[fio.git] / engines / rbd.c

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

#include <rbd/librbd.h>

#include "../fio.h"

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

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

struct rbd_options {
	void *pad;
	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;

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

	rbd = calloc(1, sizeof(struct rbd_data));
	if (!rbd)
		goto failed;

	rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *));
	if (!rbd->aio_events)
		goto failed;

	rbd->sort_events = calloc(td->o.iodepth, sizeof(struct io_u *));
	if (!rbd->sort_events)
		goto failed;

	*rbd_data_ptr = rbd;
	return 0;

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

}

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

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

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

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

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

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

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

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

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

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

	if (rbd->cluster) {
		rados_shutdown(rbd->cluster);
		rbd->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;

	fri->io_complete = 1;
}

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

	return rbd->aio_events[event];
}

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

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

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

	return 0;
}

static inline int rbd_io_u_seen(struct io_u *io_u)
{
	struct fio_rbd_iou *fri = io_u->engine_data;

	return fri->io_seen;
}

static void rbd_io_u_wait_complete(struct io_u *io_u)
{
	struct fio_rbd_iou *fri = io_u->engine_data;

	rbd_aio_wait_for_complete(fri->completion);
}

static int rbd_io_u_cmp(const void *p1, const void *p2)
{
	const struct io_u **a = (const struct io_u **) p1;
	const struct io_u **b = (const struct io_u **) p2;
	uint64_t at, bt;

	at = utime_since_now(&(*a)->start_time);
	bt = utime_since_now(&(*b)->start_time);

	if (at < bt)
		return -1;
	else if (at == bt)
		return 0;
	else
		return 1;
}

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

	sidx = 0;
	io_u_qiter(&td->io_u_all, io_u, i) {
		if (!(io_u->flags & IO_U_F_FLIGHT))
			continue;
		if (rbd_io_u_seen(io_u))
			continue;

		if (fri_check_complete(rbd, io_u, events))
			this_events++;
		else if (wait)
			rbd->sort_events[sidx++] = io_u;
	}

	if (!wait || !sidx)
		return this_events;

	/*
	 * Sort events, oldest issue first, then wait on as many as we
	 * need in order of age. If we have enough events, stop waiting,
	 * and just check if any of the older ones are done.
	 */
	if (sidx > 1)
		qsort(rbd->sort_events, sidx, sizeof(struct io_u *), rbd_io_u_cmp);

	for (i = 0; i < sidx; i++) {
		io_u = rbd->sort_events[i];

		if (fri_check_complete(rbd, io_u, events)) {
			this_events++;
			continue;
		}

		/*
		 * Stop waiting when we have enough, but continue checking
		 * all pending IOs if they are complete.
		 */
		if (*events >= min_evts)
			continue;

		rbd_io_u_wait_complete(io_u);

		if (fri_check_complete(rbd, io_u, events))
			this_events++;
	}

	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 = td->io_ops->data;
	struct fio_rbd_iou *fri = io_u->engine_data;
	int r = -1;

	fio_ro_check(td, io_u);

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

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

	if (io_u->ddir == DDIR_WRITE) {
		r = rbd_aio_write(rbd->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->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->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->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 = td->io_ops->data;

	if (rbd) {
		_fio_rbd_disconnect(rbd);
		free(rbd->aio_events);
		free(rbd->sort_events);
		free(rbd);
	}
}

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

	/* 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);
	if (r) {
		log_err("fio_setup_rbd_data failed.\n");
		goto cleanup;
	}
	td->io_ops->data = rbd;

	/* 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->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);
	return 0;

disconnect:
	_fio_rbd_disconnect(rbd);
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 = td->io_ops->data;

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

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

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

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

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

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

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

static void fio_exit fio_rbd_unregister(void)
{
	unregister_ioengine(&ioengine);
}
Commit	Line	Data
fc5c0345 DG	1	/*
	2	* rbd engine
	3	*
	4	* IO engine using Ceph's librbd to test RADOS Block Devices.
	5	*
	6	*/
	7
	8	#include <rbd/librbd.h>
	9
	10	#include "../fio.h"
	11
	12	struct fio_rbd_iou {
	13	struct io_u *io_u;
d8b64af2	14	rbd_completion_t completion;
d8b64af2	15	int io_seen;
b8ecbef6	16	int io_complete;
fc5c0345 DG	17	};
	18
	19	struct rbd_data {
	20	rados_t cluster;
	21	rados_ioctx_t io_ctx;
	22	rbd_image_t image;
	23	struct io_u **aio_events;
6f9961ac	24	struct io_u **sort_events;
fc5c0345 DG	25	};
	26
	27	struct rbd_options {
a1f871c7	28	void *pad;
fc5c0345 DG	29	char *rbd_name;
	30	char *pool_name;
	31	char *client_name;
d7d702c7	32	int busy_poll;
fc5c0345 DG	33	};
	34
	35	static struct fio_option options[] = {
	36	{
d8b64af2 JA	37	.name = "rbdname",
	38	.lname = "rbd engine rbdname",
	39	.type = FIO_OPT_STR_STORE,
	40	.help = "RBD name for RBD engine",
	41	.off1 = offsetof(struct rbd_options, rbd_name),
	42	.category = FIO_OPT_C_ENGINE,
	43	.group = FIO_OPT_G_RBD,
	44	},
fc5c0345	45	{
d7d702c7 JA	46	.name = "pool",
	47	.lname = "rbd engine pool",
	48	.type = FIO_OPT_STR_STORE,
	49	.help = "Name of the pool hosting the RBD for the RBD engine",
	50	.off1 = offsetof(struct rbd_options, pool_name),
	51	.category = FIO_OPT_C_ENGINE,
	52	.group = FIO_OPT_G_RBD,
d8b64af2	53	},
fc5c0345	54	{
d7d702c7 JA	55	.name = "clientname",
	56	.lname = "rbd engine clientname",
	57	.type = FIO_OPT_STR_STORE,
	58	.help = "Name of the ceph client to access the RBD for the RBD engine",
	59	.off1 = offsetof(struct rbd_options, client_name),
	60	.category = FIO_OPT_C_ENGINE,
	61	.group = FIO_OPT_G_RBD,
	62	},
	63	{
	64	.name = "busy_poll",
	65	.lname = "Busy poll",
	66	.type = FIO_OPT_BOOL,
	67	.help = "Busy poll for completions instead of sleeping",
fea585d4	68	.off1 = offsetof(struct rbd_options, busy_poll),
d7d702c7 JA	69	.def = "0",
	70	.category = FIO_OPT_C_ENGINE,
	71	.group = FIO_OPT_G_RBD,
d8b64af2	72	},
fc5c0345	73	{
d8b64af2 JA	74	.name = NULL,
d8b64af2 JA	75	},
fc5c0345 DG	76	};
	77
	78	static int _fio_setup_rbd_data(struct thread_data *td,
	79	struct rbd_data **rbd_data_ptr)
	80	{
6f9961ac	81	struct rbd_data *rbd;
fc5c0345 DG	82
	83	if (td->io_ops->data)
	84	return 0;
	85
6f9961ac JA	86	rbd = calloc(1, sizeof(struct rbd_data));
6f9961ac JA	87	if (!rbd)
fc5c0345 DG	88	goto failed;
fc5c0345 DG	89
6f9961ac JA	90	rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *));
6f9961ac JA	91	if (!rbd->aio_events)
fc5c0345 DG	92	goto failed;
fc5c0345 DG	93
6f9961ac JA	94	rbd->sort_events = calloc(td->o.iodepth, sizeof(struct io_u *));
	95	if (!rbd->sort_events)
	96	goto failed;
fc5c0345	97
6f9961ac	98	*rbd_data_ptr = rbd;
fc5c0345 DG	99	return 0;
	100
	101	failed:
6f9961ac JA	102	if (rbd)
6f9961ac JA	103	free(rbd);
fc5c0345 DG	104	return 1;
	105
	106	}
	107
	108	static int _fio_rbd_connect(struct thread_data *td)
	109	{
6f9961ac	110	struct rbd_data *rbd = td->io_ops->data;
fc5c0345 DG	111	struct rbd_options *o = td->eo;
	112	int r;
	113
6f9961ac	114	r = rados_create(&rbd->cluster, o->client_name);
fc5c0345 DG	115	if (r < 0) {
	116	log_err("rados_create failed.\n");
	117	goto failed_early;
	118	}
	119
6f9961ac	120	r = rados_conf_read_file(rbd->cluster, NULL);
fc5c0345 DG	121	if (r < 0) {
	122	log_err("rados_conf_read_file failed.\n");
	123	goto failed_early;
	124	}
	125
6f9961ac	126	r = rados_connect(rbd->cluster);
fc5c0345 DG	127	if (r < 0) {
	128	log_err("rados_connect failed.\n");
	129	goto failed_shutdown;
	130	}
	131
6f9961ac	132	r = rados_ioctx_create(rbd->cluster, o->pool_name, &rbd->io_ctx);
fc5c0345 DG	133	if (r < 0) {
	134	log_err("rados_ioctx_create failed.\n");
	135	goto failed_shutdown;
	136	}
	137
6f9961ac	138	r = rbd_open(rbd->io_ctx, o->rbd_name, &rbd->image, NULL /snap / );
fc5c0345 DG	139	if (r < 0) {
	140	log_err("rbd_open failed.\n");
	141	goto failed_open;
	142	}
	143	return 0;
	144
	145	failed_open:
6f9961ac JA	146	rados_ioctx_destroy(rbd->io_ctx);
6f9961ac JA	147	rbd->io_ctx = NULL;
fc5c0345	148	failed_shutdown:
6f9961ac JA	149	rados_shutdown(rbd->cluster);
6f9961ac JA	150	rbd->cluster = NULL;
fc5c0345 DG	151	failed_early:
	152	return 1;
	153	}
	154
6f9961ac	155	static void _fio_rbd_disconnect(struct rbd_data *rbd)
fc5c0345	156	{
6f9961ac	157	if (!rbd)
fc5c0345 DG	158	return;
	159
	160	/* shutdown everything */
6f9961ac JA	161	if (rbd->image) {
	162	rbd_close(rbd->image);
	163	rbd->image = NULL;
fc5c0345 DG	164	}
fc5c0345 DG	165
6f9961ac JA	166	if (rbd->io_ctx) {
	167	rados_ioctx_destroy(rbd->io_ctx);
	168	rbd->io_ctx = NULL;
fc5c0345 DG	169	}
fc5c0345 DG	170
6f9961ac JA	171	if (rbd->cluster) {
	172	rados_shutdown(rbd->cluster);
	173	rbd->cluster = NULL;
fc5c0345 DG	174	}
	175	}
	176
d8b64af2	177	static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
fc5c0345	178	{
dbf388d2 JA	179	struct fio_rbd_iou *fri = data;
dbf388d2 JA	180	struct io_u *io_u = fri->io_u;
d8b64af2	181	ssize_t ret;
fc5c0345	182
d8b64af2 JA	183	/*
	184	* Looks like return value is 0 for success, or < 0 for
	185	* a specific error. So we have to assume that it can't do
	186	* partial completions.
	187	*/
	188	ret = rbd_aio_get_return_value(fri->completion);
	189	if (ret < 0) {
	190	io_u->error = ret;
	191	io_u->resid = io_u->xfer_buflen;
	192	} else
	193	io_u->error = 0;
20cf5aab JD	194
20cf5aab JD	195	fri->io_complete = 1;
d8b64af2	196	}
fc5c0345	197
d8b64af2 JA	198	static struct io_u fio_rbd_event(struct thread_data td, int event)
d8b64af2 JA	199	{
6f9961ac	200	struct rbd_data *rbd = td->io_ops->data;
fc5c0345	201
6f9961ac	202	return rbd->aio_events[event];
fc5c0345 DG	203	}
fc5c0345 DG	204
6f9961ac	205	static inline int fri_check_complete(struct rbd_data rbd, struct io_u io_u,
d8b64af2	206	unsigned int *events)
fc5c0345	207	{
d8b64af2	208	struct fio_rbd_iou *fri = io_u->engine_data;
fc5c0345	209
b8ecbef6	210	if (fri->io_complete) {
d8b64af2	211	fri->io_seen = 1;
6f9961ac	212	rbd->aio_events[*events] = io_u;
d8b64af2	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
6f9961ac JA	222	static inline int rbd_io_u_seen(struct io_u *io_u)
	223	{
	224	struct fio_rbd_iou *fri = io_u->engine_data;
	225
	226	return fri->io_seen;
	227	}
	228
	229	static void rbd_io_u_wait_complete(struct io_u *io_u)
	230	{
	231	struct fio_rbd_iou *fri = io_u->engine_data;
	232
	233	rbd_aio_wait_for_complete(fri->completion);
	234	}
	235
	236	static int rbd_io_u_cmp(const void p1, const void p2)
	237	{
	238	const struct io_u a = (const struct io_u ) p1;
	239	const struct io_u b = (const struct io_u ) p2;
	240	uint64_t at, bt;
	241
	242	at = utime_since_now(&(*a)->start_time);
	243	bt = utime_since_now(&(*b)->start_time);
	244
	245	if (at < bt)
	246	return -1;
	247	else if (at == bt)
	248	return 0;
	249	else
	250	return 1;
	251	}
	252
d8b64af2 JA	253	static int rbd_iter_events(struct thread_data td, unsigned int events,
d8b64af2 JA	254	unsigned int min_evts, int wait)
82340a9f	255	{
6f9961ac	256	struct rbd_data *rbd = td->io_ops->data;
d8b64af2 JA	257	unsigned int this_events = 0;
d8b64af2 JA	258	struct io_u *io_u;
6f9961ac	259	int i, sidx;
82340a9f	260
6f9961ac	261	sidx = 0;
d8b64af2	262	io_u_qiter(&td->io_u_all, io_u, i) {
d8b64af2 JA	263	if (!(io_u->flags & IO_U_F_FLIGHT))
d8b64af2 JA	264	continue;
6f9961ac	265	if (rbd_io_u_seen(io_u))
d8b64af2	266	continue;
82340a9f	267
6f9961ac	268	if (fri_check_complete(rbd, io_u, events))
d8b64af2	269	this_events++;
6f9961ac JA	270	else if (wait)
	271	rbd->sort_events[sidx++] = io_u;
	272	}
82340a9f	273
6f9961ac JA	274	if (!wait \|\| !sidx)
	275	return this_events;
	276
	277	/*
	278	* Sort events, oldest issue first, then wait on as many as we
	279	* need in order of age. If we have enough events, stop waiting,
	280	* and just check if any of the older ones are done.
	281	*/
	282	if (sidx > 1)
	283	qsort(rbd->sort_events, sidx, sizeof(struct io_u *), rbd_io_u_cmp);
	284
	285	for (i = 0; i < sidx; i++) {
	286	io_u = rbd->sort_events[i];
	287
	288	if (fri_check_complete(rbd, io_u, events)) {
	289	this_events++;
	290	continue;
d8b64af2	291	}
6f9961ac JA	292
	293	/*
	294	* Stop waiting when we have enough, but continue checking
	295	* all pending IOs if they are complete.
	296	*/
d8b64af2	297	if (*events >= min_evts)
6f9961ac JA	298	continue;
	299
	300	rbd_io_u_wait_complete(io_u);
	301
	302	if (fri_check_complete(rbd, io_u, events))
	303	this_events++;
d8b64af2	304	}
fc5c0345	305
d8b64af2	306	return this_events;
fc5c0345 DG	307	}
	308
	309	static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
1f440ece	310	unsigned int max, const struct timespec *t)
fc5c0345	311	{
d8b64af2	312	unsigned int this_events, events = 0;
d7d702c7	313	struct rbd_options *o = td->eo;
d8b64af2	314	int wait = 0;
fc5c0345 DG	315
fc5c0345 DG	316	do {
d8b64af2	317	this_events = rbd_iter_events(td, &events, min, wait);
fc5c0345	318
d8b64af2	319	if (events >= min)
fc5c0345	320	break;
d8b64af2 JA	321	if (this_events)
d8b64af2 JA	322	continue;
fc5c0345	323
d7d702c7 JA	324	if (!o->busy_poll)
	325	wait = 1;
	326	else
	327	nop;
fc5c0345 DG	328	} while (1);
	329
	330	return events;
	331	}
	332
	333	static int fio_rbd_queue(struct thread_data td, struct io_u io_u)
	334	{
6f9961ac	335	struct rbd_data *rbd = td->io_ops->data;
d8b64af2 JA	336	struct fio_rbd_iou *fri = io_u->engine_data;
d8b64af2 JA	337	int r = -1;
fc5c0345 DG	338
	339	fio_ro_check(td, io_u);
	340
d8b64af2	341	fri->io_seen = 0;
b8ecbef6	342	fri->io_complete = 0;
d8b64af2	343
dbf388d2	344	r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb,
d8b64af2	345	&fri->completion);
dbf388d2 JA	346	if (r < 0) {
	347	log_err("rbd_aio_create_completion failed.\n");
	348	goto failed;
	349	}
fc5c0345	350
dbf388d2	351	if (io_u->ddir == DDIR_WRITE) {
6f9961ac JA	352	r = rbd_aio_write(rbd->image, io_u->offset, io_u->xfer_buflen,
6f9961ac JA	353	io_u->xfer_buf, fri->completion);
fc5c0345 DG	354	if (r < 0) {
fc5c0345 DG	355	log_err("rbd_aio_write failed.\n");
dbf388d2	356	goto failed_comp;
fc5c0345 DG	357	}
	358
	359	} else if (io_u->ddir == DDIR_READ) {
6f9961ac JA	360	r = rbd_aio_read(rbd->image, io_u->offset, io_u->xfer_buflen,
6f9961ac JA	361	io_u->xfer_buf, fri->completion);
fc5c0345 DG	362
	363	if (r < 0) {
	364	log_err("rbd_aio_read failed.\n");
dbf388d2	365	goto failed_comp;
fc5c0345	366	}
dbf388d2	367	} else if (io_u->ddir == DDIR_TRIM) {
6f9961ac JA	368	r = rbd_aio_discard(rbd->image, io_u->offset,
6f9961ac JA	369	io_u->xfer_buflen, fri->completion);
82340a9f	370	if (r < 0) {
dbf388d2 JA	371	log_err("rbd_aio_discard failed.\n");
dbf388d2 JA	372	goto failed_comp;
82340a9f	373	}
dbf388d2	374	} else if (io_u->ddir == DDIR_SYNC) {
6f9961ac	375	r = rbd_aio_flush(rbd->image, fri->completion);
fc5c0345 DG	376	if (r < 0) {
fc5c0345 DG	377	log_err("rbd_flush failed.\n");
dbf388d2	378	goto failed_comp;
fc5c0345	379	}
fc5c0345 DG	380	} else {
	381	dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
	382	io_u->ddir);
dbf388d2	383	goto failed_comp;
fc5c0345 DG	384	}
	385
	386	return FIO_Q_QUEUED;
dbf388d2 JA	387	failed_comp:
dbf388d2 JA	388	rbd_aio_release(fri->completion);
fc5c0345 DG	389	failed:
	390	io_u->error = r;
	391	td_verror(td, io_u->error, "xfer");
	392	return FIO_Q_COMPLETED;
	393	}
	394
	395	static int fio_rbd_init(struct thread_data *td)
	396	{
	397	int r;
	398
	399	r = _fio_rbd_connect(td);
	400	if (r) {
	401	log_err("fio_rbd_connect failed, return code: %d .\n", r);
	402	goto failed;
	403	}
	404
	405	return 0;
	406
	407	failed:
	408	return 1;
fc5c0345 DG	409	}
	410
	411	static void fio_rbd_cleanup(struct thread_data *td)
	412	{
6f9961ac	413	struct rbd_data *rbd = td->io_ops->data;
fc5c0345	414
6f9961ac JA	415	if (rbd) {
	416	_fio_rbd_disconnect(rbd);
	417	free(rbd->aio_events);
	418	free(rbd->sort_events);
	419	free(rbd);
fc5c0345	420	}
fc5c0345 DG	421	}
	422
	423	static int fio_rbd_setup(struct thread_data *td)
	424	{
fc5c0345 DG	425	rbd_image_info_t info;
fc5c0345 DG	426	struct fio_file *f;
6f9961ac	427	struct rbd_data *rbd = NULL;
fc5c0345	428	int major, minor, extra;
6f9961ac	429	int r;
fc5c0345 DG	430
	431	/* log version of librbd. No cluster connection required. */
	432	rbd_version(&major, &minor, &extra);
	433	log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra);
	434
	435	/* allocate engine specific structure to deal with librbd. */
6f9961ac	436	r = _fio_setup_rbd_data(td, &rbd);
fc5c0345 DG	437	if (r) {
	438	log_err("fio_setup_rbd_data failed.\n");
	439	goto cleanup;
	440	}
6f9961ac	441	td->io_ops->data = rbd;
fc5c0345	442
d8b64af2 JA	443	/* librbd does not allow us to run first in the main thread and later
	444	* in a fork child. It needs to be the same process context all the
	445	* time.
fc5c0345 DG	446	*/
	447	td->o.use_thread = 1;
	448
	449	/* connect in the main thread to determine to determine
	450	* the size of the given RADOS block device. And disconnect
	451	* later on.
	452	*/
	453	r = _fio_rbd_connect(td);
	454	if (r) {
	455	log_err("fio_rbd_connect failed.\n");
	456	goto cleanup;
	457	}
	458
	459	/* get size of the RADOS block device */
6f9961ac	460	r = rbd_stat(rbd->image, &info, sizeof(info));
fc5c0345 DG	461	if (r < 0) {
	462	log_err("rbd_status failed.\n");
	463	goto disconnect;
	464	}
	465	dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size);
	466
	467	/* taken from "net" engine. Pretend we deal with files,
	468	* even if we do not have any ideas about files.
	469	* The size of the RBD is set instead of a artificial file.
	470	*/
	471	if (!td->files_index) {
5903e7b7	472	add_file(td, td->o.filename ? : "rbd", 0, 0);
fc5c0345	473	td->o.nr_files = td->o.nr_files ? : 1;
b53f2c54	474	td->o.open_files++;
fc5c0345 DG	475	}
	476	f = td->files[0];
	477	f->real_file_size = info.size;
	478
	479	/* disconnect, then we were only connected to determine
	480	* the size of the RBD.
	481	*/
6f9961ac	482	_fio_rbd_disconnect(rbd);
fc5c0345 DG	483	return 0;
	484
	485	disconnect:
6f9961ac	486	_fio_rbd_disconnect(rbd);
fc5c0345 DG	487	cleanup:
	488	fio_rbd_cleanup(td);
	489	return r;
	490	}
	491
	492	static int fio_rbd_open(struct thread_data td, struct fio_file f)
	493	{
	494	return 0;
	495	}
	496
d9b100fc JA	497	static int fio_rbd_invalidate(struct thread_data td, struct fio_file f)
d9b100fc JA	498	{
903b2812	499	#if defined(CONFIG_RBD_INVAL)
6f9961ac	500	struct rbd_data *rbd = td->io_ops->data;
903b2812	501
6f9961ac	502	return rbd_invalidate_cache(rbd->image);
903b2812	503	#else
d9b100fc	504	return 0;
903b2812	505	#endif
d9b100fc JA	506	}
d9b100fc JA	507
fc5c0345 DG	508	static void fio_rbd_io_u_free(struct thread_data td, struct io_u io_u)
fc5c0345 DG	509	{
d8b64af2	510	struct fio_rbd_iou *fri = io_u->engine_data;
fc5c0345	511
d8b64af2	512	if (fri) {
fc5c0345	513	io_u->engine_data = NULL;
d8b64af2	514	free(fri);
fc5c0345 DG	515	}
	516	}
	517
	518	static int fio_rbd_io_u_init(struct thread_data td, struct io_u io_u)
	519	{
d8b64af2	520	struct fio_rbd_iou *fri;
fc5c0345	521
d8b64af2 JA	522	fri = calloc(1, sizeof(*fri));
	523	fri->io_u = io_u;
	524	io_u->engine_data = fri;
fc5c0345 DG	525	return 0;
	526	}
	527
10aa136b	528	static struct ioengine_ops ioengine = {
d9b100fc JA	529	.name = "rbd",
	530	.version = FIO_IOOPS_VERSION,
	531	.setup = fio_rbd_setup,
	532	.init = fio_rbd_init,
	533	.queue = fio_rbd_queue,
	534	.getevents = fio_rbd_getevents,
	535	.event = fio_rbd_event,
	536	.cleanup = fio_rbd_cleanup,
	537	.open_file = fio_rbd_open,
	538	.invalidate = fio_rbd_invalidate,
	539	.options = options,
	540	.io_u_init = fio_rbd_io_u_init,
	541	.io_u_free = fio_rbd_io_u_free,
	542	.option_struct_size = sizeof(struct rbd_options),
fc5c0345 DG	543	};
	544
	545	static void fio_init fio_rbd_register(void)
	546	{
	547	register_ioengine(&ioengine);
	548	}
	549
	550	static void fio_exit fio_rbd_unregister(void)
	551	{
	552	unregister_ioengine(&ioengine);
	553	}