#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
+#include <poll.h>
#include <stdlib.h>
+#include <strings.h>
+#include <sys/epoll.h>
#include <sys/stat.h>
#include <unistd.h>
enum { N_UPCALL_SOCKS = 16 };
BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS));
+BUILD_ASSERT_DECL(N_UPCALL_SOCKS <= 32); /* We use a 32-bit word as a mask. */
/* This ethtool flag was introduced in Linux 2.6.24, so it might be
* missing if we have old headers. */
/* Upcall messages. */
struct nl_sock *upcall_socks[N_UPCALL_SOCKS];
- int last_read_upcall;
- unsigned int listen_mask;
+ uint32_t ready_mask; /* 1-bit for each sock with unread messages. */
+ unsigned int listen_mask; /* Mask of DPIF_UC_* bits. */
+ int epoll_fd; /* epoll fd that includes the upcall socks. */
/* Change notification. */
struct sset changed_ports; /* Ports that have changed. */
dpif = xzalloc(sizeof *dpif);
dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
dpif);
+ dpif->epoll_fd = -1;
dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
dp->dp_ifindex, dp->dp_ifindex);
{
int i;
+ if (dpif->epoll_fd >= 0) {
+ close(dpif->epoll_fd);
+ dpif->epoll_fd = -1;
+ }
for (i = 0; i < N_UPCALL_SOCKS; i++) {
nl_sock_destroy(dpif->upcall_socks[i]);
dpif->upcall_socks[i] = NULL;
int i;
int error;
+ dpif->epoll_fd = epoll_create(N_UPCALL_SOCKS);
+ if (dpif->epoll_fd < 0) {
+ return errno;
+ }
+
for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ struct epoll_event event;
+
error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]);
if (error) {
destroy_upcall_socks(dpif);
return error;
}
+
+ event.events = EPOLLIN;
+ event.data.u32 = i;
+ if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD,
+ nl_sock_fd(dpif->upcall_socks[i]), &event) < 0) {
+ error = errno;
+ destroy_upcall_socks(dpif);
+ return error;
+ }
}
+
+ dpif->ready_mask = 0;
}
dpif->listen_mask = listen_mask;
dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int i;
int read_tries = 0;
if (!dpif->listen_mask) {
return EAGAIN;
}
- for (i = 0; i < N_UPCALL_SOCKS; i++) {
- struct nl_sock *upcall_sock;
- dpif->last_read_upcall = (dpif->last_read_upcall + 1) &
- (N_UPCALL_SOCKS - 1);
- upcall_sock = dpif->upcall_socks[dpif->last_read_upcall];
+ if (!dpif->ready_mask) {
+ struct epoll_event events[N_UPCALL_SOCKS];
+ int retval;
+ int i;
- if (nl_sock_woke(upcall_sock)) {
- int dp_ifindex;
+ do {
+ retval = epoll_wait(dpif->epoll_fd, events, N_UPCALL_SOCKS, 0);
+ } while (retval < 0 && errno == EINTR);
+ if (retval < 0) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
+ }
- for (;;) {
- struct ofpbuf *buf;
- int error;
+ for (i = 0; i < retval; i++) {
+ dpif->ready_mask |= 1u << events[i].data.u32;
+ }
+ }
- if (++read_tries > 50) {
- return EAGAIN;
- }
+ while (dpif->ready_mask) {
+ int indx = ffs(dpif->ready_mask) - 1;
+ struct nl_sock *upcall_sock = dpif->upcall_socks[indx];
- error = nl_sock_recv(upcall_sock, &buf, false);
- if (error == EAGAIN) {
- break;
- } else if (error) {
- return error;
- }
+ dpif->ready_mask &= ~(1u << indx);
- error = parse_odp_packet(buf, upcall, &dp_ifindex);
- if (!error
- && dp_ifindex == dpif->dp_ifindex
- && dpif->listen_mask & (1u << upcall->type)) {
- return 0;
- }
+ for (;;) {
+ struct ofpbuf *buf;
+ int dp_ifindex;
+ int error;
- ofpbuf_delete(buf);
- if (error) {
- return error;
- }
+ if (++read_tries > 50) {
+ return EAGAIN;
+ }
+
+ error = nl_sock_recv(upcall_sock, &buf, false);
+ if (error == EAGAIN) {
+ break;
+ } else if (error) {
+ return error;
+ }
+
+ error = parse_odp_packet(buf, upcall, &dp_ifindex);
+ if (!error
+ && dp_ifindex == dpif->dp_ifindex
+ && dpif->listen_mask & (1u << upcall->type)) {
+ return 0;
+ }
+
+ ofpbuf_delete(buf);
+ if (error) {
+ return error;
}
}
}
dpif_linux_recv_wait(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int i;
if (!dpif->listen_mask) {
return;
}
- for (i = 0; i < N_UPCALL_SOCKS; i++) {
- nl_sock_wait(dpif->upcall_socks[i], POLLIN);
- }
+ poll_fd_wait(dpif->epoll_fd, POLLIN);
}
static void