Home | History | Annotate | Download | only in engines
      1 /*
      2  * net engine
      3  *
      4  * IO engine that reads/writes to/from sockets.
      5  *
      6  */
      7 #include <stdio.h>
      8 #include <stdlib.h>
      9 #include <unistd.h>
     10 #include <signal.h>
     11 #include <errno.h>
     12 #include <assert.h>
     13 #include <netinet/in.h>
     14 #include <netinet/tcp.h>
     15 #include <arpa/inet.h>
     16 #include <netdb.h>
     17 #include <sys/poll.h>
     18 #include <sys/types.h>
     19 #include <sys/stat.h>
     20 #include <sys/socket.h>
     21 #include <sys/un.h>
     22 
     23 #include "../fio.h"
     24 
     25 struct netio_data {
     26 	int listenfd;
     27 	int use_splice;
     28 	int pipes[2];
     29 	struct sockaddr_in addr;
     30 	struct sockaddr_in6 addr6;
     31 	struct sockaddr_un addr_un;
     32 };
     33 
     34 struct netio_options {
     35 	struct thread_data *td;
     36 	unsigned int port;
     37 	unsigned int proto;
     38 	unsigned int listen;
     39 	unsigned int pingpong;
     40 	unsigned int nodelay;
     41 	unsigned int ttl;
     42 	char *intfc;
     43 };
     44 
     45 struct udp_close_msg {
     46 	uint32_t magic;
     47 	uint32_t cmd;
     48 };
     49 
     50 enum {
     51 	FIO_LINK_CLOSE = 0x89,
     52 	FIO_LINK_OPEN_CLOSE_MAGIC = 0x6c696e6b,
     53 	FIO_LINK_OPEN = 0x98,
     54 
     55 	FIO_TYPE_TCP	= 1,
     56 	FIO_TYPE_UDP	= 2,
     57 	FIO_TYPE_UNIX	= 3,
     58 	FIO_TYPE_TCP_V6	= 4,
     59 	FIO_TYPE_UDP_V6	= 5,
     60 };
     61 
     62 static int str_hostname_cb(void *data, const char *input);
     63 static struct fio_option options[] = {
     64 	{
     65 		.name	= "hostname",
     66 		.lname	= "net engine hostname",
     67 		.type	= FIO_OPT_STR_STORE,
     68 		.cb	= str_hostname_cb,
     69 		.help	= "Hostname for net IO engine",
     70 		.category = FIO_OPT_C_ENGINE,
     71 		.group	= FIO_OPT_G_NETIO,
     72 	},
     73 	{
     74 		.name	= "port",
     75 		.lname	= "net engine port",
     76 		.type	= FIO_OPT_INT,
     77 		.off1	= offsetof(struct netio_options, port),
     78 		.minval	= 1,
     79 		.maxval	= 65535,
     80 		.help	= "Port to use for TCP or UDP net connections",
     81 		.category = FIO_OPT_C_ENGINE,
     82 		.group	= FIO_OPT_G_NETIO,
     83 	},
     84 	{
     85 		.name	= "protocol",
     86 		.lname	= "net engine protocol",
     87 		.alias	= "proto",
     88 		.type	= FIO_OPT_STR,
     89 		.off1	= offsetof(struct netio_options, proto),
     90 		.help	= "Network protocol to use",
     91 		.def	= "tcp",
     92 		.posval = {
     93 			  { .ival = "tcp",
     94 			    .oval = FIO_TYPE_TCP,
     95 			    .help = "Transmission Control Protocol",
     96 			  },
     97 #ifdef CONFIG_IPV6
     98 			  { .ival = "tcpv6",
     99 			    .oval = FIO_TYPE_TCP_V6,
    100 			    .help = "Transmission Control Protocol V6",
    101 			  },
    102 #endif
    103 			  { .ival = "udp",
    104 			    .oval = FIO_TYPE_UDP,
    105 			    .help = "User Datagram Protocol",
    106 			  },
    107 #ifdef CONFIG_IPV6
    108 			  { .ival = "udpv6",
    109 			    .oval = FIO_TYPE_UDP_V6,
    110 			    .help = "User Datagram Protocol V6",
    111 			  },
    112 #endif
    113 			  { .ival = "unix",
    114 			    .oval = FIO_TYPE_UNIX,
    115 			    .help = "UNIX domain socket",
    116 			  },
    117 		},
    118 		.category = FIO_OPT_C_ENGINE,
    119 		.group	= FIO_OPT_G_NETIO,
    120 	},
    121 #ifdef CONFIG_TCP_NODELAY
    122 	{
    123 		.name	= "nodelay",
    124 		.type	= FIO_OPT_BOOL,
    125 		.off1	= offsetof(struct netio_options, nodelay),
    126 		.help	= "Use TCP_NODELAY on TCP connections",
    127 		.category = FIO_OPT_C_ENGINE,
    128 		.group	= FIO_OPT_G_NETIO,
    129 	},
    130 #endif
    131 	{
    132 		.name	= "listen",
    133 		.lname	= "net engine listen",
    134 		.type	= FIO_OPT_STR_SET,
    135 		.off1	= offsetof(struct netio_options, listen),
    136 		.help	= "Listen for incoming TCP connections",
    137 		.category = FIO_OPT_C_ENGINE,
    138 		.group	= FIO_OPT_G_NETIO,
    139 	},
    140 	{
    141 		.name	= "pingpong",
    142 		.type	= FIO_OPT_STR_SET,
    143 		.off1	= offsetof(struct netio_options, pingpong),
    144 		.help	= "Ping-pong IO requests",
    145 		.category = FIO_OPT_C_ENGINE,
    146 		.group	= FIO_OPT_G_NETIO,
    147 	},
    148 	{
    149 		.name	= "interface",
    150 		.lname	= "net engine interface",
    151 		.type	= FIO_OPT_STR_STORE,
    152 		.off1	= offsetof(struct netio_options, intfc),
    153 		.help	= "Network interface to use",
    154 		.category = FIO_OPT_C_ENGINE,
    155 		.group	= FIO_OPT_G_NETIO,
    156 	},
    157 	{
    158 		.name	= "ttl",
    159 		.lname	= "net engine multicast ttl",
    160 		.type	= FIO_OPT_INT,
    161 		.off1	= offsetof(struct netio_options, ttl),
    162 		.def    = "1",
    163 		.minval	= 0,
    164 		.help	= "Time-to-live value for outgoing UDP multicast packets",
    165 		.category = FIO_OPT_C_ENGINE,
    166 		.group	= FIO_OPT_G_NETIO,
    167 	},
    168 	{
    169 		.name	= NULL,
    170 	},
    171 };
    172 
    173 static inline int is_udp(struct netio_options *o)
    174 {
    175 	return o->proto == FIO_TYPE_UDP || o->proto == FIO_TYPE_UDP_V6;
    176 }
    177 
    178 static inline int is_tcp(struct netio_options *o)
    179 {
    180 	return o->proto == FIO_TYPE_TCP || o->proto == FIO_TYPE_TCP_V6;
    181 }
    182 
    183 static inline int is_ipv6(struct netio_options *o)
    184 {
    185 	return o->proto == FIO_TYPE_UDP_V6 || o->proto == FIO_TYPE_TCP_V6;
    186 }
    187 
    188 /*
    189  * Return -1 for error and 'nr events' for a positive number
    190  * of events
    191  */
    192 static int poll_wait(struct thread_data *td, int fd, short events)
    193 {
    194 	struct pollfd pfd;
    195 	int ret;
    196 
    197 	while (!td->terminate) {
    198 		pfd.fd = fd;
    199 		pfd.events = events;
    200 		ret = poll(&pfd, 1, -1);
    201 		if (ret < 0) {
    202 			if (errno == EINTR)
    203 				break;
    204 
    205 			td_verror(td, errno, "poll");
    206 			return -1;
    207 		} else if (!ret)
    208 			continue;
    209 
    210 		break;
    211 	}
    212 
    213 	if (pfd.revents & events)
    214 		return 1;
    215 
    216 	return -1;
    217 }
    218 
    219 static int fio_netio_is_multicast(const char *mcaddr)
    220 {
    221 	in_addr_t addr = inet_network(mcaddr);
    222 	if (addr == -1)
    223 		return 0;
    224 
    225 	if (inet_network("224.0.0.0") <= addr &&
    226 	    inet_network("239.255.255.255") >= addr)
    227 		return 1;
    228 
    229 	return 0;
    230 }
    231 
    232 
    233 static int fio_netio_prep(struct thread_data *td, struct io_u *io_u)
    234 {
    235 	struct netio_options *o = td->eo;
    236 
    237 	/*
    238 	 * Make sure we don't see spurious reads to a receiver, and vice versa
    239 	 */
    240 	if (is_tcp(o))
    241 		return 0;
    242 
    243 	if ((o->listen && io_u->ddir == DDIR_WRITE) ||
    244 	    (!o->listen && io_u->ddir == DDIR_READ)) {
    245 		td_verror(td, EINVAL, "bad direction");
    246 		return 1;
    247 	}
    248 
    249 	return 0;
    250 }
    251 
    252 #ifdef CONFIG_LINUX_SPLICE
    253 static int splice_io_u(int fdin, int fdout, unsigned int len)
    254 {
    255 	int bytes = 0;
    256 
    257 	while (len) {
    258 		int ret = splice(fdin, NULL, fdout, NULL, len, 0);
    259 
    260 		if (ret < 0) {
    261 			if (!bytes)
    262 				bytes = ret;
    263 
    264 			break;
    265 		} else if (!ret)
    266 			break;
    267 
    268 		bytes += ret;
    269 		len -= ret;
    270 	}
    271 
    272 	return bytes;
    273 }
    274 
    275 /*
    276  * Receive bytes from a socket and fill them into the internal pipe
    277  */
    278 static int splice_in(struct thread_data *td, struct io_u *io_u)
    279 {
    280 	struct netio_data *nd = td->io_ops->data;
    281 
    282 	return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen);
    283 }
    284 
    285 /*
    286  * Transmit 'len' bytes from the internal pipe
    287  */
    288 static int splice_out(struct thread_data *td, struct io_u *io_u,
    289 		      unsigned int len)
    290 {
    291 	struct netio_data *nd = td->io_ops->data;
    292 
    293 	return splice_io_u(nd->pipes[0], io_u->file->fd, len);
    294 }
    295 
    296 static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len)
    297 {
    298 	struct iovec iov = {
    299 		.iov_base = io_u->xfer_buf,
    300 		.iov_len = len,
    301 	};
    302 	int bytes = 0;
    303 
    304 	while (iov.iov_len) {
    305 		int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE);
    306 
    307 		if (ret < 0) {
    308 			if (!bytes)
    309 				bytes = ret;
    310 			break;
    311 		} else if (!ret)
    312 			break;
    313 
    314 		iov.iov_len -= ret;
    315 		iov.iov_base += ret;
    316 		bytes += ret;
    317 	}
    318 
    319 	return bytes;
    320 
    321 }
    322 
    323 /*
    324  * vmsplice() pipe to io_u buffer
    325  */
    326 static int vmsplice_io_u_out(struct thread_data *td, struct io_u *io_u,
    327 			     unsigned int len)
    328 {
    329 	struct netio_data *nd = td->io_ops->data;
    330 
    331 	return vmsplice_io_u(io_u, nd->pipes[0], len);
    332 }
    333 
    334 /*
    335  * vmsplice() io_u to pipe
    336  */
    337 static int vmsplice_io_u_in(struct thread_data *td, struct io_u *io_u)
    338 {
    339 	struct netio_data *nd = td->io_ops->data;
    340 
    341 	return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen);
    342 }
    343 
    344 /*
    345  * splice receive - transfer socket data into a pipe using splice, then map
    346  * that pipe data into the io_u using vmsplice.
    347  */
    348 static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
    349 {
    350 	int ret;
    351 
    352 	ret = splice_in(td, io_u);
    353 	if (ret > 0)
    354 		return vmsplice_io_u_out(td, io_u, ret);
    355 
    356 	return ret;
    357 }
    358 
    359 /*
    360  * splice transmit - map data from the io_u into a pipe by using vmsplice,
    361  * then transfer that pipe to a socket using splice.
    362  */
    363 static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u)
    364 {
    365 	int ret;
    366 
    367 	ret = vmsplice_io_u_in(td, io_u);
    368 	if (ret > 0)
    369 		return splice_out(td, io_u, ret);
    370 
    371 	return ret;
    372 }
    373 #else
    374 static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
    375 {
    376 	errno = EOPNOTSUPP;
    377 	return -1;
    378 }
    379 
    380 static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u)
    381 {
    382 	errno = EOPNOTSUPP;
    383 	return -1;
    384 }
    385 #endif
    386 
    387 static int fio_netio_send(struct thread_data *td, struct io_u *io_u)
    388 {
    389 	struct netio_data *nd = td->io_ops->data;
    390 	struct netio_options *o = td->eo;
    391 	int ret, flags = 0;
    392 
    393 	do {
    394 		if (is_udp(o)) {
    395 			const struct sockaddr *to;
    396 			socklen_t len;
    397 
    398 			if (is_ipv6(o)) {
    399 				to = (struct sockaddr *) &nd->addr6;
    400 				len = sizeof(nd->addr6);
    401 			} else {
    402 				to = (struct sockaddr *) &nd->addr;
    403 				len = sizeof(nd->addr);
    404 			}
    405 
    406 			ret = sendto(io_u->file->fd, io_u->xfer_buf,
    407 					io_u->xfer_buflen, flags, to, len);
    408 		} else {
    409 			/*
    410 			 * if we are going to write more, set MSG_MORE
    411 			 */
    412 #ifdef MSG_MORE
    413 			if ((td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen <
    414 			    td->o.size) && !o->pingpong)
    415 				flags |= MSG_MORE;
    416 #endif
    417 			ret = send(io_u->file->fd, io_u->xfer_buf,
    418 					io_u->xfer_buflen, flags);
    419 		}
    420 		if (ret > 0)
    421 			break;
    422 
    423 		ret = poll_wait(td, io_u->file->fd, POLLOUT);
    424 		if (ret <= 0)
    425 			break;
    426 	} while (1);
    427 
    428 	return ret;
    429 }
    430 
    431 static int is_udp_close(struct io_u *io_u, int len)
    432 {
    433 	struct udp_close_msg *msg;
    434 
    435 	if (len != sizeof(struct udp_close_msg))
    436 		return 0;
    437 
    438 	msg = io_u->xfer_buf;
    439 	if (ntohl(msg->magic) != FIO_LINK_OPEN_CLOSE_MAGIC)
    440 		return 0;
    441 	if (ntohl(msg->cmd) != FIO_LINK_CLOSE)
    442 		return 0;
    443 
    444 	return 1;
    445 }
    446 
    447 static int fio_netio_recv(struct thread_data *td, struct io_u *io_u)
    448 {
    449 	struct netio_data *nd = td->io_ops->data;
    450 	struct netio_options *o = td->eo;
    451 	int ret, flags = 0;
    452 
    453 	do {
    454 		if (is_udp(o)) {
    455 			struct sockaddr *from;
    456 			socklen_t l, *len = &l;
    457 
    458 			if (o->listen) {
    459 				if (!is_ipv6(o)) {
    460 					from = (struct sockaddr *) &nd->addr;
    461 					*len = sizeof(nd->addr);
    462 				} else {
    463 					from = (struct sockaddr *) &nd->addr6;
    464 					*len = sizeof(nd->addr6);
    465 				}
    466 			} else {
    467 				from = NULL;
    468 				len = NULL;
    469 			}
    470 
    471 			ret = recvfrom(io_u->file->fd, io_u->xfer_buf,
    472 					io_u->xfer_buflen, flags, from, len);
    473 			if (is_udp_close(io_u, ret)) {
    474 				td->done = 1;
    475 				return 0;
    476 			}
    477 		} else {
    478 			ret = recv(io_u->file->fd, io_u->xfer_buf,
    479 					io_u->xfer_buflen, flags);
    480 		}
    481 		if (ret > 0)
    482 			break;
    483 		else if (!ret && (flags & MSG_WAITALL))
    484 			break;
    485 
    486 		ret = poll_wait(td, io_u->file->fd, POLLIN);
    487 		if (ret <= 0)
    488 			break;
    489 		flags |= MSG_WAITALL;
    490 	} while (1);
    491 
    492 	return ret;
    493 }
    494 
    495 static int __fio_netio_queue(struct thread_data *td, struct io_u *io_u,
    496 			     enum fio_ddir ddir)
    497 {
    498 	struct netio_data *nd = td->io_ops->data;
    499 	struct netio_options *o = td->eo;
    500 	int ret;
    501 
    502 	if (ddir == DDIR_WRITE) {
    503 		if (!nd->use_splice || is_udp(o) ||
    504 		    o->proto == FIO_TYPE_UNIX)
    505 			ret = fio_netio_send(td, io_u);
    506 		else
    507 			ret = fio_netio_splice_out(td, io_u);
    508 	} else if (ddir == DDIR_READ) {
    509 		if (!nd->use_splice || is_udp(o) ||
    510 		    o->proto == FIO_TYPE_UNIX)
    511 			ret = fio_netio_recv(td, io_u);
    512 		else
    513 			ret = fio_netio_splice_in(td, io_u);
    514 	} else
    515 		ret = 0;	/* must be a SYNC */
    516 
    517 	if (ret != (int) io_u->xfer_buflen) {
    518 		if (ret >= 0) {
    519 			io_u->resid = io_u->xfer_buflen - ret;
    520 			io_u->error = 0;
    521 			return FIO_Q_COMPLETED;
    522 		} else {
    523 			int err = errno;
    524 
    525 			if (ddir == DDIR_WRITE && err == EMSGSIZE)
    526 				return FIO_Q_BUSY;
    527 
    528 			io_u->error = err;
    529 		}
    530 	}
    531 
    532 	if (io_u->error)
    533 		td_verror(td, io_u->error, "xfer");
    534 
    535 	return FIO_Q_COMPLETED;
    536 }
    537 
    538 static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
    539 {
    540 	struct netio_options *o = td->eo;
    541 	int ret;
    542 
    543 	fio_ro_check(td, io_u);
    544 
    545 	ret = __fio_netio_queue(td, io_u, io_u->ddir);
    546 	if (!o->pingpong || ret != FIO_Q_COMPLETED)
    547 		return ret;
    548 
    549 	/*
    550 	 * For ping-pong mode, receive or send reply as needed
    551 	 */
    552 	if (td_read(td) && io_u->ddir == DDIR_READ)
    553 		ret = __fio_netio_queue(td, io_u, DDIR_WRITE);
    554 	else if (td_write(td) && io_u->ddir == DDIR_WRITE)
    555 		ret = __fio_netio_queue(td, io_u, DDIR_READ);
    556 
    557 	return ret;
    558 }
    559 
    560 static int fio_netio_connect(struct thread_data *td, struct fio_file *f)
    561 {
    562 	struct netio_data *nd = td->io_ops->data;
    563 	struct netio_options *o = td->eo;
    564 	int type, domain;
    565 
    566 	if (o->proto == FIO_TYPE_TCP) {
    567 		domain = AF_INET;
    568 		type = SOCK_STREAM;
    569 	} else if (o->proto == FIO_TYPE_TCP_V6) {
    570 		domain = AF_INET6;
    571 		type = SOCK_STREAM;
    572 	} else if (o->proto == FIO_TYPE_UDP) {
    573 		domain = AF_INET;
    574 		type = SOCK_DGRAM;
    575 	} else if (o->proto == FIO_TYPE_UDP_V6) {
    576 		domain = AF_INET6;
    577 		type = SOCK_DGRAM;
    578 	} else if (o->proto == FIO_TYPE_UNIX) {
    579 		domain = AF_UNIX;
    580 		type = SOCK_STREAM;
    581 	} else {
    582 		log_err("fio: bad network type %d\n", o->proto);
    583 		f->fd = -1;
    584 		return 1;
    585 	}
    586 
    587 	f->fd = socket(domain, type, 0);
    588 	if (f->fd < 0) {
    589 		td_verror(td, errno, "socket");
    590 		return 1;
    591 	}
    592 
    593 #ifdef CONFIG_TCP_NODELAY
    594 	if (o->nodelay && is_tcp(o)) {
    595 		int optval = 1;
    596 
    597 		if (setsockopt(f->fd, IPPROTO_TCP, TCP_NODELAY, (void *) &optval, sizeof(int)) < 0) {
    598 			log_err("fio: cannot set TCP_NODELAY option on socket (%s), disable with 'nodelay=0'\n", strerror(errno));
    599 			return 1;
    600 		}
    601 	}
    602 #endif
    603 
    604 	if (is_udp(o)) {
    605 		if (!fio_netio_is_multicast(td->o.filename))
    606 			return 0;
    607 		if (is_ipv6(o)) {
    608 			log_err("fio: multicast not supported on IPv6\n");
    609 			close(f->fd);
    610 			return 1;
    611 		}
    612 
    613 		if (o->intfc) {
    614 			struct in_addr interface_addr;
    615 
    616 			if (inet_aton(o->intfc, &interface_addr) == 0) {
    617 				log_err("fio: interface not valid interface IP\n");
    618 				close(f->fd);
    619 				return 1;
    620 			}
    621 			if (setsockopt(f->fd, IPPROTO_IP, IP_MULTICAST_IF, (const char*)&interface_addr, sizeof(interface_addr)) < 0) {
    622 				td_verror(td, errno, "setsockopt IP_MULTICAST_IF");
    623 				close(f->fd);
    624 				return 1;
    625 			}
    626 		}
    627 		if (setsockopt(f->fd, IPPROTO_IP, IP_MULTICAST_TTL, (const char*)&o->ttl, sizeof(o->ttl)) < 0) {
    628 			td_verror(td, errno, "setsockopt IP_MULTICAST_TTL");
    629 			close(f->fd);
    630 			return 1;
    631 		}
    632 		return 0;
    633 	} else if (o->proto == FIO_TYPE_TCP) {
    634 		socklen_t len = sizeof(nd->addr);
    635 
    636 		if (connect(f->fd, (struct sockaddr *) &nd->addr, len) < 0) {
    637 			td_verror(td, errno, "connect");
    638 			close(f->fd);
    639 			return 1;
    640 		}
    641 	} else if (o->proto == FIO_TYPE_TCP_V6) {
    642 		socklen_t len = sizeof(nd->addr6);
    643 
    644 		if (connect(f->fd, (struct sockaddr *) &nd->addr6, len) < 0) {
    645 			td_verror(td, errno, "connect");
    646 			close(f->fd);
    647 			return 1;
    648 		}
    649 
    650 	} else {
    651 		struct sockaddr_un *addr = &nd->addr_un;
    652 		socklen_t len;
    653 
    654 		len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
    655 
    656 		if (connect(f->fd, (struct sockaddr *) addr, len) < 0) {
    657 			td_verror(td, errno, "connect");
    658 			close(f->fd);
    659 			return 1;
    660 		}
    661 	}
    662 
    663 	return 0;
    664 }
    665 
    666 static int fio_netio_accept(struct thread_data *td, struct fio_file *f)
    667 {
    668 	struct netio_data *nd = td->io_ops->data;
    669 	struct netio_options *o = td->eo;
    670 	socklen_t socklen;
    671 	int state;
    672 
    673 	if (is_udp(o)) {
    674 		f->fd = nd->listenfd;
    675 		return 0;
    676 	}
    677 
    678 	state = td->runstate;
    679 	td_set_runstate(td, TD_SETTING_UP);
    680 
    681 	log_info("fio: waiting for connection\n");
    682 
    683 	if (poll_wait(td, nd->listenfd, POLLIN) < 0)
    684 		goto err;
    685 
    686 	if (o->proto == FIO_TYPE_TCP) {
    687 		socklen = sizeof(nd->addr);
    688 		f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
    689 	} else {
    690 		socklen = sizeof(nd->addr6);
    691 		f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr6, &socklen);
    692 	}
    693 
    694 	if (f->fd < 0) {
    695 		td_verror(td, errno, "accept");
    696 		goto err;
    697 	}
    698 
    699 #ifdef CONFIG_TCP_NODELAY
    700 	if (o->nodelay && is_tcp(o)) {
    701 		int optval = 1;
    702 
    703 		if (setsockopt(f->fd, IPPROTO_TCP, TCP_NODELAY, (void *) &optval, sizeof(int)) < 0) {
    704 			log_err("fio: cannot set TCP_NODELAY option on socket (%s), disable with 'nodelay=0'\n", strerror(errno));
    705 			return 1;
    706 		}
    707 	}
    708 #endif
    709 
    710 	reset_all_stats(td);
    711 	td_set_runstate(td, state);
    712 	return 0;
    713 err:
    714 	td_set_runstate(td, state);
    715 	return 1;
    716 }
    717 
    718 static void fio_netio_udp_close(struct thread_data *td, struct fio_file *f)
    719 {
    720 	struct netio_data *nd = td->io_ops->data;
    721 	struct netio_options *o = td->eo;
    722 	struct udp_close_msg msg;
    723 	struct sockaddr *to;
    724 	socklen_t len;
    725 	int ret;
    726 
    727 	if (is_ipv6(o)) {
    728 		to = (struct sockaddr *) &nd->addr6;
    729 		len = sizeof(nd->addr6);
    730 	} else {
    731 		to = (struct sockaddr *) &nd->addr;
    732 		len = sizeof(nd->addr);
    733 	}
    734 
    735 	msg.magic = htonl(FIO_LINK_OPEN_CLOSE_MAGIC);
    736 	msg.cmd = htonl(FIO_LINK_CLOSE);
    737 
    738 	ret = sendto(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, len);
    739 	if (ret < 0)
    740 		td_verror(td, errno, "sendto udp link close");
    741 }
    742 
    743 static int fio_netio_close_file(struct thread_data *td, struct fio_file *f)
    744 {
    745 	struct netio_options *o = td->eo;
    746 
    747 	/*
    748 	 * If this is an UDP connection, notify the receiver that we are
    749 	 * closing down the link
    750 	 */
    751 	if (is_udp(o))
    752 		fio_netio_udp_close(td, f);
    753 
    754 	return generic_close_file(td, f);
    755 }
    756 
    757 static int fio_netio_udp_recv_open(struct thread_data *td, struct fio_file *f)
    758 {
    759 	struct netio_data *nd = td->io_ops->data;
    760 	struct netio_options *o = td->eo;
    761 	struct udp_close_msg msg;
    762 	struct sockaddr *to;
    763 	socklen_t len;
    764 	int ret;
    765 
    766 	if (is_ipv6(o)) {
    767 		len = sizeof(nd->addr6);
    768 		to = (struct sockaddr *) &nd->addr6;
    769 	} else {
    770 		len = sizeof(nd->addr);
    771 		to = (struct sockaddr *) &nd->addr;
    772 	}
    773 
    774 	ret = recvfrom(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, &len);
    775 	if (ret < 0) {
    776 		td_verror(td, errno, "recvfrom udp link open");
    777 		return ret;
    778 	}
    779 
    780 	if (ntohl(msg.magic) != FIO_LINK_OPEN_CLOSE_MAGIC ||
    781 	    ntohl(msg.cmd) != FIO_LINK_OPEN) {
    782 		log_err("fio: bad udp open magic %x/%x\n", ntohl(msg.magic),
    783 								ntohl(msg.cmd));
    784 		return -1;
    785 	}
    786 
    787 	return 0;
    788 }
    789 
    790 static int fio_netio_udp_send_open(struct thread_data *td, struct fio_file *f)
    791 {
    792 	struct netio_data *nd = td->io_ops->data;
    793 	struct netio_options *o = td->eo;
    794 	struct udp_close_msg msg;
    795 	struct sockaddr *to;
    796 	socklen_t len;
    797 	int ret;
    798 
    799 	if (is_ipv6(o)) {
    800 		len = sizeof(nd->addr6);
    801 		to = (struct sockaddr *) &nd->addr6;
    802 	} else {
    803 		len = sizeof(nd->addr);
    804 		to = (struct sockaddr *) &nd->addr;
    805 	}
    806 
    807 	msg.magic = htonl(FIO_LINK_OPEN_CLOSE_MAGIC);
    808 	msg.cmd = htonl(FIO_LINK_OPEN);
    809 
    810 	ret = sendto(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, len);
    811 	if (ret < 0) {
    812 		td_verror(td, errno, "sendto udp link open");
    813 		return ret;
    814 	}
    815 
    816 	return 0;
    817 }
    818 
    819 static int fio_netio_open_file(struct thread_data *td, struct fio_file *f)
    820 {
    821 	int ret;
    822 	struct netio_options *o = td->eo;
    823 
    824 	if (o->listen)
    825 		ret = fio_netio_accept(td, f);
    826 	else
    827 		ret = fio_netio_connect(td, f);
    828 
    829 	if (ret) {
    830 		f->fd = -1;
    831 		return ret;
    832 	}
    833 
    834 	if (is_udp(o)) {
    835 		if (td_write(td))
    836 			ret = fio_netio_udp_send_open(td, f);
    837 		else {
    838 			int state;
    839 
    840 			state = td->runstate;
    841 			td_set_runstate(td, TD_SETTING_UP);
    842 			ret = fio_netio_udp_recv_open(td, f);
    843 			td_set_runstate(td, state);
    844 		}
    845 	}
    846 
    847 	if (ret)
    848 		fio_netio_close_file(td, f);
    849 
    850 	return ret;
    851 }
    852 
    853 static int fio_fill_addr(struct thread_data *td, const char *host, int af,
    854 			 void *dst, struct addrinfo **res)
    855 {
    856 	struct netio_options *o = td->eo;
    857 	struct addrinfo hints;
    858 	int ret;
    859 
    860 	if (inet_pton(af, host, dst))
    861 		return 0;
    862 
    863 	memset(&hints, 0, sizeof(hints));
    864 
    865 	if (is_tcp(o))
    866 		hints.ai_socktype = SOCK_STREAM;
    867 	else
    868 		hints.ai_socktype = SOCK_DGRAM;
    869 
    870 	if (is_ipv6(o))
    871 		hints.ai_family = AF_INET6;
    872 	else
    873 		hints.ai_family = AF_INET;
    874 
    875 	ret = getaddrinfo(host, NULL, &hints, res);
    876 	if (ret) {
    877 		int e = EINVAL;
    878 		char str[128];
    879 
    880 		if (ret == EAI_SYSTEM)
    881 			e = errno;
    882 
    883 		snprintf(str, sizeof(str), "getaddrinfo: %s", gai_strerror(ret));
    884 		td_verror(td, e, str);
    885 		return 1;
    886 	}
    887 
    888 	return 0;
    889 }
    890 
    891 static int fio_netio_setup_connect_inet(struct thread_data *td,
    892 					const char *host, unsigned short port)
    893 {
    894 	struct netio_data *nd = td->io_ops->data;
    895 	struct netio_options *o = td->eo;
    896 	struct addrinfo *res = NULL;
    897 	void *dst, *src;
    898 	int af, len;
    899 
    900 	if (!host) {
    901 		log_err("fio: connect with no host to connect to.\n");
    902 		if (td_read(td))
    903 			log_err("fio: did you forget to set 'listen'?\n");
    904 
    905 		td_verror(td, EINVAL, "no hostname= set");
    906 		return 1;
    907 	}
    908 
    909 	nd->addr.sin_family = AF_INET;
    910 	nd->addr.sin_port = htons(port);
    911 	nd->addr6.sin6_family = AF_INET6;
    912 	nd->addr6.sin6_port = htons(port);
    913 
    914 	if (is_ipv6(o)) {
    915 		af = AF_INET6;
    916 		dst = &nd->addr6.sin6_addr;
    917 	} else {
    918 		af = AF_INET;
    919 		dst = &nd->addr.sin_addr;
    920 	}
    921 
    922 	if (fio_fill_addr(td, host, af, dst, &res))
    923 		return 1;
    924 
    925 	if (!res)
    926 		return 0;
    927 
    928 	if (is_ipv6(o)) {
    929 		len = sizeof(nd->addr6.sin6_addr);
    930 		src = &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr;
    931 	} else {
    932 		len = sizeof(nd->addr.sin_addr);
    933 		src = &((struct sockaddr_in *) res->ai_addr)->sin_addr;
    934 	}
    935 
    936 	memcpy(dst, src, len);
    937 	freeaddrinfo(res);
    938 	return 0;
    939 }
    940 
    941 static int fio_netio_setup_connect_unix(struct thread_data *td,
    942 					const char *path)
    943 {
    944 	struct netio_data *nd = td->io_ops->data;
    945 	struct sockaddr_un *soun = &nd->addr_un;
    946 
    947 	soun->sun_family = AF_UNIX;
    948 	memset(soun->sun_path, 0, sizeof(soun->sun_path));
    949 	strncpy(soun->sun_path, path, sizeof(soun->sun_path) - 1);
    950 	return 0;
    951 }
    952 
    953 static int fio_netio_setup_connect(struct thread_data *td)
    954 {
    955 	struct netio_options *o = td->eo;
    956 
    957 	if (is_udp(o) || is_tcp(o))
    958 		return fio_netio_setup_connect_inet(td, td->o.filename,o->port);
    959 	else
    960 		return fio_netio_setup_connect_unix(td, td->o.filename);
    961 }
    962 
    963 static int fio_netio_setup_listen_unix(struct thread_data *td, const char *path)
    964 {
    965 	struct netio_data *nd = td->io_ops->data;
    966 	struct sockaddr_un *addr = &nd->addr_un;
    967 	mode_t mode;
    968 	int len, fd;
    969 
    970 	fd = socket(AF_UNIX, SOCK_STREAM, 0);
    971 	if (fd < 0) {
    972 		log_err("fio: socket: %s\n", strerror(errno));
    973 		return -1;
    974 	}
    975 
    976 	mode = umask(000);
    977 
    978 	memset(addr, 0, sizeof(*addr));
    979 	addr->sun_family = AF_UNIX;
    980 	strncpy(addr->sun_path, path, sizeof(addr->sun_path) - 1);
    981 	unlink(path);
    982 
    983 	len = sizeof(addr->sun_family) + strlen(path) + 1;
    984 
    985 	if (bind(fd, (struct sockaddr *) addr, len) < 0) {
    986 		log_err("fio: bind: %s\n", strerror(errno));
    987 		close(fd);
    988 		return -1;
    989 	}
    990 
    991 	umask(mode);
    992 	nd->listenfd = fd;
    993 	return 0;
    994 }
    995 
    996 static int fio_netio_setup_listen_inet(struct thread_data *td, short port)
    997 {
    998 	struct netio_data *nd = td->io_ops->data;
    999 	struct netio_options *o = td->eo;
   1000 	struct ip_mreq mr;
   1001 	struct sockaddr_in sin;
   1002 	struct sockaddr *saddr;
   1003 	int fd, opt, type, domain;
   1004 	socklen_t len;
   1005 
   1006 	memset(&sin, 0, sizeof(sin));
   1007 
   1008 	if (o->proto == FIO_TYPE_TCP) {
   1009 		type = SOCK_STREAM;
   1010 		domain = AF_INET;
   1011 	} else if (o->proto == FIO_TYPE_TCP_V6) {
   1012 		type = SOCK_STREAM;
   1013 		domain = AF_INET6;
   1014 	} else if (o->proto == FIO_TYPE_UDP) {
   1015 		type = SOCK_DGRAM;
   1016 		domain = AF_INET;
   1017 	} else if (o->proto == FIO_TYPE_UDP_V6) {
   1018 		type = SOCK_DGRAM;
   1019 		domain = AF_INET6;
   1020 	} else {
   1021 		log_err("fio: unknown proto %d\n", o->proto);
   1022 		return 1;
   1023 	}
   1024 
   1025 	fd = socket(domain, type, 0);
   1026 	if (fd < 0) {
   1027 		td_verror(td, errno, "socket");
   1028 		return 1;
   1029 	}
   1030 
   1031 	opt = 1;
   1032 	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (void *) &opt, sizeof(opt)) < 0) {
   1033 		td_verror(td, errno, "setsockopt");
   1034 		close(fd);
   1035 		return 1;
   1036 	}
   1037 #ifdef SO_REUSEPORT
   1038 	if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, (void *) &opt, sizeof(opt)) < 0) {
   1039 		td_verror(td, errno, "setsockopt");
   1040 		close(fd);
   1041 		return 1;
   1042 	}
   1043 #endif
   1044 
   1045 	if (td->o.filename) {
   1046 		if (!is_udp(o) || !fio_netio_is_multicast(td->o.filename)) {
   1047 			log_err("fio: hostname not valid for non-multicast inbound network IO\n");
   1048 			close(fd);
   1049 			return 1;
   1050 		}
   1051 		if (is_ipv6(o)) {
   1052 			log_err("fio: IPv6 not supported for multicast network IO");
   1053 			close(fd);
   1054 			return 1;
   1055 		}
   1056 
   1057 		inet_aton(td->o.filename, &sin.sin_addr);
   1058 
   1059 		mr.imr_multiaddr = sin.sin_addr;
   1060 		if (o->intfc) {
   1061 			if (inet_aton(o->intfc, &mr.imr_interface) == 0) {
   1062 				log_err("fio: interface not valid interface IP\n");
   1063 				close(fd);
   1064 				return 1;
   1065 			}
   1066 		} else {
   1067 			mr.imr_interface.s_addr = htonl(INADDR_ANY);
   1068 		}
   1069 
   1070 		if (setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP, (const char*)&mr, sizeof(mr)) < 0) {
   1071 			td_verror(td, errno, "setsockopt IP_ADD_MEMBERSHIP");
   1072 			close(fd);
   1073 			return 1;
   1074 		}
   1075 	}
   1076 
   1077 	if (!is_ipv6(o)) {
   1078 		saddr = (struct sockaddr *) &nd->addr;
   1079 		len = sizeof(nd->addr);
   1080 
   1081 		nd->addr.sin_family = AF_INET;
   1082 		nd->addr.sin_addr.s_addr = sin.sin_addr.s_addr ? sin.sin_addr.s_addr : htonl(INADDR_ANY);
   1083 		nd->addr.sin_port = htons(port);
   1084 	} else {
   1085 		saddr = (struct sockaddr *) &nd->addr6;
   1086 		len = sizeof(nd->addr6);
   1087 
   1088 		nd->addr6.sin6_family = AF_INET6;
   1089 		nd->addr6.sin6_addr = in6addr_any;
   1090 		nd->addr6.sin6_port = htons(port);
   1091 	}
   1092 
   1093 	if (bind(fd, saddr, len) < 0) {
   1094 		close(fd);
   1095 		td_verror(td, errno, "bind");
   1096 		return 1;
   1097 	}
   1098 
   1099 	nd->listenfd = fd;
   1100 	return 0;
   1101 }
   1102 
   1103 static int fio_netio_setup_listen(struct thread_data *td)
   1104 {
   1105 	struct netio_data *nd = td->io_ops->data;
   1106 	struct netio_options *o = td->eo;
   1107 	int ret;
   1108 
   1109 	if (is_udp(o) || is_tcp(o))
   1110 		ret = fio_netio_setup_listen_inet(td, o->port);
   1111 	else
   1112 		ret = fio_netio_setup_listen_unix(td, td->o.filename);
   1113 
   1114 	if (ret)
   1115 		return ret;
   1116 	if (is_udp(o))
   1117 		return 0;
   1118 
   1119 	if (listen(nd->listenfd, 10) < 0) {
   1120 		td_verror(td, errno, "listen");
   1121 		nd->listenfd = -1;
   1122 		return 1;
   1123 	}
   1124 
   1125 	return 0;
   1126 }
   1127 
   1128 static int fio_netio_init(struct thread_data *td)
   1129 {
   1130 	struct netio_options *o = td->eo;
   1131 	int ret;
   1132 
   1133 #ifdef WIN32
   1134 	WSADATA wsd;
   1135 	WSAStartup(MAKEWORD(2,2), &wsd);
   1136 #endif
   1137 
   1138 	if (td_random(td)) {
   1139 		log_err("fio: network IO can't be random\n");
   1140 		return 1;
   1141 	}
   1142 
   1143 	if (o->proto == FIO_TYPE_UNIX && o->port) {
   1144 		log_err("fio: network IO port not valid with unix socket\n");
   1145 		return 1;
   1146 	} else if (o->proto != FIO_TYPE_UNIX && !o->port) {
   1147 		log_err("fio: network IO requires port for tcp or udp\n");
   1148 		return 1;
   1149 	}
   1150 
   1151 	if (!is_tcp(o)) {
   1152 		if (o->listen) {
   1153 			log_err("fio: listen only valid for TCP proto IO\n");
   1154 			return 1;
   1155 		}
   1156 		if (td_rw(td)) {
   1157 			log_err("fio: datagram network connections must be"
   1158 				   " read OR write\n");
   1159 			return 1;
   1160 		}
   1161 		if (o->proto == FIO_TYPE_UNIX && !td->o.filename) {
   1162 			log_err("fio: UNIX sockets need host/filename\n");
   1163 			return 1;
   1164 		}
   1165 		o->listen = td_read(td);
   1166 	}
   1167 
   1168 	if (o->listen)
   1169 		ret = fio_netio_setup_listen(td);
   1170 	else
   1171 		ret = fio_netio_setup_connect(td);
   1172 
   1173 	return ret;
   1174 }
   1175 
   1176 static void fio_netio_cleanup(struct thread_data *td)
   1177 {
   1178 	struct netio_data *nd = td->io_ops->data;
   1179 
   1180 	if (nd) {
   1181 		if (nd->listenfd != -1)
   1182 			close(nd->listenfd);
   1183 		if (nd->pipes[0] != -1)
   1184 			close(nd->pipes[0]);
   1185 		if (nd->pipes[1] != -1)
   1186 			close(nd->pipes[1]);
   1187 
   1188 		free(nd);
   1189 	}
   1190 }
   1191 
   1192 static int fio_netio_setup(struct thread_data *td)
   1193 {
   1194 	struct netio_data *nd;
   1195 
   1196 	if (!td->files_index) {
   1197 		add_file(td, td->o.filename ?: "net", 0, 0);
   1198 		td->o.nr_files = td->o.nr_files ?: 1;
   1199 		td->o.open_files++;
   1200 	}
   1201 
   1202 	if (!td->io_ops->data) {
   1203 		nd = malloc(sizeof(*nd));;
   1204 
   1205 		memset(nd, 0, sizeof(*nd));
   1206 		nd->listenfd = -1;
   1207 		nd->pipes[0] = nd->pipes[1] = -1;
   1208 		td->io_ops->data = nd;
   1209 	}
   1210 
   1211 	return 0;
   1212 }
   1213 
   1214 static void fio_netio_terminate(struct thread_data *td)
   1215 {
   1216 	kill(td->pid, SIGUSR2);
   1217 }
   1218 
   1219 #ifdef CONFIG_LINUX_SPLICE
   1220 static int fio_netio_setup_splice(struct thread_data *td)
   1221 {
   1222 	struct netio_data *nd;
   1223 
   1224 	fio_netio_setup(td);
   1225 
   1226 	nd = td->io_ops->data;
   1227 	if (nd) {
   1228 		if (pipe(nd->pipes) < 0)
   1229 			return 1;
   1230 
   1231 		nd->use_splice = 1;
   1232 		return 0;
   1233 	}
   1234 
   1235 	return 1;
   1236 }
   1237 
   1238 static struct ioengine_ops ioengine_splice = {
   1239 	.name			= "netsplice",
   1240 	.version		= FIO_IOOPS_VERSION,
   1241 	.prep			= fio_netio_prep,
   1242 	.queue			= fio_netio_queue,
   1243 	.setup			= fio_netio_setup_splice,
   1244 	.init			= fio_netio_init,
   1245 	.cleanup		= fio_netio_cleanup,
   1246 	.open_file		= fio_netio_open_file,
   1247 	.close_file		= fio_netio_close_file,
   1248 	.terminate		= fio_netio_terminate,
   1249 	.options		= options,
   1250 	.option_struct_size	= sizeof(struct netio_options),
   1251 	.flags			= FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
   1252 				  FIO_PIPEIO,
   1253 };
   1254 #endif
   1255 
   1256 static struct ioengine_ops ioengine_rw = {
   1257 	.name			= "net",
   1258 	.version		= FIO_IOOPS_VERSION,
   1259 	.prep			= fio_netio_prep,
   1260 	.queue			= fio_netio_queue,
   1261 	.setup			= fio_netio_setup,
   1262 	.init			= fio_netio_init,
   1263 	.cleanup		= fio_netio_cleanup,
   1264 	.open_file		= fio_netio_open_file,
   1265 	.close_file		= fio_netio_close_file,
   1266 	.terminate		= fio_netio_terminate,
   1267 	.options		= options,
   1268 	.option_struct_size	= sizeof(struct netio_options),
   1269 	.flags			= FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
   1270 				  FIO_PIPEIO | FIO_BIT_BASED,
   1271 };
   1272 
   1273 static int str_hostname_cb(void *data, const char *input)
   1274 {
   1275 	struct netio_options *o = data;
   1276 
   1277 	if (o->td->o.filename)
   1278 		free(o->td->o.filename);
   1279 	o->td->o.filename = strdup(input);
   1280 	return 0;
   1281 }
   1282 
   1283 static void fio_init fio_netio_register(void)
   1284 {
   1285 	register_ioengine(&ioengine_rw);
   1286 #ifdef CONFIG_LINUX_SPLICE
   1287 	register_ioengine(&ioengine_splice);
   1288 #endif
   1289 }
   1290 
   1291 static void fio_exit fio_netio_unregister(void)
   1292 {
   1293 	unregister_ioengine(&ioengine_rw);
   1294 #ifdef CONFIG_LINUX_SPLICE
   1295 	unregister_ioengine(&ioengine_splice);
   1296 #endif
   1297 }
   1298