[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 "../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;
	BOOL iothread_running;
};

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

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, const 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;
	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);
		}
	}

	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,
				    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 (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;
	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;
		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 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);
	}
}

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;
}

static struct ioengine_ops ioengine = {
	.name		= "windowsaio",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_windowsaio_init,
	.queue		= fio_windowsaio_queue,
	.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
ecc314ba	1	/*
03244c19 BC	2	* windowsaio engine
	3	*
	4	* IO engine using Windows IO Completion Ports.
ecc314ba BC	5	*/
ecc314ba BC	6
ecc314ba BC	7	#include <stdio.h>
	8	#include <stdlib.h>
	9	#include <unistd.h>
	10	#include <signal.h>
	11	#include <errno.h>
ecc314ba BC	12
	13	#include "../fio.h"
	14
ea4500d8 BC	15	typedef BOOL (WINAPI *CANCELIOEX)(HANDLE hFile, LPOVERLAPPED lpOverlapped);
ea4500d8 BC	16
66c098b8 BC	17	int geterrno_from_win_error (DWORD code, int deferrno);
66c098b8 BC	18
67897036 BC	19	struct fio_overlapped {
	20	OVERLAPPED o;
	21	struct io_u *io_u;
	22	BOOL io_complete;
67897036	23	};
ecc314ba	24
9b836561	25	struct windowsaio_data {
9b836561	26	struct io_u **aio_events;
5a90bb5f	27	HANDLE iocp;
67897036	28	HANDLE iothread;
9b836561	29	HANDLE iocomplete_event;
67897036	30	BOOL iothread_running;
9b836561 BC	31	};
9b836561 BC	32
ecc314ba	33	struct thread_ctx {
9b836561	34	HANDLE iocp;
ecc314ba BC	35	struct windowsaio_data *wd;
	36	};
	37
67897036	38	static BOOL timeout_expired(DWORD start_count, DWORD end_count);
ecc314ba	39	static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
1f440ece	40	unsigned int max, const struct timespec *t);
ecc314ba BC	41	static struct io_u fio_windowsaio_event(struct thread_data td, int event);
ecc314ba BC	42	static int fio_windowsaio_queue(struct thread_data *td,
03244c19	43	struct io_u *io_u);
ecc314ba BC	44	static void fio_windowsaio_cleanup(struct thread_data *td);
	45	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter);
	46	static int fio_windowsaio_init(struct thread_data *td);
	47	static int fio_windowsaio_open_file(struct thread_data td, struct fio_file f);
	48	static int fio_windowsaio_close_file(struct thread_data fio_unused td, struct fio_file f);
	49
ecc314ba BC	50	static int fio_windowsaio_init(struct thread_data *td)
ecc314ba BC	51	{
ecc314ba	52	struct windowsaio_data *wd;
9b836561	53	int rc = 0;
ecc314ba	54
03244c19 BC	55	wd = calloc(1, sizeof(struct windowsaio_data));
	56	if (wd == NULL) {
	57	log_err("windowsaio: failed to allocate memory for engine data\n");
9b836561	58	rc = 1;
03244c19	59	}
ecc314ba	60
9b836561 BC	61	if (!rc) {
9b836561 BC	62	wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*));
03244c19 BC	63	if (wd->aio_events == NULL) {
03244c19 BC	64	log_err("windowsaio: failed to allocate memory for aio events list\n");
9b836561	65	rc = 1;
03244c19	66	}
e4db9fec BC	67	}
e4db9fec BC	68
9b836561 BC	69	if (!rc) {
	70	/* Create an auto-reset event */
	71	wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL);
03244c19 BC	72	if (wd->iocomplete_event == NULL) {
03244c19 BC	73	log_err("windowsaio: failed to create io complete event handle\n");
9b836561	74	rc = 1;
03244c19	75	}
9b836561 BC	76	}
9b836561 BC	77
9b836561	78	if (rc) {
9b836561	79	if (wd != NULL) {
9b836561 BC	80	if (wd->aio_events != NULL)
	81	free(wd->aio_events);
	82
	83	free(wd);
	84	}
	85	}
ecc314ba	86
565e784d	87	td->io_ops_data = wd;
93bcfd20	88
5a90bb5f BC	89	if (!rc) {
	90	struct thread_ctx *ctx;
	91	struct windowsaio_data *wd;
	92	HANDLE hFile;
	93
	94	hFile = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 0);
03244c19 BC	95	if (hFile == INVALID_HANDLE_VALUE) {
03244c19 BC	96	log_err("windowsaio: failed to create io completion port\n");
5a90bb5f	97	rc = 1;
03244c19	98	}
5a90bb5f	99
565e784d	100	wd = td->io_ops_data;
5a90bb5f BC	101	wd->iothread_running = TRUE;
	102	wd->iocp = hFile;
	103
	104	if (!rc)
	105	ctx = malloc(sizeof(struct thread_ctx));
	106
	107	if (!rc && ctx == NULL)
	108	{
03244c19	109	log_err("windowsaio: failed to allocate memory for thread context structure\n");
5a90bb5f BC	110	CloseHandle(hFile);
	111	rc = 1;
	112	}
	113
	114	if (!rc)
	115	{
	116	ctx->iocp = hFile;
	117	ctx->wd = wd;
	118	wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, NULL);
03244c19 BC	119	if (wd->iothread == NULL)
03244c19 BC	120	log_err("windowsaio: failed to create io completion thread\n");
5a90bb5f BC	121	}
	122
	123	if (rc \|\| wd->iothread == NULL)
	124	rc = 1;
	125	}
	126
4e79098f	127	return rc;
ecc314ba BC	128	}
ecc314ba BC	129
67897036 BC	130	static void fio_windowsaio_cleanup(struct thread_data *td)
67897036 BC	131	{
67897036 BC	132	struct windowsaio_data *wd;
67897036 BC	133
565e784d	134	wd = td->io_ops_data;
67897036 BC	135
67897036 BC	136	if (wd != NULL) {
40c5db35 JA	137	wd->iothread_running = FALSE;
40c5db35 JA	138	WaitForSingleObject(wd->iothread, INFINITE);
67897036 BC	139
	140	CloseHandle(wd->iothread);
	141	CloseHandle(wd->iocomplete_event);
	142
67897036	143	free(wd->aio_events);
67897036 BC	144	free(wd);
67897036 BC	145
565e784d	146	td->io_ops_data = NULL;
67897036 BC	147	}
	148	}
	149
	150
ecc314ba BC	151	static int fio_windowsaio_open_file(struct thread_data td, struct fio_file f)
	152	{
	153	int rc = 0;
4e79098f	154	DWORD flags = FILE_FLAG_POSIX_SEMANTICS \| FILE_FLAG_OVERLAPPED;
ecc314ba BC	155	DWORD sharemode = FILE_SHARE_READ \| FILE_SHARE_WRITE;
	156	DWORD openmode = OPEN_ALWAYS;
	157	DWORD access;
	158
	159	dprint(FD_FILE, "fd open %s\n", f->file_name);
	160
ecc314ba	161	if (f->filetype == FIO_TYPE_PIPE) {
03244c19	162	log_err("windowsaio: pipes are not supported\n");
ecc314ba BC	163	return 1;
	164	}
	165
	166	if (!strcmp(f->file_name, "-")) {
03244c19	167	log_err("windowsaio: can't read/write to stdin/out\n");
ecc314ba BC	168	return 1;
	169	}
	170
	171	if (td->o.odirect)
	172	flags \|= FILE_FLAG_NO_BUFFERING;
	173	if (td->o.sync_io)
	174	flags \|= FILE_FLAG_WRITE_THROUGH;
	175
93bcfd20 BC	176	/*
	177	* Inform Windows whether we're going to be doing sequential or
	178	* random io so it can tune the Cache Manager
	179	*/
ecc314ba	180	if (td->o.td_ddir == TD_DDIR_READ \|\|
ea4500d8	181	td->o.td_ddir == TD_DDIR_WRITE)
ecc314ba	182	flags \|= FILE_FLAG_SEQUENTIAL_SCAN;
ecc314ba	183	else
ecc314ba	184	flags \|= FILE_FLAG_RANDOM_ACCESS;
ecc314ba	185
ea4500d8	186	if (!td_write(td) \|\| read_only)
ecc314ba BC	187	access = GENERIC_READ;
	188	else
	189	access = (GENERIC_READ \| GENERIC_WRITE);
	190
93bcfd20	191	if (td->o.create_on_open)
ecc314ba BC	192	openmode = OPEN_ALWAYS;
	193	else
	194	openmode = OPEN_EXISTING;
	195
	196	f->hFile = CreateFile(f->file_name, access, sharemode,
	197	NULL, openmode, flags, NULL);
	198
03244c19 BC	199	if (f->hFile == INVALID_HANDLE_VALUE) {
03244c19 BC	200	log_err("windowsaio: failed to open file \"%s\"\n", f->file_name);
ecc314ba	201	rc = 1;
03244c19	202	}
ecc314ba	203
93bcfd20	204	/* Only set up the completion port and thread if we're not just
ecc314ba	205	* querying the device size */
565e784d	206	if (!rc && td->io_ops_data != NULL) {
40c5db35	207	struct windowsaio_data *wd;
ecc314ba	208
565e784d	209	wd = td->io_ops_data;
ecc314ba	210
03244c19 BC	211	if (CreateIoCompletionPort(f->hFile, wd->iocp, 0, 0) == NULL) {
03244c19 BC	212	log_err("windowsaio: failed to create io completion port\n");
ecc314ba	213	rc = 1;
03244c19	214	}
ecc314ba BC	215	}
ecc314ba BC	216
ecc314ba BC	217	return rc;
	218	}
	219
	220	static int fio_windowsaio_close_file(struct thread_data fio_unused td, struct fio_file f)
	221	{
4e79098f	222	int rc = 0;
1e42cc3d	223
9b836561 BC	224	dprint(FD_FILE, "fd close %s\n", f->file_name);
9b836561 BC	225
ecc314ba	226	if (f->hFile != INVALID_HANDLE_VALUE) {
03244c19 BC	227	if (!CloseHandle(f->hFile)) {
03244c19 BC	228	log_info("windowsaio: failed to close file handle for \"%s\"\n", f->file_name);
4e79098f	229	rc = 1;
03244c19	230	}
ecc314ba BC	231	}
	232
	233	f->hFile = INVALID_HANDLE_VALUE;
4e79098f	234	return rc;
ecc314ba BC	235	}
ecc314ba BC	236
67897036 BC	237	static BOOL timeout_expired(DWORD start_count, DWORD end_count)
	238	{
	239	BOOL expired = FALSE;
	240	DWORD current_time;
	241
	242	current_time = GetTickCount();
	243
	244	if ((end_count > start_count) && current_time >= end_count)
	245	expired = TRUE;
	246	else if (current_time < start_count && current_time > end_count)
	247	expired = TRUE;
	248
	249	return expired;
	250	}
	251
	252	static struct io_u* fio_windowsaio_event(struct thread_data *td, int event)
	253	{
565e784d	254	struct windowsaio_data *wd = td->io_ops_data;
67897036 BC	255	return wd->aio_events[event];
	256	}
	257
	258	static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
1f440ece JA	259	unsigned int max,
1f440ece JA	260	const struct timespec *t)
67897036	261	{
565e784d	262	struct windowsaio_data *wd = td->io_ops_data;
67897036 BC	263	unsigned int dequeued = 0;
67897036 BC	264	struct io_u *io_u;
7b2dfab1	265	int i;
67897036 BC	266	struct fio_overlapped *fov;
	267	DWORD start_count = 0;
	268	DWORD end_count = 0;
	269	DWORD status;
	270	DWORD mswait = 250;
	271
	272	if (t != NULL) {
	273	mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	274	start_count = GetTickCount();
	275	end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	276	}
	277
	278	do {
7b2dfab1 BC	279	io_u_qiter(&td->io_u_all, io_u, i) {
	280	if (!(io_u->flags & IO_U_F_FLIGHT))
	281	continue;
	282
67897036 BC	283	fov = (struct fio_overlapped*)io_u->engine_data;
	284
	285	if (fov->io_complete) {
40c5db35	286	fov->io_complete = FALSE;
67897036 BC	287	wd->aio_events[dequeued] = io_u;
	288	dequeued++;
	289	}
	290
67897036	291	}
77e7b330 JR	292	if (dequeued >= min)
77e7b330 JR	293	break;
67897036	294
40c5db35	295	if (dequeued < min) {
67897036	296	status = WaitForSingleObject(wd->iocomplete_event, mswait);
f9a58c2a	297	if (status != WAIT_OBJECT_0 && dequeued >= min)
03244c19	298	break;
67897036 BC	299	}
	300
	301	if (dequeued >= min \|\| (t != NULL && timeout_expired(start_count, end_count)))
	302	break;
	303	} while (1);
	304
	305	return dequeued;
	306	}
	307
40c5db35	308	static int fio_windowsaio_queue(struct thread_data td, struct io_u io_u)
67897036	309	{
c73ed246 JA	310	struct fio_overlapped *o = io_u->engine_data;
c73ed246 JA	311	LPOVERLAPPED lpOvl = &o->o;
93bcfd20	312	BOOL success = FALSE;
67897036 BC	313	int rc = FIO_Q_COMPLETED;
	314
	315	fio_ro_check(td, io_u);
	316
77e7b330	317	lpOvl->Internal = 0;
4e79098f BC	318	lpOvl->InternalHigh = 0;
	319	lpOvl->Offset = io_u->offset & 0xFFFFFFFF;
	320	lpOvl->OffsetHigh = io_u->offset >> 32;
67897036 BC	321
67897036 BC	322	switch (io_u->ddir) {
40c5db35	323	case DDIR_WRITE:
77e7b330	324	success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, NULL, lpOvl);
67897036 BC	325	break;
67897036 BC	326	case DDIR_READ:
77e7b330	327	success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, NULL, lpOvl);
67897036 BC	328	break;
	329	case DDIR_SYNC:
	330	case DDIR_DATASYNC:
	331	case DDIR_SYNC_FILE_RANGE:
	332	success = FlushFileBuffers(io_u->file->hFile);
03244c19 BC	333	if (!success) {
	334	log_err("windowsaio: failed to flush file buffers\n");
	335	io_u->error = win_to_posix_error(GetLastError());
	336	}
67897036 BC	337
	338	return FIO_Q_COMPLETED;
	339	break;
	340	case DDIR_TRIM:
03244c19	341	log_err("windowsaio: manual TRIM isn't supported on Windows\n");
67897036 BC	342	io_u->error = 1;
	343	io_u->resid = io_u->xfer_buflen;
	344	return FIO_Q_COMPLETED;
	345	break;
	346	default:
	347	assert(0);
93bcfd20	348	break;
67897036 BC	349	}
67897036 BC	350
40c5db35	351	if (success \|\| GetLastError() == ERROR_IO_PENDING)
67897036	352	rc = FIO_Q_QUEUED;
40c5db35	353	else {
2277d5d5	354	io_u->error = win_to_posix_error(GetLastError());
67897036 BC	355	io_u->resid = io_u->xfer_buflen;
	356	}
	357
	358	return rc;
	359	}
	360
	361	/* Runs as a thread and waits for queued IO to complete */
	362	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter)
	363	{
	364	OVERLAPPED *ovl;
	365	struct fio_overlapped *fov;
	366	struct io_u *io_u;
	367	struct windowsaio_data *wd;
	368	struct thread_ctx *ctx;
	369	ULONG_PTR ulKey = 0;
	370	DWORD bytes;
	371
	372	ctx = (struct thread_ctx*)lpParameter;
	373	wd = ctx->wd;
	374
	375	do {
66c098b8	376	if (!GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey, &ovl, 250) && ovl == NULL)
67897036 BC	377	continue;
	378
	379	fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o);
	380	io_u = fov->io_u;
	381
	382	if (ovl->Internal == ERROR_SUCCESS) {
	383	io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
	384	io_u->error = 0;
	385	} else {
	386	io_u->resid = io_u->xfer_buflen;
2277d5d5	387	io_u->error = win_to_posix_error(GetLastError());
67897036 BC	388	}
67897036 BC	389
40c5db35	390	fov->io_complete = TRUE;
67897036 BC	391	SetEvent(wd->iocomplete_event);
	392	} while (ctx->wd->iothread_running);
	393
	394	CloseHandle(ctx->iocp);
	395	free(ctx);
	396	return 0;
	397	}
	398
c73ed246 JA	399	static void fio_windowsaio_io_u_free(struct thread_data td, struct io_u io_u)
	400	{
	401	struct fio_overlapped *o = io_u->engine_data;
	402
	403	if (o) {
c73ed246 JA	404	io_u->engine_data = NULL;
	405	free(o);
	406	}
	407	}
	408
	409	static int fio_windowsaio_io_u_init(struct thread_data td, struct io_u io_u)
	410	{
	411	struct fio_overlapped *o;
	412
	413	o = malloc(sizeof(*o));
b0106419	414	o->io_complete = FALSE;
c73ed246	415	o->io_u = io_u;
77e7b330	416	o->o.hEvent = NULL;
c73ed246 JA	417	io_u->engine_data = o;
	418	return 0;
	419	}
	420
ecc314ba BC	421	static struct ioengine_ops ioengine = {
	422	.name = "windowsaio",
	423	.version = FIO_IOOPS_VERSION,
	424	.init = fio_windowsaio_init,
	425	.queue = fio_windowsaio_queue,
ecc314ba BC	426	.getevents = fio_windowsaio_getevents,
	427	.event = fio_windowsaio_event,
	428	.cleanup = fio_windowsaio_cleanup,
	429	.open_file = fio_windowsaio_open_file,
	430	.close_file = fio_windowsaio_close_file,
c73ed246 JA	431	.get_file_size = generic_get_file_size,
	432	.io_u_init = fio_windowsaio_io_u_init,
	433	.io_u_free = fio_windowsaio_io_u_free,
ecc314ba BC	434	};
ecc314ba BC	435
c874d188	436	static void fio_init fio_windowsaio_register(void)
ecc314ba BC	437	{
	438	register_ioengine(&ioengine);
	439	}
	440
c874d188	441	static void fio_exit fio_windowsaio_unregister(void)
ecc314ba BC	442	{
	443	unregister_ioengine(&ioengine);
	444	}