[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;
};

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	= 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;

	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;
bb707dff	42	unsigned int uncached;
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	},
bb707dff JA	67	{
	68	.name = "uncached",
	69	.lname = "Uncached",
	70	.type = FIO_OPT_INT,
	71	.off1 = offsetof(struct psyncv2_options, uncached),
	72	.help = "Use RWF_UNCACHED for buffered read/writes",
	73	.category = FIO_OPT_C_ENGINE,
	74	.group = FIO_OPT_G_INVALID,
	75	},
2cafffbe JA	76	{
	77	.name = NULL,
	78	},
	79	};
	80	#endif
	81
2866c82d JA	82	static int fio_syncio_prep(struct thread_data td, struct io_u io_u)
2866c82d JA	83	{
53cdc686 JA	84	struct fio_file *f = io_u->file;
53cdc686 JA	85
ff58fced	86	if (!ddir_rw(io_u->ddir))
87dc1ab1 JA	87	return 0;
87dc1ab1 JA	88
ef5f5a3a	89	if (LAST_POS(f) != -1ULL && LAST_POS(f) == io_u->offset)
e943b878 JA	90	return 0;
e943b878 JA	91
53cdc686	92	if (lseek(f->fd, io_u->offset, SEEK_SET) == -1) {
e1161c32	93	td_verror(td, errno, "lseek");
2866c82d JA	94	return 1;
	95	}
	96
	97	return 0;
	98	}
	99
2bd3eabc	100	static int fio_io_end(struct thread_data td, struct io_u io_u, int ret)
2866c82d	101	{
ff58fced	102	if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
ef5f5a3a	103	LAST_POS(io_u->file) = io_u->offset + ret;
e943b878	104
cec6b55d	105	if (ret != (int) io_u->xfer_buflen) {
22819ec2	106	if (ret >= 0) {
cec6b55d JA	107	io_u->resid = io_u->xfer_buflen - ret;
cec6b55d JA	108	io_u->error = 0;
36167d82	109	return FIO_Q_COMPLETED;
2866c82d JA	110	} else
	111	io_u->error = errno;
	112	}
	113
bfd197cb JA	114	if (io_u->error) {
bfd197cb JA	115	io_u_log_error(td, io_u);
e1161c32	116	td_verror(td, io_u->error, "xfer");
bfd197cb	117	}
2866c82d	118
36167d82	119	return FIO_Q_COMPLETED;
2866c82d JA	120	}
2866c82d JA	121
07fc0acd	122	#ifdef CONFIG_PWRITEV
2e4ef4fb JA	123	static enum fio_q_status fio_pvsyncio_queue(struct thread_data *td,
2e4ef4fb JA	124	struct io_u *io_u)
07fc0acd	125	{
565e784d	126	struct syncio_data *sd = td->io_ops_data;
07fc0acd JA	127	struct iovec *iov = &sd->iovecs[0];
	128	struct fio_file *f = io_u->file;
	129	int ret;
	130
	131	fio_ro_check(td, io_u);
	132
	133	iov->iov_base = io_u->xfer_buf;
	134	iov->iov_len = io_u->xfer_buflen;
	135
	136	if (io_u->ddir == DDIR_READ)
	137	ret = preadv(f->fd, iov, 1, io_u->offset);
	138	else if (io_u->ddir == DDIR_WRITE)
	139	ret = pwritev(f->fd, iov, 1, io_u->offset);
	140	else if (io_u->ddir == DDIR_TRIM) {
	141	do_io_u_trim(td, io_u);
	142	return FIO_Q_COMPLETED;
	143	} else
	144	ret = do_io_u_sync(td, io_u);
	145
	146	return fio_io_end(td, io_u, ret);
	147	}
	148	#endif
	149
6562685f	150	#ifdef FIO_HAVE_PWRITEV2
2e4ef4fb JA	151	static enum fio_q_status fio_pvsyncio2_queue(struct thread_data *td,
2e4ef4fb JA	152	struct io_u *io_u)
2cafffbe	153	{
565e784d	154	struct syncio_data *sd = td->io_ops_data;
2cafffbe JA	155	struct psyncv2_options *o = td->eo;
	156	struct iovec *iov = &sd->iovecs[0];
	157	struct fio_file *f = io_u->file;
	158	int ret, flags = 0;
	159
	160	fio_ro_check(td, io_u);
	161
a0679ce5	162	if (o->hipri &&
1bd5d213	163	(rand_between(&sd->rand_state, 1, 100) <= o->hipri_percentage))
2cafffbe	164	flags \|= RWF_HIPRI;
bb707dff JA	165	if (!td->o.odirect && o->uncached)
bb707dff JA	166	flags \|= RWF_UNCACHED;
2cafffbe JA	167
	168	iov->iov_base = io_u->xfer_buf;
	169	iov->iov_len = io_u->xfer_buflen;
	170
	171	if (io_u->ddir == DDIR_READ)
	172	ret = preadv2(f->fd, iov, 1, io_u->offset, flags);
	173	else if (io_u->ddir == DDIR_WRITE)
	174	ret = pwritev2(f->fd, iov, 1, io_u->offset, flags);
	175	else if (io_u->ddir == DDIR_TRIM) {
	176	do_io_u_trim(td, io_u);
	177	return FIO_Q_COMPLETED;
	178	} else
	179	ret = do_io_u_sync(td, io_u);
	180
	181	return fio_io_end(td, io_u, ret);
	182	}
	183	#endif
	184
2e4ef4fb JA	185	static enum fio_q_status fio_psyncio_queue(struct thread_data *td,
2e4ef4fb JA	186	struct io_u *io_u)
a31041ea	187	{
2bd3eabc JA	188	struct fio_file *f = io_u->file;
	189	int ret;
	190
	191	fio_ro_check(td, io_u);
	192
	193	if (io_u->ddir == DDIR_READ)
	194	ret = pread(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
	195	else if (io_u->ddir == DDIR_WRITE)
	196	ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
6eaf09d6 SL	197	else if (io_u->ddir == DDIR_TRIM) {
	198	do_io_u_trim(td, io_u);
	199	return FIO_Q_COMPLETED;
	200	} else
0a28ecda	201	ret = do_io_u_sync(td, io_u);
2bd3eabc JA	202
	203	return fio_io_end(td, io_u, ret);
	204	}
	205
2e4ef4fb JA	206	static enum fio_q_status fio_syncio_queue(struct thread_data *td,
2e4ef4fb JA	207	struct io_u *io_u)
2bd3eabc JA	208	{
	209	struct fio_file *f = io_u->file;
	210	int ret;
	211
	212	fio_ro_check(td, io_u);
	213
	214	if (io_u->ddir == DDIR_READ)
	215	ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
	216	else if (io_u->ddir == DDIR_WRITE)
	217	ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
6eaf09d6 SL	218	else if (io_u->ddir == DDIR_TRIM) {
	219	do_io_u_trim(td, io_u);
	220	return FIO_Q_COMPLETED;
	221	} else
0a28ecda	222	ret = do_io_u_sync(td, io_u);
2bd3eabc JA	223
2bd3eabc JA	224	return fio_io_end(td, io_u, ret);
a31041ea	225	}
a31041ea	226
1d2af02a JA	227	static int fio_vsyncio_getevents(struct thread_data *td, unsigned int min,
1d2af02a JA	228	unsigned int max,
1f440ece	229	const struct timespec fio_unused *t)
1d2af02a	230	{
565e784d	231	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	232	int ret;
	233
	234	if (min) {
e51cf72c JA	235	ret = sd->events;
e51cf72c JA	236	sd->events = 0;
1d2af02a JA	237	} else
	238	ret = 0;
	239
	240	dprint(FD_IO, "vsyncio_getevents: min=%d,max=%d: %d\n", min, max, ret);
	241	return ret;
	242	}
	243
	244	static struct io_u fio_vsyncio_event(struct thread_data td, int event)
	245	{
565e784d	246	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	247
	248	return sd->io_us[event];
	249	}
	250
	251	static int fio_vsyncio_append(struct thread_data td, struct io_u io_u)
	252	{
565e784d	253	struct syncio_data *sd = td->io_ops_data;
1d2af02a	254
5f9099ea	255	if (ddir_sync(io_u->ddir))
1d2af02a JA	256	return 0;
	257
	258	if (io_u->offset == sd->last_offset && io_u->file == sd->last_file &&
	259	io_u->ddir == sd->last_ddir)
	260	return 1;
	261
	262	return 0;
	263	}
	264
	265	static void fio_vsyncio_set_iov(struct syncio_data sd, struct io_u io_u,
2b13e716	266	int idx)
1d2af02a	267	{
2b13e716 JA	268	sd->io_us[idx] = io_u;
	269	sd->iovecs[idx].iov_base = io_u->xfer_buf;
	270	sd->iovecs[idx].iov_len = io_u->xfer_buflen;
1d2af02a JA	271	sd->last_offset = io_u->offset + io_u->xfer_buflen;
	272	sd->last_file = io_u->file;
	273	sd->last_ddir = io_u->ddir;
	274	sd->queued_bytes += io_u->xfer_buflen;
	275	sd->queued++;
	276	}
	277
2e4ef4fb JA	278	static enum fio_q_status fio_vsyncio_queue(struct thread_data *td,
2e4ef4fb JA	279	struct io_u *io_u)
1d2af02a	280	{
565e784d	281	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	282
	283	fio_ro_check(td, io_u);
	284
	285	if (!fio_vsyncio_append(td, io_u)) {
	286	dprint(FD_IO, "vsyncio_queue: no append (%d)\n", sd->queued);
	287	/*
	288	* If we can't append and have stuff queued, tell fio to
	289	* commit those first and then retry this io
	290	*/
	291	if (sd->queued)
	292	return FIO_Q_BUSY;
0a28ecda JA	293	if (ddir_sync(io_u->ddir)) {
	294	int ret = do_io_u_sync(td, io_u);
	295
	296	return fio_io_end(td, io_u, ret);
	297	}
cc9159c3	298
1d2af02a JA	299	sd->queued = 0;
	300	sd->queued_bytes = 0;
	301	fio_vsyncio_set_iov(sd, io_u, 0);
	302	} else {
	303	if (sd->queued == td->o.iodepth) {
	304	dprint(FD_IO, "vsyncio_queue: max depth %d\n", sd->queued);
	305	return FIO_Q_BUSY;
	306	}
	307
	308	dprint(FD_IO, "vsyncio_queue: append\n");
	309	fio_vsyncio_set_iov(sd, io_u, sd->queued);
	310	}
	311
	312	dprint(FD_IO, "vsyncio_queue: depth now %d\n", sd->queued);
	313	return FIO_Q_QUEUED;
	314	}
	315
	316	/*
	317	* Check that we transferred all bytes, or saw an error, etc
	318	*/
	319	static int fio_vsyncio_end(struct thread_data *td, ssize_t bytes)
	320	{
565e784d	321	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	322	struct io_u *io_u;
	323	unsigned int i;
	324	int err;
	325
	326	/*
	327	* transferred everything, perfect
	328	*/
	329	if (bytes == sd->queued_bytes)
	330	return 0;
	331
	332	err = errno;
	333	for (i = 0; i < sd->queued; i++) {
	334	io_u = sd->io_us[i];
	335
	336	if (bytes == -1) {
	337	io_u->error = err;
	338	} else {
	339	unsigned int this_io;
	340
	341	this_io = bytes;
	342	if (this_io > io_u->xfer_buflen)
	343	this_io = io_u->xfer_buflen;
	344
	345	io_u->resid = io_u->xfer_buflen - this_io;
	346	io_u->error = 0;
	347	bytes -= this_io;
	348	}
	349	}
	350
	351	if (bytes == -1) {
	352	td_verror(td, err, "xfer vsync");
	353	return -err;
	354	}
	355
	356	return 0;
	357	}
	358
	359	static int fio_vsyncio_commit(struct thread_data *td)
	360	{
565e784d	361	struct syncio_data *sd = td->io_ops_data;
1d2af02a JA	362	struct fio_file *f;
	363	ssize_t ret;
	364
	365	if (!sd->queued)
	366	return 0;
	367
838bc709	368	io_u_mark_submit(td, sd->queued);
1d2af02a JA	369	f = sd->last_file;
	370
	371	if (lseek(f->fd, sd->io_us[0]->offset, SEEK_SET) == -1) {
	372	int err = -errno;
	373
	374	td_verror(td, errno, "lseek");
	375	return err;
	376	}
	377
	378	if (sd->last_ddir == DDIR_READ)
	379	ret = readv(f->fd, sd->iovecs, sd->queued);
	380	else
	381	ret = writev(f->fd, sd->iovecs, sd->queued);
	382
	383	dprint(FD_IO, "vsyncio_commit: %d\n", (int) ret);
e51cf72c JA	384	sd->events = sd->queued;
e51cf72c JA	385	sd->queued = 0;
1d2af02a JA	386	return fio_vsyncio_end(td, ret);
	387	}
	388
	389	static int fio_vsyncio_init(struct thread_data *td)
	390	{
	391	struct syncio_data *sd;
	392
	393	sd = malloc(sizeof(*sd));
	394	memset(sd, 0, sizeof(*sd));
	395	sd->last_offset = -1ULL;
	396	sd->iovecs = malloc(td->o.iodepth * sizeof(struct iovec));
	397	sd->io_us = malloc(td->o.iodepth * sizeof(struct io_u *));
a0679ce5	398	init_rand(&sd->rand_state, 0);
1d2af02a	399
565e784d	400	td->io_ops_data = sd;
1d2af02a JA	401	return 0;
	402	}
	403
	404	static void fio_vsyncio_cleanup(struct thread_data *td)
	405	{
565e784d	406	struct syncio_data *sd = td->io_ops_data;
1d2af02a	407
02a3d83f TK	408	if (sd) {
	409	free(sd->iovecs);
	410	free(sd->io_us);
	411	free(sd);
	412	}
1d2af02a JA	413	}
1d2af02a JA	414
a31041ea	415	static struct ioengine_ops ioengine_rw = {
2866c82d JA	416	.name = "sync",
2866c82d JA	417	.version = FIO_IOOPS_VERSION,
2866c82d JA	418	.prep = fio_syncio_prep,
2866c82d JA	419	.queue = fio_syncio_queue,
b5af8293 JA	420	.open_file = generic_open_file,
b5af8293 JA	421	.close_file = generic_close_file,
df9c26b1	422	.get_file_size = generic_get_file_size,
2866c82d JA	423	.flags = FIO_SYNCIO,
2866c82d JA	424	};
5f350952	425
a31041ea	426	static struct ioengine_ops ioengine_prw = {
	427	.name = "psync",
	428	.version = FIO_IOOPS_VERSION,
2bd3eabc	429	.queue = fio_psyncio_queue,
a31041ea	430	.open_file = generic_open_file,
a31041ea	431	.close_file = generic_close_file,
df9c26b1	432	.get_file_size = generic_get_file_size,
a31041ea	433	.flags = FIO_SYNCIO,
	434	};
	435
1d2af02a JA	436	static struct ioengine_ops ioengine_vrw = {
	437	.name = "vsync",
	438	.version = FIO_IOOPS_VERSION,
	439	.init = fio_vsyncio_init,
	440	.cleanup = fio_vsyncio_cleanup,
	441	.queue = fio_vsyncio_queue,
	442	.commit = fio_vsyncio_commit,
	443	.event = fio_vsyncio_event,
	444	.getevents = fio_vsyncio_getevents,
	445	.open_file = generic_open_file,
	446	.close_file = generic_close_file,
df9c26b1	447	.get_file_size = generic_get_file_size,
1d2af02a JA	448	.flags = FIO_SYNCIO,
	449	};
	450
07fc0acd JA	451	#ifdef CONFIG_PWRITEV
	452	static struct ioengine_ops ioengine_pvrw = {
	453	.name = "pvsync",
	454	.version = FIO_IOOPS_VERSION,
	455	.init = fio_vsyncio_init,
	456	.cleanup = fio_vsyncio_cleanup,
	457	.queue = fio_pvsyncio_queue,
	458	.open_file = generic_open_file,
	459	.close_file = generic_close_file,
	460	.get_file_size = generic_get_file_size,
	461	.flags = FIO_SYNCIO,
	462	};
	463	#endif
	464
6562685f	465	#ifdef FIO_HAVE_PWRITEV2
2cafffbe JA	466	static struct ioengine_ops ioengine_pvrw2 = {
	467	.name = "pvsync2",
	468	.version = FIO_IOOPS_VERSION,
	469	.init = fio_vsyncio_init,
	470	.cleanup = fio_vsyncio_cleanup,
	471	.queue = fio_pvsyncio2_queue,
	472	.open_file = generic_open_file,
	473	.close_file = generic_close_file,
	474	.get_file_size = generic_get_file_size,
	475	.flags = FIO_SYNCIO,
	476	.options = options,
	477	.option_struct_size = sizeof(struct psyncv2_options),
	478	};
	479	#endif
	480
5f350952 JA	481	static void fio_init fio_syncio_register(void)
5f350952 JA	482	{
a31041ea	483	register_ioengine(&ioengine_rw);
a31041ea	484	register_ioengine(&ioengine_prw);
1d2af02a	485	register_ioengine(&ioengine_vrw);
9a0ced5a	486	#ifdef CONFIG_PWRITEV
07fc0acd	487	register_ioengine(&ioengine_pvrw);
9a0ced5a	488	#endif
6562685f	489	#ifdef FIO_HAVE_PWRITEV2
526c403d JD	490	register_ioengine(&ioengine_pvrw2);
526c403d JD	491	#endif
5f350952 JA	492	}
	493
	494	static void fio_exit fio_syncio_unregister(void)
	495	{
a31041ea	496	unregister_ioengine(&ioengine_rw);
a31041ea	497	unregister_ioengine(&ioengine_prw);
1d2af02a	498	unregister_ioengine(&ioengine_vrw);
9a0ced5a	499	#ifdef CONFIG_PWRITEV
07fc0acd	500	unregister_ioengine(&ioengine_pvrw);
9a0ced5a	501	#endif
6562685f	502	#ifdef FIO_HAVE_PWRITEV2
526c403d JD	503	unregister_ioengine(&ioengine_pvrw2);
526c403d JD	504	#endif
5f350952	505	}