X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=engines%2Fnet.c;h=afb3265c6ada813179fa12853ec121d6632ab378;hp=17d24374f4ed7742f676bfc83aa83b6068c8e0ab;hb=35651820abe54473964cc1832fdfe23d7af3209a;hpb=b5af82930ccfd7dda6a1b11794efb452eb76d8dc

diff --git a/engines/net.c b/engines/net.c
index 17d24374..afb3265c 100644
--- a/engines/net.c
+++ b/engines/net.c
@@ -1,5 +1,8 @@
 /*
- * Transfer data over the net.
+ * net engine
+ *
+ * IO engine that reads/writes to/from sockets.
+ *
  */
 #include <stdio.h>
 #include <stdlib.h>
@@ -10,21 +13,57 @@
 #include <arpa/inet.h>
 #include <netdb.h>
 #include <sys/poll.h>
+#include <sys/types.h>
+#include <sys/socket.h>
 
 #include "../fio.h"
-#include "../os.h"
 
 struct netio_data {
 	int listenfd;
 	int send_to_net;
+	int use_splice;
+	int net_protocol;
+	int pipes[2];
 	char host[64];
 	struct sockaddr_in addr;
 };
 
+/*
+ * Return -1 for error and 'nr events' for a positive number
+ * of events
+ */
+static int poll_wait(struct thread_data *td, int fd, short events)
+{
+	struct pollfd pfd;
+	int ret;
+
+	while (!td->terminate) {
+		pfd.fd = fd;
+		pfd.events = events;
+		ret = poll(&pfd, 1, -1);
+		if (ret < 0) {
+			if (errno == EINTR)
+				continue;
+
+			td_verror(td, errno, "poll");
+			return -1;
+		} else if (!ret)
+			continue;
+
+		break;
+	}
+
+	if (pfd.revents & events)
+		return 1;
+	else if (td->terminate)
+		return 1;
+
+	return -1;
+}
+
 static int fio_netio_prep(struct thread_data *td, struct io_u *io_u)
 {
 	struct netio_data *nd = td->io_ops->data;
-	struct fio_file *f = io_u->file;
 
 	/*
 	 * Make sure we don't see spurious reads to a receiver, and vice versa
@@ -35,36 +74,207 @@ static int fio_netio_prep(struct thread_data *td, struct io_u *io_u)
 		return 1;
 	}
 		
-	if (io_u->ddir == DDIR_SYNC)
-		return 0;
-	if (io_u->offset == f->last_completed_pos)
-		return 0;
+	return 0;
+}
+
+#ifdef FIO_HAVE_SPLICE
+static int splice_io_u(int fdin, int fdout, unsigned int len)
+{
+	int bytes = 0;
+
+	while (len) {
+		int ret = splice(fdin, NULL, fdout, NULL, len, 0);
+
+		if (ret < 0) {
+			if (!bytes)
+				bytes = ret;
+
+			break;
+		} else if (!ret)
+			break;
+
+		bytes += ret;
+		len -= ret;
+	}
+
+	return bytes;
+}
+
+/*
+ * Receive bytes from a socket and fill them into the internal pipe
+ */
+static int splice_in(struct thread_data *td, struct io_u *io_u)
+{
+	struct netio_data *nd = td->io_ops->data;
+
+	return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen);
+}
+
+/*
+ * Transmit 'len' bytes from the internal pipe
+ */
+static int splice_out(struct thread_data *td, struct io_u *io_u,
+		      unsigned int len)
+{
+	struct netio_data *nd = td->io_ops->data;
+
+	return splice_io_u(nd->pipes[0], io_u->file->fd, len);
+}
+
+static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len)
+{
+	struct iovec iov = {
+		.iov_base = io_u->xfer_buf,
+		.iov_len = len,
+	};
+	int bytes = 0;
+
+	while (iov.iov_len) {
+		int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE);
+
+		if (ret < 0) {
+			if (!bytes)
+				bytes = ret;
+			break;
+		} else if (!ret)
+			break;
+
+		iov.iov_len -= ret;
+		iov.iov_base += ret;
+		bytes += ret;
+	}
+
+	return bytes;
+
+}
+
+/*
+ * vmsplice() pipe to io_u buffer
+ */
+static int vmsplice_io_u_out(struct thread_data *td, struct io_u *io_u,
+			     unsigned int len)
+{
+	struct netio_data *nd = td->io_ops->data;
+
+	return vmsplice_io_u(io_u, nd->pipes[0], len);
+}
+
+/*
+ * vmsplice() io_u to pipe
+ */
+static int vmsplice_io_u_in(struct thread_data *td, struct io_u *io_u)
+{
+	struct netio_data *nd = td->io_ops->data;
+
+	return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen);
+}
+
+/*
+ * splice receive - transfer socket data into a pipe using splice, then map
+ * that pipe data into the io_u using vmsplice.
+ */
+static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
+{
+	int ret;
+
+	ret = splice_in(td, io_u);
+	if (ret > 0)
+		return vmsplice_io_u_out(td, io_u, ret);
+
+	return ret;
+}
+
+/*
+ * splice transmit - map data from the io_u into a pipe by using vmsplice,
+ * then transfer that pipe to a socket using splice.
+ */
+static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u)
+{
+	int ret;
+
+	ret = vmsplice_io_u_in(td, io_u);
+	if (ret > 0)
+		return splice_out(td, io_u, ret);
+
+	return ret;
+}
+#else
+static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
+{
+	errno = EOPNOTSUPP;
+	return -1;
+}
+
+static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u)
+{
+	errno = EOPNOTSUPP;
+	return -1;
+}
+#endif
+
+static int fio_netio_send(struct thread_data *td, struct io_u *io_u)
+{
+	struct netio_data *nd = td->io_ops->data;
+	int ret, flags = 0;
+
+	ret = poll_wait(td, io_u->file->fd, POLLOUT);
+	if (ret <= 0)
+		return ret;
 
 	/*
-	 * If offset is different from last end position, it's a seek.
-	 * As network io is purely sequential, we don't allow seeks.
+	 * if we are going to write more, set MSG_MORE
 	 */
-	td_verror(td, EINVAL, "cannot seek");
-	return 1;
+#ifdef MSG_MORE
+	if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size)
+		flags = MSG_MORE;
+#endif
+
+	if (nd->net_protocol == IPPROTO_UDP) {
+		return sendto(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen,
+				0, &nd->addr, sizeof(nd->addr));
+	} else {
+		return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen,
+				flags);
+	}
+}
+
+static int fio_netio_recv(struct thread_data *td, struct io_u *io_u)
+{
+	struct netio_data *nd = td->io_ops->data;
+	int ret, flags = MSG_WAITALL;
+
+	ret = poll_wait(td, io_u->file->fd, POLLIN);
+	if (ret <= 0)
+		return ret;
+
+	if (nd->net_protocol == IPPROTO_UDP) {
+		socklen_t len = sizeof(nd->addr);
+
+		return recvfrom(io_u->file->fd, io_u->xfer_buf,
+				io_u->xfer_buflen, 0, &nd->addr, &len);
+	} else {
+		return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen,
+				flags);
+	}
 }
 
 static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
 {
-	struct fio_file *f = io_u->file;
-	int ret, flags = 0;
+	struct netio_data *nd = td->io_ops->data;
+	int ret;
+
+	fio_ro_check(td, io_u);
 
 	if (io_u->ddir == DDIR_WRITE) {
-		/*
-		 * if we are going to write more, set MSG_MORE
-		 */
-		if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen <
-		    td->io_size)
-			flags = MSG_MORE;
-
-		ret = send(f->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
+		if (!nd->use_splice || nd->net_protocol == IPPROTO_UDP)
+			ret = fio_netio_send(td, io_u);
+		else
+			ret = fio_netio_splice_out(td, io_u);
 	} else if (io_u->ddir == DDIR_READ) {
-		flags = MSG_WAITALL;
-		ret = recv(f->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
+		if (!nd->use_splice || nd->net_protocol == IPPROTO_UDP)
+			ret = fio_netio_recv(td, io_u);
+		else
+			ret = fio_netio_splice_in(td, io_u);
 	} else
 		ret = 0;	/* must be a SYNC */
 
@@ -73,8 +283,14 @@ static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
 			io_u->resid = io_u->xfer_buflen - ret;
 			io_u->error = 0;
 			return FIO_Q_COMPLETED;
-		} else
-			io_u->error = errno;
+		} else {
+			int err = errno;
+
+			if (io_u->ddir == DDIR_WRITE && err == EMSGSIZE)
+				return FIO_Q_BUSY;
+
+			io_u->error = err;
+		}
 	}
 
 	if (io_u->error)
@@ -86,13 +302,22 @@ static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
 static int fio_netio_connect(struct thread_data *td, struct fio_file *f)
 {
 	struct netio_data *nd = td->io_ops->data;
+	int type;
+
+	if (nd->net_protocol == IPPROTO_TCP)
+		type = SOCK_STREAM;
+	else
+		type = SOCK_DGRAM;
 
-	f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
+	f->fd = socket(AF_INET, type, nd->net_protocol);
 	if (f->fd < 0) {
 		td_verror(td, errno, "socket");
 		return 1;
 	}
 
+	if (nd->net_protocol == IPPROTO_UDP)
+		return 0;
+
 	if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
 		td_verror(td, errno, "connect");
 		return 1;
@@ -105,47 +330,26 @@ static int fio_netio_accept(struct thread_data *td, struct fio_file *f)
 {
 	struct netio_data *nd = td->io_ops->data;
 	socklen_t socklen = sizeof(nd->addr);
-	struct pollfd pfd;
-	int ret;
 
-	fprintf(f_out, "fio: waiting for connection\n");
-
-	/*
-	 * Accept loop. poll for incoming events, accept them. Repeat until we
-	 * have all connections.
-	 */
-	while (!td->terminate) {
-		pfd.fd = nd->listenfd;
-		pfd.events = POLLIN;
-
-		ret = poll(&pfd, 1, -1);
-		if (ret < 0) {
-			if (errno == EINTR)
-				continue;
+	if (nd->net_protocol == IPPROTO_UDP) {
+		f->fd = nd->listenfd;
+		return 0;
+	}
 
-			td_verror(td, errno, "poll");
-			break;
-		} else if (!ret)
-			continue;
+	log_info("fio: waiting for connection\n");
 
-		/*
-		 * should be impossible
-		 */
-		if (!(pfd.revents & POLLIN))
-			continue;
+	if (poll_wait(td, nd->listenfd, POLLIN) < 0)
+		return 1;
 
-		f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
-		if (f->fd < 0) {
-			td_verror(td, errno, "accept");
-			return 1;
-		}
-		break;
+	f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
+	if (f->fd < 0) {
+		td_verror(td, errno, "accept");
+		return 1;
 	}
 
 	return 0;
 }
 
-
 static int fio_netio_open_file(struct thread_data *td, struct fio_file *f)
 {
 	if (td_read(td))
@@ -180,9 +384,14 @@ static int fio_netio_setup_connect(struct thread_data *td, const char *host,
 static int fio_netio_setup_listen(struct thread_data *td, short port)
 {
 	struct netio_data *nd = td->io_ops->data;
-	int fd, opt;
+	int fd, opt, type;
 
-	fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
+	if (nd->net_protocol == IPPROTO_TCP)
+		type = SOCK_STREAM;
+	else
+		type = SOCK_DGRAM;
+
+	fd = socket(AF_INET, type, nd->net_protocol);
 	if (fd < 0) {
 		td_verror(td, errno, "socket");
 		return 1;
@@ -208,7 +417,7 @@ static int fio_netio_setup_listen(struct thread_data *td, short port)
 		td_verror(td, errno, "bind");
 		return 1;
 	}
-	if (listen(fd, 1) < 0) {
+	if (nd->net_protocol == IPPROTO_TCP && listen(fd, 1) < 0) {
 		td_verror(td, errno, "listen");
 		return 1;
 	}
@@ -220,34 +429,55 @@ static int fio_netio_setup_listen(struct thread_data *td, short port)
 static int fio_netio_init(struct thread_data *td)
 {
 	struct netio_data *nd = td->io_ops->data;
-	unsigned short port;
-	struct fio_file *f;
+	unsigned int port;
 	char host[64], buf[128];
-	char *sep;
-	int ret, i;
-
-	if (!td->total_file_size) {
-		log_err("fio: need size= set\n");
-		return 1;
-	}
+	char *sep, *portp, *modep;
+	int ret;
 
 	if (td_rw(td)) {
 		log_err("fio: network connections must be read OR write\n");
 		return 1;
 	}
-
-	strcpy(buf, td->filename);
-
-	sep = strchr(buf, ':');
-	if (!sep) {
-		log_err("fio: bad network host:port <<%s>>\n", td->filename);
+	if (td_random(td)) {
+		log_err("fio: network IO can't be random\n");
 		return 1;
 	}
 
+	strcpy(buf, td->o.filename);
+
+	sep = strchr(buf, '/');
+	if (!sep)
+		goto bad_host;
+
 	*sep = '\0';
 	sep++;
 	strcpy(host, buf);
-	port = atoi(sep);
+	if (!strlen(host))
+		goto bad_host;
+
+	modep = NULL;
+	portp = sep;
+	sep = strchr(portp, '/');
+	if (sep) {
+		*sep = '\0';
+		modep = sep + 1;
+	}
+		
+	port = strtol(portp, NULL, 10);
+	if (!port || port > 65535)
+		goto bad_host;
+
+	if (modep) {
+		if (!strncmp("tcp", modep, strlen(modep)) ||
+		    !strncmp("TCP", modep, strlen(modep)))
+			nd->net_protocol = IPPROTO_TCP;
+		else if (!strncmp("udp", modep, strlen(modep)) ||
+			 !strncmp("UDP", modep, strlen(modep)))
+			nd->net_protocol = IPPROTO_UDP;
+		else
+			goto bad_host;
+	} else
+		nd->net_protocol = IPPROTO_TCP;
 
 	if (td_read(td)) {
 		nd->send_to_net = 0;
@@ -257,19 +487,10 @@ static int fio_netio_init(struct thread_data *td)
 		ret = fio_netio_setup_connect(td, host, port);
 	}
 
-	if (ret)
-		return ret;
-
-	td->io_size = td->total_file_size;
-	td->total_io_size = td->io_size;
-
-	for_each_file(td, f, i) {
-		f->file_size = td->total_file_size / td->nr_files;
-		f->real_file_size = f->file_size;
-	}
-
-	td->nr_open_files = td->nr_files;
-	return 0;
+	return ret;
+bad_host:
+	log_err("fio: bad network host/port/protocol: %s\n", td->o.filename);
+	return 1;
 }
 
 static void fio_netio_cleanup(struct thread_data *td)
@@ -277,22 +498,68 @@ static void fio_netio_cleanup(struct thread_data *td)
 	struct netio_data *nd = td->io_ops->data;
 
 	if (nd) {
+		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]);
+
 		free(nd);
-		td->io_ops->data = NULL;
 	}
 }
 
 static int fio_netio_setup(struct thread_data *td)
 {
-	struct netio_data *nd = malloc(sizeof(*nd));
+	struct netio_data *nd;
+
+	if (!td->io_ops->data) {
+		nd = malloc(sizeof(*nd));;
+
+		memset(nd, 0, sizeof(*nd));
+		nd->listenfd = -1;
+		nd->pipes[0] = nd->pipes[1] = -1;
+		td->io_ops->data = nd;
+	}
 
-	memset(nd, 0, sizeof(*nd));
-	nd->listenfd = -1;
-	td->io_ops->data = nd;
 	return 0;
 }
 
-static struct ioengine_ops ioengine = {
+#ifdef FIO_HAVE_SPLICE
+static int fio_netio_setup_splice(struct thread_data *td)
+{
+	struct netio_data *nd;
+
+	fio_netio_setup(td);
+
+	nd = td->io_ops->data;
+	if (nd) {
+		if (pipe(nd->pipes) < 0)
+			return 1;
+
+		nd->use_splice = 1;
+		return 0;
+	}
+
+	return 1;
+}
+
+static struct ioengine_ops ioengine_splice = {
+	.name		= "netsplice",
+	.version	= FIO_IOOPS_VERSION,
+	.prep		= fio_netio_prep,
+	.queue		= fio_netio_queue,
+	.setup		= fio_netio_setup_splice,
+	.init		= fio_netio_init,
+	.cleanup	= fio_netio_cleanup,
+	.open_file	= fio_netio_open_file,
+	.close_file	= generic_close_file,
+	.flags		= FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
+			  FIO_SIGQUIT,
+};
+#endif
+
+static struct ioengine_ops ioengine_rw = {
 	.name		= "net",
 	.version	= FIO_IOOPS_VERSION,
 	.prep		= fio_netio_prep,
@@ -302,15 +569,22 @@ static struct ioengine_ops ioengine = {
 	.cleanup	= fio_netio_cleanup,
 	.open_file	= fio_netio_open_file,
 	.close_file	= generic_close_file,
-	.flags		= FIO_SYNCIO | FIO_DISKLESSIO,
+	.flags		= FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
+			  FIO_SIGQUIT,
 };
 
 static void fio_init fio_netio_register(void)
 {
-	register_ioengine(&ioengine);
+	register_ioengine(&ioengine_rw);
+#ifdef FIO_HAVE_SPLICE
+	register_ioengine(&ioengine_splice);
+#endif
 }
 
 static void fio_exit fio_netio_unregister(void)
 {
-	unregister_ioengine(&ioengine);
+	unregister_ioengine(&ioengine_rw);
+#ifdef FIO_HAVE_SPLICE
+	unregister_ioengine(&ioengine_splice);
+#endif
 }