[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"
#include "../optgroup.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;
};

struct windowsaio_options {
	struct thread_data *td;
	unsigned int no_completion_thread;
};

static struct fio_option options[] = {
	{
		.name	= "no_completion_thread",
		.lname	= "No completion polling thread",
		.type	= FIO_OPT_STR_SET,
		.off1	= offsetof(struct windowsaio_options, no_completion_thread),
		.help	= "Use to avoid separate completion polling thread",
		.category = FIO_OPT_C_ENGINE,
		.group	= FIO_OPT_G_WINDOWSAIO,
	},
	{
		.name	= NULL,
	},
};

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;
		struct windowsaio_options *o = td->eo;

		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 (o->no_completion_thread == 0) {
			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)
					log_err("windowsaio: failed to create io completion thread\n");
				else if (fio_option_is_set(&td->o, cpumask))
					fio_setaffinity(threadid, td->o.cpumask);
			}
			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 windowsaio_invalidate_cache(struct fio_file *f)
{
	DWORD error;
	DWORD isharemode = (FILE_SHARE_DELETE | FILE_SHARE_READ |
				FILE_SHARE_WRITE);
	HANDLE ihFile;
	int rc = 0;

	/*
	 * 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 %lu\n",
				 f->file_name, error);
			rc = 1;
		}
	} else {
		error = GetLastError();
		if (error != ERROR_FILE_NOT_FOUND) {
			log_info("windowsaio: cache invalidation of %s failed: error %lu\n",
				 f->file_name, error);
			rc = 1;
		}
	}

	return rc;
}

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) && !(td->flags & TD_F_SYNCS)) || read_only)
		access = GENERIC_READ;
	else
		access = (GENERIC_READ | GENERIC_WRITE);

	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) {
		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];
}

/* dequeue completion entrees directly (no separate completion thread) */
static int fio_windowsaio_getevents_nothread(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;
	DWORD start_count = 0;
	DWORD end_count = 0;
	DWORD mswait = 250;
	struct fio_overlapped *fov;

	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 {
		BOOL ret;
		OVERLAPPED *ovl;

		ULONG entries = min(16, max-dequeued);
		OVERLAPPED_ENTRY oe[16];
		ret = GetQueuedCompletionStatusEx(wd->iocp, oe, 16, &entries, mswait, 0);
		if (ret && entries) {
			int entry_num;

			for (entry_num=0; entry_num<entries; entry_num++) {
				ovl = oe[entry_num].lpOverlapped;
				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 = FALSE;
				wd->aio_events[dequeued] = io_u;
				dequeued++;
			}
		}

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

/* dequeue completion entrees creates by separate IoCompletionRoutine thread */
static int fio_windowaio_getevents_thread(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_getevents(struct thread_data *td, unsigned int min,
				    unsigned int max, const struct timespec *t)
{
	struct windowsaio_options *o = td->eo;

	if (o->no_completion_thread)
		return fio_windowsaio_getevents_nothread(td, min, max, t);
	return fio_windowaio_getevents_thread(td, min, max, t);
}

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;
}

/* 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 {
		BOOL ret;

		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);
	}
}

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,
	.options	= options,
	.option_struct_size	= sizeof(struct windowsaio_options),
};

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
ecc314ba BC	13	#include "../fio.h"
1388e473	14	#include "../optgroup.h"
ecc314ba	15
ea4500d8 BC	16	typedef BOOL (WINAPI *CANCELIOEX)(HANDLE hFile, LPOVERLAPPED lpOverlapped);
ea4500d8 BC	17
66c098b8 BC	18	int geterrno_from_win_error (DWORD code, int deferrno);
66c098b8 BC	19
67897036 BC	20	struct fio_overlapped {
	21	OVERLAPPED o;
	22	struct io_u *io_u;
	23	BOOL io_complete;
67897036	24	};
ecc314ba	25
9b836561	26	struct windowsaio_data {
9b836561	27	struct io_u **aio_events;
5a90bb5f	28	HANDLE iocp;
67897036	29	HANDLE iothread;
9b836561	30	HANDLE iocomplete_event;
67897036	31	BOOL iothread_running;
9b836561 BC	32	};
9b836561 BC	33
ecc314ba	34	struct thread_ctx {
9b836561	35	HANDLE iocp;
ecc314ba BC	36	struct windowsaio_data *wd;
	37	};
	38
1388e473	39	struct windowsaio_options {
	40	struct thread_data *td;
	41	unsigned int no_completion_thread;
	42	};
	43
	44	static struct fio_option options[] = {
	45	{
	46	.name = "no_completion_thread",
	47	.lname = "No completion polling thread",
	48	.type = FIO_OPT_STR_SET,
	49	.off1 = offsetof(struct windowsaio_options, no_completion_thread),
	50	.help = "Use to avoid separate completion polling thread",
	51	.category = FIO_OPT_C_ENGINE,
	52	.group = FIO_OPT_G_WINDOWSAIO,
	53	},
	54	{
	55	.name = NULL,
	56	},
	57	};
	58
ecc314ba	59	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter);
ecc314ba	60
ecc314ba BC	61	static int fio_windowsaio_init(struct thread_data *td)
ecc314ba BC	62	{
ecc314ba	63	struct windowsaio_data *wd;
9b836561	64	int rc = 0;
ecc314ba	65
03244c19 BC	66	wd = calloc(1, sizeof(struct windowsaio_data));
	67	if (wd == NULL) {
	68	log_err("windowsaio: failed to allocate memory for engine data\n");
9b836561	69	rc = 1;
03244c19	70	}
ecc314ba	71
9b836561 BC	72	if (!rc) {
9b836561 BC	73	wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*));
03244c19 BC	74	if (wd->aio_events == NULL) {
03244c19 BC	75	log_err("windowsaio: failed to allocate memory for aio events list\n");
9b836561	76	rc = 1;
03244c19	77	}
e4db9fec BC	78	}
e4db9fec BC	79
9b836561 BC	80	if (!rc) {
	81	/* Create an auto-reset event */
	82	wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL);
03244c19 BC	83	if (wd->iocomplete_event == NULL) {
03244c19 BC	84	log_err("windowsaio: failed to create io complete event handle\n");
9b836561	85	rc = 1;
03244c19	86	}
9b836561 BC	87	}
9b836561 BC	88
9b836561	89	if (rc) {
9b836561	90	if (wd != NULL) {
9b836561 BC	91	if (wd->aio_events != NULL)
	92	free(wd->aio_events);
	93
	94	free(wd);
	95	}
	96	}
ecc314ba	97
565e784d	98	td->io_ops_data = wd;
93bcfd20	99
5a90bb5f BC	100	if (!rc) {
	101	struct thread_ctx *ctx;
	102	struct windowsaio_data *wd;
	103	HANDLE hFile;
1388e473	104	struct windowsaio_options *o = td->eo;
5a90bb5f BC	105
5a90bb5f BC	106	hFile = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 0);
03244c19 BC	107	if (hFile == INVALID_HANDLE_VALUE) {
03244c19 BC	108	log_err("windowsaio: failed to create io completion port\n");
5a90bb5f	109	rc = 1;
03244c19	110	}
5a90bb5f	111
565e784d	112	wd = td->io_ops_data;
5a90bb5f BC	113	wd->iothread_running = TRUE;
	114	wd->iocp = hFile;
	115
1388e473	116	if (o->no_completion_thread == 0) {
	117	if (!rc)
	118	ctx = malloc(sizeof(struct thread_ctx));
5a90bb5f	119
1388e473	120	if (!rc && ctx == NULL) {
	121	log_err("windowsaio: failed to allocate memory for thread context structure\n");
	122	CloseHandle(hFile);
	123	rc = 1;
	124	}
5a90bb5f	125
1388e473	126	if (!rc) {
1388e473	127	DWORD threadid;
438bb1cf	128
1388e473	129	ctx->iocp = hFile;
	130	ctx->wd = wd;
	131	wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, &threadid);
	132	if (!wd->iothread)
	133	log_err("windowsaio: failed to create io completion thread\n");
	134	else if (fio_option_is_set(&td->o, cpumask))
	135	fio_setaffinity(threadid, td->o.cpumask);
	136	}
	137	if (rc \|\| wd->iothread == NULL)
	138	rc = 1;
5a90bb5f	139	}
5a90bb5f BC	140	}
5a90bb5f BC	141
4e79098f	142	return rc;
ecc314ba BC	143	}
ecc314ba BC	144
67897036 BC	145	static void fio_windowsaio_cleanup(struct thread_data *td)
67897036 BC	146	{
67897036 BC	147	struct windowsaio_data *wd;
67897036 BC	148
565e784d	149	wd = td->io_ops_data;
67897036 BC	150
67897036 BC	151	if (wd != NULL) {
40c5db35 JA	152	wd->iothread_running = FALSE;
40c5db35 JA	153	WaitForSingleObject(wd->iothread, INFINITE);
67897036 BC	154
	155	CloseHandle(wd->iothread);
	156	CloseHandle(wd->iocomplete_event);
	157
67897036	158	free(wd->aio_events);
67897036 BC	159	free(wd);
67897036 BC	160
565e784d	161	td->io_ops_data = NULL;
67897036 BC	162	}
	163	}
	164
8300eba5 SW	165	static int windowsaio_invalidate_cache(struct fio_file *f)
	166	{
	167	DWORD error;
	168	DWORD isharemode = (FILE_SHARE_DELETE \| FILE_SHARE_READ \|
1633aa61	169	FILE_SHARE_WRITE);
8300eba5 SW	170	HANDLE ihFile;
	171	int rc = 0;
	172
	173	/*
	174	* Encourage Windows to drop cached parts of a file by temporarily
	175	* opening it for non-buffered access. Note: this will only work when
	176	* the following is the only thing with the file open on the whole
	177	* system.
	178	*/
	179	dprint(FD_IO, "windowaio: attempt invalidate cache for %s\n",
	180	f->file_name);
	181	ihFile = CreateFile(f->file_name, 0, isharemode, NULL, OPEN_EXISTING,
	182	FILE_FLAG_NO_BUFFERING, NULL);
	183
	184	if (ihFile != INVALID_HANDLE_VALUE) {
	185	if (!CloseHandle(ihFile)) {
	186	error = GetLastError();
4937100f SW	187	log_info("windowsaio: invalidation fd close %s failed: error %lu\n",
4937100f SW	188	f->file_name, error);
8300eba5 SW	189	rc = 1;
	190	}
	191	} else {
	192	error = GetLastError();
	193	if (error != ERROR_FILE_NOT_FOUND) {
4937100f SW	194	log_info("windowsaio: cache invalidation of %s failed: error %lu\n",
4937100f SW	195	f->file_name, error);
8300eba5 SW	196	rc = 1;
	197	}
	198	}
	199
	200	return rc;
	201	}
	202
ecc314ba BC	203	static int fio_windowsaio_open_file(struct thread_data td, struct fio_file f)
	204	{
	205	int rc = 0;
4e79098f	206	DWORD flags = FILE_FLAG_POSIX_SEMANTICS \| FILE_FLAG_OVERLAPPED;
ecc314ba BC	207	DWORD sharemode = FILE_SHARE_READ \| FILE_SHARE_WRITE;
	208	DWORD openmode = OPEN_ALWAYS;
	209	DWORD access;
	210
	211	dprint(FD_FILE, "fd open %s\n", f->file_name);
	212
ecc314ba	213	if (f->filetype == FIO_TYPE_PIPE) {
03244c19	214	log_err("windowsaio: pipes are not supported\n");
ecc314ba BC	215	return 1;
	216	}
	217
	218	if (!strcmp(f->file_name, "-")) {
03244c19	219	log_err("windowsaio: can't read/write to stdin/out\n");
ecc314ba BC	220	return 1;
	221	}
	222
	223	if (td->o.odirect)
	224	flags \|= FILE_FLAG_NO_BUFFERING;
	225	if (td->o.sync_io)
	226	flags \|= FILE_FLAG_WRITE_THROUGH;
	227
93bcfd20 BC	228	/*
93bcfd20 BC	229	* Inform Windows whether we're going to be doing sequential or
f20a86a7	230	* random IO so it can tune the Cache Manager
93bcfd20	231	*/
f20a86a7 SW	232	switch (td->o.fadvise_hint) {
	233	case F_ADV_TYPE:
	234	if (td_random(td))
	235	flags \|= FILE_FLAG_RANDOM_ACCESS;
	236	else
	237	flags \|= FILE_FLAG_SEQUENTIAL_SCAN;
	238	break;
	239	case F_ADV_RANDOM:
ecc314ba	240	flags \|= FILE_FLAG_RANDOM_ACCESS;
f20a86a7 SW	241	break;
	242	case F_ADV_SEQUENTIAL:
	243	flags \|= FILE_FLAG_SEQUENTIAL_SCAN;
	244	break;
	245	case F_ADV_NONE:
	246	break;
	247	default:
	248	log_err("fio: unknown fadvise type %d\n", td->o.fadvise_hint);
	249	}
ecc314ba	250
c3773c17	251	if ((!td_write(td) && !(td->flags & TD_F_SYNCS)) \|\| read_only)
ecc314ba BC	252	access = GENERIC_READ;
	253	else
	254	access = (GENERIC_READ \| GENERIC_WRITE);
	255
93bcfd20	256	if (td->o.create_on_open)
ecc314ba BC	257	openmode = OPEN_ALWAYS;
	258	else
	259	openmode = OPEN_EXISTING;
	260
8300eba5 SW	261	/* If we're going to use direct I/O, Windows will try and invalidate
8300eba5 SW	262	* its cache at that point so there's no need to do it here */
3e9ae304	263	if (td->o.invalidate_cache && !td->o.odirect)
8300eba5	264	windowsaio_invalidate_cache(f);
8300eba5	265
ecc314ba BC	266	f->hFile = CreateFile(f->file_name, access, sharemode,
	267	NULL, openmode, flags, NULL);
	268
03244c19 BC	269	if (f->hFile == INVALID_HANDLE_VALUE) {
03244c19 BC	270	log_err("windowsaio: failed to open file \"%s\"\n", f->file_name);
ecc314ba	271	rc = 1;
03244c19	272	}
ecc314ba	273
93bcfd20	274	/* Only set up the completion port and thread if we're not just
ecc314ba	275	* querying the device size */
565e784d	276	if (!rc && td->io_ops_data != NULL) {
40c5db35	277	struct windowsaio_data *wd;
ecc314ba	278
565e784d	279	wd = td->io_ops_data;
ecc314ba	280
03244c19 BC	281	if (CreateIoCompletionPort(f->hFile, wd->iocp, 0, 0) == NULL) {
03244c19 BC	282	log_err("windowsaio: failed to create io completion port\n");
ecc314ba	283	rc = 1;
03244c19	284	}
ecc314ba BC	285	}
ecc314ba BC	286
ecc314ba BC	287	return rc;
	288	}
	289
	290	static int fio_windowsaio_close_file(struct thread_data fio_unused td, struct fio_file f)
	291	{
4e79098f	292	int rc = 0;
1e42cc3d	293
9b836561 BC	294	dprint(FD_FILE, "fd close %s\n", f->file_name);
9b836561 BC	295
ecc314ba	296	if (f->hFile != INVALID_HANDLE_VALUE) {
03244c19 BC	297	if (!CloseHandle(f->hFile)) {
03244c19 BC	298	log_info("windowsaio: failed to close file handle for \"%s\"\n", f->file_name);
4e79098f	299	rc = 1;
03244c19	300	}
ecc314ba BC	301	}
	302
	303	f->hFile = INVALID_HANDLE_VALUE;
4e79098f	304	return rc;
ecc314ba BC	305	}
ecc314ba BC	306
67897036 BC	307	static BOOL timeout_expired(DWORD start_count, DWORD end_count)
	308	{
	309	BOOL expired = FALSE;
	310	DWORD current_time;
	311
	312	current_time = GetTickCount();
	313
	314	if ((end_count > start_count) && current_time >= end_count)
	315	expired = TRUE;
	316	else if (current_time < start_count && current_time > end_count)
	317	expired = TRUE;
	318
	319	return expired;
	320	}
	321
	322	static struct io_u* fio_windowsaio_event(struct thread_data *td, int event)
	323	{
565e784d	324	struct windowsaio_data *wd = td->io_ops_data;
67897036 BC	325	return wd->aio_events[event];
	326	}
	327
1388e473	328	/* dequeue completion entrees directly (no separate completion thread) */
	329	static int fio_windowsaio_getevents_nothread(struct thread_data *td, unsigned int min,
	330	unsigned int max, const struct timespec *t)
	331	{
	332	struct windowsaio_data *wd = td->io_ops_data;
	333	unsigned int dequeued = 0;
	334	struct io_u *io_u;
	335	DWORD start_count = 0;
	336	DWORD end_count = 0;
	337	DWORD mswait = 250;
	338	struct fio_overlapped *fov;
	339
	340	if (t != NULL) {
	341	mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	342	start_count = GetTickCount();
	343	end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	344	}
	345
	346	do {
	347	BOOL ret;
	348	OVERLAPPED *ovl;
	349
	350	ULONG entries = min(16, max-dequeued);
	351	OVERLAPPED_ENTRY oe[16];
	352	ret = GetQueuedCompletionStatusEx(wd->iocp, oe, 16, &entries, mswait, 0);
	353	if (ret && entries) {
	354	int entry_num;
	355
	356	for (entry_num=0; entry_num<entries; entry_num++) {
	357	ovl = oe[entry_num].lpOverlapped;
	358	fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o);
	359	io_u = fov->io_u;
	360
	361	if (ovl->Internal == ERROR_SUCCESS) {
	362	io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
	363	io_u->error = 0;
	364	} else {
	365	io_u->resid = io_u->xfer_buflen;
	366	io_u->error = win_to_posix_error(GetLastError());
	367	}
	368
	369	fov->io_complete = FALSE;
	370	wd->aio_events[dequeued] = io_u;
	371	dequeued++;
	372	}
	373	}
	374
	375	if (dequeued >= min \|\|
	376	(t != NULL && timeout_expired(start_count, end_count)))
	377	break;
	378	} while (1);
	379	return dequeued;
	380	}
	381
	382	/* dequeue completion entrees creates by separate IoCompletionRoutine thread */
	383	static int fio_windowaio_getevents_thread(struct thread_data *td, unsigned int min,
	384	unsigned int max, const struct timespec *t)
67897036	385	{
565e784d	386	struct windowsaio_data *wd = td->io_ops_data;
67897036 BC	387	unsigned int dequeued = 0;
67897036 BC	388	struct io_u *io_u;
7b2dfab1	389	int i;
67897036 BC	390	struct fio_overlapped *fov;
	391	DWORD start_count = 0;
	392	DWORD end_count = 0;
	393	DWORD status;
	394	DWORD mswait = 250;
	395
	396	if (t != NULL) {
	397	mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	398	start_count = GetTickCount();
	399	end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000);
	400	}
	401
	402	do {
7b2dfab1 BC	403	io_u_qiter(&td->io_u_all, io_u, i) {
	404	if (!(io_u->flags & IO_U_F_FLIGHT))
	405	continue;
	406
67897036 BC	407	fov = (struct fio_overlapped*)io_u->engine_data;
	408
	409	if (fov->io_complete) {
40c5db35	410	fov->io_complete = FALSE;
67897036 BC	411	wd->aio_events[dequeued] = io_u;
	412	dequeued++;
	413	}
67897036	414	}
77e7b330 JR	415	if (dequeued >= min)
77e7b330 JR	416	break;
67897036	417
40c5db35	418	if (dequeued < min) {
67897036	419	status = WaitForSingleObject(wd->iocomplete_event, mswait);
f9a58c2a	420	if (status != WAIT_OBJECT_0 && dequeued >= min)
03244c19	421	break;
67897036 BC	422	}
67897036 BC	423
1633aa61 JA	424	if (dequeued >= min \|\|
1633aa61 JA	425	(t != NULL && timeout_expired(start_count, end_count)))
67897036 BC	426	break;
	427	} while (1);
	428
	429	return dequeued;
	430	}
	431
1388e473	432	static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
	433	unsigned int max, const struct timespec *t)
	434	{
	435	struct windowsaio_options *o = td->eo;
	436
	437	if (o->no_completion_thread)
	438	return fio_windowsaio_getevents_nothread(td, min, max, t);
	439	return fio_windowaio_getevents_thread(td, min, max, t);
	440	}
	441
2e4ef4fb JA	442	static enum fio_q_status fio_windowsaio_queue(struct thread_data *td,
2e4ef4fb JA	443	struct io_u *io_u)
67897036	444	{
c73ed246 JA	445	struct fio_overlapped *o = io_u->engine_data;
c73ed246 JA	446	LPOVERLAPPED lpOvl = &o->o;
93bcfd20	447	BOOL success = FALSE;
67897036 BC	448	int rc = FIO_Q_COMPLETED;
	449
	450	fio_ro_check(td, io_u);
	451
77e7b330	452	lpOvl->Internal = 0;
4e79098f BC	453	lpOvl->InternalHigh = 0;
	454	lpOvl->Offset = io_u->offset & 0xFFFFFFFF;
	455	lpOvl->OffsetHigh = io_u->offset >> 32;
67897036 BC	456
67897036 BC	457	switch (io_u->ddir) {
40c5db35	458	case DDIR_WRITE:
1633aa61 JA	459	success = WriteFile(io_u->file->hFile, io_u->xfer_buf,
1633aa61 JA	460	io_u->xfer_buflen, NULL, lpOvl);
67897036 BC	461	break;
67897036 BC	462	case DDIR_READ:
1633aa61 JA	463	success = ReadFile(io_u->file->hFile, io_u->xfer_buf,
1633aa61 JA	464	io_u->xfer_buflen, NULL, lpOvl);
67897036 BC	465	break;
	466	case DDIR_SYNC:
	467	case DDIR_DATASYNC:
	468	case DDIR_SYNC_FILE_RANGE:
	469	success = FlushFileBuffers(io_u->file->hFile);
03244c19 BC	470	if (!success) {
	471	log_err("windowsaio: failed to flush file buffers\n");
	472	io_u->error = win_to_posix_error(GetLastError());
	473	}
67897036 BC	474
67897036 BC	475	return FIO_Q_COMPLETED;
67897036	476	case DDIR_TRIM:
03244c19	477	log_err("windowsaio: manual TRIM isn't supported on Windows\n");
67897036 BC	478	io_u->error = 1;
	479	io_u->resid = io_u->xfer_buflen;
	480	return FIO_Q_COMPLETED;
67897036 BC	481	default:
67897036 BC	482	assert(0);
93bcfd20	483	break;
67897036 BC	484	}
67897036 BC	485
40c5db35	486	if (success \|\| GetLastError() == ERROR_IO_PENDING)
67897036	487	rc = FIO_Q_QUEUED;
40c5db35	488	else {
2277d5d5	489	io_u->error = win_to_posix_error(GetLastError());
67897036 BC	490	io_u->resid = io_u->xfer_buflen;
	491	}
	492
	493	return rc;
	494	}
	495
	496	/* Runs as a thread and waits for queued IO to complete */
	497	static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter)
	498	{
	499	OVERLAPPED *ovl;
	500	struct fio_overlapped *fov;
	501	struct io_u *io_u;
	502	struct windowsaio_data *wd;
	503	struct thread_ctx *ctx;
	504	ULONG_PTR ulKey = 0;
	505	DWORD bytes;
	506
	507	ctx = (struct thread_ctx*)lpParameter;
	508	wd = ctx->wd;
	509
	510	do {
1633aa61 JA	511	BOOL ret;
	512
	513	ret = GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey,
	514	&ovl, 250);
	515	if (!ret && ovl == NULL)
67897036 BC	516	continue;
	517
	518	fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o);
	519	io_u = fov->io_u;
	520
	521	if (ovl->Internal == ERROR_SUCCESS) {
	522	io_u->resid = io_u->xfer_buflen - ovl->InternalHigh;
	523	io_u->error = 0;
	524	} else {
	525	io_u->resid = io_u->xfer_buflen;
2277d5d5	526	io_u->error = win_to_posix_error(GetLastError());
67897036 BC	527	}
67897036 BC	528
40c5db35	529	fov->io_complete = TRUE;
67897036 BC	530	SetEvent(wd->iocomplete_event);
	531	} while (ctx->wd->iothread_running);
	532
	533	CloseHandle(ctx->iocp);
	534	free(ctx);
	535	return 0;
	536	}
	537
c73ed246 JA	538	static void fio_windowsaio_io_u_free(struct thread_data td, struct io_u io_u)
	539	{
	540	struct fio_overlapped *o = io_u->engine_data;
	541
	542	if (o) {
c73ed246 JA	543	io_u->engine_data = NULL;
	544	free(o);
	545	}
	546	}
	547
	548	static int fio_windowsaio_io_u_init(struct thread_data td, struct io_u io_u)
	549	{
	550	struct fio_overlapped *o;
	551
	552	o = malloc(sizeof(*o));
b0106419	553	o->io_complete = FALSE;
c73ed246	554	o->io_u = io_u;
77e7b330	555	o->o.hEvent = NULL;
c73ed246 JA	556	io_u->engine_data = o;
	557	return 0;
	558	}
	559
ecc314ba BC	560	static struct ioengine_ops ioengine = {
	561	.name = "windowsaio",
	562	.version = FIO_IOOPS_VERSION,
	563	.init = fio_windowsaio_init,
	564	.queue = fio_windowsaio_queue,
ecc314ba BC	565	.getevents = fio_windowsaio_getevents,
	566	.event = fio_windowsaio_event,
	567	.cleanup = fio_windowsaio_cleanup,
	568	.open_file = fio_windowsaio_open_file,
	569	.close_file = fio_windowsaio_close_file,
c73ed246 JA	570	.get_file_size = generic_get_file_size,
	571	.io_u_init = fio_windowsaio_io_u_init,
	572	.io_u_free = fio_windowsaio_io_u_free,
1388e473	573	.options = options,
1388e473	574	.option_struct_size = sizeof(struct windowsaio_options),
ecc314ba BC	575	};
ecc314ba BC	576
c874d188	577	static void fio_init fio_windowsaio_register(void)
ecc314ba BC	578	{
	579	register_ioengine(&ioengine);
	580	}
	581
c874d188	582	static void fio_exit fio_windowsaio_unregister(void)
ecc314ba BC	583	{
	584	unregister_ioengine(&ioengine);
	585	}