lacp: Sort slaves in appctl output.

[openvswitch] / lib / dpif-linux.c

diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c
index 4b4ac55be78fb6c834437ff3c298a372717643df..1f6c2c028cd641c5653cc0434c5427bcb7c2ab4d 100644 (file)
--- a/lib/dpif-linux.c
+++ b/lib/dpif-linux.c
@@ -28,7 +28,10 @@
 #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>
 
@@ -63,6 +66,7 @@ BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
 
 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. */
@@ -135,8 +139,9 @@ struct dpif_linux {
 
     /* 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. */
@@ -271,6 +276,7 @@ open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
     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);
@@ -291,6 +297,10 @@ destroy_upcall_socks(struct dpif_linux *dpif)
 {
     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;
@@ -1002,13 +1012,31 @@ dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
         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;
@@ -1088,24 +1116,39 @@ static int
 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;
-        int dp_ifindex;
+    if (!dpif->ready_mask) {
+        struct epoll_event events[N_UPCALL_SOCKS];
+        int retval;
+        int i;
+
+        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 (i = 0; i < retval; i++) {
+            dpif->ready_mask |= 1u << events[i].data.u32;
+        }
+    }
 
-        dpif->last_read_upcall = (dpif->last_read_upcall + 1) &
-                                 (N_UPCALL_SOCKS - 1);
-        upcall_sock = dpif->upcall_socks[dpif->last_read_upcall];
+    while (dpif->ready_mask) {
+        int indx = ffs(dpif->ready_mask) - 1;
+        struct nl_sock *upcall_sock = dpif->upcall_socks[indx];
 
+        dpif->ready_mask &= ~(1u << indx);
 
         for (;;) {
             struct ofpbuf *buf;
+            int dp_ifindex;
             int error;
 
             if (++read_tries > 50) {
@@ -1140,15 +1183,12 @@ static void
 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