[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 <assert.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
2866c82d	1	/*
a31041ea	2	* sync/psync engine
da751ca9 JA	3	*
da751ca9 JA	4	* IO engine that does regular read(2)/write(2) with lseek(2) to transfer
a31041ea	5	* data and IO engine that does regular pread(2)/pwrite(2) to transfer data.
2866c82d JA	6	*
	7	*/
	8	#include <stdio.h>
	9	#include <stdlib.h>
	10	#include <unistd.h>
5921e80c	11	#include <sys/uio.h>
2866c82d JA	12	#include <errno.h>
2866c82d JA	13	#include <assert.h>
5f350952 JA	14
5f350952 JA	15	#include "../fio.h"
2cafffbe	16	#include "../optgroup.h"
a0679ce5	17	#include "../lib/rand.h"
2866c82d	18
ef5f5a3a JA	19	/*
	20	* Sync engine uses engine_data to store last offset
	21	*/
710bf9c5	22	#define LAST_POS(f) ((f)->engine_pos)
ef5f5a3a	23
1d2af02a JA	24	struct syncio_data {
	25	struct iovec *iovecs;
	26	struct io_u **io_us;
	27	unsigned int queued;
e51cf72c	28	unsigned int events;
1d2af02a JA	29	unsigned long queued_bytes;
	30
	31	unsigned long long last_offset;
	32	struct fio_file *last_file;
	33	enum fio_ddir last_ddir;
a0679ce5 SB	34
a0679ce5 SB	35	struct frand_state rand_state;
1d2af02a JA	36	};
1d2af02a JA	37
6562685f	38	#ifdef FIO_HAVE_PWRITEV2
2cafffbe JA	39	struct psyncv2_options {
	40	void *pad;
	41	unsigned int hipri;
a0679ce5	42	unsigned int hipri_percentage;
2cafffbe JA	43	};
	44
	45	static struct fio_option options[] = {
	46	{
	47	.name = "hipri",
	48	.lname = "RWF_HIPRI",
	49	.type = FIO_OPT_STR_SET,
	50	.off1 = offsetof(struct psyncv2_options, hipri),
	51	.help = "Set RWF_HIPRI for pwritev2/preadv2",
	52	.category = FIO_OPT_C_ENGINE,
	53	.group = FIO_OPT_G_INVALID,
	54	},
a0679ce5 SB	55	{
	56	.name = "hipri_percentage",
	57	.lname = "RWF_HIPRI_PERCENTAGE",
	58	.type = FIO_OPT_INT,
	59	.off1 = offsetof(struct psyncv2_options, hipri_percentage),
	60	.minval = 0,
	61	.maxval = 100,
	62	.def = "100",
	63	.help = "Probabilistically set RWF_HIPRI for pwritev2/preadv2",
	64	.category = FIO_OPT_C_ENGINE,
	65	.group = FIO_OPT_G_INVALID,
	66	},
2cafffbe JA	67	{
	68	.name = NULL,
	69	},
	70	};
	71	#endif
	72
2866c82d JA	73	static int fio_syncio_prep(struct thread_data td, struct io_u io_u)
2866c82d JA	74	{
53cdc686 JA	75	struct fio_file *f = io_u->file;
53cdc686 JA	76
ff58fced	77	if (!ddir_rw(io_u->ddir))
87dc1ab1 JA	78	return 0;
87dc1ab1 JA	79
ef5f5a3a	80	if (LAST_POS(f) != -1ULL && LAST_POS(f) == io_u->offset)
e943b878 JA	81	return 0;
e943b878 JA	82
53cdc686	83	if (lseek(f->fd, io_u->offset, SEEK_SET) == -1) {
e1161c32	84	td_verror(td, errno, "lseek");
2866c82d JA	85	return 1;
	86	}
	87
	88	return 0;
	89	}
	90
2bd3eabc	91	static int fio_io_end(struct thread_data td, struct io_u io_u, int ret)
2866c82d	92	{
ff58fced	93	if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
ef5f5a3a	94	LAST_POS(io_u->file) = io_u->offset + ret;
e943b878	95
cec6b55d	96	if (ret != (int) io_u->xfer_buflen) {
22819ec2	97	if (ret >= 0) {
cec6b55d JA	98	io_u->resid = io_u->xfer_buflen - ret;
cec6b55d JA	99	io_u->error = 0;
36167d82	100	return FIO_Q_COMPLETED;
2866c82d JA	101	} else
	102	io_u->error = errno;
	103	}
	104
bfd197cb JA	105	if (io_u->error) {
bfd197cb JA	106	io_u_log_error(td, io_u);
e1161c32	107	td_verror(td, io_u->error, "xfer");
bfd197cb	108	}
2866c82d	109
36167d82	110	return FIO_Q_COMPLETED;
2866c82d JA	111	}
2866c82d JA	112
07fc0acd JA	113	#ifdef CONFIG_PWRITEV
	114	static int fio_pvsyncio_queue(struct thread_data td, struct io_u io_u)
	115	{
565e784d	116	struct syncio_data *sd = td->io_ops_data;
07fc0acd JA	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
6562685f	140	#ifdef FIO_HAVE_PWRITEV2
2cafffbe JA	141	static int fio_pvsyncio2_queue(struct thread_data td, struct io_u io_u)
2cafffbe JA	142	{
565e784d	143	struct syncio_data *sd = td->io_ops_data;
2cafffbe JA	144	struct psyncv2_options *o = td->eo;
	145	struct iovec *iov = &sd->iovecs[0];
	146	struct fio_file *f = io_u->file;
	147	int ret, flags = 0;
	148
	149	fio_ro_check(td, io_u);
	150
a0679ce5 SB	151	if (o->hipri &&
a0679ce5 SB	152	(rand32_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
2cafffbe JA	153	flags \|= RWF_HIPRI;
	154
	155	iov->iov_base = io_u->xfer_buf;
	156	iov->iov_len = io_u->xfer_buflen;
	157
	158	if (io_u->ddir == DDIR_READ)
	159	ret = preadv2(f->fd, iov, 1, io_u->offset, flags);
	160	else if (io_u->ddir == DDIR_WRITE)
	161	ret = pwritev2(f->fd, iov, 1, io_u->offset, flags);
	162	else if (io_u->ddir == DDIR_TRIM) {
	163	do_io_u_trim(td, io_u);
	164	return FIO_Q_COMPLETED;
	165	} else
	166	ret = do_io_u_sync(td, io_u);
	167
	168	return fio_io_end(td, io_u, ret);
	169	}
	170	#endif
	171
	172
2bd3eabc	173	static int fio_psyncio_queue(struct thread_data td, struct io_u io_u)
a31041ea	174	{
2bd3eabc JA	175	struct fio_file *f = io_u->file;
	176	int ret;
	177
	178	fio_ro_check(td, io_u);
	179
	180	if (io_u->ddir == DDIR_READ)
	181	ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	182	else if (io_u->ddir == DDIR_WRITE)
	183	ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
6eaf09d6 SL	184	else if (io_u->ddir == DDIR_TRIM) {
	185	do_io_u_trim(td, io_u);
	186	return FIO_Q_COMPLETED;
	187	} else
0a28ecda	188	ret = do_io_u_sync(td, io_u);
2bd3eabc JA	189
	190	return fio_io_end(td, io_u, ret);
	191	}
	192
	193	static int fio_syncio_queue(struct thread_data td, struct io_u io_u)
	194	{
	195	struct fio_file *f = io_u->file;
	196	int ret;
	197
	198	fio_ro_check(td, io_u);
	199
	200	if (io_u->ddir == DDIR_READ)
	201	ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	202	else if (io_u->ddir == DDIR_WRITE)
	203	ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
6eaf09d6 SL	204	else if (io_u->ddir == DDIR_TRIM) {
	205	do_io_u_trim(td, io_u);
	206	return FIO_Q_COMPLETED;
	207	} else
0a28ecda	208	ret = do_io_u_sync(td, io_u);
2bd3eabc JA	209
2bd3eabc JA	210	return fio_io_end(td, io_u, ret);
a31041ea	211	}
a31041ea	212
1d2af02a JA	213	static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
1d2af02a JA	214	unsigned int max,
1f440ece	215	const struct timespec fio_unused *t)
1d2af02a	216	{
565e784d	217	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	218	int ret;
	219
	220	if (min) {
e51cf72c JA	221	ret = sd->events;
e51cf72c JA	222	sd->events = 0;
1d2af02a JA	223	} else
	224	ret = 0;
	225
	226	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	227	return ret;
	228	}
	229
	230	static struct io_u fio_vsyncio_event(struct thread_data td, int event)
	231	{
565e784d	232	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	233
	234	return sd->io_us[event];
	235	}
	236
	237	static int fio_vsyncio_append(struct thread_data td, struct io_u io_u)
	238	{
565e784d	239	struct syncio_data *sd = td->io_ops_data;
1d2af02a	240
5f9099ea	241	if (ddir_sync(io_u->ddir))
1d2af02a JA	242	return 0;
	243
	244	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	245	io_u->ddir == sd->last_ddir)
	246	return 1;
	247
	248	return 0;
	249	}
	250
	251	static void fio_vsyncio_set_iov(struct syncio_data sd, struct io_u io_u,
2b13e716	252	int idx)
1d2af02a	253	{
2b13e716 JA	254	sd->io_us[idx] = io_u;
	255	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	256	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
1d2af02a JA	257	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	258	sd->last_file = io_u->file;
	259	sd->last_ddir = io_u->ddir;
	260	sd->queued_bytes += io_u->xfer_buflen;
	261	sd->queued++;
	262	}
	263
	264	static int fio_vsyncio_queue(struct thread_data td, struct io_u io_u)
	265	{
565e784d	266	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	267
	268	fio_ro_check(td, io_u);
	269
	270	if (!fio_vsyncio_append(td, io_u)) {
	271	dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
	272	/*
	273	* If we can't append and have stuff queued, tell fio to
	274	* commit those first and then retry this io
	275	*/
	276	if (sd->queued)
	277	return FIO_Q_BUSY;
0a28ecda JA	278	if (ddir_sync(io_u->ddir)) {
	279	int ret = do_io_u_sync(td, io_u);
	280
	281	return fio_io_end(td, io_u, ret);
	282	}
cc9159c3	283
1d2af02a JA	284	sd->queued = 0;
	285	sd->queued_bytes = 0;
	286	fio_vsyncio_set_iov(sd, io_u, 0);
	287	} else {
	288	if (sd->queued == td->o.iodepth) {
	289	dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
	290	return FIO_Q_BUSY;
	291	}
	292
	293	dprint(FD_IO, "vsyncio_queue: append\n");
	294	fio_vsyncio_set_iov(sd, io_u, sd->queued);
	295	}
	296
	297	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	298	return FIO_Q_QUEUED;
	299	}
	300
	301	/*
	302	* Check that we transferred all bytes, or saw an error, etc
	303	*/
	304	static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
	305	{
565e784d	306	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	307	struct io_u *io_u;
	308	unsigned int i;
	309	int err;
	310
	311	/*
	312	* transferred everything, perfect
	313	*/
	314	if (bytes == sd->queued_bytes)
	315	return 0;
	316
	317	err = errno;
	318	for (i = 0; i < sd->queued; i++) {
	319	io_u = sd->io_us[i];
	320
	321	if (bytes == -1) {
	322	io_u->error = err;
	323	} else {
	324	unsigned int this_io;
	325
	326	this_io = bytes;
	327	if (this_io > io_u->xfer_buflen)
	328	this_io = io_u->xfer_buflen;
	329
	330	io_u->resid = io_u->xfer_buflen - this_io;
	331	io_u->error = 0;
	332	bytes -= this_io;
	333	}
	334	}
	335
	336	if (bytes == -1) {
	337	td_verror(td, err, "xfer vsync");
	338	return -err;
	339	}
	340
	341	return 0;
	342	}
	343
	344	static int fio_vsyncio_commit(struct thread_data *td)
	345	{
565e784d	346	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	347	struct fio_file *f;
	348	ssize_t ret;
	349
	350	if (!sd->queued)
	351	return 0;
	352
838bc709	353	io_u_mark_submit(td, sd->queued);
1d2af02a JA	354	f = sd->last_file;
	355
	356	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
	357	int err = -errno;
	358
	359	td_verror(td, errno, "lseek");
	360	return err;
	361	}
	362
	363	if (sd->last_ddir == DDIR_READ)
	364	ret = readv(f->fd, sd->iovecs, sd->queued);
	365	else
	366	ret = writev(f->fd, sd->iovecs, sd->queued);
	367
	368	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
e51cf72c JA	369	sd->events = sd->queued;
e51cf72c JA	370	sd->queued = 0;
1d2af02a JA	371	return fio_vsyncio_end(td, ret);
	372	}
	373
	374	static int fio_vsyncio_init(struct thread_data *td)
	375	{
	376	struct syncio_data *sd;
	377
	378	sd = malloc(sizeof(*sd));
	379	memset(sd, 0, sizeof(*sd));
	380	sd->last_offset = -1ULL;
	381	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	382	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
a0679ce5	383	init_rand(&sd->rand_state, 0);
1d2af02a	384
565e784d	385	td->io_ops_data = sd;
1d2af02a JA	386	return 0;
	387	}
	388
	389	static void fio_vsyncio_cleanup(struct thread_data *td)
	390	{
565e784d	391	struct syncio_data *sd = td->io_ops_data;
1d2af02a	392
02a3d83f TK	393	if (sd) {
	394	free(sd->iovecs);
	395	free(sd->io_us);
	396	free(sd);
	397	}
1d2af02a JA	398	}
1d2af02a JA	399
a31041ea	400	static struct ioengine_ops ioengine_rw = {
2866c82d JA	401	.name = "sync",
2866c82d JA	402	.version = FIO_IOOPS_VERSION,
2866c82d JA	403	.prep = fio_syncio_prep,
2866c82d JA	404	.queue = fio_syncio_queue,
b5af8293 JA	405	.open_file = generic_open_file,
b5af8293 JA	406	.close_file = generic_close_file,
df9c26b1	407	.get_file_size = generic_get_file_size,
2866c82d JA	408	.flags = FIO_SYNCIO,
2866c82d JA	409	};
5f350952	410
a31041ea	411	static struct ioengine_ops ioengine_prw = {
	412	.name = "psync",
	413	.version = FIO_IOOPS_VERSION,
2bd3eabc	414	.queue = fio_psyncio_queue,
a31041ea	415	.open_file = generic_open_file,
a31041ea	416	.close_file = generic_close_file,
df9c26b1	417	.get_file_size = generic_get_file_size,
a31041ea	418	.flags = FIO_SYNCIO,
	419	};
	420
1d2af02a JA	421	static struct ioengine_ops ioengine_vrw = {
	422	.name = "vsync",
	423	.version = FIO_IOOPS_VERSION,
	424	.init = fio_vsyncio_init,
	425	.cleanup = fio_vsyncio_cleanup,
	426	.queue = fio_vsyncio_queue,
	427	.commit = fio_vsyncio_commit,
	428	.event = fio_vsyncio_event,
	429	.getevents = fio_vsyncio_getevents,
	430	.open_file = generic_open_file,
	431	.close_file = generic_close_file,
df9c26b1	432	.get_file_size = generic_get_file_size,
1d2af02a JA	433	.flags = FIO_SYNCIO,
	434	};
	435
07fc0acd JA	436	#ifdef CONFIG_PWRITEV
	437	static struct ioengine_ops ioengine_pvrw = {
	438	.name = "pvsync",
	439	.version = FIO_IOOPS_VERSION,
	440	.init = fio_vsyncio_init,
	441	.cleanup = fio_vsyncio_cleanup,
	442	.queue = fio_pvsyncio_queue,
	443	.open_file = generic_open_file,
	444	.close_file = generic_close_file,
	445	.get_file_size = generic_get_file_size,
	446	.flags = FIO_SYNCIO,
	447	};
	448	#endif
	449
6562685f	450	#ifdef FIO_HAVE_PWRITEV2
2cafffbe JA	451	static struct ioengine_ops ioengine_pvrw2 = {
	452	.name = "pvsync2",
	453	.version = FIO_IOOPS_VERSION,
	454	.init = fio_vsyncio_init,
	455	.cleanup = fio_vsyncio_cleanup,
	456	.queue = fio_pvsyncio2_queue,
	457	.open_file = generic_open_file,
	458	.close_file = generic_close_file,
	459	.get_file_size = generic_get_file_size,
	460	.flags = FIO_SYNCIO,
	461	.options = options,
	462	.option_struct_size = sizeof(struct psyncv2_options),
	463	};
	464	#endif
	465
5f350952 JA	466	static void fio_init fio_syncio_register(void)
5f350952 JA	467	{
a31041ea	468	register_ioengine(&ioengine_rw);
a31041ea	469	register_ioengine(&ioengine_prw);
1d2af02a	470	register_ioengine(&ioengine_vrw);
9a0ced5a	471	#ifdef CONFIG_PWRITEV
07fc0acd	472	register_ioengine(&ioengine_pvrw);
9a0ced5a	473	#endif
6562685f	474	#ifdef FIO_HAVE_PWRITEV2
526c403d JD	475	register_ioengine(&ioengine_pvrw2);
526c403d JD	476	#endif
5f350952 JA	477	}
	478
	479	static void fio_exit fio_syncio_unregister(void)
	480	{
a31041ea	481	unregister_ioengine(&ioengine_rw);
a31041ea	482	unregister_ioengine(&ioengine_prw);
1d2af02a	483	unregister_ioengine(&ioengine_vrw);
9a0ced5a	484	#ifdef CONFIG_PWRITEV
07fc0acd	485	unregister_ioengine(&ioengine_pvrw);
9a0ced5a	486	#endif
6562685f	487	#ifdef FIO_HAVE_PWRITEV2
526c403d JD	488	unregister_ioengine(&ioengine_pvrw2);
526c403d JD	489	#endif
5f350952	490	}