X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=engines%2Frdma.c;h=da00cba8b66b3f6db0bcd3a9cba8b45b9362ffa7;hp=87b061a2e61788563fff7ac05d69e06b1b8e3f29;hb=9acb08a9957b1111a06fbca6af113fa0c98dbd7c;hpb=ea6209ad989dfb437aec487d26d4209bcf755d2c

diff --git a/engines/rdma.c b/engines/rdma.c
index 87b061a2..da00cba8 100644
--- a/engines/rdma.c
+++ b/engines/rdma.c
@@ -5,12 +5,7 @@
  * Supports both RDMA memory semantics and channel semantics
  *   for the InfiniBand, RoCE and iWARP protocols.
  *
- * This I/O engine is disabled by default. To enable it, execute:
- *
- * $ export EXTFLAGS+=" -DFIO_HAVE_RDMA "
- * $ export EXTLIBS+=" -libverbs -lrdmacm "
- *
- * before running make. You will need the Linux RDMA software as well, either
+ * You will need the Linux RDMA software installed, either
  * from your Linux distributor or directly from openfabrics.org:
  *
  * http://www.openfabrics.org/downloads/OFED/
@@ -18,13 +13,13 @@
  * Exchanging steps of RDMA ioengine control messages:
  *	1. client side sends test mode (RDMA_WRITE/RDMA_READ/SEND)
  *	   to server side.
- * 	2. server side parses test mode, and sends back confirmation
+ *	2. server side parses test mode, and sends back confirmation
  *	   to client side. In RDMA WRITE/READ test, this confirmation
- * 	   includes memory information, such as rkey, address.
+ *	   includes memory information, such as rkey, address.
  *	3. client side initiates test loop.
- * 	4. In RDMA WRITE/READ test, client side sends a completion
+ *	4. In RDMA WRITE/READ test, client side sends a completion
  *	   notification to server side. Server side updates its
- * 	   td->done as true.
+ *	   td->done as true.
  *
  */
 #include <stdio.h>
@@ -41,16 +36,14 @@
 #include <sys/time.h>
 #include <sys/resource.h>
 
-#include <byteswap.h>
 #include <pthread.h>
 #include <inttypes.h>
 
 #include "../fio.h"
-
-#ifdef FIO_HAVE_RDMA
+#include "../hash.h"
+#include "../optgroup.h"
 
 #include <rdma/rdma_cma.h>
-#include <infiniband/arch.h>
 
 #define FIO_RDMA_MAX_IO_DEPTH    512
 
@@ -62,6 +55,77 @@ enum rdma_io_mode {
 	FIO_RDMA_CHA_RECV
 };
 
+struct rdmaio_options {
+	struct thread_data *td;
+	unsigned int port;
+	enum rdma_io_mode verb;
+};
+
+static int str_hostname_cb(void *data, const char *input)
+{
+	struct rdmaio_options *o = data;
+
+	if (o->td->o.filename)
+		free(o->td->o.filename);
+	o->td->o.filename = strdup(input);
+	return 0;
+}
+
+static struct fio_option options[] = {
+	{
+		.name	= "hostname",
+		.lname	= "rdma engine hostname",
+		.type	= FIO_OPT_STR_STORE,
+		.cb	= str_hostname_cb,
+		.help	= "Hostname for RDMA IO engine",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_RDMA,
+	},
+	{
+		.name	= "port",
+		.lname	= "rdma engine port",
+		.type	= FIO_OPT_INT,
+		.off1	= offsetof(struct rdmaio_options, port),
+		.minval	= 1,
+		.maxval	= 65535,
+		.help	= "Port to use for RDMA connections",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_RDMA,
+	},
+	{
+		.name	= "verb",
+		.lname	= "RDMA engine verb",
+		.alias	= "proto",
+		.type	= FIO_OPT_STR,
+		.off1	= offsetof(struct rdmaio_options, verb),
+		.help	= "RDMA engine verb",
+		.def	= "write",
+		.posval = {
+			  { .ival = "write",
+			    .oval = FIO_RDMA_MEM_WRITE,
+			    .help = "Memory Write",
+			  },
+			  { .ival = "read",
+			    .oval = FIO_RDMA_MEM_READ,
+			    .help = "Memory Read",
+			  },
+			  { .ival = "send",
+			    .oval = FIO_RDMA_CHA_SEND,
+			    .help = "Posted Send",
+			  },
+			  { .ival = "recv",
+			    .oval = FIO_RDMA_CHA_RECV,
+			    .help = "Posted Receive",
+			  },
+		},
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_RDMA,
+	},
+	{
+		.name	= NULL,
+	},
+};
+
 struct remote_u {
 	uint64_t buf;
 	uint32_t rkey;
@@ -73,6 +137,7 @@ struct rdma_info_blk {
 	uint32_t nr;		/* client: io depth
 				   server: number of records for memory semantic
 				 */
+	uint32_t max_bs;        /* maximum block size */
 	struct remote_u rmt_us[FIO_RDMA_MAX_IO_DEPTH];
 };
 
@@ -125,13 +190,22 @@ struct rdmaio_data {
 
 static int client_recv(struct thread_data *td, struct ibv_wc *wc)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
+	unsigned int max_bs;
 
 	if (wc->byte_len != sizeof(rd->recv_buf)) {
 		log_err("Received bogus data, size %d\n", wc->byte_len);
 		return 1;
 	}
 
+	max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
+	if (max_bs > ntohl(rd->recv_buf.max_bs)) {
+		log_err("fio: Server's block size (%d) must be greater than or "
+			"equal to the client's block size (%d)!\n",
+			ntohl(rd->recv_buf.max_bs), max_bs);
+		return 1;
+	}
+
 	/* store mr info for MEMORY semantic */
 	if ((rd->rdma_protocol == FIO_RDMA_MEM_WRITE) ||
 	    (rd->rdma_protocol == FIO_RDMA_MEM_READ)) {
@@ -141,7 +215,7 @@ static int client_recv(struct thread_data *td, struct ibv_wc *wc)
 		rd->rmt_nr = ntohl(rd->recv_buf.nr);
 
 		for (i = 0; i < rd->rmt_nr; i++) {
-			rd->rmt_us[i].buf = ntohll(rd->recv_buf.rmt_us[i].buf);
+			rd->rmt_us[i].buf = be64_to_cpu(rd->recv_buf.rmt_us[i].buf);
 			rd->rmt_us[i].rkey = ntohl(rd->recv_buf.rmt_us[i].rkey);
 			rd->rmt_us[i].size = ntohl(rd->recv_buf.rmt_us[i].size);
 
@@ -157,7 +231,8 @@ static int client_recv(struct thread_data *td, struct ibv_wc *wc)
 
 static int server_recv(struct thread_data *td, struct ibv_wc *wc)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
+	unsigned int max_bs;
 
 	if (wc->wr_id == FIO_RDMA_MAX_IO_DEPTH) {
 		rd->rdma_protocol = ntohl(rd->recv_buf.mode);
@@ -165,6 +240,15 @@ static int server_recv(struct thread_data *td, struct ibv_wc *wc)
 		/* CHANNEL semantic, do nothing */
 		if (rd->rdma_protocol == FIO_RDMA_CHA_SEND)
 			rd->rdma_protocol = FIO_RDMA_CHA_RECV;
+
+		max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
+		if (max_bs < ntohl(rd->recv_buf.max_bs)) {
+			log_err("fio: Server's block size (%d) must be greater than or "
+				"equal to the client's block size (%d)!\n",
+				ntohl(rd->recv_buf.max_bs), max_bs);
+			return 1;
+		}
+
 	}
 
 	return 0;
@@ -172,7 +256,7 @@ static int server_recv(struct thread_data *td, struct ibv_wc *wc)
 
 static int cq_event_handler(struct thread_data *td, enum ibv_wc_opcode opcode)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_wc wc;
 	struct rdma_io_u_data *r_io_u_d;
 	int ret;
@@ -193,9 +277,12 @@ static int cq_event_handler(struct thread_data *td, enum ibv_wc_opcode opcode)
 
 		case IBV_WC_RECV:
 			if (rd->is_client == 1)
-				client_recv(td, &wc);
+				ret = client_recv(td, &wc);
 			else
-				server_recv(td, &wc);
+				ret = server_recv(td, &wc);
+
+			if (ret)
+				return -1;
 
 			if (wc.wr_id == FIO_RDMA_MAX_IO_DEPTH)
 				break;
@@ -265,6 +352,7 @@ static int cq_event_handler(struct thread_data *td, enum ibv_wc_opcode opcode)
 		}
 		rd->cq_event_num++;
 	}
+
 	if (ret) {
 		log_err("fio: poll error %d\n", ret);
 		return 1;
@@ -279,7 +367,7 @@ static int cq_event_handler(struct thread_data *td, enum ibv_wc_opcode opcode)
  */
 static int rdma_poll_wait(struct thread_data *td, enum ibv_wc_opcode opcode)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_cq *ev_cq;
 	void *ev_ctx;
 	int ret;
@@ -304,7 +392,7 @@ again:
 	}
 
 	ret = cq_event_handler(td, opcode);
-	if (ret < 1)
+	if (ret == 0)
 		goto again;
 
 	ibv_ack_cq_events(rd->cq, ret);
@@ -316,7 +404,7 @@ again:
 
 static int fio_rdmaio_setup_qp(struct thread_data *td)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_qp_init_attr init_attr;
 	int qp_depth = td->o.iodepth * 2;	/* 2 times of io depth */
 
@@ -324,8 +412,9 @@ static int fio_rdmaio_setup_qp(struct thread_data *td)
 		rd->pd = ibv_alloc_pd(rd->child_cm_id->verbs);
 	else
 		rd->pd = ibv_alloc_pd(rd->cm_id->verbs);
+
 	if (rd->pd == NULL) {
-		log_err("fio: ibv_alloc_pd fail\n");
+		log_err("fio: ibv_alloc_pd fail: %m\n");
 		return 1;
 	}
 
@@ -334,7 +423,7 @@ static int fio_rdmaio_setup_qp(struct thread_data *td)
 	else
 		rd->channel = ibv_create_comp_channel(rd->cm_id->verbs);
 	if (rd->channel == NULL) {
-		log_err("fio: ibv_create_comp_channel fail\n");
+		log_err("fio: ibv_create_comp_channel fail: %m\n");
 		goto err1;
 	}
 
@@ -348,12 +437,12 @@ static int fio_rdmaio_setup_qp(struct thread_data *td)
 		rd->cq = ibv_create_cq(rd->cm_id->verbs,
 				       qp_depth, rd, rd->channel, 0);
 	if (rd->cq == NULL) {
-		log_err("fio: ibv_create_cq failed\n");
+		log_err("fio: ibv_create_cq failed: %m\n");
 		goto err2;
 	}
 
 	if (ibv_req_notify_cq(rd->cq, 0) != 0) {
-		log_err("fio: ibv_create_cq failed\n");
+		log_err("fio: ibv_req_notify_cq failed: %m\n");
 		goto err3;
 	}
 
@@ -369,13 +458,13 @@ static int fio_rdmaio_setup_qp(struct thread_data *td)
 
 	if (rd->is_client == 0) {
 		if (rdma_create_qp(rd->child_cm_id, rd->pd, &init_attr) != 0) {
-			log_err("fio: rdma_create_qp failed\n");
+			log_err("fio: rdma_create_qp failed: %m\n");
 			goto err3;
 		}
 		rd->qp = rd->child_cm_id->qp;
 	} else {
 		if (rdma_create_qp(rd->cm_id, rd->pd, &init_attr) != 0) {
-			log_err("fio: rdma_create_qp failed\n");
+			log_err("fio: rdma_create_qp failed: %m\n");
 			goto err3;
 		}
 		rd->qp = rd->cm_id->qp;
@@ -395,19 +484,19 @@ err1:
 
 static int fio_rdmaio_setup_control_msg_buffers(struct thread_data *td)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 
 	rd->recv_mr = ibv_reg_mr(rd->pd, &rd->recv_buf, sizeof(rd->recv_buf),
 				 IBV_ACCESS_LOCAL_WRITE);
 	if (rd->recv_mr == NULL) {
-		log_err("fio: recv_buf reg_mr failed\n");
+		log_err("fio: recv_buf reg_mr failed: %m\n");
 		return 1;
 	}
 
 	rd->send_mr = ibv_reg_mr(rd->pd, &rd->send_buf, sizeof(rd->send_buf),
 				 0);
 	if (rd->send_mr == NULL) {
-		log_err("fio: send_buf reg_mr failed\n");
+		log_err("fio: send_buf reg_mr failed: %m\n");
 		ibv_dereg_mr(rd->recv_mr);
 		return 1;
 	}
@@ -415,7 +504,7 @@ static int fio_rdmaio_setup_control_msg_buffers(struct thread_data *td)
 	/* setup work request */
 	/* recv wq */
 	rd->recv_sgl.addr = (uint64_t) (unsigned long)&rd->recv_buf;
-	rd->recv_sgl.length = sizeof rd->recv_buf;
+	rd->recv_sgl.length = sizeof(rd->recv_buf);
 	rd->recv_sgl.lkey = rd->recv_mr->lkey;
 	rd->rq_wr.sg_list = &rd->recv_sgl;
 	rd->rq_wr.num_sge = 1;
@@ -423,7 +512,7 @@ static int fio_rdmaio_setup_control_msg_buffers(struct thread_data *td)
 
 	/* send wq */
 	rd->send_sgl.addr = (uint64_t) (unsigned long)&rd->send_buf;
-	rd->send_sgl.length = sizeof rd->send_buf;
+	rd->send_sgl.length = sizeof(rd->send_buf);
 	rd->send_sgl.lkey = rd->send_mr->lkey;
 
 	rd->sq_wr.opcode = IBV_WR_SEND;
@@ -439,14 +528,13 @@ static int get_next_channel_event(struct thread_data *td,
 				  struct rdma_event_channel *channel,
 				  enum rdma_cm_event_type wait_event)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
-
-	int ret;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct rdma_cm_event *event;
+	int ret;
 
 	ret = rdma_get_cm_event(channel, &event);
 	if (ret) {
-		log_err("fio: rdma_get_cm_event");
+		log_err("fio: rdma_get_cm_event: %d\n", ret);
 		return 1;
 	}
 
@@ -472,7 +560,7 @@ static int get_next_channel_event(struct thread_data *td,
 
 static int fio_rdmaio_prep(struct thread_data *td, struct io_u *io_u)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct rdma_io_u_data *r_io_u_d;
 
 	r_io_u_d = io_u->engine_data;
@@ -515,14 +603,14 @@ static int fio_rdmaio_prep(struct thread_data *td, struct io_u *io_u)
 
 static struct io_u *fio_rdmaio_event(struct thread_data *td, int event)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct io_u *io_u;
 	int i;
 
 	io_u = rd->io_us_completed[0];
-	for (i = 0; i < rd->io_u_completed_nr - 1; i++) {
+	for (i = 0; i < rd->io_u_completed_nr - 1; i++)
 		rd->io_us_completed[i] = rd->io_us_completed[i + 1];
-	}
+
 	rd->io_u_completed_nr--;
 
 	dprint_io_u(io_u, "fio_rdmaio_event");
@@ -531,17 +619,14 @@ static struct io_u *fio_rdmaio_event(struct thread_data *td, int event)
 }
 
 static int fio_rdmaio_getevents(struct thread_data *td, unsigned int min,
-				unsigned int max, struct timespec *t)
+				unsigned int max, const struct timespec *t)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
-	int r;
+	struct rdmaio_data *rd = td->io_ops_data;
 	enum ibv_wc_opcode comp_opcode;
-	comp_opcode = IBV_WC_RDMA_WRITE;
 	struct ibv_cq *ev_cq;
 	void *ev_ctx;
-	int ret;
-
-	r = 0;
+	int ret, r = 0;
+	comp_opcode = IBV_WC_RDMA_WRITE;
 
 	switch (rd->rdma_protocol) {
 	case FIO_RDMA_MEM_WRITE:
@@ -598,7 +683,7 @@ again:
 static int fio_rdmaio_send(struct thread_data *td, struct io_u **io_us,
 			   unsigned int nr)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_send_wr *bad_wr;
 #if 0
 	enum ibv_wc_opcode comp_opcode;
@@ -616,10 +701,7 @@ static int fio_rdmaio_send(struct thread_data *td, struct io_u **io_us,
 		case FIO_RDMA_MEM_WRITE:
 			/* compose work request */
 			r_io_u_d = io_us[i]->engine_data;
-			if (td->o.use_os_rand)
-				index = os_random_long(&td->random_state) % rd->rmt_nr;
-			else
-				index = __rand(&rd->rand_state) % rd->rmt_nr;
+			index = __rand(&rd->rand_state) % rd->rmt_nr;
 			r_io_u_d->sq_wr.opcode = IBV_WR_RDMA_WRITE;
 			r_io_u_d->sq_wr.wr.rdma.rkey = rd->rmt_us[index].rkey;
 			r_io_u_d->sq_wr.wr.rdma.remote_addr = \
@@ -629,10 +711,7 @@ static int fio_rdmaio_send(struct thread_data *td, struct io_u **io_us,
 		case FIO_RDMA_MEM_READ:
 			/* compose work request */
 			r_io_u_d = io_us[i]->engine_data;
-			if (td->o.use_os_rand)
-				index = os_random_long(&td->random_state) % rd->rmt_nr;
-			else
-				index = __rand(&rd->rand_state) % rd->rmt_nr;
+			index = __rand(&rd->rand_state) % rd->rmt_nr;
 			r_io_u_d->sq_wr.opcode = IBV_WR_RDMA_READ;
 			r_io_u_d->sq_wr.wr.rdma.rkey = rd->rmt_us[index].rkey;
 			r_io_u_d->sq_wr.wr.rdma.remote_addr = \
@@ -651,7 +730,7 @@ static int fio_rdmaio_send(struct thread_data *td, struct io_u **io_us,
 		}
 
 		if (ibv_post_send(rd->qp, &r_io_u_d->sq_wr, &bad_wr) != 0) {
-			log_err("fio: ibv_post_send fail\n");
+			log_err("fio: ibv_post_send fail: %m\n");
 			return -1;
 		}
 
@@ -667,7 +746,7 @@ static int fio_rdmaio_send(struct thread_data *td, struct io_u **io_us,
 static int fio_rdmaio_recv(struct thread_data *td, struct io_u **io_us,
 			   unsigned int nr)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_recv_wr *bad_wr;
 	struct rdma_io_u_data *r_io_u_d;
 	int i;
@@ -679,7 +758,7 @@ static int fio_rdmaio_recv(struct thread_data *td, struct io_u **io_us,
 			r_io_u_d = io_us[i]->engine_data;
 			if (ibv_post_recv(rd->qp, &r_io_u_d->rq_wr, &bad_wr) !=
 			    0) {
-				log_err("fio: ibv_post_recv fail\n");
+				log_err("fio: ibv_post_recv fail: %m\n");
 				return 1;
 			}
 		}
@@ -687,7 +766,7 @@ static int fio_rdmaio_recv(struct thread_data *td, struct io_u **io_us,
 		   || (rd->rdma_protocol == FIO_RDMA_MEM_WRITE)) {
 		/* re-post the rq_wr */
 		if (ibv_post_recv(rd->qp, &rd->rq_wr, &bad_wr) != 0) {
-			log_err("fio: ibv_post_recv fail\n");
+			log_err("fio: ibv_post_recv fail: %m\n");
 			return 1;
 		}
 
@@ -703,7 +782,7 @@ static int fio_rdmaio_recv(struct thread_data *td, struct io_u **io_us,
 
 static int fio_rdmaio_queue(struct thread_data *td, struct io_u *io_u)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 
 	fio_ro_check(td, io_u);
 
@@ -721,8 +800,8 @@ static int fio_rdmaio_queue(struct thread_data *td, struct io_u *io_u)
 static void fio_rdmaio_queued(struct thread_data *td, struct io_u **io_us,
 			      unsigned int nr)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
-	struct timeval now;
+	struct rdmaio_data *rd = td->io_ops_data;
+	struct timespec now;
 	unsigned int i;
 
 	if (!fio_fill_issue_time(td))
@@ -744,7 +823,7 @@ static void fio_rdmaio_queued(struct thread_data *td, struct io_u **io_us,
 
 static int fio_rdmaio_commit(struct thread_data *td)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct io_u **io_us;
 	int ret;
 
@@ -754,11 +833,11 @@ static int fio_rdmaio_commit(struct thread_data *td)
 	io_us = rd->io_us_queued;
 	do {
 		/* RDMA_WRITE or RDMA_READ */
-		if (rd->is_client) {
+		if (rd->is_client)
 			ret = fio_rdmaio_send(td, io_us, rd->io_u_queued_nr);
-		} else if (!rd->is_client) {
+		else if (!rd->is_client)
 			ret = fio_rdmaio_recv(td, io_us, rd->io_u_queued_nr);
-		} else
+		else
 			ret = 0;	/* must be a SYNC */
 
 		if (ret > 0) {
@@ -776,17 +855,17 @@ static int fio_rdmaio_commit(struct thread_data *td)
 
 static int fio_rdmaio_connect(struct thread_data *td, struct fio_file *f)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct rdma_conn_param conn_param;
 	struct ibv_send_wr *bad_wr;
 
-	memset(&conn_param, 0, sizeof conn_param);
+	memset(&conn_param, 0, sizeof(conn_param));
 	conn_param.responder_resources = 1;
 	conn_param.initiator_depth = 1;
 	conn_param.retry_count = 10;
 
 	if (rdma_connect(rd->cm_id, &conn_param) != 0) {
-		log_err("fio: rdma_connect fail\n");
+		log_err("fio: rdma_connect fail: %m\n");
 		return 1;
 	}
 
@@ -801,19 +880,21 @@ static int fio_rdmaio_connect(struct thread_data *td, struct fio_file *f)
 	rd->send_buf.nr = htonl(td->o.iodepth);
 
 	if (ibv_post_send(rd->qp, &rd->sq_wr, &bad_wr) != 0) {
-		log_err("fio: ibv_post_send fail");
+		log_err("fio: ibv_post_send fail: %m\n");
 		return 1;
 	}
 
-	rdma_poll_wait(td, IBV_WC_SEND);
+	if (rdma_poll_wait(td, IBV_WC_SEND) < 0)
+		return 1;
 
 	/* wait for remote MR info from server side */
-	rdma_poll_wait(td, IBV_WC_RECV);
+	if (rdma_poll_wait(td, IBV_WC_RECV) < 0)
+		return 1;
 
 	/* In SEND/RECV test, it's a good practice to setup the iodepth of
 	 * of the RECV side deeper than that of the SEND side to
 	 * avoid RNR (receiver not ready) error. The
-	 * SEND side may send so many unsolicited message before 
+	 * SEND side may send so many unsolicited message before
 	 * RECV side commits sufficient recv buffers into recv queue.
 	 * This may lead to RNR error. Here, SEND side pauses for a while
 	 * during which RECV side commits sufficient recv buffers.
@@ -825,17 +906,18 @@ static int fio_rdmaio_connect(struct thread_data *td, struct fio_file *f)
 
 static int fio_rdmaio_accept(struct thread_data *td, struct fio_file *f)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct rdma_conn_param conn_param;
 	struct ibv_send_wr *bad_wr;
+	int ret = 0;
 
 	/* rdma_accept() - then wait for accept success */
-	memset(&conn_param, 0, sizeof conn_param);
+	memset(&conn_param, 0, sizeof(conn_param));
 	conn_param.responder_resources = 1;
 	conn_param.initiator_depth = 1;
 
 	if (rdma_accept(rd->child_cm_id, &conn_param) != 0) {
-		log_err("fio: rdma_accept\n");
+		log_err("fio: rdma_accept: %m\n");
 		return 1;
 	}
 
@@ -846,16 +928,17 @@ static int fio_rdmaio_accept(struct thread_data *td, struct fio_file *f)
 	}
 
 	/* wait for request */
-	rdma_poll_wait(td, IBV_WC_RECV);
+	ret = rdma_poll_wait(td, IBV_WC_RECV) < 0;
 
 	if (ibv_post_send(rd->qp, &rd->sq_wr, &bad_wr) != 0) {
-		log_err("fio: ibv_post_send fail");
+		log_err("fio: ibv_post_send fail: %m\n");
 		return 1;
 	}
 
-	rdma_poll_wait(td, IBV_WC_SEND);
+	if (rdma_poll_wait(td, IBV_WC_SEND) < 0)
+		return 1;
 
-	return 0;
+	return ret;
 }
 
 static int fio_rdmaio_open_file(struct thread_data *td, struct fio_file *f)
@@ -868,7 +951,7 @@ static int fio_rdmaio_open_file(struct thread_data *td, struct fio_file *f)
 
 static int fio_rdmaio_close_file(struct thread_data *td, struct fio_file *f)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_send_wr *bad_wr;
 
 	/* unregister rdma buffer */
@@ -881,11 +964,11 @@ static int fio_rdmaio_close_file(struct thread_data *td, struct fio_file *f)
 				     || (rd->rdma_protocol ==
 					 FIO_RDMA_MEM_READ))) {
 		if (ibv_post_send(rd->qp, &rd->sq_wr, &bad_wr) != 0) {
-			log_err("fio: ibv_post_send fail");
+			log_err("fio: ibv_post_send fail: %m\n");
 			return 1;
 		}
 
-		dprint(FD_IO, "fio: close infomation sent success\n");
+		dprint(FD_IO, "fio: close information sent success\n");
 		rdma_poll_wait(td, IBV_WC_SEND);
 	}
 
@@ -893,14 +976,17 @@ static int fio_rdmaio_close_file(struct thread_data *td, struct fio_file *f)
 		rdma_disconnect(rd->cm_id);
 	else {
 		rdma_disconnect(rd->child_cm_id);
-/*        rdma_disconnect(rd->cm_id); */
+#if 0
+		rdma_disconnect(rd->cm_id);
+#endif
 	}
 
-/*    if (get_next_channel_event(td, rd->cm_channel, RDMA_CM_EVENT_DISCONNECTED) != 0)
-    {
-        log_err("fio: wait for RDMA_CM_EVENT_DISCONNECTED\n");
-        return 1;
-    }*/
+#if 0
+	if (get_next_channel_event(td, rd->cm_channel, RDMA_CM_EVENT_DISCONNECTED) != 0) {
+		log_err("fio: wait for RDMA_CM_EVENT_DISCONNECTED\n");
+		return 1;
+	}
+#endif
 
 	ibv_destroy_cq(rd->cq);
 	ibv_destroy_qp(rd->qp);
@@ -921,8 +1007,9 @@ static int fio_rdmaio_close_file(struct thread_data *td, struct fio_file *f)
 static int fio_rdmaio_setup_connect(struct thread_data *td, const char *host,
 				    unsigned short port)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_recv_wr *bad_wr;
+	int err;
 
 	rd->addr.sin_family = AF_INET;
 	rd->addr.sin_port = htons(port);
@@ -940,28 +1027,28 @@ static int fio_rdmaio_setup_connect(struct thread_data *td, const char *host,
 	}
 
 	/* resolve route */
-	if (rdma_resolve_addr(rd->cm_id, NULL,
-			      (struct sockaddr *)&rd->addr, 2000) != 0) {
-		log_err("fio: rdma_resolve_addr");
+	err = rdma_resolve_addr(rd->cm_id, NULL, (struct sockaddr *)&rd->addr, 2000);
+	if (err != 0) {
+		log_err("fio: rdma_resolve_addr: %d\n", err);
 		return 1;
 	}
 
-	if (get_next_channel_event
-	    (td, rd->cm_channel, RDMA_CM_EVENT_ADDR_RESOLVED)
-	    != 0) {
-		log_err("fio: get_next_channel_event");
+	err = get_next_channel_event(td, rd->cm_channel, RDMA_CM_EVENT_ADDR_RESOLVED);
+	if (err != 0) {
+		log_err("fio: get_next_channel_event: %d\n", err);
 		return 1;
 	}
 
 	/* resolve route */
-	if (rdma_resolve_route(rd->cm_id, 2000) != 0) {
-		log_err("fio: rdma_resolve_route");
+	err = rdma_resolve_route(rd->cm_id, 2000);
+	if (err != 0) {
+		log_err("fio: rdma_resolve_route: %d\n", err);
 		return 1;
 	}
 
-	if (get_next_channel_event
-	    (td, rd->cm_channel, RDMA_CM_EVENT_ROUTE_RESOLVED) != 0) {
-		log_err("fio: get_next_channel_event");
+	err = get_next_channel_event(td, rd->cm_channel, RDMA_CM_EVENT_ROUTE_RESOLVED);
+	if (err != 0) {
+		log_err("fio: get_next_channel_event: %d\n", err);
 		return 1;
 	}
 
@@ -973,8 +1060,9 @@ static int fio_rdmaio_setup_connect(struct thread_data *td, const char *host,
 		return 1;
 
 	/* post recv buf */
-	if (ibv_post_recv(rd->qp, &rd->rq_wr, &bad_wr) != 0) {
-		log_err("fio: ibv_post_recv fail\n");
+	err = ibv_post_recv(rd->qp, &rd->rq_wr, &bad_wr);
+	if (err != 0) {
+		log_err("fio: ibv_post_recv fail: %d\n", err);
 		return 1;
 	}
 
@@ -983,8 +1071,11 @@ static int fio_rdmaio_setup_connect(struct thread_data *td, const char *host,
 
 static int fio_rdmaio_setup_listen(struct thread_data *td, short port)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
+	struct rdmaio_data *rd = td->io_ops_data;
 	struct ibv_recv_wr *bad_wr;
+	int state = td->runstate;
+
+	td_set_runstate(td, TD_SETTING_UP);
 
 	rd->addr.sin_family = AF_INET;
 	rd->addr.sin_addr.s_addr = htonl(INADDR_ANY);
@@ -992,15 +1083,17 @@ static int fio_rdmaio_setup_listen(struct thread_data *td, short port)
 
 	/* rdma_listen */
 	if (rdma_bind_addr(rd->cm_id, (struct sockaddr *)&rd->addr) != 0) {
-		log_err("fio: rdma_bind_addr fail\n");
+		log_err("fio: rdma_bind_addr fail: %m\n");
 		return 1;
 	}
 
 	if (rdma_listen(rd->cm_id, 3) != 0) {
-		log_err("fio: rdma_listen fail\n");
+		log_err("fio: rdma_listen fail: %m\n");
 		return 1;
 	}
 
+	log_info("fio: waiting for connection\n");
+
 	/* wait for CONNECT_REQUEST */
 	if (get_next_channel_event
 	    (td, rd->cm_channel, RDMA_CM_EVENT_CONNECT_REQUEST) != 0) {
@@ -1016,31 +1109,19 @@ static int fio_rdmaio_setup_listen(struct thread_data *td, short port)
 
 	/* post recv buf */
 	if (ibv_post_recv(rd->qp, &rd->rq_wr, &bad_wr) != 0) {
-		log_err("fio: ibv_post_recv fail\n");
+		log_err("fio: ibv_post_recv fail: %m\n");
 		return 1;
 	}
 
+	td_set_runstate(td, state);
 	return 0;
 }
 
-static int fio_rdmaio_init(struct thread_data *td)
+static int check_set_rlimits(struct thread_data *td)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
-	unsigned int port;
-	char host[64], buf[128];
-	char *sep, *portp, *modep;
-	int ret;
+#ifdef CONFIG_RLIMIT_MEMLOCK
 	struct rlimit rl;
 
-	if (td_rw(td)) {
-		log_err("fio: rdma connections must be read OR write\n");
-		return 1;
-	}
-	if (td_random(td)) {
-		log_err("fio: RDMA network IO can't be random\n");
-		return 1;
-	}
-
 	/* check RLIMIT_MEMLOCK */
 	if (getrlimit(RLIMIT_MEMLOCK, &rl) != 0) {
 		log_err("fio: getrlimit fail: %d(%s)\n",
@@ -1065,57 +1146,104 @@ static int fio_rdmaio_init(struct thread_data *td)
 			return 1;
 		}
 	}
+#endif
 
-	strcpy(buf, td->o.filename);
+	return 0;
+}
 
-	sep = strchr(buf, '/');
-	if (!sep)
-		goto bad_host;
+static int compat_options(struct thread_data *td)
+{
+	// The original RDMA engine had an ugly / seperator
+	// on the filename for it's options. This function
+	// retains backwards compatibility with it.100
 
-	*sep = '\0';
-	sep++;
-	strcpy(host, buf);
-	if (!strlen(host))
-		goto bad_host;
+	struct rdmaio_options *o = td->eo;
+	char *modep, *portp;
+	char *filename = td->o.filename;
 
-	modep = NULL;
-	portp = sep;
-	sep = strchr(portp, '/');
-	if (sep) {
-		*sep = '\0';
-		modep = sep + 1;
-	}
+	if (!filename)
+		return 0;
 
-	port = strtol(portp, NULL, 10);
-	if (!port || port > 65535)
+	portp = strchr(filename, '/');
+	if (portp == NULL)
+		return 0;
+
+	*portp = '\0';
+	portp++;
+
+	o->port = strtol(portp, NULL, 10);
+	if (!o->port || o->port > 65535)
 		goto bad_host;
 
+	modep = strchr(portp, '/');
+	if (modep != NULL) {
+		*modep = '\0';
+		modep++;
+	}
+
 	if (modep) {
 		if (!strncmp("rdma_write", modep, strlen(modep)) ||
 		    !strncmp("RDMA_WRITE", modep, strlen(modep)))
-			rd->rdma_protocol = FIO_RDMA_MEM_WRITE;
+			o->verb = FIO_RDMA_MEM_WRITE;
 		else if (!strncmp("rdma_read", modep, strlen(modep)) ||
 			 !strncmp("RDMA_READ", modep, strlen(modep)))
-			rd->rdma_protocol = FIO_RDMA_MEM_READ;
+			o->verb = FIO_RDMA_MEM_READ;
 		else if (!strncmp("send", modep, strlen(modep)) ||
 			 !strncmp("SEND", modep, strlen(modep)))
-			rd->rdma_protocol = FIO_RDMA_CHA_SEND;
+			o->verb = FIO_RDMA_CHA_SEND;
 		else
 			goto bad_host;
 	} else
-		rd->rdma_protocol = FIO_RDMA_MEM_WRITE;
+		o->verb = FIO_RDMA_MEM_WRITE;
+
+
+	return 0;
+
+bad_host:
+	log_err("fio: bad rdma host/port/protocol: %s\n", td->o.filename);
+	return 1;
+}
+
+static int fio_rdmaio_init(struct thread_data *td)
+{
+	struct rdmaio_data *rd = td->io_ops_data;
+	struct rdmaio_options *o = td->eo;
+	unsigned int max_bs;
+	int ret, i;
+
+	if (td_rw(td)) {
+		log_err("fio: rdma connections must be read OR write\n");
+		return 1;
+	}
+	if (td_random(td)) {
+		log_err("fio: RDMA network IO can't be random\n");
+		return 1;
+	}
+
+	if (compat_options(td))
+		return 1;
+
+	if (!o->port) {
+		log_err("fio: no port has been specified which is required "
+			"for the rdma engine\n");
+		return 1;
+	}
+
+	if (check_set_rlimits(td))
+		return 1;
 
+	rd->rdma_protocol = o->verb;
 	rd->cq_event_num = 0;
 
 	rd->cm_channel = rdma_create_event_channel();
 	if (!rd->cm_channel) {
-		log_err("fio: rdma_create_event_channel fail\n");
+		log_err("fio: rdma_create_event_channel fail: %m\n");
 		return 1;
 	}
 
 	ret = rdma_create_id(rd->cm_channel, &rd->cm_id, rd, RDMA_PS_TCP);
 	if (ret) {
-		log_err("fio: rdma_create_id fail\n");
+		log_err("fio: rdma_create_id fail: %m\n");
 		return 1;
 	}
 
@@ -1142,20 +1270,20 @@ static int fio_rdmaio_init(struct thread_data *td)
 
 	if (td_read(td)) {	/* READ as the server */
 		rd->is_client = 0;
+		td->flags |= TD_F_NO_PROGRESS;
 		/* server rd->rdma_buf_len will be setup after got request */
-		ret = fio_rdmaio_setup_listen(td, port);
+		ret = fio_rdmaio_setup_listen(td, o->port);
 	} else {		/* WRITE as the client */
 		rd->is_client = 1;
-		ret = fio_rdmaio_setup_connect(td, host, port);
+		ret = fio_rdmaio_setup_connect(td, td->o.filename, o->port);
 	}
 
-	struct flist_head *entry;
-	unsigned int max_bs;
 	max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
+	rd->send_buf.max_bs = htonl(max_bs);
+
 	/* register each io_u in the free list */
-	int i = 0;
-	flist_for_each(entry, &td->io_u_freelist) {
-		struct io_u *io_u = flist_entry(entry, struct io_u, list);
+	for (i = 0; i < td->io_u_freelist.nr; i++) {
+		struct io_u *io_u = td->io_u_freelist.io_us[i];
 
 		io_u->engine_data = malloc(sizeof(struct rdma_io_u_data));
 		memset(io_u->engine_data, 0, sizeof(struct rdma_io_u_data));
@@ -1166,118 +1294,72 @@ static int fio_rdmaio_init(struct thread_data *td)
 				      IBV_ACCESS_REMOTE_READ |
 				      IBV_ACCESS_REMOTE_WRITE);
 		if (io_u->mr == NULL) {
-			log_err("fio: ibv_reg_mr io_u failed\n");
+			log_err("fio: ibv_reg_mr io_u failed: %m\n");
 			return 1;
 		}
 
 		rd->send_buf.rmt_us[i].buf =
-		    htonll((uint64_t) (unsigned long)io_u->buf);
+		    cpu_to_be64((uint64_t) (unsigned long)io_u->buf);
 		rd->send_buf.rmt_us[i].rkey = htonl(io_u->mr->rkey);
 		rd->send_buf.rmt_us[i].size = htonl(max_bs);
 
-/*    log_info("fio: Send rkey %x addr %" PRIx64 " len %d to client\n",
-          io_u->mr->rkey, io_u->buf, max_bs); */
-		i++;
+#if 0
+		log_info("fio: Send rkey %x addr %" PRIx64 " len %d to client\n", io_u->mr->rkey, io_u->buf, max_bs); */
+#endif
 	}
 
 	rd->send_buf.nr = htonl(i);
 
 	return ret;
-bad_host:
-	log_err("fio: bad rdma host/port/protocol: %s\n", td->o.filename);
-	return 1;
 }
 
 static void fio_rdmaio_cleanup(struct thread_data *td)
 {
-	struct rdmaio_data *rd = td->io_ops->data;
-
-	if (rd) {
-/*        if (nd->listenfd != -1)
-            close(nd->listenfd);
-        if (nd->pipes[0] != -1)
-            close(nd->pipes[0]);
-        if (nd->pipes[1] != -1)
-            close(nd->pipes[1]);
-*/
+	struct rdmaio_data *rd = td->io_ops_data;
+
+	if (rd)
 		free(rd);
-	}
 }
 
 static int fio_rdmaio_setup(struct thread_data *td)
 {
 	struct rdmaio_data *rd;
 
-	if (!td->io_ops->data) {
-		rd = malloc(sizeof(*rd));;
-
-		memset(rd, 0, sizeof(*rd));
-		init_rand_seed(&rd->rand_state, GOLDEN_RATIO_PRIME);
-		td->io_ops->data = rd;
+	if (!td->files_index) {
+		add_file(td, td->o.filename ?: "rdma", 0, 0);
+		td->o.nr_files = td->o.nr_files ?: 1;
+		td->o.open_files++;
 	}
 
-	return 0;
-}
-
-static struct ioengine_ops ioengine_rw = {
-	.name		= "rdma",
-	.version	= FIO_IOOPS_VERSION,
-	.setup		= fio_rdmaio_setup,
-	.init		= fio_rdmaio_init,
-	.prep		= fio_rdmaio_prep,
-	.queue		= fio_rdmaio_queue,
-	.commit		= fio_rdmaio_commit,
-	.getevents	= fio_rdmaio_getevents,
-	.event		= fio_rdmaio_event,
-	.cleanup	= fio_rdmaio_cleanup,
-	.open_file	= fio_rdmaio_open_file,
-	.close_file	= fio_rdmaio_close_file,
-	.flags		= FIO_DISKLESSIO | FIO_UNIDIR | FIO_PIPEIO,
-};
-
-#else /* FIO_HAVE_RDMA */
+	if (!td->io_ops_data) {
+		rd = malloc(sizeof(*rd));
 
-static int fio_rdmaio_open_file(struct thread_data *td, struct fio_file *f)
-{
-	return 0;
-}
+		memset(rd, 0, sizeof(*rd));
+		init_rand_seed(&rd->rand_state, (unsigned int) GOLDEN_RATIO_PRIME, 0);
+		td->io_ops_data = rd;
+	}
 
-static int fio_rdmaio_close_file(struct thread_data *td, struct fio_file *f)
-{
 	return 0;
 }
 
-static int fio_rdmaio_queue(struct thread_data *td, struct io_u *io_u)
-{
-	return FIO_Q_COMPLETED;
-}
-
-static int fio_rdmaio_init(struct thread_data fio_unused * td)
-{
-	log_err("fio: rdma(librdmacm libibverbs) not available\n");
-	log_err("     You haven't compiled rdma ioengine into fio.\n");
-	log_err("     If you want to try rdma ioengine,\n");
-	log_err("     make sure OFED is installed,\n");
-	log_err("     $ ofed_info\n");
-	log_err("     then try to make fio as follows:\n");
-	log_err("     $ export EXTFLAGS+=\" -DFIO_HAVE_RDMA \"\n");
-	log_err("     $ export EXTLIBS+=\" -libverbs -lrdmacm \"\n");
-	log_err("     $ make clean && make\n");
-	return 1;
-}
-
 static struct ioengine_ops ioengine_rw = {
-	.name		= "rdma",
-	.version	= FIO_IOOPS_VERSION,
-	.init		= fio_rdmaio_init,
-	.queue		= fio_rdmaio_queue,
-	.open_file	= fio_rdmaio_open_file,
-	.close_file	= fio_rdmaio_close_file,
-	.flags		= FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR | FIO_PIPEIO,
+	.name			= "rdma",
+	.version		= FIO_IOOPS_VERSION,
+	.setup			= fio_rdmaio_setup,
+	.init			= fio_rdmaio_init,
+	.prep			= fio_rdmaio_prep,
+	.queue			= fio_rdmaio_queue,
+	.commit			= fio_rdmaio_commit,
+	.getevents		= fio_rdmaio_getevents,
+	.event			= fio_rdmaio_event,
+	.cleanup		= fio_rdmaio_cleanup,
+	.open_file		= fio_rdmaio_open_file,
+	.close_file		= fio_rdmaio_close_file,
+	.flags			= FIO_DISKLESSIO | FIO_UNIDIR | FIO_PIPEIO,
+	.options		= options,
+	.option_struct_size	= sizeof(struct rdmaio_options),
 };
 
-#endif
-
 static void fio_init fio_rdmaio_register(void)
 {
 	register_ioengine(&ioengine_rw);