X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=c33405fd33cf9b9f53db8ff0f5e4188c51703fea;hb=8b681e6fdffe4ebd68dc259544abc87d4cccf0cb;hp=fbca67601e4cb725ba245148b89146834229ccf9;hpb=46097491e4ab6b957bb11690d86df24af19b5655;p=openvswitch

diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
index fbca6760..c33405fd 100644
--- a/lib/netdev-linux.c
+++ b/lib/netdev-linux.c
@@ -15,9 +15,6 @@
  */
 
 #include <config.h>
-
-#include "netdev-linux.h"
-
 #include <assert.h>
 #include <errno.h>
 #include <fcntl.h>
@@ -53,6 +50,7 @@
 #include "openflow/openflow.h"
 #include "packets.h"
 #include "poll-loop.h"
+#include "rtnetlink.h"
 #include "socket-util.h"
 #include "shash.h"
 #include "svec.h"
@@ -87,7 +85,8 @@ enum {
     VALID_IN4 = 1 << 2,
     VALID_IN6 = 1 << 3,
     VALID_MTU = 1 << 4,
-    VALID_CARRIER = 1 << 5
+    VALID_CARRIER = 1 << 5,
+    VALID_IS_INTERNAL = 1 << 6
 };
 
 /* Cached network device information. */
@@ -98,26 +97,16 @@ struct netdev_linux_cache {
 
     int ifindex;
     uint8_t etheraddr[ETH_ADDR_LEN];
-    struct in_addr in4;
+    struct in_addr address, netmask;
     struct in6_addr in6;
     int mtu;
     int carrier;
+    bool is_internal;
 };
 
 static struct shash cache_map = SHASH_INITIALIZER(&cache_map);
 static struct rtnetlink_notifier netdev_linux_cache_notifier;
 
-/* Policy for RTNLGRP_LINK messages.
- *
- * There are *many* more fields in these messages, but currently we only care
- * about interface names. */
-static const struct nl_policy rtnlgrp_link_policy[] = {
-    [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
-    [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
-    [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
-                     .min_len = sizeof(struct rtnl_link_stats) },
-};
-
 /* An AF_INET socket (used for ioctl operations). */
 static int af_inet_sock = -1;
 
@@ -138,6 +127,8 @@ static int netdev_linux_do_ethtool(struct netdev *, struct ethtool_cmd *,
                                    int cmd, const char *cmd_name);
 static int netdev_linux_do_ioctl(const struct netdev *, struct ifreq *,
                                  int cmd, const char *cmd_name);
+static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
+                                 int cmd, const char *cmd_name);
 static int get_flags(const struct netdev *, int *flagsp);
 static int set_flags(struct netdev *, int flags);
 static int do_get_ifindex(const char *netdev_name);
@@ -212,7 +203,7 @@ netdev_linux_open(const char *name, char *suffix, int ethertype,
     int error;
 
     /* Allocate network device. */
-    netdev = xcalloc(1, sizeof *netdev);
+    netdev = xzalloc(sizeof *netdev);
     netdev_init(&netdev->netdev, suffix, &netdev_linux_class);
     netdev->netdev_fd = -1;
     netdev->tap_fd = -1;
@@ -248,9 +239,11 @@ netdev_linux_open(const char *name, char *suffix, int ethertype,
 
         /* Create tap device. */
         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
-        error = netdev_linux_do_ioctl(&netdev->netdev, &ifr,
-                                      TUNSETIFF, "TUNSETIFF");
-        if (error) {
+        strncpy(ifr.ifr_name, suffix, sizeof ifr.ifr_name);
+        if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
+            VLOG_WARN("%s: creating tap device failed: %s", suffix,
+                      strerror(errno));
+            error = errno;
             goto error;
         }
 
@@ -377,7 +370,7 @@ netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
 
     if (netdev->tap_fd < 0) {
         /* Device was opened with NETDEV_ETH_TYPE_NONE. */
-        return EAGAIN;
+        return -EAGAIN;
     }
 
     for (;;) {
@@ -389,7 +382,7 @@ netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
                 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
                              strerror(errno), netdev_get_name(netdev_));
             }
-            return errno;
+            return -errno;
         }
     }
 }
@@ -499,9 +492,17 @@ netdev_linux_set_etheraddr(struct netdev *netdev_,
                            const uint8_t mac[ETH_ADDR_LEN])
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
-    int error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
-    if (!error) {
-        memcpy(netdev->cache->etheraddr, mac, ETH_ADDR_LEN);
+    int error;
+
+    if (!(netdev->cache->valid & VALID_ETHERADDR)
+        || !eth_addr_equals(netdev->cache->etheraddr, mac)) {
+        error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
+        if (!error) {
+            netdev->cache->valid |= VALID_ETHERADDR;
+            memcpy(netdev->cache->etheraddr, mac, ETH_ADDR_LEN);
+        }
+    } else {
+        error = 0;
     }
     return error;
 }
@@ -636,25 +637,83 @@ check_for_working_netlink_stats(void)
  * XXX All of the members of struct netdev_stats are 64 bits wide, but on
  * 32-bit architectures the Linux network stats are only 32 bits. */
 static int
-netdev_linux_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
+netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats)
 {
+    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
     static int use_netlink_stats = -1;
     int error;
+    struct netdev_stats raw_stats;
+    struct netdev_stats *collect_stats = stats;
 
     COVERAGE_INC(netdev_get_stats);
+
+    if (!(netdev->cache->valid & VALID_IS_INTERNAL)) {
+        netdev->cache->is_internal = (netdev->tap_fd != -1);
+
+        if (!netdev->cache->is_internal) {
+            struct ethtool_drvinfo drvinfo;
+
+            memset(&drvinfo, 0, sizeof drvinfo);
+            error = netdev_linux_do_ethtool(&netdev->netdev,
+                                            (struct ethtool_cmd *)&drvinfo,
+                                            ETHTOOL_GDRVINFO,
+                                            "ETHTOOL_GDRVINFO");
+
+            if (!error) {
+                netdev->cache->is_internal = !strcmp(drvinfo.driver,
+                                                     "openvswitch");
+            }
+        }
+
+        netdev->cache->valid |= VALID_IS_INTERNAL;
+    }
+
+    if (netdev->cache->is_internal) {
+        collect_stats = &raw_stats;
+    }
+
     if (use_netlink_stats < 0) {
         use_netlink_stats = check_for_working_netlink_stats();
     }
     if (use_netlink_stats) {
         int ifindex;
 
-        error = get_ifindex(netdev, &ifindex);
+        error = get_ifindex(&netdev->netdev, &ifindex);
         if (!error) {
-            error = get_stats_via_netlink(ifindex, stats);
+            error = get_stats_via_netlink(ifindex, collect_stats);
         }
     } else {
-        error = get_stats_via_proc(netdev->name, stats);
+        error = get_stats_via_proc(netdev->netdev.name, collect_stats);
+    }
+
+    /* If this port is an internal port then the transmit and receive stats
+     * will appear to be swapped relative to the other ports since we are the
+     * one sending the data, not a remote computer.  For consistency, we swap
+     * them back here. */
+    if (netdev->cache->is_internal) {
+        stats->rx_packets = raw_stats.tx_packets;
+        stats->tx_packets = raw_stats.rx_packets;
+        stats->rx_bytes = raw_stats.tx_bytes;
+        stats->tx_bytes = raw_stats.rx_bytes;
+        stats->rx_errors = raw_stats.tx_errors;
+        stats->tx_errors = raw_stats.rx_errors;
+        stats->rx_dropped = raw_stats.tx_dropped;
+        stats->tx_dropped = raw_stats.rx_dropped;
+        stats->multicast = raw_stats.multicast;
+        stats->collisions = raw_stats.collisions;
+        stats->rx_length_errors = 0;
+        stats->rx_over_errors = 0;
+        stats->rx_crc_errors = 0;
+        stats->rx_frame_errors = 0;
+        stats->rx_fifo_errors = 0;
+        stats->rx_missed_errors = 0;
+        stats->tx_aborted_errors = 0;
+        stats->tx_carrier_errors = 0;
+        stats->tx_fifo_errors = 0;
+        stats->tx_heartbeat_errors = 0;
+        stats->tx_window_errors = 0;
     }
+
     return error;
 }
 
@@ -948,49 +1007,48 @@ netdev_linux_set_policing(struct netdev *netdev,
     return 0;
 }
 
-/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if
- * 'in4' is non-null) and returns true.  Otherwise, returns false. */
 static int
-netdev_linux_get_in4(const struct netdev *netdev_, struct in_addr *in4)
+netdev_linux_get_in4(const struct netdev *netdev_,
+                     struct in_addr *address, struct in_addr *netmask)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
     if (!(netdev->cache->valid & VALID_IN4)) {
-        const struct sockaddr_in *sin;
-        struct ifreq ifr;
         int error;
 
-        ifr.ifr_addr.sa_family = AF_INET;
-        error = netdev_linux_do_ioctl(netdev_, &ifr,
+        error = netdev_linux_get_ipv4(netdev_, &netdev->cache->address,
                                       SIOCGIFADDR, "SIOCGIFADDR");
         if (error) {
             return error;
         }
 
-        sin = (struct sockaddr_in *) &ifr.ifr_addr;
-        netdev->cache->in4 = sin->sin_addr;
+        error = netdev_linux_get_ipv4(netdev_, &netdev->cache->netmask,
+                                      SIOCGIFNETMASK, "SIOCGIFNETMASK");
+        if (error) {
+            return error;
+        }
+
         netdev->cache->valid |= VALID_IN4;
     }
-    *in4 = netdev->cache->in4;
-    return in4->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
+    *address = netdev->cache->address;
+    *netmask = netdev->cache->netmask;
+    return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
 }
 
-/* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask.  If
- * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared.  Returns a
- * positive errno value. */
 static int
-netdev_linux_set_in4(struct netdev *netdev_, struct in_addr addr,
-                     struct in_addr mask)
+netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
+                     struct in_addr netmask)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
     int error;
 
-    error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", addr);
+    error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
     if (!error) {
         netdev->cache->valid |= VALID_IN4;
-        netdev->cache->in4 = addr;
-        if (addr.s_addr != INADDR_ANY) {
+        netdev->cache->address = address;
+        netdev->cache->netmask = netmask;
+        if (address.s_addr != INADDR_ANY) {
             error = do_set_addr(netdev_, SIOCSIFNETMASK,
-                                "SIOCSIFNETMASK", mask);
+                                "SIOCSIFNETMASK", netmask);
         }
     }
     return error;
@@ -1089,6 +1147,67 @@ netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router)
     return error;
 }
 
+static int
+netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
+                          char **netdev_name)
+{
+    static const char fn[] = "/proc/net/route";
+    FILE *stream;
+    char line[256];
+    int ln;
+
+    *netdev_name = NULL;
+    stream = fopen(fn, "r");
+    if (stream == NULL) {
+        VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
+        return errno;
+    }
+
+    ln = 0;
+    while (fgets(line, sizeof line, stream)) {
+        if (++ln >= 2) {
+            char iface[17];
+            uint32_t dest, gateway, mask;
+            int refcnt, metric, mtu;
+            unsigned int flags, use, window, irtt;
+
+            if (sscanf(line,
+                       "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
+                       " %d %u %u\n",
+                       iface, &dest, &gateway, &flags, &refcnt,
+                       &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
+
+                VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", 
+                        fn, ln, line);
+                continue;
+            }
+            if (!(flags & RTF_UP)) {
+                /* Skip routes that aren't up. */
+                continue;
+            }
+
+            /* The output of 'dest', 'mask', and 'gateway' were given in
+             * network byte order, so we don't need need any endian 
+             * conversions here. */
+            if ((dest & mask) == (host->s_addr & mask)) {
+                if (!gateway) {
+                    /* The host is directly reachable. */
+                    next_hop->s_addr = 0;
+                } else {
+                    /* To reach the host, we must go through a gateway. */
+                    next_hop->s_addr = gateway;
+                }
+                *netdev_name = xstrdup(iface);
+                fclose(stream);
+                return 0;
+            }
+        }
+    }
+
+    fclose(stream);
+    return ENXIO;
+}
+
 /* Looks up the ARP table entry for 'ip' on 'netdev'.  If one exists and can be
  * successfully retrieved, it stores the corresponding MAC address in 'mac' and
  * returns 0.  Otherwise, it returns a positive errno value; in particular,
@@ -1282,6 +1401,7 @@ const struct netdev_class netdev_linux_class = {
     netdev_linux_set_in4,
     netdev_linux_get_in6,
     netdev_linux_add_router,
+    netdev_linux_get_next_hop,
     netdev_linux_arp_lookup,
 
     netdev_linux_update_flags,
@@ -1325,6 +1445,7 @@ const struct netdev_class netdev_tap_class = {
     netdev_linux_set_in4,
     netdev_linux_get_in6,
     netdev_linux_add_router,
+    netdev_linux_get_next_hop,
     netdev_linux_arp_lookup,
 
     netdev_linux_update_flags,
@@ -1336,6 +1457,17 @@ const struct netdev_class netdev_tap_class = {
 static int
 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
 {
+    /* Policy for RTNLGRP_LINK messages.
+     *
+     * There are *many* more fields in these messages, but currently we only
+     * care about these fields. */
+    static const struct nl_policy rtnlgrp_link_policy[] = {
+        [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
+        [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
+                         .min_len = sizeof(struct rtnl_link_stats) },
+    };
+
+
     static struct nl_sock *rtnl_sock;
     struct ofpbuf request;
     struct ofpbuf *reply;
@@ -1374,6 +1506,7 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
 
     if (!attrs[IFLA_STATS]) {
         VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
+        ofpbuf_delete(reply);
         return EPROTO;
     }
 
@@ -1400,6 +1533,8 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
     stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
     stats->tx_window_errors = rtnl_stats->tx_window_errors;
 
+    ofpbuf_delete(reply);
+
     return 0;
 }
 
@@ -1593,131 +1728,19 @@ netdev_linux_do_ioctl(const struct netdev *netdev, struct ifreq *ifr,
     }
     return 0;
 }
-
-/* rtnetlink socket. */
-static struct nl_sock *notify_sock;
-
-/* All registered notifiers. */
-static struct list all_notifiers = LIST_INITIALIZER(&all_notifiers);
-
-static void rtnetlink_report_change(const struct nlmsghdr *,
-                                    const struct ifinfomsg *,
-                                    struct nlattr *attrs[]);
-static void rtnetlink_report_notify_error(void);
-
-int
-rtnetlink_notifier_register(struct rtnetlink_notifier *notifier,
-                            rtnetlink_notify_func *cb, void *aux)
-{
-    if (!notify_sock) {
-        int error = nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0,
-                                   &notify_sock);
-        if (error) {
-            VLOG_WARN("could not create rtnetlink socket: %s",
-                      strerror(error));
-            return error;
-        }
-    } else {
-        /* Catch up on notification work so that the new notifier won't
-         * receive any stale notifications. */
-        rtnetlink_notifier_run();
-    }
-
-    list_push_back(&all_notifiers, &notifier->node);
-    notifier->cb = cb;
-    notifier->aux = aux;
-    return 0;
-}
-
-void
-rtnetlink_notifier_unregister(struct rtnetlink_notifier *notifier)
-{
-    list_remove(&notifier->node);
-    if (list_is_empty(&all_notifiers)) {
-        nl_sock_destroy(notify_sock);
-        notify_sock = NULL;
-    }
-}
-
-void
-rtnetlink_notifier_run(void)
-{
-    static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
-
-    if (!notify_sock) {
-        return;
-    }
-
-    for (;;) {
-        struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
-        struct ofpbuf *buf;
-        int error;
 
-        error = nl_sock_recv(notify_sock, &buf, false);
-        if (!error) {
-            if (nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
-                                rtnlgrp_link_policy,
-                                attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
-                struct ifinfomsg *ifinfo;
-
-                ifinfo = (void *) ((char *) buf->data + NLMSG_HDRLEN);
-                rtnetlink_report_change(buf->data, ifinfo, attrs);
-            } else {
-                VLOG_WARN_RL(&rl, "received bad rtnl message");
-                rtnetlink_report_notify_error();
-            }
-            ofpbuf_delete(buf);
-        } else if (error == EAGAIN) {
-            return;
-        } else {
-            if (error == ENOBUFS) {
-                VLOG_WARN_RL(&rl, "rtnetlink receive buffer overflowed");
-            } else {
-                VLOG_WARN_RL(&rl, "error reading rtnetlink socket: %s",
-                             strerror(error));
-            }
-            rtnetlink_report_notify_error();
-        }
-    }
-}
-
-void
-rtnetlink_notifier_wait(void)
-{
-    if (notify_sock) {
-        nl_sock_wait(notify_sock, POLLIN);
-    }
-}
-
-static void
-rtnetlink_report_change(const struct nlmsghdr *nlmsg,
-                           const struct ifinfomsg *ifinfo,
-                           struct nlattr *attrs[])
-{
-    struct rtnetlink_notifier *notifier;
-    struct rtnetlink_change change;
-
-    COVERAGE_INC(rtnetlink_changed);
-
-    change.nlmsg_type = nlmsg->nlmsg_type;
-    change.ifi_index = ifinfo->ifi_index;
-    change.ifname = nl_attr_get_string(attrs[IFLA_IFNAME]);
-    change.master_ifindex = (attrs[IFLA_MASTER]
-                             ? nl_attr_get_u32(attrs[IFLA_MASTER]) : 0);
-
-    LIST_FOR_EACH (notifier, struct rtnetlink_notifier, node,
-                   &all_notifiers) {
-        notifier->cb(&change, notifier->aux);
-    }
-}
-
-static void
-rtnetlink_report_notify_error(void)
+static int
+netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
+                      int cmd, const char *cmd_name)
 {
-    struct rtnetlink_notifier *notifier;
+    struct ifreq ifr;
+    int error;
 
-    LIST_FOR_EACH (notifier, struct rtnetlink_notifier, node,
-                   &all_notifiers) {
-        notifier->cb(NULL, notifier->aux);
+    ifr.ifr_addr.sa_family = AF_INET;
+    error = netdev_linux_do_ioctl(netdev, &ifr, cmd, cmd_name);
+    if (!error) {
+        const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
+        *ip = sin->sin_addr;
     }
+    return error;
 }