[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:
	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_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, const 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
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;
	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:
d9cb70a8 JA	89	if (rbd_data)
d9cb70a8 JA	90	free(rbd_data);
fc5c0345 DG	91	return 1;
	92
	93	}
	94
	95	static int _fio_rbd_connect(struct thread_data *td)
	96	{
	97	struct rbd_data *rbd_data = td->io_ops->data;
	98	struct rbd_options *o = td->eo;
	99	int r;
	100
	101	r = rados_create(&(rbd_data->cluster), o->client_name);
	102	if (r < 0) {
	103	log_err("rados_create failed.\n");
	104	goto failed_early;
	105	}
	106
	107	r = rados_conf_read_file(rbd_data->cluster, NULL);
	108	if (r < 0) {
	109	log_err("rados_conf_read_file failed.\n");
	110	goto failed_early;
	111	}
	112
	113	r = rados_connect(rbd_data->cluster);
	114	if (r < 0) {
	115	log_err("rados_connect failed.\n");
	116	goto failed_shutdown;
	117	}
	118
	119	r = rados_ioctx_create(rbd_data->cluster, o->pool_name,
	120	&(rbd_data->io_ctx));
	121	if (r < 0) {
	122	log_err("rados_ioctx_create failed.\n");
	123	goto failed_shutdown;
	124	}
	125
	126	r = rbd_open(rbd_data->io_ctx, o->rbd_name, &(rbd_data->image),
	127	NULL /snap / );
	128	if (r < 0) {
	129	log_err("rbd_open failed.\n");
	130	goto failed_open;
	131	}
	132	return 0;
	133
	134	failed_open:
	135	rados_ioctx_destroy(rbd_data->io_ctx);
086f9935	136	rbd_data->io_ctx = NULL;
fc5c0345 DG	137	failed_shutdown:
fc5c0345 DG	138	rados_shutdown(rbd_data->cluster);
086f9935	139	rbd_data->cluster = NULL;
fc5c0345 DG	140	failed_early:
	141	return 1;
	142	}
	143
	144	static void _fio_rbd_disconnect(struct rbd_data *rbd_data)
	145	{
	146	if (!rbd_data)
	147	return;
	148
	149	/* shutdown everything */
	150	if (rbd_data->image) {
	151	rbd_close(rbd_data->image);
	152	rbd_data->image = NULL;
	153	}
	154
	155	if (rbd_data->io_ctx) {
	156	rados_ioctx_destroy(rbd_data->io_ctx);
	157	rbd_data->io_ctx = NULL;
	158	}
	159
	160	if (rbd_data->cluster) {
	161	rados_shutdown(rbd_data->cluster);
	162	rbd_data->cluster = NULL;
	163	}
	164	}
	165
	166	static void _fio_rbd_finish_write_aiocb(rbd_completion_t comp, void *data)
	167	{
	168	struct io_u io_u = (struct io_u )data;
	169	struct fio_rbd_iou *fio_rbd_iou =
	170	(struct fio_rbd_iou *)io_u->engine_data;
	171
	172	fio_rbd_iou->io_complete = 1;
	173
	174	/* if write needs to be verified - we should not release comp here
	175	without fetching the result */
	176
	177	rbd_aio_release(comp);
	178	/* TODO handle error */
	179
	180	return;
	181	}
	182
	183	static void _fio_rbd_finish_read_aiocb(rbd_completion_t comp, void *data)
	184	{
	185	struct io_u io_u = (struct io_u )data;
	186	struct fio_rbd_iou *fio_rbd_iou =
	187	(struct fio_rbd_iou *)io_u->engine_data;
	188
	189	fio_rbd_iou->io_complete = 1;
	190
	191	/* if read needs to be verified - we should not release comp here
	192	without fetching the result */
	193	rbd_aio_release(comp);
	194
	195	/* TODO handle error */
	196
	197	return;
	198	}
	199
82340a9f HW	200	static void _fio_rbd_finish_sync_aiocb(rbd_completion_t comp, void *data)
	201	{
	202	struct io_u io_u = (struct io_u )data;
	203	struct fio_rbd_iou *fio_rbd_iou =
	204	(struct fio_rbd_iou *)io_u->engine_data;
	205
	206	fio_rbd_iou->io_complete = 1;
	207
	208	/* if sync needs to be verified - we should not release comp here
	209	without fetching the result */
	210	rbd_aio_release(comp);
	211
	212	/* TODO handle error */
	213
	214	return;
	215	}
	216
fc5c0345 DG	217	static struct io_u fio_rbd_event(struct thread_data td, int event)
	218	{
	219	struct rbd_data *rbd_data = td->io_ops->data;
	220
	221	return rbd_data->aio_events[event];
	222	}
	223
	224	static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
1f440ece	225	unsigned int max, const struct timespec *t)
fc5c0345 DG	226	{
	227	struct rbd_data *rbd_data = td->io_ops->data;
	228	unsigned int events = 0;
	229	struct io_u *io_u;
	230	int i;
	231	struct fio_rbd_iou *fov;
	232
	233	do {
	234	io_u_qiter(&td->io_u_all, io_u, i) {
	235	if (!(io_u->flags & IO_U_F_FLIGHT))
	236	continue;
	237
	238	fov = (struct fio_rbd_iou *)io_u->engine_data;
	239
	240	if (fov->io_complete) {
	241	fov->io_complete = 0;
	242	rbd_data->aio_events[events] = io_u;
	243	events++;
	244	}
	245
	246	}
	247	if (events < min)
	248	usleep(100);
	249	else
	250	break;
	251
	252	} while (1);
	253
	254	return events;
	255	}
	256
	257	static int fio_rbd_queue(struct thread_data td, struct io_u io_u)
	258	{
	259	int r = -1;
	260	struct rbd_data *rbd_data = td->io_ops->data;
	261	rbd_completion_t comp;
	262
	263	fio_ro_check(td, io_u);
	264
	265	if (io_u->ddir == DDIR_WRITE) {
	266	r = rbd_aio_create_completion(io_u,
	267	(rbd_callback_t)
	268	_fio_rbd_finish_write_aiocb,
	269	&comp);
	270	if (r < 0) {
	271	log_err
	272	("rbd_aio_create_completion for DDIR_WRITE failed.\n");
	273	goto failed;
	274	}
	275
	276	r = rbd_aio_write(rbd_data->image, io_u->offset,
	277	io_u->xfer_buflen, io_u->xfer_buf, comp);
	278	if (r < 0) {
	279	log_err("rbd_aio_write failed.\n");
	280	goto failed;
	281	}
	282
	283	} else if (io_u->ddir == DDIR_READ) {
	284	r = rbd_aio_create_completion(io_u,
	285	(rbd_callback_t)
	286	_fio_rbd_finish_read_aiocb,
	287	&comp);
	288	if (r < 0) {
	289	log_err
290	("rbd_aio_create_completion for DDIR_READ failed.\n");
291	goto failed;
292	}
293
294	r = rbd_aio_read(rbd_data->image, io_u->offset,
295	io_u->xfer_buflen, io_u->xfer_buf, comp);
296
297	if (r < 0) {
298	log_err("rbd_aio_read failed.\n");
299	goto failed;
300	}
301
302	} else if (io_u->ddir == DDIR_SYNC) {
82340a9f HW	303	r = rbd_aio_create_completion(io_u,
	304	(rbd_callback_t)
	305	_fio_rbd_finish_sync_aiocb,
	306	&comp);
	307	if (r < 0) {
	308	log_err
	309	("rbd_aio_create_completion for DDIR_SYNC failed.\n");
	310	goto failed;
	311	}
	312
	313	r = rbd_aio_flush(rbd_data->image, comp);
fc5c0345 DG	314	if (r < 0) {
	315	log_err("rbd_flush failed.\n");
	316	goto failed;
	317	}
	318
fc5c0345 DG	319	} else {
	320	dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
	321	io_u->ddir);
	322	return FIO_Q_COMPLETED;
	323	}
	324
	325	return FIO_Q_QUEUED;
	326
	327	failed:
	328	io_u->error = r;
	329	td_verror(td, io_u->error, "xfer");
	330	return FIO_Q_COMPLETED;
	331	}
	332
	333	static int fio_rbd_init(struct thread_data *td)
	334	{
	335	int r;
	336
	337	r = _fio_rbd_connect(td);
	338	if (r) {
	339	log_err("fio_rbd_connect failed, return code: %d .\n", r);
	340	goto failed;
	341	}
	342
	343	return 0;
	344
	345	failed:
	346	return 1;
	347
	348	}
	349
	350	static void fio_rbd_cleanup(struct thread_data *td)
	351	{
	352	struct rbd_data *rbd_data = td->io_ops->data;
	353
	354	if (rbd_data) {
	355	_fio_rbd_disconnect(rbd_data);
	356	free(rbd_data->aio_events);
	357	free(rbd_data);
	358	}
	359
	360	}
	361
	362	static int fio_rbd_setup(struct thread_data *td)
	363	{
	364	int r = 0;
	365	rbd_image_info_t info;
	366	struct fio_file *f;
	367	struct rbd_data *rbd_data = NULL;
	368	int major, minor, extra;
	369
	370	/* log version of librbd. No cluster connection required. */
	371	rbd_version(&major, &minor, &extra);
	372	log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra);
	373
	374	/* allocate engine specific structure to deal with librbd. */
	375	r = _fio_setup_rbd_data(td, &rbd_data);
	376	if (r) {
	377	log_err("fio_setup_rbd_data failed.\n");
	378	goto cleanup;
	379	}
	380	td->io_ops->data = rbd_data;
	381
	382	/* librbd does not allow us to run first in the main thread and later in a
383	* fork child. It needs to be the same process context all the time.
384	*/
385	td->o.use_thread = 1;
386
387	/* connect in the main thread to determine to determine
388	* the size of the given RADOS block device. And disconnect
389	* later on.
390	*/
391	r = _fio_rbd_connect(td);
392	if (r) {
393	log_err("fio_rbd_connect failed.\n");
394	goto cleanup;
395	}
396
397	/* get size of the RADOS block device */
398	r = rbd_stat(rbd_data->image, &info, sizeof(info));
399	if (r < 0) {
400	log_err("rbd_status failed.\n");
401	goto disconnect;
402	}
403	dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size);
404
405	/* taken from "net" engine. Pretend we deal with files,
406	* even if we do not have any ideas about files.
407	* The size of the RBD is set instead of a artificial file.
408	*/
409	if (!td->files_index) {
5903e7b7	410	add_file(td, td->o.filename ? : "rbd", 0, 0);
fc5c0345	411	td->o.nr_files = td->o.nr_files ? : 1;
b53f2c54	412	td->o.open_files++;
fc5c0345 DG	413	}
	414	f = td->files[0];
	415	f->real_file_size = info.size;
	416
	417	/* disconnect, then we were only connected to determine
	418	* the size of the RBD.
	419	*/
	420	_fio_rbd_disconnect(rbd_data);
	421	return 0;
	422
	423	disconnect:
	424	_fio_rbd_disconnect(rbd_data);
	425	cleanup:
	426	fio_rbd_cleanup(td);
	427	return r;
	428	}
	429
	430	static int fio_rbd_open(struct thread_data td, struct fio_file f)
	431	{
	432	return 0;
	433	}
	434
d9b100fc JA	435	static int fio_rbd_invalidate(struct thread_data td, struct fio_file f)
	436	{
	437	return 0;
	438	}
	439
fc5c0345 DG	440	static void fio_rbd_io_u_free(struct thread_data td, struct io_u io_u)
	441	{
	442	struct fio_rbd_iou *o = io_u->engine_data;
	443
	444	if (o) {
	445	io_u->engine_data = NULL;
	446	free(o);
	447	}
	448	}
	449
	450	static int fio_rbd_io_u_init(struct thread_data td, struct io_u io_u)
	451	{
	452	struct fio_rbd_iou *o;
	453
	454	o = malloc(sizeof(*o));
	455	o->io_complete = 0;
	456	o->io_u = io_u;
	457	io_u->engine_data = o;
	458	return 0;
	459	}
	460
10aa136b	461	static struct ioengine_ops ioengine = {
d9b100fc JA	462	.name = "rbd",
	463	.version = FIO_IOOPS_VERSION,
	464	.setup = fio_rbd_setup,
	465	.init = fio_rbd_init,
	466	.queue = fio_rbd_queue,
	467	.getevents = fio_rbd_getevents,
	468	.event = fio_rbd_event,
	469	.cleanup = fio_rbd_cleanup,
	470	.open_file = fio_rbd_open,
	471	.invalidate = fio_rbd_invalidate,
	472	.options = options,
	473	.io_u_init = fio_rbd_io_u_init,
	474	.io_u_free = fio_rbd_io_u_free,
	475	.option_struct_size = sizeof(struct rbd_options),
fc5c0345 DG	476	};
	477
	478	static void fio_init fio_rbd_register(void)
	479	{
	480	register_ioengine(&ioengine);
	481	}
	482
	483	static void fio_exit fio_rbd_unregister(void)
	484	{
	485	unregister_ioengine(&ioengine);
	486	}