[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 int 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 int 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 &&
	    (rand32_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 int 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 int 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 int 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 int fio_pvsyncio_queue(struct thread_data td, struct io_u io_u)
	114	{
	115	struct syncio_data *sd = td->io_ops_data;
	116	struct iovec *iov = &sd->iovecs[0];
	117	struct fio_file *f = io_u->file;
	118	int ret;
	119
	120	fio_ro_check(td, io_u);
	121
	122	iov->iov_base = io_u->xfer_buf;
	123	iov->iov_len = io_u->xfer_buflen;
	124
	125	if (io_u->ddir == DDIR_READ)
	126	ret = preadv(f->fd, iov, 1, io_u->offset);
	127	else if (io_u->ddir == DDIR_WRITE)
	128	ret = pwritev(f->fd, iov, 1, io_u->offset);
	129	else if (io_u->ddir == DDIR_TRIM) {
	130	do_io_u_trim(td, io_u);
	131	return FIO_Q_COMPLETED;
	132	} else
	133	ret = do_io_u_sync(td, io_u);
	134
	135	return fio_io_end(td, io_u, ret);
	136	}
	137	#endif
	138
	139	#ifdef FIO_HAVE_PWRITEV2
	140	static int fio_pvsyncio2_queue(struct thread_data td, struct io_u io_u)
	141	{
	142	struct syncio_data *sd = td->io_ops_data;
	143	struct psyncv2_options *o = td->eo;
	144	struct iovec *iov = &sd->iovecs[0];
	145	struct fio_file *f = io_u->file;
	146	int ret, flags = 0;
	147
	148	fio_ro_check(td, io_u);
	149
	150	if (o->hipri &&
	151	(rand32_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
	152	flags \|= RWF_HIPRI;
	153
	154	iov->iov_base = io_u->xfer_buf;
	155	iov->iov_len = io_u->xfer_buflen;
	156
	157	if (io_u->ddir == DDIR_READ)
	158	ret = preadv2(f->fd, iov, 1, io_u->offset, flags);
	159	else if (io_u->ddir == DDIR_WRITE)
	160	ret = pwritev2(f->fd, iov, 1, io_u->offset, flags);
	161	else if (io_u->ddir == DDIR_TRIM) {
	162	do_io_u_trim(td, io_u);
	163	return FIO_Q_COMPLETED;
	164	} else
	165	ret = do_io_u_sync(td, io_u);
	166
	167	return fio_io_end(td, io_u, ret);
	168	}
	169	#endif
	170
	171
	172	static int fio_psyncio_queue(struct thread_data td, struct io_u io_u)
	173	{
	174	struct fio_file *f = io_u->file;
	175	int ret;
	176
	177	fio_ro_check(td, io_u);
	178
	179	if (io_u->ddir == DDIR_READ)
	180	ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	181	else if (io_u->ddir == DDIR_WRITE)
	182	ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	183	else if (io_u->ddir == DDIR_TRIM) {
	184	do_io_u_trim(td, io_u);
	185	return FIO_Q_COMPLETED;
	186	} else
	187	ret = do_io_u_sync(td, io_u);
	188
	189	return fio_io_end(td, io_u, ret);
	190	}
	191
	192	static int fio_syncio_queue(struct thread_data td, struct io_u io_u)
	193	{
	194	struct fio_file *f = io_u->file;
	195	int ret;
	196
	197	fio_ro_check(td, io_u);
	198
	199	if (io_u->ddir == DDIR_READ)
	200	ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	201	else if (io_u->ddir == DDIR_WRITE)
	202	ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	203	else if (io_u->ddir == DDIR_TRIM) {
	204	do_io_u_trim(td, io_u);
	205	return FIO_Q_COMPLETED;
	206	} else
	207	ret = do_io_u_sync(td, io_u);
	208
	209	return fio_io_end(td, io_u, ret);
	210	}
	211
	212	static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
	213	unsigned int max,
	214	const struct timespec fio_unused *t)
	215	{
	216	struct syncio_data *sd = td->io_ops_data;
	217	int ret;
	218
	219	if (min) {
	220	ret = sd->events;
	221	sd->events = 0;
	222	} else
	223	ret = 0;
	224
	225	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	226	return ret;
	227	}
	228
	229	static struct io_u fio_vsyncio_event(struct thread_data td, int event)
	230	{
	231	struct syncio_data *sd = td->io_ops_data;
	232
	233	return sd->io_us[event];
	234	}
	235
	236	static int fio_vsyncio_append(struct thread_data td, struct io_u io_u)
	237	{
	238	struct syncio_data *sd = td->io_ops_data;
	239
	240	if (ddir_sync(io_u->ddir))
	241	return 0;
	242
	243	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	244	io_u->ddir == sd->last_ddir)
	245	return 1;
	246
	247	return 0;
	248	}
	249
	250	static void fio_vsyncio_set_iov(struct syncio_data sd, struct io_u io_u,
	251	int idx)
	252	{
	253	sd->io_us[idx] = io_u;
	254	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	255	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
	256	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	257	sd->last_file = io_u->file;
	258	sd->last_ddir = io_u->ddir;
	259	sd->queued_bytes += io_u->xfer_buflen;
	260	sd->queued++;
	261	}
	262
	263	static int fio_vsyncio_queue(struct thread_data td, struct io_u io_u)
	264	{
	265	struct syncio_data *sd = td->io_ops_data;
	266
	267	fio_ro_check(td, io_u);
	268
	269	if (!fio_vsyncio_append(td, io_u)) {
	270	dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
	271	/*
	272	* If we can't append and have stuff queued, tell fio to
	273	* commit those first and then retry this io
	274	*/
	275	if (sd->queued)
	276	return FIO_Q_BUSY;
	277	if (ddir_sync(io_u->ddir)) {
	278	int ret = do_io_u_sync(td, io_u);
	279
	280	return fio_io_end(td, io_u, ret);
	281	}
	282
	283	sd->queued = 0;
	284	sd->queued_bytes = 0;
	285	fio_vsyncio_set_iov(sd, io_u, 0);
	286	} else {
	287	if (sd->queued == td->o.iodepth) {
	288	dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
	289	return FIO_Q_BUSY;
	290	}
	291
	292	dprint(FD_IO, "vsyncio_queue: append\n");
	293	fio_vsyncio_set_iov(sd, io_u, sd->queued);
	294	}
	295
	296	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	297	return FIO_Q_QUEUED;
	298	}
	299
	300	/*
	301	* Check that we transferred all bytes, or saw an error, etc
	302	*/
	303	static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
	304	{
	305	struct syncio_data *sd = td->io_ops_data;
	306	struct io_u *io_u;
	307	unsigned int i;
	308	int err;
	309
	310	/*
	311	* transferred everything, perfect
	312	*/
	313	if (bytes == sd->queued_bytes)
	314	return 0;
	315
	316	err = errno;
	317	for (i = 0; i < sd->queued; i++) {
	318	io_u = sd->io_us[i];
	319
	320	if (bytes == -1) {
	321	io_u->error = err;
	322	} else {
	323	unsigned int this_io;
	324
	325	this_io = bytes;
	326	if (this_io > io_u->xfer_buflen)
	327	this_io = io_u->xfer_buflen;
	328
	329	io_u->resid = io_u->xfer_buflen - this_io;
	330	io_u->error = 0;
	331	bytes -= this_io;
	332	}
	333	}
	334
	335	if (bytes == -1) {
	336	td_verror(td, err, "xfer vsync");
	337	return -err;
	338	}
	339
	340	return 0;
	341	}
	342
	343	static int fio_vsyncio_commit(struct thread_data *td)
	344	{
	345	struct syncio_data *sd = td->io_ops_data;
	346	struct fio_file *f;
	347	ssize_t ret;
	348
	349	if (!sd->queued)
	350	return 0;
	351
	352	io_u_mark_submit(td, sd->queued);
	353	f = sd->last_file;
	354
	355	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
	356	int err = -errno;
	357
	358	td_verror(td, errno, "lseek");
	359	return err;
	360	}
	361
	362	if (sd->last_ddir == DDIR_READ)
	363	ret = readv(f->fd, sd->iovecs, sd->queued);
	364	else
	365	ret = writev(f->fd, sd->iovecs, sd->queued);
	366
	367	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
	368	sd->events = sd->queued;
	369	sd->queued = 0;
	370	return fio_vsyncio_end(td, ret);
	371	}
	372
	373	static int fio_vsyncio_init(struct thread_data *td)
	374	{
	375	struct syncio_data *sd;
	376
	377	sd = malloc(sizeof(*sd));
	378	memset(sd, 0, sizeof(*sd));
	379	sd->last_offset = -1ULL;
	380	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	381	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
	382	init_rand(&sd->rand_state, 0);
	383
	384	td->io_ops_data = sd;
	385	return 0;
	386	}
	387
	388	static void fio_vsyncio_cleanup(struct thread_data *td)
	389	{
	390	struct syncio_data *sd = td->io_ops_data;
	391
	392	if (sd) {
	393	free(sd->iovecs);
	394	free(sd->io_us);
	395	free(sd);
	396	}
	397	}
	398
	399	static struct ioengine_ops ioengine_rw = {
	400	.name = "sync",
	401	.version = FIO_IOOPS_VERSION,
	402	.prep = fio_syncio_prep,
	403	.queue = fio_syncio_queue,
	404	.open_file = generic_open_file,
	405	.close_file = generic_close_file,
	406	.get_file_size = generic_get_file_size,
	407	.flags = FIO_SYNCIO,
	408	};
	409
	410	static struct ioengine_ops ioengine_prw = {
	411	.name = "psync",
	412	.version = FIO_IOOPS_VERSION,
	413	.queue = fio_psyncio_queue,
	414	.open_file = generic_open_file,
	415	.close_file = generic_close_file,
	416	.get_file_size = generic_get_file_size,
	417	.flags = FIO_SYNCIO,
	418	};
	419
	420	static struct ioengine_ops ioengine_vrw = {
	421	.name = "vsync",
	422	.version = FIO_IOOPS_VERSION,
	423	.init = fio_vsyncio_init,
	424	.cleanup = fio_vsyncio_cleanup,
	425	.queue = fio_vsyncio_queue,
	426	.commit = fio_vsyncio_commit,
	427	.event = fio_vsyncio_event,
	428	.getevents = fio_vsyncio_getevents,
	429	.open_file = generic_open_file,
	430	.close_file = generic_close_file,
	431	.get_file_size = generic_get_file_size,
	432	.flags = FIO_SYNCIO,
	433	};
	434
	435	#ifdef CONFIG_PWRITEV
	436	static struct ioengine_ops ioengine_pvrw = {
	437	.name = "pvsync",
	438	.version = FIO_IOOPS_VERSION,
	439	.init = fio_vsyncio_init,
	440	.cleanup = fio_vsyncio_cleanup,
	441	.queue = fio_pvsyncio_queue,
	442	.open_file = generic_open_file,
	443	.close_file = generic_close_file,
	444	.get_file_size = generic_get_file_size,
	445	.flags = FIO_SYNCIO,
	446	};
	447	#endif
	448
	449	#ifdef FIO_HAVE_PWRITEV2
	450	static struct ioengine_ops ioengine_pvrw2 = {
	451	.name = "pvsync2",
	452	.version = FIO_IOOPS_VERSION,
	453	.init = fio_vsyncio_init,
	454	.cleanup = fio_vsyncio_cleanup,
	455	.queue = fio_pvsyncio2_queue,
	456	.open_file = generic_open_file,
	457	.close_file = generic_close_file,
	458	.get_file_size = generic_get_file_size,
	459	.flags = FIO_SYNCIO,
	460	.options = options,
	461	.option_struct_size = sizeof(struct psyncv2_options),
	462	};
	463	#endif
	464
	465	static void fio_init fio_syncio_register(void)
	466	{
	467	register_ioengine(&ioengine_rw);
	468	register_ioengine(&ioengine_prw);
	469	register_ioengine(&ioengine_vrw);
	470	#ifdef CONFIG_PWRITEV
	471	register_ioengine(&ioengine_pvrw);
	472	#endif
	473	#ifdef FIO_HAVE_PWRITEV2
	474	register_ioengine(&ioengine_pvrw2);
	475	#endif
	476	}
	477
	478	static void fio_exit fio_syncio_unregister(void)
	479	{
	480	unregister_ioengine(&ioengine_rw);
	481	unregister_ioengine(&ioengine_prw);
	482	unregister_ioengine(&ioengine_vrw);
	483	#ifdef CONFIG_PWRITEV
	484	unregister_ioengine(&ioengine_pvrw);
	485	#endif
	486	#ifdef FIO_HAVE_PWRITEV2
	487	unregister_ioengine(&ioengine_pvrw2);
	488	#endif
	489	}