[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 <sys/types.h>
#include <sys/socket.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;

	/*
	 * 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;
	}
		
	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)
{
	int flags = 0;

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

	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;

	fio_ro_check(td, io_u);

	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 int 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)
		goto bad_host;

	*sep = '\0';
	sep++;
	strcpy(host, buf);
	if (!strlen(host))
		goto bad_host;

	port = strtol(sep, NULL, 10);
	if (!port || port > 65535)
		goto bad_host;

	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;
bad_host:
	log_err("fio: bad network host/port: %s\n", td->o.filename);
	return 1;
}

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);
	}
}

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;
}

#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,
	.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 |
			  FIO_SIGQUIT,
};

static void fio_init fio_netio_register(void)
{
	register_ioengine(&ioengine_rw);
#ifdef FIO_HAVE_SPLICE
	register_ioengine(&ioengine_splice);
#endif
}

static void fio_exit fio_netio_unregister(void)
{
	unregister_ioengine(&ioengine_rw);
#ifdef FIO_HAVE_SPLICE
	unregister_ioengine(&ioengine_splice);
#endif
}
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>
7292056a JA	16	#include <sys/types.h>
7292056a JA	17	#include <sys/socket.h>
ed92ac0c JA	18
ed92ac0c JA	19	#include "../fio.h"
ed92ac0c	20
b5af8293 JA	21	struct netio_data {
	22	int listenfd;
	23	int send_to_net;
9cce02e8 JA	24	int use_splice;
9cce02e8 JA	25	int pipes[2];
b5af8293 JA	26	char host[64];
	27	struct sockaddr_in addr;
	28	};
ed92ac0c JA	29
	30	static int fio_netio_prep(struct thread_data td, struct io_u io_u)
	31	{
b5af8293	32	struct netio_data *nd = td->io_ops->data;
ed92ac0c	33
7a6499da JA	34	/*
	35	* Make sure we don't see spurious reads to a receiver, and vice versa
	36	*/
b5af8293 JA	37	if ((nd->send_to_net && io_u->ddir == DDIR_READ) \|\|
b5af8293 JA	38	(!nd->send_to_net && io_u->ddir == DDIR_WRITE)) {
e1161c32	39	td_verror(td, EINVAL, "bad direction");
7a6499da	40	return 1;
ed92ac0c	41	}
7a6499da	42
f85ac25a	43	return 0;
ed92ac0c JA	44	}
ed92ac0c JA	45
5921e80c	46	#ifdef FIO_HAVE_SPLICE
cd963e18	47	static int splice_io_u(int fdin, int fdout, unsigned int len)
ed92ac0c	48	{
9cce02e8	49	int bytes = 0;
7a6499da	50
9cce02e8	51	while (len) {
cd963e18	52	int ret = splice(fdin, NULL, fdout, NULL, len, 0);
9cce02e8 JA	53
	54	if (ret < 0) {
	55	if (!bytes)
	56	bytes = ret;
	57
	58	break;
	59	} else if (!ret)
	60	break;
	61
	62	bytes += ret;
f657a2fb	63	len -= ret;
9cce02e8 JA	64	}
	65
	66	return bytes;
	67	}
	68
	69	/*
cd963e18	70	* Receive bytes from a socket and fill them into the internal pipe
9cce02e8	71	*/
cd963e18	72	static int splice_in(struct thread_data td, struct io_u io_u)
9cce02e8 JA	73	{
9cce02e8 JA	74	struct netio_data *nd = td->io_ops->data;
9cce02e8	75
cd963e18	76	return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen);
9cce02e8 JA	77	}
	78
	79	/*
cd963e18	80	* Transmit 'len' bytes from the internal pipe
9cce02e8	81	*/
cd963e18 JA	82	static int splice_out(struct thread_data td, struct io_u io_u,
cd963e18 JA	83	unsigned int len)
9cce02e8 JA	84	{
9cce02e8 JA	85	struct netio_data *nd = td->io_ops->data;
cd963e18 JA	86
	87	return splice_io_u(nd->pipes[0], io_u->file->fd, len);
	88	}
	89
	90	static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len)
	91	{
9cce02e8 JA	92	struct iovec iov = {
	93	.iov_base = io_u->xfer_buf,
	94	.iov_len = len,
	95	};
	96	int bytes = 0;
	97
	98	while (iov.iov_len) {
cd963e18	99	int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE);
9cce02e8 JA	100
	101	if (ret < 0) {
	102	if (!bytes)
	103	bytes = ret;
	104	break;
	105	} else if (!ret)
	106	break;
	107
	108	iov.iov_len -= ret;
cd963e18	109	iov.iov_base += ret;
f657a2fb	110	bytes += ret;
9cce02e8 JA	111	}
	112
	113	return bytes;
cd963e18	114
9cce02e8 JA	115	}
	116
	117	/*
cd963e18	118	* vmsplice() pipe to io_u buffer
9cce02e8	119	*/
cd963e18 JA	120	static int vmsplice_io_u_out(struct thread_data td, struct io_u io_u,
cd963e18 JA	121	unsigned int len)
9cce02e8 JA	122	{
9cce02e8 JA	123	struct netio_data *nd = td->io_ops->data;
9cce02e8	124
cd963e18 JA	125	return vmsplice_io_u(io_u, nd->pipes[0], len);
cd963e18 JA	126	}
9cce02e8	127
cd963e18 JA	128	/*
	129	* vmsplice() io_u to pipe
	130	*/
	131	static int vmsplice_io_u_in(struct thread_data td, struct io_u io_u)
	132	{
	133	struct netio_data *nd = td->io_ops->data;
ed92ac0c	134
cd963e18	135	return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen);
9cce02e8 JA	136	}
9cce02e8 JA	137
cd963e18 JA	138	/*
	139	* splice receive - transfer socket data into a pipe using splice, then map
	140	* that pipe data into the io_u using vmsplice.
	141	*/
9cce02e8 JA	142	static int fio_netio_splice_in(struct thread_data td, struct io_u io_u)
	143	{
	144	int ret;
	145
	146	ret = splice_in(td, io_u);
cd963e18 JA	147	if (ret > 0)
cd963e18 JA	148	return vmsplice_io_u_out(td, io_u, ret);
9cce02e8	149
cd963e18	150	return ret;
9cce02e8 JA	151	}
9cce02e8 JA	152
cd963e18 JA	153	/*
	154	* splice transmit - map data from the io_u into a pipe by using vmsplice,
	155	* then transfer that pipe to a socket using splice.
	156	*/
9cce02e8 JA	157	static int fio_netio_splice_out(struct thread_data td, struct io_u io_u)
	158	{
	159	int ret;
	160
	161	ret = vmsplice_io_u_in(td, io_u);
cd963e18 JA	162	if (ret > 0)
cd963e18 JA	163	return splice_out(td, io_u, ret);
9cce02e8	164
cd963e18	165	return ret;
9cce02e8	166	}
5921e80c JA	167	#else
	168	static int fio_netio_splice_in(struct thread_data td, struct io_u io_u)
	169	{
af8771b9	170	errno = EOPNOTSUPP;
5921e80c JA	171	return -1;
	172	}
	173
	174	static int fio_netio_splice_out(struct thread_data td, struct io_u io_u)
	175	{
af8771b9	176	errno = EOPNOTSUPP;
5921e80c JA	177	return -1;
	178	}
	179	#endif
9cce02e8 JA	180
	181	static int fio_netio_send(struct thread_data td, struct io_u io_u)
	182	{
	183	int flags = 0;
	184
	185	/*
	186	* if we are going to write more, set MSG_MORE
	187	*/
5921e80c	188	#ifdef MSG_MORE
9cce02e8 JA	189	if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size)
9cce02e8 JA	190	flags = MSG_MORE;
5921e80c	191	#endif
9cce02e8 JA	192
	193	return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
	194	}
	195
	196	static int fio_netio_recv(struct io_u *io_u)
	197	{
	198	int flags = MSG_WAITALL;
	199
	200	return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
	201	}
	202
	203	static int fio_netio_queue(struct thread_data td, struct io_u io_u)
	204	{
	205	struct netio_data *nd = td->io_ops->data;
	206	int ret;
	207
7101d9c2 JA	208	fio_ro_check(td, io_u);
7101d9c2 JA	209
9cce02e8 JA	210	if (io_u->ddir == DDIR_WRITE) {
	211	if (nd->use_splice)
	212	ret = fio_netio_splice_out(td, io_u);
	213	else
	214	ret = fio_netio_send(td, io_u);
d4f12dd0	215	} else if (io_u->ddir == DDIR_READ) {
9cce02e8 JA	216	if (nd->use_splice)
	217	ret = fio_netio_splice_in(td, io_u);
	218	else
	219	ret = fio_netio_recv(io_u);
d4f12dd0	220	} else
7a6499da	221	ret = 0; /* must be a SYNC */
ed92ac0c	222
cec6b55d	223	if (ret != (int) io_u->xfer_buflen) {
22819ec2	224	if (ret >= 0) {
cec6b55d JA	225	io_u->resid = io_u->xfer_buflen - ret;
cec6b55d JA	226	io_u->error = 0;
36167d82	227	return FIO_Q_COMPLETED;
ed92ac0c JA	228	} else
	229	io_u->error = errno;
	230	}
	231
36167d82	232	if (io_u->error)
e1161c32	233	td_verror(td, io_u->error, "xfer");
ed92ac0c	234
36167d82	235	return FIO_Q_COMPLETED;
ed92ac0c JA	236	}
ed92ac0c JA	237
b5af8293	238	static int fio_netio_connect(struct thread_data td, struct fio_file f)
ed92ac0c	239	{
b5af8293	240	struct netio_data *nd = td->io_ops->data;
ed92ac0c	241
b5af8293 JA	242	f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
	243	if (f->fd < 0) {
	244	td_verror(td, errno, "socket");
	245	return 1;
ed92ac0c JA	246	}
ed92ac0c JA	247
b5af8293 JA	248	if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
	249	td_verror(td, errno, "connect");
	250	return 1;
ed92ac0c JA	251	}
	252
	253	return 0;
ed92ac0c JA	254	}
ed92ac0c JA	255
b5af8293	256	static int fio_netio_accept(struct thread_data td, struct fio_file f)
5fdd124a	257	{
b5af8293 JA	258	struct netio_data *nd = td->io_ops->data;
b5af8293 JA	259	socklen_t socklen = sizeof(nd->addr);
5fdd124a	260	struct pollfd pfd;
b5af8293	261	int ret;
5fdd124a	262
6d86144d	263	log_info("fio: waiting for connection\n");
5fdd124a JA	264
	265	/*
	266	* Accept loop. poll for incoming events, accept them. Repeat until we
	267	* have all connections.
	268	*/
b5af8293 JA	269	while (!td->terminate) {
b5af8293 JA	270	pfd.fd = nd->listenfd;
5fdd124a JA	271	pfd.events = POLLIN;
	272
	273	ret = poll(&pfd, 1, -1);
	274	if (ret < 0) {
	275	if (errno == EINTR)
	276	continue;
	277
e1161c32	278	td_verror(td, errno, "poll");
5fdd124a JA	279	break;
	280	} else if (!ret)
	281	continue;
	282
0c09442b JA	283	/*
	284	* should be impossible
	285	*/
	286	if (!(pfd.revents & POLLIN))
	287	continue;
	288
b5af8293 JA	289	f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
	290	if (f->fd < 0) {
	291	td_verror(td, errno, "accept");
	292	return 1;
	293	}
	294	break;
	295	}
5fdd124a	296
b5af8293 JA	297	return 0;
	298	}
	299
b5af8293 JA	300	static int fio_netio_open_file(struct thread_data td, struct fio_file f)
	301	{
	302	if (td_read(td))
	303	return fio_netio_accept(td, f);
	304	else
	305	return fio_netio_connect(td, f);
	306	}
	307
	308	static int fio_netio_setup_connect(struct thread_data td, const char host,
	309	unsigned short port)
	310	{
	311	struct netio_data *nd = td->io_ops->data;
	312
	313	nd->addr.sin_family = AF_INET;
	314	nd->addr.sin_port = htons(port);
	315
	316	if (inet_aton(host, &nd->addr.sin_addr) != 1) {
	317	struct hostent *hent;
	318
	319	hent = gethostbyname(host);
	320	if (!hent) {
	321	td_verror(td, errno, "gethostbyname");
	322	return 1;
5fdd124a	323	}
b5af8293 JA	324
b5af8293 JA	325	memcpy(&nd->addr.sin_addr, hent->h_addr, 4);
5fdd124a JA	326	}
	327
	328	return 0;
	329	}
	330
b5af8293	331	static int fio_netio_setup_listen(struct thread_data *td, short port)
ed92ac0c	332	{
b5af8293	333	struct netio_data *nd = td->io_ops->data;
5fdd124a	334	int fd, opt;
ed92ac0c	335
6bedbfaf	336	fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
ed92ac0c	337	if (fd < 0) {
e1161c32	338	td_verror(td, errno, "socket");
ed92ac0c JA	339	return 1;
	340	}
	341
	342	opt = 1;
	343	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
e1161c32	344	td_verror(td, errno, "setsockopt");
ed92ac0c JA	345	return 1;
ed92ac0c JA	346	}
6bedbfaf JA	347	#ifdef SO_REUSEPORT
6bedbfaf JA	348	if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
e1161c32	349	td_verror(td, errno, "setsockopt");
6bedbfaf JA	350	return 1;
	351	}
	352	#endif
ed92ac0c	353
b5af8293 JA	354	nd->addr.sin_family = AF_INET;
	355	nd->addr.sin_addr.s_addr = htonl(INADDR_ANY);
	356	nd->addr.sin_port = htons(port);
ed92ac0c	357
b5af8293	358	if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
e1161c32	359	td_verror(td, errno, "bind");
ed92ac0c JA	360	return 1;
	361	}
	362	if (listen(fd, 1) < 0) {
e1161c32	363	td_verror(td, errno, "listen");
ed92ac0c JA	364	return 1;
	365	}
	366
b5af8293 JA	367	nd->listenfd = fd;
b5af8293 JA	368	return 0;
ed92ac0c JA	369	}
ed92ac0c JA	370
9bec88e1	371	static int fio_netio_init(struct thread_data *td)
ed92ac0c	372	{
b5af8293	373	struct netio_data *nd = td->io_ops->data;
443662ef	374	unsigned int port;
b5af8293	375	char host[64], buf[128];
ed92ac0c	376	char *sep;
af52b345	377	int ret;
ed92ac0c	378
413dd459	379	if (td_rw(td)) {
ed92ac0c JA	380	log_err("fio: network connections must be read OR write\n");
	381	return 1;
	382	}
16d55aae JA	383	if (td_random(td)) {
	384	log_err("fio: network IO can't be random\n");
	385	return 1;
	386	}
ed92ac0c	387
2dc1bbeb	388	strcpy(buf, td->o.filename);
ed92ac0c	389
9f9214f2	390	sep = strchr(buf, '/');
443662ef JA	391	if (!sep)
443662ef JA	392	goto bad_host;
ed92ac0c JA	393
	394	*sep = '\0';
	395	sep++;
	396	strcpy(host, buf);
443662ef JA	397	if (!strlen(host))
	398	goto bad_host;
	399
	400	port = strtol(sep, NULL, 10);
	401	if (!port \|\| port > 65535)
	402	goto bad_host;
ed92ac0c	403
413dd459	404	if (td_read(td)) {
b5af8293	405	nd->send_to_net = 0;
ed92ac0c JA	406	ret = fio_netio_setup_listen(td, port);
ed92ac0c JA	407	} else {
b5af8293	408	nd->send_to_net = 1;
ed92ac0c JA	409	ret = fio_netio_setup_connect(td, host, port);
	410	}
	411
7bb48f84	412	return ret;
443662ef JA	413	bad_host:
	414	log_err("fio: bad network host/port: %s\n", td->o.filename);
	415	return 1;
ed92ac0c JA	416	}
ed92ac0c JA	417
b5af8293	418	static void fio_netio_cleanup(struct thread_data *td)
9bec88e1	419	{
b5af8293 JA	420	struct netio_data *nd = td->io_ops->data;
	421
	422	if (nd) {
64b24cd8 JA	423	if (nd->listenfd != -1)
	424	close(nd->listenfd);
	425	if (nd->pipes[0] != -1)
	426	close(nd->pipes[0]);
	427	if (nd->pipes[1] != -1)
	428	close(nd->pipes[1]);
	429
b5af8293	430	free(nd);
b5af8293 JA	431	}
	432	}
	433
	434	static int fio_netio_setup(struct thread_data *td)
	435	{
7bb48f84	436	struct netio_data *nd;
7bb48f84 JA	437
	438	if (!td->io_ops->data) {
	439	nd = malloc(sizeof(*nd));;
	440
	441	memset(nd, 0, sizeof(*nd));
	442	nd->listenfd = -1;
64b24cd8	443	nd->pipes[0] = nd->pipes[1] = -1;
7bb48f84	444	td->io_ops->data = nd;
7bb48f84	445	}
b5af8293	446
9bec88e1 JA	447	return 0;
	448	}
	449
5921e80c	450	#ifdef FIO_HAVE_SPLICE
9cce02e8 JA	451	static int fio_netio_setup_splice(struct thread_data *td)
	452	{
	453	struct netio_data *nd;
	454
	455	fio_netio_setup(td);
	456
	457	nd = td->io_ops->data;
	458	if (nd) {
	459	if (pipe(nd->pipes) < 0)
	460	return 1;
	461
	462	nd->use_splice = 1;
	463	return 0;
	464	}
	465
	466	return 1;
	467	}
	468
5921e80c JA	469	static struct ioengine_ops ioengine_splice = {
5921e80c JA	470	.name = "netsplice",
ed92ac0c	471	.version = FIO_IOOPS_VERSION,
ed92ac0c JA	472	.prep = fio_netio_prep,
ed92ac0c JA	473	.queue = fio_netio_queue,
5921e80c	474	.setup = fio_netio_setup_splice,
9bec88e1	475	.init = fio_netio_init,
b5af8293 JA	476	.cleanup = fio_netio_cleanup,
	477	.open_file = fio_netio_open_file,
	478	.close_file = generic_close_file,
ad830ed7 JA	479	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_UNIDIR \|
ad830ed7 JA	480	FIO_SIGQUIT,
ed92ac0c	481	};
5921e80c	482	#endif
ed92ac0c	483
5921e80c JA	484	static struct ioengine_ops ioengine_rw = {
5921e80c JA	485	.name = "net",
9cce02e8 JA	486	.version = FIO_IOOPS_VERSION,
	487	.prep = fio_netio_prep,
	488	.queue = fio_netio_queue,
5921e80c	489	.setup = fio_netio_setup,
9cce02e8 JA	490	.init = fio_netio_init,
	491	.cleanup = fio_netio_cleanup,
	492	.open_file = fio_netio_open_file,
	493	.close_file = generic_close_file,
ad830ed7 JA	494	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_UNIDIR \|
ad830ed7 JA	495	FIO_SIGQUIT,
9cce02e8 JA	496	};
9cce02e8 JA	497
ed92ac0c JA	498	static void fio_init fio_netio_register(void)
ed92ac0c JA	499	{
9cce02e8	500	register_ioengine(&ioengine_rw);
5921e80c	501	#ifdef FIO_HAVE_SPLICE
9cce02e8	502	register_ioengine(&ioengine_splice);
5921e80c	503	#endif
ed92ac0c JA	504	}
	505
	506	static void fio_exit fio_netio_unregister(void)
	507	{
9cce02e8	508	unregister_ioengine(&ioengine_rw);
5921e80c	509	#ifdef FIO_HAVE_SPLICE
9cce02e8	510	unregister_ioengine(&ioengine_splice);
5921e80c	511	#endif
ed92ac0c	512	}