X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=engines%2Frbd.c;h=2f25889ac877a09730bf6e037f5c9a3d7042dc64;hb=9dc528b1638b625b5e167983a74de4e85c5859ea;hp=02fe655202ca6320617493e994df80693b6a3cf9;hpb=53b6c9794fcf64a53b4b51ab992af723b84f1aa5;p=fio.git

diff --git a/engines/rbd.c b/engines/rbd.c
index 02fe6552..2f25889a 100644
--- a/engines/rbd.c
+++ b/engines/rbd.c
@@ -9,9 +9,6 @@
 
 #include "../fio.h"
 #include "../optgroup.h"
-#ifdef CONFIG_RBD_BLKIN
-#include <zipkin_c.h>
-#endif
 
 #ifdef CONFIG_RBD_POLL
 /* add for poll */
@@ -24,9 +21,6 @@ struct fio_rbd_iou {
 	rbd_completion_t completion;
 	int io_seen;
 	int io_complete;
-#ifdef CONFIG_RBD_BLKIN
-	struct blkin_trace_info info;
-#endif
 };
 
 struct rbd_data {
@@ -35,9 +29,8 @@ struct rbd_data {
 	rbd_image_t image;
 	struct io_u **aio_events;
 	struct io_u **sort_events;
-#ifdef CONFIG_RBD_POLL
 	int fd; /* add for poll */
-#endif
+	bool connected;
 };
 
 struct rbd_options {
@@ -113,10 +106,10 @@ static int _fio_setup_rbd_data(struct thread_data *td,
 	if (!rbd)
 		goto failed;
 
-#ifdef CONFIG_RBD_POLL
+	rbd->connected = false;
+
 	/* add for poll, init fd: -1 */
 	rbd->fd = -1;
-#endif
 
 	rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *));
 	if (!rbd->aio_events)
@@ -141,6 +134,35 @@ failed:
 
 }
 
+#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_SEMAPHORE);
+	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 = td->io_ops_data;
@@ -151,7 +173,7 @@ static int _fio_rbd_connect(struct thread_data *td)
 		char *client_name = NULL; 
 
 		/*
-		 * If we specify cluser name, the rados_create2
+		 * If we specify cluster name, the rados_create2
 		 * will not assume 'client.'. name is considered
 		 * as a full type.id namestr
 		 */
@@ -178,6 +200,14 @@ static int _fio_rbd_connect(struct thread_data *td)
 		log_err("rados_create failed.\n");
 		goto failed_early;
 	}
+	if (o->pool_name == NULL) {
+		log_err("rbd pool name must be provided.\n");
+		goto failed_early;
+	}
+	if (!o->rbd_name) {
+		log_err("rbdname must be provided.\n");
+		goto failed_early;
+	}
 
 	r = rados_conf_read_file(rbd->cluster, NULL);
 	if (r < 0) {
@@ -197,35 +227,38 @@ static int _fio_rbd_connect(struct thread_data *td)
 		goto failed_shutdown;
 	}
 
+        if (td->o.odirect) {
+		r = rados_conf_set(rbd->cluster, "rbd_cache", "false");
+		if (r < 0) {
+			log_info("failed to disable RBD in-memory cache\n");
+		}
+	}
+
 	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;
 	}
 
-#ifdef CONFIG_RBD_POLL
-	/* add for rbd poll */
-	rbd->fd = eventfd(0, EFD_NONBLOCK);
-	if (rbd->fd < 0) {
-		log_err("eventfd failed.\n");
-		goto failed_open;
-	}
-    
-	r = rbd_set_image_notification(rbd->image, rbd->fd, EVENT_TYPE_EVENTFD);
-	if (r < 0) {
-		log_err("rbd_set_image_notification failed.\n");
-		goto failed_notify;
+	if (!td->o.odirect) {
+		/*
+		 * ensure cache enables writeback/around mode unless explicitly
+		 * configured for writethrough mode
+		 */
+		r = rbd_flush(rbd->image);
+		if (r < 0) {
+			log_info("rbd: failed to issue initial flush\n");
+		}
 	}
-#endif
 
-	return 0;
+	if (!_fio_rbd_setup_poll(rbd))
+		goto failed_poll;
 
-#ifdef CONFIG_RBD_POLL
-failed_notify:
-	close(rbd->fd);
-	rbd->fd = -1;
-#endif
+	return 0;
 
+failed_poll:
+	rbd_close(rbd->image);
+	rbd->image = NULL;
 failed_open:
 	rados_ioctx_destroy(rbd->io_ctx);
 	rbd->io_ctx = NULL;
@@ -241,13 +274,11 @@ static void _fio_rbd_disconnect(struct rbd_data *rbd)
 	if (!rbd)
 		return;
 
-#ifdef CONFIG_RBD_POLL
 	/* close eventfd */
-	if (rbd->fd >= 0) {
+	if (rbd->fd != -1) {
 		close(rbd->fd);
 		rbd->fd = -1;
 	}
-#endif
 
 	/* shutdown everything */
 	if (rbd->image) {
@@ -279,7 +310,7 @@ static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data)
 	 */
 	ret = rbd_aio_get_return_value(fri->completion);
 	if (ret < 0) {
-		io_u->error = ret;
+		io_u->error = -ret;
 		io_u->resid = io_u->xfer_buflen;
 	} else
 		io_u->error = 0;
@@ -311,12 +342,14 @@ static inline int fri_check_complete(struct rbd_data *rbd, struct io_u *io_u,
 	return 0;
 }
 
+#ifndef CONFIG_RBD_POLL
 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;
 }
+#endif
 
 static void rbd_io_u_wait_complete(struct io_u *io_u)
 {
@@ -355,26 +388,37 @@ static int rbd_iter_events(struct thread_data *td, unsigned int *events,
 	int event_num = 0;
 	struct fio_rbd_iou *fri = NULL;
 	rbd_completion_t comps[min_evts];
+	uint64_t counter;
+	bool completed;
 
 	struct pollfd pfd;
 	pfd.fd = rbd->fd;
 	pfd.events = POLLIN;
 
-	ret = poll(&pfd, 1, -1);
-	if (ret <= 0) {
+	ret = poll(&pfd, 1, wait ? -1 : 0);
+	if (ret <= 0)
+		return 0;
+	if (!(pfd.revents & POLLIN))
 		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;
+
+		/* best effort to decrement the semaphore */
+		ret = read(rbd->fd, &counter, sizeof(counter));
+		if (ret <= 0)
+			log_err("rbd_iter_events failed to decrement semaphore.\n");
+
+		completed = fri_check_complete(rbd, io_u, events);
+		assert(completed);
+
+		this_events++;
+	}
 #else
 	io_u_qiter(&td->io_u_all, io_u, i) {
-#endif
 		if (!(io_u->flags & IO_U_F_FLIGHT))
 			continue;
 		if (rbd_io_u_seen(io_u))
@@ -385,6 +429,7 @@ static int rbd_iter_events(struct thread_data *td, unsigned int *events,
 		else if (wait)
 			rbd->sort_events[sidx++] = io_u;
 	}
+#endif
 
 	if (!wait || !sidx)
 		return this_events;
@@ -445,7 +490,8 @@ static int fio_rbd_getevents(struct thread_data *td, unsigned int min,
 	return events;
 }
 
-static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u)
+static enum fio_q_status 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;
@@ -464,28 +510,16 @@ static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u)
 	}
 
 	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");
 			goto failed_comp;
 		}
 
 	} else if (io_u->ddir == DDIR_READ) {
-#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");
@@ -507,6 +541,7 @@ static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u)
 	} else {
 		dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__,
 		       io_u->ddir);
+		r = -EINVAL;
 		goto failed_comp;
 	}
 
@@ -514,7 +549,7 @@ static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u)
 failed_comp:
 	rbd_aio_release(fri->completion);
 failed:
-	io_u->error = r;
+	io_u->error = -r;
 	td_verror(td, io_u->error, "xfer");
 	return FIO_Q_COMPLETED;
 }
@@ -522,6 +557,10 @@ failed:
 static int fio_rbd_init(struct thread_data *td)
 {
 	int r;
+	struct rbd_data *rbd = td->io_ops_data;
+
+	if (rbd->connected)
+		return 0;
 
 	r = _fio_rbd_connect(td);
 	if (r) {
@@ -552,13 +591,8 @@ 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) {
@@ -582,18 +616,24 @@ static int fio_rbd_setup(struct thread_data *td)
 		log_err("fio_rbd_connect failed.\n");
 		goto cleanup;
 	}
+	rbd->connected = true;
 
 	/* 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;
+		goto cleanup;
+	} else if (info.size == 0) {
+		log_err("image size should be larger than zero.\n");
+		r = -EINVAL;
+		goto cleanup;
 	}
-	dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size);
+
+	dprint(FD_IO, "rbd-engine: image size: %" PRIu64 "\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.
+	 * The size of the RBD is set instead of an artificial file.
 	 */
 	if (!td->files_index) {
 		add_file(td, td->o.filename ? : "rbd", 0, 0);
@@ -603,14 +643,8 @@ static int fio_rbd_setup(struct thread_data *td)
 	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;
@@ -652,7 +686,7 @@ static int fio_rbd_io_u_init(struct thread_data *td, struct io_u *io_u)
 	return 0;
 }
 
-static struct ioengine_ops ioengine = {
+FIO_STATIC struct ioengine_ops ioengine = {
 	.name			= "rbd",
 	.version		= FIO_IOOPS_VERSION,
 	.setup			= fio_rbd_setup,