X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=lib%2Fnetdev-linux.c;h=e86a160c8e17da10d506ab277950e75e0faf1448;hb=7103dec49eb569c3196239da6c178a29c3003e2b;hp=e5e4adb17e5805b92e36a36d62ec92256b9e9493;hpb=149f577a25508779b756515be1f1bdcefa3710fa;p=openvswitch

diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
index e5e4adb1..e86a160c 100644
--- a/lib/netdev-linux.c
+++ b/lib/netdev-linux.c
@@ -50,6 +50,7 @@
 #include "netlink.h"
 #include "ofpbuf.h"
 #include "openflow/openflow.h"
+#include "openvswitch/internal_dev.h"
 #include "openvswitch/gre.h"
 #include "packets.h"
 #include "poll-loop.h"
@@ -58,10 +59,6 @@
 #include "shash.h"
 #include "svec.h"
 
-#ifndef GRE_IOCTL_ONLY
-#include <linux/if_link.h>
-#endif
-
 #define THIS_MODULE VLM_netdev_linux
 #include "vlog.h"
 
@@ -75,7 +72,7 @@
 #endif
 
 static struct rtnetlink_notifier netdev_linux_cache_notifier;
-static struct shash cache_map = SHASH_INITIALIZER(&cache_map);
+static int cache_notifier_refcount;
 
 enum {
     VALID_IFINDEX = 1 << 0,
@@ -84,64 +81,48 @@ enum {
     VALID_IN6 = 1 << 3,
     VALID_MTU = 1 << 4,
     VALID_CARRIER = 1 << 5,
-    VALID_IS_INTERNAL = 1 << 6
+    VALID_IS_PSEUDO = 1 << 6       /* Represents is_internal and is_tap. */
 };
 
 struct tap_state {
     int fd;
 };
 
+struct patch_state {
+    char *peer;
+};
+
 struct netdev_dev_linux {
     struct netdev_dev netdev_dev;
 
     struct shash_node *shash_node;
     unsigned int cache_valid;
 
+    /* The following are figured out "on demand" only.  They are only valid
+     * when the corresponding VALID_* bit in 'cache_valid' is set. */
     int ifindex;
     uint8_t etheraddr[ETH_ADDR_LEN];
     struct in_addr address, netmask;
     struct in6_addr in6;
     int mtu;
     int carrier;
-    bool is_internal;
+    bool is_internal;           /* Is this an openvswitch internal device? */
+    bool is_tap;                /* Is this a tuntap device? */
 
     union {
         struct tap_state tap;
+        struct patch_state patch;
     } state;
 };
 
 struct netdev_linux {
     struct netdev netdev;
-
-    /* File descriptors.  For ordinary network devices, the two fds below are
-     * the same; for tap devices, they differ. */
-    int netdev_fd;              /* Network device. */
-    int tap_fd;                 /* TAP character device, if any, otherwise the
-                                 * network device. */
+    int fd;
 };
 
 /* An AF_INET socket (used for ioctl operations). */
 static int af_inet_sock = -1;
 
-struct gre_config {
-    uint32_t local_ip;
-    uint32_t remote_ip;
-    uint32_t in_key;
-    uint32_t out_key;
-    bool have_in_key;
-    bool have_out_key;
-    bool in_csum;
-    bool out_csum;
-};
-
-static struct {
-    union {
-        struct nl_sock *nl_sock;
-        int ioctl_fd;
-    };
-    bool use_ioctl;
-} gre_descriptors;
-
 struct netdev_linux_notifier {
     struct netdev_notifier notifier;
     struct list node;
@@ -155,7 +136,9 @@ static struct rtnetlink_notifier netdev_linux_poll_notifier;
  * additional log messages. */
 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
 
-static int netdev_linux_do_ethtool(const struct netdev *, struct ethtool_cmd *,
+static int netdev_linux_init(void);
+
+static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
                                    int cmd, const char *cmd_name);
 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
                                  const char *cmd_name);
@@ -174,21 +157,28 @@ static int set_etheraddr(const char *netdev_name, int hwaddr_family,
 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
 
+static bool
+is_netdev_linux_class(const struct netdev_class *netdev_class)
+{
+    return netdev_class->init == netdev_linux_init;
+}
+
 static struct netdev_dev_linux *
 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
 {
-    const char *type = netdev_dev_get_type(netdev_dev);
-    assert(!strcmp(type, "system") || !strcmp(type, "tap")
-            || !strcmp(type, "gre"));
+    const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
+    assert(is_netdev_linux_class(netdev_class));
+
     return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
 }
 
 static struct netdev_linux *
 netdev_linux_cast(const struct netdev *netdev)
 {
-    const char *type = netdev_get_type(netdev);
-    assert(!strcmp(type, "system") || !strcmp(type, "tap")
-            || !strcmp(type, "gre"));
+    struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
+    const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
+    assert(is_netdev_linux_class(netdev_class));
+
     return CONTAINER_OF(netdev, struct netdev_linux, netdev);
 }
 
@@ -220,244 +210,154 @@ netdev_linux_wait(void)
 
 static void
 netdev_linux_cache_cb(const struct rtnetlink_change *change,
-                      void *aux UNUSED)
+                      void *aux OVS_UNUSED)
 {
     struct netdev_dev_linux *dev;
     if (change) {
-        dev = shash_find_data(&cache_map, change->ifname);
-        if (dev) {
-            dev->cache_valid = 0;
+        struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
+        if (base_dev) {
+            const struct netdev_class *netdev_class =
+                                                netdev_dev_get_class(base_dev);
+
+            if (is_netdev_linux_class(netdev_class)) {
+                dev = netdev_dev_linux_cast(base_dev);
+                dev->cache_valid = 0;
+            }
         }
     } else {
+        struct shash device_shash;
         struct shash_node *node;
-        SHASH_FOR_EACH (node, &cache_map) {
+
+        shash_init(&device_shash);
+        netdev_dev_get_devices(&netdev_linux_class, &device_shash);
+        SHASH_FOR_EACH (node, &device_shash) {
             dev = node->data;
             dev->cache_valid = 0;
         }
+        shash_destroy(&device_shash);
     }
 }
 
-/* The arguments are marked as unused to prevent warnings on platforms where
- * the Netlink interface isn't supported. */
 static int
-setup_gre_netlink(const char *name UNUSED, struct gre_config *config UNUSED,
-                  bool create UNUSED)
+if_up(const char *name)
 {
-#ifdef GRE_IOCTL_ONLY
-    return EOPNOTSUPP;
-#else
-    int error;
-    struct ofpbuf request, *reply;
-    unsigned int nl_flags;
-    struct ifinfomsg ifinfomsg;
-    struct nlattr *linkinfo_hdr;
-    struct nlattr *info_data_hdr;
-    uint16_t iflags = 0;
-    uint16_t oflags = 0;
-    uint8_t pmtudisc = 0;
-
-    if (!gre_descriptors.nl_sock) {
-        error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0,
-                               &gre_descriptors.nl_sock);
-        if (error) {
-            VLOG_WARN("couldn't create netlink socket: %s\n", strerror(error));
-            gre_descriptors.nl_sock = NULL;
-            goto error;
-        }
-    }
+    struct ifreq ifr;
 
-    ofpbuf_init(&request, 0);
+    strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
+    ifr.ifr_flags = IFF_UP;
 
-    nl_flags = NLM_F_REQUEST;
-    if (create) {
-        nl_flags |= NLM_F_CREATE|NLM_F_EXCL;
+    if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
+        VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
+                    name, strerror(errno));
+        return errno;
     }
 
-    /* We over-reserve space, because we do some pointer arithmetic
-     * and don't want the buffer address shifting under us. */
-    nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 2048, RTM_NEWLINK,
-                        nl_flags);
-
-    memset(&ifinfomsg, 0, sizeof ifinfomsg);
-    ifinfomsg.ifi_family = AF_UNSPEC;
-    nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
-
-    linkinfo_hdr = ofpbuf_tail(&request);
-    nl_msg_put_unspec(&request, IFLA_LINKINFO, NULL, 0);
-
-    nl_msg_put_unspec(&request, IFLA_INFO_KIND, "gretap", 6);
-
-    info_data_hdr = ofpbuf_tail(&request);
-    nl_msg_put_unspec(&request, IFLA_INFO_DATA, NULL, 0);
+    return 0;
+}
 
-    /* Set flags */
-    if (config->have_in_key) {
-        iflags |= GRE_KEY;
-    }
-    if (config->have_out_key) {
-        oflags |= GRE_KEY;
-    }
+/* A veth may be created using the 'command' "+<name>,<peer>". A veth may 
+ * be destroyed by using the 'command' "-<name>", where <name> can be 
+ * either side of the device.
+ */
+static int
+modify_veth(const char *format, ...)
+{
+    FILE *veth_file;
+    va_list args;
+    int retval;
 
-    if (config->in_csum) {
-        iflags |= GRE_CSUM;
+    veth_file = fopen("/sys/class/net/veth_pairs", "w");
+    if (!veth_file) {
+        VLOG_WARN_RL(&rl, "could not open veth device.  Are you running a "
+                "supported XenServer with the kernel module loaded?");
+        return ENODEV;
     }
-    if (config->out_csum) {
-        oflags |= GRE_CSUM;
-    }
-
-    /* Add options */
-    nl_msg_put_u32(&request, IFLA_GRE_IKEY, config->in_key);
-    nl_msg_put_u32(&request, IFLA_GRE_OKEY, config->out_key);
-    nl_msg_put_u16(&request, IFLA_GRE_IFLAGS, iflags);
-    nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags);
-    nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip);
-    nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip);
-    nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, pmtudisc);
-    nl_msg_put_u8(&request, IFLA_GRE_TTL, 0);
-    nl_msg_put_u8(&request, IFLA_GRE_TOS, 0);
-
-    info_data_hdr->nla_len = (char *)ofpbuf_tail(&request)
-                                - (char *)info_data_hdr;
-    linkinfo_hdr->nla_len = (char *)ofpbuf_tail(&request)
-                                - (char *)linkinfo_hdr;
+    setvbuf(veth_file, NULL, _IONBF, 0);
 
-    nl_msg_put_string(&request, IFLA_IFNAME, name);
+    va_start(args, format);
+    retval = vfprintf(veth_file, format, args);
+    va_end(args);
 
-    error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
-    ofpbuf_uninit(&request);
-    if (error) {
-        VLOG_WARN("couldn't transact netlink socket: %s\n", strerror(error));
-        goto error;
+    fclose(veth_file);
+    if (retval < 0) {
+        VLOG_WARN_RL(&rl, "could not destroy patch: %s", strerror(errno));
+        return errno;
     }
-    ofpbuf_delete(reply);
 
-error:
-    return error;
-#endif
+    return 0;
 }
 
 static int
-setup_gre_ioctl(const char *name, struct gre_config *config, bool create)
+create_patch(const char *name, const char *peer)
 {
-    struct ip_tunnel_parm p;
-    struct ifreq ifr;
-
-    memset(&p, 0, sizeof p);
-
-    strncpy(p.name, name, IFNAMSIZ);
+    int retval;
+    struct netdev_dev *peer_nd;
 
-    p.iph.version = 4;
-    p.iph.ihl = 5;
-    p.iph.protocol = IPPROTO_GRE;
-    p.iph.saddr = config->local_ip;
-    p.iph.daddr = config->remote_ip;
 
-    if (config->have_in_key) {
-        p.i_flags |= GRE_KEY;
-        p.i_key = config->in_key;
-    }
-    if (config->have_out_key) {
-        p.o_flags |= GRE_KEY;
-        p.o_key = config->out_key;
+    /* Only create the veth if the peer didn't already do it. */
+    peer_nd = netdev_dev_from_name(peer);
+    if (peer_nd) {
+        if (!strcmp("patch", netdev_dev_get_type(peer_nd))) {
+            struct netdev_dev_linux *ndl = netdev_dev_linux_cast(peer_nd);
+            if (!strcmp(name, ndl->state.patch.peer)) {
+                return 0;
+            } else {
+                VLOG_WARN_RL(&rl, "peer '%s' already paired with '%s'", 
+                        peer, ndl->state.patch.peer);
+                return EINVAL;
+            }
+        } else {
+            VLOG_WARN_RL(&rl, "peer '%s' exists and is not a patch", peer);
+            return EINVAL;
+        }
     }
 
-    if (config->in_csum) {
-        p.i_flags |= GRE_CSUM;
-    }
-    if (config->out_csum) {
-        p.o_flags |= GRE_CSUM;
+    retval = modify_veth("+%s,%s", name, peer);
+    if (retval) {
+        return retval;
     }
 
-    strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
-    ifr.ifr_ifru.ifru_data = (void *)&p;
-
-    if (!gre_descriptors.ioctl_fd) {
-        gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
-        if (gre_descriptors.ioctl_fd < 0) {
-            VLOG_WARN("couldn't create gre ioctl socket: %s\n", strerror(errno));
-            gre_descriptors.ioctl_fd = 0;
-            return errno;
-        }
+    retval = if_up(name);
+    if (retval) {
+        return retval;
     }
 
-    if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
-              &ifr) < 0) {
-        VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
-        return errno;
+    retval = if_up(peer);
+    if (retval) {
+        return retval;
     }
 
     return 0;
 }
 
 static int
-setup_gre(const char *name, const struct shash *args, bool create)
+setup_patch(const char *name, const struct shash *args, char **peer_)
 {
-    int error;
-    struct in_addr in_addr;
-    struct shash_node *node;
-    struct gre_config config;
-
-    memset(&config, 0, sizeof config);
-    config.in_csum = true;
-    config.out_csum = true;
-
-    SHASH_FOR_EACH (node, args) {
-        if (!strcmp(node->name, "remote_ip")) {
-            if (lookup_ip(node->data, &in_addr)) {
-                VLOG_WARN("bad 'remote_ip' for gre device %s ", name);
-            } else {
-                config.remote_ip = in_addr.s_addr;
-            }
-        } else if (!strcmp(node->name, "local_ip")) {
-            if (lookup_ip(node->data, &in_addr)) {
-                VLOG_WARN("bad 'local_ip' for gre device %s ", name);
-            } else {
-                config.local_ip = in_addr.s_addr;
-            }
-        } else if (!strcmp(node->name, "key")) {
-            config.have_in_key = true;
-            config.have_out_key = true;
-            config.in_key = htonl(atoi(node->data));
-            config.out_key = htonl(atoi(node->data));
-        } else if (!strcmp(node->name, "in_key")) {
-            config.have_in_key = true;
-            config.in_key = htonl(atoi(node->data));
-        } else if (!strcmp(node->name, "out_key")) {
-            config.have_out_key = true;
-            config.out_key = htonl(atoi(node->data));
-        } else if (!strcmp(node->name, "csum")) {
-            if (!strcmp(node->data, "false")) {
-                config.in_csum = false;
-                config.out_csum = false;
-            }
-        } else {
-            VLOG_WARN("unknown gre argument '%s'", node->name);
-        }
-    }
+    const char *peer;
 
-    if (!config.remote_ip) {
-        VLOG_WARN("gre type requires valid 'remote_ip' argument");
-        error = EINVAL;
-        goto error;
+    peer = shash_find_data(args, "peer");
+    if (!peer) {
+        VLOG_WARN("patch type requires valid 'peer' argument");
+        return EINVAL;
     }
 
-    if (!gre_descriptors.use_ioctl) {
-        error = setup_gre_netlink(name, &config, create);
-        if (error == EOPNOTSUPP) {
-            gre_descriptors.use_ioctl = true;
-        }
+    if (shash_count(args) > 1) {
+        VLOG_WARN("patch type takes only a 'peer' argument");
+        return EINVAL;
     }
-    if (gre_descriptors.use_ioctl) {
-        error = setup_gre_ioctl(name, &config, create);
+
+    if (strlen(peer) >= IFNAMSIZ) {
+        VLOG_WARN_RL(&rl, "patch 'peer' arg too long");
+        return EINVAL;
     }
 
-error:
-    return error;
+    *peer_ = xstrdup(peer);
+    return create_patch(name, peer);
 }
 
 /* Creates the netdev device of 'type' with 'name'. */
 static int
-netdev_linux_create_system(const char *name, const char *type UNUSED,
+netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
                     const struct shash *args, struct netdev_dev **netdev_devp)
 {
     struct netdev_dev_linux *netdev_dev;
@@ -467,24 +367,30 @@ netdev_linux_create_system(const char *name, const char *type UNUSED,
         VLOG_WARN("%s: arguments for system devices should be empty", name);
     }
 
-    if (shash_is_empty(&cache_map)) {
+    if (!cache_notifier_refcount) {
         error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
                                             netdev_linux_cache_cb, NULL);
         if (error) {
             return error;
         }
     }
+    cache_notifier_refcount++;
 
     netdev_dev = xzalloc(sizeof *netdev_dev);
-    netdev_dev->shash_node = shash_add(&cache_map, name, &netdev_dev);
-
     netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
+
     *netdev_devp = &netdev_dev->netdev_dev;
     return 0;
 }
 
+/* For most types of netdevs we open the device for each call of
+ * netdev_open().  However, this is not the case with tap devices,
+ * since it is only possible to open the device once.  In this
+ * situation we share a single file descriptor, and consequently
+ * buffers, across all readers.  Therefore once data is read it will
+ * be unavailable to other reads for tap devices. */
 static int
-netdev_linux_create_tap(const char *name, const char *type UNUSED,
+netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
                     const struct shash *args, struct netdev_dev **netdev_devp)
 {
     struct netdev_dev_linux *netdev_dev;
@@ -534,129 +440,48 @@ error:
 }
 
 static int
-if_up(const char *name)
-{
-    struct ifreq ifr;
-
-    strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
-    ifr.ifr_flags = IFF_UP;
-
-    if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
-        VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
-                    name, strerror(errno));
-        return errno;
-    }
-
-    return 0;
-}
-
-static int
-netdev_linux_create_gre(const char *name, const char *type UNUSED,
+netdev_linux_create_patch(const char *name, const char *type OVS_UNUSED,
                     const struct shash *args, struct netdev_dev **netdev_devp)
 {
     struct netdev_dev_linux *netdev_dev;
+    char *peer = NULL;
     int error;
 
-    netdev_dev = xzalloc(sizeof *netdev_dev);
-
-    error = setup_gre(name, args, true);
-    if (error) {
-        goto error;
-    }
-
-    error = if_up(name);
+    error = setup_patch(name, args, &peer);
     if (error) {
-        goto error;
+        free(peer);
+        return error;
     }
 
-    netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_gre_class);
+    netdev_dev = xzalloc(sizeof *netdev_dev);
+    netdev_dev->state.patch.peer = peer;
+    netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_patch_class);
     *netdev_devp = &netdev_dev->netdev_dev;
-    return 0;
-
-error:
-    free(netdev_dev);
-    return error;
-}
-
-static int
-netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
-                             const struct shash *args)
-{
-    const char *name = netdev_dev_get_name(netdev_dev_);
-
-    return setup_gre(name, args, false);
-}
-
-/* The arguments are marked as unused to prevent warnings on platforms where
- * the Netlink interface isn't supported. */
-static int
-destroy_gre_netlink(struct netdev_dev_linux *netdev_dev UNUSED)
-{
-#ifdef GRE_IOCTL_ONLY
-    return EOPNOTSUPP;
-#else
-    const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
-    int error;
-    struct ofpbuf request, *reply;
-    struct ifinfomsg ifinfomsg;
-    int ifindex;
-
-    ofpbuf_init(&request, 0);
-    
-    nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK,
-                        NLM_F_REQUEST);
-
-    memset(&ifinfomsg, 0, sizeof ifinfomsg);
-    ifinfomsg.ifi_family = AF_UNSPEC;
-    nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
-
-    ifindex = do_get_ifindex(name);
-    nl_msg_put_u32(&request, IFLA_LINK, ifindex);
-
-    nl_msg_put_string(&request, IFLA_IFNAME, name);
-
-    error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
-    ofpbuf_uninit(&request);
-    if (error) {
-        VLOG_WARN("couldn't transact netlink socket: %s\n", strerror(error));
-        goto error;
-    }
-    ofpbuf_delete(reply);
 
-error:
     return 0;
-#endif
 }
 
-static int
-destroy_gre_ioctl(struct netdev_dev_linux *netdev_dev)
+static void
+destroy_tap(struct netdev_dev_linux *netdev_dev)
 {
-    const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
-    struct ip_tunnel_parm p;
-    struct ifreq ifr;
-
-    memset(&p, 0, sizeof p);
-    strncpy(p.name, name, IFNAMSIZ);
-
-    strncpy(ifr.ifr_name, name, IFNAMSIZ);
-    ifr.ifr_ifru.ifru_data = (void *)&p;
+    struct tap_state *state = &netdev_dev->state.tap;
 
-    if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) {
-        VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
-        return errno;
+    if (state->fd >= 0) {
+        close(state->fd);
     }
-
-    return 0;
 }
 
 static void
-destroy_tap(struct netdev_dev_linux *netdev_dev)
+destroy_patch(struct netdev_dev_linux *netdev_dev)
 {
-    struct tap_state *state = &netdev_dev->state.tap;
+    const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
+    struct patch_state *state = &netdev_dev->state.patch;
 
-    if (state->fd >= 0) {
-        close(state->fd);
+    /* Only destroy veth if 'peer' doesn't exist as an existing netdev. */
+    if (!netdev_dev_from_name(state->peer)) {
+        modify_veth("-%s", name);
     }
+    free(state->peer);
 }
 
 /* Destroys the netdev device 'netdev_dev_'. */
@@ -667,75 +492,42 @@ netdev_linux_destroy(struct netdev_dev *netdev_dev_)
     const char *type = netdev_dev_get_type(netdev_dev_);
 
     if (!strcmp(type, "system")) {
-        shash_delete(&cache_map, netdev_dev->shash_node);
+        cache_notifier_refcount--;
 
-        if (shash_is_empty(&cache_map)) {
+        if (!cache_notifier_refcount) {
             rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
         }
     } else if (!strcmp(type, "tap")) {
         destroy_tap(netdev_dev);
-    } else if (!strcmp(type, "gre")) {
-        if (gre_descriptors.use_ioctl) {
-            destroy_gre_ioctl(netdev_dev);
-        } else {
-            destroy_gre_netlink(netdev_dev);
-        }
+    } else if (!strcmp(type, "patch")) {
+        destroy_patch(netdev_dev);
     }
 
-    free(netdev_dev_);
+    free(netdev_dev);
 }
 
 static int
-netdev_linux_open(struct netdev_dev *netdev_dev, int ethertype,
+netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
                   struct netdev **netdevp)
 {
+    struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
     struct netdev_linux *netdev;
     enum netdev_flags flags;
     int error;
 
     /* Allocate network device. */
     netdev = xzalloc(sizeof *netdev);
-    netdev_init(&netdev->netdev, netdev_dev);
-    netdev->netdev_fd = -1;
-    netdev->tap_fd = -1;
-
-    if (!strcmp(netdev_dev_get_type(netdev_dev), "tap")) {
-        static const char tap_dev[] = "/dev/net/tun";
-        struct ifreq ifr;
-
-        /* Open tap device. */
-        netdev->tap_fd = open(tap_dev, O_RDWR);
-        if (netdev->tap_fd < 0) {
-            error = errno;
-            VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
-            goto error;
-        }
-
-        /* Create tap device. */
-        ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
-        strncpy(ifr.ifr_name, netdev_dev_get_name(netdev_dev),
-                sizeof ifr.ifr_name);
-        if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
-            VLOG_WARN("%s: creating tap device failed: %s",
-                      netdev_dev_get_name(netdev_dev),
-                      strerror(errno));
-            error = errno;
-            goto error;
-        }
-
-        /* Make non-blocking. */
-        error = set_nonblocking(netdev->tap_fd);
-        if (error) {
-            goto error;
-        }
-    }
+    netdev->fd = -1;
+    netdev_init(&netdev->netdev, netdev_dev_);
 
     error = netdev_get_flags(&netdev->netdev, &flags);
     if (error == ENODEV) {
         goto error;
     }
 
-    if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) {
+    if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
+        netdev->fd = netdev_dev->state.tap.fd;
+    } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
         struct sockaddr_ll sll;
         int protocol;
         int ifindex;
@@ -744,17 +536,14 @@ netdev_linux_open(struct netdev_dev *netdev_dev, int ethertype,
         protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
                     : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
                     : ethertype);
-        netdev->netdev_fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
-        if (netdev->netdev_fd < 0) {
+        netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
+        if (netdev->fd < 0) {
             error = errno;
             goto error;
         }
-        if (netdev->tap_fd < 0) {
-            netdev->tap_fd = netdev->netdev_fd;
-        }
 
         /* Set non-blocking mode. */
-        error = set_nonblocking(netdev->netdev_fd);
+        error = set_nonblocking(netdev->fd);
         if (error) {
             goto error;
         }
@@ -769,10 +558,10 @@ netdev_linux_open(struct netdev_dev *netdev_dev, int ethertype,
         memset(&sll, 0, sizeof sll);
         sll.sll_family = AF_PACKET;
         sll.sll_ifindex = ifindex;
-        if (bind(netdev->netdev_fd,
+        if (bind(netdev->fd,
                  (struct sockaddr *) &sll, sizeof sll) < 0) {
             error = errno;
-            VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev),
+            VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
                      strerror(error));
             goto error;
         }
@@ -781,7 +570,7 @@ netdev_linux_open(struct netdev_dev *netdev_dev, int ethertype,
          * packets of the requested type on all system interfaces.  We do not
          * want to receive that data, but there is no way to avoid it.  So we
          * must now drain out the receive queue. */
-        error = drain_rcvbuf(netdev->netdev_fd);
+        error = drain_rcvbuf(netdev->fd);
         if (error) {
             goto error;
         }
@@ -801,11 +590,8 @@ netdev_linux_close(struct netdev *netdev_)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 
-    if (netdev->netdev_fd >= 0) {
-        close(netdev->netdev_fd);
-    }
-    if (netdev->tap_fd >= 0 && netdev->netdev_fd != netdev->tap_fd) {
-        close(netdev->tap_fd);
+    if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
+        close(netdev->fd);
     }
     free(netdev);
 }
@@ -837,13 +623,13 @@ netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 
-    if (netdev->tap_fd < 0) {
+    if (netdev->fd < 0) {
         /* Device was opened with NETDEV_ETH_TYPE_NONE. */
         return -EAGAIN;
     }
 
     for (;;) {
-        ssize_t retval = read(netdev->tap_fd, data, size);
+        ssize_t retval = read(netdev->fd, data, size);
         if (retval >= 0) {
             return retval;
         } else if (errno != EINTR) {
@@ -862,8 +648,8 @@ static void
 netdev_linux_recv_wait(struct netdev *netdev_)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
-    if (netdev->tap_fd >= 0) {
-        poll_fd_wait(netdev->tap_fd, POLLIN);
+    if (netdev->fd >= 0) {
+        poll_fd_wait(netdev->fd, POLLIN);
     }
 }
 
@@ -872,19 +658,19 @@ static int
 netdev_linux_drain(struct netdev *netdev_)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
-    if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
+    if (netdev->fd < 0) {
         return 0;
-    } else if (netdev->tap_fd != netdev->netdev_fd) {
+    } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
         struct ifreq ifr;
         int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
                                           SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
         if (error) {
             return error;
         }
-        drain_fd(netdev->tap_fd, ifr.ifr_qlen);
+        drain_fd(netdev->fd, ifr.ifr_qlen);
         return 0;
     } else {
-        return drain_rcvbuf(netdev->netdev_fd);
+        return drain_rcvbuf(netdev->fd);
     }
 }
 
@@ -904,12 +690,12 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
 
     /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
      */
-    if (netdev->tap_fd < 0) {
+    if (netdev->fd < 0) {
         return EPIPE;
     }
 
     for (;;) {
-        ssize_t retval = write(netdev->tap_fd, data, size);
+        ssize_t retval = write(netdev->fd, data, size);
         if (retval < 0) {
             /* The Linux AF_PACKET implementation never blocks waiting for room
              * for packets, instead returning ENOBUFS.  Translate this into
@@ -944,10 +730,10 @@ static void
 netdev_linux_send_wait(struct netdev *netdev_)
 {
     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
-    if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
+    if (netdev->fd < 0) {
         /* Nothing to do. */
-    } else if (netdev->tap_fd == netdev->netdev_fd) {
-        poll_fd_wait(netdev->tap_fd, POLLOUT);
+    } else if (strcmp(netdev_get_type(netdev_), "tap")) {
+        poll_fd_wait(netdev->fd, POLLOUT);
     } else {
         /* TAP device always accepts packets.*/
         poll_immediate_wake();
@@ -1118,6 +904,35 @@ check_for_working_netlink_stats(void)
     }
 }
 
+/* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
+static void
+netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
+{
+    if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
+        const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
+        const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
+        
+        netdev_dev->is_tap = !strcmp(type, "tap");
+        netdev_dev->is_internal = false;
+        if (!netdev_dev->is_tap) {
+            struct ethtool_drvinfo drvinfo;
+            int error;
+
+            memset(&drvinfo, 0, sizeof drvinfo);
+            error = netdev_linux_do_ethtool(name,
+                                            (struct ethtool_cmd *)&drvinfo,
+                                            ETHTOOL_GDRVINFO,
+                                            "ETHTOOL_GDRVINFO");
+
+            if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
+                netdev_dev->is_internal = true;
+            }
+        }
+
+        netdev_dev->cache_valid |= VALID_IS_PSEUDO;
+    }
+}
+
 /* Retrieves current device stats for 'netdev'.
  *
  * XXX All of the members of struct netdev_stats are 64 bits wide, but on
@@ -1135,28 +950,7 @@ netdev_linux_get_stats(const struct netdev *netdev_,
 
     COVERAGE_INC(netdev_get_stats);
 
-    if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
-        netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_),
-                                                "tap");
-
-        if (!netdev_dev->is_internal) {
-            struct ethtool_drvinfo drvinfo;
-
-            memset(&drvinfo, 0, sizeof drvinfo);
-            error = netdev_linux_do_ethtool(netdev_,
-                                            (struct ethtool_cmd *)&drvinfo,
-                                            ETHTOOL_GDRVINFO,
-                                            "ETHTOOL_GDRVINFO");
-
-            if (!error) {
-                netdev_dev->is_internal = !strcmp(drvinfo.driver,
-                                                        "openvswitch");
-            }
-        }
-
-        netdev_dev->cache_valid |= VALID_IS_INTERNAL;
-    }
-
+    netdev_linux_update_is_pseudo(netdev_dev);
     if (netdev_dev->is_internal) {
         collect_stats = &raw_stats;
     }
@@ -1179,7 +973,7 @@ netdev_linux_get_stats(const struct netdev *netdev_,
      * 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_dev->is_internal) {
+    if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) {
         stats->rx_packets = raw_stats.tx_packets;
         stats->tx_packets = raw_stats.rx_packets;
         stats->rx_bytes = raw_stats.tx_bytes;
@@ -1206,6 +1000,41 @@ netdev_linux_get_stats(const struct netdev *netdev_,
     return error;
 }
 
+static int
+netdev_linux_set_stats(struct netdev *netdev,
+                       const struct netdev_stats *stats)
+{
+    struct netdev_dev_linux *netdev_dev =
+        netdev_dev_linux_cast(netdev_get_dev(netdev));
+    struct internal_dev_stats dp_dev_stats;
+    struct ifreq ifr;
+
+    /* We must reject this call if 'netdev' is not an Open vSwitch internal
+     * port, because the ioctl that we are about to execute is in the "device
+     * private ioctls" range, which means that executing it on a device that
+     * is not the type we expect could do any random thing.
+     *
+     * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE
+     * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h.  I guess
+     * DaveM is a little behind on that.) */
+    netdev_linux_update_is_pseudo(netdev_dev);
+    if (!netdev_dev->is_internal) {
+        return EOPNOTSUPP;
+    }
+
+    /* This actually only sets the *offset* that the dp_dev applies, but in our
+     * usage for fake bond devices the dp_dev never has any traffic of it own
+     * so it has the same effect. */
+    dp_dev_stats.rx_packets = stats->rx_packets;
+    dp_dev_stats.rx_bytes = stats->rx_bytes;
+    dp_dev_stats.tx_packets = stats->tx_packets;
+    dp_dev_stats.tx_bytes = stats->tx_bytes;
+    ifr.ifr_data = (void *) &dp_dev_stats;
+    return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr,
+                                 INTERNAL_DEV_SET_STATS,
+                                 "INTERNAL_DEV_SET_STATS");
+}
+
 /* Stores the features supported by 'netdev' into each of '*current',
  * '*advertised', '*supported', and '*peer' that are non-null.  Each value is a
  * bitmap of "enum ofp_port_features" bits, in host byte order.  Returns 0 if
@@ -1219,7 +1048,7 @@ netdev_linux_get_features(struct netdev *netdev,
     int error;
 
     memset(&ecmd, 0, sizeof ecmd);
-    error = netdev_linux_do_ethtool(netdev, &ecmd,
+    error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
                                     ETHTOOL_GSET, "ETHTOOL_GSET");
     if (error) {
         return error;
@@ -1340,7 +1169,7 @@ netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
     int error;
 
     memset(&ecmd, 0, sizeof ecmd);
-    error = netdev_linux_do_ethtool(netdev, &ecmd,
+    error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
                                     ETHTOOL_GSET, "ETHTOOL_GSET");
     if (error) {
         return error;
@@ -1383,7 +1212,7 @@ netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
     if (advertise & OFPPF_PAUSE_ASYM) {
         ecmd.advertising |= ADVERTISED_Asym_Pause;
     }
-    return netdev_linux_do_ethtool(netdev, &ecmd,
+    return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
                                    ETHTOOL_SSET, "ETHTOOL_SSET");
 }
 
@@ -1461,8 +1290,8 @@ netdev_linux_set_policing(struct netdev *netdev,
     COVERAGE_INC(netdev_set_policing);
     if (kbits_rate) {
         if (!kbits_burst) {
-            /* Default to 10 kilobits if not specified. */
-            kbits_burst = 10;
+            /* Default to 1000 kilobits if not specified. */
+            kbits_burst = 1000;
         }
 
         /* xxx This should be more careful about only adding if it
@@ -1622,7 +1451,7 @@ do_set_addr(struct netdev *netdev,
 
 /* Adds 'router' as a default IP gateway. */
 static int
-netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router)
+netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
 {
     struct in_addr any = { INADDR_ANY };
     struct rtentry rt;
@@ -1789,7 +1618,7 @@ poll_notify(struct list *list)
 
 static void
 netdev_linux_poll_cb(const struct rtnetlink_change *change,
-                     void *aux UNUSED)
+                     void *aux OVS_UNUSED)
 {
     if (change) {
         struct list *list = shash_find_data(&netdev_linux_notifiers,
@@ -1889,6 +1718,7 @@ const struct netdev_class netdev_linux_class = {
     netdev_linux_get_ifindex,
     netdev_linux_get_carrier,
     netdev_linux_get_stats,
+    netdev_linux_set_stats,
 
     netdev_linux_get_features,
     netdev_linux_set_advertisements,
@@ -1937,6 +1767,7 @@ const struct netdev_class netdev_tap_class = {
     netdev_linux_get_ifindex,
     netdev_linux_get_carrier,
     netdev_linux_get_stats,
+    NULL,                       /* set_stats */
 
     netdev_linux_get_features,
     netdev_linux_set_advertisements,
@@ -1956,16 +1787,16 @@ const struct netdev_class netdev_tap_class = {
     netdev_linux_poll_remove,
 };
 
-const struct netdev_class netdev_gre_class = {
-    "gre",
+const struct netdev_class netdev_patch_class = {
+    "patch",
 
     netdev_linux_init,
     netdev_linux_run,
     netdev_linux_wait,
 
-    netdev_linux_create_gre,
+    netdev_linux_create_patch,
     netdev_linux_destroy,
-    netdev_linux_reconfigure_gre,
+    NULL,                       /* reconfigure */
 
     netdev_linux_open,
     netdev_linux_close,
@@ -1985,6 +1816,7 @@ const struct netdev_class netdev_gre_class = {
     netdev_linux_get_ifindex,
     netdev_linux_get_carrier,
     netdev_linux_get_stats,
+    NULL,                       /* set_stats */
 
     netdev_linux_get_features,
     netdev_linux_set_advertisements,
@@ -2003,6 +1835,7 @@ const struct netdev_class netdev_gre_class = {
     netdev_linux_poll_add,
     netdev_linux_poll_remove,
 };
+
 
 static int
 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
@@ -2243,13 +2076,13 @@ set_etheraddr(const char *netdev_name, int hwaddr_family,
 }
 
 static int
-netdev_linux_do_ethtool(const struct netdev *netdev, struct ethtool_cmd *ecmd,
+netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
                         int cmd, const char *cmd_name)
 {
     struct ifreq ifr;
 
     memset(&ifr, 0, sizeof ifr);
-    strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
+    strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
     ifr.ifr_data = (caddr_t) ecmd;
 
     ecmd->cmd = cmd;
@@ -2259,8 +2092,7 @@ netdev_linux_do_ethtool(const struct netdev *netdev, struct ethtool_cmd *ecmd,
     } else {
         if (errno != EOPNOTSUPP) {
             VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
-                         "failed: %s", cmd_name, netdev_get_name(netdev),
-                         strerror(errno));
+                         "failed: %s", cmd_name, name, strerror(errno));
         } else {
             /* The device doesn't support this operation.  That's pretty
              * common, so there's no point in logging anything. */