[fio.git] / engines / net.c

/*
 * net engine
 *
 * IO engine that reads/writes to/from sockets.
 *
 */
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <assert.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netdb.h>
#include <sys/poll.h>

#include "../fio.h"

struct netio_data {
	int listenfd;
	int send_to_net;
	int use_splice;
	int pipes[2];
	char host[64];
	struct sockaddr_in addr;
};

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
	 */
	if ((nd->send_to_net && io_u->ddir == DDIR_READ) ||
	    (!nd->send_to_net && io_u->ddir == DDIR_WRITE)) {
		td_verror(td, EINVAL, "bad direction");
		return 1;
	}
		
	if (io_u->ddir == DDIR_SYNC)
		return 0;
	if (io_u->offset == f->last_completed_pos)
		return 0;

	/*
	 * If offset is different from last end position, it's a seek.
	 * As network io is purely sequential, we don't allow seeks.
	 */
	td_verror(td, EINVAL, "cannot seek");
	return 1;
}

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;
}

static int fio_netio_send(struct thread_data *td, struct io_u *io_u)
{
	int flags = 0;

	/*
	 * if we are going to write more, set MSG_MORE
	 */
	if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size)
		flags = MSG_MORE;

	return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
}

static int fio_netio_recv(struct io_u *io_u)
{
	int flags = MSG_WAITALL;

	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 netio_data *nd = td->io_ops->data;
	int ret;

	if (io_u->ddir == DDIR_WRITE) {
		if (nd->use_splice)
			ret = fio_netio_splice_out(td, io_u);
		else
			ret = fio_netio_send(td, io_u);
	} else if (io_u->ddir == DDIR_READ) {
		if (nd->use_splice)
			ret = fio_netio_splice_in(td, io_u);
		else
			ret = fio_netio_recv(io_u);
	} else
		ret = 0;	/* must be a SYNC */

	if (ret != (int) io_u->xfer_buflen) {
		if (ret >= 0) {
			io_u->resid = io_u->xfer_buflen - ret;
			io_u->error = 0;
			return FIO_Q_COMPLETED;
		} else
			io_u->error = errno;
	}

	if (io_u->error)
		td_verror(td, io_u->error, "xfer");

	return FIO_Q_COMPLETED;
}

static int fio_netio_connect(struct thread_data *td, struct fio_file *f)
{
	struct netio_data *nd = td->io_ops->data;

	f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
	if (f->fd < 0) {
		td_verror(td, errno, "socket");
		return 1;
	}

	if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
		td_verror(td, errno, "connect");
		return 1;
	}

	return 0;
}

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;

	log_info("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;

			td_verror(td, errno, "poll");
			break;
		} else if (!ret)
			continue;

		/*
		 * should be impossible
		 */
		if (!(pfd.revents & POLLIN))
			continue;

		f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
		if (f->fd < 0) {
			td_verror(td, errno, "accept");
			return 1;
		}
		break;
	}

	return 0;
}

static int fio_netio_open_file(struct thread_data *td, struct fio_file *f)
{
	if (td_read(td))
		return fio_netio_accept(td, f);
	else
		return fio_netio_connect(td, f);
}

static int fio_netio_setup_connect(struct thread_data *td, const char *host,
				   unsigned short port)
{
	struct netio_data *nd = td->io_ops->data;

	nd->addr.sin_family = AF_INET;
	nd->addr.sin_port = htons(port);

	if (inet_aton(host, &nd->addr.sin_addr) != 1) {
		struct hostent *hent;

		hent = gethostbyname(host);
		if (!hent) {
			td_verror(td, errno, "gethostbyname");
			return 1;
		}

		memcpy(&nd->addr.sin_addr, hent->h_addr, 4);
	}

	return 0;
}

static int fio_netio_setup_listen(struct thread_data *td, short port)
{
	struct netio_data *nd = td->io_ops->data;
	int fd, opt;

	fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
	if (fd < 0) {
		td_verror(td, errno, "socket");
		return 1;
	}

	opt = 1;
	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
		td_verror(td, errno, "setsockopt");
		return 1;
	}
#ifdef SO_REUSEPORT
	if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
		td_verror(td, errno, "setsockopt");
		return 1;
	}
#endif

	nd->addr.sin_family = AF_INET;
	nd->addr.sin_addr.s_addr = htonl(INADDR_ANY);
	nd->addr.sin_port = htons(port);

	if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
		td_verror(td, errno, "bind");
		return 1;
	}
	if (listen(fd, 1) < 0) {
		td_verror(td, errno, "listen");
		return 1;
	}

	nd->listenfd = fd;
	return 0;
}

static int fio_netio_init(struct thread_data *td)
{
	struct netio_data *nd = td->io_ops->data;
	unsigned short port;
	char host[64], buf[128];
	char *sep;
	int ret;

	if (td_rw(td)) {
		log_err("fio: network connections must be read OR write\n");
		return 1;
	}
	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) {
		log_err("fio: bad network host/port <<%s>>\n", td->o.filename);
		return 1;
	}

	*sep = '\0';
	sep++;
	strcpy(host, buf);
	port = atoi(sep);

	if (td_read(td)) {
		nd->send_to_net = 0;
		ret = fio_netio_setup_listen(td, port);
	} else {
		nd->send_to_net = 1;
		ret = fio_netio_setup_connect(td, host, port);
	}

	return ret;
}

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;

	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;
	}

	return 0;
}

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_rw = {
	.name		= "net",
	.version	= FIO_IOOPS_VERSION,
	.prep		= fio_netio_prep,
	.queue		= fio_netio_queue,
	.setup		= fio_netio_setup,
	.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,
};

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,
};

static void fio_init fio_netio_register(void)
{
	register_ioengine(&ioengine_rw);
	register_ioengine(&ioengine_splice);
}

static void fio_exit fio_netio_unregister(void)
{
	unregister_ioengine(&ioengine_rw);
	unregister_ioengine(&ioengine_splice);
}
Commit	Line	Data
ed92ac0c	1	/*
da751ca9 JA	2	* net engine
	3	*
	4	* IO engine that reads/writes to/from sockets.
	5	*
ed92ac0c JA	6	*/
	7	#include <stdio.h>
	8	#include <stdlib.h>
	9	#include <unistd.h>
	10	#include <errno.h>
	11	#include <assert.h>
	12	#include <netinet/in.h>
	13	#include <arpa/inet.h>
	14	#include <netdb.h>
5fdd124a	15	#include <sys/poll.h>
ed92ac0c JA	16
ed92ac0c JA	17	#include "../fio.h"
ed92ac0c	18
b5af8293 JA	19	struct netio_data {
	20	int listenfd;
	21	int send_to_net;
9cce02e8 JA	22	int use_splice;
9cce02e8 JA	23	int pipes[2];
b5af8293 JA	24	char host[64];
	25	struct sockaddr_in addr;
	26	};
ed92ac0c JA	27
	28	static int fio_netio_prep(struct thread_data td, struct io_u io_u)
	29	{
b5af8293	30	struct netio_data *nd = td->io_ops->data;
ed92ac0c JA	31	struct fio_file *f = io_u->file;
ed92ac0c JA	32
7a6499da JA	33	/*
	34	* Make sure we don't see spurious reads to a receiver, and vice versa
	35	*/
b5af8293 JA	36	if ((nd->send_to_net && io_u->ddir == DDIR_READ) \|\|
b5af8293 JA	37	(!nd->send_to_net && io_u->ddir == DDIR_WRITE)) {
e1161c32	38	td_verror(td, EINVAL, "bad direction");
7a6499da	39	return 1;
ed92ac0c	40	}
7a6499da	41
ed92ac0c JA	42	if (io_u->ddir == DDIR_SYNC)
	43	return 0;
	44	if (io_u->offset == f->last_completed_pos)
	45	return 0;
	46
e01547d2 JA	47	/*
	48	* If offset is different from last end position, it's a seek.
	49	* As network io is purely sequential, we don't allow seeks.
	50	*/
e1161c32	51	td_verror(td, EINVAL, "cannot seek");
ed92ac0c JA	52	return 1;
	53	}
	54
cd963e18	55	static int splice_io_u(int fdin, int fdout, unsigned int len)
ed92ac0c	56	{
9cce02e8	57	int bytes = 0;
7a6499da	58
9cce02e8	59	while (len) {
cd963e18	60	int ret = splice(fdin, NULL, fdout, NULL, len, 0);
9cce02e8 JA	61
	62	if (ret < 0) {
	63	if (!bytes)
	64	bytes = ret;
	65
	66	break;
	67	} else if (!ret)
	68	break;
	69
	70	bytes += ret;
f657a2fb	71	len -= ret;
9cce02e8 JA	72	}
	73
	74	return bytes;
	75	}
	76
	77	/*
cd963e18	78	* Receive bytes from a socket and fill them into the internal pipe
9cce02e8	79	*/
cd963e18	80	static int splice_in(struct thread_data td, struct io_u io_u)
9cce02e8 JA	81	{
9cce02e8 JA	82	struct netio_data *nd = td->io_ops->data;
9cce02e8	83
cd963e18	84	return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen);
9cce02e8 JA	85	}
	86
	87	/*
cd963e18	88	* Transmit 'len' bytes from the internal pipe
9cce02e8	89	*/
cd963e18 JA	90	static int splice_out(struct thread_data td, struct io_u io_u,
cd963e18 JA	91	unsigned int len)
9cce02e8 JA	92	{
9cce02e8 JA	93	struct netio_data *nd = td->io_ops->data;
cd963e18 JA	94
	95	return splice_io_u(nd->pipes[0], io_u->file->fd, len);
	96	}
	97
	98	static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len)
	99	{
9cce02e8 JA	100	struct iovec iov = {
	101	.iov_base = io_u->xfer_buf,
	102	.iov_len = len,
	103	};
	104	int bytes = 0;
	105
	106	while (iov.iov_len) {
cd963e18	107	int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE);
9cce02e8 JA	108
	109	if (ret < 0) {
	110	if (!bytes)
	111	bytes = ret;
	112	break;
	113	} else if (!ret)
	114	break;
	115
	116	iov.iov_len -= ret;
cd963e18	117	iov.iov_base += ret;
f657a2fb	118	bytes += ret;
9cce02e8 JA	119	}
	120
	121	return bytes;
cd963e18	122
9cce02e8 JA	123	}
	124
	125	/*
cd963e18	126	* vmsplice() pipe to io_u buffer
9cce02e8	127	*/
cd963e18 JA	128	static int vmsplice_io_u_out(struct thread_data td, struct io_u io_u,
cd963e18 JA	129	unsigned int len)
9cce02e8 JA	130	{
9cce02e8 JA	131	struct netio_data *nd = td->io_ops->data;
9cce02e8	132
cd963e18 JA	133	return vmsplice_io_u(io_u, nd->pipes[0], len);
cd963e18 JA	134	}
9cce02e8	135
cd963e18 JA	136	/*
	137	* vmsplice() io_u to pipe
	138	*/
	139	static int vmsplice_io_u_in(struct thread_data td, struct io_u io_u)
	140	{
	141	struct netio_data *nd = td->io_ops->data;
ed92ac0c	142
cd963e18	143	return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen);
9cce02e8 JA	144	}
9cce02e8 JA	145
cd963e18 JA	146	/*
	147	* splice receive - transfer socket data into a pipe using splice, then map
	148	* that pipe data into the io_u using vmsplice.
	149	*/
9cce02e8 JA	150	static int fio_netio_splice_in(struct thread_data td, struct io_u io_u)
	151	{
	152	int ret;
	153
	154	ret = splice_in(td, io_u);
cd963e18 JA	155	if (ret > 0)
cd963e18 JA	156	return vmsplice_io_u_out(td, io_u, ret);
9cce02e8	157
cd963e18	158	return ret;
9cce02e8 JA	159	}
9cce02e8 JA	160
cd963e18 JA	161	/*
	162	* splice transmit - map data from the io_u into a pipe by using vmsplice,
	163	* then transfer that pipe to a socket using splice.
	164	*/
9cce02e8 JA	165	static int fio_netio_splice_out(struct thread_data td, struct io_u io_u)
	166	{
	167	int ret;
	168
	169	ret = vmsplice_io_u_in(td, io_u);
cd963e18 JA	170	if (ret > 0)
cd963e18 JA	171	return splice_out(td, io_u, ret);
9cce02e8	172
cd963e18	173	return ret;
9cce02e8 JA	174	}
	175
	176	static int fio_netio_send(struct thread_data td, struct io_u io_u)
	177	{
	178	int flags = 0;
	179
	180	/*
	181	* if we are going to write more, set MSG_MORE
	182	*/
	183	if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size)
	184	flags = MSG_MORE;
	185
	186	return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
	187	}
	188
	189	static int fio_netio_recv(struct io_u *io_u)
	190	{
	191	int flags = MSG_WAITALL;
	192
	193	return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
	194	}
	195
	196	static int fio_netio_queue(struct thread_data td, struct io_u io_u)
	197	{
	198	struct netio_data *nd = td->io_ops->data;
	199	int ret;
	200
	201	if (io_u->ddir == DDIR_WRITE) {
	202	if (nd->use_splice)
	203	ret = fio_netio_splice_out(td, io_u);
	204	else
	205	ret = fio_netio_send(td, io_u);
d4f12dd0	206	} else if (io_u->ddir == DDIR_READ) {
9cce02e8 JA	207	if (nd->use_splice)
	208	ret = fio_netio_splice_in(td, io_u);
	209	else
	210	ret = fio_netio_recv(io_u);
d4f12dd0	211	} else
7a6499da	212	ret = 0; /* must be a SYNC */
ed92ac0c	213
cec6b55d	214	if (ret != (int) io_u->xfer_buflen) {
22819ec2	215	if (ret >= 0) {
cec6b55d JA	216	io_u->resid = io_u->xfer_buflen - ret;
cec6b55d JA	217	io_u->error = 0;
36167d82	218	return FIO_Q_COMPLETED;
ed92ac0c JA	219	} else
	220	io_u->error = errno;
	221	}
	222
36167d82	223	if (io_u->error)
e1161c32	224	td_verror(td, io_u->error, "xfer");
ed92ac0c	225
36167d82	226	return FIO_Q_COMPLETED;
ed92ac0c JA	227	}
ed92ac0c JA	228
b5af8293	229	static int fio_netio_connect(struct thread_data td, struct fio_file f)
ed92ac0c	230	{
b5af8293	231	struct netio_data *nd = td->io_ops->data;
ed92ac0c	232
b5af8293 JA	233	f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
	234	if (f->fd < 0) {
	235	td_verror(td, errno, "socket");
	236	return 1;
ed92ac0c JA	237	}
ed92ac0c JA	238
b5af8293 JA	239	if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
	240	td_verror(td, errno, "connect");
	241	return 1;
ed92ac0c JA	242	}
	243
	244	return 0;
ed92ac0c JA	245	}
ed92ac0c JA	246
b5af8293	247	static int fio_netio_accept(struct thread_data td, struct fio_file f)
5fdd124a	248	{
b5af8293 JA	249	struct netio_data *nd = td->io_ops->data;
b5af8293 JA	250	socklen_t socklen = sizeof(nd->addr);
5fdd124a	251	struct pollfd pfd;
b5af8293	252	int ret;
5fdd124a	253
6d86144d	254	log_info("fio: waiting for connection\n");
5fdd124a JA	255
	256	/*
	257	* Accept loop. poll for incoming events, accept them. Repeat until we
	258	* have all connections.
	259	*/
b5af8293 JA	260	while (!td->terminate) {
b5af8293 JA	261	pfd.fd = nd->listenfd;
5fdd124a JA	262	pfd.events = POLLIN;
	263
	264	ret = poll(&pfd, 1, -1);
	265	if (ret < 0) {
	266	if (errno == EINTR)
	267	continue;
	268
e1161c32	269	td_verror(td, errno, "poll");
5fdd124a JA	270	break;
	271	} else if (!ret)
	272	continue;
	273
0c09442b JA	274	/*
	275	* should be impossible
	276	*/
	277	if (!(pfd.revents & POLLIN))
	278	continue;
	279
b5af8293 JA	280	f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
	281	if (f->fd < 0) {
	282	td_verror(td, errno, "accept");
	283	return 1;
	284	}
	285	break;
	286	}
5fdd124a	287
b5af8293 JA	288	return 0;
	289	}
	290
b5af8293 JA	291	static int fio_netio_open_file(struct thread_data td, struct fio_file f)
	292	{
	293	if (td_read(td))
	294	return fio_netio_accept(td, f);
	295	else
	296	return fio_netio_connect(td, f);
	297	}
	298
	299	static int fio_netio_setup_connect(struct thread_data td, const char host,
	300	unsigned short port)
	301	{
	302	struct netio_data *nd = td->io_ops->data;
	303
	304	nd->addr.sin_family = AF_INET;
	305	nd->addr.sin_port = htons(port);
	306
	307	if (inet_aton(host, &nd->addr.sin_addr) != 1) {
	308	struct hostent *hent;
	309
	310	hent = gethostbyname(host);
	311	if (!hent) {
	312	td_verror(td, errno, "gethostbyname");
	313	return 1;
5fdd124a	314	}
b5af8293 JA	315
b5af8293 JA	316	memcpy(&nd->addr.sin_addr, hent->h_addr, 4);
5fdd124a JA	317	}
	318
	319	return 0;
	320	}
	321
b5af8293	322	static int fio_netio_setup_listen(struct thread_data *td, short port)
ed92ac0c	323	{
b5af8293	324	struct netio_data *nd = td->io_ops->data;
5fdd124a	325	int fd, opt;
ed92ac0c	326
6bedbfaf	327	fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
ed92ac0c	328	if (fd < 0) {
e1161c32	329	td_verror(td, errno, "socket");
ed92ac0c JA	330	return 1;
	331	}
	332
	333	opt = 1;
	334	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
e1161c32	335	td_verror(td, errno, "setsockopt");
ed92ac0c JA	336	return 1;
ed92ac0c JA	337	}
6bedbfaf JA	338	#ifdef SO_REUSEPORT
6bedbfaf JA	339	if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
e1161c32	340	td_verror(td, errno, "setsockopt");
6bedbfaf JA	341	return 1;
	342	}
	343	#endif
ed92ac0c	344
b5af8293 JA	345	nd->addr.sin_family = AF_INET;
	346	nd->addr.sin_addr.s_addr = htonl(INADDR_ANY);
	347	nd->addr.sin_port = htons(port);
ed92ac0c	348
b5af8293	349	if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
e1161c32	350	td_verror(td, errno, "bind");
ed92ac0c JA	351	return 1;
	352	}
	353	if (listen(fd, 1) < 0) {
e1161c32	354	td_verror(td, errno, "listen");
ed92ac0c JA	355	return 1;
	356	}
	357
b5af8293 JA	358	nd->listenfd = fd;
b5af8293 JA	359	return 0;
ed92ac0c JA	360	}
ed92ac0c JA	361
9bec88e1	362	static int fio_netio_init(struct thread_data *td)
ed92ac0c	363	{
b5af8293	364	struct netio_data *nd = td->io_ops->data;
e01547d2	365	unsigned short port;
b5af8293	366	char host[64], buf[128];
ed92ac0c	367	char *sep;
af52b345	368	int ret;
ed92ac0c	369
413dd459	370	if (td_rw(td)) {
ed92ac0c JA	371	log_err("fio: network connections must be read OR write\n");
	372	return 1;
	373	}
16d55aae JA	374	if (td_random(td)) {
	375	log_err("fio: network IO can't be random\n");
	376	return 1;
	377	}
ed92ac0c	378
2dc1bbeb	379	strcpy(buf, td->o.filename);
ed92ac0c	380
9f9214f2	381	sep = strchr(buf, '/');
ed92ac0c	382	if (!sep) {
2dc1bbeb	383	log_err("fio: bad network host/port <<%s>>\n", td->o.filename);
ed92ac0c JA	384	return 1;
	385	}
	386
	387	*sep = '\0';
	388	sep++;
	389	strcpy(host, buf);
e01547d2	390	port = atoi(sep);
ed92ac0c	391
413dd459	392	if (td_read(td)) {
b5af8293	393	nd->send_to_net = 0;
ed92ac0c JA	394	ret = fio_netio_setup_listen(td, port);
ed92ac0c JA	395	} else {
b5af8293	396	nd->send_to_net = 1;
ed92ac0c JA	397	ret = fio_netio_setup_connect(td, host, port);
	398	}
	399
7bb48f84	400	return ret;
ed92ac0c JA	401	}
ed92ac0c JA	402
b5af8293	403	static void fio_netio_cleanup(struct thread_data *td)
9bec88e1	404	{
b5af8293 JA	405	struct netio_data *nd = td->io_ops->data;
	406
	407	if (nd) {
64b24cd8 JA	408	if (nd->listenfd != -1)
	409	close(nd->listenfd);
	410	if (nd->pipes[0] != -1)
	411	close(nd->pipes[0]);
	412	if (nd->pipes[1] != -1)
	413	close(nd->pipes[1]);
	414
b5af8293 JA	415	free(nd);
	416	td->io_ops->data = NULL;
	417	}
	418	}
	419
	420	static int fio_netio_setup(struct thread_data *td)
	421	{
7bb48f84	422	struct netio_data *nd;
7bb48f84 JA	423
	424	if (!td->io_ops->data) {
	425	nd = malloc(sizeof(*nd));;
	426
	427	memset(nd, 0, sizeof(*nd));
	428	nd->listenfd = -1;
64b24cd8	429	nd->pipes[0] = nd->pipes[1] = -1;
7bb48f84	430	td->io_ops->data = nd;
7bb48f84	431	}
b5af8293	432
9bec88e1 JA	433	return 0;
	434	}
	435
9cce02e8 JA	436	static int fio_netio_setup_splice(struct thread_data *td)
	437	{
	438	struct netio_data *nd;
	439
	440	fio_netio_setup(td);
	441
	442	nd = td->io_ops->data;
	443	if (nd) {
	444	if (pipe(nd->pipes) < 0)
	445	return 1;
	446
	447	nd->use_splice = 1;
	448	return 0;
	449	}
	450
	451	return 1;
	452	}
	453
	454	static struct ioengine_ops ioengine_rw = {
ed92ac0c JA	455	.name = "net",
ed92ac0c JA	456	.version = FIO_IOOPS_VERSION,
ed92ac0c JA	457	.prep = fio_netio_prep,
ed92ac0c JA	458	.queue = fio_netio_queue,
ed92ac0c	459	.setup = fio_netio_setup,
9bec88e1	460	.init = fio_netio_init,
b5af8293 JA	461	.cleanup = fio_netio_cleanup,
	462	.open_file = fio_netio_open_file,
	463	.close_file = generic_close_file,
dc372f07	464	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_UNIDIR,
ed92ac0c JA	465	};
ed92ac0c JA	466
9cce02e8 JA	467	static struct ioengine_ops ioengine_splice = {
	468	.name = "netsplice",
	469	.version = FIO_IOOPS_VERSION,
	470	.prep = fio_netio_prep,
	471	.queue = fio_netio_queue,
	472	.setup = fio_netio_setup_splice,
	473	.init = fio_netio_init,
	474	.cleanup = fio_netio_cleanup,
	475	.open_file = fio_netio_open_file,
	476	.close_file = generic_close_file,
b67740d3	477	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_UNIDIR,
9cce02e8 JA	478	};
9cce02e8 JA	479
ed92ac0c JA	480	static void fio_init fio_netio_register(void)
ed92ac0c JA	481	{
9cce02e8 JA	482	register_ioengine(&ioengine_rw);
9cce02e8 JA	483	register_ioengine(&ioengine_splice);
ed92ac0c JA	484	}
	485
	486	static void fio_exit fio_netio_unregister(void)
	487	{
9cce02e8 JA	488	unregister_ioengine(&ioengine_rw);
9cce02e8 JA	489	unregister_ioengine(&ioengine_splice);
ed92ac0c	490	}