[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;
	unsigned int uncached;
	unsigned int nowait;
};

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	= "uncached",
		.lname	= "Uncached",
		.type	= FIO_OPT_INT,
		.off1	= offsetof(struct psyncv2_options, uncached),
		.help	= "Use RWF_UNCACHED for buffered read/writes",
		.category = FIO_OPT_C_ENGINE,
		.group	= FIO_OPT_G_INVALID,
	},
	{
		.name	= "nowait",
		.lname	= "RWF_NOWAIT",
		.type	= FIO_OPT_BOOL,
		.off1	= offsetof(struct psyncv2_options, nowait),
		.help	= "Set RWF_NOWAIT 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;
	if (!td->o.odirect && o->uncached)
		flags |= RWF_UNCACHED;
	if (o->nowait)
		flags |= RWF_NOWAIT;

	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 = calloc(1, 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	unsigned int uncached;
	43	unsigned int nowait;
	44	};
	45
	46	static struct fio_option options[] = {
	47	{
	48	.name = "hipri",
	49	.lname = "RWF_HIPRI",
	50	.type = FIO_OPT_STR_SET,
	51	.off1 = offsetof(struct psyncv2_options, hipri),
	52	.help = "Set RWF_HIPRI for pwritev2/preadv2",
	53	.category = FIO_OPT_C_ENGINE,
	54	.group = FIO_OPT_G_INVALID,
	55	},
	56	{
	57	.name = "hipri_percentage",
	58	.lname = "RWF_HIPRI_PERCENTAGE",
	59	.type = FIO_OPT_INT,
	60	.off1 = offsetof(struct psyncv2_options, hipri_percentage),
	61	.minval = 0,
	62	.maxval = 100,
	63	.def = "100",
	64	.help = "Probabilistically set RWF_HIPRI for pwritev2/preadv2",
	65	.category = FIO_OPT_C_ENGINE,
	66	.group = FIO_OPT_G_INVALID,
	67	},
	68	{
	69	.name = "uncached",
	70	.lname = "Uncached",
	71	.type = FIO_OPT_INT,
	72	.off1 = offsetof(struct psyncv2_options, uncached),
	73	.help = "Use RWF_UNCACHED for buffered read/writes",
	74	.category = FIO_OPT_C_ENGINE,
	75	.group = FIO_OPT_G_INVALID,
	76	},
	77	{
	78	.name = "nowait",
	79	.lname = "RWF_NOWAIT",
	80	.type = FIO_OPT_BOOL,
	81	.off1 = offsetof(struct psyncv2_options, nowait),
	82	.help = "Set RWF_NOWAIT for pwritev2/preadv2",
	83	.category = FIO_OPT_C_ENGINE,
	84	.group = FIO_OPT_G_INVALID,
	85	},
	86	{
	87	.name = NULL,
	88	},
	89	};
	90	#endif
	91
	92	static int fio_syncio_prep(struct thread_data td, struct io_u io_u)
	93	{
	94	struct fio_file *f = io_u->file;
	95
	96	if (!ddir_rw(io_u->ddir))
	97	return 0;
	98
	99	if (LAST_POS(f) != -1ULL && LAST_POS(f) == io_u->offset)
	100	return 0;
	101
	102	if (lseek(f->fd, io_u->offset, SEEK_SET) == -1) {
	103	td_verror(td, errno, "lseek");
	104	return 1;
	105	}
	106
	107	return 0;
	108	}
	109
	110	static int fio_io_end(struct thread_data td, struct io_u io_u, int ret)
	111	{
	112	if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
	113	LAST_POS(io_u->file) = io_u->offset + ret;
	114
	115	if (ret != (int) io_u->xfer_buflen) {
	116	if (ret >= 0) {
	117	io_u->resid = io_u->xfer_buflen - ret;
	118	io_u->error = 0;
	119	return FIO_Q_COMPLETED;
	120	} else
	121	io_u->error = errno;
	122	}
	123
	124	if (io_u->error) {
	125	io_u_log_error(td, io_u);
	126	td_verror(td, io_u->error, "xfer");
	127	}
	128
	129	return FIO_Q_COMPLETED;
	130	}
	131
	132	#ifdef CONFIG_PWRITEV
	133	static enum fio_q_status fio_pvsyncio_queue(struct thread_data *td,
	134	struct io_u *io_u)
	135	{
	136	struct syncio_data *sd = td->io_ops_data;
	137	struct iovec *iov = &sd->iovecs[0];
	138	struct fio_file *f = io_u->file;
	139	int ret;
	140
	141	fio_ro_check(td, io_u);
	142
	143	iov->iov_base = io_u->xfer_buf;
	144	iov->iov_len = io_u->xfer_buflen;
	145
	146	if (io_u->ddir == DDIR_READ)
	147	ret = preadv(f->fd, iov, 1, io_u->offset);
	148	else if (io_u->ddir == DDIR_WRITE)
	149	ret = pwritev(f->fd, iov, 1, io_u->offset);
	150	else if (io_u->ddir == DDIR_TRIM) {
	151	do_io_u_trim(td, io_u);
	152	return FIO_Q_COMPLETED;
	153	} else
	154	ret = do_io_u_sync(td, io_u);
	155
	156	return fio_io_end(td, io_u, ret);
	157	}
	158	#endif
	159
	160	#ifdef FIO_HAVE_PWRITEV2
	161	static enum fio_q_status fio_pvsyncio2_queue(struct thread_data *td,
	162	struct io_u *io_u)
	163	{
	164	struct syncio_data *sd = td->io_ops_data;
	165	struct psyncv2_options *o = td->eo;
	166	struct iovec *iov = &sd->iovecs[0];
	167	struct fio_file *f = io_u->file;
	168	int ret, flags = 0;
	169
	170	fio_ro_check(td, io_u);
	171
	172	if (o->hipri &&
	173	(rand_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
	174	flags \|= RWF_HIPRI;
	175	if (!td->o.odirect && o->uncached)
	176	flags \|= RWF_UNCACHED;
	177	if (o->nowait)
	178	flags \|= RWF_NOWAIT;
	179
	180	iov->iov_base = io_u->xfer_buf;
	181	iov->iov_len = io_u->xfer_buflen;
	182
	183	if (io_u->ddir == DDIR_READ)
	184	ret = preadv2(f->fd, iov, 1, io_u->offset, flags);
	185	else if (io_u->ddir == DDIR_WRITE)
	186	ret = pwritev2(f->fd, iov, 1, io_u->offset, flags);
	187	else if (io_u->ddir == DDIR_TRIM) {
	188	do_io_u_trim(td, io_u);
	189	return FIO_Q_COMPLETED;
	190	} else
	191	ret = do_io_u_sync(td, io_u);
	192
	193	return fio_io_end(td, io_u, ret);
	194	}
	195	#endif
	196
	197	static enum fio_q_status fio_psyncio_queue(struct thread_data *td,
	198	struct io_u *io_u)
	199	{
	200	struct fio_file *f = io_u->file;
	201	int ret;
	202
	203	fio_ro_check(td, io_u);
	204
	205	if (io_u->ddir == DDIR_READ)
	206	ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	207	else if (io_u->ddir == DDIR_WRITE)
	208	ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	209	else if (io_u->ddir == DDIR_TRIM) {
	210	do_io_u_trim(td, io_u);
	211	return FIO_Q_COMPLETED;
	212	} else
	213	ret = do_io_u_sync(td, io_u);
	214
	215	return fio_io_end(td, io_u, ret);
	216	}
	217
	218	static enum fio_q_status fio_syncio_queue(struct thread_data *td,
	219	struct io_u *io_u)
	220	{
	221	struct fio_file *f = io_u->file;
	222	int ret;
	223
	224	fio_ro_check(td, io_u);
	225
	226	if (io_u->ddir == DDIR_READ)
	227	ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	228	else if (io_u->ddir == DDIR_WRITE)
	229	ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	230	else if (io_u->ddir == DDIR_TRIM) {
	231	do_io_u_trim(td, io_u);
	232	return FIO_Q_COMPLETED;
	233	} else
	234	ret = do_io_u_sync(td, io_u);
	235
	236	return fio_io_end(td, io_u, ret);
	237	}
	238
	239	static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
	240	unsigned int max,
	241	const struct timespec fio_unused *t)
	242	{
	243	struct syncio_data *sd = td->io_ops_data;
	244	int ret;
	245
	246	if (min) {
	247	ret = sd->events;
	248	sd->events = 0;
	249	} else
	250	ret = 0;
	251
	252	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	253	return ret;
	254	}
	255
	256	static struct io_u fio_vsyncio_event(struct thread_data td, int event)
	257	{
	258	struct syncio_data *sd = td->io_ops_data;
	259
	260	return sd->io_us[event];
	261	}
	262
	263	static int fio_vsyncio_append(struct thread_data td, struct io_u io_u)
	264	{
	265	struct syncio_data *sd = td->io_ops_data;
	266
	267	if (ddir_sync(io_u->ddir))
	268	return 0;
	269
	270	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	271	io_u->ddir == sd->last_ddir)
	272	return 1;
	273
	274	return 0;
	275	}
	276
	277	static void fio_vsyncio_set_iov(struct syncio_data sd, struct io_u io_u,
	278	int idx)
	279	{
	280	sd->io_us[idx] = io_u;
	281	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	282	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
	283	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	284	sd->last_file = io_u->file;
	285	sd->last_ddir = io_u->ddir;
	286	sd->queued_bytes += io_u->xfer_buflen;
	287	sd->queued++;
	288	}
	289
	290	static enum fio_q_status fio_vsyncio_queue(struct thread_data *td,
	291	struct io_u *io_u)
	292	{
	293	struct syncio_data *sd = td->io_ops_data;
	294
	295	fio_ro_check(td, io_u);
	296
	297	if (!fio_vsyncio_append(td, io_u)) {
	298	dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
	299	/*
	300	* If we can't append and have stuff queued, tell fio to
	301	* commit those first and then retry this io
	302	*/
	303	if (sd->queued)
	304	return FIO_Q_BUSY;
	305	if (ddir_sync(io_u->ddir)) {
	306	int ret = do_io_u_sync(td, io_u);
	307
	308	return fio_io_end(td, io_u, ret);
	309	}
	310
	311	sd->queued = 0;
	312	sd->queued_bytes = 0;
	313	fio_vsyncio_set_iov(sd, io_u, 0);
	314	} else {
	315	if (sd->queued == td->o.iodepth) {
	316	dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
	317	return FIO_Q_BUSY;
	318	}
	319
	320	dprint(FD_IO, "vsyncio_queue: append\n");
	321	fio_vsyncio_set_iov(sd, io_u, sd->queued);
	322	}
	323
	324	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	325	return FIO_Q_QUEUED;
	326	}
	327
	328	/*
	329	* Check that we transferred all bytes, or saw an error, etc
	330	*/
	331	static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
	332	{
	333	struct syncio_data *sd = td->io_ops_data;
	334	struct io_u *io_u;
	335	unsigned int i;
	336	int err;
	337
	338	/*
	339	* transferred everything, perfect
	340	*/
	341	if (bytes == sd->queued_bytes)
	342	return 0;
	343
	344	err = errno;
	345	for (i = 0; i < sd->queued; i++) {
	346	io_u = sd->io_us[i];
	347
	348	if (bytes == -1) {
	349	io_u->error = err;
	350	} else {
	351	unsigned int this_io;
	352
	353	this_io = bytes;
	354	if (this_io > io_u->xfer_buflen)
	355	this_io = io_u->xfer_buflen;
	356
	357	io_u->resid = io_u->xfer_buflen - this_io;
	358	io_u->error = 0;
	359	bytes -= this_io;
	360	}
	361	}
	362
	363	if (bytes == -1) {
	364	td_verror(td, err, "xfer vsync");
	365	return -err;
	366	}
	367
	368	return 0;
	369	}
	370
	371	static int fio_vsyncio_commit(struct thread_data *td)
	372	{
	373	struct syncio_data *sd = td->io_ops_data;
	374	struct fio_file *f;
	375	ssize_t ret;
	376
	377	if (!sd->queued)
	378	return 0;
	379
	380	io_u_mark_submit(td, sd->queued);
	381	f = sd->last_file;
	382
	383	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
	384	int err = -errno;
	385
	386	td_verror(td, errno, "lseek");
	387	return err;
	388	}
	389
	390	if (sd->last_ddir == DDIR_READ)
	391	ret = readv(f->fd, sd->iovecs, sd->queued);
	392	else
	393	ret = writev(f->fd, sd->iovecs, sd->queued);
	394
	395	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
	396	sd->events = sd->queued;
	397	sd->queued = 0;
	398	return fio_vsyncio_end(td, ret);
	399	}
	400
	401	static int fio_vsyncio_init(struct thread_data *td)
	402	{
	403	struct syncio_data *sd;
	404
	405	sd = calloc(1, sizeof(*sd));
	406	sd->last_offset = -1ULL;
	407	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	408	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
	409	init_rand(&sd->rand_state, 0);
	410
	411	td->io_ops_data = sd;
	412	return 0;
	413	}
	414
	415	static void fio_vsyncio_cleanup(struct thread_data *td)
	416	{
	417	struct syncio_data *sd = td->io_ops_data;
	418
	419	if (sd) {
	420	free(sd->iovecs);
	421	free(sd->io_us);
	422	free(sd);
	423	}
	424	}
	425
	426	static struct ioengine_ops ioengine_rw = {
	427	.name = "sync",
	428	.version = FIO_IOOPS_VERSION,
	429	.prep = fio_syncio_prep,
	430	.queue = fio_syncio_queue,
	431	.open_file = generic_open_file,
	432	.close_file = generic_close_file,
	433	.get_file_size = generic_get_file_size,
	434	.flags = FIO_SYNCIO,
	435	};
	436
	437	static struct ioengine_ops ioengine_prw = {
	438	.name = "psync",
	439	.version = FIO_IOOPS_VERSION,
	440	.queue = fio_psyncio_queue,
	441	.open_file = generic_open_file,
	442	.close_file = generic_close_file,
	443	.get_file_size = generic_get_file_size,
	444	.flags = FIO_SYNCIO,
	445	};
	446
	447	static struct ioengine_ops ioengine_vrw = {
	448	.name = "vsync",
	449	.version = FIO_IOOPS_VERSION,
	450	.init = fio_vsyncio_init,
	451	.cleanup = fio_vsyncio_cleanup,
	452	.queue = fio_vsyncio_queue,
	453	.commit = fio_vsyncio_commit,
	454	.event = fio_vsyncio_event,
	455	.getevents = fio_vsyncio_getevents,
	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	};
	461
	462	#ifdef CONFIG_PWRITEV
	463	static struct ioengine_ops ioengine_pvrw = {
	464	.name = "pvsync",
	465	.version = FIO_IOOPS_VERSION,
	466	.init = fio_vsyncio_init,
	467	.cleanup = fio_vsyncio_cleanup,
	468	.queue = fio_pvsyncio_queue,
	469	.open_file = generic_open_file,
	470	.close_file = generic_close_file,
	471	.get_file_size = generic_get_file_size,
	472	.flags = FIO_SYNCIO,
	473	};
	474	#endif
	475
	476	#ifdef FIO_HAVE_PWRITEV2
	477	static struct ioengine_ops ioengine_pvrw2 = {
	478	.name = "pvsync2",
	479	.version = FIO_IOOPS_VERSION,
	480	.init = fio_vsyncio_init,
	481	.cleanup = fio_vsyncio_cleanup,
	482	.queue = fio_pvsyncio2_queue,
	483	.open_file = generic_open_file,
	484	.close_file = generic_close_file,
	485	.get_file_size = generic_get_file_size,
	486	.flags = FIO_SYNCIO,
	487	.options = options,
	488	.option_struct_size = sizeof(struct psyncv2_options),
	489	};
	490	#endif
	491
	492	static void fio_init fio_syncio_register(void)
	493	{
	494	register_ioengine(&ioengine_rw);
	495	register_ioengine(&ioengine_prw);
	496	register_ioengine(&ioengine_vrw);
	497	#ifdef CONFIG_PWRITEV
	498	register_ioengine(&ioengine_pvrw);
	499	#endif
	500	#ifdef FIO_HAVE_PWRITEV2
	501	register_ioengine(&ioengine_pvrw2);
	502	#endif
	503	}
	504
	505	static void fio_exit fio_syncio_unregister(void)
	506	{
	507	unregister_ioengine(&ioengine_rw);
	508	unregister_ioengine(&ioengine_prw);
	509	unregister_ioengine(&ioengine_vrw);
	510	#ifdef CONFIG_PWRITEV
	511	unregister_ioengine(&ioengine_pvrw);
	512	#endif
	513	#ifdef FIO_HAVE_PWRITEV2
	514	unregister_ioengine(&ioengine_pvrw2);
	515	#endif
	516	}