[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;
};

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.
	 */
	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 (rbd_aio_is_complete(fri->completion)) {
		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;

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