4 * IO engine that reads/writes to/from sockets.
13 #include <netinet/in.h>
14 #include <netinet/tcp.h>
15 #include <arpa/inet.h>
18 #include <sys/types.h>
20 #include <sys/socket.h>
29 struct sockaddr_in addr;
30 struct sockaddr_un addr_un;
33 struct netio_options {
34 struct thread_data *td;
38 unsigned int pingpong;
42 struct udp_close_msg {
48 FIO_LINK_CLOSE = 0x89,
49 FIO_LINK_OPEN_CLOSE_MAGIC = 0x6c696e6b,
57 static int str_hostname_cb(void *data, const char *input);
58 static struct fio_option options[] = {
61 .lname = "net engine hostname",
62 .type = FIO_OPT_STR_STORE,
63 .cb = str_hostname_cb,
64 .help = "Hostname for net IO engine",
65 .category = FIO_OPT_C_IO,
69 .lname = "net engine port",
71 .off1 = offsetof(struct netio_options, port),
74 .help = "Port to use for TCP or UDP net connections",
75 .category = FIO_OPT_C_IO,
79 .lname = "net engine protocol",
82 .off1 = offsetof(struct netio_options, proto),
83 .help = "Network protocol to use",
85 .category = FIO_OPT_C_IO,
89 .help = "Transmission Control Protocol",
93 .help = "User Datagram Protocol",
96 .oval = FIO_TYPE_UNIX,
97 .help = "UNIX domain socket",
101 #ifdef CONFIG_TCP_NODELAY
104 .type = FIO_OPT_BOOL,
105 .off1 = offsetof(struct netio_options, nodelay),
106 .help = "Use TCP_NODELAY on TCP connections",
111 .lname = "net engine listen",
112 .type = FIO_OPT_STR_SET,
113 .off1 = offsetof(struct netio_options, listen),
114 .help = "Listen for incoming TCP connections",
115 .category = FIO_OPT_C_IO,
119 .type = FIO_OPT_STR_SET,
120 .off1 = offsetof(struct netio_options, pingpong),
121 .help = "Ping-pong IO requests",
129 * Return -1 for error and 'nr events' for a positive number
132 static int poll_wait(struct thread_data *td, int fd, short events)
137 while (!td->terminate) {
140 ret = poll(&pfd, 1, -1);
145 td_verror(td, errno, "poll");
153 if (pfd.revents & events)
159 static int fio_netio_prep(struct thread_data *td, struct io_u *io_u)
161 struct netio_options *o = td->eo;
164 * Make sure we don't see spurious reads to a receiver, and vice versa
166 if (o->proto == FIO_TYPE_TCP)
169 if ((o->listen && io_u->ddir == DDIR_WRITE) ||
170 (!o->listen && io_u->ddir == DDIR_READ)) {
171 td_verror(td, EINVAL, "bad direction");
178 #ifdef CONFIG_LINUX_SPLICE
179 static int splice_io_u(int fdin, int fdout, unsigned int len)
184 int ret = splice(fdin, NULL, fdout, NULL, len, 0);
202 * Receive bytes from a socket and fill them into the internal pipe
204 static int splice_in(struct thread_data *td, struct io_u *io_u)
206 struct netio_data *nd = td->io_ops->data;
208 return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen);
212 * Transmit 'len' bytes from the internal pipe
214 static int splice_out(struct thread_data *td, struct io_u *io_u,
217 struct netio_data *nd = td->io_ops->data;
219 return splice_io_u(nd->pipes[0], io_u->file->fd, len);
222 static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len)
225 .iov_base = io_u->xfer_buf,
230 while (iov.iov_len) {
231 int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE);
250 * vmsplice() pipe to io_u buffer
252 static int vmsplice_io_u_out(struct thread_data *td, struct io_u *io_u,
255 struct netio_data *nd = td->io_ops->data;
257 return vmsplice_io_u(io_u, nd->pipes[0], len);
261 * vmsplice() io_u to pipe
263 static int vmsplice_io_u_in(struct thread_data *td, struct io_u *io_u)
265 struct netio_data *nd = td->io_ops->data;
267 return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen);
271 * splice receive - transfer socket data into a pipe using splice, then map
272 * that pipe data into the io_u using vmsplice.
274 static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
278 ret = splice_in(td, io_u);
280 return vmsplice_io_u_out(td, io_u, ret);
286 * splice transmit - map data from the io_u into a pipe by using vmsplice,
287 * then transfer that pipe to a socket using splice.
289 static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u)
293 ret = vmsplice_io_u_in(td, io_u);
295 return splice_out(td, io_u, ret);
300 static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u)
306 static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u)
313 static int fio_netio_send(struct thread_data *td, struct io_u *io_u)
315 struct netio_data *nd = td->io_ops->data;
316 struct netio_options *o = td->eo;
320 if (o->proto == FIO_TYPE_UDP) {
321 struct sockaddr *to = (struct sockaddr *) &nd->addr;
323 ret = sendto(io_u->file->fd, io_u->xfer_buf,
324 io_u->xfer_buflen, flags, to,
328 * if we are going to write more, set MSG_MORE
331 if ((td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen <
332 td->o.size) && !o->pingpong)
335 ret = send(io_u->file->fd, io_u->xfer_buf,
336 io_u->xfer_buflen, flags);
341 ret = poll_wait(td, io_u->file->fd, POLLOUT);
349 static int is_udp_close(struct io_u *io_u, int len)
351 struct udp_close_msg *msg;
353 if (len != sizeof(struct udp_close_msg))
356 msg = io_u->xfer_buf;
357 if (ntohl(msg->magic) != FIO_LINK_OPEN_CLOSE_MAGIC)
359 if (ntohl(msg->cmd) != FIO_LINK_CLOSE)
365 static int fio_netio_recv(struct thread_data *td, struct io_u *io_u)
367 struct netio_data *nd = td->io_ops->data;
368 struct netio_options *o = td->eo;
372 if (o->proto == FIO_TYPE_UDP) {
373 socklen_t len = sizeof(nd->addr);
374 struct sockaddr *from = (struct sockaddr *) &nd->addr;
376 ret = recvfrom(io_u->file->fd, io_u->xfer_buf,
377 io_u->xfer_buflen, flags, from, &len);
378 if (is_udp_close(io_u, ret)) {
383 ret = recv(io_u->file->fd, io_u->xfer_buf,
384 io_u->xfer_buflen, flags);
388 else if (!ret && (flags & MSG_WAITALL))
391 ret = poll_wait(td, io_u->file->fd, POLLIN);
394 flags |= MSG_WAITALL;
400 static int __fio_netio_queue(struct thread_data *td, struct io_u *io_u,
403 struct netio_data *nd = td->io_ops->data;
404 struct netio_options *o = td->eo;
407 if (ddir == DDIR_WRITE) {
408 if (!nd->use_splice || o->proto == FIO_TYPE_UDP ||
409 o->proto == FIO_TYPE_UNIX)
410 ret = fio_netio_send(td, io_u);
412 ret = fio_netio_splice_out(td, io_u);
413 } else if (ddir == DDIR_READ) {
414 if (!nd->use_splice || o->proto == FIO_TYPE_UDP ||
415 o->proto == FIO_TYPE_UNIX)
416 ret = fio_netio_recv(td, io_u);
418 ret = fio_netio_splice_in(td, io_u);
420 ret = 0; /* must be a SYNC */
422 if (ret != (int) io_u->xfer_buflen) {
424 io_u->resid = io_u->xfer_buflen - ret;
426 return FIO_Q_COMPLETED;
430 if (ddir == DDIR_WRITE && err == EMSGSIZE)
438 td_verror(td, io_u->error, "xfer");
440 return FIO_Q_COMPLETED;
443 static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
445 struct netio_options *o = td->eo;
448 fio_ro_check(td, io_u);
450 ret = __fio_netio_queue(td, io_u, io_u->ddir);
451 if (!o->pingpong || ret != FIO_Q_COMPLETED)
455 * For ping-pong mode, receive or send reply as needed
457 if (td_read(td) && io_u->ddir == DDIR_READ)
458 ret = __fio_netio_queue(td, io_u, DDIR_WRITE);
459 else if (td_write(td) && io_u->ddir == DDIR_WRITE)
460 ret = __fio_netio_queue(td, io_u, DDIR_READ);
465 static int fio_netio_connect(struct thread_data *td, struct fio_file *f)
467 struct netio_data *nd = td->io_ops->data;
468 struct netio_options *o = td->eo;
469 int type, domain, optval;
471 if (o->proto == FIO_TYPE_TCP) {
474 } else if (o->proto == FIO_TYPE_UDP) {
477 } else if (o->proto == FIO_TYPE_UNIX) {
481 log_err("fio: bad network type %d\n", o->proto);
486 f->fd = socket(domain, type, 0);
488 td_verror(td, errno, "socket");
492 #ifdef CONFIG_TCP_NODELAY
493 if (o->nodelay && o->proto == FIO_TYPE_TCP) {
495 if (setsockopt(f->fd, IPPROTO_TCP, TCP_NODELAY, (void *) &optval, sizeof(int)) < 0) {
496 log_err("fio: cannot set TCP_NODELAY option on socket (%s), disable with 'nodelay=0'\n", strerror(errno));
502 if (o->proto == FIO_TYPE_UDP)
504 else if (o->proto == FIO_TYPE_TCP) {
505 socklen_t len = sizeof(nd->addr);
507 if (connect(f->fd, (struct sockaddr *) &nd->addr, len) < 0) {
508 td_verror(td, errno, "connect");
513 struct sockaddr_un *addr = &nd->addr_un;
516 len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
518 if (connect(f->fd, (struct sockaddr *) addr, len) < 0) {
519 td_verror(td, errno, "connect");
528 static int fio_netio_accept(struct thread_data *td, struct fio_file *f)
530 struct netio_data *nd = td->io_ops->data;
531 struct netio_options *o = td->eo;
532 socklen_t socklen = sizeof(nd->addr);
535 if (o->proto == FIO_TYPE_UDP) {
536 f->fd = nd->listenfd;
540 state = td->runstate;
541 td_set_runstate(td, TD_SETTING_UP);
543 log_info("fio: waiting for connection\n");
545 if (poll_wait(td, nd->listenfd, POLLIN) < 0)
548 f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
550 td_verror(td, errno, "accept");
554 #ifdef CONFIG_TCP_NODELAY
555 if (o->nodelay && o->proto == FIO_TYPE_TCP) {
557 if (setsockopt(f->fd, IPPROTO_TCP, TCP_NODELAY, (void *) &optval, sizeof(int)) < 0) {
558 log_err("fio: cannot set TCP_NODELAY option on socket (%s), disable with 'nodelay=0'\n", strerror(errno));
565 td_set_runstate(td, state);
568 td_set_runstate(td, state);
572 static void fio_netio_udp_close(struct thread_data *td, struct fio_file *f)
574 struct netio_data *nd = td->io_ops->data;
575 struct udp_close_msg msg;
576 struct sockaddr *to = (struct sockaddr *) &nd->addr;
579 msg.magic = htonl(FIO_LINK_OPEN_CLOSE_MAGIC);
580 msg.cmd = htonl(FIO_LINK_CLOSE);
582 ret = sendto(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to,
585 td_verror(td, errno, "sendto udp link close");
588 static int fio_netio_close_file(struct thread_data *td, struct fio_file *f)
590 struct netio_options *o = td->eo;
593 * If this is an UDP connection, notify the receiver that we are
594 * closing down the link
596 if (o->proto == FIO_TYPE_UDP)
597 fio_netio_udp_close(td, f);
599 return generic_close_file(td, f);
602 static int fio_netio_udp_recv_open(struct thread_data *td, struct fio_file *f)
604 struct netio_data *nd = td->io_ops->data;
605 struct udp_close_msg msg;
606 struct sockaddr *to = (struct sockaddr *) &nd->addr;
607 socklen_t len = sizeof(nd->addr);
610 ret = recvfrom(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to, &len);
612 td_verror(td, errno, "sendto udp link open");
616 if (ntohl(msg.magic) != FIO_LINK_OPEN_CLOSE_MAGIC ||
617 ntohl(msg.cmd) != FIO_LINK_OPEN) {
618 log_err("fio: bad udp open magic %x/%x\n", ntohl(msg.magic),
626 static int fio_netio_udp_send_open(struct thread_data *td, struct fio_file *f)
628 struct netio_data *nd = td->io_ops->data;
629 struct udp_close_msg msg;
630 struct sockaddr *to = (struct sockaddr *) &nd->addr;
633 msg.magic = htonl(FIO_LINK_OPEN_CLOSE_MAGIC);
634 msg.cmd = htonl(FIO_LINK_OPEN);
636 ret = sendto(f->fd, (void *) &msg, sizeof(msg), MSG_WAITALL, to,
639 td_verror(td, errno, "sendto udp link open");
646 static int fio_netio_open_file(struct thread_data *td, struct fio_file *f)
649 struct netio_options *o = td->eo;
652 ret = fio_netio_accept(td, f);
654 ret = fio_netio_connect(td, f);
661 if (o->proto == FIO_TYPE_UDP) {
663 ret = fio_netio_udp_send_open(td, f);
667 state = td->runstate;
668 td_set_runstate(td, TD_SETTING_UP);
669 ret = fio_netio_udp_recv_open(td, f);
670 td_set_runstate(td, state);
675 fio_netio_close_file(td, f);
680 static int fio_netio_setup_connect_inet(struct thread_data *td,
681 const char *host, unsigned short port)
683 struct netio_data *nd = td->io_ops->data;
686 log_err("fio: connect with no host to connect to.\n");
688 log_err("fio: did you forget to set 'listen'?\n");
690 td_verror(td, EINVAL, "no hostname= set");
694 nd->addr.sin_family = AF_INET;
695 nd->addr.sin_port = htons(port);
697 if (inet_aton(host, &nd->addr.sin_addr) != 1) {
698 struct hostent *hent;
700 hent = gethostbyname(host);
702 td_verror(td, errno, "gethostbyname");
706 memcpy(&nd->addr.sin_addr, hent->h_addr, 4);
712 static int fio_netio_setup_connect_unix(struct thread_data *td,
715 struct netio_data *nd = td->io_ops->data;
716 struct sockaddr_un *soun = &nd->addr_un;
718 soun->sun_family = AF_UNIX;
719 strcpy(soun->sun_path, path);
723 static int fio_netio_setup_connect(struct thread_data *td)
725 struct netio_options *o = td->eo;
727 if (o->proto == FIO_TYPE_UDP || o->proto == FIO_TYPE_TCP)
728 return fio_netio_setup_connect_inet(td, td->o.filename,o->port);
730 return fio_netio_setup_connect_unix(td, td->o.filename);
733 static int fio_netio_setup_listen_unix(struct thread_data *td, const char *path)
735 struct netio_data *nd = td->io_ops->data;
736 struct sockaddr_un *addr = &nd->addr_un;
740 fd = socket(AF_UNIX, SOCK_STREAM, 0);
742 log_err("fio: socket: %s\n", strerror(errno));
748 memset(addr, 0, sizeof(*addr));
749 addr->sun_family = AF_UNIX;
750 strcpy(addr->sun_path, path);
753 len = sizeof(addr->sun_family) + strlen(path) + 1;
755 if (bind(fd, (struct sockaddr *) addr, len) < 0) {
756 log_err("fio: bind: %s\n", strerror(errno));
766 static int fio_netio_setup_listen_inet(struct thread_data *td, short port)
768 struct netio_data *nd = td->io_ops->data;
769 struct netio_options *o = td->eo;
772 if (o->proto == FIO_TYPE_TCP)
777 fd = socket(AF_INET, type, 0);
779 td_verror(td, errno, "socket");
784 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (void *) &opt, sizeof(opt)) < 0) {
785 td_verror(td, errno, "setsockopt");
789 if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, (void *) &opt, sizeof(opt)) < 0) {
790 td_verror(td, errno, "setsockopt");
795 nd->addr.sin_family = AF_INET;
796 nd->addr.sin_addr.s_addr = htonl(INADDR_ANY);
797 nd->addr.sin_port = htons(port);
799 if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
800 td_verror(td, errno, "bind");
808 static int fio_netio_setup_listen(struct thread_data *td)
810 struct netio_data *nd = td->io_ops->data;
811 struct netio_options *o = td->eo;
814 if (o->proto == FIO_TYPE_UDP || o->proto == FIO_TYPE_TCP)
815 ret = fio_netio_setup_listen_inet(td, o->port);
817 ret = fio_netio_setup_listen_unix(td, td->o.filename);
821 if (o->proto == FIO_TYPE_UDP)
824 if (listen(nd->listenfd, 10) < 0) {
825 td_verror(td, errno, "listen");
833 static int fio_netio_init(struct thread_data *td)
835 struct netio_options *o = td->eo;
840 WSAStartup(MAKEWORD(2,2), &wsd);
844 log_err("fio: network IO can't be random\n");
848 if (o->proto == FIO_TYPE_UNIX && o->port) {
849 log_err("fio: network IO port not valid with unix socket\n");
851 } else if (o->proto != FIO_TYPE_UNIX && !o->port) {
852 log_err("fio: network IO requires port for tcp or udp\n");
856 if (o->proto != FIO_TYPE_TCP) {
858 log_err("fio: listen only valid for TCP proto IO\n");
862 log_err("fio: datagram network connections must be"
866 if (o->proto == FIO_TYPE_UNIX && !td->o.filename) {
867 log_err("fio: UNIX sockets need host/filename\n");
870 o->listen = td_read(td);
873 if (o->proto != FIO_TYPE_UNIX && o->listen && td->o.filename) {
874 log_err("fio: hostname not valid for inbound network IO\n");
879 ret = fio_netio_setup_listen(td);
881 ret = fio_netio_setup_connect(td);
886 static void fio_netio_cleanup(struct thread_data *td)
888 struct netio_data *nd = td->io_ops->data;
891 if (nd->listenfd != -1)
893 if (nd->pipes[0] != -1)
895 if (nd->pipes[1] != -1)
902 static int fio_netio_setup(struct thread_data *td)
904 struct netio_data *nd;
906 if (!td->files_index) {
907 add_file(td, td->o.filename ?: "net");
908 td->o.nr_files = td->o.nr_files ?: 1;
911 if (!td->io_ops->data) {
912 nd = malloc(sizeof(*nd));;
914 memset(nd, 0, sizeof(*nd));
916 nd->pipes[0] = nd->pipes[1] = -1;
917 td->io_ops->data = nd;
923 static void fio_netio_terminate(struct thread_data *td)
925 kill(td->pid, SIGUSR2);
928 #ifdef CONFIG_LINUX_SPLICE
929 static int fio_netio_setup_splice(struct thread_data *td)
931 struct netio_data *nd;
935 nd = td->io_ops->data;
937 if (pipe(nd->pipes) < 0)
947 static struct ioengine_ops ioengine_splice = {
949 .version = FIO_IOOPS_VERSION,
950 .prep = fio_netio_prep,
951 .queue = fio_netio_queue,
952 .setup = fio_netio_setup_splice,
953 .init = fio_netio_init,
954 .cleanup = fio_netio_cleanup,
955 .open_file = fio_netio_open_file,
956 .close_file = fio_netio_close_file,
957 .terminate = fio_netio_terminate,
959 .option_struct_size = sizeof(struct netio_options),
960 .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
965 static struct ioengine_ops ioengine_rw = {
967 .version = FIO_IOOPS_VERSION,
968 .prep = fio_netio_prep,
969 .queue = fio_netio_queue,
970 .setup = fio_netio_setup,
971 .init = fio_netio_init,
972 .cleanup = fio_netio_cleanup,
973 .open_file = fio_netio_open_file,
974 .close_file = fio_netio_close_file,
975 .terminate = fio_netio_terminate,
977 .option_struct_size = sizeof(struct netio_options),
978 .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR |
982 static int str_hostname_cb(void *data, const char *input)
984 struct netio_options *o = data;
986 if (o->td->o.filename)
987 free(o->td->o.filename);
988 o->td->o.filename = strdup(input);
992 static void fio_init fio_netio_register(void)
994 register_ioengine(&ioengine_rw);
995 #ifdef CONFIG_LINUX_SPLICE
996 register_ioengine(&ioengine_splice);
1000 static void fio_exit fio_netio_unregister(void)
1002 unregister_ioengine(&ioengine_rw);
1003 #ifdef CONFIG_LINUX_SPLICE
1004 unregister_ioengine(&ioengine_splice);