[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;
	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;
};

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 = 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:
	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_write_aiocb(rbd_completion_t comp, void *data)
{
	struct io_u *io_u = (struct io_u *)data;
	struct fio_rbd_iou *fio_rbd_iou =
	    (struct fio_rbd_iou *)io_u->engine_data;

	fio_rbd_iou->io_complete = 1;

	/* if write needs to be verified - we should not release comp here
	   without fetching the result */

	rbd_aio_release(comp);
	/* TODO handle error */

	return;
}

static void _fio_rbd_finish_read_aiocb(rbd_completion_t comp, void *data)
{
	struct io_u *io_u = (struct io_u *)data;
	struct fio_rbd_iou *fio_rbd_iou =
	    (struct fio_rbd_iou *)io_u->engine_data;

	fio_rbd_iou->io_complete = 1;

	/* if read needs to be verified - we should not release comp here
	   without fetching the result */
	rbd_aio_release(comp);

	/* TODO handle error */

	return;
}

static void _fio_rbd_finish_sync_aiocb(rbd_completion_t comp, void *data)
{
	struct io_u *io_u = (struct io_u *)data;
	struct fio_rbd_iou *fio_rbd_iou =
	    (struct fio_rbd_iou *)io_u->engine_data;

	fio_rbd_iou->io_complete = 1;

	/* if sync needs to be verified - we should not release comp here
	   without fetching the result */
	rbd_aio_release(comp);

	/* TODO handle error */

	return;
}

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 int fio_rbd_getevents(struct thread_data *td, unsigned int min,
			     unsigned int max, struct timespec *t)
{
	struct rbd_data *rbd_data = td->io_ops->data;
	unsigned int events = 0;
	struct io_u *io_u;
	int i;
	struct fio_rbd_iou *fov;

	do {
		io_u_qiter(&td->io_u_all, io_u, i) {
			if (!(io_u->flags & IO_U_F_FLIGHT))
				continue;

			fov = (struct fio_rbd_iou *)io_u->engine_data;

			if (fov->io_complete) {
				fov->io_complete = 0;
				rbd_data->aio_events[events] = io_u;
				events++;
			}

		}
		if (events < min)
			usleep(100);
		else
			break;

	} while (1);

	return events;
}

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

	fio_ro_check(td, io_u);

	if (io_u->ddir == DDIR_WRITE) {
		r = rbd_aio_create_completion(io_u,
					      (rbd_callback_t)
					      _fio_rbd_finish_write_aiocb,
					      &comp);
		if (r < 0) {
			log_err
			    ("rbd_aio_create_completion for DDIR_WRITE failed.\n");
			goto failed;
		}

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

	} else if (io_u->ddir == DDIR_READ) {
		r = rbd_aio_create_completion(io_u,
					      (rbd_callback_t)
					      _fio_rbd_finish_read_aiocb,
					      &comp);
		if (r < 0) {
			log_err
			    ("rbd_aio_create_completion for DDIR_READ failed.\n");
			goto failed;
		}

		r = rbd_aio_read(rbd_data->image, io_u->offset,
				 io_u->xfer_buflen, io_u->xfer_buf, comp);

		if (r < 0) {
			log_err("rbd_aio_read failed.\n");
			goto failed;
		}

	} else if (io_u->ddir == DDIR_SYNC) {
		r = rbd_aio_create_completion(io_u,
					      (rbd_callback_t)
					      _fio_rbd_finish_sync_aiocb,
					      &comp);
		if (r < 0) {
			log_err
			    ("rbd_aio_create_completion for DDIR_SYNC failed.\n");
			goto failed;
		}

		r = rbd_aio_flush(rbd_data->image, comp);
		if (r < 0) {
			log_err("rbd_flush failed.\n");
			goto failed;
		}

	} else {
		dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
		       io_u->ddir);
		return FIO_Q_COMPLETED;
	}

	return FIO_Q_QUEUED;

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)
{
	return 0;
}

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

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

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

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