[fio.git] / engines / sync.c

/*
 * sync/psync engine
 *
 * IO engine that does regular read(2)/write(2) with lseek(2) to transfer
 * data and IO engine that does regular pread(2)/pwrite(2) to transfer data.
 *
 */
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/uio.h>
#include <errno.h>

#include "../fio.h"
#include "../optgroup.h"
#include "../lib/rand.h"

/*
 * Sync engine uses engine_data to store last offset
 */
#define LAST_POS(f)	((f)->engine_pos)

struct syncio_data {
	struct iovec *iovecs;
	struct io_u **io_us;
	unsigned int queued;
	unsigned int events;
	unsigned long queued_bytes;

	unsigned long long last_offset;
	struct fio_file *last_file;
	enum fio_ddir last_ddir;

	struct frand_state rand_state;
};

#ifdef FIO_HAVE_PWRITEV2
struct psyncv2_options {
	void *pad;
	unsigned int hipri;
	unsigned int hipri_percentage;
};

static struct fio_option options[] = {
	{
		.name	= "hipri",
		.lname	= "RWF_HIPRI",
		.type	= FIO_OPT_STR_SET,
		.off1	= offsetof(struct psyncv2_options, hipri),
		.help	= "Set RWF_HIPRI for pwritev2/preadv2",
		.category = FIO_OPT_C_ENGINE,
		.group	= FIO_OPT_G_INVALID,
	},
	{
		.name	= "hipri_percentage",
		.lname	= "RWF_HIPRI_PERCENTAGE",
		.type	= FIO_OPT_INT,
		.off1	= offsetof(struct psyncv2_options, hipri_percentage),
		.minval	= 0,
		.maxval	= 100,
		.def    = "100",
		.help	= "Probabilistically set RWF_HIPRI for pwritev2/preadv2",
		.category = FIO_OPT_C_ENGINE,
		.group	= FIO_OPT_G_INVALID,
	},
	{
		.name	= NULL,
	},
};
#endif

static int fio_syncio_prep(struct thread_data *td, struct io_u *io_u)
{
	struct fio_file *f = io_u->file;

	if (!ddir_rw(io_u->ddir))
		return 0;

	if (LAST_POS(f) != -1ULL && LAST_POS(f) == io_u->offset)
		return 0;

	if (lseek(f->fd, io_u->offset, SEEK_SET) == -1) {
		td_verror(td, errno, "lseek");
		return 1;
	}

	return 0;
}

static int fio_io_end(struct thread_data *td, struct io_u *io_u, int ret)
{
	if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
		LAST_POS(io_u->file) = io_u->offset + ret;

	if (ret != (int) io_u->xfer_buflen) {
		if (ret >= 0) {
			io_u->resid = io_u->xfer_buflen - ret;
			io_u->error = 0;
			return FIO_Q_COMPLETED;
		} else
			io_u->error = errno;
	}

	if (io_u->error) {
		io_u_log_error(td, io_u);
		td_verror(td, io_u->error, "xfer");
	}

	return FIO_Q_COMPLETED;
}

#ifdef CONFIG_PWRITEV
static enum fio_q_status fio_pvsyncio_queue(struct thread_data *td,
					    struct io_u *io_u)
{
	struct syncio_data *sd = td->io_ops_data;
	struct iovec *iov = &sd->iovecs[0];
	struct fio_file *f = io_u->file;
	int ret;

	fio_ro_check(td, io_u);

	iov->iov_base = io_u->xfer_buf;
	iov->iov_len = io_u->xfer_buflen;

	if (io_u->ddir == DDIR_READ)
		ret = preadv(f->fd, iov, 1, io_u->offset);
	else if (io_u->ddir == DDIR_WRITE)
		ret = pwritev(f->fd, iov, 1, io_u->offset);
	else if (io_u->ddir == DDIR_TRIM) {
		do_io_u_trim(td, io_u);
		return FIO_Q_COMPLETED;
	} else
		ret = do_io_u_sync(td, io_u);

	return fio_io_end(td, io_u, ret);
}
#endif

#ifdef FIO_HAVE_PWRITEV2
static enum fio_q_status fio_pvsyncio2_queue(struct thread_data *td,
					     struct io_u *io_u)
{
	struct syncio_data *sd = td->io_ops_data;
	struct psyncv2_options *o = td->eo;
	struct iovec *iov = &sd->iovecs[0];
	struct fio_file *f = io_u->file;
	int ret, flags = 0;

	fio_ro_check(td, io_u);

	if (o->hipri &&
	    (rand_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
		flags |= RWF_HIPRI;

	iov->iov_base = io_u->xfer_buf;
	iov->iov_len = io_u->xfer_buflen;

	if (io_u->ddir == DDIR_READ)
		ret = preadv2(f->fd, iov, 1, io_u->offset, flags);
	else if (io_u->ddir == DDIR_WRITE)
		ret = pwritev2(f->fd, iov, 1, io_u->offset, flags);
	else if (io_u->ddir == DDIR_TRIM) {
		do_io_u_trim(td, io_u);
		return FIO_Q_COMPLETED;
	} else
		ret = do_io_u_sync(td, io_u);

	return fio_io_end(td, io_u, ret);
}
#endif

static enum fio_q_status fio_psyncio_queue(struct thread_data *td,
					   struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	int ret;

	fio_ro_check(td, io_u);

	if (io_u->ddir == DDIR_READ)
		ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	else if (io_u->ddir == DDIR_WRITE)
		ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	else if (io_u->ddir == DDIR_TRIM) {
		do_io_u_trim(td, io_u);
		return FIO_Q_COMPLETED;
	} else
		ret = do_io_u_sync(td, io_u);

	return fio_io_end(td, io_u, ret);
}

static enum fio_q_status fio_syncio_queue(struct thread_data *td,
					  struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	int ret;

	fio_ro_check(td, io_u);

	if (io_u->ddir == DDIR_READ)
		ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	else if (io_u->ddir == DDIR_WRITE)
		ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	else if (io_u->ddir == DDIR_TRIM) {
		do_io_u_trim(td, io_u);
		return FIO_Q_COMPLETED;
	} else
		ret = do_io_u_sync(td, io_u);

	return fio_io_end(td, io_u, ret);
}

static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
				 unsigned int max,
				 const struct timespec fio_unused *t)
{
	struct syncio_data *sd = td->io_ops_data;
	int ret;

	if (min) {
		ret = sd->events;
		sd->events = 0;
	} else
		ret = 0;

	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	return ret;
}

static struct io_u *fio_vsyncio_event(struct thread_data *td, int event)
{
	struct syncio_data *sd = td->io_ops_data;

	return sd->io_us[event];
}

static int fio_vsyncio_append(struct thread_data *td, struct io_u *io_u)
{
	struct syncio_data *sd = td->io_ops_data;

	if (ddir_sync(io_u->ddir))
		return 0;

	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	    io_u->ddir == sd->last_ddir)
		return 1;

	return 0;
}

static void fio_vsyncio_set_iov(struct syncio_data *sd, struct io_u *io_u,
				int idx)
{
	sd->io_us[idx] = io_u;
	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	sd->last_file = io_u->file;
	sd->last_ddir = io_u->ddir;
	sd->queued_bytes += io_u->xfer_buflen;
	sd->queued++;
}

static enum fio_q_status fio_vsyncio_queue(struct thread_data *td,
					   struct io_u *io_u)
{
	struct syncio_data *sd = td->io_ops_data;

	fio_ro_check(td, io_u);

	if (!fio_vsyncio_append(td, io_u)) {
		dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
		/*
		 * If we can't append and have stuff queued, tell fio to
		 * commit those first and then retry this io
		 */
		if (sd->queued)
			return FIO_Q_BUSY;
		if (ddir_sync(io_u->ddir)) {
			int ret = do_io_u_sync(td, io_u);

			return fio_io_end(td, io_u, ret);
		}

		sd->queued = 0;
		sd->queued_bytes = 0;
		fio_vsyncio_set_iov(sd, io_u, 0);
	} else {
		if (sd->queued == td->o.iodepth) {
			dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
			return FIO_Q_BUSY;
		}

		dprint(FD_IO, "vsyncio_queue: append\n");
		fio_vsyncio_set_iov(sd, io_u, sd->queued);
	}

	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	return FIO_Q_QUEUED;
}

/*
 * Check that we transferred all bytes, or saw an error, etc
 */
static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
{
	struct syncio_data *sd = td->io_ops_data;
	struct io_u *io_u;
	unsigned int i;
	int err;

	/*
	 * transferred everything, perfect
	 */
	if (bytes == sd->queued_bytes)
		return 0;

	err = errno;
	for (i = 0; i < sd->queued; i++) {
		io_u = sd->io_us[i];

		if (bytes == -1) {
			io_u->error = err;
		} else {
			unsigned int this_io;

			this_io = bytes;
			if (this_io > io_u->xfer_buflen)
				this_io = io_u->xfer_buflen;

			io_u->resid = io_u->xfer_buflen - this_io;
			io_u->error = 0;
			bytes -= this_io;
		}
	}

	if (bytes == -1) {
		td_verror(td, err, "xfer vsync");
		return -err;
	}

	return 0;
}

static int fio_vsyncio_commit(struct thread_data *td)
{
	struct syncio_data *sd = td->io_ops_data;
	struct fio_file *f;
	ssize_t ret;

	if (!sd->queued)
		return 0;

	io_u_mark_submit(td, sd->queued);
	f = sd->last_file;

	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
		int err = -errno;

		td_verror(td, errno, "lseek");
		return err;
	}

	if (sd->last_ddir == DDIR_READ)
		ret = readv(f->fd, sd->iovecs, sd->queued);
	else
		ret = writev(f->fd, sd->iovecs, sd->queued);

	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
	sd->events = sd->queued;
	sd->queued = 0;
	return fio_vsyncio_end(td, ret);
}

static int fio_vsyncio_init(struct thread_data *td)
{
	struct syncio_data *sd;

	sd = malloc(sizeof(*sd));
	memset(sd, 0, sizeof(*sd));
	sd->last_offset = -1ULL;
	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
	init_rand(&sd->rand_state, 0);

	td->io_ops_data = sd;
	return 0;
}

static void fio_vsyncio_cleanup(struct thread_data *td)
{
	struct syncio_data *sd = td->io_ops_data;

	if (sd) {
		free(sd->iovecs);
		free(sd->io_us);
		free(sd);
	}
}

static struct ioengine_ops ioengine_rw = {
	.name		= "sync",
	.version	= FIO_IOOPS_VERSION,
	.prep		= fio_syncio_prep,
	.queue		= fio_syncio_queue,
	.open_file	= generic_open_file,
	.close_file	= generic_close_file,
	.get_file_size	= generic_get_file_size,
	.flags		= FIO_SYNCIO,
};

static struct ioengine_ops ioengine_prw = {
	.name		= "psync",
	.version	= FIO_IOOPS_VERSION,
	.queue		= fio_psyncio_queue,
	.open_file	= generic_open_file,
	.close_file	= generic_close_file,
	.get_file_size	= generic_get_file_size,
	.flags		= FIO_SYNCIO,
};

static struct ioengine_ops ioengine_vrw = {
	.name		= "vsync",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_vsyncio_init,
	.cleanup	= fio_vsyncio_cleanup,
	.queue		= fio_vsyncio_queue,
	.commit		= fio_vsyncio_commit,
	.event		= fio_vsyncio_event,
	.getevents	= fio_vsyncio_getevents,
	.open_file	= generic_open_file,
	.close_file	= generic_close_file,
	.get_file_size	= generic_get_file_size,
	.flags		= FIO_SYNCIO,
};

#ifdef CONFIG_PWRITEV
static struct ioengine_ops ioengine_pvrw = {
	.name		= "pvsync",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_vsyncio_init,
	.cleanup	= fio_vsyncio_cleanup,
	.queue		= fio_pvsyncio_queue,
	.open_file	= generic_open_file,
	.close_file	= generic_close_file,
	.get_file_size	= generic_get_file_size,
	.flags		= FIO_SYNCIO,
};
#endif

#ifdef FIO_HAVE_PWRITEV2
static struct ioengine_ops ioengine_pvrw2 = {
	.name		= "pvsync2",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_vsyncio_init,
	.cleanup	= fio_vsyncio_cleanup,
	.queue		= fio_pvsyncio2_queue,
	.open_file	= generic_open_file,
	.close_file	= generic_close_file,
	.get_file_size	= generic_get_file_size,
	.flags		= FIO_SYNCIO,
	.options	= options,
	.option_struct_size	= sizeof(struct psyncv2_options),
};
#endif

static void fio_init fio_syncio_register(void)
{
	register_ioengine(&ioengine_rw);
	register_ioengine(&ioengine_prw);
	register_ioengine(&ioengine_vrw);
#ifdef CONFIG_PWRITEV
	register_ioengine(&ioengine_pvrw);
#endif
#ifdef FIO_HAVE_PWRITEV2
	register_ioengine(&ioengine_pvrw2);
#endif
}

static void fio_exit fio_syncio_unregister(void)
{
	unregister_ioengine(&ioengine_rw);
	unregister_ioengine(&ioengine_prw);
	unregister_ioengine(&ioengine_vrw);
#ifdef CONFIG_PWRITEV
	unregister_ioengine(&ioengine_pvrw);
#endif
#ifdef FIO_HAVE_PWRITEV2
	unregister_ioengine(&ioengine_pvrw2);
#endif
}
Commit	Line	Data
	1	/*
	2	* sync/psync engine
	3	*
	4	* IO engine that does regular read(2)/write(2) with lseek(2) to transfer
	5	* data and IO engine that does regular pread(2)/pwrite(2) to transfer data.
	6	*
	7	*/
	8	#include <stdio.h>
	9	#include <stdlib.h>
	10	#include <unistd.h>
	11	#include <sys/uio.h>
	12	#include <errno.h>
	13
	14	#include "../fio.h"
	15	#include "../optgroup.h"
	16	#include "../lib/rand.h"
	17
	18	/*
	19	* Sync engine uses engine_data to store last offset
	20	*/
	21	#define LAST_POS(f) ((f)->engine_pos)
	22
	23	struct syncio_data {
	24	struct iovec *iovecs;
	25	struct io_u **io_us;
	26	unsigned int queued;
	27	unsigned int events;
	28	unsigned long queued_bytes;
	29
	30	unsigned long long last_offset;
	31	struct fio_file *last_file;
	32	enum fio_ddir last_ddir;
	33
	34	struct frand_state rand_state;
	35	};
	36
	37	#ifdef FIO_HAVE_PWRITEV2
	38	struct psyncv2_options {
	39	void *pad;
	40	unsigned int hipri;
	41	unsigned int hipri_percentage;
	42	};
	43
	44	static struct fio_option options[] = {
	45	{
	46	.name = "hipri",
	47	.lname = "RWF_HIPRI",
	48	.type = FIO_OPT_STR_SET,
	49	.off1 = offsetof(struct psyncv2_options, hipri),
	50	.help = "Set RWF_HIPRI for pwritev2/preadv2",
	51	.category = FIO_OPT_C_ENGINE,
	52	.group = FIO_OPT_G_INVALID,
	53	},
	54	{
	55	.name = "hipri_percentage",
	56	.lname = "RWF_HIPRI_PERCENTAGE",
	57	.type = FIO_OPT_INT,
	58	.off1 = offsetof(struct psyncv2_options, hipri_percentage),
	59	.minval = 0,
	60	.maxval = 100,
	61	.def = "100",
	62	.help = "Probabilistically set RWF_HIPRI for pwritev2/preadv2",
	63	.category = FIO_OPT_C_ENGINE,
	64	.group = FIO_OPT_G_INVALID,
	65	},
	66	{
	67	.name = NULL,
	68	},
	69	};
	70	#endif
	71
	72	static int fio_syncio_prep(struct thread_data td, struct io_u io_u)
	73	{
	74	struct fio_file *f = io_u->file;
	75
	76	if (!ddir_rw(io_u->ddir))
	77	return 0;
	78
	79	if (LAST_POS(f) != -1ULL && LAST_POS(f) == io_u->offset)
	80	return 0;
	81
	82	if (lseek(f->fd, io_u->offset, SEEK_SET) == -1) {
	83	td_verror(td, errno, "lseek");
	84	return 1;
	85	}
	86
	87	return 0;
	88	}
	89
	90	static int fio_io_end(struct thread_data td, struct io_u io_u, int ret)
	91	{
	92	if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
	93	LAST_POS(io_u->file) = io_u->offset + ret;
	94
	95	if (ret != (int) io_u->xfer_buflen) {
	96	if (ret >= 0) {
	97	io_u->resid = io_u->xfer_buflen - ret;
	98	io_u->error = 0;
	99	return FIO_Q_COMPLETED;
	100	} else
	101	io_u->error = errno;
	102	}
	103
	104	if (io_u->error) {
	105	io_u_log_error(td, io_u);
	106	td_verror(td, io_u->error, "xfer");
	107	}
	108
	109	return FIO_Q_COMPLETED;
	110	}
	111
	112	#ifdef CONFIG_PWRITEV
	113	static enum fio_q_status fio_pvsyncio_queue(struct thread_data *td,
	114	struct io_u *io_u)
	115	{
	116	struct syncio_data *sd = td->io_ops_data;
	117	struct iovec *iov = &sd->iovecs[0];
	118	struct fio_file *f = io_u->file;
	119	int ret;
	120
	121	fio_ro_check(td, io_u);
	122
	123	iov->iov_base = io_u->xfer_buf;
	124	iov->iov_len = io_u->xfer_buflen;
	125
	126	if (io_u->ddir == DDIR_READ)
	127	ret = preadv(f->fd, iov, 1, io_u->offset);
	128	else if (io_u->ddir == DDIR_WRITE)
	129	ret = pwritev(f->fd, iov, 1, io_u->offset);
	130	else if (io_u->ddir == DDIR_TRIM) {
	131	do_io_u_trim(td, io_u);
	132	return FIO_Q_COMPLETED;
	133	} else
	134	ret = do_io_u_sync(td, io_u);
	135
	136	return fio_io_end(td, io_u, ret);
	137	}
	138	#endif
	139
	140	#ifdef FIO_HAVE_PWRITEV2
	141	static enum fio_q_status fio_pvsyncio2_queue(struct thread_data *td,
	142	struct io_u *io_u)
	143	{
	144	struct syncio_data *sd = td->io_ops_data;
	145	struct psyncv2_options *o = td->eo;
	146	struct iovec *iov = &sd->iovecs[0];
	147	struct fio_file *f = io_u->file;
	148	int ret, flags = 0;
	149
	150	fio_ro_check(td, io_u);
	151
	152	if (o->hipri &&
	153	(rand_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
	154	flags \|= RWF_HIPRI;
	155
	156	iov->iov_base = io_u->xfer_buf;
	157	iov->iov_len = io_u->xfer_buflen;
	158
	159	if (io_u->ddir == DDIR_READ)
	160	ret = preadv2(f->fd, iov, 1, io_u->offset, flags);
	161	else if (io_u->ddir == DDIR_WRITE)
	162	ret = pwritev2(f->fd, iov, 1, io_u->offset, flags);
	163	else if (io_u->ddir == DDIR_TRIM) {
	164	do_io_u_trim(td, io_u);
	165	return FIO_Q_COMPLETED;
	166	} else
	167	ret = do_io_u_sync(td, io_u);
	168
	169	return fio_io_end(td, io_u, ret);
	170	}
	171	#endif
	172
	173	static enum fio_q_status fio_psyncio_queue(struct thread_data *td,
	174	struct io_u *io_u)
	175	{
	176	struct fio_file *f = io_u->file;
	177	int ret;
	178
	179	fio_ro_check(td, io_u);
	180
	181	if (io_u->ddir == DDIR_READ)
	182	ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	183	else if (io_u->ddir == DDIR_WRITE)
	184	ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	185	else if (io_u->ddir == DDIR_TRIM) {
	186	do_io_u_trim(td, io_u);
	187	return FIO_Q_COMPLETED;
	188	} else
	189	ret = do_io_u_sync(td, io_u);
	190
	191	return fio_io_end(td, io_u, ret);
	192	}
	193
	194	static enum fio_q_status fio_syncio_queue(struct thread_data *td,
	195	struct io_u *io_u)
	196	{
	197	struct fio_file *f = io_u->file;
	198	int ret;
	199
	200	fio_ro_check(td, io_u);
	201
	202	if (io_u->ddir == DDIR_READ)
	203	ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	204	else if (io_u->ddir == DDIR_WRITE)
	205	ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	206	else if (io_u->ddir == DDIR_TRIM) {
	207	do_io_u_trim(td, io_u);
	208	return FIO_Q_COMPLETED;
	209	} else
	210	ret = do_io_u_sync(td, io_u);
	211
	212	return fio_io_end(td, io_u, ret);
	213	}
	214
	215	static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
	216	unsigned int max,
	217	const struct timespec fio_unused *t)
	218	{
	219	struct syncio_data *sd = td->io_ops_data;
	220	int ret;
	221
	222	if (min) {
	223	ret = sd->events;
	224	sd->events = 0;
	225	} else
	226	ret = 0;
	227
	228	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	229	return ret;
	230	}
	231
	232	static struct io_u fio_vsyncio_event(struct thread_data td, int event)
	233	{
	234	struct syncio_data *sd = td->io_ops_data;
	235
	236	return sd->io_us[event];
	237	}
	238
	239	static int fio_vsyncio_append(struct thread_data td, struct io_u io_u)
	240	{
	241	struct syncio_data *sd = td->io_ops_data;
	242
	243	if (ddir_sync(io_u->ddir))
	244	return 0;
	245
	246	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	247	io_u->ddir == sd->last_ddir)
	248	return 1;
	249
	250	return 0;
	251	}
	252
	253	static void fio_vsyncio_set_iov(struct syncio_data sd, struct io_u io_u,
	254	int idx)
	255	{
	256	sd->io_us[idx] = io_u;
	257	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	258	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
	259	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	260	sd->last_file = io_u->file;
	261	sd->last_ddir = io_u->ddir;
	262	sd->queued_bytes += io_u->xfer_buflen;
	263	sd->queued++;
	264	}
	265
	266	static enum fio_q_status fio_vsyncio_queue(struct thread_data *td,
	267	struct io_u *io_u)
	268	{
	269	struct syncio_data *sd = td->io_ops_data;
	270
	271	fio_ro_check(td, io_u);
	272
	273	if (!fio_vsyncio_append(td, io_u)) {
	274	dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
	275	/*
	276	* If we can't append and have stuff queued, tell fio to
	277	* commit those first and then retry this io
	278	*/
	279	if (sd->queued)
	280	return FIO_Q_BUSY;
	281	if (ddir_sync(io_u->ddir)) {
	282	int ret = do_io_u_sync(td, io_u);
	283
	284	return fio_io_end(td, io_u, ret);
	285	}
	286
	287	sd->queued = 0;
	288	sd->queued_bytes = 0;
	289	fio_vsyncio_set_iov(sd, io_u, 0);
	290	} else {
	291	if (sd->queued == td->o.iodepth) {
	292	dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
	293	return FIO_Q_BUSY;
	294	}
	295
	296	dprint(FD_IO, "vsyncio_queue: append\n");
	297	fio_vsyncio_set_iov(sd, io_u, sd->queued);
	298	}
	299
	300	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	301	return FIO_Q_QUEUED;
	302	}
	303
	304	/*
	305	* Check that we transferred all bytes, or saw an error, etc
	306	*/
	307	static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
	308	{
	309	struct syncio_data *sd = td->io_ops_data;
	310	struct io_u *io_u;
	311	unsigned int i;
	312	int err;
	313
	314	/*
	315	* transferred everything, perfect
	316	*/
	317	if (bytes == sd->queued_bytes)
	318	return 0;
	319
	320	err = errno;
	321	for (i = 0; i < sd->queued; i++) {
	322	io_u = sd->io_us[i];
	323
	324	if (bytes == -1) {
	325	io_u->error = err;
	326	} else {
	327	unsigned int this_io;
	328
	329	this_io = bytes;
	330	if (this_io > io_u->xfer_buflen)
	331	this_io = io_u->xfer_buflen;
	332
	333	io_u->resid = io_u->xfer_buflen - this_io;
	334	io_u->error = 0;
	335	bytes -= this_io;
	336	}
	337	}
	338
	339	if (bytes == -1) {
	340	td_verror(td, err, "xfer vsync");
	341	return -err;
	342	}
	343
	344	return 0;
	345	}
	346
	347	static int fio_vsyncio_commit(struct thread_data *td)
	348	{
	349	struct syncio_data *sd = td->io_ops_data;
	350	struct fio_file *f;
	351	ssize_t ret;
	352
	353	if (!sd->queued)
	354	return 0;
	355
	356	io_u_mark_submit(td, sd->queued);
	357	f = sd->last_file;
	358
	359	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
	360	int err = -errno;
	361
	362	td_verror(td, errno, "lseek");
	363	return err;
	364	}
	365
	366	if (sd->last_ddir == DDIR_READ)
	367	ret = readv(f->fd, sd->iovecs, sd->queued);
	368	else
	369	ret = writev(f->fd, sd->iovecs, sd->queued);
	370
	371	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
	372	sd->events = sd->queued;
	373	sd->queued = 0;
	374	return fio_vsyncio_end(td, ret);
	375	}
	376
	377	static int fio_vsyncio_init(struct thread_data *td)
	378	{
	379	struct syncio_data *sd;
	380
	381	sd = malloc(sizeof(*sd));
	382	memset(sd, 0, sizeof(*sd));
	383	sd->last_offset = -1ULL;
	384	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	385	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
	386	init_rand(&sd->rand_state, 0);
	387
	388	td->io_ops_data = sd;
	389	return 0;
	390	}
	391
	392	static void fio_vsyncio_cleanup(struct thread_data *td)
	393	{
	394	struct syncio_data *sd = td->io_ops_data;
	395
	396	if (sd) {
	397	free(sd->iovecs);
	398	free(sd->io_us);
	399	free(sd);
	400	}
	401	}
	402
	403	static struct ioengine_ops ioengine_rw = {
	404	.name = "sync",
	405	.version = FIO_IOOPS_VERSION,
	406	.prep = fio_syncio_prep,
	407	.queue = fio_syncio_queue,
	408	.open_file = generic_open_file,
	409	.close_file = generic_close_file,
	410	.get_file_size = generic_get_file_size,
	411	.flags = FIO_SYNCIO,
	412	};
	413
	414	static struct ioengine_ops ioengine_prw = {
	415	.name = "psync",
	416	.version = FIO_IOOPS_VERSION,
	417	.queue = fio_psyncio_queue,
	418	.open_file = generic_open_file,
	419	.close_file = generic_close_file,
	420	.get_file_size = generic_get_file_size,
	421	.flags = FIO_SYNCIO,
	422	};
	423
	424	static struct ioengine_ops ioengine_vrw = {
	425	.name = "vsync",
	426	.version = FIO_IOOPS_VERSION,
	427	.init = fio_vsyncio_init,
	428	.cleanup = fio_vsyncio_cleanup,
	429	.queue = fio_vsyncio_queue,
	430	.commit = fio_vsyncio_commit,
	431	.event = fio_vsyncio_event,
	432	.getevents = fio_vsyncio_getevents,
	433	.open_file = generic_open_file,
	434	.close_file = generic_close_file,
	435	.get_file_size = generic_get_file_size,
	436	.flags = FIO_SYNCIO,
	437	};
	438
	439	#ifdef CONFIG_PWRITEV
	440	static struct ioengine_ops ioengine_pvrw = {
	441	.name = "pvsync",
	442	.version = FIO_IOOPS_VERSION,
	443	.init = fio_vsyncio_init,
	444	.cleanup = fio_vsyncio_cleanup,
	445	.queue = fio_pvsyncio_queue,
	446	.open_file = generic_open_file,
	447	.close_file = generic_close_file,
	448	.get_file_size = generic_get_file_size,
	449	.flags = FIO_SYNCIO,
	450	};
	451	#endif
	452
	453	#ifdef FIO_HAVE_PWRITEV2
	454	static struct ioengine_ops ioengine_pvrw2 = {
	455	.name = "pvsync2",
	456	.version = FIO_IOOPS_VERSION,
	457	.init = fio_vsyncio_init,
	458	.cleanup = fio_vsyncio_cleanup,
	459	.queue = fio_pvsyncio2_queue,
	460	.open_file = generic_open_file,
	461	.close_file = generic_close_file,
	462	.get_file_size = generic_get_file_size,
	463	.flags = FIO_SYNCIO,
	464	.options = options,
	465	.option_struct_size = sizeof(struct psyncv2_options),
	466	};
	467	#endif
	468
	469	static void fio_init fio_syncio_register(void)
	470	{
	471	register_ioengine(&ioengine_rw);
	472	register_ioengine(&ioengine_prw);
	473	register_ioengine(&ioengine_vrw);
	474	#ifdef CONFIG_PWRITEV
	475	register_ioengine(&ioengine_pvrw);
	476	#endif
	477	#ifdef FIO_HAVE_PWRITEV2
	478	register_ioengine(&ioengine_pvrw2);
	479	#endif
	480	}
	481
	482	static void fio_exit fio_syncio_unregister(void)
	483	{
	484	unregister_ioengine(&ioengine_rw);
	485	unregister_ioengine(&ioengine_prw);
	486	unregister_ioengine(&ioengine_vrw);
	487	#ifdef CONFIG_PWRITEV
	488	unregister_ioengine(&ioengine_pvrw);
	489	#endif
	490	#ifdef FIO_HAVE_PWRITEV2
	491	unregister_ioengine(&ioengine_pvrw2);
	492	#endif
	493	}