[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 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)
				break;

			td_verror(td, errno, "poll");
			return -1;
		} else if (!ret)
			continue;

		break;
	}

	if (pfd.revents & events)
		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;

	/*
	 * 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)
{
	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 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

	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 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 || 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) {
		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 */

	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 {
			int err = errno;

			if (io_u->ddir == DDIR_WRITE && err == EMSGSIZE)
				return FIO_Q_BUSY;

			io_u->error = err;
		}
	}

	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;
	int type;

	if (nd->net_protocol == IPPROTO_TCP)
		type = SOCK_STREAM;
	else
		type = SOCK_DGRAM;

	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;
	}

	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);

	if (nd->net_protocol == IPPROTO_UDP) {
		f->fd = nd->listenfd;
		return 0;
	}

	log_info("fio: waiting for connection\n");

	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;
	}

	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, type;

	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;
	}

	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 (nd->net_protocol == IPPROTO_TCP && 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, *portp, *modep;
	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;

	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;
		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/protocol: %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	24	int use_splice;
414c2a3e	25	int net_protocol;
9cce02e8	26	int pipes[2];
b5af8293 JA	27	char host[64];
	28	struct sockaddr_in addr;
	29	};
ed92ac0c	30
371d456c JA	31	/*
	32	* Return -1 for error and 'nr events' for a positive number
	33	* of events
	34	*/
	35	static int poll_wait(struct thread_data *td, int fd, short events)
	36	{
	37	struct pollfd pfd;
	38	int ret;
	39
	40	while (!td->terminate) {
	41	pfd.fd = fd;
	42	pfd.events = events;
	43	ret = poll(&pfd, 1, -1);
	44	if (ret < 0) {
	45	if (errno == EINTR)
d5b388a5	46	break;
371d456c JA	47
	48	td_verror(td, errno, "poll");
	49	return -1;
	50	} else if (!ret)
	51	continue;
	52
	53	break;
	54	}
	55
	56	if (pfd.revents & events)
	57	return 1;
371d456c JA	58
	59	return -1;
	60	}
	61
ed92ac0c JA	62	static int fio_netio_prep(struct thread_data td, struct io_u io_u)
ed92ac0c JA	63	{
b5af8293	64	struct netio_data *nd = td->io_ops->data;
ed92ac0c	65
7a6499da JA	66	/*
	67	* Make sure we don't see spurious reads to a receiver, and vice versa
	68	*/
b5af8293 JA	69	if ((nd->send_to_net && io_u->ddir == DDIR_READ) \|\|
b5af8293 JA	70	(!nd->send_to_net && io_u->ddir == DDIR_WRITE)) {
e1161c32	71	td_verror(td, EINVAL, "bad direction");
7a6499da	72	return 1;
ed92ac0c	73	}
7a6499da	74
f85ac25a	75	return 0;
ed92ac0c JA	76	}
ed92ac0c JA	77
5921e80c	78	#ifdef FIO_HAVE_SPLICE
cd963e18	79	static int splice_io_u(int fdin, int fdout, unsigned int len)
ed92ac0c	80	{
9cce02e8	81	int bytes = 0;
7a6499da	82
9cce02e8	83	while (len) {
cd963e18	84	int ret = splice(fdin, NULL, fdout, NULL, len, 0);
9cce02e8 JA	85
	86	if (ret < 0) {
	87	if (!bytes)
	88	bytes = ret;
	89
	90	break;
	91	} else if (!ret)
	92	break;
	93
	94	bytes += ret;
f657a2fb	95	len -= ret;
9cce02e8 JA	96	}
	97
	98	return bytes;
	99	}
	100
	101	/*
cd963e18	102	* Receive bytes from a socket and fill them into the internal pipe
9cce02e8	103	*/
cd963e18	104	static int splice_in(struct thread_data td, struct io_u io_u)
9cce02e8 JA	105	{
9cce02e8 JA	106	struct netio_data *nd = td->io_ops->data;
9cce02e8	107
cd963e18	108	return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen);
9cce02e8 JA	109	}
	110
	111	/*
cd963e18	112	* Transmit 'len' bytes from the internal pipe
9cce02e8	113	*/
cd963e18 JA	114	static int splice_out(struct thread_data td, struct io_u io_u,
cd963e18 JA	115	unsigned int len)
9cce02e8 JA	116	{
9cce02e8 JA	117	struct netio_data *nd = td->io_ops->data;
cd963e18 JA	118
	119	return splice_io_u(nd->pipes[0], io_u->file->fd, len);
	120	}
	121
	122	static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len)
	123	{
9cce02e8 JA	124	struct iovec iov = {
	125	.iov_base = io_u->xfer_buf,
	126	.iov_len = len,
	127	};
	128	int bytes = 0;
	129
	130	while (iov.iov_len) {
cd963e18	131	int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE);
9cce02e8 JA	132
	133	if (ret < 0) {
	134	if (!bytes)
	135	bytes = ret;
	136	break;
	137	} else if (!ret)
	138	break;
	139
	140	iov.iov_len -= ret;
cd963e18	141	iov.iov_base += ret;
f657a2fb	142	bytes += ret;
9cce02e8 JA	143	}
	144
	145	return bytes;
cd963e18	146
9cce02e8 JA	147	}
	148
	149	/*
cd963e18	150	* vmsplice() pipe to io_u buffer
9cce02e8	151	*/
cd963e18 JA	152	static int vmsplice_io_u_out(struct thread_data td, struct io_u io_u,
cd963e18 JA	153	unsigned int len)
9cce02e8 JA	154	{
9cce02e8 JA	155	struct netio_data *nd = td->io_ops->data;
9cce02e8	156
cd963e18 JA	157	return vmsplice_io_u(io_u, nd->pipes[0], len);
cd963e18 JA	158	}
9cce02e8	159
cd963e18 JA	160	/*
	161	* vmsplice() io_u to pipe
	162	*/
	163	static int vmsplice_io_u_in(struct thread_data td, struct io_u io_u)
	164	{
	165	struct netio_data *nd = td->io_ops->data;
ed92ac0c	166
cd963e18	167	return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen);
9cce02e8 JA	168	}
9cce02e8 JA	169
cd963e18 JA	170	/*
	171	* splice receive - transfer socket data into a pipe using splice, then map
	172	* that pipe data into the io_u using vmsplice.
	173	*/
9cce02e8 JA	174	static int fio_netio_splice_in(struct thread_data td, struct io_u io_u)
	175	{
	176	int ret;
	177
	178	ret = splice_in(td, io_u);
cd963e18 JA	179	if (ret > 0)
cd963e18 JA	180	return vmsplice_io_u_out(td, io_u, ret);
9cce02e8	181
cd963e18	182	return ret;
9cce02e8 JA	183	}
9cce02e8 JA	184
cd963e18 JA	185	/*
	186	* splice transmit - map data from the io_u into a pipe by using vmsplice,
	187	* then transfer that pipe to a socket using splice.
	188	*/
9cce02e8 JA	189	static int fio_netio_splice_out(struct thread_data td, struct io_u io_u)
	190	{
	191	int ret;
	192
	193	ret = vmsplice_io_u_in(td, io_u);
cd963e18 JA	194	if (ret > 0)
cd963e18 JA	195	return splice_out(td, io_u, ret);
9cce02e8	196
cd963e18	197	return ret;
9cce02e8	198	}
5921e80c JA	199	#else
	200	static int fio_netio_splice_in(struct thread_data td, struct io_u io_u)
	201	{
af8771b9	202	errno = EOPNOTSUPP;
5921e80c JA	203	return -1;
	204	}
	205
	206	static int fio_netio_splice_out(struct thread_data td, struct io_u io_u)
	207	{
af8771b9	208	errno = EOPNOTSUPP;
5921e80c JA	209	return -1;
	210	}
	211	#endif
9cce02e8 JA	212
	213	static int fio_netio_send(struct thread_data td, struct io_u io_u)
	214	{
414c2a3e	215	struct netio_data *nd = td->io_ops->data;
371d456c JA	216	int ret, flags = 0;
	217
	218	ret = poll_wait(td, io_u->file->fd, POLLOUT);
	219	if (ret <= 0)
	220	return ret;
9cce02e8 JA	221
	222	/*
	223	* if we are going to write more, set MSG_MORE
	224	*/
5921e80c	225	#ifdef MSG_MORE
9cce02e8 JA	226	if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size)
9cce02e8 JA	227	flags = MSG_MORE;
5921e80c	228	#endif
9cce02e8	229
414c2a3e JA	230	if (nd->net_protocol == IPPROTO_UDP) {
414c2a3e JA	231	return sendto(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen,
35651820	232	0, &nd->addr, sizeof(nd->addr));
414c2a3e JA	233	} else {
	234	return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen,
	235	flags);
	236	}
9cce02e8 JA	237	}
9cce02e8 JA	238
414c2a3e	239	static int fio_netio_recv(struct thread_data td, struct io_u io_u)
9cce02e8	240	{
414c2a3e	241	struct netio_data *nd = td->io_ops->data;
371d456c JA	242	int ret, flags = MSG_WAITALL;
	243
	244	ret = poll_wait(td, io_u->file->fd, POLLIN);
	245	if (ret <= 0)
	246	return ret;
9cce02e8	247
414c2a3e JA	248	if (nd->net_protocol == IPPROTO_UDP) {
	249	socklen_t len = sizeof(nd->addr);
	250
	251	return recvfrom(io_u->file->fd, io_u->xfer_buf,
	252	io_u->xfer_buflen, 0, &nd->addr, &len);
	253	} else {
	254	return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen,
	255	flags);
	256	}
9cce02e8 JA	257	}
	258
	259	static int fio_netio_queue(struct thread_data td, struct io_u io_u)
	260	{
	261	struct netio_data *nd = td->io_ops->data;
	262	int ret;
	263
7101d9c2 JA	264	fio_ro_check(td, io_u);
7101d9c2 JA	265
9cce02e8	266	if (io_u->ddir == DDIR_WRITE) {
414c2a3e	267	if (!nd->use_splice \|\| nd->net_protocol == IPPROTO_UDP)
9cce02e8	268	ret = fio_netio_send(td, io_u);
414c2a3e JA	269	else
414c2a3e JA	270	ret = fio_netio_splice_out(td, io_u);
d4f12dd0	271	} else if (io_u->ddir == DDIR_READ) {
414c2a3e JA	272	if (!nd->use_splice \|\| nd->net_protocol == IPPROTO_UDP)
414c2a3e JA	273	ret = fio_netio_recv(td, io_u);
9cce02e8	274	else
414c2a3e	275	ret = fio_netio_splice_in(td, io_u);
d4f12dd0	276	} else
7a6499da	277	ret = 0; /* must be a SYNC */
ed92ac0c	278
cec6b55d	279	if (ret != (int) io_u->xfer_buflen) {
22819ec2	280	if (ret >= 0) {
cec6b55d JA	281	io_u->resid = io_u->xfer_buflen - ret;
cec6b55d JA	282	io_u->error = 0;
36167d82	283	return FIO_Q_COMPLETED;
414c2a3e JA	284	} else {
	285	int err = errno;
	286
	287	if (io_u->ddir == DDIR_WRITE && err == EMSGSIZE)
	288	return FIO_Q_BUSY;
	289
	290	io_u->error = err;
	291	}
ed92ac0c JA	292	}
ed92ac0c JA	293
36167d82	294	if (io_u->error)
e1161c32	295	td_verror(td, io_u->error, "xfer");
ed92ac0c	296
36167d82	297	return FIO_Q_COMPLETED;
ed92ac0c JA	298	}
ed92ac0c JA	299
b5af8293	300	static int fio_netio_connect(struct thread_data td, struct fio_file f)
ed92ac0c	301	{
b5af8293	302	struct netio_data *nd = td->io_ops->data;
414c2a3e JA	303	int type;
	304
	305	if (nd->net_protocol == IPPROTO_TCP)
	306	type = SOCK_STREAM;
	307	else
	308	type = SOCK_DGRAM;
ed92ac0c	309
414c2a3e	310	f->fd = socket(AF_INET, type, nd->net_protocol);
b5af8293 JA	311	if (f->fd < 0) {
	312	td_verror(td, errno, "socket");
	313	return 1;
ed92ac0c JA	314	}
ed92ac0c JA	315
414c2a3e JA	316	if (nd->net_protocol == IPPROTO_UDP)
	317	return 0;
	318
b5af8293 JA	319	if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
	320	td_verror(td, errno, "connect");
	321	return 1;
ed92ac0c JA	322	}
	323
	324	return 0;
ed92ac0c JA	325	}
ed92ac0c JA	326
b5af8293	327	static int fio_netio_accept(struct thread_data td, struct fio_file f)
5fdd124a	328	{
b5af8293 JA	329	struct netio_data *nd = td->io_ops->data;
b5af8293 JA	330	socklen_t socklen = sizeof(nd->addr);
5fdd124a	331
414c2a3e JA	332	if (nd->net_protocol == IPPROTO_UDP) {
	333	f->fd = nd->listenfd;
	334	return 0;
	335	}
	336
6d86144d	337	log_info("fio: waiting for connection\n");
5fdd124a	338
371d456c JA	339	if (poll_wait(td, nd->listenfd, POLLIN) < 0)
371d456c JA	340	return 1;
0c09442b	341
371d456c JA	342	f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
	343	if (f->fd < 0) {
	344	td_verror(td, errno, "accept");
	345	return 1;
b5af8293	346	}
5fdd124a	347
b5af8293 JA	348	return 0;
	349	}
	350
b5af8293 JA	351	static int fio_netio_open_file(struct thread_data td, struct fio_file f)
	352	{
	353	if (td_read(td))
	354	return fio_netio_accept(td, f);
	355	else
	356	return fio_netio_connect(td, f);
	357	}
	358
	359	static int fio_netio_setup_connect(struct thread_data td, const char host,
	360	unsigned short port)
	361	{
	362	struct netio_data *nd = td->io_ops->data;
	363
	364	nd->addr.sin_family = AF_INET;
	365	nd->addr.sin_port = htons(port);
	366
	367	if (inet_aton(host, &nd->addr.sin_addr) != 1) {
	368	struct hostent *hent;
	369
	370	hent = gethostbyname(host);
	371	if (!hent) {
	372	td_verror(td, errno, "gethostbyname");
	373	return 1;
5fdd124a	374	}
b5af8293 JA	375
b5af8293 JA	376	memcpy(&nd->addr.sin_addr, hent->h_addr, 4);
5fdd124a JA	377	}
	378
	379	return 0;
	380	}
	381
b5af8293	382	static int fio_netio_setup_listen(struct thread_data *td, short port)
ed92ac0c	383	{
b5af8293	384	struct netio_data *nd = td->io_ops->data;
414c2a3e	385	int fd, opt, type;
ed92ac0c	386
414c2a3e JA	387	if (nd->net_protocol == IPPROTO_TCP)
	388	type = SOCK_STREAM;
	389	else
	390	type = SOCK_DGRAM;
	391
	392	fd = socket(AF_INET, type, nd->net_protocol);
ed92ac0c	393	if (fd < 0) {
e1161c32	394	td_verror(td, errno, "socket");
ed92ac0c JA	395	return 1;
	396	}
	397
	398	opt = 1;
	399	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
e1161c32	400	td_verror(td, errno, "setsockopt");
ed92ac0c JA	401	return 1;
ed92ac0c JA	402	}
6bedbfaf JA	403	#ifdef SO_REUSEPORT
6bedbfaf JA	404	if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
e1161c32	405	td_verror(td, errno, "setsockopt");
6bedbfaf JA	406	return 1;
	407	}
	408	#endif
ed92ac0c	409
b5af8293 JA	410	nd->addr.sin_family = AF_INET;
	411	nd->addr.sin_addr.s_addr = htonl(INADDR_ANY);
	412	nd->addr.sin_port = htons(port);
ed92ac0c	413
b5af8293	414	if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
e1161c32	415	td_verror(td, errno, "bind");
ed92ac0c JA	416	return 1;
ed92ac0c JA	417	}
414c2a3e	418	if (nd->net_protocol == IPPROTO_TCP && listen(fd, 1) < 0) {
e1161c32	419	td_verror(td, errno, "listen");
ed92ac0c JA	420	return 1;
	421	}
	422
b5af8293 JA	423	nd->listenfd = fd;
b5af8293 JA	424	return 0;
ed92ac0c JA	425	}
ed92ac0c JA	426
9bec88e1	427	static int fio_netio_init(struct thread_data *td)
ed92ac0c	428	{
b5af8293	429	struct netio_data *nd = td->io_ops->data;
443662ef	430	unsigned int port;
b5af8293	431	char host[64], buf[128];
414c2a3e	432	char sep, portp, *modep;
af52b345	433	int ret;
ed92ac0c	434
413dd459	435	if (td_rw(td)) {
ed92ac0c JA	436	log_err("fio: network connections must be read OR write\n");
	437	return 1;
	438	}
16d55aae JA	439	if (td_random(td)) {
	440	log_err("fio: network IO can't be random\n");
	441	return 1;
	442	}
ed92ac0c	443
2dc1bbeb	444	strcpy(buf, td->o.filename);
ed92ac0c	445
9f9214f2	446	sep = strchr(buf, '/');
443662ef JA	447	if (!sep)
443662ef JA	448	goto bad_host;
ed92ac0c JA	449
	450	*sep = '\0';
	451	sep++;
	452	strcpy(host, buf);
443662ef JA	453	if (!strlen(host))
	454	goto bad_host;
	455
414c2a3e JA	456	modep = NULL;
	457	portp = sep;
	458	sep = strchr(portp, '/');
	459	if (sep) {
	460	*sep = '\0';
	461	modep = sep + 1;
	462	}
	463
	464	port = strtol(portp, NULL, 10);
443662ef JA	465	if (!port \|\| port > 65535)
443662ef JA	466	goto bad_host;
ed92ac0c	467
414c2a3e	468	if (modep) {
3f8fc5ad JA	469	if (!strncmp("tcp", modep, strlen(modep)) \|\|
3f8fc5ad JA	470	!strncmp("TCP", modep, strlen(modep)))
414c2a3e	471	nd->net_protocol = IPPROTO_TCP;
3f8fc5ad JA	472	else if (!strncmp("udp", modep, strlen(modep)) \|\|
3f8fc5ad JA	473	!strncmp("UDP", modep, strlen(modep)))
414c2a3e JA	474	nd->net_protocol = IPPROTO_UDP;
	475	else
	476	goto bad_host;
	477	} else
	478	nd->net_protocol = IPPROTO_TCP;
	479
413dd459	480	if (td_read(td)) {
b5af8293	481	nd->send_to_net = 0;
ed92ac0c JA	482	ret = fio_netio_setup_listen(td, port);
ed92ac0c JA	483	} else {
b5af8293	484	nd->send_to_net = 1;
ed92ac0c JA	485	ret = fio_netio_setup_connect(td, host, port);
	486	}
	487
7bb48f84	488	return ret;
443662ef	489	bad_host:
414c2a3e	490	log_err("fio: bad network host/port/protocol: %s\n", td->o.filename);
443662ef	491	return 1;
ed92ac0c JA	492	}
ed92ac0c JA	493
b5af8293	494	static void fio_netio_cleanup(struct thread_data *td)
9bec88e1	495	{
b5af8293 JA	496	struct netio_data *nd = td->io_ops->data;
	497
	498	if (nd) {
64b24cd8 JA	499	if (nd->listenfd != -1)
	500	close(nd->listenfd);
	501	if (nd->pipes[0] != -1)
	502	close(nd->pipes[0]);
	503	if (nd->pipes[1] != -1)
	504	close(nd->pipes[1]);
	505
b5af8293	506	free(nd);
b5af8293 JA	507	}
	508	}
	509
	510	static int fio_netio_setup(struct thread_data *td)
	511	{
7bb48f84	512	struct netio_data *nd;
7bb48f84 JA	513
	514	if (!td->io_ops->data) {
	515	nd = malloc(sizeof(*nd));;
	516
	517	memset(nd, 0, sizeof(*nd));
	518	nd->listenfd = -1;
64b24cd8	519	nd->pipes[0] = nd->pipes[1] = -1;
7bb48f84	520	td->io_ops->data = nd;
7bb48f84	521	}
b5af8293	522
9bec88e1 JA	523	return 0;
	524	}
	525
5921e80c	526	#ifdef FIO_HAVE_SPLICE
9cce02e8 JA	527	static int fio_netio_setup_splice(struct thread_data *td)
	528	{
	529	struct netio_data *nd;
	530
	531	fio_netio_setup(td);
	532
	533	nd = td->io_ops->data;
	534	if (nd) {
	535	if (pipe(nd->pipes) < 0)
	536	return 1;
	537
	538	nd->use_splice = 1;
	539	return 0;
	540	}
	541
	542	return 1;
	543	}
	544
5921e80c JA	545	static struct ioengine_ops ioengine_splice = {
5921e80c JA	546	.name = "netsplice",
ed92ac0c	547	.version = FIO_IOOPS_VERSION,
ed92ac0c JA	548	.prep = fio_netio_prep,
ed92ac0c JA	549	.queue = fio_netio_queue,
5921e80c	550	.setup = fio_netio_setup_splice,
9bec88e1	551	.init = fio_netio_init,
b5af8293 JA	552	.cleanup = fio_netio_cleanup,
	553	.open_file = fio_netio_open_file,
	554	.close_file = generic_close_file,
ad830ed7 JA	555	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_UNIDIR \|
ad830ed7 JA	556	FIO_SIGQUIT,
ed92ac0c	557	};
5921e80c	558	#endif
ed92ac0c	559
5921e80c JA	560	static struct ioengine_ops ioengine_rw = {
5921e80c JA	561	.name = "net",
9cce02e8 JA	562	.version = FIO_IOOPS_VERSION,
	563	.prep = fio_netio_prep,
	564	.queue = fio_netio_queue,
5921e80c	565	.setup = fio_netio_setup,
9cce02e8 JA	566	.init = fio_netio_init,
	567	.cleanup = fio_netio_cleanup,
	568	.open_file = fio_netio_open_file,
	569	.close_file = generic_close_file,
ad830ed7 JA	570	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_UNIDIR \|
ad830ed7 JA	571	FIO_SIGQUIT,
9cce02e8 JA	572	};
9cce02e8 JA	573
ed92ac0c JA	574	static void fio_init fio_netio_register(void)
ed92ac0c JA	575	{
9cce02e8	576	register_ioengine(&ioengine_rw);
5921e80c	577	#ifdef FIO_HAVE_SPLICE
9cce02e8	578	register_ioengine(&ioengine_splice);
5921e80c	579	#endif
ed92ac0c JA	580	}
	581
	582	static void fio_exit fio_netio_unregister(void)
	583	{
9cce02e8	584	unregister_ioengine(&ioengine_rw);
5921e80c	585	#ifdef FIO_HAVE_SPLICE
9cce02e8	586	unregister_ioengine(&ioengine_splice);
5921e80c	587	#endif
ed92ac0c	588	}