[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 &&
	    (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 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
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	12	#include <errno.h>
5f350952 JA	13
5f350952 JA	14	#include "../fio.h"
2cafffbe	15	#include "../optgroup.h"
a0679ce5	16	#include "../lib/rand.h"
2866c82d	17
ef5f5a3a JA	18	/*
	19	* Sync engine uses engine_data to store last offset
	20	*/
710bf9c5	21	#define LAST_POS(f) ((f)->engine_pos)
ef5f5a3a	22
1d2af02a JA	23	struct syncio_data {
	24	struct iovec *iovecs;
	25	struct io_u **io_us;
	26	unsigned int queued;
e51cf72c	27	unsigned int events;
1d2af02a JA	28	unsigned long queued_bytes;
	29
	30	unsigned long long last_offset;
	31	struct fio_file *last_file;
	32	enum fio_ddir last_ddir;
a0679ce5 SB	33
a0679ce5 SB	34	struct frand_state rand_state;
1d2af02a JA	35	};
1d2af02a JA	36
6562685f	37	#ifdef FIO_HAVE_PWRITEV2
2cafffbe JA	38	struct psyncv2_options {
	39	void *pad;
	40	unsigned int hipri;
a0679ce5	41	unsigned int hipri_percentage;
2cafffbe JA	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	},
a0679ce5 SB	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	},
2cafffbe JA	66	{
	67	.name = NULL,
	68	},
	69	};
	70	#endif
	71
2866c82d JA	72	static int fio_syncio_prep(struct thread_data td, struct io_u io_u)
2866c82d JA	73	{
53cdc686 JA	74	struct fio_file *f = io_u->file;
53cdc686 JA	75
ff58fced	76	if (!ddir_rw(io_u->ddir))
87dc1ab1 JA	77	return 0;
87dc1ab1 JA	78
ef5f5a3a	79	if (LAST_POS(f) != -1ULL && LAST_POS(f) == io_u->offset)
e943b878 JA	80	return 0;
e943b878 JA	81
53cdc686	82	if (lseek(f->fd, io_u->offset, SEEK_SET) == -1) {
e1161c32	83	td_verror(td, errno, "lseek");
2866c82d JA	84	return 1;
	85	}
	86
	87	return 0;
	88	}
	89
2bd3eabc	90	static int fio_io_end(struct thread_data td, struct io_u io_u, int ret)
2866c82d	91	{
ff58fced	92	if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
ef5f5a3a	93	LAST_POS(io_u->file) = io_u->offset + ret;
e943b878	94
cec6b55d	95	if (ret != (int) io_u->xfer_buflen) {
22819ec2	96	if (ret >= 0) {
cec6b55d JA	97	io_u->resid = io_u->xfer_buflen - ret;
cec6b55d JA	98	io_u->error = 0;
36167d82	99	return FIO_Q_COMPLETED;
2866c82d JA	100	} else
	101	io_u->error = errno;
	102	}
	103
bfd197cb JA	104	if (io_u->error) {
bfd197cb JA	105	io_u_log_error(td, io_u);
e1161c32	106	td_verror(td, io_u->error, "xfer");
bfd197cb	107	}
2866c82d	108
36167d82	109	return FIO_Q_COMPLETED;
2866c82d JA	110	}
2866c82d JA	111
07fc0acd	112	#ifdef CONFIG_PWRITEV
d3b07186 BVA	113	static enum fio_q_status
d3b07186 BVA	114	fio_pvsyncio_queue(struct thread_data td, struct io_u io_u)
07fc0acd	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
d3b07186 BVA	141	static enum fio_q_status
d3b07186 BVA	142	fio_pvsyncio2_queue(struct thread_data td, struct io_u io_u)
2cafffbe	143	{
565e784d	144	struct syncio_data *sd = td->io_ops_data;
2cafffbe JA	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
a0679ce5 SB	152	if (o->hipri &&
a0679ce5 SB	153	(rand32_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
2cafffbe JA	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
d3b07186 BVA	174	static enum fio_q_status
d3b07186 BVA	175	fio_psyncio_queue(struct thread_data td, struct io_u io_u)
a31041ea	176	{
2bd3eabc JA	177	struct fio_file *f = io_u->file;
	178	int ret;
	179
	180	fio_ro_check(td, io_u);
	181
	182	if (io_u->ddir == DDIR_READ)
	183	ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	184	else if (io_u->ddir == DDIR_WRITE)
	185	ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
6eaf09d6 SL	186	else if (io_u->ddir == DDIR_TRIM) {
	187	do_io_u_trim(td, io_u);
	188	return FIO_Q_COMPLETED;
	189	} else
0a28ecda	190	ret = do_io_u_sync(td, io_u);
2bd3eabc JA	191
	192	return fio_io_end(td, io_u, ret);
	193	}
	194
d3b07186 BVA	195	static enum fio_q_status
d3b07186 BVA	196	fio_syncio_queue(struct thread_data td, struct io_u io_u)
2bd3eabc JA	197	{
	198	struct fio_file *f = io_u->file;
	199	int ret;
	200
	201	fio_ro_check(td, io_u);
	202
	203	if (io_u->ddir == DDIR_READ)
	204	ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	205	else if (io_u->ddir == DDIR_WRITE)
	206	ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
6eaf09d6 SL	207	else if (io_u->ddir == DDIR_TRIM) {
	208	do_io_u_trim(td, io_u);
	209	return FIO_Q_COMPLETED;
	210	} else
0a28ecda	211	ret = do_io_u_sync(td, io_u);
2bd3eabc JA	212
2bd3eabc JA	213	return fio_io_end(td, io_u, ret);
a31041ea	214	}
a31041ea	215
1d2af02a JA	216	static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
1d2af02a JA	217	unsigned int max,
1f440ece	218	const struct timespec fio_unused *t)
1d2af02a	219	{
565e784d	220	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	221	int ret;
	222
	223	if (min) {
e51cf72c JA	224	ret = sd->events;
e51cf72c JA	225	sd->events = 0;
1d2af02a JA	226	} else
	227	ret = 0;
	228
	229	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	230	return ret;
	231	}
	232
	233	static struct io_u fio_vsyncio_event(struct thread_data td, int event)
	234	{
565e784d	235	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	236
	237	return sd->io_us[event];
	238	}
	239
	240	static int fio_vsyncio_append(struct thread_data td, struct io_u io_u)
	241	{
565e784d	242	struct syncio_data *sd = td->io_ops_data;
1d2af02a	243
5f9099ea	244	if (ddir_sync(io_u->ddir))
1d2af02a JA	245	return 0;
	246
	247	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	248	io_u->ddir == sd->last_ddir)
	249	return 1;
	250
	251	return 0;
	252	}
	253
	254	static void fio_vsyncio_set_iov(struct syncio_data sd, struct io_u io_u,
2b13e716	255	int idx)
1d2af02a	256	{
2b13e716 JA	257	sd->io_us[idx] = io_u;
	258	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	259	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
1d2af02a JA	260	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	261	sd->last_file = io_u->file;
	262	sd->last_ddir = io_u->ddir;
	263	sd->queued_bytes += io_u->xfer_buflen;
	264	sd->queued++;
	265	}
	266
d3b07186 BVA	267	static enum fio_q_status
d3b07186 BVA	268	fio_vsyncio_queue(struct thread_data td, struct io_u io_u)
1d2af02a	269	{
565e784d	270	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	271
	272	fio_ro_check(td, io_u);
	273
	274	if (!fio_vsyncio_append(td, io_u)) {
	275	dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
	276	/*
	277	* If we can't append and have stuff queued, tell fio to
	278	* commit those first and then retry this io
	279	*/
	280	if (sd->queued)
	281	return FIO_Q_BUSY;
0a28ecda JA	282	if (ddir_sync(io_u->ddir)) {
	283	int ret = do_io_u_sync(td, io_u);
	284
	285	return fio_io_end(td, io_u, ret);
	286	}
cc9159c3	287
1d2af02a JA	288	sd->queued = 0;
	289	sd->queued_bytes = 0;
	290	fio_vsyncio_set_iov(sd, io_u, 0);
	291	} else {
	292	if (sd->queued == td->o.iodepth) {
	293	dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
	294	return FIO_Q_BUSY;
	295	}
	296
	297	dprint(FD_IO, "vsyncio_queue: append\n");
	298	fio_vsyncio_set_iov(sd, io_u, sd->queued);
	299	}
	300
	301	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	302	return FIO_Q_QUEUED;
	303	}
	304
	305	/*
	306	* Check that we transferred all bytes, or saw an error, etc
	307	*/
	308	static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
	309	{
565e784d	310	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	311	struct io_u *io_u;
	312	unsigned int i;
	313	int err;
	314
	315	/*
	316	* transferred everything, perfect
	317	*/
	318	if (bytes == sd->queued_bytes)
	319	return 0;
	320
	321	err = errno;
	322	for (i = 0; i < sd->queued; i++) {
	323	io_u = sd->io_us[i];
	324
	325	if (bytes == -1) {
	326	io_u->error = err;
	327	} else {
	328	unsigned int this_io;
	329
	330	this_io = bytes;
	331	if (this_io > io_u->xfer_buflen)
	332	this_io = io_u->xfer_buflen;
	333
	334	io_u->resid = io_u->xfer_buflen - this_io;
	335	io_u->error = 0;
	336	bytes -= this_io;
	337	}
	338	}
	339
	340	if (bytes == -1) {
	341	td_verror(td, err, "xfer vsync");
	342	return -err;
	343	}
	344
	345	return 0;
	346	}
	347
	348	static int fio_vsyncio_commit(struct thread_data *td)
	349	{
565e784d	350	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	351	struct fio_file *f;
	352	ssize_t ret;
	353
	354	if (!sd->queued)
	355	return 0;
	356
838bc709	357	io_u_mark_submit(td, sd->queued);
1d2af02a JA	358	f = sd->last_file;
	359
	360	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
	361	int err = -errno;
	362
	363	td_verror(td, errno, "lseek");
	364	return err;
	365	}
	366
	367	if (sd->last_ddir == DDIR_READ)
	368	ret = readv(f->fd, sd->iovecs, sd->queued);
	369	else
	370	ret = writev(f->fd, sd->iovecs, sd->queued);
	371
	372	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
e51cf72c JA	373	sd->events = sd->queued;
e51cf72c JA	374	sd->queued = 0;
1d2af02a JA	375	return fio_vsyncio_end(td, ret);
	376	}
	377
	378	static int fio_vsyncio_init(struct thread_data *td)
	379	{
	380	struct syncio_data *sd;
	381
	382	sd = malloc(sizeof(*sd));
	383	memset(sd, 0, sizeof(*sd));
	384	sd->last_offset = -1ULL;
	385	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	386	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
a0679ce5	387	init_rand(&sd->rand_state, 0);
1d2af02a	388
565e784d	389	td->io_ops_data = sd;
1d2af02a JA	390	return 0;
	391	}
	392
	393	static void fio_vsyncio_cleanup(struct thread_data *td)
	394	{
565e784d	395	struct syncio_data *sd = td->io_ops_data;
1d2af02a	396
02a3d83f TK	397	if (sd) {
	398	free(sd->iovecs);
	399	free(sd->io_us);
	400	free(sd);
	401	}
1d2af02a JA	402	}
1d2af02a JA	403
a31041ea	404	static struct ioengine_ops ioengine_rw = {
2866c82d JA	405	.name = "sync",
2866c82d JA	406	.version = FIO_IOOPS_VERSION,
2866c82d JA	407	.prep = fio_syncio_prep,
2866c82d JA	408	.queue = fio_syncio_queue,
b5af8293 JA	409	.open_file = generic_open_file,
b5af8293 JA	410	.close_file = generic_close_file,
df9c26b1	411	.get_file_size = generic_get_file_size,
2866c82d JA	412	.flags = FIO_SYNCIO,
2866c82d JA	413	};
5f350952	414
a31041ea	415	static struct ioengine_ops ioengine_prw = {
	416	.name = "psync",
	417	.version = FIO_IOOPS_VERSION,
2bd3eabc	418	.queue = fio_psyncio_queue,
a31041ea	419	.open_file = generic_open_file,
a31041ea	420	.close_file = generic_close_file,
df9c26b1	421	.get_file_size = generic_get_file_size,
a31041ea	422	.flags = FIO_SYNCIO,
	423	};
	424
1d2af02a JA	425	static struct ioengine_ops ioengine_vrw = {
	426	.name = "vsync",
	427	.version = FIO_IOOPS_VERSION,
	428	.init = fio_vsyncio_init,
	429	.cleanup = fio_vsyncio_cleanup,
	430	.queue = fio_vsyncio_queue,
	431	.commit = fio_vsyncio_commit,
	432	.event = fio_vsyncio_event,
	433	.getevents = fio_vsyncio_getevents,
	434	.open_file = generic_open_file,
	435	.close_file = generic_close_file,
df9c26b1	436	.get_file_size = generic_get_file_size,
1d2af02a JA	437	.flags = FIO_SYNCIO,
	438	};
	439
07fc0acd JA	440	#ifdef CONFIG_PWRITEV
	441	static struct ioengine_ops ioengine_pvrw = {
	442	.name = "pvsync",
	443	.version = FIO_IOOPS_VERSION,
	444	.init = fio_vsyncio_init,
	445	.cleanup = fio_vsyncio_cleanup,
	446	.queue = fio_pvsyncio_queue,
	447	.open_file = generic_open_file,
	448	.close_file = generic_close_file,
	449	.get_file_size = generic_get_file_size,
	450	.flags = FIO_SYNCIO,
	451	};
	452	#endif
	453
6562685f	454	#ifdef FIO_HAVE_PWRITEV2
2cafffbe JA	455	static struct ioengine_ops ioengine_pvrw2 = {
	456	.name = "pvsync2",
	457	.version = FIO_IOOPS_VERSION,
	458	.init = fio_vsyncio_init,
	459	.cleanup = fio_vsyncio_cleanup,
	460	.queue = fio_pvsyncio2_queue,
	461	.open_file = generic_open_file,
	462	.close_file = generic_close_file,
	463	.get_file_size = generic_get_file_size,
	464	.flags = FIO_SYNCIO,
	465	.options = options,
	466	.option_struct_size = sizeof(struct psyncv2_options),
	467	};
	468	#endif
	469
5f350952 JA	470	static void fio_init fio_syncio_register(void)
5f350952 JA	471	{
a31041ea	472	register_ioengine(&ioengine_rw);
a31041ea	473	register_ioengine(&ioengine_prw);
1d2af02a	474	register_ioengine(&ioengine_vrw);
9a0ced5a	475	#ifdef CONFIG_PWRITEV
07fc0acd	476	register_ioengine(&ioengine_pvrw);
9a0ced5a	477	#endif
6562685f	478	#ifdef FIO_HAVE_PWRITEV2
526c403d JD	479	register_ioengine(&ioengine_pvrw2);
526c403d JD	480	#endif
5f350952 JA	481	}
	482
	483	static void fio_exit fio_syncio_unregister(void)
	484	{
a31041ea	485	unregister_ioengine(&ioengine_rw);
a31041ea	486	unregister_ioengine(&ioengine_prw);
1d2af02a	487	unregister_ioengine(&ioengine_vrw);
9a0ced5a	488	#ifdef CONFIG_PWRITEV
07fc0acd	489	unregister_ioengine(&ioengine_pvrw);
9a0ced5a	490	#endif
6562685f	491	#ifdef FIO_HAVE_PWRITEV2
526c403d JD	492	unregister_ioengine(&ioengine_pvrw2);
526c403d JD	493	#endif
5f350952	494	}