[fio.git] / engines / rbd.c

/*
 * rbd engine
 *
 * IO engine using Ceph's librbd to test RADOS Block Devices.
 *
 */

#include <rbd/librbd.h>

#include "../fio.h"

struct fio_rbd_iou {
	struct io_u *io_u;
	rbd_completion_t completion;
	int io_seen;
	int io_complete;
};

struct rbd_data {
	rados_t cluster;
	rados_ioctx_t io_ctx;
	rbd_image_t image;
	struct io_u **aio_events;
};

struct rbd_options {
	struct thread_data *td;
	char *rbd_name;
	char *pool_name;
	char *client_name;
	int busy_poll;
};

static struct fio_option options[] = {
	{
		.name		= "rbdname",
		.lname		= "rbd engine rbdname",
		.type		= FIO_OPT_STR_STORE,
		.help		= "RBD name for RBD engine",
		.off1		= offsetof(struct rbd_options, rbd_name),
		.category	= FIO_OPT_C_ENGINE,
		.group		= FIO_OPT_G_RBD,
	},
	{
		.name		= "pool",
		.lname		= "rbd engine pool",
		.type		= FIO_OPT_STR_STORE,
		.help		= "Name of the pool hosting the RBD for the RBD engine",
		.off1		= offsetof(struct rbd_options, pool_name),
		.category	= FIO_OPT_C_ENGINE,
		.group		= FIO_OPT_G_RBD,
	},
	{
		.name		= "clientname",
		.lname		= "rbd engine clientname",
		.type		= FIO_OPT_STR_STORE,
		.help		= "Name of the ceph client to access the RBD for the RBD engine",
		.off1		= offsetof(struct rbd_options, client_name),
		.category	= FIO_OPT_C_ENGINE,
		.group		= FIO_OPT_G_RBD,
	},
	{
		.name		= "busy_poll",
		.lname		= "Busy poll",
		.type		= FIO_OPT_BOOL,
		.help		= "Busy poll for completions instead of sleeping",
		.off1		= offsetof(struct rbd_options, busy_poll),
		.def		= "0",
		.category	= FIO_OPT_C_ENGINE,
		.group		= FIO_OPT_G_RBD,
	},
	{
		.name = NULL,
	},
};

static int _fio_setup_rbd_data(struct thread_data *td,
			       struct rbd_data **rbd_data_ptr)
{
	struct rbd_data *rbd_data;

	if (td->io_ops->data)
		return 0;

	rbd_data = malloc(sizeof(struct rbd_data));
	if (!rbd_data)
		goto failed;

	memset(rbd_data, 0, sizeof(struct rbd_data));

	rbd_data->aio_events = malloc(td->o.iodepth * sizeof(struct io_u *));
	if (!rbd_data->aio_events)
		goto failed;

	memset(rbd_data->aio_events, 0, td->o.iodepth * sizeof(struct io_u *));

	*rbd_data_ptr = rbd_data;

	return 0;

failed:
	if (rbd_data)
		free(rbd_data);
	return 1;

}

static int _fio_rbd_connect(struct thread_data *td)
{
	struct rbd_data *rbd_data = td->io_ops->data;
	struct rbd_options *o = td->eo;
	int r;

	r = rados_create(&rbd_data->cluster, o->client_name);
	if (r < 0) {
		log_err("rados_create failed.\n");
		goto failed_early;
	}

	r = rados_conf_read_file(rbd_data->cluster, NULL);
	if (r < 0) {
		log_err("rados_conf_read_file failed.\n");
		goto failed_early;
	}

	r = rados_connect(rbd_data->cluster);
	if (r < 0) {
		log_err("rados_connect failed.\n");
		goto failed_shutdown;
	}

	r = rados_ioctx_create(rbd_data->cluster, o->pool_name,
			       &rbd_data->io_ctx);
	if (r < 0) {
		log_err("rados_ioctx_create failed.\n");
		goto failed_shutdown;
	}

	r = rbd_open(rbd_data->io_ctx, o->rbd_name, &rbd_data->image,
		     NULL /*snap */ );
	if (r < 0) {
		log_err("rbd_open failed.\n");
		goto failed_open;
	}
	return 0;

failed_open:
	rados_ioctx_destroy(rbd_data->io_ctx);
	rbd_data->io_ctx = NULL;
failed_shutdown:
	rados_shutdown(rbd_data->cluster);
	rbd_data->cluster = NULL;
failed_early:
	return 1;
}

static void _fio_rbd_disconnect(struct rbd_data *rbd_data)
{
	if (!rbd_data)
		return;

	/* shutdown everything */
	if (rbd_data->image) {
		rbd_close(rbd_data->image);
		rbd_data->image = NULL;
	}

	if (rbd_data->io_ctx) {
		rados_ioctx_destroy(rbd_data->io_ctx);
		rbd_data->io_ctx = NULL;
	}

	if (rbd_data->cluster) {
		rados_shutdown(rbd_data->cluster);
		rbd_data->cluster = NULL;
	}
}

static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
{
	struct fio_rbd_iou *fri = data;
	struct io_u *io_u = fri->io_u;
	ssize_t ret;

	/*
	 * Looks like return value is 0 for success, or < 0 for
	 * a specific error. So we have to assume that it can't do
	 * partial completions.
	 */
	fri->io_complete = 1;
	
	ret = rbd_aio_get_return_value(fri->completion);
	if (ret < 0) {
		io_u->error = ret;
		io_u->resid = io_u->xfer_buflen;
	} else
		io_u->error = 0;
}

static struct io_u *fio_rbd_event(struct thread_data *td, int event)
{
	struct rbd_data *rbd_data = td->io_ops->data;

	return rbd_data->aio_events[event];
}

static inline int fri_check_complete(struct rbd_data *rbd_data,
				     struct io_u *io_u,
				     unsigned int *events)
{
	struct fio_rbd_iou *fri = io_u->engine_data;

	if (fri->io_complete) {
		fri->io_seen = 1;
		rbd_data->aio_events[*events] = io_u;
		(*events)++;

		rbd_aio_release(fri->completion);
		return 1;
	}

	return 0;
}

static int rbd_iter_events(struct thread_data *td, unsigned int *events,
			   unsigned int min_evts, int wait)
{
	struct rbd_data *rbd_data = td->io_ops->data;
	unsigned int this_events = 0;
	struct io_u *io_u;
	int i;

	io_u_qiter(&td->io_u_all, io_u, i) {
		struct fio_rbd_iou *fri = io_u->engine_data;

		if (!(io_u->flags & IO_U_F_FLIGHT))
			continue;
		if (fri->io_seen)
			continue;

		if (fri_check_complete(rbd_data, io_u, events))
			this_events++;
		else if (wait) {
			rbd_aio_wait_for_complete(fri->completion);

			if (fri_check_complete(rbd_data, io_u, events))
				this_events++;
		}
		if (*events >= min_evts)
			break;
	}

	return this_events;
}

static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
			     unsigned int max, const struct timespec *t)
{
	unsigned int this_events, events = 0;
	struct rbd_options *o = td->eo;
	int wait = 0;

	do {
		this_events = rbd_iter_events(td, &events, min, wait);

		if (events >= min)
			break;
		if (this_events)
			continue;

		if (!o->busy_poll)
			wait = 1;
		else
			nop;
	} while (1);

	return events;
}

static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u)
{
	struct rbd_data *rbd_data = td->io_ops->data;
	struct fio_rbd_iou *fri = io_u->engine_data;
	int r = -1;

	fio_ro_check(td, io_u);

	fri->io_seen = 0;
	fri->io_complete = 0;

	r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb,
						&fri->completion);
	if (r < 0) {
		log_err("rbd_aio_create_completion failed.\n");
		goto failed;
	}

	if (io_u->ddir == DDIR_WRITE) {
		r = rbd_aio_write(rbd_data->image, io_u->offset,
				  io_u->xfer_buflen, io_u->xfer_buf,
				  fri->completion);
		if (r < 0) {
			log_err("rbd_aio_write failed.\n");
			goto failed_comp;
		}

	} else if (io_u->ddir == DDIR_READ) {
		r = rbd_aio_read(rbd_data->image, io_u->offset,
				 io_u->xfer_buflen, io_u->xfer_buf,
				 fri->completion);

		if (r < 0) {
			log_err("rbd_aio_read failed.\n");
			goto failed_comp;
		}
	} else if (io_u->ddir == DDIR_TRIM) {
		r = rbd_aio_discard(rbd_data->image, io_u->offset,
				 io_u->xfer_buflen, fri->completion);
		if (r < 0) {
			log_err("rbd_aio_discard failed.\n");
			goto failed_comp;
		}
	} else if (io_u->ddir == DDIR_SYNC) {
		r = rbd_aio_flush(rbd_data->image, fri->completion);
		if (r < 0) {
			log_err("rbd_flush failed.\n");
			goto failed_comp;
		}
	} else {
		dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
		       io_u->ddir);
		goto failed_comp;
	}

	return FIO_Q_QUEUED;
failed_comp:
	rbd_aio_release(fri->completion);
failed:
	io_u->error = r;
	td_verror(td, io_u->error, "xfer");
	return FIO_Q_COMPLETED;
}

static int fio_rbd_init(struct thread_data *td)
{
	int r;

	r = _fio_rbd_connect(td);
	if (r) {
		log_err("fio_rbd_connect failed, return code: %d .\n", r);
		goto failed;
	}

	return 0;

failed:
	return 1;
}

static void fio_rbd_cleanup(struct thread_data *td)
{
	struct rbd_data *rbd_data = td->io_ops->data;

	if (rbd_data) {
		_fio_rbd_disconnect(rbd_data);
		free(rbd_data->aio_events);
		free(rbd_data);
	}

}

static int fio_rbd_setup(struct thread_data *td)
{
	int r = 0;
	rbd_image_info_t info;
	struct fio_file *f;
	struct rbd_data *rbd_data = NULL;
	int major, minor, extra;

	/* log version of librbd. No cluster connection required. */
	rbd_version(&major, &minor, &extra);
	log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra);

	/* allocate engine specific structure to deal with librbd. */
	r = _fio_setup_rbd_data(td, &rbd_data);
	if (r) {
		log_err("fio_setup_rbd_data failed.\n");
		goto cleanup;
	}
	td->io_ops->data = rbd_data;

	/* librbd does not allow us to run first in the main thread and later
	 * in a fork child. It needs to be the same process context all the
	 * time. 
	 */
	td->o.use_thread = 1;

	/* connect in the main thread to determine to determine
	 * the size of the given RADOS block device. And disconnect
	 * later on.
	 */
	r = _fio_rbd_connect(td);
	if (r) {
		log_err("fio_rbd_connect failed.\n");
		goto cleanup;
	}

	/* get size of the RADOS block device */
	r = rbd_stat(rbd_data->image, &info, sizeof(info));
	if (r < 0) {
		log_err("rbd_status failed.\n");
		goto disconnect;
	}
	dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size);

	/* taken from "net" engine. Pretend we deal with files,
	 * even if we do not have any ideas about files.
	 * The size of the RBD is set instead of a artificial file.
	 */
	if (!td->files_index) {
		add_file(td, td->o.filename ? : "rbd", 0, 0);
		td->o.nr_files = td->o.nr_files ? : 1;
		td->o.open_files++;
	}
	f = td->files[0];
	f->real_file_size = info.size;

	/* disconnect, then we were only connected to determine
	 * the size of the RBD.
	 */
	_fio_rbd_disconnect(rbd_data);
	return 0;

disconnect:
	_fio_rbd_disconnect(rbd_data);
cleanup:
	fio_rbd_cleanup(td);
	return r;
}

static int fio_rbd_open(struct thread_data *td, struct fio_file *f)
{
	return 0;
}

static int fio_rbd_invalidate(struct thread_data *td, struct fio_file *f)
{
#if defined(CONFIG_RBD_INVAL)
	struct rbd_data *rbd_data = td->io_ops->data;

	return rbd_invalidate_cache(rbd_data->image);
#else
	return 0;
#endif
}

static void fio_rbd_io_u_free(struct thread_data *td, struct io_u *io_u)
{
	struct fio_rbd_iou *fri = io_u->engine_data;

	if (fri) {
		io_u->engine_data = NULL;
		free(fri);
	}
}

static int fio_rbd_io_u_init(struct thread_data *td, struct io_u *io_u)
{
	struct fio_rbd_iou *fri;

	fri = calloc(1, sizeof(*fri));
	fri->io_u = io_u;
	io_u->engine_data = fri;
	return 0;
}

static struct ioengine_ops ioengine = {
	.name			= "rbd",
	.version		= FIO_IOOPS_VERSION,
	.setup			= fio_rbd_setup,
	.init			= fio_rbd_init,
	.queue			= fio_rbd_queue,
	.getevents		= fio_rbd_getevents,
	.event			= fio_rbd_event,
	.cleanup		= fio_rbd_cleanup,
	.open_file		= fio_rbd_open,
	.invalidate		= fio_rbd_invalidate,
	.options		= options,
	.io_u_init		= fio_rbd_io_u_init,
	.io_u_free		= fio_rbd_io_u_free,
	.option_struct_size	= sizeof(struct rbd_options),
};

static void fio_init fio_rbd_register(void)
{
	register_ioengine(&ioengine);
}

static void fio_exit fio_rbd_unregister(void)
{
	unregister_ioengine(&ioengine);
}
Commit	Line	Data
fc5c0345 DG	1	/*
	2	* rbd engine
	3	*
	4	* IO engine using Ceph's librbd to test RADOS Block Devices.
	5	*
	6	*/
	7
	8	#include <rbd/librbd.h>
	9
	10	#include "../fio.h"
	11
	12	struct fio_rbd_iou {
	13	struct io_u *io_u;
d8b64af2	14	rbd_completion_t completion;
d8b64af2	15	int io_seen;
b8ecbef6	16	int io_complete;
fc5c0345 DG	17	};
	18
	19	struct rbd_data {
	20	rados_t cluster;
	21	rados_ioctx_t io_ctx;
	22	rbd_image_t image;
	23	struct io_u **aio_events;
	24	};
	25
	26	struct rbd_options {
fea585d4	27	struct thread_data *td;
fc5c0345 DG	28	char *rbd_name;
	29	char *pool_name;
	30	char *client_name;
d7d702c7	31	int busy_poll;
fc5c0345 DG	32	};
	33
	34	static struct fio_option options[] = {
	35	{
d8b64af2 JA	36	.name = "rbdname",
	37	.lname = "rbd engine rbdname",
	38	.type = FIO_OPT_STR_STORE,
	39	.help = "RBD name for RBD engine",
	40	.off1 = offsetof(struct rbd_options, rbd_name),
	41	.category = FIO_OPT_C_ENGINE,
	42	.group = FIO_OPT_G_RBD,
	43	},
fc5c0345	44	{
d7d702c7 JA	45	.name = "pool",
	46	.lname = "rbd engine pool",
	47	.type = FIO_OPT_STR_STORE,
	48	.help = "Name of the pool hosting the RBD for the RBD engine",
	49	.off1 = offsetof(struct rbd_options, pool_name),
	50	.category = FIO_OPT_C_ENGINE,
	51	.group = FIO_OPT_G_RBD,
d8b64af2	52	},
fc5c0345	53	{
d7d702c7 JA	54	.name = "clientname",
	55	.lname = "rbd engine clientname",
	56	.type = FIO_OPT_STR_STORE,
	57	.help = "Name of the ceph client to access the RBD for the RBD engine",
	58	.off1 = offsetof(struct rbd_options, client_name),
	59	.category = FIO_OPT_C_ENGINE,
	60	.group = FIO_OPT_G_RBD,
	61	},
	62	{
	63	.name = "busy_poll",
	64	.lname = "Busy poll",
	65	.type = FIO_OPT_BOOL,
	66	.help = "Busy poll for completions instead of sleeping",
fea585d4	67	.off1 = offsetof(struct rbd_options, busy_poll),
d7d702c7 JA	68	.def = "0",
	69	.category = FIO_OPT_C_ENGINE,
	70	.group = FIO_OPT_G_RBD,
d8b64af2	71	},
fc5c0345	72	{
d8b64af2 JA	73	.name = NULL,
d8b64af2 JA	74	},
fc5c0345 DG	75	};
	76
	77	static int _fio_setup_rbd_data(struct thread_data *td,
	78	struct rbd_data **rbd_data_ptr)
	79	{
	80	struct rbd_data *rbd_data;
	81
	82	if (td->io_ops->data)
	83	return 0;
	84
	85	rbd_data = malloc(sizeof(struct rbd_data));
	86	if (!rbd_data)
	87	goto failed;
	88
	89	memset(rbd_data, 0, sizeof(struct rbd_data));
	90
	91	rbd_data->aio_events = malloc(td->o.iodepth * sizeof(struct io_u *));
	92	if (!rbd_data->aio_events)
	93	goto failed;
	94
	95	memset(rbd_data->aio_events, 0, td->o.iodepth * sizeof(struct io_u *));
	96
	97	*rbd_data_ptr = rbd_data;
	98
	99	return 0;
	100
	101	failed:
d9cb70a8 JA	102	if (rbd_data)
d9cb70a8 JA	103	free(rbd_data);
fc5c0345 DG	104	return 1;
	105
	106	}
	107
	108	static int _fio_rbd_connect(struct thread_data *td)
	109	{
	110	struct rbd_data *rbd_data = td->io_ops->data;
	111	struct rbd_options *o = td->eo;
	112	int r;
	113
dbf388d2	114	r = rados_create(&rbd_data->cluster, o->client_name);
fc5c0345 DG	115	if (r < 0) {
	116	log_err("rados_create failed.\n");
	117	goto failed_early;
	118	}
	119
	120	r = rados_conf_read_file(rbd_data->cluster, NULL);
	121	if (r < 0) {
	122	log_err("rados_conf_read_file failed.\n");
	123	goto failed_early;
	124	}
	125
	126	r = rados_connect(rbd_data->cluster);
	127	if (r < 0) {
	128	log_err("rados_connect failed.\n");
	129	goto failed_shutdown;
	130	}
	131
	132	r = rados_ioctx_create(rbd_data->cluster, o->pool_name,
dbf388d2	133	&rbd_data->io_ctx);
fc5c0345 DG	134	if (r < 0) {
	135	log_err("rados_ioctx_create failed.\n");
	136	goto failed_shutdown;
	137	}
	138
dbf388d2	139	r = rbd_open(rbd_data->io_ctx, o->rbd_name, &rbd_data->image,
fc5c0345 DG	140	NULL /snap / );
	141	if (r < 0) {
	142	log_err("rbd_open failed.\n");
	143	goto failed_open;
	144	}
	145	return 0;
	146
	147	failed_open:
	148	rados_ioctx_destroy(rbd_data->io_ctx);
086f9935	149	rbd_data->io_ctx = NULL;
fc5c0345 DG	150	failed_shutdown:
fc5c0345 DG	151	rados_shutdown(rbd_data->cluster);
086f9935	152	rbd_data->cluster = NULL;
fc5c0345 DG	153	failed_early:
	154	return 1;
	155	}
	156
	157	static void _fio_rbd_disconnect(struct rbd_data *rbd_data)
	158	{
	159	if (!rbd_data)
	160	return;
	161
	162	/* shutdown everything */
	163	if (rbd_data->image) {
	164	rbd_close(rbd_data->image);
	165	rbd_data->image = NULL;
	166	}
	167
	168	if (rbd_data->io_ctx) {
	169	rados_ioctx_destroy(rbd_data->io_ctx);
	170	rbd_data->io_ctx = NULL;
	171	}
	172
	173	if (rbd_data->cluster) {
	174	rados_shutdown(rbd_data->cluster);
	175	rbd_data->cluster = NULL;
	176	}
	177	}
	178
d8b64af2	179	static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
fc5c0345	180	{
dbf388d2 JA	181	struct fio_rbd_iou *fri = data;
dbf388d2 JA	182	struct io_u *io_u = fri->io_u;
d8b64af2	183	ssize_t ret;
fc5c0345	184
d8b64af2 JA	185	/*
	186	* Looks like return value is 0 for success, or < 0 for
	187	* a specific error. So we have to assume that it can't do
	188	* partial completions.
	189	*/
b8ecbef6 KM	190	fri->io_complete = 1;
b8ecbef6 KM	191
d8b64af2 JA	192	ret = rbd_aio_get_return_value(fri->completion);
	193	if (ret < 0) {
	194	io_u->error = ret;
	195	io_u->resid = io_u->xfer_buflen;
	196	} else
	197	io_u->error = 0;
	198	}
fc5c0345	199
d8b64af2 JA	200	static struct io_u fio_rbd_event(struct thread_data td, int event)
	201	{
	202	struct rbd_data *rbd_data = td->io_ops->data;
fc5c0345	203
d8b64af2	204	return rbd_data->aio_events[event];
fc5c0345 DG	205	}
fc5c0345 DG	206
d8b64af2 JA	207	static inline int fri_check_complete(struct rbd_data *rbd_data,
	208	struct io_u *io_u,
	209	unsigned int *events)
fc5c0345	210	{
d8b64af2	211	struct fio_rbd_iou *fri = io_u->engine_data;
fc5c0345	212
b8ecbef6	213	if (fri->io_complete) {
d8b64af2 JA	214	fri->io_seen = 1;
	215	rbd_data->aio_events[*events] = io_u;
	216	(*events)++;
fc5c0345	217
d8b64af2 JA	218	rbd_aio_release(fri->completion);
	219	return 1;
	220	}
fc5c0345	221
d8b64af2	222	return 0;
fc5c0345 DG	223	}
fc5c0345 DG	224
d8b64af2 JA	225	static int rbd_iter_events(struct thread_data td, unsigned int events,
d8b64af2 JA	226	unsigned int min_evts, int wait)
82340a9f	227	{
d8b64af2 JA	228	struct rbd_data *rbd_data = td->io_ops->data;
	229	unsigned int this_events = 0;
	230	struct io_u *io_u;
	231	int i;
82340a9f	232
d8b64af2 JA	233	io_u_qiter(&td->io_u_all, io_u, i) {
d8b64af2 JA	234	struct fio_rbd_iou *fri = io_u->engine_data;
82340a9f	235
d8b64af2 JA	236	if (!(io_u->flags & IO_U_F_FLIGHT))
	237	continue;
	238	if (fri->io_seen)
	239	continue;
82340a9f	240
d8b64af2 JA	241	if (fri_check_complete(rbd_data, io_u, events))
	242	this_events++;
	243	else if (wait) {
	244	rbd_aio_wait_for_complete(fri->completion);
82340a9f	245
d8b64af2 JA	246	if (fri_check_complete(rbd_data, io_u, events))
	247	this_events++;
	248	}
	249	if (*events >= min_evts)
	250	break;
	251	}
fc5c0345	252
d8b64af2	253	return this_events;
fc5c0345 DG	254	}
	255
	256	static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
1f440ece	257	unsigned int max, const struct timespec *t)
fc5c0345	258	{
d8b64af2	259	unsigned int this_events, events = 0;
d7d702c7	260	struct rbd_options *o = td->eo;
d8b64af2	261	int wait = 0;
fc5c0345 DG	262
fc5c0345 DG	263	do {
d8b64af2	264	this_events = rbd_iter_events(td, &events, min, wait);
fc5c0345	265
d8b64af2	266	if (events >= min)
fc5c0345	267	break;
d8b64af2 JA	268	if (this_events)
d8b64af2 JA	269	continue;
fc5c0345	270
d7d702c7 JA	271	if (!o->busy_poll)
	272	wait = 1;
	273	else
	274	nop;
fc5c0345 DG	275	} while (1);
	276
	277	return events;
	278	}
	279
	280	static int fio_rbd_queue(struct thread_data td, struct io_u io_u)
	281	{
fc5c0345	282	struct rbd_data *rbd_data = td->io_ops->data;
d8b64af2 JA	283	struct fio_rbd_iou *fri = io_u->engine_data;
d8b64af2 JA	284	int r = -1;
fc5c0345 DG	285
	286	fio_ro_check(td, io_u);
	287
d8b64af2	288	fri->io_seen = 0;
b8ecbef6	289	fri->io_complete = 0;
d8b64af2	290
dbf388d2	291	r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb,
d8b64af2	292	&fri->completion);
dbf388d2 JA	293	if (r < 0) {
	294	log_err("rbd_aio_create_completion failed.\n");
	295	goto failed;
	296	}
fc5c0345	297
dbf388d2	298	if (io_u->ddir == DDIR_WRITE) {
fc5c0345	299	r = rbd_aio_write(rbd_data->image, io_u->offset,
d8b64af2 JA	300	io_u->xfer_buflen, io_u->xfer_buf,
d8b64af2 JA	301	fri->completion);
fc5c0345 DG	302	if (r < 0) {
fc5c0345 DG	303	log_err("rbd_aio_write failed.\n");
dbf388d2	304	goto failed_comp;
fc5c0345 DG	305	}
	306
	307	} else if (io_u->ddir == DDIR_READ) {
fc5c0345	308	r = rbd_aio_read(rbd_data->image, io_u->offset,
d8b64af2 JA	309	io_u->xfer_buflen, io_u->xfer_buf,
d8b64af2 JA	310	fri->completion);
fc5c0345 DG	311
	312	if (r < 0) {
	313	log_err("rbd_aio_read failed.\n");
dbf388d2	314	goto failed_comp;
fc5c0345	315	}
dbf388d2 JA	316	} else if (io_u->ddir == DDIR_TRIM) {
	317	r = rbd_aio_discard(rbd_data->image, io_u->offset,
	318	io_u->xfer_buflen, fri->completion);
82340a9f	319	if (r < 0) {
dbf388d2 JA	320	log_err("rbd_aio_discard failed.\n");
dbf388d2 JA	321	goto failed_comp;
82340a9f	322	}
dbf388d2	323	} else if (io_u->ddir == DDIR_SYNC) {
d8b64af2	324	r = rbd_aio_flush(rbd_data->image, fri->completion);
fc5c0345 DG	325	if (r < 0) {
fc5c0345 DG	326	log_err("rbd_flush failed.\n");
dbf388d2	327	goto failed_comp;
fc5c0345	328	}
fc5c0345 DG	329	} else {
	330	dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
	331	io_u->ddir);
dbf388d2	332	goto failed_comp;
fc5c0345 DG	333	}
	334
	335	return FIO_Q_QUEUED;
dbf388d2 JA	336	failed_comp:
dbf388d2 JA	337	rbd_aio_release(fri->completion);
fc5c0345 DG	338	failed:
	339	io_u->error = r;
	340	td_verror(td, io_u->error, "xfer");
	341	return FIO_Q_COMPLETED;
	342	}
	343
	344	static int fio_rbd_init(struct thread_data *td)
	345	{
	346	int r;
	347
	348	r = _fio_rbd_connect(td);
	349	if (r) {
	350	log_err("fio_rbd_connect failed, return code: %d .\n", r);
	351	goto failed;
	352	}
	353
	354	return 0;
	355
	356	failed:
	357	return 1;
fc5c0345 DG	358	}
	359
	360	static void fio_rbd_cleanup(struct thread_data *td)
	361	{
	362	struct rbd_data *rbd_data = td->io_ops->data;
	363
	364	if (rbd_data) {
	365	_fio_rbd_disconnect(rbd_data);
	366	free(rbd_data->aio_events);
	367	free(rbd_data);
	368	}
	369
	370	}
	371
	372	static int fio_rbd_setup(struct thread_data *td)
	373	{
	374	int r = 0;
	375	rbd_image_info_t info;
	376	struct fio_file *f;
	377	struct rbd_data *rbd_data = NULL;
	378	int major, minor, extra;
	379
	380	/* log version of librbd. No cluster connection required. */
	381	rbd_version(&major, &minor, &extra);
	382	log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra);
	383
	384	/* allocate engine specific structure to deal with librbd. */
	385	r = _fio_setup_rbd_data(td, &rbd_data);
	386	if (r) {
	387	log_err("fio_setup_rbd_data failed.\n");
	388	goto cleanup;
	389	}
	390	td->io_ops->data = rbd_data;
	391
d8b64af2 JA	392	/* librbd does not allow us to run first in the main thread and later
	393	* in a fork child. It needs to be the same process context all the
	394	* time.
fc5c0345 DG	395	*/
	396	td->o.use_thread = 1;
	397
	398	/* connect in the main thread to determine to determine
	399	* the size of the given RADOS block device. And disconnect
	400	* later on.
	401	*/
	402	r = _fio_rbd_connect(td);
	403	if (r) {
	404	log_err("fio_rbd_connect failed.\n");
	405	goto cleanup;
	406	}
	407
	408	/* get size of the RADOS block device */
	409	r = rbd_stat(rbd_data->image, &info, sizeof(info));
	410	if (r < 0) {
	411	log_err("rbd_status failed.\n");
	412	goto disconnect;
	413	}
	414	dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size);
	415
	416	/* taken from "net" engine. Pretend we deal with files,
	417	* even if we do not have any ideas about files.
	418	* The size of the RBD is set instead of a artificial file.
	419	*/
	420	if (!td->files_index) {
5903e7b7	421	add_file(td, td->o.filename ? : "rbd", 0, 0);
fc5c0345	422	td->o.nr_files = td->o.nr_files ? : 1;
b53f2c54	423	td->o.open_files++;
fc5c0345 DG	424	}
	425	f = td->files[0];
	426	f->real_file_size = info.size;
	427
	428	/* disconnect, then we were only connected to determine
	429	* the size of the RBD.
	430	*/
	431	_fio_rbd_disconnect(rbd_data);
	432	return 0;
	433
	434	disconnect:
	435	_fio_rbd_disconnect(rbd_data);
	436	cleanup:
	437	fio_rbd_cleanup(td);
	438	return r;
	439	}
	440
	441	static int fio_rbd_open(struct thread_data td, struct fio_file f)
	442	{
	443	return 0;
	444	}
	445
d9b100fc JA	446	static int fio_rbd_invalidate(struct thread_data td, struct fio_file f)
d9b100fc JA	447	{
903b2812 JA	448	#if defined(CONFIG_RBD_INVAL)
	449	struct rbd_data *rbd_data = td->io_ops->data;
	450
	451	return rbd_invalidate_cache(rbd_data->image);
	452	#else
d9b100fc	453	return 0;
903b2812	454	#endif
d9b100fc JA	455	}
d9b100fc JA	456
fc5c0345 DG	457	static void fio_rbd_io_u_free(struct thread_data td, struct io_u io_u)
fc5c0345 DG	458	{
d8b64af2	459	struct fio_rbd_iou *fri = io_u->engine_data;
fc5c0345	460
d8b64af2	461	if (fri) {
fc5c0345	462	io_u->engine_data = NULL;
d8b64af2	463	free(fri);
fc5c0345 DG	464	}
	465	}
	466
	467	static int fio_rbd_io_u_init(struct thread_data td, struct io_u io_u)
	468	{
d8b64af2	469	struct fio_rbd_iou *fri;
fc5c0345	470
d8b64af2 JA	471	fri = calloc(1, sizeof(*fri));
	472	fri->io_u = io_u;
	473	io_u->engine_data = fri;
fc5c0345 DG	474	return 0;
	475	}
	476
10aa136b	477	static struct ioengine_ops ioengine = {
d9b100fc JA	478	.name = "rbd",
	479	.version = FIO_IOOPS_VERSION,
	480	.setup = fio_rbd_setup,
	481	.init = fio_rbd_init,
	482	.queue = fio_rbd_queue,
	483	.getevents = fio_rbd_getevents,
	484	.event = fio_rbd_event,
	485	.cleanup = fio_rbd_cleanup,
	486	.open_file = fio_rbd_open,
	487	.invalidate = fio_rbd_invalidate,
	488	.options = options,
	489	.io_u_init = fio_rbd_io_u_init,
	490	.io_u_free = fio_rbd_io_u_free,
	491	.option_struct_size = sizeof(struct rbd_options),
fc5c0345 DG	492	};
	493
	494	static void fio_init fio_rbd_register(void)
	495	{
	496	register_ioengine(&ioengine);
	497	}
	498
	499	static void fio_exit fio_rbd_unregister(void)
	500	{
	501	unregister_ioengine(&ioengine);
	502	}