[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 DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter);

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)
		{
			DWORD threadid;

			ctx->iocp = hFile;
			ctx->wd = wd;
			wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, &threadid);

			if (wd->iothread != NULL)
				fio_setaffinity(threadid, td->o.cpumask);
			else
				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
	 */
	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_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
ecc314ba	38	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter);
ecc314ba	39
ecc314ba BC	40	static int fio_windowsaio_init(struct thread_data *td)
ecc314ba BC	41	{
ecc314ba	42	struct windowsaio_data *wd;
9b836561	43	int rc = 0;
ecc314ba	44
03244c19 BC	45	wd = calloc(1, sizeof(struct windowsaio_data));
	46	if (wd == NULL) {
	47	log_err("windowsaio: failed to allocate memory for engine data\n");
9b836561	48	rc = 1;
03244c19	49	}
ecc314ba	50
9b836561 BC	51	if (!rc) {
9b836561 BC	52	wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*));
03244c19 BC	53	if (wd->aio_events == NULL) {
03244c19 BC	54	log_err("windowsaio: failed to allocate memory for aio events list\n");
9b836561	55	rc = 1;
03244c19	56	}
e4db9fec BC	57	}
e4db9fec BC	58
9b836561 BC	59	if (!rc) {
	60	/* Create an auto-reset event */
	61	wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL);
03244c19 BC	62	if (wd->iocomplete_event == NULL) {
03244c19 BC	63	log_err("windowsaio: failed to create io complete event handle\n");
9b836561	64	rc = 1;
03244c19	65	}
9b836561 BC	66	}
9b836561 BC	67
9b836561	68	if (rc) {
9b836561	69	if (wd != NULL) {
9b836561 BC	70	if (wd->aio_events != NULL)
	71	free(wd->aio_events);
	72
	73	free(wd);
	74	}
	75	}
ecc314ba	76
565e784d	77	td->io_ops_data = wd;
93bcfd20	78
5a90bb5f BC	79	if (!rc) {
	80	struct thread_ctx *ctx;
	81	struct windowsaio_data *wd;
	82	HANDLE hFile;
	83
	84	hFile = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 0);
03244c19 BC	85	if (hFile == INVALID_HANDLE_VALUE) {
03244c19 BC	86	log_err("windowsaio: failed to create io completion port\n");
5a90bb5f	87	rc = 1;
03244c19	88	}
5a90bb5f	89
565e784d	90	wd = td->io_ops_data;
5a90bb5f BC	91	wd->iothread_running = TRUE;
	92	wd->iocp = hFile;
	93
	94	if (!rc)
	95	ctx = malloc(sizeof(struct thread_ctx));
	96
	97	if (!rc && ctx == NULL)
	98	{
03244c19	99	log_err("windowsaio: failed to allocate memory for thread context structure\n");
5a90bb5f BC	100	CloseHandle(hFile);
	101	rc = 1;
	102	}
	103
	104	if (!rc)
	105	{
438bb1cf BC	106	DWORD threadid;
438bb1cf BC	107
5a90bb5f BC	108	ctx->iocp = hFile;
5a90bb5f BC	109	ctx->wd = wd;
438bb1cf BC	110	wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, &threadid);
	111
	112	if (wd->iothread != NULL)
	113	fio_setaffinity(threadid, td->o.cpumask);
	114	else
03244c19	115	log_err("windowsaio: failed to create io completion thread\n");
5a90bb5f BC	116	}
	117
	118	if (rc \|\| wd->iothread == NULL)
	119	rc = 1;
	120	}
	121
4e79098f	122	return rc;
ecc314ba BC	123	}
ecc314ba BC	124
67897036 BC	125	static void fio_windowsaio_cleanup(struct thread_data *td)
67897036 BC	126	{
67897036 BC	127	struct windowsaio_data *wd;
67897036 BC	128
565e784d	129	wd = td->io_ops_data;
67897036 BC	130
67897036 BC	131	if (wd != NULL) {
40c5db35 JA	132	wd->iothread_running = FALSE;
40c5db35 JA	133	WaitForSingleObject(wd->iothread, INFINITE);
67897036 BC	134
	135	CloseHandle(wd->iothread);
	136	CloseHandle(wd->iocomplete_event);
	137
67897036	138	free(wd->aio_events);
67897036 BC	139	free(wd);
67897036 BC	140
565e784d	141	td->io_ops_data = NULL;
67897036 BC	142	}
	143	}
	144
ecc314ba BC	145	static int fio_windowsaio_open_file(struct thread_data td, struct fio_file f)
	146	{
	147	int rc = 0;
4e79098f	148	DWORD flags = FILE_FLAG_POSIX_SEMANTICS \| FILE_FLAG_OVERLAPPED;
ecc314ba BC	149	DWORD sharemode = FILE_SHARE_READ \| FILE_SHARE_WRITE;
	150	DWORD openmode = OPEN_ALWAYS;
	151	DWORD access;
	152
	153	dprint(FD_FILE, "fd open %s\n", f->file_name);
	154
ecc314ba	155	if (f->filetype == FIO_TYPE_PIPE) {
03244c19	156	log_err("windowsaio: pipes are not supported\n");
ecc314ba BC	157	return 1;
	158	}
	159
	160	if (!strcmp(f->file_name, "-")) {
03244c19	161	log_err("windowsaio: can't read/write to stdin/out\n");
ecc314ba BC	162	return 1;
	163	}
	164
	165	if (td->o.odirect)
	166	flags \|= FILE_FLAG_NO_BUFFERING;
	167	if (td->o.sync_io)
	168	flags \|= FILE_FLAG_WRITE_THROUGH;
	169
93bcfd20 BC	170	/*
93bcfd20 BC	171	* Inform Windows whether we're going to be doing sequential or
f20a86a7	172	* random IO so it can tune the Cache Manager
93bcfd20	173	*/
f20a86a7 SW	174	switch (td->o.fadvise_hint) {
	175	case F_ADV_TYPE:
	176	if (td_random(td))
	177	flags \|= FILE_FLAG_RANDOM_ACCESS;
	178	else
	179	flags \|= FILE_FLAG_SEQUENTIAL_SCAN;
	180	break;
	181	case F_ADV_RANDOM:
ecc314ba	182	flags \|= FILE_FLAG_RANDOM_ACCESS;
f20a86a7 SW	183	break;
	184	case F_ADV_SEQUENTIAL:
	185	flags \|= FILE_FLAG_SEQUENTIAL_SCAN;
	186	break;
	187	case F_ADV_NONE:
	188	break;
	189	default:
	190	log_err("fio: unknown fadvise type %d\n", td->o.fadvise_hint);
	191	}
ecc314ba	192
ea4500d8	193	if (!td_write(td) \|\| read_only)
ecc314ba BC	194	access = GENERIC_READ;
	195	else
	196	access = (GENERIC_READ \| GENERIC_WRITE);
	197
93bcfd20	198	if (td->o.create_on_open)
ecc314ba BC	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
03244c19 BC	206	if (f->hFile == INVALID_HANDLE_VALUE) {
03244c19 BC	207	log_err("windowsaio: failed to open file \"%s\"\n", f->file_name);
ecc314ba	208	rc = 1;
03244c19	209	}
ecc314ba	210
93bcfd20	211	/* Only set up the completion port and thread if we're not just
ecc314ba	212	* querying the device size */
565e784d	213	if (!rc && td->io_ops_data != NULL) {
40c5db35	214	struct windowsaio_data *wd;
ecc314ba	215
565e784d	216	wd = td->io_ops_data;
ecc314ba	217
03244c19 BC	218	if (CreateIoCompletionPort(f->hFile, wd->iocp, 0, 0) == NULL) {
03244c19 BC	219	log_err("windowsaio: failed to create io completion port\n");
ecc314ba	220	rc = 1;
03244c19	221	}
ecc314ba BC	222	}
ecc314ba BC	223
ecc314ba BC	224	return rc;
	225	}
	226
	227	static int fio_windowsaio_close_file(struct thread_data fio_unused td, struct fio_file f)
	228	{
4e79098f	229	int rc = 0;
1e42cc3d	230
9b836561 BC	231	dprint(FD_FILE, "fd close %s\n", f->file_name);
9b836561 BC	232
ecc314ba	233	if (f->hFile != INVALID_HANDLE_VALUE) {
03244c19 BC	234	if (!CloseHandle(f->hFile)) {
03244c19 BC	235	log_info("windowsaio: failed to close file handle for \"%s\"\n", f->file_name);
4e79098f	236	rc = 1;
03244c19	237	}
ecc314ba BC	238	}
	239
	240	f->hFile = INVALID_HANDLE_VALUE;
4e79098f	241	return rc;
ecc314ba BC	242	}
ecc314ba BC	243
67897036 BC	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	{
565e784d	261	struct windowsaio_data *wd = td->io_ops_data;
67897036 BC	262	return wd->aio_events[event];
	263	}
	264
	265	static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
1f440ece JA	266	unsigned int max,
1f440ece JA	267	const struct timespec *t)
67897036	268	{
565e784d	269	struct windowsaio_data *wd = td->io_ops_data;
67897036 BC	270	unsigned int dequeued = 0;
67897036 BC	271	struct io_u *io_u;
7b2dfab1	272	int i;
67897036 BC	273	struct fio_overlapped *fov;
	274	DWORD start_count = 0;
	275	DWORD end_count = 0;
	276	DWORD status;
	277	DWORD mswait = 250;
	278
	279	if (t != NULL) {
	280	mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	281	start_count = GetTickCount();
	282	end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	283	}
	284
	285	do {
7b2dfab1 BC	286	io_u_qiter(&td->io_u_all, io_u, i) {
	287	if (!(io_u->flags & IO_U_F_FLIGHT))
	288	continue;
	289
67897036 BC	290	fov = (struct fio_overlapped*)io_u->engine_data;
	291
	292	if (fov->io_complete) {
40c5db35	293	fov->io_complete = FALSE;
67897036 BC	294	wd->aio_events[dequeued] = io_u;
	295	dequeued++;
	296	}
	297
67897036	298	}
77e7b330 JR	299	if (dequeued >= min)
77e7b330 JR	300	break;
67897036	301
40c5db35	302	if (dequeued < min) {
67897036	303	status = WaitForSingleObject(wd->iocomplete_event, mswait);
f9a58c2a	304	if (status != WAIT_OBJECT_0 && dequeued >= min)
03244c19	305	break;
67897036 BC	306	}
	307
	308	if (dequeued >= min \|\| (t != NULL && timeout_expired(start_count, end_count)))
	309	break;
	310	} while (1);
	311
	312	return dequeued;
	313	}
	314
40c5db35	315	static int fio_windowsaio_queue(struct thread_data td, struct io_u io_u)
67897036	316	{
c73ed246 JA	317	struct fio_overlapped *o = io_u->engine_data;
c73ed246 JA	318	LPOVERLAPPED lpOvl = &o->o;
93bcfd20	319	BOOL success = FALSE;
67897036 BC	320	int rc = FIO_Q_COMPLETED;
	321
	322	fio_ro_check(td, io_u);
	323
77e7b330	324	lpOvl->Internal = 0;
4e79098f BC	325	lpOvl->InternalHigh = 0;
	326	lpOvl->Offset = io_u->offset & 0xFFFFFFFF;
	327	lpOvl->OffsetHigh = io_u->offset >> 32;
67897036 BC	328
67897036 BC	329	switch (io_u->ddir) {
40c5db35	330	case DDIR_WRITE:
77e7b330	331	success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, NULL, lpOvl);
67897036 BC	332	break;
67897036 BC	333	case DDIR_READ:
77e7b330	334	success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, NULL, lpOvl);
67897036 BC	335	break;
	336	case DDIR_SYNC:
	337	case DDIR_DATASYNC:
	338	case DDIR_SYNC_FILE_RANGE:
	339	success = FlushFileBuffers(io_u->file->hFile);
03244c19 BC	340	if (!success) {
	341	log_err("windowsaio: failed to flush file buffers\n");
	342	io_u->error = win_to_posix_error(GetLastError());
	343	}
67897036 BC	344
	345	return FIO_Q_COMPLETED;
	346	break;
	347	case DDIR_TRIM:
03244c19	348	log_err("windowsaio: manual TRIM isn't supported on Windows\n");
67897036 BC	349	io_u->error = 1;
	350	io_u->resid = io_u->xfer_buflen;
	351	return FIO_Q_COMPLETED;
	352	break;
	353	default:
	354	assert(0);
93bcfd20	355	break;
67897036 BC	356	}
67897036 BC	357
40c5db35	358	if (success \|\| GetLastError() == ERROR_IO_PENDING)
67897036	359	rc = FIO_Q_QUEUED;
40c5db35	360	else {
2277d5d5	361	io_u->error = win_to_posix_error(GetLastError());
67897036 BC	362	io_u->resid = io_u->xfer_buflen;
	363	}
	364
	365	return rc;
	366	}
	367
	368	/* Runs as a thread and waits for queued IO to complete */
	369	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter)
	370	{
	371	OVERLAPPED *ovl;
	372	struct fio_overlapped *fov;
	373	struct io_u *io_u;
	374	struct windowsaio_data *wd;
	375	struct thread_ctx *ctx;
	376	ULONG_PTR ulKey = 0;
	377	DWORD bytes;
	378
	379	ctx = (struct thread_ctx*)lpParameter;
	380	wd = ctx->wd;
	381
	382	do {
66c098b8	383	if (!GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey, &ovl, 250) && ovl == NULL)
67897036 BC	384	continue;
	385
	386	fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o);
	387	io_u = fov->io_u;
	388
	389	if (ovl->Internal == ERROR_SUCCESS) {
	390	io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
	391	io_u->error = 0;
	392	} else {
	393	io_u->resid = io_u->xfer_buflen;
2277d5d5	394	io_u->error = win_to_posix_error(GetLastError());
67897036 BC	395	}
67897036 BC	396
40c5db35	397	fov->io_complete = TRUE;
67897036 BC	398	SetEvent(wd->iocomplete_event);
	399	} while (ctx->wd->iothread_running);
	400
	401	CloseHandle(ctx->iocp);
	402	free(ctx);
	403	return 0;
	404	}
	405
c73ed246 JA	406	static void fio_windowsaio_io_u_free(struct thread_data td, struct io_u io_u)
	407	{
	408	struct fio_overlapped *o = io_u->engine_data;
	409
	410	if (o) {
c73ed246 JA	411	io_u->engine_data = NULL;
	412	free(o);
	413	}
	414	}
	415
	416	static int fio_windowsaio_io_u_init(struct thread_data td, struct io_u io_u)
	417	{
	418	struct fio_overlapped *o;
	419
	420	o = malloc(sizeof(*o));
b0106419	421	o->io_complete = FALSE;
c73ed246	422	o->io_u = io_u;
77e7b330	423	o->o.hEvent = NULL;
c73ed246 JA	424	io_u->engine_data = o;
	425	return 0;
	426	}
	427
ecc314ba BC	428	static struct ioengine_ops ioengine = {
	429	.name = "windowsaio",
	430	.version = FIO_IOOPS_VERSION,
	431	.init = fio_windowsaio_init,
	432	.queue = fio_windowsaio_queue,
ecc314ba BC	433	.getevents = fio_windowsaio_getevents,
	434	.event = fio_windowsaio_event,
	435	.cleanup = fio_windowsaio_cleanup,
	436	.open_file = fio_windowsaio_open_file,
	437	.close_file = fio_windowsaio_close_file,
c73ed246 JA	438	.get_file_size = generic_get_file_size,
	439	.io_u_init = fio_windowsaio_io_u_init,
	440	.io_u_free = fio_windowsaio_io_u_free,
ecc314ba BC	441	};
ecc314ba BC	442
c874d188	443	static void fio_init fio_windowsaio_register(void)
ecc314ba BC	444	{
	445	register_ioengine(&ioengine);
	446	}
	447
c874d188	448	static void fio_exit fio_windowsaio_unregister(void)
ecc314ba BC	449	{
	450	unregister_ioengine(&ioengine);
	451	}