/*
- * Transfer data over the net.
+ * net engine
+ *
+ * IO engine that reads/writes to/from sockets.
+ *
*/
#include <stdio.h>
#include <stdlib.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netdb.h>
+#include <sys/poll.h>
#include "../fio.h"
-#include "../os.h"
-struct net_data {
+struct netio_data {
+ int listenfd;
int send_to_net;
- struct io_u *last_io_u;
+ int use_splice;
+ int pipes[2];
+ char host[64];
+ struct sockaddr_in addr;
};
-static int fio_netio_getevents(struct thread_data *td, int fio_unused min,
- int max, struct timespec fio_unused *t)
-{
- assert(max <= 1);
-
- /*
- * we can only have one finished io_u for sync io, since the depth
- * is always 1
- */
- if (list_empty(&td->io_u_busylist))
- return 0;
-
- return 1;
-}
-
-static struct io_u *fio_netio_event(struct thread_data *td, int event)
-{
- struct net_data *nd = td->io_ops->data;
-
- assert(event == 0);
-
- return nd->last_io_u;
-}
-
static int fio_netio_prep(struct thread_data *td, struct io_u *io_u)
{
- struct net_data *nd = td->io_ops->data;
+ struct netio_data *nd = td->io_ops->data;
struct fio_file *f = io_u->file;
/*
*/
if ((nd->send_to_net && io_u->ddir == DDIR_READ) ||
(!nd->send_to_net && io_u->ddir == DDIR_WRITE)) {
- printf("boo!\n");
- td_verror(td, EINVAL);
+ td_verror(td, EINVAL, "bad direction");
return 1;
}
* If offset is different from last end position, it's a seek.
* As network io is purely sequential, we don't allow seeks.
*/
- td_verror(td, EINVAL);
+ td_verror(td, EINVAL, "cannot seek");
return 1;
}
+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;
+}
+
+static int fio_netio_send(struct thread_data *td, struct io_u *io_u)
+{
+ int flags = 0;
+
+ /*
+ * if we are going to write more, set MSG_MORE
+ */
+ if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size)
+ flags = MSG_MORE;
+
+ return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
+}
+
+static int fio_netio_recv(struct io_u *io_u)
+{
+ int flags = MSG_WAITALL;
+
+ return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
+}
+
static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
{
- struct net_data *nd = td->io_ops->data;
- struct fio_file *f = io_u->file;
- int ret, flags = 0;
+ struct netio_data *nd = td->io_ops->data;
+ int ret;
if (io_u->ddir == DDIR_WRITE) {
- /*
- * if we are going to write more, set MSG_MORE
- */
- if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen <
- td->io_size)
- flags = MSG_MORE;
-
- ret = send(f->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
+ if (nd->use_splice)
+ ret = fio_netio_splice_out(td, io_u);
+ else
+ ret = fio_netio_send(td, io_u);
} else if (io_u->ddir == DDIR_READ) {
- flags = MSG_WAITALL;
- ret = recv(f->fd, io_u->xfer_buf, io_u->xfer_buflen, flags);
+ if (nd->use_splice)
+ ret = fio_netio_splice_in(td, io_u);
+ else
+ ret = fio_netio_recv(io_u);
} else
ret = 0; /* must be a SYNC */
if (ret != (int) io_u->xfer_buflen) {
- if (ret > 0) {
+ if (ret >= 0) {
io_u->resid = io_u->xfer_buflen - ret;
io_u->error = 0;
- return ret;
+ return FIO_Q_COMPLETED;
} else
io_u->error = errno;
}
- if (!io_u->error)
- nd->last_io_u = io_u;
+ if (io_u->error)
+ td_verror(td, io_u->error, "xfer");
- return io_u->error;
+ return FIO_Q_COMPLETED;
}
-static int fio_netio_setup_connect(struct thread_data *td, const char *host,
- unsigned short port)
+static int fio_netio_connect(struct thread_data *td, struct fio_file *f)
{
- struct sockaddr_in addr;
- struct fio_file *f;
- int i;
+ struct netio_data *nd = td->io_ops->data;
- memset(&addr, 0, sizeof(addr));
- addr.sin_family = AF_INET;
- addr.sin_port = htons(port);
+ f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
+ if (f->fd < 0) {
+ td_verror(td, errno, "socket");
+ return 1;
+ }
- if (inet_aton(host, &addr.sin_addr) != 1) {
- struct hostent *hent;
+ if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
+ td_verror(td, errno, "connect");
+ return 1;
+ }
- hent = gethostbyname(host);
- if (!hent) {
- td_verror(td, errno);
- return 1;
- }
+ return 0;
+}
- memcpy(&addr.sin_addr, hent->h_addr, 4);
- }
+static int fio_netio_accept(struct thread_data *td, struct fio_file *f)
+{
+ struct netio_data *nd = td->io_ops->data;
+ socklen_t socklen = sizeof(nd->addr);
+ struct pollfd pfd;
+ int ret;
- for_each_file(td, f, i) {
- f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
+ log_info("fio: waiting for connection\n");
+
+ /*
+ * Accept loop. poll for incoming events, accept them. Repeat until we
+ * have all connections.
+ */
+ while (!td->terminate) {
+ pfd.fd = nd->listenfd;
+ pfd.events = POLLIN;
+
+ ret = poll(&pfd, 1, -1);
+ if (ret < 0) {
+ if (errno == EINTR)
+ continue;
+
+ td_verror(td, errno, "poll");
+ break;
+ } else if (!ret)
+ continue;
+
+ /*
+ * should be impossible
+ */
+ if (!(pfd.revents & POLLIN))
+ continue;
+
+ f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen);
if (f->fd < 0) {
- td_verror(td, errno);
+ td_verror(td, errno, "accept");
return 1;
}
+ break;
+ }
- if (connect(f->fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
- td_verror(td, errno);
+ 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, unsigned short port)
+static int fio_netio_setup_listen(struct thread_data *td, short port)
{
- struct sockaddr_in addr;
- socklen_t socklen;
- struct fio_file *f;
- int fd, opt, i;
+ struct netio_data *nd = td->io_ops->data;
+ int fd, opt;
fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
if (fd < 0) {
- td_verror(td, errno);
+ td_verror(td, errno, "socket");
return 1;
}
opt = 1;
if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
- td_verror(td, errno);
+ 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);
+ td_verror(td, errno, "setsockopt");
return 1;
}
#endif
- memset(&addr, 0, sizeof(addr));
- addr.sin_family = AF_INET;
- addr.sin_addr.s_addr = htonl(INADDR_ANY);
- addr.sin_port = htons(port);
+ 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 *) &addr, sizeof(addr)) < 0) {
- td_verror(td, errno);
+ if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) {
+ td_verror(td, errno, "bind");
return 1;
}
if (listen(fd, 1) < 0) {
- td_verror(td, errno);
+ td_verror(td, errno, "listen");
return 1;
}
- fprintf(f_out, "fio: waiting for %u connections\n", td->nr_files);
-
- socklen = sizeof(addr);
- for_each_file(td, f, i) {
- f->fd = accept(fd, (struct sockaddr *) &addr, &socklen);
- if (f->fd < 0) {
- td_verror(td, errno);
- return 1;
- }
- }
-
+ nd->listenfd = fd;
return 0;
}
-static int fio_netio_setup(struct thread_data *td)
+static int fio_netio_init(struct thread_data *td)
{
- char host[64], buf[128];
- struct net_data *nd;
+ struct netio_data *nd = td->io_ops->data;
unsigned short port;
- struct fio_file *f;
+ char host[64], buf[128];
char *sep;
- int ret, i;
+ int ret;
- if (!td->total_file_size) {
- log_err("fio: need size= set\n");
+ if (td_rw(td)) {
+ log_err("fio: network connections must be read OR write\n");
return 1;
}
-
- /*
- * work around for late init call
- */
- if (td->io_ops->init(td))
- return 1;
-
- nd = td->io_ops->data;
-
- if (td->iomix) {
- log_err("fio: network connections must be read OR write\n");
+ if (td_random(td)) {
+ log_err("fio: network IO can't be random\n");
return 1;
}
- strcpy(buf, td->filename);
+ strcpy(buf, td->o.filename);
- sep = strchr(buf, ':');
+ sep = strchr(buf, '/');
if (!sep) {
- log_err("fio: bad network host:port <<%s>>\n", td->filename);
+ log_err("fio: bad network host/port <<%s>>\n", td->o.filename);
return 1;
}
strcpy(host, buf);
port = atoi(sep);
- if (td->ddir == READ) {
+ if (td_read(td)) {
nd->send_to_net = 0;
ret = fio_netio_setup_listen(td, port);
} else {
ret = fio_netio_setup_connect(td, host, port);
}
- if (ret)
- return ret;
+ return ret;
+}
- td->io_size = td->total_file_size;
- td->total_io_size = td->io_size;
+static void fio_netio_cleanup(struct thread_data *td)
+{
+ struct netio_data *nd = td->io_ops->data;
- for_each_file(td, f, i) {
- f->file_size = td->total_file_size / td->nr_files;
- f->real_file_size = f->file_size;
- }
+ 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]);
- return 0;
+ free(nd);
+ td->io_ops->data = NULL;
+ }
}
-static void fio_netio_cleanup(struct thread_data *td)
+static int fio_netio_setup(struct thread_data *td)
{
- if (td->io_ops->data) {
- free(td->io_ops->data);
- td->io_ops->data = NULL;
+ 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;
}
-static int fio_netio_init(struct thread_data *td)
+static int fio_netio_setup_splice(struct thread_data *td)
{
- struct net_data *nd;
+ struct netio_data *nd;
- /*
- * Hack to work-around the ->setup() function calling init on its
- * own, since it needs ->io_ops->data to be set up.
- */
- if (td->io_ops->data)
+ fio_netio_setup(td);
+
+ nd = td->io_ops->data;
+ if (nd) {
+ if (pipe(nd->pipes) < 0)
+ return 1;
+
+ nd->use_splice = 1;
return 0;
+ }
- nd = malloc(sizeof(*nd));
- nd->last_io_u = NULL;
- td->io_ops->data = nd;
- return 0;
+ return 1;
}
-static struct ioengine_ops ioengine = {
+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,
+};
+
+static struct ioengine_ops ioengine_splice = {
+ .name = "netsplice",
+ .version = FIO_IOOPS_VERSION,
.prep = fio_netio_prep,
.queue = fio_netio_queue,
- .getevents = fio_netio_getevents,
- .event = fio_netio_event,
+ .setup = fio_netio_setup_splice,
+ .init = fio_netio_init,
.cleanup = fio_netio_cleanup,
- .setup = fio_netio_setup,
- .flags = FIO_SYNCIO | FIO_NETIO,
+ .open_file = fio_netio_open_file,
+ .close_file = generic_close_file,
+ .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR,
};
static void fio_init fio_netio_register(void)
{
- register_ioengine(&ioengine);
+ register_ioengine(&ioengine_rw);
+ register_ioengine(&ioengine_splice);
}
static void fio_exit fio_netio_unregister(void)
{
- unregister_ioengine(&ioengine);
+ unregister_ioengine(&ioengine_rw);
+ unregister_ioengine(&ioengine_splice);
}
projects / fio.git / blobdiff