net: fix error reported on job exit and full residual
[fio.git] / engines / net.c
... / ...
CommitLineData
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
25struct 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
34struct 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
45struct udp_close_msg {
46 uint32_t magic;
47 uint32_t cmd;
48};
49
50enum {
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
62static int str_hostname_cb(void *data, const char *input);
63static 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
173static inline int is_udp(struct netio_options *o)
174{
175 return o->proto == FIO_TYPE_UDP || o->proto == FIO_TYPE_UDP_V6;
176}
177
178static inline int is_tcp(struct netio_options *o)
179{
180 return o->proto == FIO_TYPE_TCP || o->proto == FIO_TYPE_TCP_V6;
181}
182
183static 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 */
192static 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
219static 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
233static 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
253static 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 */
278static 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 */
288static 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
296static 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 */
326static 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 */
337static 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 */
348static 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 */
363static 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
374static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
375{
376 errno = EOPNOTSUPP;
377 return -1;
378}
379
380static 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
387static 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
431static 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
447static 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
495static 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 if (!ret)
523 return FIO_Q_BUSY;
524 else {
525 int err = errno;
526
527 if (ddir == DDIR_WRITE && err == EMSGSIZE)
528 return FIO_Q_BUSY;
529
530 io_u->error = err;
531 }
532 }
533
534 if (io_u->error)
535 td_verror(td, io_u->error, "xfer");
536
537 return FIO_Q_COMPLETED;
538}
539
540static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
541{
542 struct netio_options *o = td->eo;
543 int ret;
544
545 fio_ro_check(td, io_u);
546
547 ret = __fio_netio_queue(td, io_u, io_u->ddir);
548 if (!o->pingpong || ret != FIO_Q_COMPLETED)
549 return ret;
550
551 /*
552 * For ping-pong mode, receive or send reply as needed
553 */
554 if (td_read(td) && io_u->ddir == DDIR_READ)
555 ret = __fio_netio_queue(td, io_u, DDIR_WRITE);
556 else if (td_write(td) && io_u->ddir == DDIR_WRITE)
557 ret = __fio_netio_queue(td, io_u, DDIR_READ);
558
559 return ret;
560}
561
562static int fio_netio_connect(struct thread_data *td, struct fio_file *f)
563{
564 struct netio_data *nd = td->io_ops->data;
565 struct netio_options *o = td->eo;
566 int type, domain;
567
568 if (o->proto == FIO_TYPE_TCP) {
569 domain = AF_INET;
570 type = SOCK_STREAM;
571 } else if (o->proto == FIO_TYPE_TCP_V6) {
572 domain = AF_INET6;
573 type = SOCK_STREAM;
574 } else if (o->proto == FIO_TYPE_UDP) {
575 domain = AF_INET;
576 type = SOCK_DGRAM;
577 } else if (o->proto == FIO_TYPE_UDP_V6) {
578 domain = AF_INET6;
579 type = SOCK_DGRAM;
580 } else if (o->proto == FIO_TYPE_UNIX) {
581 domain = AF_UNIX;
582 type = SOCK_STREAM;
583 } else {
584 log_err("fio: bad network type %d\n", o->proto);
585 f->fd = -1;
586 return 1;
587 }
588
589 f->fd = socket(domain, type, 0);
590 if (f->fd < 0) {
591 td_verror(td, errno, "socket");
592 return 1;
593 }
594
595#ifdef CONFIG_TCP_NODELAY
596 if (o->nodelay && is_tcp(o)) {
597 int optval = 1;
598
599 if (setsockopt(f->fd, IPPROTO_TCP, TCP_NODELAY, (void *) &optval, sizeof(int)) < 0) {
600 log_err("fio: cannot set TCP_NODELAY option on socket (%s), disable with 'nodelay=0'\n", strerror(errno));
601 return 1;
602 }
603 }
604#endif
605
606 if (is_udp(o)) {
607 if (!fio_netio_is_multicast(td->o.filename))
608 return 0;
609 if (is_ipv6(o)) {
610 log_err("fio: multicast not supported on IPv6\n");
611 close(f->fd);
612 return 1;
613 }
614
615 if (o->intfc) {
616 struct in_addr interface_addr;
617
618 if (inet_aton(o->intfc, &interface_addr) == 0) {
619 log_err("fio: interface not valid interface IP\n");
620 close(f->fd);
621 return 1;
622 }
623 if (setsockopt(f->fd, IPPROTO_IP, IP_MULTICAST_IF, (const char*)&interface_addr, sizeof(interface_addr)) < 0) {
624 td_verror(td, errno, "setsockopt IP_MULTICAST_IF");
625 close(f->fd);
626 return 1;
627 }
628 }
629 if (setsockopt(f->fd, IPPROTO_IP, IP_MULTICAST_TTL, (const char*)&o->ttl, sizeof(o->ttl)) < 0) {
630 td_verror(td, errno, "setsockopt IP_MULTICAST_TTL");
631 close(f->fd);
632 return 1;
633 }
634 return 0;
635 } else if (o->proto == FIO_TYPE_TCP) {
636 socklen_t len = sizeof(nd->addr);
637
638 if (connect(f->fd, (struct sockaddr *) &nd->addr, len) < 0) {
639 td_verror(td, errno, "connect");
640 close(f->fd);
641 return 1;
642 }
643 } else if (o->proto == FIO_TYPE_TCP_V6) {
644 socklen_t len = sizeof(nd->addr6);
645
646 if (connect(f->fd, (struct sockaddr *) &nd->addr6, len) < 0) {
647 td_verror(td, errno, "connect");
648 close(f->fd);
649 return 1;
650 }
651
652 } else {
653 struct sockaddr_un *addr = &nd->addr_un;
654 socklen_t len;
655
656 len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
657
658 if (connect(f->fd, (struct sockaddr *) addr, len) < 0) {
659 td_verror(td, errno, "connect");
660 close(f->fd);
661 return 1;
662 }
663 }
664
665 return 0;
666}
667
668static int fio_netio_accept(struct thread_data *td, struct fio_file *f)
669{
670 struct netio_data *nd = td->io_ops->data;
671 struct netio_options *o = td->eo;
672 socklen_t socklen;
673 int state;
674
675 if (is_udp(o)) {
676 f->fd = nd->listenfd;
677 return 0;
678 }
679
680 state = td->runstate;
681 td_set_runstate(td, TD_SETTING_UP);
682
683 log_info("fio: waiting for connection\n");
684
685 if (poll_wait(td, nd->listenfd, POLLIN) < 0)
686 goto err;
687
688 if (o->proto == FIO_TYPE_TCP) {
689 socklen = sizeof(nd->addr);
690 f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
691 } else {
692 socklen = sizeof(nd->addr6);
693 f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr6, &socklen);
694 }
695
696 if (f->fd < 0) {
697 td_verror(td, errno, "accept");
698 goto err;
699 }
700
701#ifdef CONFIG_TCP_NODELAY
702 if (o->nodelay && is_tcp(o)) {
703 int optval = 1;
704
705 if (setsockopt(f->fd, IPPROTO_TCP, TCP_NODELAY, (void *) &optval, sizeof(int)) < 0) {
706 log_err("fio: cannot set TCP_NODELAY option on socket (%s), disable with 'nodelay=0'\n", strerror(errno));
707 return 1;
708 }
709 }
710#endif
711
712 reset_all_stats(td);
713 td_set_runstate(td, state);
714 return 0;
715err:
716 td_set_runstate(td, state);
717 return 1;
718}
719
720static void fio_netio_udp_close(struct thread_data *td, struct fio_file *f)
721{
722 struct netio_data *nd = td->io_ops->data;
723 struct netio_options *o = td->eo;
724 struct udp_close_msg msg;
725 struct sockaddr *to;
726 socklen_t len;
727 int ret;
728
729 if (is_ipv6(o)) {
730 to = (struct sockaddr *) &nd->addr6;
731 len = sizeof(nd->addr6);
732 } else {
733 to = (struct sockaddr *) &nd->addr;
734 len = sizeof(nd->addr);
735 }
736
737 msg.magic = htonl(FIO_LINK_OPEN_CLOSE_MAGIC);
738 msg.cmd = htonl(FIO_LINK_CLOSE);
739
740 ret = sendto(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, len);
741 if (ret < 0)
742 td_verror(td, errno, "sendto udp link close");
743}
744
745static int fio_netio_close_file(struct thread_data *td, struct fio_file *f)
746{
747 struct netio_options *o = td->eo;
748
749 /*
750 * If this is an UDP connection, notify the receiver that we are
751 * closing down the link
752 */
753 if (is_udp(o))
754 fio_netio_udp_close(td, f);
755
756 return generic_close_file(td, f);
757}
758
759static int fio_netio_udp_recv_open(struct thread_data *td, struct fio_file *f)
760{
761 struct netio_data *nd = td->io_ops->data;
762 struct netio_options *o = td->eo;
763 struct udp_close_msg msg;
764 struct sockaddr *to;
765 socklen_t len;
766 int ret;
767
768 if (is_ipv6(o)) {
769 len = sizeof(nd->addr6);
770 to = (struct sockaddr *) &nd->addr6;
771 } else {
772 len = sizeof(nd->addr);
773 to = (struct sockaddr *) &nd->addr;
774 }
775
776 ret = recvfrom(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, &len);
777 if (ret < 0) {
778 td_verror(td, errno, "recvfrom udp link open");
779 return ret;
780 }
781
782 if (ntohl(msg.magic) != FIO_LINK_OPEN_CLOSE_MAGIC ||
783 ntohl(msg.cmd) != FIO_LINK_OPEN) {
784 log_err("fio: bad udp open magic %x/%x\n", ntohl(msg.magic),
785 ntohl(msg.cmd));
786 return -1;
787 }
788
789 return 0;
790}
791
792static int fio_netio_udp_send_open(struct thread_data *td, struct fio_file *f)
793{
794 struct netio_data *nd = td->io_ops->data;
795 struct netio_options *o = td->eo;
796 struct udp_close_msg msg;
797 struct sockaddr *to;
798 socklen_t len;
799 int ret;
800
801 if (is_ipv6(o)) {
802 len = sizeof(nd->addr6);
803 to = (struct sockaddr *) &nd->addr6;
804 } else {
805 len = sizeof(nd->addr);
806 to = (struct sockaddr *) &nd->addr;
807 }
808
809 msg.magic = htonl(FIO_LINK_OPEN_CLOSE_MAGIC);
810 msg.cmd = htonl(FIO_LINK_OPEN);
811
812 ret = sendto(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, len);
813 if (ret < 0) {
814 td_verror(td, errno, "sendto udp link open");
815 return ret;
816 }
817
818 return 0;
819}
820
821static int fio_netio_open_file(struct thread_data *td, struct fio_file *f)
822{
823 int ret;
824 struct netio_options *o = td->eo;
825
826 if (o->listen)
827 ret = fio_netio_accept(td, f);
828 else
829 ret = fio_netio_connect(td, f);
830
831 if (ret) {
832 f->fd = -1;
833 return ret;
834 }
835
836 if (is_udp(o)) {
837 if (td_write(td))
838 ret = fio_netio_udp_send_open(td, f);
839 else {
840 int state;
841
842 state = td->runstate;
843 td_set_runstate(td, TD_SETTING_UP);
844 ret = fio_netio_udp_recv_open(td, f);
845 td_set_runstate(td, state);
846 }
847 }
848
849 if (ret)
850 fio_netio_close_file(td, f);
851
852 return ret;
853}
854
855static int fio_fill_addr(struct thread_data *td, const char *host, int af,
856 void *dst, struct addrinfo **res)
857{
858 struct netio_options *o = td->eo;
859 struct addrinfo hints;
860 int ret;
861
862 if (inet_pton(af, host, dst))
863 return 0;
864
865 memset(&hints, 0, sizeof(hints));
866
867 if (is_tcp(o))
868 hints.ai_socktype = SOCK_STREAM;
869 else
870 hints.ai_socktype = SOCK_DGRAM;
871
872 if (is_ipv6(o))
873 hints.ai_family = AF_INET6;
874 else
875 hints.ai_family = AF_INET;
876
877 ret = getaddrinfo(host, NULL, &hints, res);
878 if (ret) {
879 int e = EINVAL;
880 char str[128];
881
882 if (ret == EAI_SYSTEM)
883 e = errno;
884
885 snprintf(str, sizeof(str), "getaddrinfo: %s", gai_strerror(ret));
886 td_verror(td, e, str);
887 return 1;
888 }
889
890 return 0;
891}
892
893static int fio_netio_setup_connect_inet(struct thread_data *td,
894 const char *host, unsigned short port)
895{
896 struct netio_data *nd = td->io_ops->data;
897 struct netio_options *o = td->eo;
898 struct addrinfo *res = NULL;
899 void *dst, *src;
900 int af, len;
901
902 if (!host) {
903 log_err("fio: connect with no host to connect to.\n");
904 if (td_read(td))
905 log_err("fio: did you forget to set 'listen'?\n");
906
907 td_verror(td, EINVAL, "no hostname= set");
908 return 1;
909 }
910
911 nd->addr.sin_family = AF_INET;
912 nd->addr.sin_port = htons(port);
913 nd->addr6.sin6_family = AF_INET6;
914 nd->addr6.sin6_port = htons(port);
915
916 if (is_ipv6(o)) {
917 af = AF_INET6;
918 dst = &nd->addr6.sin6_addr;
919 } else {
920 af = AF_INET;
921 dst = &nd->addr.sin_addr;
922 }
923
924 if (fio_fill_addr(td, host, af, dst, &res))
925 return 1;
926
927 if (!res)
928 return 0;
929
930 if (is_ipv6(o)) {
931 len = sizeof(nd->addr6.sin6_addr);
932 src = &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr;
933 } else {
934 len = sizeof(nd->addr.sin_addr);
935 src = &((struct sockaddr_in *) res->ai_addr)->sin_addr;
936 }
937
938 memcpy(dst, src, len);
939 freeaddrinfo(res);
940 return 0;
941}
942
943static int fio_netio_setup_connect_unix(struct thread_data *td,
944 const char *path)
945{
946 struct netio_data *nd = td->io_ops->data;
947 struct sockaddr_un *soun = &nd->addr_un;
948
949 soun->sun_family = AF_UNIX;
950 memset(soun->sun_path, 0, sizeof(soun->sun_path));
951 strncpy(soun->sun_path, path, sizeof(soun->sun_path) - 1);
952 return 0;
953}
954
955static int fio_netio_setup_connect(struct thread_data *td)
956{
957 struct netio_options *o = td->eo;
958
959 if (is_udp(o) || is_tcp(o))
960 return fio_netio_setup_connect_inet(td, td->o.filename,o->port);
961 else
962 return fio_netio_setup_connect_unix(td, td->o.filename);
963}
964
965static int fio_netio_setup_listen_unix(struct thread_data *td, const char *path)
966{
967 struct netio_data *nd = td->io_ops->data;
968 struct sockaddr_un *addr = &nd->addr_un;
969 mode_t mode;
970 int len, fd;
971
972 fd = socket(AF_UNIX, SOCK_STREAM, 0);
973 if (fd < 0) {
974 log_err("fio: socket: %s\n", strerror(errno));
975 return -1;
976 }
977
978 mode = umask(000);
979
980 memset(addr, 0, sizeof(*addr));
981 addr->sun_family = AF_UNIX;
982 strncpy(addr->sun_path, path, sizeof(addr->sun_path) - 1);
983 unlink(path);
984
985 len = sizeof(addr->sun_family) + strlen(path) + 1;
986
987 if (bind(fd, (struct sockaddr *) addr, len) < 0) {
988 log_err("fio: bind: %s\n", strerror(errno));
989 close(fd);
990 return -1;
991 }
992
993 umask(mode);
994 nd->listenfd = fd;
995 return 0;
996}
997
998static int fio_netio_setup_listen_inet(struct thread_data *td, short port)
999{
1000 struct netio_data *nd = td->io_ops->data;
1001 struct netio_options *o = td->eo;
1002 struct ip_mreq mr;
1003 struct sockaddr_in sin;
1004 struct sockaddr *saddr;
1005 int fd, opt, type, domain;
1006 socklen_t len;
1007
1008 memset(&sin, 0, sizeof(sin));
1009
1010 if (o->proto == FIO_TYPE_TCP) {
1011 type = SOCK_STREAM;
1012 domain = AF_INET;
1013 } else if (o->proto == FIO_TYPE_TCP_V6) {
1014 type = SOCK_STREAM;
1015 domain = AF_INET6;
1016 } else if (o->proto == FIO_TYPE_UDP) {
1017 type = SOCK_DGRAM;
1018 domain = AF_INET;
1019 } else if (o->proto == FIO_TYPE_UDP_V6) {
1020 type = SOCK_DGRAM;
1021 domain = AF_INET6;
1022 } else {
1023 log_err("fio: unknown proto %d\n", o->proto);
1024 return 1;
1025 }
1026
1027 fd = socket(domain, type, 0);
1028 if (fd < 0) {
1029 td_verror(td, errno, "socket");
1030 return 1;
1031 }
1032
1033 opt = 1;
1034 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (void *) &opt, sizeof(opt)) < 0) {
1035 td_verror(td, errno, "setsockopt");
1036 close(fd);
1037 return 1;
1038 }
1039#ifdef SO_REUSEPORT
1040 if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, (void *) &opt, sizeof(opt)) < 0) {
1041 td_verror(td, errno, "setsockopt");
1042 close(fd);
1043 return 1;
1044 }
1045#endif
1046
1047 if (td->o.filename) {
1048 if (!is_udp(o) || !fio_netio_is_multicast(td->o.filename)) {
1049 log_err("fio: hostname not valid for non-multicast inbound network IO\n");
1050 close(fd);
1051 return 1;
1052 }
1053 if (is_ipv6(o)) {
1054 log_err("fio: IPv6 not supported for multicast network IO");
1055 close(fd);
1056 return 1;
1057 }
1058
1059 inet_aton(td->o.filename, &sin.sin_addr);
1060
1061 mr.imr_multiaddr = sin.sin_addr;
1062 if (o->intfc) {
1063 if (inet_aton(o->intfc, &mr.imr_interface) == 0) {
1064 log_err("fio: interface not valid interface IP\n");
1065 close(fd);
1066 return 1;
1067 }
1068 } else {
1069 mr.imr_interface.s_addr = htonl(INADDR_ANY);
1070 }
1071
1072 if (setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP, (const char*)&mr, sizeof(mr)) < 0) {
1073 td_verror(td, errno, "setsockopt IP_ADD_MEMBERSHIP");
1074 close(fd);
1075 return 1;
1076 }
1077 }
1078
1079 if (!is_ipv6(o)) {
1080 saddr = (struct sockaddr *) &nd->addr;
1081 len = sizeof(nd->addr);
1082
1083 nd->addr.sin_family = AF_INET;
1084 nd->addr.sin_addr.s_addr = sin.sin_addr.s_addr ? sin.sin_addr.s_addr : htonl(INADDR_ANY);
1085 nd->addr.sin_port = htons(port);
1086 } else {
1087 saddr = (struct sockaddr *) &nd->addr6;
1088 len = sizeof(nd->addr6);
1089
1090 nd->addr6.sin6_family = AF_INET6;
1091 nd->addr6.sin6_addr = in6addr_any;
1092 nd->addr6.sin6_port = htons(port);
1093 }
1094
1095 if (bind(fd, saddr, len) < 0) {
1096 close(fd);
1097 td_verror(td, errno, "bind");
1098 return 1;
1099 }
1100
1101 nd->listenfd = fd;
1102 return 0;
1103}
1104
1105static int fio_netio_setup_listen(struct thread_data *td)
1106{
1107 struct netio_data *nd = td->io_ops->data;
1108 struct netio_options *o = td->eo;
1109 int ret;
1110
1111 if (is_udp(o) || is_tcp(o))
1112 ret = fio_netio_setup_listen_inet(td, o->port);
1113 else
1114 ret = fio_netio_setup_listen_unix(td, td->o.filename);
1115
1116 if (ret)
1117 return ret;
1118 if (is_udp(o))
1119 return 0;
1120
1121 if (listen(nd->listenfd, 10) < 0) {
1122 td_verror(td, errno, "listen");
1123 nd->listenfd = -1;
1124 return 1;
1125 }
1126
1127 return 0;
1128}
1129
1130static int fio_netio_init(struct thread_data *td)
1131{
1132 struct netio_options *o = td->eo;
1133 int ret;
1134
1135#ifdef WIN32
1136 WSADATA wsd;
1137 WSAStartup(MAKEWORD(2,2), &wsd);
1138#endif
1139
1140 if (td_random(td)) {
1141 log_err("fio: network IO can't be random\n");
1142 return 1;
1143 }
1144
1145 if (o->proto == FIO_TYPE_UNIX && o->port) {
1146 log_err("fio: network IO port not valid with unix socket\n");
1147 return 1;
1148 } else if (o->proto != FIO_TYPE_UNIX && !o->port) {
1149 log_err("fio: network IO requires port for tcp or udp\n");
1150 return 1;
1151 }
1152
1153 if (!is_tcp(o)) {
1154 if (o->listen) {
1155 log_err("fio: listen only valid for TCP proto IO\n");
1156 return 1;
1157 }
1158 if (td_rw(td)) {
1159 log_err("fio: datagram network connections must be"
1160 " read OR write\n");
1161 return 1;
1162 }
1163 if (o->proto == FIO_TYPE_UNIX && !td->o.filename) {
1164 log_err("fio: UNIX sockets need host/filename\n");
1165 return 1;
1166 }
1167 o->listen = td_read(td);
1168 }
1169
1170 if (o->listen)
1171 ret = fio_netio_setup_listen(td);
1172 else
1173 ret = fio_netio_setup_connect(td);
1174
1175 return ret;
1176}
1177
1178static void fio_netio_cleanup(struct thread_data *td)
1179{
1180 struct netio_data *nd = td->io_ops->data;
1181
1182 if (nd) {
1183 if (nd->listenfd != -1)
1184 close(nd->listenfd);
1185 if (nd->pipes[0] != -1)
1186 close(nd->pipes[0]);
1187 if (nd->pipes[1] != -1)
1188 close(nd->pipes[1]);
1189
1190 free(nd);
1191 }
1192}
1193
1194static int fio_netio_setup(struct thread_data *td)
1195{
1196 struct netio_data *nd;
1197
1198 if (!td->files_index) {
1199 add_file(td, td->o.filename ?: "net", 0, 0);
1200 td->o.nr_files = td->o.nr_files ?: 1;
1201 td->o.open_files++;
1202 }
1203
1204 if (!td->io_ops->data) {
1205 nd = malloc(sizeof(*nd));;
1206
1207 memset(nd, 0, sizeof(*nd));
1208 nd->listenfd = -1;
1209 nd->pipes[0] = nd->pipes[1] = -1;
1210 td->io_ops->data = nd;
1211 }
1212
1213 return 0;
1214}
1215
1216static void fio_netio_terminate(struct thread_data *td)
1217{
1218 kill(td->pid, SIGTERM);
1219}
1220
1221#ifdef CONFIG_LINUX_SPLICE
1222static int fio_netio_setup_splice(struct thread_data *td)
1223{
1224 struct netio_data *nd;
1225
1226 fio_netio_setup(td);
1227
1228 nd = td->io_ops->data;
1229 if (nd) {
1230 if (pipe(nd->pipes) < 0)
1231 return 1;
1232
1233 nd->use_splice = 1;
1234 return 0;
1235 }
1236
1237 return 1;
1238}
1239
1240static struct ioengine_ops ioengine_splice = {
1241 .name = "netsplice",
1242 .version = FIO_IOOPS_VERSION,
1243 .prep = fio_netio_prep,
1244 .queue = fio_netio_queue,
1245 .setup = fio_netio_setup_splice,
1246 .init = fio_netio_init,
1247 .cleanup = fio_netio_cleanup,
1248 .open_file = fio_netio_open_file,
1249 .close_file = fio_netio_close_file,
1250 .terminate = fio_netio_terminate,
1251 .options = options,
1252 .option_struct_size = sizeof(struct netio_options),
1253 .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
1254 FIO_PIPEIO,
1255};
1256#endif
1257
1258static struct ioengine_ops ioengine_rw = {
1259 .name = "net",
1260 .version = FIO_IOOPS_VERSION,
1261 .prep = fio_netio_prep,
1262 .queue = fio_netio_queue,
1263 .setup = fio_netio_setup,
1264 .init = fio_netio_init,
1265 .cleanup = fio_netio_cleanup,
1266 .open_file = fio_netio_open_file,
1267 .close_file = fio_netio_close_file,
1268 .terminate = fio_netio_terminate,
1269 .options = options,
1270 .option_struct_size = sizeof(struct netio_options),
1271 .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
1272 FIO_PIPEIO | FIO_BIT_BASED,
1273};
1274
1275static int str_hostname_cb(void *data, const char *input)
1276{
1277 struct netio_options *o = data;
1278
1279 if (o->td->o.filename)
1280 free(o->td->o.filename);
1281 o->td->o.filename = strdup(input);
1282 return 0;
1283}
1284
1285static void fio_init fio_netio_register(void)
1286{
1287 register_ioengine(&ioengine_rw);
1288#ifdef CONFIG_LINUX_SPLICE
1289 register_ioengine(&ioengine_splice);
1290#endif
1291}
1292
1293static void fio_exit fio_netio_unregister(void)
1294{
1295 unregister_ioengine(&ioengine_rw);
1296#ifdef CONFIG_LINUX_SPLICE
1297 unregister_ioengine(&ioengine_splice);
1298#endif
1299}