[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;
};

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_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;
	int wait = 0;

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

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

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