X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev.c;h=e4e7ab132eea510f1dce5a48ca16ada912c2ad7a;hb=e0edde6fee279cdbbf3c179f5f50adaf0c7c7f1e;hp=f1eb29b521fa83022e706c34e27b69cef2c48be0;hpb=0b3f40f37183df18eb70255a9cc96d57d96b9a3e;p=sliver-openvswitch.git

diff --git a/lib/netdev.c b/lib/netdev.c
index f1eb29b52..e4e7ab132 100644
--- a/lib/netdev.c
+++ b/lib/netdev.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2009 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -28,26 +28,30 @@
 #include "coverage.h"
 #include "dynamic-string.h"
 #include "fatal-signal.h"
+#include "hash.h"
 #include "list.h"
 #include "netdev-provider.h"
+#include "netdev-vport.h"
 #include "ofpbuf.h"
 #include "openflow/openflow.h"
 #include "packets.h"
 #include "poll-loop.h"
 #include "shash.h"
+#include "sset.h"
 #include "svec.h"
-
-#define THIS_MODULE VLM_netdev
 #include "vlog.h"
 
-static const struct netdev_class *netdev_classes[] = {
-    &netdev_linux_class,
-    &netdev_tap_class,
-};
-static int n_netdev_classes = ARRAY_SIZE(netdev_classes);
+VLOG_DEFINE_THIS_MODULE(netdev);
+
+COVERAGE_DEFINE(netdev_received);
+COVERAGE_DEFINE(netdev_sent);
+COVERAGE_DEFINE(netdev_add_router);
+COVERAGE_DEFINE(netdev_get_stats);
+
+static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
 
 /* All created network devices. */
-static struct shash netdev_obj_shash = SHASH_INITIALIZER(&netdev_obj_shash);
+static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
 
 /* All open network devices. */
 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
@@ -56,45 +60,27 @@ static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
  * additional log messages. */
 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
 
-static void restore_all_flags(void *aux);
+static void close_all_netdevs(void *aux OVS_UNUSED);
 static int restore_flags(struct netdev *netdev);
+void update_device_args(struct netdev_dev *, const struct shash *args);
 
-/* Attempts to initialize the netdev module.  Returns 0 if successful,
- * otherwise a positive errno value.
- *
- * Calling this function is optional.  If not called explicitly, it will
- * automatically be called upon the first attempt to open or create a 
- * network device. */
-int
+static void
 netdev_initialize(void)
 {
-    static int status = -1;
-    if (status < 0) {
-        int i, j;
-
-        fatal_signal_add_hook(restore_all_flags, NULL, true);
-
-        status = 0;
-        for (i = j = 0; i < n_netdev_classes; i++) {
-            const struct netdev_class *class = netdev_classes[i];
-            if (class->init) {
-                int retval = class->init();
-                if (!retval) {
-                    netdev_classes[j++] = class;
-                } else {
-                    VLOG_ERR("failed to initialize %s network device "
-                             "class: %s", class->type, strerror(retval));
-                    if (!status) {
-                        status = retval;
-                    }
-                }
-            } else {
-                netdev_classes[j++] = class;
-            }
-        }
-        n_netdev_classes = j;
+    static bool inited;
+
+    if (!inited) {
+        inited = true;
+
+        fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
+
+#ifdef HAVE_NETLINK
+        netdev_register_provider(&netdev_linux_class);
+        netdev_register_provider(&netdev_internal_class);
+        netdev_register_provider(&netdev_tap_class);
+        netdev_vport_register();
+#endif
     }
-    return status;
 }
 
 /* Performs periodic work needed by all the various kinds of netdevs.
@@ -104,11 +90,11 @@ netdev_initialize(void)
 void
 netdev_run(void)
 {
-    int i;
-    for (i = 0; i < n_netdev_classes; i++) {
-        const struct netdev_class *class = netdev_classes[i];
-        if (class->run) {
-            class->run();
+    struct shash_node *node;
+    SHASH_FOR_EACH(node, &netdev_classes) {
+        const struct netdev_class *netdev_class = node->data;
+        if (netdev_class->run) {
+            netdev_class->run();
         }
     }
 }
@@ -120,146 +106,182 @@ netdev_run(void)
 void
 netdev_wait(void)
 {
-    int i;
-    for (i = 0; i < n_netdev_classes; i++) {
-        const struct netdev_class *class = netdev_classes[i];
-        if (class->wait) {
-            class->wait();
+    struct shash_node *node;
+    SHASH_FOR_EACH(node, &netdev_classes) {
+        const struct netdev_class *netdev_class = node->data;
+        if (netdev_class->wait) {
+            netdev_class->wait();
         }
     }
 }
 
-/* Attempts to create a network device object of 'type' with 'name'.  'type' 
- * corresponds to the 'type' field used in the netdev_class * structure.  
- * Arguments for creation are provided in 'args', which may be empty or NULL 
- * if none are needed. */
+/* Initializes and registers a new netdev provider.  After successful
+ * registration, new netdevs of that type can be opened using netdev_open(). */
 int
-netdev_create(const char *name, const char *type, const struct shash *args)
+netdev_register_provider(const struct netdev_class *new_class)
 {
-    struct shash empty_args = SHASH_INITIALIZER(&empty_args);
-    int i;
-
-    netdev_initialize();
-
-    if (!args) {
-        args = &empty_args;
-    }
-
-    if (shash_find(&netdev_obj_shash, name)) {
-        VLOG_WARN("attempted to create a netdev object with bound name: %s",
-                name);
+    if (shash_find(&netdev_classes, new_class->type)) {
+        VLOG_WARN("attempted to register duplicate netdev provider: %s",
+                   new_class->type);
         return EEXIST;
     }
 
-    for (i = 0; i < n_netdev_classes; i++) {
-        const struct netdev_class *class = netdev_classes[i];
-        if (!strcmp(type, class->type)) {
-            return class->create(name, type, args, true);
+    if (new_class->init) {
+        int error = new_class->init();
+        if (error) {
+            VLOG_ERR("failed to initialize %s network device class: %s",
+                     new_class->type, strerror(error));
+            return error;
         }
     }
 
-    VLOG_WARN("could not create netdev object of unknown type: %s", type);
+    shash_add(&netdev_classes, new_class->type, new_class);
 
-    return EINVAL;
+    return 0;
 }
 
-/* Destroys netdev object 'name'.  Netdev objects maintain a reference count
- * which is incremented on netdev_open() and decremented on netdev_close().  
- * If 'name' has a non-zero reference count, it will not destroy the object 
- * and return EBUSY. */
+/* Unregisters a netdev provider.  'type' must have been previously
+ * registered and not currently be in use by any netdevs.  After unregistration
+ * new netdevs of that type cannot be opened using netdev_open(). */
 int
-netdev_destroy(const char *name)
+netdev_unregister_provider(const char *type)
 {
-    struct shash_node *node;
-    struct netdev_obj *netdev_obj;
+    struct shash_node *del_node, *netdev_dev_node;
 
-    node = shash_find(&netdev_obj_shash, name);
-    if (!node) {
-        return ENODEV;
+    del_node = shash_find(&netdev_classes, type);
+    if (!del_node) {
+        VLOG_WARN("attempted to unregister a netdev provider that is not "
+                  "registered: %s", type);
+        return EAFNOSUPPORT;
     }
 
-    netdev_obj = node->data;
-    if (netdev_obj->ref_cnt != 0) {
-        VLOG_WARN("attempt to destroy open netdev object (%d): %s", 
-                netdev_obj->ref_cnt, name);
-        return EBUSY;
+    SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
+        struct netdev_dev *netdev_dev = netdev_dev_node->data;
+        if (!strcmp(netdev_dev->netdev_class->type, type)) {
+            VLOG_WARN("attempted to unregister in use netdev provider: %s",
+                      type);
+            return EBUSY;
+        }
     }
 
-    shash_delete(&netdev_obj_shash, node);
-    netdev_obj->netdev_class->destroy(netdev_obj);
+    shash_delete(&netdev_classes, del_node);
 
     return 0;
 }
 
-/* Reconfigures the device object 'name' with 'args'.  'args' may be empty 
- * or NULL if none are needed. */
+const struct netdev_class *
+netdev_lookup_provider(const char *type)
+{
+    netdev_initialize();
+    return shash_find_data(&netdev_classes, type && type[0] ? type : "system");
+}
+
+/* Clears 'types' and enumerates the types of all currently registered netdev
+ * providers into it.  The caller must first initialize the sset. */
+void
+netdev_enumerate_types(struct sset *types)
+{
+    struct shash_node *node;
+
+    netdev_initialize();
+    sset_clear(types);
+
+    SHASH_FOR_EACH(node, &netdev_classes) {
+        const struct netdev_class *netdev_class = node->data;
+        sset_add(types, netdev_class->type);
+    }
+}
+
+/* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
+ * (e.g. "system") and returns zero if successful, otherwise a positive errno
+ * value.  On success, sets '*netdevp' to the new network device, otherwise to
+ * null.
+ *
+ * Some network devices may need to be configured (with netdev_set_config())
+ * before they can be used. */
 int
-netdev_reconfigure(const char *name, const struct shash *args)
+netdev_open(const char *name, const char *type, struct netdev **netdevp)
 {
-    struct shash empty_args = SHASH_INITIALIZER(&empty_args);
-    struct netdev_obj *netdev_obj;
+    struct netdev_dev *netdev_dev;
+    int error;
+
+    *netdevp = NULL;
+    netdev_initialize();
+
+    netdev_dev = shash_find_data(&netdev_dev_shash, name);
+
+    if (!netdev_dev) {
+        const struct netdev_class *class;
+
+        class = netdev_lookup_provider(type);
+        if (!class) {
+            VLOG_WARN("could not create netdev %s of unknown type %s",
+                      name, type);
+            return EAFNOSUPPORT;
+        }
+        error = class->create(class, name, &netdev_dev);
+        if (error) {
+            return error;
+        }
+        assert(netdev_dev->netdev_class == class);
 
-    if (!args) {
-        args = &empty_args;
     }
 
-    netdev_obj = shash_find_data(&netdev_obj_shash, name);
-    if (!netdev_obj) {
-        return ENODEV;
+    error = netdev_dev->netdev_class->open(netdev_dev, netdevp);
+
+    if (!error) {
+        netdev_dev->ref_cnt++;
+    } else {
+        if (!netdev_dev->ref_cnt) {
+            netdev_dev_uninit(netdev_dev, true);
+        }
     }
 
-    if (netdev_obj->netdev_class->reconfigure) {
-        return netdev_obj->netdev_class->reconfigure(netdev_obj, args);
+    return error;
+}
+
+/* Reconfigures the device 'netdev' with 'args'.  'args' may be empty
+ * or NULL if none are needed. */
+int
+netdev_set_config(struct netdev *netdev, const struct shash *args)
+{
+    struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
+
+    if (netdev_dev->netdev_class->set_config) {
+        struct shash no_args = SHASH_INITIALIZER(&no_args);
+        return netdev_dev->netdev_class->set_config(netdev_dev,
+                                                    args ? args : &no_args);
+    } else if (args && !shash_is_empty(args)) {
+        VLOG_WARN("%s: arguments provided to device that is not configurable",
+                  netdev_get_name(netdev));
     }
 
     return 0;
 }
 
-/* Opens the network device named 'name' (e.g. "eth0") and returns zero if
- * successful, otherwise a positive errno value.  On success, sets '*netdevp'
- * to the new network device, otherwise to null.
+/* Returns the current configuration for 'netdev' in 'args'.  The caller must
+ * have already initialized 'args' with shash_init().  Returns 0 on success, in
+ * which case 'args' will be filled with 'netdev''s configuration.  On failure
+ * returns a positive errno value, in which case 'args' will be empty.
  *
- * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
- * capture frames of that type received on the device.  It may also be one of
- * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
- * categories. */
+ * The caller owns 'args' and its contents and must eventually free them with
+ * shash_destroy_free_data(). */
 int
-netdev_open(const char *name, int ethertype, struct netdev **netdevp)
+netdev_get_config(const struct netdev *netdev, struct shash *args)
 {
-    struct netdev_obj *netdev_obj;
-    struct netdev *netdev = NULL;
+    struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
     int error;
-    int i;
-
-    netdev_initialize();
 
-    netdev_obj = shash_find_data(&netdev_obj_shash, name);
-    if (netdev_obj) {
-        netdev_obj->ref_cnt++;
-        error = netdev_obj->netdev_class->open(name, ethertype, &netdev);
-    } else {
-        /* Default to "system". */
-        error = EAFNOSUPPORT;
-        for (i = 0; i < n_netdev_classes; i++) {
-            const struct netdev_class *class = netdev_classes[i];
-            if (!strcmp(class->type, "system")) {
-                struct shash empty_args = SHASH_INITIALIZER(&empty_args);
-
-                /* Dynamically create the netdev object, but indicate
-                 * that it should be destroyed when the the last user
-                 * closes its handle. */
-                error = class->create(name, "system", &empty_args, false);
-                if (!error) {
-                    netdev_obj = shash_find_data(&netdev_obj_shash, name);
-                    netdev_obj->ref_cnt++;
-                    error = class->open(name, ethertype, &netdev);
-                }
-                break;
-            }
+    shash_clear_free_data(args);
+    if (netdev_dev->netdev_class->get_config) {
+        error = netdev_dev->netdev_class->get_config(netdev_dev, args);
+        if (error) {
+            shash_clear_free_data(args);
         }
+    } else {
+        error = 0;
     }
 
-    *netdevp = error ? NULL : netdev;
     return error;
 }
 
@@ -268,37 +290,16 @@ void
 netdev_close(struct netdev *netdev)
 {
     if (netdev) {
-        struct netdev_obj *netdev_obj;
-        char *name = netdev->name;
-        int error;
-
-        netdev_obj = shash_find_data(&netdev_obj_shash, name);
-        assert(netdev_obj);
-        if (netdev_obj->ref_cnt > 0) {
-            netdev_obj->ref_cnt--;
-        } else {
-            VLOG_WARN("netdev %s closed too many times", name);
-        }
+        struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
 
-        /* If the reference count for the netdev object is zero, and it
-         * was dynamically created by netdev_open(), destroy it. */
-        if (!netdev_obj->ref_cnt && !netdev_obj->created) {
-            netdev_destroy(name);
-        }
+        assert(netdev_dev->ref_cnt);
+        netdev_dev->ref_cnt--;
+        netdev_uninit(netdev, true);
 
-        /* Restore flags that we changed, if any. */
-        fatal_signal_block();
-        error = restore_flags(netdev);
-        list_remove(&netdev->node);
-        fatal_signal_unblock();
-        if (error) {
-            VLOG_WARN("failed to restore network device flags on %s: %s",
-                      name, strerror(error));
+        /* If the reference count for the netdev device is zero, destroy it. */
+        if (!netdev_dev->ref_cnt) {
+            netdev_dev_uninit(netdev_dev, true);
         }
-
-        /* Free. */
-        netdev->netdev_class->close(netdev);
-        free(name);
     }
 }
 
@@ -310,7 +311,7 @@ netdev_exists(const char *name)
     struct netdev *netdev;
     int error;
 
-    error = netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev);
+    error = netdev_open(name, "system", &netdev);
     if (!error) {
         netdev_close(netdev);
         return true;
@@ -323,32 +324,44 @@ netdev_exists(const char *name)
     }
 }
 
-/* Initializes 'svec' with a list of the names of all known network devices. */
-int
-netdev_enumerate(struct svec *svec)
+/* Returns true if a network device named 'name' is currently opened,
+ * otherwise false. */
+bool
+netdev_is_open(const char *name)
 {
-    int error;
-    int i;
+    return !!shash_find_data(&netdev_dev_shash, name);
+}
 
-    svec_init(svec);
+/* Parses 'netdev_name_', which is of the form [type@]name into its component
+ * pieces.  'name' and 'type' must be freed by the caller. */
+void
+netdev_parse_name(const char *netdev_name_, char **name, char **type)
+{
+    char *netdev_name = xstrdup(netdev_name_);
+    char *separator;
+
+    separator = strchr(netdev_name, '@');
+    if (separator) {
+        *separator = '\0';
+        *type = netdev_name;
+        *name = xstrdup(separator + 1);
+    } else {
+        *name = netdev_name;
+        *type = xstrdup("system");
+    }
+}
 
-    netdev_initialize();
+/* Attempts to set up 'netdev' for receiving packets with netdev_recv().
+ * Returns 0 if successful, otherwise a positive errno value.  EOPNOTSUPP
+ * indicates that the network device does not implement packet reception
+ * through this interface. */
+int
+netdev_listen(struct netdev *netdev)
+{
+    int (*listen)(struct netdev *);
 
-    error = 0;
-    for (i = 0; i < n_netdev_classes; i++) {
-        const struct netdev_class *class = netdev_classes[i];
-        if (class->enumerate) {
-            int retval = class->enumerate(svec);
-            if (retval) {
-                VLOG_WARN("failed to enumerate %s network devices: %s",
-                          class->type, strerror(retval));
-                if (!error) {
-                    error = retval;
-                }
-            }
-        }
-    }
-    return error;
+    listen = netdev_get_dev(netdev)->netdev_class->listen;
+    return listen ? (listen)(netdev) : EOPNOTSUPP;
 }
 
 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
@@ -358,21 +371,29 @@ netdev_enumerate(struct svec *svec)
  * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
  * need not be included.)
  *
+ * This function can only be expected to return a packet if ->listen() has
+ * been called successfully.
+ *
  * If a packet is successfully retrieved, returns 0.  In this case 'buffer' is
  * guaranteed to contain at least ETH_TOTAL_MIN bytes.  Otherwise, returns a
  * positive errno value.  Returns EAGAIN immediately if no packet is ready to
  * be returned.
- */
+ *
+ * Some network devices may not implement support for this function.  In such
+ * cases this function will always return EOPNOTSUPP. */
 int
 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
 {
+    int (*recv)(struct netdev *, void *, size_t);
     int retval;
 
     assert(buffer->size == 0);
     assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
 
-    retval = netdev->netdev_class->recv(netdev,
-                                        buffer->data, ofpbuf_tailroom(buffer));
+    recv = netdev_get_dev(netdev)->netdev_class->recv;
+    retval = (recv
+              ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
+              : -EOPNOTSUPP);
     if (retval >= 0) {
         COVERAGE_INC(netdev_received);
         buffer->size += retval;
@@ -390,14 +411,22 @@ netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
 void
 netdev_recv_wait(struct netdev *netdev)
 {
-    netdev->netdev_class->recv_wait(netdev);
+    void (*recv_wait)(struct netdev *);
+
+    recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
+    if (recv_wait) {
+        recv_wait(netdev);
+    }
 }
 
 /* Discards all packets waiting to be received from 'netdev'. */
 int
 netdev_drain(struct netdev *netdev)
 {
-    return netdev->netdev_class->drain(netdev);
+    int (*drain)(struct netdev *);
+
+    drain = netdev_get_dev(netdev)->netdev_class->drain;
+    return drain ? drain(netdev) : 0;
 }
 
 /* Sends 'buffer' on 'netdev'.  Returns 0 if successful, otherwise a positive
@@ -408,11 +437,18 @@ netdev_drain(struct netdev *netdev)
  * The caller retains ownership of 'buffer' in all cases.
  *
  * The kernel maintains a packet transmission queue, so the caller is not
- * expected to do additional queuing of packets. */
+ * expected to do additional queuing of packets.
+ *
+ * Some network devices may not implement support for this function.  In such
+ * cases this function will always return EOPNOTSUPP. */
 int
 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
 {
-    int error = netdev->netdev_class->send(netdev, buffer->data, buffer->size);
+    int (*send)(struct netdev *, const void *, size_t);
+    int error;
+
+    send = netdev_get_dev(netdev)->netdev_class->send;
+    error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
     if (!error) {
         COVERAGE_INC(netdev_sent);
     }
@@ -429,7 +465,12 @@ netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
 void
 netdev_send_wait(struct netdev *netdev)
 {
-    return netdev->netdev_class->send_wait(netdev);
+    void (*send_wait)(struct netdev *);
+
+    send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
+    if (send_wait) {
+        send_wait(netdev);
+    }
 }
 
 /* Attempts to set 'netdev''s MAC address to 'mac'.  Returns 0 if successful,
@@ -437,7 +478,7 @@ netdev_send_wait(struct netdev *netdev)
 int
 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
 {
-    return netdev->netdev_class->set_etheraddr(netdev, mac);
+    return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
 }
 
 /* Retrieves 'netdev''s MAC address.  If successful, returns 0 and copies the
@@ -446,7 +487,7 @@ netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
 int
 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
 {
-    return netdev->netdev_class->get_etheraddr(netdev, mac);
+    return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
 }
 
 /* Returns the name of the network device that 'netdev' represents,
@@ -454,25 +495,52 @@ netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
 const char *
 netdev_get_name(const struct netdev *netdev)
 {
-    return netdev->name;
+    return netdev_get_dev(netdev)->name;
 }
 
 /* Retrieves the MTU of 'netdev'.  The MTU is the maximum size of transmitted
  * (and received) packets, in bytes, not including the hardware header; thus,
  * this is typically 1500 bytes for Ethernet devices.
  *
- * If successful, returns 0 and stores the MTU size in '*mtup'.  On failure,
- * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
- * '*mtup'. */
+ * If successful, returns 0 and stores the MTU size in '*mtup'.  Returns
+ * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
+ * On other failure, returns a positive errno value.  On failure, sets '*mtup'
+ * to 0. */
 int
 netdev_get_mtu(const struct netdev *netdev, int *mtup)
 {
-    int error = netdev->netdev_class->get_mtu(netdev, mtup);
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    int error;
+
+    error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
     if (error) {
-        VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
+        *mtup = 0;
+        if (error != EOPNOTSUPP) {
+            VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
+                         "%s", netdev_get_name(netdev), strerror(error));
+        }
+    }
+    return error;
+}
+
+/* Sets the MTU of 'netdev'.  The MTU is the maximum size of transmitted
+ * (and received) packets, in bytes.
+ *
+ * If successful, returns 0.  Returns EOPNOTSUPP if 'netdev' does not have an
+ * MTU (as e.g. some tunnels do not).  On other failure, returns a positive
+ * errno value. */
+int
+netdev_set_mtu(const struct netdev *netdev, int mtu)
+{
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    int error;
+
+    error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
+    if (error && error != EOPNOTSUPP) {
+        VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
                      netdev_get_name(netdev), strerror(error));
-        *mtup = ETH_PAYLOAD_MAX;
     }
+
     return error;
 }
 
@@ -484,24 +552,41 @@ netdev_get_mtu(const struct netdev *netdev, int *mtup)
  * value should be unique within a host and remain stable at least until
  * reboot.  SNMP says an ifindex "ranges between 1 and the value of ifNumber"
  * but many systems do not follow this rule anyhow.
+ *
+ * Some network devices may not implement support for this function.  In such
+ * cases this function will always return -EOPNOTSUPP.
  */
 int
 netdev_get_ifindex(const struct netdev *netdev)
 {
-    return netdev->netdev_class->get_ifindex(netdev);
+    int (*get_ifindex)(const struct netdev *);
+
+    get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
+
+    return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
 }
 
 /* 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
  * successful, otherwise a positive errno value.  On failure, all of the
- * passed-in values are set to 0. */
+ * passed-in values are set to 0.
+ *
+ * Some network devices may not implement support for this function.  In such
+ * cases this function will always return EOPNOTSUPP. */
 int
-netdev_get_features(struct netdev *netdev,
-                    uint32_t *current, uint32_t *advertised,
-                    uint32_t *supported, uint32_t *peer)
+netdev_get_features(const struct netdev *netdev,
+                    enum netdev_features *current,
+                    enum netdev_features *advertised,
+                    enum netdev_features *supported,
+                    enum netdev_features *peer)
 {
-    uint32_t dummy[4];
+    int (*get_features)(const struct netdev *netdev,
+                        enum netdev_features *current,
+                        enum netdev_features *advertised,
+                        enum netdev_features *supported,
+                        enum netdev_features *peer);
+    enum netdev_features dummy[4];
     int error;
 
     if (!current) {
@@ -517,50 +602,62 @@ netdev_get_features(struct netdev *netdev,
         peer = &dummy[3];
     }
 
-    error = netdev->netdev_class->get_features(netdev, current, advertised,
-                                               supported, peer);
+    get_features = netdev_get_dev(netdev)->netdev_class->get_features;
+    error = get_features
+                    ? get_features(netdev, current, advertised, supported,
+                                   peer)
+                    : EOPNOTSUPP;
     if (error) {
         *current = *advertised = *supported = *peer = 0;
     }
     return error;
 }
 
-/* Returns the maximum speed of a network connection that has the "enum
- * ofp_port_features" bits in 'features', in bits per second.  If no bits that
- * indicate a speed are set in 'features', assumes 100Mbps. */
+/* Returns the maximum speed of a network connection that has the NETDEV_F_*
+ * bits in 'features', in bits per second.  If no bits that indicate a speed
+ * are set in 'features', assumes 100Mbps. */
 uint64_t
-netdev_features_to_bps(uint32_t features)
+netdev_features_to_bps(enum netdev_features features)
 {
     enum {
-        F_10000MB = OFPPF_10GB_FD,
-        F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
-        F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
-        F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
+        F_1000000MB = NETDEV_F_1TB_FD,
+        F_100000MB = NETDEV_F_100GB_FD,
+        F_40000MB = NETDEV_F_40GB_FD,
+        F_10000MB = NETDEV_F_10GB_FD,
+        F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
+        F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
+        F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
     };
 
-    return (  features & F_10000MB  ? UINT64_C(10000000000)
-            : features & F_1000MB   ? UINT64_C(1000000000)
-            : features & F_100MB    ? UINT64_C(100000000)
-            : features & F_10MB     ? UINT64_C(10000000)
-                                    : UINT64_C(100000000));
+    return (  features & F_1000000MB ? UINT64_C(1000000000000)
+            : features & F_100000MB  ? UINT64_C(100000000000)
+            : features & F_40000MB   ? UINT64_C(40000000000)
+            : features & F_10000MB   ? UINT64_C(10000000000)
+            : features & F_1000MB    ? UINT64_C(1000000000)
+            : features & F_100MB     ? UINT64_C(100000000)
+            : features & F_10MB      ? UINT64_C(10000000)
+                                     : UINT64_C(100000000));
 }
 
-/* Returns true if any of the "enum ofp_port_features" bits that indicate a
- * full-duplex link are set in 'features', otherwise false. */
+/* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
+ * are set in 'features', otherwise false. */
 bool
-netdev_features_is_full_duplex(uint32_t features)
+netdev_features_is_full_duplex(enum netdev_features features)
 {
-    return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
-                        | OFPPF_10GB_FD)) != 0;
+    return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
+                        | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
+                        | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
 }
 
 /* Set the features advertised by 'netdev' to 'advertise'.  Returns 0 if
  * successful, otherwise a positive errno value. */
 int
-netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
+netdev_set_advertisements(struct netdev *netdev,
+                          enum netdev_features advertise)
 {
-    return (netdev->netdev_class->set_advertisements
-            ? netdev->netdev_class->set_advertisements(netdev, advertise)
+    return (netdev_get_dev(netdev)->netdev_class->set_advertisements
+            ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
+                    netdev, advertise)
             : EOPNOTSUPP);
 }
 
@@ -574,8 +671,8 @@ netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
  *
  *   - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
  *
- * 'address' or 'netmask' or both may be null, in which case the address or netmask
- * is not reported. */
+ * 'address' or 'netmask' or both may be null, in which case the address or
+ * netmask is not reported. */
 int
 netdev_get_in4(const struct netdev *netdev,
                struct in_addr *address_, struct in_addr *netmask_)
@@ -584,8 +681,9 @@ netdev_get_in4(const struct netdev *netdev,
     struct in_addr netmask;
     int error;
 
-    error = (netdev->netdev_class->get_in4
-             ? netdev->netdev_class->get_in4(netdev, &address, &netmask)
+    error = (netdev_get_dev(netdev)->netdev_class->get_in4
+             ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
+                    &address, &netmask)
              : EOPNOTSUPP);
     if (address_) {
         address_->s_addr = error ? 0 : address.s_addr;
@@ -602,19 +700,39 @@ netdev_get_in4(const struct netdev *netdev,
 int
 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
 {
-    return (netdev->netdev_class->set_in4
-            ? netdev->netdev_class->set_in4(netdev, addr, mask)
+    return (netdev_get_dev(netdev)->netdev_class->set_in4
+            ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
             : EOPNOTSUPP);
 }
 
+/* Obtains ad IPv4 address from device name and save the address in
+ * in4.  Returns 0 if successful, otherwise a positive errno value.
+ */
+int
+netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
+{
+    struct netdev *netdev;
+    int error;
+
+    error = netdev_open(device_name, "system", &netdev);
+    if (error) {
+        in4->s_addr = htonl(0);
+        return error;
+    }
+
+    error = netdev_get_in4(netdev, in4, NULL);
+    netdev_close(netdev);
+    return error;
+}
+
 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
  * to 'netdev'. */
 int
 netdev_add_router(struct netdev *netdev, struct in_addr router)
 {
     COVERAGE_INC(netdev_add_router);
-    return (netdev->netdev_class->add_router
-            ? netdev->netdev_class->add_router(netdev, router)
+    return (netdev_get_dev(netdev)->netdev_class->add_router
+            ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
             : EOPNOTSUPP);
 }
 
@@ -630,9 +748,9 @@ netdev_get_next_hop(const struct netdev *netdev,
                     const struct in_addr *host, struct in_addr *next_hop,
                     char **netdev_name)
 {
-    int error = (netdev->netdev_class->get_next_hop
-                 ? netdev->netdev_class->get_next_hop(host, next_hop,
-                                                      netdev_name)
+    int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
+                 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
+                        host, next_hop, netdev_name)
                  : EOPNOTSUPP);
     if (error) {
         next_hop->s_addr = 0;
@@ -641,6 +759,21 @@ netdev_get_next_hop(const struct netdev *netdev,
     return error;
 }
 
+/* Populates 'sh' with status information.
+ *
+ * Populates 'sh' with 'netdev' specific status information.  This information
+ * may be used to populate the status column of the Interface table as defined
+ * in ovs-vswitchd.conf.db(5). */
+int
+netdev_get_drv_info(const struct netdev *netdev, struct shash *sh)
+{
+    struct netdev_dev *dev = netdev_get_dev(netdev);
+
+    return (dev->netdev_class->get_drv_info
+            ? dev->netdev_class->get_drv_info(netdev, sh)
+            : EOPNOTSUPP);
+}
+
 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
  * returns 0.  Otherwise, returns a positive errno value and sets '*in6' to
  * all-zero-bits (in6addr_any).
@@ -658,8 +791,9 @@ netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
     struct in6_addr dummy;
     int error;
 
-    error = (netdev->netdev_class->get_in6
-             ? netdev->netdev_class->get_in6(netdev, in6 ? in6 : &dummy)
+    error = (netdev_get_dev(netdev)->netdev_class->get_in6
+             ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
+                    in6 ? in6 : &dummy)
              : EOPNOTSUPP);
     if (error && in6) {
         memset(in6, 0, sizeof *in6);
@@ -679,8 +813,8 @@ do_update_flags(struct netdev *netdev, enum netdev_flags off,
     enum netdev_flags old_flags;
     int error;
 
-    error = netdev->netdev_class->update_flags(netdev, off & ~on,
-                                               on, &old_flags);
+    error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
+                off & ~on, on, &old_flags);
     if (error) {
         VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
                      off || on ? "set" : "get", netdev_get_name(netdev),
@@ -751,10 +885,11 @@ netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
  * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
 int
 netdev_arp_lookup(const struct netdev *netdev,
-                  uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
+                  ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
 {
-    int error = (netdev->netdev_class->arp_lookup
-                 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
+    int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
+                 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
+                        ip, mac)
                  : EOPNOTSUPP);
     if (error) {
         memset(mac, 0, ETH_ADDR_LEN);
@@ -762,18 +897,58 @@ netdev_arp_lookup(const struct netdev *netdev,
     return error;
 }
 
-/* Sets 'carrier' to true if carrier is active (link light is on) on
- * 'netdev'. */
-int
-netdev_get_carrier(const struct netdev *netdev, bool *carrier)
+/* Returns true if carrier is active (link light is on) on 'netdev'. */
+bool
+netdev_get_carrier(const struct netdev *netdev)
 {
-    int error = (netdev->netdev_class->get_carrier
-                 ? netdev->netdev_class->get_carrier(netdev, carrier)
-                 : EOPNOTSUPP);
+    int error;
+    enum netdev_flags flags;
+    bool carrier;
+
+    netdev_get_flags(netdev, &flags);
+    if (!(flags & NETDEV_UP)) {
+        return false;
+    }
+
+    if (!netdev_get_dev(netdev)->netdev_class->get_carrier) {
+        return true;
+    }
+
+    error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
+                                                              &carrier);
     if (error) {
-        *carrier = false;
+        VLOG_DBG("%s: failed to get network device carrier status, assuming "
+                 "down: %s", netdev_get_name(netdev), strerror(error));
+        carrier = false;
     }
-    return error;
+
+    return carrier;
+}
+
+/* Returns the number of times 'netdev''s carrier has changed. */
+long long int
+netdev_get_carrier_resets(const struct netdev *netdev)
+{
+    return (netdev_get_dev(netdev)->netdev_class->get_carrier_resets
+            ? netdev_get_dev(netdev)->netdev_class->get_carrier_resets(netdev)
+            : 0);
+}
+
+/* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
+ * link status instead of checking 'netdev''s carrier.  'netdev''s MII
+ * registers will be polled once ever 'interval' milliseconds.  If 'netdev'
+ * does not support MII, another method may be used as a fallback.  If
+ * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
+ * its normal behavior.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+int
+netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
+{
+    struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
+    return (netdev_dev->netdev_class->set_miimon_interval
+            ? netdev_dev->netdev_class->set_miimon_interval(netdev, interval)
+            : EOPNOTSUPP);
 }
 
 /* Retrieves current device stats for 'netdev'. */
@@ -783,8 +958,8 @@ netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
     int error;
 
     COVERAGE_INC(netdev_get_stats);
-    error = (netdev->netdev_class->get_stats
-             ? netdev->netdev_class->get_stats(netdev, stats)
+    error = (netdev_get_dev(netdev)->netdev_class->get_stats
+             ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
              : EOPNOTSUPP);
     if (error) {
         memset(stats, 0xff, sizeof *stats);
@@ -792,6 +967,19 @@ netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
     return error;
 }
 
+/* Attempts to change the stats for 'netdev' to those provided in 'stats'.
+ * Returns 0 if successful, otherwise a positive errno value.
+ *
+ * This will probably fail for most network devices.  Some devices might only
+ * allow setting their stats to 0. */
+int
+netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
+{
+    return (netdev_get_dev(netdev)->netdev_class->set_stats
+             ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
+             : EOPNOTSUPP);
+}
+
 /* Attempts to set input rate limiting (policing) policy, such that up to
  * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
  * size of 'kbits' kb. */
@@ -799,237 +987,448 @@ int
 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
                     uint32_t kbits_burst)
 {
-    return (netdev->netdev_class->set_policing
-            ? netdev->netdev_class->set_policing(netdev,
-                                                 kbits_rate, kbits_burst)
+    return (netdev_get_dev(netdev)->netdev_class->set_policing
+            ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
+                    kbits_rate, kbits_burst)
             : EOPNOTSUPP);
 }
 
-/* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
- * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
- * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
- * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
- * -1. */
+/* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
+ * empty if 'netdev' does not support QoS.  Any names added to 'types' should
+ * be documented as valid for the "type" column in the "QoS" table in
+ * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
+ *
+ * Every network device supports disabling QoS with a type of "", but this type
+ * will not be added to 'types'.
+ *
+ * The caller must initialize 'types' (e.g. with sset_init()) before calling
+ * this function.  The caller is responsible for destroying 'types' (e.g. with
+ * sset_destroy()) when it is no longer needed.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
 int
-netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
+netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
 {
-    int error = (netdev->netdev_class->get_vlan_vid
-                 ? netdev->netdev_class->get_vlan_vid(netdev, vlan_vid)
-                 : ENOENT);
-    if (error) {
-        *vlan_vid = 0;
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    return (class->get_qos_types
+            ? class->get_qos_types(netdev, types)
+            : 0);
+}
+
+/* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
+ * which should be "" or one of the types returned by netdev_get_qos_types()
+ * for 'netdev'.  Returns 0 if successful, otherwise a positive errno value.
+ * On success, initializes 'caps' with the QoS capabilities; on failure, clears
+ * 'caps' to all zeros. */
+int
+netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
+                            struct netdev_qos_capabilities *caps)
+{
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+
+    if (*type) {
+        int retval = (class->get_qos_capabilities
+                      ? class->get_qos_capabilities(netdev, type, caps)
+                      : EOPNOTSUPP);
+        if (retval) {
+            memset(caps, 0, sizeof *caps);
+        }
+        return retval;
+    } else {
+        /* Every netdev supports turning off QoS. */
+        memset(caps, 0, sizeof *caps);
+        return 0;
     }
-    return error;
 }
 
-/* Returns a network device that has 'in4' as its IP address, if one exists,
- * otherwise a null pointer. */
-struct netdev *
-netdev_find_dev_by_in4(const struct in_addr *in4)
+/* Obtains the number of queues supported by 'netdev' for the specified 'type'
+ * of QoS.  Returns 0 if successful, otherwise a positive errno value.  Stores
+ * the number of queues (zero on failure) in '*n_queuesp'.
+ *
+ * This is just a simple wrapper around netdev_get_qos_capabilities(). */
+int
+netdev_get_n_queues(const struct netdev *netdev,
+                    const char *type, unsigned int *n_queuesp)
 {
-    struct netdev *netdev;
-    struct svec dev_list;
-    size_t i;
-
-    netdev_enumerate(&dev_list);
-    for (i = 0; i < dev_list.n; i++) {
-        const char *name = dev_list.names[i];
-        struct in_addr dev_in4;
-
-        if (!netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev)
-            && !netdev_get_in4(netdev, &dev_in4, NULL)
-            && dev_in4.s_addr == in4->s_addr) {
-            goto exit;
+    struct netdev_qos_capabilities caps;
+    int retval;
+
+    retval = netdev_get_qos_capabilities(netdev, type, &caps);
+    *n_queuesp = caps.n_queues;
+    return retval;
+}
+
+/* Queries 'netdev' about its currently configured form of QoS.  If successful,
+ * stores the name of the current form of QoS into '*typep', stores any details
+ * of configuration as string key-value pairs in 'details', and returns 0.  On
+ * failure, sets '*typep' to NULL and returns a positive errno value.
+ *
+ * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
+ *
+ * The caller must initialize 'details' as an empty shash (e.g. with
+ * shash_init()) before calling this function.  The caller must free 'details',
+ * including 'data' members, when it is no longer needed (e.g. with
+ * shash_destroy_free_data()).
+ *
+ * The caller must not modify or free '*typep'.
+ *
+ * '*typep' will be one of the types returned by netdev_get_qos_types() for
+ * 'netdev'.  The contents of 'details' should be documented as valid for
+ * '*typep' in the "other_config" column in the "QoS" table in
+ * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
+int
+netdev_get_qos(const struct netdev *netdev,
+               const char **typep, struct shash *details)
+{
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    int retval;
+
+    if (class->get_qos) {
+        retval = class->get_qos(netdev, typep, details);
+        if (retval) {
+            *typep = NULL;
+            shash_clear_free_data(details);
         }
-        netdev_close(netdev);
+        return retval;
+    } else {
+        /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
+        *typep = "";
+        return 0;
+    }
+}
+
+/* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
+ * with details of configuration from 'details'.  Returns 0 if successful,
+ * otherwise a positive errno value.  On error, the previous QoS configuration
+ * is retained.
+ *
+ * When this function changes the type of QoS (not just 'details'), this also
+ * resets all queue configuration for 'netdev' to their defaults (which depend
+ * on the specific type of QoS).  Otherwise, the queue configuration for
+ * 'netdev' is unchanged.
+ *
+ * 'type' should be "" (to disable QoS) or one of the types returned by
+ * netdev_get_qos_types() for 'netdev'.  The contents of 'details' should be
+ * documented as valid for the given 'type' in the "other_config" column in the
+ * "QoS" table in vswitchd/vswitch.xml (which is built as
+ * ovs-vswitchd.conf.db(8)).
+ *
+ * NULL may be specified for 'details' if there are no configuration
+ * details. */
+int
+netdev_set_qos(struct netdev *netdev,
+               const char *type, const struct shash *details)
+{
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+
+    if (!type) {
+        type = "";
     }
-    netdev = NULL;
 
-exit:
-    svec_destroy(&dev_list);
-    return netdev;
+    if (class->set_qos) {
+        if (!details) {
+            static struct shash empty = SHASH_INITIALIZER(&empty);
+            details = &empty;
+        }
+        return class->set_qos(netdev, type, details);
+    } else {
+        return *type ? EOPNOTSUPP : 0;
+    }
 }
-
-/* Initializes 'netdev_obj' as a netdev object named 'name' of the 
- * specified 'netdev_class'.
+
+/* Queries 'netdev' for information about the queue numbered 'queue_id'.  If
+ * successful, adds that information as string key-value pairs to 'details'.
+ * Returns 0 if successful, otherwise a positive errno value.
  *
- * This function adds 'netdev_obj' to a netdev-owned shash, so it is
- * very important that 'netdev_obj' only be freed after calling
- * netdev_destroy().  */
-void
-netdev_obj_init(struct netdev_obj *netdev_obj, const char *name,
-                const struct netdev_class *netdev_class, bool created)
+ * 'queue_id' must be less than the number of queues supported by 'netdev' for
+ * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
+ *
+ * The returned contents of 'details' should be documented as valid for the
+ * given 'type' in the "other_config" column in the "Queue" table in
+ * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
+ *
+ * The caller must initialize 'details' (e.g. with shash_init()) before calling
+ * this function.  The caller must free 'details', including 'data' members,
+ * when it is no longer needed (e.g. with shash_destroy_free_data()). */
+int
+netdev_get_queue(const struct netdev *netdev,
+                 unsigned int queue_id, struct shash *details)
 {
-    assert(!shash_find(&netdev_obj_shash, name));
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    int retval;
 
-    netdev_obj->netdev_class = netdev_class;
-    netdev_obj->ref_cnt = 0;
-    netdev_obj->created = created;
-    shash_add(&netdev_obj_shash, name, netdev_obj);
+    retval = (class->get_queue
+              ? class->get_queue(netdev, queue_id, details)
+              : EOPNOTSUPP);
+    if (retval) {
+        shash_clear_free_data(details);
+    }
+    return retval;
 }
 
-/* Initializes 'netdev' as a netdev named 'name' of the specified
- * 'netdev_class'.
+/* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
+ * string pairs in 'details'.  The contents of 'details' should be documented
+ * as valid for the given 'type' in the "other_config" column in the "Queue"
+ * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
+ * Returns 0 if successful, otherwise a positive errno value.  On failure, the
+ * given queue's configuration should be unmodified.
  *
- * This function adds 'netdev' to a netdev-owned linked list, so it is very
- * important that 'netdev' only be freed after calling netdev_close(). */
-void
-netdev_init(struct netdev *netdev, const char *name,
-            const struct netdev_class *netdev_class)
+ * 'queue_id' must be less than the number of queues supported by 'netdev' for
+ * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
+ *
+ * This function does not modify 'details', and the caller retains ownership of
+ * it. */
+int
+netdev_set_queue(struct netdev *netdev,
+                 unsigned int queue_id, const struct shash *details)
 {
-    netdev->netdev_class = netdev_class;
-    netdev->name = xstrdup(name);
-    netdev->save_flags = 0;
-    netdev->changed_flags = 0;
-    list_push_back(&netdev_list, &netdev->node);
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    return (class->set_queue
+            ? class->set_queue(netdev, queue_id, details)
+            : EOPNOTSUPP);
 }
 
-/* Returns the class type of 'netdev'.  
+/* Attempts to delete the queue numbered 'queue_id' from 'netdev'.  Some kinds
+ * of QoS may have a fixed set of queues, in which case attempts to delete them
+ * will fail with EOPNOTSUPP.
  *
- * The caller must not free the returned value. */
-const char *netdev_get_type(const struct netdev *netdev)
+ * Returns 0 if successful, otherwise a positive errno value.  On failure, the
+ * given queue will be unmodified.
+ *
+ * 'queue_id' must be less than the number of queues supported by 'netdev' for
+ * the current form of QoS (e.g. as returned by
+ * netdev_get_n_queues(netdev)). */
+int
+netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
 {
-    return netdev->netdev_class->type;
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    return (class->delete_queue
+            ? class->delete_queue(netdev, queue_id)
+            : EOPNOTSUPP);
 }
 
-/* Initializes 'notifier' as a netdev notifier for 'netdev', for which
- * notification will consist of calling 'cb', with auxiliary data 'aux'. */
-void
-netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
-                     void (*cb)(struct netdev_notifier *), void *aux)
+/* Obtains statistics about 'queue_id' on 'netdev'.  On success, returns 0 and
+ * fills 'stats' with the queue's statistics; individual members of 'stats' may
+ * be set to all-1-bits if the statistic is unavailable.  On failure, returns a
+ * positive errno value and fills 'stats' with all-1-bits. */
+int
+netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
+                       struct netdev_queue_stats *stats)
 {
-    notifier->netdev = netdev;
-    notifier->cb = cb;
-    notifier->aux = aux;
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    int retval;
+
+    retval = (class->get_queue_stats
+              ? class->get_queue_stats(netdev, queue_id, stats)
+              : EOPNOTSUPP);
+    if (retval) {
+        memset(stats, 0xff, sizeof *stats);
+    }
+    return retval;
+}
+
+/* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
+ * its configuration, and the 'aux' specified by the caller.  The order of
+ * iteration is unspecified, but (when successful) each queue is visited
+ * exactly once.
+ *
+ * Calling this function may be more efficient than calling netdev_get_queue()
+ * for every queue.
+ *
+ * 'cb' must not modify or free the 'details' argument passed in.  It may
+ * delete or modify the queue passed in as its 'queue_id' argument.  It may
+ * modify but must not delete any other queue within 'netdev'.  'cb' should not
+ * add new queues because this may cause some queues to be visited twice or not
+ * at all.
+ *
+ * Returns 0 if successful, otherwise a positive errno value.  On error, some
+ * configured queues may not have been included in the iteration. */
+int
+netdev_dump_queues(const struct netdev *netdev,
+                   netdev_dump_queues_cb *cb, void *aux)
+{
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    return (class->dump_queues
+            ? class->dump_queues(netdev, cb, aux)
+            : EOPNOTSUPP);
 }
-
-/* Tracks changes in the status of a set of network devices. */
-struct netdev_monitor {
-    struct shash polled_netdevs;
-    struct shash changed_netdevs;
-};
 
-/* Creates and returns a new structure for monitor changes in the status of
- * network devices. */
-struct netdev_monitor *
-netdev_monitor_create(void)
+/* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
+ * its statistics, and the 'aux' specified by the caller.  The order of
+ * iteration is unspecified, but (when successful) each queue is visited
+ * exactly once.
+ *
+ * Calling this function may be more efficient than calling
+ * netdev_get_queue_stats() for every queue.
+ *
+ * 'cb' must not modify or free the statistics passed in.
+ *
+ * Returns 0 if successful, otherwise a positive errno value.  On error, some
+ * configured queues may not have been included in the iteration. */
+int
+netdev_dump_queue_stats(const struct netdev *netdev,
+                        netdev_dump_queue_stats_cb *cb, void *aux)
+{
+    const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
+    return (class->dump_queue_stats
+            ? class->dump_queue_stats(netdev, cb, aux)
+            : EOPNOTSUPP);
+}
+
+/* Returns a sequence number which indicates changes in one of 'netdev''s
+ * properties.  The returned sequence will be nonzero so that callers have a
+ * value which they may use as a reset when tracking 'netdev'.
+ *
+ * The returned sequence number will change whenever 'netdev''s flags,
+ * features, ethernet address, or carrier changes.  It may change for other
+ * reasons as well, or no reason at all. */
+unsigned int
+netdev_change_seq(const struct netdev *netdev)
+{
+    return netdev_get_dev(netdev)->netdev_class->change_seq(netdev);
+}
+
+/* Initializes 'netdev_dev' as a netdev device named 'name' of the specified
+ * 'netdev_class'.  This function is ordinarily called from a netdev provider's
+ * 'create' function.
+ *
+ * This function adds 'netdev_dev' to a netdev-owned shash, so it is
+ * very important that 'netdev_dev' only be freed after calling
+ * the refcount drops to zero.  */
+void
+netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
+                const struct netdev_class *netdev_class)
 {
-    struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
-    shash_init(&monitor->polled_netdevs);
-    shash_init(&monitor->changed_netdevs);
-    return monitor;
+    assert(!shash_find(&netdev_dev_shash, name));
+
+    memset(netdev_dev, 0, sizeof *netdev_dev);
+    netdev_dev->netdev_class = netdev_class;
+    netdev_dev->name = xstrdup(name);
+    netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
 }
 
-/* Destroys 'monitor'. */
+/* Undoes the results of initialization.
+ *
+ * Normally this function does not need to be called as netdev_close has
+ * the same effect when the refcount drops to zero.
+ * However, it may be called by providers due to an error on creation
+ * that occurs after initialization.  It this case netdev_close() would
+ * never be called. */
 void
-netdev_monitor_destroy(struct netdev_monitor *monitor)
+netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
 {
-    if (monitor) {
-        struct shash_node *node;
+    char *name = netdev_dev->name;
 
-        SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
-            struct netdev_notifier *notifier = node->data;
-            notifier->netdev->netdev_class->poll_remove(notifier);
-        }
+    assert(!netdev_dev->ref_cnt);
+
+    shash_delete(&netdev_dev_shash, netdev_dev->node);
 
-        shash_destroy(&monitor->polled_netdevs);
-        shash_destroy(&monitor->changed_netdevs);
-        free(monitor);
+    if (destroy) {
+        netdev_dev->netdev_class->destroy(netdev_dev);
     }
+    free(name);
 }
 
-static void
-netdev_monitor_cb(struct netdev_notifier *notifier)
+/* Returns the class type of 'netdev_dev'.
+ *
+ * The caller must not free the returned value. */
+const char *
+netdev_dev_get_type(const struct netdev_dev *netdev_dev)
 {
-    struct netdev_monitor *monitor = notifier->aux;
-    const char *name = netdev_get_name(notifier->netdev);
-    if (!shash_find(&monitor->changed_netdevs, name)) {
-        shash_add(&monitor->changed_netdevs, name, NULL);
-    }
+    return netdev_dev->netdev_class->type;
 }
 
-/* Attempts to add 'netdev' as a netdev monitored by 'monitor'.  Returns 0 if
- * successful, otherwise a positive errno value.
+/* Returns the class associated with 'netdev_dev'. */
+const struct netdev_class *
+netdev_dev_get_class(const struct netdev_dev *netdev_dev)
+{
+    return netdev_dev->netdev_class;
+}
+
+/* Returns the name of 'netdev_dev'.
  *
- * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
- * it once. */
-int
-netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
-{
-    const char *netdev_name = netdev_get_name(netdev);
-    int error = 0;
-    if (!shash_find(&monitor->polled_netdevs, netdev_name)
-        && netdev->netdev_class->poll_add)
-    {
-        struct netdev_notifier *notifier;
-        error = netdev->netdev_class->poll_add(netdev, netdev_monitor_cb,
-                                               monitor, &notifier);
-        if (!error) {
-            assert(notifier->netdev == netdev);
-            shash_add(&monitor->polled_netdevs, netdev_name, notifier);
-        }
-    }
-    return error;
+ * The caller must not free the returned value. */
+const char *
+netdev_dev_get_name(const struct netdev_dev *netdev_dev)
+{
+    return netdev_dev->name;
+}
+
+/* Returns the netdev_dev with 'name' or NULL if there is none.
+ *
+ * The caller must not free the returned value. */
+struct netdev_dev *
+netdev_dev_from_name(const char *name)
+{
+    return shash_find_data(&netdev_dev_shash, name);
 }
 
-/* Removes 'netdev' from the set of netdevs monitored by 'monitor'.  (This has
- * no effect if 'netdev' is not in the set of devices monitored by
- * 'monitor'.) */
+/* Fills 'device_list' with devices that match 'netdev_class'.
+ *
+ * The caller is responsible for initializing and destroying 'device_list'
+ * but the contained netdev_devs must not be freed. */
 void
-netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
+netdev_dev_get_devices(const struct netdev_class *netdev_class,
+                       struct shash *device_list)
 {
-    const char *netdev_name = netdev_get_name(netdev);
     struct shash_node *node;
+    SHASH_FOR_EACH (node, &netdev_dev_shash) {
+        struct netdev_dev *dev = node->data;
 
-    node = shash_find(&monitor->polled_netdevs, netdev_name);
-    if (node) {
-        /* Cancel future notifications. */
-        struct netdev_notifier *notifier = node->data;
-        netdev->netdev_class->poll_remove(notifier);
-        shash_delete(&monitor->polled_netdevs, node);
-
-        /* Drop any pending notification. */
-        node = shash_find(&monitor->changed_netdevs, netdev_name);
-        if (node) {
-            shash_delete(&monitor->changed_netdevs, node);
+        if (dev->netdev_class == netdev_class) {
+            shash_add(device_list, node->name, node->data);
         }
     }
 }
 
-/* Checks for changes to netdevs in the set monitored by 'monitor'.  If any of
- * the attributes (Ethernet address, carrier status, speed or peer-advertised
- * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
- * sets '*devnamep' to the name of a device that has changed and returns 0.
- * The caller is responsible for freeing '*devnamep' (with free()).
+/* Initializes 'netdev' as a instance of the netdev_dev.
  *
- * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN.
- */
-int
-netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
+ * This function adds 'netdev' to a netdev-owned linked list, so it is very
+ * important that 'netdev' only be freed after calling netdev_close(). */
+void
+netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
 {
-    struct shash_node *node = shash_first(&monitor->changed_netdevs);
-    if (!node) {
-        *devnamep = NULL;
-        return EAGAIN;
-    } else {
-        *devnamep = xstrdup(node->name);
-        shash_delete(&monitor->changed_netdevs, node);
-        return 0;
-    }
+    memset(netdev, 0, sizeof *netdev);
+    netdev->netdev_dev = netdev_dev;
+    list_push_back(&netdev_list, &netdev->node);
 }
 
-/* Registers with the poll loop to wake up from the next call to poll_block()
- * when netdev_monitor_poll(monitor) would indicate that a device has
- * changed. */
+/* Undoes the results of initialization.
+ *
+ * Normally this function only needs to be called from netdev_close().
+ * However, it may be called by providers due to an error on opening
+ * that occurs after initialization.  It this case netdev_close() would
+ * never be called. */
 void
-netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
+netdev_uninit(struct netdev *netdev, bool close)
 {
-    if (!shash_is_empty(&monitor->changed_netdevs)) {
-        poll_immediate_wake();
-    } else {
-        /* XXX Nothing needed here for netdev_linux, but maybe other netdev
-         * classes need help. */
+    /* Restore flags that we changed, if any. */
+    int error = restore_flags(netdev);
+    list_remove(&netdev->node);
+    if (error) {
+        VLOG_WARN("failed to restore network device flags on %s: %s",
+                  netdev_get_name(netdev), strerror(error));
     }
+
+    if (close) {
+        netdev_get_dev(netdev)->netdev_class->close(netdev);
+    }
+}
+
+
+/* Returns the class type of 'netdev'.
+ *
+ * The caller must not free the returned value. */
+const char *
+netdev_get_type(const struct netdev *netdev)
+{
+    return netdev_get_dev(netdev)->netdev_class->type;
+}
+
+struct netdev_dev *
+netdev_get_dev(const struct netdev *netdev)
+{
+    return netdev->netdev_dev;
 }
 
 /* Restore the network device flags on 'netdev' to those that were active
@@ -1043,21 +1442,20 @@ restore_flags(struct netdev *netdev)
     if (netdev->changed_flags) {
         enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
         enum netdev_flags old_flags;
-        return netdev->netdev_class->update_flags(netdev,
-                                                  netdev->changed_flags
-                                                  & ~restore,
-                                                  restore, &old_flags);
+        return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
+                                           netdev->changed_flags & ~restore,
+                                           restore, &old_flags);
     }
     return 0;
 }
 
-/* Retores all the flags on all network devices that we modified.  Called from
- * a signal handler, so it does not attempt to report error conditions. */
+/* Close all netdevs on shutdown so they can do any needed cleanup such as
+ * destroying devices, restoring flags, etc. */
 static void
-restore_all_flags(void *aux UNUSED)
+close_all_netdevs(void *aux OVS_UNUSED)
 {
-    struct netdev *netdev;
-    LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
-        restore_flags(netdev);
+    struct netdev *netdev, *next;
+    LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
+        netdev_close(netdev);
     }
 }