[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"
#include "../optgroup.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 *cluster_name;
	char *rbd_name;
	char *pool_name;
	char *client_name;
	int busy_poll;
};

static struct fio_option options[] = {
        {
		.name		= "clustername",
		.lname		= "ceph cluster name",
		.type		= FIO_OPT_STR_STORE,
		.help		= "Cluster name for ceph",
		.off1		= offsetof(struct rbd_options, cluster_name),
		.category	= FIO_OPT_C_ENGINE,
		.group		= FIO_OPT_G_RBD,
        },
	{
		.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;

	if (o->cluster_name) {
		char *client_name = NULL; 

		/*
		 * If we specify cluser name, the rados_creat2
		 * will not assume 'client.'. name is considered
		 * as a full type.id namestr
		 */
		if (!index(o->client_name, '.')) {
			client_name = calloc(1, strlen("client.") +
						strlen(o->client_name) + 1);
			strcat(client_name, "client.");
			o->client_name = strcat(client_name, o->client_name);
		}
		r = rados_create2(&rbd->cluster, o->cluster_name,
					o->client_name, 0);
	} else
		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"
d220c761	11	#include "../optgroup.h"
fc5c0345 DG	12
	13	struct fio_rbd_iou {
	14	struct io_u *io_u;
d8b64af2	15	rbd_completion_t completion;
d8b64af2	16	int io_seen;
b8ecbef6	17	int io_complete;
fc5c0345 DG	18	};
	19
	20	struct rbd_data {
	21	rados_t cluster;
	22	rados_ioctx_t io_ctx;
	23	rbd_image_t image;
	24	struct io_u **aio_events;
6f9961ac	25	struct io_u **sort_events;
fc5c0345 DG	26	};
	27
	28	struct rbd_options {
a1f871c7	29	void *pad;
6e20c6e7	30	char *cluster_name;
fc5c0345 DG	31	char *rbd_name;
	32	char *pool_name;
	33	char *client_name;
d7d702c7	34	int busy_poll;
fc5c0345 DG	35	};
	36
	37	static struct fio_option options[] = {
6e20c6e7 T	38	{
	39	.name = "clustername",
	40	.lname = "ceph cluster name",
	41	.type = FIO_OPT_STR_STORE,
	42	.help = "Cluster name for ceph",
	43	.off1 = offsetof(struct rbd_options, cluster_name),
	44	.category = FIO_OPT_C_ENGINE,
	45	.group = FIO_OPT_G_RBD,
	46	},
fc5c0345	47	{
d8b64af2 JA	48	.name = "rbdname",
	49	.lname = "rbd engine rbdname",
	50	.type = FIO_OPT_STR_STORE,
	51	.help = "RBD name for RBD engine",
	52	.off1 = offsetof(struct rbd_options, rbd_name),
	53	.category = FIO_OPT_C_ENGINE,
	54	.group = FIO_OPT_G_RBD,
	55	},
fc5c0345	56	{
d7d702c7 JA	57	.name = "pool",
	58	.lname = "rbd engine pool",
	59	.type = FIO_OPT_STR_STORE,
	60	.help = "Name of the pool hosting the RBD for the RBD engine",
	61	.off1 = offsetof(struct rbd_options, pool_name),
	62	.category = FIO_OPT_C_ENGINE,
	63	.group = FIO_OPT_G_RBD,
d8b64af2	64	},
fc5c0345	65	{
d7d702c7 JA	66	.name = "clientname",
	67	.lname = "rbd engine clientname",
	68	.type = FIO_OPT_STR_STORE,
	69	.help = "Name of the ceph client to access the RBD for the RBD engine",
	70	.off1 = offsetof(struct rbd_options, client_name),
	71	.category = FIO_OPT_C_ENGINE,
	72	.group = FIO_OPT_G_RBD,
	73	},
	74	{
	75	.name = "busy_poll",
	76	.lname = "Busy poll",
	77	.type = FIO_OPT_BOOL,
	78	.help = "Busy poll for completions instead of sleeping",
fea585d4	79	.off1 = offsetof(struct rbd_options, busy_poll),
d7d702c7 JA	80	.def = "0",
	81	.category = FIO_OPT_C_ENGINE,
	82	.group = FIO_OPT_G_RBD,
d8b64af2	83	},
fc5c0345	84	{
d8b64af2 JA	85	.name = NULL,
d8b64af2 JA	86	},
fc5c0345 DG	87	};
	88
	89	static int _fio_setup_rbd_data(struct thread_data *td,
	90	struct rbd_data **rbd_data_ptr)
	91	{
6f9961ac	92	struct rbd_data *rbd;
fc5c0345 DG	93
	94	if (td->io_ops->data)
	95	return 0;
	96
6f9961ac JA	97	rbd = calloc(1, sizeof(struct rbd_data));
6f9961ac JA	98	if (!rbd)
fc5c0345 DG	99	goto failed;
fc5c0345 DG	100
6f9961ac JA	101	rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *));
6f9961ac JA	102	if (!rbd->aio_events)
fc5c0345 DG	103	goto failed;
fc5c0345 DG	104
6f9961ac JA	105	rbd->sort_events = calloc(td->o.iodepth, sizeof(struct io_u *));
	106	if (!rbd->sort_events)
	107	goto failed;
fc5c0345	108
6f9961ac	109	*rbd_data_ptr = rbd;
fc5c0345 DG	110	return 0;
	111
	112	failed:
6f9961ac JA	113	if (rbd)
6f9961ac JA	114	free(rbd);
fc5c0345 DG	115	return 1;
	116
	117	}
	118
	119	static int _fio_rbd_connect(struct thread_data *td)
	120	{
6f9961ac	121	struct rbd_data *rbd = td->io_ops->data;
fc5c0345 DG	122	struct rbd_options *o = td->eo;
	123	int r;
	124
6e20c6e7 T	125	if (o->cluster_name) {
	126	char *client_name = NULL;
	127
	128	/*
	129	* If we specify cluser name, the rados_creat2
	130	* will not assume 'client.'. name is considered
	131	* as a full type.id namestr
	132	*/
	133	if (!index(o->client_name, '.')) {
	134	client_name = calloc(1, strlen("client.") +
	135	strlen(o->client_name) + 1);
	136	strcat(client_name, "client.");
	137	o->client_name = strcat(client_name, o->client_name);
	138	}
	139	r = rados_create2(&rbd->cluster, o->cluster_name,
	140	o->client_name, 0);
	141	} else
	142	r = rados_create(&rbd->cluster, o->client_name);
	143
fc5c0345 DG	144	if (r < 0) {
	145	log_err("rados_create failed.\n");
	146	goto failed_early;
	147	}
	148
6f9961ac	149	r = rados_conf_read_file(rbd->cluster, NULL);
fc5c0345 DG	150	if (r < 0) {
	151	log_err("rados_conf_read_file failed.\n");
	152	goto failed_early;
	153	}
	154
6f9961ac	155	r = rados_connect(rbd->cluster);
fc5c0345 DG	156	if (r < 0) {
	157	log_err("rados_connect failed.\n");
	158	goto failed_shutdown;
	159	}
	160
6f9961ac	161	r = rados_ioctx_create(rbd->cluster, o->pool_name, &rbd->io_ctx);
fc5c0345 DG	162	if (r < 0) {
	163	log_err("rados_ioctx_create failed.\n");
	164	goto failed_shutdown;
	165	}
	166
6f9961ac	167	r = rbd_open(rbd->io_ctx, o->rbd_name, &rbd->image, NULL /snap / );
fc5c0345 DG	168	if (r < 0) {
	169	log_err("rbd_open failed.\n");
	170	goto failed_open;
	171	}
	172	return 0;
	173
	174	failed_open:
6f9961ac JA	175	rados_ioctx_destroy(rbd->io_ctx);
6f9961ac JA	176	rbd->io_ctx = NULL;
fc5c0345	177	failed_shutdown:
6f9961ac JA	178	rados_shutdown(rbd->cluster);
6f9961ac JA	179	rbd->cluster = NULL;
fc5c0345 DG	180	failed_early:
	181	return 1;
	182	}
	183
6f9961ac	184	static void _fio_rbd_disconnect(struct rbd_data *rbd)
fc5c0345	185	{
6f9961ac	186	if (!rbd)
fc5c0345 DG	187	return;
	188
	189	/* shutdown everything */
6f9961ac JA	190	if (rbd->image) {
	191	rbd_close(rbd->image);
	192	rbd->image = NULL;
fc5c0345 DG	193	}
fc5c0345 DG	194
6f9961ac JA	195	if (rbd->io_ctx) {
	196	rados_ioctx_destroy(rbd->io_ctx);
	197	rbd->io_ctx = NULL;
fc5c0345 DG	198	}
fc5c0345 DG	199
6f9961ac JA	200	if (rbd->cluster) {
	201	rados_shutdown(rbd->cluster);
	202	rbd->cluster = NULL;
fc5c0345 DG	203	}
	204	}
	205
d8b64af2	206	static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
fc5c0345	207	{
dbf388d2 JA	208	struct fio_rbd_iou *fri = data;
dbf388d2 JA	209	struct io_u *io_u = fri->io_u;
d8b64af2	210	ssize_t ret;
fc5c0345	211
d8b64af2 JA	212	/*
	213	* Looks like return value is 0 for success, or < 0 for
	214	* a specific error. So we have to assume that it can't do
	215	* partial completions.
	216	*/
	217	ret = rbd_aio_get_return_value(fri->completion);
	218	if (ret < 0) {
	219	io_u->error = ret;
	220	io_u->resid = io_u->xfer_buflen;
	221	} else
	222	io_u->error = 0;
20cf5aab JD	223
20cf5aab JD	224	fri->io_complete = 1;
d8b64af2	225	}
fc5c0345	226
d8b64af2 JA	227	static struct io_u fio_rbd_event(struct thread_data td, int event)
d8b64af2 JA	228	{
6f9961ac	229	struct rbd_data *rbd = td->io_ops->data;
fc5c0345	230
6f9961ac	231	return rbd->aio_events[event];
fc5c0345 DG	232	}
fc5c0345 DG	233
6f9961ac	234	static inline int fri_check_complete(struct rbd_data rbd, struct io_u io_u,
d8b64af2	235	unsigned int *events)
fc5c0345	236	{
d8b64af2	237	struct fio_rbd_iou *fri = io_u->engine_data;
fc5c0345	238
b8ecbef6	239	if (fri->io_complete) {
d8b64af2	240	fri->io_seen = 1;
6f9961ac	241	rbd->aio_events[*events] = io_u;
d8b64af2	242	(*events)++;
fc5c0345	243
d8b64af2 JA	244	rbd_aio_release(fri->completion);
	245	return 1;
	246	}
fc5c0345	247
d8b64af2	248	return 0;
fc5c0345 DG	249	}
fc5c0345 DG	250
6f9961ac JA	251	static inline int rbd_io_u_seen(struct io_u *io_u)
	252	{
	253	struct fio_rbd_iou *fri = io_u->engine_data;
	254
	255	return fri->io_seen;
	256	}
	257
	258	static void rbd_io_u_wait_complete(struct io_u *io_u)
	259	{
	260	struct fio_rbd_iou *fri = io_u->engine_data;
	261
	262	rbd_aio_wait_for_complete(fri->completion);
	263	}
	264
	265	static int rbd_io_u_cmp(const void p1, const void p2)
	266	{
	267	const struct io_u a = (const struct io_u ) p1;
	268	const struct io_u b = (const struct io_u ) p2;
	269	uint64_t at, bt;
	270
	271	at = utime_since_now(&(*a)->start_time);
	272	bt = utime_since_now(&(*b)->start_time);
	273
	274	if (at < bt)
	275	return -1;
	276	else if (at == bt)
	277	return 0;
	278	else
	279	return 1;
	280	}
	281
d8b64af2 JA	282	static int rbd_iter_events(struct thread_data td, unsigned int events,
d8b64af2 JA	283	unsigned int min_evts, int wait)
82340a9f	284	{
6f9961ac	285	struct rbd_data *rbd = td->io_ops->data;
d8b64af2 JA	286	unsigned int this_events = 0;
d8b64af2 JA	287	struct io_u *io_u;
6f9961ac	288	int i, sidx;
82340a9f	289
6f9961ac	290	sidx = 0;
d8b64af2	291	io_u_qiter(&td->io_u_all, io_u, i) {
d8b64af2 JA	292	if (!(io_u->flags & IO_U_F_FLIGHT))
d8b64af2 JA	293	continue;
6f9961ac	294	if (rbd_io_u_seen(io_u))
d8b64af2	295	continue;
82340a9f	296
6f9961ac	297	if (fri_check_complete(rbd, io_u, events))
d8b64af2	298	this_events++;
6f9961ac JA	299	else if (wait)
	300	rbd->sort_events[sidx++] = io_u;
	301	}
82340a9f	302
6f9961ac JA	303	if (!wait \|\| !sidx)
	304	return this_events;
	305
	306	/*
	307	* Sort events, oldest issue first, then wait on as many as we
	308	* need in order of age. If we have enough events, stop waiting,
	309	* and just check if any of the older ones are done.
	310	*/
	311	if (sidx > 1)
	312	qsort(rbd->sort_events, sidx, sizeof(struct io_u *), rbd_io_u_cmp);
	313
	314	for (i = 0; i < sidx; i++) {
	315	io_u = rbd->sort_events[i];
	316
	317	if (fri_check_complete(rbd, io_u, events)) {
	318	this_events++;
	319	continue;
d8b64af2	320	}
6f9961ac JA	321
	322	/*
	323	* Stop waiting when we have enough, but continue checking
	324	* all pending IOs if they are complete.
	325	*/
d8b64af2	326	if (*events >= min_evts)
6f9961ac JA	327	continue;
	328
	329	rbd_io_u_wait_complete(io_u);
	330
	331	if (fri_check_complete(rbd, io_u, events))
	332	this_events++;
d8b64af2	333	}
fc5c0345	334
d8b64af2	335	return this_events;
fc5c0345 DG	336	}
	337
	338	static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
1f440ece	339	unsigned int max, const struct timespec *t)
fc5c0345	340	{
d8b64af2	341	unsigned int this_events, events = 0;
d7d702c7	342	struct rbd_options *o = td->eo;
d8b64af2	343	int wait = 0;
fc5c0345 DG	344
fc5c0345 DG	345	do {
d8b64af2	346	this_events = rbd_iter_events(td, &events, min, wait);
fc5c0345	347
d8b64af2	348	if (events >= min)
fc5c0345	349	break;
d8b64af2 JA	350	if (this_events)
d8b64af2 JA	351	continue;
fc5c0345	352
d7d702c7 JA	353	if (!o->busy_poll)
	354	wait = 1;
	355	else
	356	nop;
fc5c0345 DG	357	} while (1);
	358
	359	return events;
	360	}
	361
	362	static int fio_rbd_queue(struct thread_data td, struct io_u io_u)
	363	{
6f9961ac	364	struct rbd_data *rbd = td->io_ops->data;
d8b64af2 JA	365	struct fio_rbd_iou *fri = io_u->engine_data;
d8b64af2 JA	366	int r = -1;
fc5c0345 DG	367
	368	fio_ro_check(td, io_u);
	369
d8b64af2	370	fri->io_seen = 0;
b8ecbef6	371	fri->io_complete = 0;
d8b64af2	372
dbf388d2	373	r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb,
d8b64af2	374	&fri->completion);
dbf388d2 JA	375	if (r < 0) {
	376	log_err("rbd_aio_create_completion failed.\n");
	377	goto failed;
	378	}
fc5c0345	379
dbf388d2	380	if (io_u->ddir == DDIR_WRITE) {
6f9961ac JA	381	r = rbd_aio_write(rbd->image, io_u->offset, io_u->xfer_buflen,
6f9961ac JA	382	io_u->xfer_buf, fri->completion);
fc5c0345 DG	383	if (r < 0) {
fc5c0345 DG	384	log_err("rbd_aio_write failed.\n");
dbf388d2	385	goto failed_comp;
fc5c0345 DG	386	}
	387
	388	} else if (io_u->ddir == DDIR_READ) {
6f9961ac JA	389	r = rbd_aio_read(rbd->image, io_u->offset, io_u->xfer_buflen,
6f9961ac JA	390	io_u->xfer_buf, fri->completion);
fc5c0345 DG	391
	392	if (r < 0) {
	393	log_err("rbd_aio_read failed.\n");
dbf388d2	394	goto failed_comp;
fc5c0345	395	}
dbf388d2	396	} else if (io_u->ddir == DDIR_TRIM) {
6f9961ac JA	397	r = rbd_aio_discard(rbd->image, io_u->offset,
6f9961ac JA	398	io_u->xfer_buflen, fri->completion);
82340a9f	399	if (r < 0) {
dbf388d2 JA	400	log_err("rbd_aio_discard failed.\n");
dbf388d2 JA	401	goto failed_comp;
82340a9f	402	}
dbf388d2	403	} else if (io_u->ddir == DDIR_SYNC) {
6f9961ac	404	r = rbd_aio_flush(rbd->image, fri->completion);
fc5c0345 DG	405	if (r < 0) {
fc5c0345 DG	406	log_err("rbd_flush failed.\n");
dbf388d2	407	goto failed_comp;
fc5c0345	408	}
fc5c0345 DG	409	} else {
	410	dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
	411	io_u->ddir);
dbf388d2	412	goto failed_comp;
fc5c0345 DG	413	}
	414
	415	return FIO_Q_QUEUED;
dbf388d2 JA	416	failed_comp:
dbf388d2 JA	417	rbd_aio_release(fri->completion);
fc5c0345 DG	418	failed:
	419	io_u->error = r;
	420	td_verror(td, io_u->error, "xfer");
	421	return FIO_Q_COMPLETED;
	422	}
	423
	424	static int fio_rbd_init(struct thread_data *td)
	425	{
	426	int r;
	427
	428	r = _fio_rbd_connect(td);
	429	if (r) {
	430	log_err("fio_rbd_connect failed, return code: %d .\n", r);
	431	goto failed;
	432	}
	433
	434	return 0;
	435
	436	failed:
	437	return 1;
fc5c0345 DG	438	}
	439
	440	static void fio_rbd_cleanup(struct thread_data *td)
	441	{
6f9961ac	442	struct rbd_data *rbd = td->io_ops->data;
fc5c0345	443
6f9961ac JA	444	if (rbd) {
	445	_fio_rbd_disconnect(rbd);
	446	free(rbd->aio_events);
	447	free(rbd->sort_events);
	448	free(rbd);
fc5c0345	449	}
fc5c0345 DG	450	}
	451
	452	static int fio_rbd_setup(struct thread_data *td)
	453	{
fc5c0345 DG	454	rbd_image_info_t info;
fc5c0345 DG	455	struct fio_file *f;
6f9961ac	456	struct rbd_data *rbd = NULL;
fc5c0345	457	int major, minor, extra;
6f9961ac	458	int r;
fc5c0345 DG	459
	460	/* log version of librbd. No cluster connection required. */
	461	rbd_version(&major, &minor, &extra);
	462	log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra);
	463
	464	/* allocate engine specific structure to deal with librbd. */
6f9961ac	465	r = _fio_setup_rbd_data(td, &rbd);
fc5c0345 DG	466	if (r) {
	467	log_err("fio_setup_rbd_data failed.\n");
	468	goto cleanup;
	469	}
6f9961ac	470	td->io_ops->data = rbd;
fc5c0345	471
d8b64af2 JA	472	/* librbd does not allow us to run first in the main thread and later
	473	* in a fork child. It needs to be the same process context all the
	474	* time.
fc5c0345 DG	475	*/
	476	td->o.use_thread = 1;
	477
	478	/* connect in the main thread to determine to determine
	479	* the size of the given RADOS block device. And disconnect
	480	* later on.
	481	*/
	482	r = _fio_rbd_connect(td);
	483	if (r) {
	484	log_err("fio_rbd_connect failed.\n");
	485	goto cleanup;
	486	}
	487
	488	/* get size of the RADOS block device */
6f9961ac	489	r = rbd_stat(rbd->image, &info, sizeof(info));
fc5c0345 DG	490	if (r < 0) {
	491	log_err("rbd_status failed.\n");
	492	goto disconnect;
	493	}
	494	dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size);
	495
	496	/* taken from "net" engine. Pretend we deal with files,
	497	* even if we do not have any ideas about files.
	498	* The size of the RBD is set instead of a artificial file.
	499	*/
	500	if (!td->files_index) {
5903e7b7	501	add_file(td, td->o.filename ? : "rbd", 0, 0);
fc5c0345	502	td->o.nr_files = td->o.nr_files ? : 1;
b53f2c54	503	td->o.open_files++;
fc5c0345 DG	504	}
	505	f = td->files[0];
	506	f->real_file_size = info.size;
	507
	508	/* disconnect, then we were only connected to determine
	509	* the size of the RBD.
	510	*/
6f9961ac	511	_fio_rbd_disconnect(rbd);
fc5c0345 DG	512	return 0;
	513
	514	disconnect:
6f9961ac	515	_fio_rbd_disconnect(rbd);
fc5c0345 DG	516	cleanup:
	517	fio_rbd_cleanup(td);
	518	return r;
	519	}
	520
	521	static int fio_rbd_open(struct thread_data td, struct fio_file f)
	522	{
	523	return 0;
	524	}
	525
d9b100fc JA	526	static int fio_rbd_invalidate(struct thread_data td, struct fio_file f)
d9b100fc JA	527	{
903b2812	528	#if defined(CONFIG_RBD_INVAL)
6f9961ac	529	struct rbd_data *rbd = td->io_ops->data;
903b2812	530
6f9961ac	531	return rbd_invalidate_cache(rbd->image);
903b2812	532	#else
d9b100fc	533	return 0;
903b2812	534	#endif
d9b100fc JA	535	}
d9b100fc JA	536
fc5c0345 DG	537	static void fio_rbd_io_u_free(struct thread_data td, struct io_u io_u)
fc5c0345 DG	538	{
d8b64af2	539	struct fio_rbd_iou *fri = io_u->engine_data;
fc5c0345	540
d8b64af2	541	if (fri) {
fc5c0345	542	io_u->engine_data = NULL;
d8b64af2	543	free(fri);
fc5c0345 DG	544	}
	545	}
	546
	547	static int fio_rbd_io_u_init(struct thread_data td, struct io_u io_u)
	548	{
d8b64af2	549	struct fio_rbd_iou *fri;
fc5c0345	550
d8b64af2 JA	551	fri = calloc(1, sizeof(*fri));
	552	fri->io_u = io_u;
	553	io_u->engine_data = fri;
fc5c0345 DG	554	return 0;
	555	}
	556
10aa136b	557	static struct ioengine_ops ioengine = {
d9b100fc JA	558	.name = "rbd",
	559	.version = FIO_IOOPS_VERSION,
	560	.setup = fio_rbd_setup,
	561	.init = fio_rbd_init,
	562	.queue = fio_rbd_queue,
	563	.getevents = fio_rbd_getevents,
	564	.event = fio_rbd_event,
	565	.cleanup = fio_rbd_cleanup,
	566	.open_file = fio_rbd_open,
	567	.invalidate = fio_rbd_invalidate,
	568	.options = options,
	569	.io_u_init = fio_rbd_io_u_init,
	570	.io_u_free = fio_rbd_io_u_free,
	571	.option_struct_size = sizeof(struct rbd_options),
fc5c0345 DG	572	};
	573
	574	static void fio_init fio_rbd_register(void)
	575	{
	576	register_ioengine(&ioengine);
	577	}
	578
	579	static void fio_exit fio_rbd_unregister(void)
	580	{
	581	unregister_ioengine(&ioengine);
	582	}