X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=engines%2Frbd.c;h=aa50c801c3c831eec13a5c212fdc836c7f85e675;hp=e4155da2f97194c961ada6d42563bf182836a9ff;hb=73ae55c3c87e0965b8d8f1daf5baac14ca316654;hpb=903b2812e9c772d30677b6768e6e00d6d16bc801

diff --git a/engines/rbd.c b/engines/rbd.c
index e4155da2..aa50c801 100644
--- a/engines/rbd.c
+++ b/engines/rbd.c
@@ -8,12 +8,25 @@
 #include <rbd/librbd.h>
 
 #include "../fio.h"
+#include "../optgroup.h"
+#ifdef CONFIG_RBD_BLKIN
+#include <zipkin_c.h>
+#endif
+
+#ifdef CONFIG_RBD_POLL
+/* add for poll */
+#include <poll.h>
+#include <sys/eventfd.h>
+#endif
 
 struct fio_rbd_iou {
 	struct io_u *io_u;
 	rbd_completion_t completion;
-	int io_complete;
 	int io_seen;
+	int io_complete;
+#ifdef CONFIG_RBD_BLKIN
+	struct blkin_trace_info info;
+#endif
 };
 
 struct rbd_data {
@@ -21,16 +34,29 @@ struct rbd_data {
 	rados_ioctx_t io_ctx;
 	rbd_image_t image;
 	struct io_u **aio_events;
+	struct io_u **sort_events;
+	int fd; /* add for poll */
 };
 
 struct rbd_options {
-	struct thread_data *td;
+	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",
@@ -41,22 +67,32 @@ static struct fio_option options[] = {
 		.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		= "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		= "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,
@@ -66,113 +102,184 @@ static struct fio_option options[] = {
 static int _fio_setup_rbd_data(struct thread_data *td,
 			       struct rbd_data **rbd_data_ptr)
 {
-	struct rbd_data *rbd_data;
+	struct rbd_data *rbd;
 
-	if (td->io_ops->data)
+	if (td->io_ops_data)
 		return 0;
 
-	rbd_data = malloc(sizeof(struct rbd_data));
-	if (!rbd_data)
+	rbd = calloc(1, sizeof(struct rbd_data));
+	if (!rbd)
 		goto failed;
 
-	memset(rbd_data, 0, sizeof(struct rbd_data));
+	/* add for poll, init fd: -1 */
+	rbd->fd = -1;
 
-	rbd_data->aio_events = malloc(td->o.iodepth * sizeof(struct io_u *));
-	if (!rbd_data->aio_events)
+	rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *));
+	if (!rbd->aio_events)
 		goto failed;
 
-	memset(rbd_data->aio_events, 0, td->o.iodepth * sizeof(struct io_u *));
-
-	*rbd_data_ptr = rbd_data;
+	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_data)
-		free(rbd_data);
+	if (rbd) {
+		if (rbd->aio_events) 
+			free(rbd->aio_events);
+		if (rbd->sort_events)
+			free(rbd->sort_events);
+		free(rbd);
+	}
 	return 1;
 
 }
 
+#ifdef CONFIG_RBD_POLL
+static bool _fio_rbd_setup_poll(struct rbd_data *rbd)
+{
+	int r;
+
+	/* add for rbd poll */
+	rbd->fd = eventfd(0, EFD_NONBLOCK);
+	if (rbd->fd < 0) {
+		log_err("eventfd failed.\n");
+		return false;
+	}
+
+	r = rbd_set_image_notification(rbd->image, rbd->fd, EVENT_TYPE_EVENTFD);
+	if (r < 0) {
+		log_err("rbd_set_image_notification failed.\n");
+		close(rbd->fd);
+		rbd->fd = -1;
+		return false;
+	}
+
+	return true;
+}
+#else
+static bool _fio_rbd_setup_poll(struct rbd_data *rbd)
+{
+	return true;
+}
+#endif
+
 static int _fio_rbd_connect(struct thread_data *td)
 {
-	struct rbd_data *rbd_data = td->io_ops->data;
+	struct rbd_data *rbd = td->io_ops_data;
 	struct rbd_options *o = td->eo;
 	int r;
 
-	r = rados_create(&(rbd_data->cluster), o->client_name);
+	if (o->cluster_name) {
+		char *client_name = NULL; 
+
+		/*
+		 * If we specify cluser name, the rados_create2
+		 * will not assume 'client.'. name is considered
+		 * as a full type.id namestr
+		 */
+		if (o->client_name) {
+			if (!index(o->client_name, '.')) {
+				client_name = calloc(1, strlen("client.") +
+						    strlen(o->client_name) + 1);
+				strcat(client_name, "client.");
+				strcat(client_name, o->client_name);
+			} else {
+				client_name = o->client_name;
+			}
+		}
+
+		r = rados_create2(&rbd->cluster, o->cluster_name,
+				 client_name, 0);
+
+		if (client_name && !index(o->client_name, '.'))
+			free(client_name);
+	} 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_data->cluster, NULL);
+	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_data->cluster);
+	r = rados_connect(rbd->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));
+	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_data->io_ctx, o->rbd_name, &(rbd_data->image),
-		     NULL /*snap */ );
+	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;
 	}
+
+	if (!_fio_rbd_setup_poll(rbd))
+		goto failed_poll;
+
 	return 0;
 
+failed_poll:
+	rbd_close(rbd->image);
+	rbd->image = NULL;
 failed_open:
-	rados_ioctx_destroy(rbd_data->io_ctx);
-	rbd_data->io_ctx = NULL;
+	rados_ioctx_destroy(rbd->io_ctx);
+	rbd->io_ctx = NULL;
 failed_shutdown:
-	rados_shutdown(rbd_data->cluster);
-	rbd_data->cluster = NULL;
+	rados_shutdown(rbd->cluster);
+	rbd->cluster = NULL;
 failed_early:
 	return 1;
 }
 
-static void _fio_rbd_disconnect(struct rbd_data *rbd_data)
+static void _fio_rbd_disconnect(struct rbd_data *rbd)
 {
-	if (!rbd_data)
+	if (!rbd)
 		return;
 
+	/* close eventfd */
+	if (rbd->fd != -1) {
+		close(rbd->fd);
+		rbd->fd = -1;
+	}
+
 	/* shutdown everything */
-	if (rbd_data->image) {
-		rbd_close(rbd_data->image);
-		rbd_data->image = NULL;
+	if (rbd->image) {
+		rbd_close(rbd->image);
+		rbd->image = NULL;
 	}
 
-	if (rbd_data->io_ctx) {
-		rados_ioctx_destroy(rbd_data->io_ctx);
-		rbd_data->io_ctx = NULL;
+	if (rbd->io_ctx) {
+		rados_ioctx_destroy(rbd->io_ctx);
+		rbd->io_ctx = NULL;
 	}
 
-	if (rbd_data->cluster) {
-		rados_shutdown(rbd_data->cluster);
-		rbd_data->cluster = NULL;
+	if (rbd->cluster) {
+		rados_shutdown(rbd->cluster);
+		rbd->cluster = NULL;
 	}
 }
 
 static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
 {
-	struct io_u *io_u = data;
-	struct fio_rbd_iou *fri = io_u->engine_data;
+	struct fio_rbd_iou *fri = data;
+	struct io_u *io_u = fri->io_u;
 	ssize_t ret;
 
-	fri->io_complete = 1;
-
 	/*
 	 * Looks like return value is 0 for success, or < 0 for
 	 * a specific error. So we have to assume that it can't do
@@ -184,25 +291,25 @@ static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
 		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_data = td->io_ops->data;
+	struct rbd_data *rbd = td->io_ops_data;
 
-	return rbd_data->aio_events[event];
+	return rbd->aio_events[event];
 }
 
-static inline int fri_check_complete(struct rbd_data *rbd_data,
-				     struct io_u *io_u,
+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_complete = 0;
 		fri->io_seen = 1;
-		rbd_data->aio_events[*events] = io_u;
+		rbd->aio_events[*events] = io_u;
 		(*events)++;
 
 		rbd_aio_release(fri->completion);
@@ -212,32 +319,110 @@ static inline int fri_check_complete(struct rbd_data *rbd_data,
 	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_data = td->io_ops->data;
+	struct rbd_data *rbd = td->io_ops_data;
 	unsigned int this_events = 0;
 	struct io_u *io_u;
-	int i;
+	int i, sidx = 0;
 
-	io_u_qiter(&td->io_u_all, io_u, i) {
-		struct fio_rbd_iou *fri = io_u->engine_data;
+#ifdef CONFIG_RBD_POLL
+	int ret = 0;
+	int event_num = 0;
+	struct fio_rbd_iou *fri = NULL;
+	rbd_completion_t comps[min_evts];
+
+	struct pollfd pfd;
+	pfd.fd = rbd->fd;
+	pfd.events = POLLIN;
+
+	ret = poll(&pfd, 1, -1);
+	if (ret <= 0)
+		return 0;
 
+	assert(pfd.revents & POLLIN);
+
+	event_num = rbd_poll_io_events(rbd->image, comps, min_evts);
+
+	for (i = 0; i < event_num; i++) {
+		fri = rbd_aio_get_arg(comps[i]);
+		io_u = fri->io_u;
+#else
+	io_u_qiter(&td->io_u_all, io_u, i) {
+#endif
 		if (!(io_u->flags & IO_U_F_FLIGHT))
 			continue;
-		if (fri->io_seen)
+		if (rbd_io_u_seen(io_u))
 			continue;
 
-		if (fri_check_complete(rbd_data, io_u, events))
+		if (fri_check_complete(rbd, io_u, events))
 			this_events++;
-		else if (wait) {
-			rbd_aio_wait_for_complete(fri->completion);
+		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);
 
-			if (fri_check_complete(rbd_data, io_u, events))
-				this_events++;
+	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)
-			break;
+			continue;
+
+		rbd_io_u_wait_complete(io_u);
+
+		if (fri_check_complete(rbd, io_u, events))
+			this_events++;
 	}
 
 	return this_events;
@@ -247,6 +432,7 @@ 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 {
@@ -257,7 +443,10 @@ static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
 		if (this_events)
 			continue;
 
-		wait = 1;
+		if (!o->busy_poll)
+			wait = 1;
+		else
+			nop;
 	} while (1);
 
 	return events;
@@ -265,76 +454,72 @@ static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
 
 static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u)
 {
-	struct rbd_data *rbd_data = td->io_ops->data;
+	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_complete = 0;
 	fri->io_seen = 0;
+	fri->io_complete = 0;
 
-	if (io_u->ddir == DDIR_WRITE) {
-		r = rbd_aio_create_completion(io_u, _fio_rbd_finish_aiocb,
+	r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb,
 						&fri->completion);
-		if (r < 0) {
-			log_err
-			    ("rbd_aio_create_completion for DDIR_WRITE failed.\n");
-			goto failed;
-		}
+	if (r < 0) {
+		log_err("rbd_aio_create_completion failed.\n");
+		goto failed;
+	}
 
-		r = rbd_aio_write(rbd_data->image, io_u->offset,
-				  io_u->xfer_buflen, io_u->xfer_buf,
-				  fri->completion);
+	if (io_u->ddir == DDIR_WRITE) {
+#ifdef CONFIG_RBD_BLKIN
+		blkin_init_trace_info(&fri->info);
+		r = rbd_aio_write_traced(rbd->image, io_u->offset, io_u->xfer_buflen,
+					 io_u->xfer_buf, fri->completion, &fri->info);
+#else
+		r = rbd_aio_write(rbd->image, io_u->offset, io_u->xfer_buflen,
+					 io_u->xfer_buf, fri->completion);
+#endif
 		if (r < 0) {
 			log_err("rbd_aio_write failed.\n");
-			rbd_aio_release(fri->completion);
-			goto failed;
+			goto failed_comp;
 		}
 
 	} else if (io_u->ddir == DDIR_READ) {
-		r = rbd_aio_create_completion(io_u, _fio_rbd_finish_aiocb,
-						&fri->completion);
-		if (r < 0) {
-			log_err
-			    ("rbd_aio_create_completion for DDIR_READ failed.\n");
-			goto failed;
-		}
-
-		r = rbd_aio_read(rbd_data->image, io_u->offset,
-				 io_u->xfer_buflen, io_u->xfer_buf,
-				 fri->completion);
+#ifdef CONFIG_RBD_BLKIN
+		blkin_init_trace_info(&fri->info);
+		r = rbd_aio_read_traced(rbd->image, io_u->offset, io_u->xfer_buflen,
+					io_u->xfer_buf, fri->completion, &fri->info);
+#else
+		r = rbd_aio_read(rbd->image, io_u->offset, io_u->xfer_buflen,
+					io_u->xfer_buf, fri->completion);
+#endif
 
 		if (r < 0) {
 			log_err("rbd_aio_read failed.\n");
-			rbd_aio_release(fri->completion);
-			goto failed;
+			goto failed_comp;
 		}
-
-	} else if (io_u->ddir == DDIR_SYNC) {
-		r = rbd_aio_create_completion(io_u, _fio_rbd_finish_aiocb,
-						&fri->completion);
+	} 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_create_completion for DDIR_SYNC failed.\n");
-			goto failed;
+			log_err("rbd_aio_discard failed.\n");
+			goto failed_comp;
 		}
-
-		r = rbd_aio_flush(rbd_data->image, fri->completion);
+	} else if (io_u->ddir == DDIR_SYNC) {
+		r = rbd_aio_flush(rbd->image, fri->completion);
 		if (r < 0) {
 			log_err("rbd_flush failed.\n");
-			rbd_aio_release(fri->completion);
-			goto failed;
+			goto failed_comp;
 		}
-
 	} else {
 		dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
 		       io_u->ddir);
-		return FIO_Q_COMPLETED;
+		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");
@@ -359,35 +544,35 @@ failed:
 
 static void fio_rbd_cleanup(struct thread_data *td)
 {
-	struct rbd_data *rbd_data = td->io_ops->data;
+	struct rbd_data *rbd = td->io_ops_data;
 
-	if (rbd_data) {
-		_fio_rbd_disconnect(rbd_data);
-		free(rbd_data->aio_events);
-		free(rbd_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)
 {
-	int r = 0;
 	rbd_image_info_t info;
 	struct fio_file *f;
-	struct rbd_data *rbd_data = NULL;
+	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_data);
+	r = _fio_setup_rbd_data(td, &rbd);
 	if (r) {
 		log_err("fio_setup_rbd_data failed.\n");
 		goto cleanup;
 	}
-	td->io_ops->data = rbd_data;
+	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
@@ -406,7 +591,7 @@ static int fio_rbd_setup(struct thread_data *td)
 	}
 
 	/* get size of the RADOS block device */
-	r = rbd_stat(rbd_data->image, &info, sizeof(info));
+	r = rbd_stat(rbd->image, &info, sizeof(info));
 	if (r < 0) {
 		log_err("rbd_status failed.\n");
 		goto disconnect;
@@ -428,11 +613,11 @@ static int fio_rbd_setup(struct thread_data *td)
 	/* disconnect, then we were only connected to determine
 	 * the size of the RBD.
 	 */
-	_fio_rbd_disconnect(rbd_data);
+	_fio_rbd_disconnect(rbd);
 	return 0;
 
 disconnect:
-	_fio_rbd_disconnect(rbd_data);
+	_fio_rbd_disconnect(rbd);
 cleanup:
 	fio_rbd_cleanup(td);
 	return r;
@@ -446,9 +631,9 @@ static int fio_rbd_open(struct thread_data *td, struct fio_file *f)
 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;
+	struct rbd_data *rbd = td->io_ops_data;
 
-	return rbd_invalidate_cache(rbd_data->image);
+	return rbd_invalidate_cache(rbd->image);
 #else
 	return 0;
 #endif