	88	free(wd);
	89	}
	90	}
	91
	92	hKernel32Dll = GetModuleHandle("kernel32.dll");
	93	wd->pCancelIoEx = (CANCELIOEX)GetProcAddress(hKernel32Dll, "CancelIoEx");
	94	td->io_ops->data = wd;
	95
	96	if (!rc) {
	97	struct thread_ctx *ctx;
	98	struct windowsaio_data *wd;
	99	HANDLE hFile;
	100
	101	hFile = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 0);
	102	if (hFile == INVALID_HANDLE_VALUE) {
	103	log_err("windowsaio: failed to create io completion port\n");
	104	rc = 1;
	105	}
	106
	107	wd = td->io_ops->data;
	108	wd->iothread_running = TRUE;
	109	wd->iocp = hFile;
	110
	111	if (!rc)
	112	ctx = malloc(sizeof(struct thread_ctx));
	113
	114	if (!rc && ctx == NULL)
	115	{
	116	log_err("windowsaio: failed to allocate memory for thread context structure\n");
	117	CloseHandle(hFile);
	118	rc = 1;
	119	}
	120
	121	if (!rc)
	122	{
	123	ctx->iocp = hFile;
	124	ctx->wd = wd;
	125	wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, NULL);
	126	if (wd->iothread == NULL)
	127	log_err("windowsaio: failed to create io completion thread\n");
	128	}
	129
	130	if (rc \|\| wd->iothread == NULL)
	131	rc = 1;
	132	}
	133
	134	return rc;
	135	}
	136
	137	static void fio_windowsaio_cleanup(struct thread_data *td)
	138	{
	139	struct windowsaio_data *wd;
	140
	141	wd = td->io_ops->data;
	142
	143	if (wd != NULL) {
	144	wd->iothread_running = FALSE;
	145	WaitForSingleObject(wd->iothread, INFINITE);
	146
	147	CloseHandle(wd->iothread);
	148	CloseHandle(wd->iocomplete_event);
	149
	150	free(wd->aio_events);
	151	free(wd);
	152
	153	td->io_ops->data = NULL;
	154	}
	155	}
	156
	157
	158	static int fio_windowsaio_open_file(struct thread_data td, struct fio_file f)
	159	{
	160	int rc = 0;
	161	DWORD flags = FILE_FLAG_POSIX_SEMANTICS \| FILE_FLAG_OVERLAPPED;
	162	DWORD sharemode = FILE_SHARE_READ \| FILE_SHARE_WRITE;
	163	DWORD openmode = OPEN_ALWAYS;
	164	DWORD access;
	165
	166	dprint(FD_FILE, "fd open %s\n", f->file_name);
	167
	168	if (f->filetype == FIO_TYPE_PIPE) {
	169	log_err("windowsaio: pipes are not supported\n");
	170	return 1;
	171	}
	172
	173	if (!strcmp(f->file_name, "-")) {
	174	log_err("windowsaio: can't read/write to stdin/out\n");
	175	return 1;
	176	}
	177
	178	if (td->o.odirect)
	179	flags \|= FILE_FLAG_NO_BUFFERING;
	180	if (td->o.sync_io)
	181	flags \|= FILE_FLAG_WRITE_THROUGH;
	182
	183	/*
	184	* Inform Windows whether we're going to be doing sequential or
	185	* random io so it can tune the Cache Manager
	186	*/
	187	if (td->o.td_ddir == TD_DDIR_READ \|\|
	188	td->o.td_ddir == TD_DDIR_WRITE)
	189	flags \|= FILE_FLAG_SEQUENTIAL_SCAN;
	190	else
	191	flags \|= FILE_FLAG_RANDOM_ACCESS;
	192
	193	if (!td_write(td) \|\| read_only)
	194	access = GENERIC_READ;
	195	else
	196	access = (GENERIC_READ \| GENERIC_WRITE);
	197
	198	if (td->o.create_on_open)
	199	openmode = OPEN_ALWAYS;
	200	else
	201	openmode = OPEN_EXISTING;
	202
	203	f->hFile = CreateFile(f->file_name, access, sharemode,
	204	NULL, openmode, flags, NULL);
	205
	206	if (f->hFile == INVALID_HANDLE_VALUE) {
	207	log_err("windowsaio: failed to open file \"%s\"\n", f->file_name);
	208	rc = 1;
	209	}
	210
	211	/* Only set up the completion port and thread if we're not just
	212	* querying the device size */
	213	if (!rc && td->io_ops->data != NULL) {
	214	struct windowsaio_data *wd;
	215
	216	wd = td->io_ops->data;
	217
	218	if (CreateIoCompletionPort(f->hFile, wd->iocp, 0, 0) == NULL) {
	219	log_err("windowsaio: failed to create io completion port\n");
	220	rc = 1;
	221	}
	222	}
	223
	224	return rc;
	225	}
	226
	227	static int fio_windowsaio_close_file(struct thread_data fio_unused td, struct fio_file f)
	228	{
	229	int rc = 0;
	230
	231	dprint(FD_FILE, "fd close %s\n", f->file_name);
	232
	233	if (f->hFile != INVALID_HANDLE_VALUE) {
	234	if (!CloseHandle(f->hFile)) {
	235	log_info("windowsaio: failed to close file handle for \"%s\"\n", f->file_name);
	236	rc = 1;
	237	}
	238	}
	239
	240	f->hFile = INVALID_HANDLE_VALUE;
	241	return rc;
	242	}
	243
	244	static BOOL timeout_expired(DWORD start_count, DWORD end_count)
	245	{
	246	BOOL expired = FALSE;
	247	DWORD current_time;
	248
	249	current_time = GetTickCount();
	250
	251	if ((end_count > start_count) && current_time >= end_count)
	252	expired = TRUE;
	253	else if (current_time < start_count && current_time > end_count)
	254	expired = TRUE;
	255
	256	return expired;
	257	}
	258
	259	static struct io_u* fio_windowsaio_event(struct thread_data *td, int event)
	260	{
	261	struct windowsaio_data *wd = td->io_ops->data;
	262	return wd->aio_events[event];
	263	}
	264
	265	static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
	266	unsigned int max, struct timespec *t)
	267	{
	268	struct windowsaio_data *wd = td->io_ops->data;
	269	struct flist_head *entry;
	270	unsigned int dequeued = 0;
	271	struct io_u *io_u;
	272	struct fio_overlapped *fov;
	273	DWORD start_count = 0;
	274	DWORD end_count = 0;
	275	DWORD status;
	276	DWORD mswait = 250;
	277
	278	if (t != NULL) {
	279	mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	280	start_count = GetTickCount();
	281	end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	282	}
	283
	284	do {
	285	flist_for_each(entry, &td->io_u_busylist) {
	286	io_u = flist_entry(entry, struct io_u, list);
	287	fov = (struct fio_overlapped*)io_u->engine_data;
	288
	289	if (fov->io_complete) {
	290	fov->io_complete = FALSE;
	291	ResetEvent(fov->o.hEvent);
	292	wd->aio_events[dequeued] = io_u;
	293	dequeued++;
	294	}
	295
	296	if (dequeued >= min)
	297	break;
	298	}
	299
	300	if (dequeued < min) {
	301	status = WaitForSingleObject(wd->iocomplete_event, mswait);
	302	if (status != WAIT_OBJECT_0 && dequeued >= min)
	303	break;
	304	}
	305
	306	if (dequeued >= min \|\| (t != NULL && timeout_expired(start_count, end_count)))
	307	break;
	308	} while (1);
	309
	310	return dequeued;
	311	}
	312
	313	static int fio_windowsaio_queue(struct thread_data td, struct io_u io_u)
	314	{
	315	struct fio_overlapped *o = io_u->engine_data;
	316	LPOVERLAPPED lpOvl = &o->o;
	317	DWORD iobytes;
	318	BOOL success = FALSE;
	319	int rc = FIO_Q_COMPLETED;
	320
	321	fio_ro_check(td, io_u);
	322
	323	lpOvl->Internal = STATUS_PENDING;
	324	lpOvl->InternalHigh = 0;
	325	lpOvl->Offset = io_u->offset & 0xFFFFFFFF;
	326	lpOvl->OffsetHigh = io_u->offset >> 32;
	327
	328	switch (io_u->ddir) {
	329	case DDIR_WRITE:
	330	success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, lpOvl);
	331	break;
	332	case DDIR_READ:
	333	success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, lpOvl);
	334	break;
	335	case DDIR_SYNC:
	336	case DDIR_DATASYNC:
	337	case DDIR_SYNC_FILE_RANGE:
	338	success = FlushFileBuffers(io_u->file->hFile);
	339	if (!success) {
	340	log_err("windowsaio: failed to flush file buffers\n");
	341	io_u->error = win_to_posix_error(GetLastError());
	342	}
	343
	344	return FIO_Q_COMPLETED;
	345	break;
	346	case DDIR_TRIM:
	347	log_err("windowsaio: manual TRIM isn't supported on Windows\n");
	348	io_u->error = 1;
	349	io_u->resid = io_u->xfer_buflen;
	350	return FIO_Q_COMPLETED;
	351	break;
	352	default:
	353	assert(0);
	354	break;
	355	}
	356
	357	if (success \|\| GetLastError() == ERROR_IO_PENDING)
	358	rc = FIO_Q_QUEUED;
	359	else {
	360	io_u->error = win_to_posix_error(GetLastError());
	361	io_u->resid = io_u->xfer_buflen;
	362	}
	363
	364	return rc;
	365	}
	366
	367	/* Runs as a thread and waits for queued IO to complete */
	368	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter)
	369	{
	370	OVERLAPPED *ovl;
	371	struct fio_overlapped *fov;
	372	struct io_u *io_u;
	373	struct windowsaio_data *wd;
	374	struct thread_ctx *ctx;
	375	ULONG_PTR ulKey = 0;
	376	DWORD bytes;
	377
	378	ctx = (struct thread_ctx*)lpParameter;
	379	wd = ctx->wd;
	380
	381	do {
	382	if (!GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey, &ovl, 250) && ovl == NULL)
	383	continue;
	384
	385	fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o);
	386	io_u = fov->io_u;
	387
	388	if (ovl->Internal == ERROR_SUCCESS) {
	389	io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
	390	io_u->error = 0;
	391	} else {
	392	io_u->resid = io_u->xfer_buflen;
	393	io_u->error = win_to_posix_error(GetLastError());
	394	}
	395
	396	fov->io_complete = TRUE;
	397	SetEvent(wd->iocomplete_event);
	398	} while (ctx->wd->iothread_running);
	399
	400	CloseHandle(ctx->iocp);
	401	free(ctx);
	402	return 0;
	403	}
	404
	405	static int fio_windowsaio_cancel(struct thread_data *td,
	406	struct io_u *io_u)
	407	{
	408	int rc = 0;
	409
	410	struct windowsaio_data *wd = td->io_ops->data;
	411
	412	/* If we're running on Vista or newer, we can cancel individual IO requests */
	413	if (wd->pCancelIoEx != NULL) {
	414	struct fio_overlapped *ovl = io_u->engine_data;
	415
	416	if (!wd->pCancelIoEx(io_u->file->hFile, &ovl->o)) {
	417	log_err("windowsaio: failed to cancel io\n");
	418	rc = 1;
	419	}
	420	} else
	421	rc = 1;
	422
	423	return rc;
	424	}
	425
	426	static void fio_windowsaio_io_u_free(struct thread_data td, struct io_u io_u)
	427	{
	428	struct fio_overlapped *o = io_u->engine_data;
	429
	430	if (o) {
	431	CloseHandle(o->o.hEvent);
	432	io_u->engine_data = NULL;
	433	free(o);
	434	}
	435	}
	436
	437	static int fio_windowsaio_io_u_init(struct thread_data td, struct io_u io_u)
	438	{
	439	struct fio_overlapped *o;
	440
	441	o = malloc(sizeof(*o));
	442	o->io_complete = FALSE;
	443	o->io_u = io_u;
	444	o->o.hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);
	445	if (o->o.hEvent == NULL) {
	446	log_err("windowsaio: failed to create event handle\n");
	447	free(o);
	448	return 1;
	449	}
	450
	451	io_u->engine_data = o;
	452	return 0;
	453	}
	454
	455	static struct ioengine_ops ioengine = {
	456	.name = "windowsaio",
	457	.version = FIO_IOOPS_VERSION,
	458	.init = fio_windowsaio_init,
	459	.queue = fio_windowsaio_queue,
	460	.cancel = fio_windowsaio_cancel,
	461	.getevents = fio_windowsaio_getevents,
	462	.event = fio_windowsaio_event,
	463	.cleanup = fio_windowsaio_cleanup,
	464	.open_file = fio_windowsaio_open_file,
	465	.close_file = fio_windowsaio_close_file,
	466	.get_file_size = generic_get_file_size,
	467	.io_u_init = fio_windowsaio_io_u_init,
	468	.io_u_free = fio_windowsaio_io_u_free,
	469	};
	470
	471	static void fio_init fio_windowsaio_register(void)
	472	{
	473	register_ioengine(&ioengine);
	474	}
	475
	476	static void fio_exit fio_windowsaio_unregister(void)
	477	{
	478	unregister_ioengine(&ioengine);
	479	}