netdev-offload-tc: Use single 'once' variable for probing tc features

[mirror_ovs.git] / lib / netdev-linux.c

/*
 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017 Nicira, Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <config.h>

#include "netdev-linux.h"
#include "netdev-linux-private.h"

#include <errno.h>
#include <fcntl.h>
#include <sys/types.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <inttypes.h>
#include <math.h>
#include <linux/filter.h>
#include <linux/gen_stats.h>
#include <linux/if_ether.h>
#include <linux/if_packet.h>
#include <linux/if_tun.h>
#include <linux/types.h>
#include <linux/ethtool.h>
#include <linux/mii.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
#include <linux/virtio_net.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/uio.h>
#include <sys/utsname.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/route.h>
#include <poll.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include "coverage.h"
#include "dp-packet.h"
#include "dpif-netlink.h"
#include "dpif-netdev.h"
#include "openvswitch/dynamic-string.h"
#include "fatal-signal.h"
#include "hash.h"
#include "openvswitch/hmap.h"
#include "netdev-afxdp.h"
#include "netdev-provider.h"
#include "netdev-vport.h"
#include "netlink-notifier.h"
#include "netlink-socket.h"
#include "netlink.h"
#include "netnsid.h"
#include "openvswitch/ofpbuf.h"
#include "openflow/openflow.h"
#include "ovs-atomic.h"
#include "ovs-numa.h"
#include "packets.h"
#include "openvswitch/poll-loop.h"
#include "rtnetlink.h"
#include "openvswitch/shash.h"
#include "socket-util.h"
#include "sset.h"
#include "tc.h"
#include "timer.h"
#include "unaligned.h"
#include "openvswitch/vlog.h"
#include "userspace-tso.h"
#include "util.h"

VLOG_DEFINE_THIS_MODULE(netdev_linux);

COVERAGE_DEFINE(netdev_set_policing);
COVERAGE_DEFINE(netdev_arp_lookup);
COVERAGE_DEFINE(netdev_get_ifindex);
COVERAGE_DEFINE(netdev_get_hwaddr);
COVERAGE_DEFINE(netdev_set_hwaddr);
COVERAGE_DEFINE(netdev_get_ethtool);
COVERAGE_DEFINE(netdev_set_ethtool);

\f
#ifndef IFLA_IF_NETNSID
#define IFLA_IF_NETNSID 0x45
#endif
/* These were introduced in Linux 2.6.14, so they might be missing if we have
 * old headers. */
#ifndef ADVERTISED_Pause
#define ADVERTISED_Pause                (1 << 13)
#endif
#ifndef ADVERTISED_Asym_Pause
#define ADVERTISED_Asym_Pause           (1 << 14)
#endif

/* These were introduced in Linux 2.6.24, so they might be missing if we
 * have old headers. */
#ifndef ETHTOOL_GFLAGS
#define ETHTOOL_GFLAGS       0x00000025 /* Get flags bitmap(ethtool_value) */
#endif
#ifndef ETHTOOL_SFLAGS
#define ETHTOOL_SFLAGS       0x00000026 /* Set flags bitmap(ethtool_value) */
#endif

/* This was introduced in Linux 2.6.25, so it might be missing if we have old
 * headers. */
#ifndef TC_RTAB_SIZE
#define TC_RTAB_SIZE 1024
#endif

/* Linux 2.6.21 introduced struct tpacket_auxdata.
 * Linux 2.6.27 added the tp_vlan_tci member.
 * Linux 3.0 defined TP_STATUS_VLAN_VALID.
 * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined
 * TP_STATUS_VLAN_TPID_VALID.
 *
 * With all this churn it's easiest to unconditionally define a replacement
 * structure that has everything we want.
 */
#ifndef PACKET_AUXDATA
#define PACKET_AUXDATA                  8
#endif
#ifndef TP_STATUS_VLAN_VALID
#define TP_STATUS_VLAN_VALID            (1 << 4)
#endif
#ifndef TP_STATUS_VLAN_TPID_VALID
#define TP_STATUS_VLAN_TPID_VALID       (1 << 6)
#endif
#undef tpacket_auxdata
#define tpacket_auxdata rpl_tpacket_auxdata
struct tpacket_auxdata {
    uint32_t tp_status;
    uint32_t tp_len;
    uint32_t tp_snaplen;
    uint16_t tp_mac;
    uint16_t tp_net;
    uint16_t tp_vlan_tci;
    uint16_t tp_vlan_tpid;
};

/* Linux 2.6.27 introduced ethtool_cmd_speed
 *
 * To avoid revisiting problems reported with using configure to detect
 * compatibility (see report at
 * https://mail.openvswitch.org/pipermail/ovs-dev/2014-October/291521.html)
 * unconditionally replace ethtool_cmd_speed. */
#define ethtool_cmd_speed rpl_ethtool_cmd_speed
static inline uint32_t rpl_ethtool_cmd_speed(const struct ethtool_cmd *ep)
{
        return ep->speed | (ep->speed_hi << 16);
}

/* Linux 2.6.30 introduced supported and advertised flags for
 * 1G base KX, and 10G base KX4, KR and R. */
#ifndef SUPPORTED_1000baseKX_Full
#define SUPPORTED_1000baseKX_Full      (1 << 17)
#define SUPPORTED_10000baseKX4_Full    (1 << 18)
#define SUPPORTED_10000baseKR_Full     (1 << 19)
#define SUPPORTED_10000baseR_FEC       (1 << 20)
#define ADVERTISED_1000baseKX_Full     (1 << 17)
#define ADVERTISED_10000baseKX4_Full   (1 << 18)
#define ADVERTISED_10000baseKR_Full    (1 << 19)
#define ADVERTISED_10000baseR_FEC      (1 << 20)
#endif

/* Linux 3.5 introduced supported and advertised flags for
 * 40G base KR4, CR4, SR4 and LR4. */
#ifndef SUPPORTED_40000baseKR4_Full
#define SUPPORTED_40000baseKR4_Full    (1 << 23)
#define SUPPORTED_40000baseCR4_Full    (1 << 24)
#define SUPPORTED_40000baseSR4_Full    (1 << 25)
#define SUPPORTED_40000baseLR4_Full    (1 << 26)
#define ADVERTISED_40000baseKR4_Full   (1 << 23)
#define ADVERTISED_40000baseCR4_Full   (1 << 24)
#define ADVERTISED_40000baseSR4_Full   (1 << 25)
#define ADVERTISED_40000baseLR4_Full   (1 << 26)
#endif

/* Linux 2.6.35 introduced IFLA_STATS64 and rtnl_link_stats64.
 *
 * Tests for rtnl_link_stats64 don't seem to consistently work, e.g. on
 * 2.6.32-431.29.2.el6.x86_64 (see report at
 * https://mail.openvswitch.org/pipermail/ovs-dev/2014-October/291521.html).
 * Maybe if_link.h is not self-contained on those kernels.  It is easiest to
 * unconditionally define a replacement. */
#ifndef IFLA_STATS64
#define IFLA_STATS64 23
#endif
#define rtnl_link_stats64 rpl_rtnl_link_stats64
struct rtnl_link_stats64 {
    uint64_t rx_packets;
    uint64_t tx_packets;
    uint64_t rx_bytes;
    uint64_t tx_bytes;
    uint64_t rx_errors;
    uint64_t tx_errors;
    uint64_t rx_dropped;
    uint64_t tx_dropped;
    uint64_t multicast;
    uint64_t collisions;

    uint64_t rx_length_errors;
    uint64_t rx_over_errors;
    uint64_t rx_crc_errors;
    uint64_t rx_frame_errors;
    uint64_t rx_fifo_errors;
    uint64_t rx_missed_errors;

    uint64_t tx_aborted_errors;
    uint64_t tx_carrier_errors;
    uint64_t tx_fifo_errors;
    uint64_t tx_heartbeat_errors;
    uint64_t tx_window_errors;

    uint64_t rx_compressed;
    uint64_t tx_compressed;
};

/* Linux 3.19 introduced virtio_types.h.  It might be missing
 * if we are using old kernel. */
#ifndef HAVE_VIRTIO_TYPES
typedef __u16 __bitwise__ __virtio16;
typedef __u32 __bitwise__ __virtio32;
typedef __u64 __bitwise__ __virtio64;
#endif

enum {
    VALID_IFINDEX           = 1 << 0,
    VALID_ETHERADDR         = 1 << 1,
    VALID_IN                = 1 << 2,
    VALID_MTU               = 1 << 3,
    VALID_POLICING          = 1 << 4,
    VALID_VPORT_STAT_ERROR  = 1 << 5,
    VALID_DRVINFO           = 1 << 6,
    VALID_FEATURES          = 1 << 7,
    VALID_NUMA_ID           = 1 << 8,
};

/* Use one for the packet buffer and another for the aux buffer to receive
 * TSO packets. */
#define IOV_STD_SIZE 1
#define IOV_TSO_SIZE 2

enum {
    IOV_PACKET = 0,
    IOV_AUXBUF = 1,
};
\f
struct linux_lag_member {
   uint32_t block_id;
   struct shash_node *node;
};

/* Protects 'lag_shash' and the mutable members of struct linux_lag_member. */
static struct ovs_mutex lag_mutex = OVS_MUTEX_INITIALIZER;

/* All members whose LAG primary interfaces are OVS network devices. */
static struct shash lag_shash OVS_GUARDED_BY(lag_mutex)
    = SHASH_INITIALIZER(&lag_shash);

/* Traffic control. */

/* An instance of a traffic control class.  Always associated with a particular
 * network device.
 *
 * Each TC implementation subclasses this with whatever additional data it
 * needs. */
struct tc {
    const struct tc_ops *ops;
    struct hmap queues;         /* Contains "struct tc_queue"s.
                                 * Read by generic TC layer.
                                 * Written only by TC implementation. */
};

#define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) }

/* One traffic control queue.
 *
 * Each TC implementation subclasses this with whatever additional data it
 * needs. */
struct tc_queue {
    struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
    unsigned int queue_id;      /* OpenFlow queue ID. */
    long long int created;      /* Time queue was created, in msecs. */
};

/* A particular kind of traffic control.  Each implementation generally maps to
 * one particular Linux qdisc class.
 *
 * The functions below return 0 if successful or a positive errno value on
 * failure, except where otherwise noted.  All of them must be provided, except
 * where otherwise noted. */
struct tc_ops {
    /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb".
     * This is null for tc_ops_default and tc_ops_other, for which there are no
     * appropriate values. */
    const char *linux_name;

    /* Name used in OVS database, e.g. "linux-htb".  Must be nonnull. */
    const char *ovs_name;

    /* Number of supported OpenFlow queues, 0 for qdiscs that have no
     * queues.  The queues are numbered 0 through n_queues - 1. */
    unsigned int n_queues;

    /* Called to install this TC class on 'netdev'.  The implementation should
     * make the Netlink calls required to set up 'netdev' with the right qdisc
     * and configure it according to 'details'.  The implementation may assume
     * that the current qdisc is the default; that is, there is no need for it
     * to delete the current qdisc before installing itself.
     *
     * The contents of 'details' should be documented as valid for 'ovs_name'
     * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
     * (which is built as ovs-vswitchd.conf.db(8)).
     *
     * This function must return 0 if and only if it sets 'netdev->tc' to an
     * initialized 'struct tc'.
     *
     * (This function is null for tc_ops_other, which cannot be installed.  For
     * other TC classes it should always be nonnull.) */
    int (*tc_install)(struct netdev *netdev, const struct smap *details);

    /* Called when the netdev code determines (through a Netlink query) that
     * this TC class's qdisc is installed on 'netdev', but we didn't install
     * it ourselves and so don't know any of the details.
     *
     * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for
     * 'netdev'.  The TCA_KIND attribute of 'nlmsg' is 'linux_name'.  The
     * implementation should parse the other attributes of 'nlmsg' as
     * necessary to determine its configuration.  If necessary it should also
     * use Netlink queries to determine the configuration of queues on
     * 'netdev'.
     *
     * This function must return 0 if and only if it sets 'netdev->tc' to an
     * initialized 'struct tc'. */
    int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg);

    /* Destroys the data structures allocated by the implementation as part of
     * 'tc'.  (This includes destroying 'tc->queues' by calling
     * tc_destroy(tc).
     *
     * The implementation should not need to perform any Netlink calls.  If
     * desirable, the caller is responsible for deconfiguring the kernel qdisc.
     * (But it may not be desirable.)
     *
     * This function may be null if 'tc' is trivial. */
    void (*tc_destroy)(struct tc *tc);

    /* Retrieves details of 'netdev->tc' configuration into 'details'.
     *
     * The implementation should not need to perform any Netlink calls, because
     * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
     * cached the configuration.
     *
     * The contents of 'details' should be documented as valid for 'ovs_name'
     * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
     * (which is built as ovs-vswitchd.conf.db(8)).
     *
     * This function may be null if 'tc' is not configurable.
     */
    int (*qdisc_get)(const struct netdev *netdev, struct smap *details);

    /* Reconfigures 'netdev->tc' according to 'details', performing any
     * required Netlink calls to complete the reconfiguration.
     *
     * The contents of 'details' should be documented as valid for 'ovs_name'
     * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
     * (which is built as ovs-vswitchd.conf.db(8)).
     *
     * This function may be null if 'tc' is not configurable.
     */
    int (*qdisc_set)(struct netdev *, const struct smap *details);

    /* Retrieves details of 'queue' on 'netdev->tc' into 'details'.  'queue' is
     * one of the 'struct tc_queue's within 'netdev->tc->queues'.
     *
     * The contents of 'details' should be documented as valid for 'ovs_name'
     * in the "other_config" column in the "Queue" table in
     * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
     *
     * The implementation should not need to perform any Netlink calls, because
     * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
     * cached the queue configuration.
     *
     * This function may be null if 'tc' does not have queues ('n_queues' is
     * 0). */
    int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
                     struct smap *details);

    /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
     * 'details', perfoming any required Netlink calls to complete the
     * reconfiguration.  The caller ensures that 'queue_id' is less than
     * 'n_queues'.
     *
     * The contents of 'details' should be documented as valid for 'ovs_name'
     * in the "other_config" column in the "Queue" table in
     * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
     *
     * This function may be null if 'tc' does not have queues or its queues are
     * not configurable. */
    int (*class_set)(struct netdev *, unsigned int queue_id,
                     const struct smap *details);

    /* Deletes 'queue' from 'netdev->tc'.  'queue' is one of the 'struct
     * tc_queue's within 'netdev->tc->queues'.
     *
     * This function may be null if 'tc' does not have queues or its queues
     * cannot be deleted. */
    int (*class_delete)(struct netdev *, struct tc_queue *queue);

    /* Obtains stats for 'queue' from 'netdev->tc'.  'queue' is one of the
     * 'struct tc_queue's within 'netdev->tc->queues'.
     *
     * On success, initializes '*stats'.
     *
     * This function may be null if 'tc' does not have queues or if it cannot
     * report queue statistics. */
    int (*class_get_stats)(const struct netdev *netdev,
                           const struct tc_queue *queue,
                           struct netdev_queue_stats *stats);

    /* Extracts queue stats from 'nlmsg', which is a response to a
     * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'.
     *
     * This function may be null if 'tc' does not have queues or if it cannot
     * report queue statistics. */
    int (*class_dump_stats)(const struct netdev *netdev,
                            const struct ofpbuf *nlmsg,
                            netdev_dump_queue_stats_cb *cb, void *aux);
};

static void
tc_init(struct tc *tc, const struct tc_ops *ops)
{
    tc->ops = ops;
    hmap_init(&tc->queues);
}

static void
tc_destroy(struct tc *tc)
{
    hmap_destroy(&tc->queues);
}

static const struct tc_ops tc_ops_htb;
static const struct tc_ops tc_ops_hfsc;
static const struct tc_ops tc_ops_codel;
static const struct tc_ops tc_ops_fqcodel;
static const struct tc_ops tc_ops_sfq;
static const struct tc_ops tc_ops_netem;
static const struct tc_ops tc_ops_default;
static const struct tc_ops tc_ops_noop;
static const struct tc_ops tc_ops_other;

static const struct tc_ops *const tcs[] = {
    &tc_ops_htb,                /* Hierarchy token bucket (see tc-htb(8)). */
    &tc_ops_hfsc,               /* Hierarchical fair service curve. */
    &tc_ops_codel,              /* Controlled delay */
    &tc_ops_fqcodel,            /* Fair queue controlled delay */
    &tc_ops_sfq,                /* Stochastic fair queueing */
    &tc_ops_netem,              /* Network Emulator */
    &tc_ops_noop,               /* Non operating qos type. */
    &tc_ops_default,            /* Default qdisc (see tc-pfifo_fast(8)). */
    &tc_ops_other,              /* Some other qdisc. */
    NULL
};

static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks);
static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size);
static unsigned int tc_buffer_per_jiffy(unsigned int rate);
static uint32_t tc_time_to_ticks(uint32_t time);

static struct tcmsg *netdev_linux_tc_make_request(const struct netdev *,
                                                  int type,
                                                  unsigned int flags,
                                                  struct ofpbuf *);
static int tc_add_policer(struct netdev *,
                          uint32_t kbits_rate, uint32_t kbits_burst);

static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
                          struct nlattr **options);
static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id,
                          struct nlattr **options,
                          struct netdev_queue_stats *);
static int tc_query_class(const struct netdev *,
                          unsigned int handle, unsigned int parent,
                          struct ofpbuf **replyp);
static int tc_delete_class(const struct netdev *, unsigned int handle);

static int tc_del_qdisc(struct netdev *netdev);
static int tc_query_qdisc(const struct netdev *netdev);

void
tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate);
static int tc_calc_cell_log(unsigned int mtu);
static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu);
static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes);
\f

/* This is set pretty low because we probably won't learn anything from the
 * additional log messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);

/* Polling miimon status for all ports causes performance degradation when
 * handling a large number of ports. If there are no devices using miimon, then
 * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait().
 *
 * Readers do not depend on this variable synchronizing with the related
 * changes in the device miimon status, so we can use atomic_count. */
static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0);

static int netdev_linux_parse_vnet_hdr(struct dp_packet *b);
static void netdev_linux_prepend_vnet_hdr(struct dp_packet *b, int mtu);
static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
                                   int cmd, const char *cmd_name);
static int get_flags(const struct netdev *, unsigned int *flags);
static int set_flags(const char *, unsigned int flags);
static int update_flags(struct netdev_linux *netdev, enum netdev_flags off,
                        enum netdev_flags on, enum netdev_flags *old_flagsp)
    OVS_REQUIRES(netdev->mutex);
static int get_ifindex(const struct netdev *, int *ifindexp);
static int do_set_addr(struct netdev *netdev,
                       int ioctl_nr, const char *ioctl_name,
                       struct in_addr addr);
static int get_etheraddr(const char *netdev_name, struct eth_addr *ea);
static int set_etheraddr(const char *netdev_name, const struct eth_addr);
static int af_packet_sock(void);
static bool netdev_linux_miimon_enabled(void);
static void netdev_linux_miimon_run(void);
static void netdev_linux_miimon_wait(void);
static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup);

static bool
is_tap_netdev(const struct netdev *netdev)
{
    return netdev_get_class(netdev) == &netdev_tap_class;
}
\f
static int
netdev_linux_netnsid_update__(struct netdev_linux *netdev)
{
    struct dpif_netlink_vport reply;
    struct ofpbuf *buf;
    int error;

    error = dpif_netlink_vport_get(netdev_get_name(&netdev->up), &reply, &buf);
    if (error) {
        if (error == ENOENT) {
            /* Assume it is local if there is no API (e.g. if the openvswitch
             * kernel module is not loaded). */
            netnsid_set_local(&netdev->netnsid);
        } else {
            netnsid_unset(&netdev->netnsid);
        }
        return error;
    }

    netnsid_set(&netdev->netnsid, reply.netnsid);
    ofpbuf_delete(buf);
    return 0;
}

static int
netdev_linux_netnsid_update(struct netdev_linux *netdev)
{
    if (netnsid_is_unset(netdev->netnsid)) {
        if (netdev_get_class(&netdev->up) == &netdev_tap_class) {
            netnsid_set_local(&netdev->netnsid);
        } else {
            return netdev_linux_netnsid_update__(netdev);
        }
    }

    return 0;
}

static bool
netdev_linux_netnsid_is_eq(struct netdev_linux *netdev, int nsid)
{
    netdev_linux_netnsid_update(netdev);
    return netnsid_eq(netdev->netnsid, nsid);
}

static bool
netdev_linux_netnsid_is_remote(struct netdev_linux *netdev)
{
    netdev_linux_netnsid_update(netdev);
    return netnsid_is_remote(netdev->netnsid);
}

static int netdev_linux_update_via_netlink(struct netdev_linux *);
static void netdev_linux_update(struct netdev_linux *netdev, int,
                                const struct rtnetlink_change *)
    OVS_REQUIRES(netdev->mutex);
static void netdev_linux_changed(struct netdev_linux *netdev,
                                 unsigned int ifi_flags, unsigned int mask)
    OVS_REQUIRES(netdev->mutex);

/* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK,
 * RTNLGRP_IPV4_IFADDR and RTNLGRP_IPV6_IFADDR changes, or NULL
 * if no such socket could be created. */
static struct nl_sock *
netdev_linux_notify_sock(void)
{
    static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
    static struct nl_sock *sock;
    unsigned int mcgroups[] = {RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR,
                                RTNLGRP_IPV6_IFADDR, RTNLGRP_IPV6_IFINFO};

    if (ovsthread_once_start(&once)) {
        int error;

        error = nl_sock_create(NETLINK_ROUTE, &sock);
        if (!error) {
            size_t i;

            for (i = 0; i < ARRAY_SIZE(mcgroups); i++) {
                error = nl_sock_join_mcgroup(sock, mcgroups[i]);
                if (error) {
                    nl_sock_destroy(sock);
                    sock = NULL;
                    break;
                }
            }
        }
        nl_sock_listen_all_nsid(sock, true);
        ovsthread_once_done(&once);
    }

    return sock;
}

static bool
netdev_linux_miimon_enabled(void)
{
    return atomic_count_get(&miimon_cnt) > 0;
}

static bool
netdev_linux_kind_is_lag(const char *kind)
{
    if (!strcmp(kind, "bond") || !strcmp(kind, "team")) {
        return true;
    }

    return false;
}

static void
netdev_linux_update_lag(struct rtnetlink_change *change)
    OVS_REQUIRES(lag_mutex)
{
    struct linux_lag_member *lag;

    if (change->sub && netdev_linux_kind_is_lag(change->sub)) {
        lag = shash_find_data(&lag_shash, change->ifname);

        if (!lag) {
            struct netdev *master_netdev;
            char master_name[IFNAMSIZ];
            uint32_t block_id;
            int error = 0;

            if_indextoname(change->master_ifindex, master_name);
            master_netdev = netdev_from_name(master_name);
            if (!master_netdev) {
                return;
            }

            if (is_netdev_linux_class(master_netdev->netdev_class)) {
                block_id = netdev_get_block_id(master_netdev);
                if (!block_id) {
                    netdev_close(master_netdev);
                    return;
                }

                lag = xmalloc(sizeof *lag);
                lag->block_id = block_id;
                lag->node = shash_add(&lag_shash, change->ifname, lag);

                /* delete ingress block in case it exists */
                tc_add_del_qdisc(change->if_index, false, 0, TC_INGRESS);
                /* LAG master is linux netdev so add member to same block. */
                error = tc_add_del_qdisc(change->if_index, true, block_id,
                                         TC_INGRESS);
                if (error) {
                    VLOG_WARN("failed to bind LAG member %s to "
                              "primary's block", change->ifname);
                    shash_delete(&lag_shash, lag->node);
                    free(lag);
                }
            }

            netdev_close(master_netdev);
        }
    } else if (change->master_ifindex == 0) {
        /* Check if this was a lag member that has been removed. */
        lag = shash_find_data(&lag_shash, change->ifname);

        if (lag) {
            tc_add_del_qdisc(change->if_index, false, lag->block_id,
                             TC_INGRESS);
            shash_delete(&lag_shash, lag->node);
            free(lag);
        }
    }
}

void
netdev_linux_run(const struct netdev_class *netdev_class OVS_UNUSED)
{
    struct nl_sock *sock;
    int error;

    if (netdev_linux_miimon_enabled()) {
        netdev_linux_miimon_run();
    }

    sock = netdev_linux_notify_sock();
    if (!sock) {
        return;
    }

    do {
        uint64_t buf_stub[4096 / 8];
        int nsid;
        struct ofpbuf buf;

        ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
        error = nl_sock_recv(sock, &buf, &nsid, false);
        if (!error) {
            struct rtnetlink_change change;

            if (rtnetlink_parse(&buf, &change)) {
                struct netdev *netdev_ = NULL;
                char dev_name[IFNAMSIZ];

                if (!change.ifname) {
                     change.ifname = if_indextoname(change.if_index, dev_name);
                }

                if (change.ifname) {
                    netdev_ = netdev_from_name(change.ifname);
                }
                if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) {
                    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

                    ovs_mutex_lock(&netdev->mutex);
                    netdev_linux_update(netdev, nsid, &change);
                    ovs_mutex_unlock(&netdev->mutex);
                }

                if (change.ifname &&
                    rtnetlink_type_is_rtnlgrp_link(change.nlmsg_type)) {

                    /* Need to try updating the LAG information. */
                    ovs_mutex_lock(&lag_mutex);
                    netdev_linux_update_lag(&change);
                    ovs_mutex_unlock(&lag_mutex);
                }
                netdev_close(netdev_);
            }
        } else if (error == ENOBUFS) {
            struct shash device_shash;
            struct shash_node *node;

            nl_sock_drain(sock);

            shash_init(&device_shash);
            netdev_get_devices(&netdev_linux_class, &device_shash);
            SHASH_FOR_EACH (node, &device_shash) {
                struct netdev *netdev_ = node->data;
                struct netdev_linux *netdev = netdev_linux_cast(netdev_);
                unsigned int flags;

                ovs_mutex_lock(&netdev->mutex);
                get_flags(netdev_, &flags);
                netdev_linux_changed(netdev, flags, 0);
                ovs_mutex_unlock(&netdev->mutex);

                netdev_close(netdev_);
            }
            shash_destroy(&device_shash);
        } else if (error != EAGAIN) {
            static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 5);
            VLOG_WARN_RL(&rll, "error reading or parsing netlink (%s)",
                         ovs_strerror(error));
        }
        ofpbuf_uninit(&buf);
    } while (!error);
}

static void
netdev_linux_wait(const struct netdev_class *netdev_class OVS_UNUSED)
{
    struct nl_sock *sock;

    if (netdev_linux_miimon_enabled()) {
        netdev_linux_miimon_wait();
    }
    sock = netdev_linux_notify_sock();
    if (sock) {
        nl_sock_wait(sock, POLLIN);
    }
}

static void
netdev_linux_changed(struct netdev_linux *dev,
                     unsigned int ifi_flags, unsigned int mask)
    OVS_REQUIRES(dev->mutex)
{
    netdev_change_seq_changed(&dev->up);

    if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
        dev->carrier_resets++;
    }
    dev->ifi_flags = ifi_flags;

    dev->cache_valid &= mask;
    if (!(mask & VALID_IN)) {
        netdev_get_addrs_list_flush();
    }
}

static void
netdev_linux_update__(struct netdev_linux *dev,
                      const struct rtnetlink_change *change)
    OVS_REQUIRES(dev->mutex)
{
    if (rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)) {
        if (change->nlmsg_type == RTM_NEWLINK) {
            /* Keep drv-info, ip addresses, and NUMA id. */
            netdev_linux_changed(dev, change->ifi_flags,
                                 VALID_DRVINFO | VALID_IN | VALID_NUMA_ID);

            /* Update netdev from rtnl-change msg. */
            if (change->mtu) {
                dev->mtu = change->mtu;
                dev->cache_valid |= VALID_MTU;
                dev->netdev_mtu_error = 0;
            }

            if (!eth_addr_is_zero(change->mac)) {
                dev->etheraddr = change->mac;
                dev->cache_valid |= VALID_ETHERADDR;
                dev->ether_addr_error = 0;

                /* The mac addr has been changed, report it now. */
                rtnetlink_report_link();
            }

            if (change->primary && netdev_linux_kind_is_lag(change->primary)) {
                dev->is_lag_master = true;
            }

            dev->ifindex = change->if_index;
            dev->cache_valid |= VALID_IFINDEX;
            dev->get_ifindex_error = 0;
            dev->present = true;
        } else {
            /* FIXME */
            netdev_linux_changed(dev, change->ifi_flags, 0);
            dev->present = false;
            netnsid_unset(&dev->netnsid);
        }
    } else if (rtnetlink_type_is_rtnlgrp_addr(change->nlmsg_type)) {
        /* Invalidates in4, in6. */
        netdev_linux_changed(dev, dev->ifi_flags, ~VALID_IN);
    } else {
        OVS_NOT_REACHED();
    }
}

static void
netdev_linux_update(struct netdev_linux *dev, int nsid,
                    const struct rtnetlink_change *change)
    OVS_REQUIRES(dev->mutex)
{
    if (netdev_linux_netnsid_is_eq(dev, nsid)) {
        netdev_linux_update__(dev, change);
    }
}

static struct netdev *
netdev_linux_alloc(void)
{
    struct netdev_linux *netdev = xzalloc(sizeof *netdev);
    return &netdev->up;
}

static int
netdev_linux_common_construct(struct netdev *netdev_)
{
    /* Prevent any attempt to create (or open) a network device named "default"
     * or "all".  These device names are effectively reserved on Linux because
     * /proc/sys/net/ipv4/conf/ always contains directories by these names.  By
     * itself this wouldn't call for any special treatment, but in practice if
     * a program tries to create devices with these names, it causes the kernel
     * to fire a "new device" notification event even though creation failed,
     * and in turn that causes OVS to wake up and try to create them again,
     * which ends up as a 100% CPU loop. */
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    const char *name = netdev_->name;
    if (!strcmp(name, "default") || !strcmp(name, "all")) {
        static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 1);
        VLOG_WARN_RL(&rll, "%s: Linux forbids network device with this name",
                     name);
        return EINVAL;
    }

    /* The device could be in the same network namespace or in another one. */
    netnsid_unset(&netdev->netnsid);
    ovs_mutex_init(&netdev->mutex);

    if (userspace_tso_enabled()) {
        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_TCP_TSO;
        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_TCP_CKSUM;
        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_UDP_CKSUM;
        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_SCTP_CKSUM;
        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_IPV4_CKSUM;
    }

    return 0;
}

/* Creates system and internal devices. */
int
netdev_linux_construct(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error = netdev_linux_common_construct(netdev_);
    if (error) {
        return error;
    }

    error = get_flags(&netdev->up, &netdev->ifi_flags);
    if (error == ENODEV) {
        if (netdev->up.netdev_class != &netdev_internal_class) {
            /* The device does not exist, so don't allow it to be opened. */
            return ENODEV;
        } else {
            /* "Internal" netdevs have to be created as netdev objects before
             * they exist in the kernel, because creating them in the kernel
             * happens by passing a netdev object to dpif_port_add().
             * Therefore, ignore the error. */
        }
    }

    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_construct_tap(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    static const char tap_dev[] = "/dev/net/tun";
    const char *name = netdev_->name;
    struct ifreq ifr;

    int error = netdev_linux_common_construct(netdev_);
    if (error) {
        return error;
    }

    /* 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, ovs_strerror(error));
        return error;
    }

    /* Create tap device. */
    get_flags(&netdev->up, &netdev->ifi_flags);
    ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
    if (userspace_tso_enabled()) {
        ifr.ifr_flags |= IFF_VNET_HDR;
    }

    ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
    if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
        VLOG_WARN("%s: creating tap device failed: %s", name,
                  ovs_strerror(errno));
        error = errno;
        goto error_close;
    }

    /* Make non-blocking. */
    error = set_nonblocking(netdev->tap_fd);
    if (error) {
        goto error_close;
    }

    if (ioctl(netdev->tap_fd, TUNSETPERSIST, 1)) {
        VLOG_WARN("%s: creating tap device failed (persist): %s", name,
                  ovs_strerror(errno));
        error = errno;
        goto error_close;
    }

    if (userspace_tso_enabled()) {
        /* Old kernels don't support TUNSETOFFLOAD. If TUNSETOFFLOAD is
         * available, it will return EINVAL when a flag is unknown.
         * Therefore, try enabling offload with no flags to check
         * if TUNSETOFFLOAD support is available or not. */
        if (ioctl(netdev->tap_fd, TUNSETOFFLOAD, 0) == 0 || errno != EINVAL) {
            unsigned long oflags = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6;

            if (ioctl(netdev->tap_fd, TUNSETOFFLOAD, oflags) == -1) {
                VLOG_WARN("%s: enabling tap offloading failed: %s", name,
                          ovs_strerror(errno));
                error = errno;
                goto error_close;
            }
        }
    }

    netdev->present = true;
    return 0;

error_close:
    close(netdev->tap_fd);
    return error;
}

static void
netdev_linux_destruct(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    if (netdev->tc && netdev->tc->ops->tc_destroy) {
        netdev->tc->ops->tc_destroy(netdev->tc);
    }

    if (netdev_get_class(netdev_) == &netdev_tap_class
        && netdev->tap_fd >= 0)
    {
        ioctl(netdev->tap_fd, TUNSETPERSIST, 0);
        close(netdev->tap_fd);
    }

    if (netdev->miimon_interval > 0) {
        atomic_count_dec(&miimon_cnt);
    }

    ovs_mutex_destroy(&netdev->mutex);
}

static void
netdev_linux_dealloc(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    free(netdev);
}

static struct netdev_rxq *
netdev_linux_rxq_alloc(void)
{
    struct netdev_rxq_linux *rx = xzalloc(sizeof *rx);
    return &rx->up;
}

static int
netdev_linux_rxq_construct(struct netdev_rxq *rxq_)
{
    struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
    struct netdev *netdev_ = rx->up.netdev;
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    rx->is_tap = is_tap_netdev(netdev_);
    if (rx->is_tap) {
        rx->fd = netdev->tap_fd;
    } else {
        struct sockaddr_ll sll;
        int ifindex, val;
        /* Result of tcpdump -dd inbound */
        static const struct sock_filter filt[] = {
            { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */
            { 0x15, 0, 1, 0x00000004 }, /* jeq #4     jt 2  jf 3 */
            { 0x6, 0, 0, 0x00000000 },  /* ret #0 */
            { 0x6, 0, 0, 0x0000ffff }   /* ret #65535 */
        };
        static const struct sock_fprog fprog = {
            ARRAY_SIZE(filt), (struct sock_filter *) filt
        };

        /* Create file descriptor. */
        rx->fd = socket(PF_PACKET, SOCK_RAW, 0);
        if (rx->fd < 0) {
            error = errno;
            VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error));
            goto error;
        }

        val = 1;
        if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) {
            error = errno;
            VLOG_ERR("%s: failed to mark socket for auxdata (%s)",
                     netdev_get_name(netdev_), ovs_strerror(error));
            goto error;
        }

        if (userspace_tso_enabled()
            && setsockopt(rx->fd, SOL_PACKET, PACKET_VNET_HDR, &val,
                          sizeof val)) {
            error = errno;
            VLOG_ERR("%s: failed to enable vnet hdr in txq raw socket: %s",
                     netdev_get_name(netdev_), ovs_strerror(errno));
            goto error;
        }

        /* Set non-blocking mode. */
        error = set_nonblocking(rx->fd);
        if (error) {
            goto error;
        }

        /* Get ethernet device index. */
        error = get_ifindex(&netdev->up, &ifindex);
        if (error) {
            goto error;
        }

        /* Bind to specific ethernet device. */
        memset(&sll, 0, sizeof sll);
        sll.sll_family = AF_PACKET;
        sll.sll_ifindex = ifindex;
        sll.sll_protocol = htons(ETH_P_ALL);
        if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
            error = errno;
            VLOG_ERR("%s: failed to bind raw socket (%s)",
                     netdev_get_name(netdev_), ovs_strerror(error));
            goto error;
        }

        /* Filter for only inbound packets. */
        error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog,
                           sizeof fprog);
        if (error) {
            error = errno;
            VLOG_ERR("%s: failed to attach filter (%s)",
                     netdev_get_name(netdev_), ovs_strerror(error));
            goto error;
        }
    }
    ovs_mutex_unlock(&netdev->mutex);

    return 0;

error:
    if (rx->fd >= 0) {
        close(rx->fd);
    }
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static void
netdev_linux_rxq_destruct(struct netdev_rxq *rxq_)
{
    struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
    int i;

    if (!rx->is_tap) {
        close(rx->fd);
    }

    for (i = 0; i < NETDEV_MAX_BURST; i++) {
        dp_packet_delete(rx->aux_bufs[i]);
    }
}

static void
netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_)
{
    struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);

    free(rx);
}

static ovs_be16
auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux, bool double_tagged)
{
    if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) {
        return htons(aux->tp_vlan_tpid);
    } else if (double_tagged) {
        return htons(ETH_TYPE_VLAN_8021AD);
    } else {
        return htons(ETH_TYPE_VLAN_8021Q);
    }
}

static bool
auxdata_has_vlan_tci(const struct tpacket_auxdata *aux)
{
    return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID;
}

/*
 * Receive packets from raw socket in batch process for better performance,
 * it can receive NETDEV_MAX_BURST packets at most once, the received
 * packets are added into *batch. The return value is 0 or errno.
 *
 * It also used recvmmsg to reduce multiple syscalls overhead;
 */
static int
netdev_linux_batch_rxq_recv_sock(struct netdev_rxq_linux *rx, int mtu,
                                 struct dp_packet_batch *batch)
{
    int iovlen;
    size_t std_len;
    ssize_t retval;
    int virtio_net_hdr_size;
    struct iovec iovs[NETDEV_MAX_BURST][IOV_TSO_SIZE];
    struct cmsghdr *cmsg;
    union {
        struct cmsghdr cmsg;
        char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
    } cmsg_buffers[NETDEV_MAX_BURST];
    struct mmsghdr mmsgs[NETDEV_MAX_BURST];
    struct dp_packet *buffers[NETDEV_MAX_BURST];
    int i;

    if (userspace_tso_enabled()) {
        /* Use the buffer from the allocated packet below to receive MTU
         * sized packets and an aux_buf for extra TSO data. */
        iovlen = IOV_TSO_SIZE;
        virtio_net_hdr_size = sizeof(struct virtio_net_hdr);
    } else {
        /* Use only the buffer from the allocated packet. */
        iovlen = IOV_STD_SIZE;
        virtio_net_hdr_size = 0;
    }

    /* The length here needs to be accounted in the same way when the
     * aux_buf is allocated so that it can be prepended to TSO buffer. */
    std_len = virtio_net_hdr_size + VLAN_ETH_HEADER_LEN + mtu;
    for (i = 0; i < NETDEV_MAX_BURST; i++) {
         buffers[i] = dp_packet_new_with_headroom(std_len, DP_NETDEV_HEADROOM);
         iovs[i][IOV_PACKET].iov_base = dp_packet_data(buffers[i]);
         iovs[i][IOV_PACKET].iov_len = std_len;
         if (iovlen == IOV_TSO_SIZE) {
             iovs[i][IOV_AUXBUF].iov_base = dp_packet_data(rx->aux_bufs[i]);
             iovs[i][IOV_AUXBUF].iov_len = dp_packet_tailroom(rx->aux_bufs[i]);
         }

         mmsgs[i].msg_hdr.msg_name = NULL;
         mmsgs[i].msg_hdr.msg_namelen = 0;
         mmsgs[i].msg_hdr.msg_iov = iovs[i];
         mmsgs[i].msg_hdr.msg_iovlen = iovlen;
         mmsgs[i].msg_hdr.msg_control = &cmsg_buffers[i];
         mmsgs[i].msg_hdr.msg_controllen = sizeof cmsg_buffers[i];
         mmsgs[i].msg_hdr.msg_flags = 0;
    }

    do {
        retval = recvmmsg(rx->fd, mmsgs, NETDEV_MAX_BURST, MSG_TRUNC, NULL);
    } while (retval < 0 && errno == EINTR);

    if (retval < 0) {
        retval = errno;
        for (i = 0; i < NETDEV_MAX_BURST; i++) {
            dp_packet_delete(buffers[i]);
        }

        return retval;
    }

    for (i = 0; i < retval; i++) {
        struct dp_packet *pkt;

        if (mmsgs[i].msg_len < ETH_HEADER_LEN) {
            struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up);
            struct netdev_linux *netdev = netdev_linux_cast(netdev_);

            dp_packet_delete(buffers[i]);
            netdev->rx_dropped += 1;
            VLOG_WARN_RL(&rl, "%s: Dropped packet: less than ether hdr size",
                         netdev_get_name(netdev_));
            continue;
        }

        if (mmsgs[i].msg_len > std_len) {
            /* Build a single linear TSO packet by prepending the data from
             * std_len buffer to the aux_buf. */
            pkt = rx->aux_bufs[i];
            dp_packet_set_size(pkt, mmsgs[i].msg_len - std_len);
            dp_packet_push(pkt, dp_packet_data(buffers[i]), std_len);
            /* The headroom should be the same in buffers[i], pkt and
             * DP_NETDEV_HEADROOM. */
            dp_packet_resize(pkt, DP_NETDEV_HEADROOM, 0);
            dp_packet_delete(buffers[i]);
            rx->aux_bufs[i] = NULL;
         } else {
            dp_packet_set_size(buffers[i], mmsgs[i].msg_len);
            pkt = buffers[i];
         }

        if (virtio_net_hdr_size && netdev_linux_parse_vnet_hdr(pkt)) {
            struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up);
            struct netdev_linux *netdev = netdev_linux_cast(netdev_);

            /* Unexpected error situation: the virtio header is not present
             * or corrupted. Drop the packet but continue in case next ones
             * are correct. */
            dp_packet_delete(pkt);
            netdev->rx_dropped += 1;
            VLOG_WARN_RL(&rl, "%s: Dropped packet: Invalid virtio net header",
                         netdev_get_name(netdev_));
            continue;
        }

        for (cmsg = CMSG_FIRSTHDR(&mmsgs[i].msg_hdr); cmsg;
                 cmsg = CMSG_NXTHDR(&mmsgs[i].msg_hdr, cmsg)) {
            const struct tpacket_auxdata *aux;

            if (cmsg->cmsg_level != SOL_PACKET
                || cmsg->cmsg_type != PACKET_AUXDATA
                || cmsg->cmsg_len <
                       CMSG_LEN(sizeof(struct tpacket_auxdata))) {
                continue;
            }

            aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg));
            if (auxdata_has_vlan_tci(aux)) {
                struct eth_header *eth;
                bool double_tagged;

                eth = dp_packet_data(pkt);
                double_tagged = eth->eth_type == htons(ETH_TYPE_VLAN_8021Q);

                eth_push_vlan(pkt,
                              auxdata_to_vlan_tpid(aux, double_tagged),
                              htons(aux->tp_vlan_tci));
                break;
            }
        }
        dp_packet_batch_add(batch, pkt);
    }

    /* Delete unused buffers. */
    for (; i < NETDEV_MAX_BURST; i++) {
        dp_packet_delete(buffers[i]);
    }

    return 0;
}

/*
 * Receive packets from tap by batch process for better performance,
 * it can receive NETDEV_MAX_BURST packets at most once, the received
 * packets are added into *batch. The return value is 0 or errno.
 */
static int
netdev_linux_batch_rxq_recv_tap(struct netdev_rxq_linux *rx, int mtu,
                                struct dp_packet_batch *batch)
{
    int virtio_net_hdr_size;
    ssize_t retval;
    size_t std_len;
    int iovlen;
    int i;

    if (userspace_tso_enabled()) {
        /* Use the buffer from the allocated packet below to receive MTU
         * sized packets and an aux_buf for extra TSO data. */
        iovlen = IOV_TSO_SIZE;
        virtio_net_hdr_size = sizeof(struct virtio_net_hdr);
    } else {
        /* Use only the buffer from the allocated packet. */
        iovlen = IOV_STD_SIZE;
        virtio_net_hdr_size = 0;
    }

    /* The length here needs to be accounted in the same way when the
     * aux_buf is allocated so that it can be prepended to TSO buffer. */
    std_len = virtio_net_hdr_size + VLAN_ETH_HEADER_LEN + mtu;
    for (i = 0; i < NETDEV_MAX_BURST; i++) {
        struct dp_packet *buffer;
        struct dp_packet *pkt;
        struct iovec iov[IOV_TSO_SIZE];

        /* Assume Ethernet port. No need to set packet_type. */
        buffer = dp_packet_new_with_headroom(std_len, DP_NETDEV_HEADROOM);
        iov[IOV_PACKET].iov_base = dp_packet_data(buffer);
        iov[IOV_PACKET].iov_len = std_len;
        if (iovlen == IOV_TSO_SIZE) {
            iov[IOV_AUXBUF].iov_base = dp_packet_data(rx->aux_bufs[i]);
            iov[IOV_AUXBUF].iov_len = dp_packet_tailroom(rx->aux_bufs[i]);
        }

        do {
            retval = readv(rx->fd, iov, iovlen);
        } while (retval < 0 && errno == EINTR);

        if (retval < 0) {
            dp_packet_delete(buffer);
            break;
        }

        if (retval > std_len) {
            /* Build a single linear TSO packet by prepending the data from
             * std_len buffer to the aux_buf. */
            pkt = rx->aux_bufs[i];
            dp_packet_set_size(pkt, retval - std_len);
            dp_packet_push(pkt, dp_packet_data(buffer), std_len);
            /* The headroom should be the same in buffers[i], pkt and
             * DP_NETDEV_HEADROOM. */
            dp_packet_resize(pkt, DP_NETDEV_HEADROOM, 0);
            dp_packet_delete(buffer);
            rx->aux_bufs[i] = NULL;
        } else {
            dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
            pkt = buffer;
        }

        if (virtio_net_hdr_size && netdev_linux_parse_vnet_hdr(pkt)) {
            struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up);
            struct netdev_linux *netdev = netdev_linux_cast(netdev_);

            /* Unexpected error situation: the virtio header is not present
             * or corrupted. Drop the packet but continue in case next ones
             * are correct. */
            dp_packet_delete(pkt);
            netdev->rx_dropped += 1;
            VLOG_WARN_RL(&rl, "%s: Dropped packet: Invalid virtio net header",
                         netdev_get_name(netdev_));
            continue;
        }

        dp_packet_batch_add(batch, pkt);
    }

    if ((i == 0) && (retval < 0)) {
        return errno;
    }

    return 0;
}

static int
netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch,
                      int *qfill)
{
    struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
    struct netdev *netdev = rx->up.netdev;
    ssize_t retval;
    int mtu;

    if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) {
        mtu = ETH_PAYLOAD_MAX;
    }

    if (userspace_tso_enabled()) {
        /* Allocate TSO packets. The packet has enough headroom to store
         * a full non-TSO packet. When a TSO packet is received, the data
         * from non-TSO buffer (std_len) is prepended to the TSO packet
         * (aux_buf). */
        size_t std_len = sizeof(struct virtio_net_hdr) + VLAN_ETH_HEADER_LEN
                         + DP_NETDEV_HEADROOM + mtu;
        size_t data_len = LINUX_RXQ_TSO_MAX_LEN - std_len;
        for (int i = 0; i < NETDEV_MAX_BURST; i++) {
            if (rx->aux_bufs[i]) {
                continue;
            }

            rx->aux_bufs[i] = dp_packet_new_with_headroom(data_len, std_len);
        }
    }

    dp_packet_batch_init(batch);
    retval = (rx->is_tap
              ? netdev_linux_batch_rxq_recv_tap(rx, mtu, batch)
              : netdev_linux_batch_rxq_recv_sock(rx, mtu, batch));

    if (retval) {
        if (retval != EAGAIN && retval != EMSGSIZE) {
            VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
                         netdev_rxq_get_name(rxq_), ovs_strerror(errno));
        }
    }

    if (qfill) {
        *qfill = -ENOTSUP;
    }

    return retval;
}

static void
netdev_linux_rxq_wait(struct netdev_rxq *rxq_)
{
    struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
    poll_fd_wait(rx->fd, POLLIN);
}

static int
netdev_linux_rxq_drain(struct netdev_rxq *rxq_)
{
    struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
    if (rx->is_tap) {
        struct ifreq ifr;
        int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr,
                                        SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
        if (error) {
            return error;
        }
        drain_fd(rx->fd, ifr.ifr_qlen);
        return 0;
    } else {
        return drain_rcvbuf(rx->fd);
    }
}

static int
netdev_linux_sock_batch_send(int sock, int ifindex, bool tso, int mtu,
                             struct dp_packet_batch *batch)
{
    const size_t size = dp_packet_batch_size(batch);
    /* We don't bother setting most fields in sockaddr_ll because the
     * kernel ignores them for SOCK_RAW. */
    struct sockaddr_ll sll = { .sll_family = AF_PACKET,
                               .sll_ifindex = ifindex };

    struct mmsghdr *mmsg = xmalloc(sizeof(*mmsg) * size);
    struct iovec *iov = xmalloc(sizeof(*iov) * size);

    struct dp_packet *packet;
    DP_PACKET_BATCH_FOR_EACH (i, packet, batch) {
        if (tso) {
            netdev_linux_prepend_vnet_hdr(packet, mtu);
        }

        iov[i].iov_base = dp_packet_data(packet);
        iov[i].iov_len = dp_packet_size(packet);
        mmsg[i].msg_hdr = (struct msghdr) { .msg_name = &sll,
                                            .msg_namelen = sizeof sll,
                                            .msg_iov = &iov[i],
                                            .msg_iovlen = 1 };
    }

    int error = 0;
    for (uint32_t ofs = 0; ofs < size; ) {
        ssize_t retval;
        do {
            retval = sendmmsg(sock, mmsg + ofs, size - ofs, 0);
            error = retval < 0 ? errno : 0;
        } while (error == EINTR);
        if (error) {
            break;
        }
        ofs += retval;
    }

    free(mmsg);
    free(iov);
    return error;
}

/* Use the tap fd to send 'batch' to tap device 'netdev'.  Using the tap fd is
 * essential, because packets sent to a tap device with an AF_PACKET socket
 * will loop back to be *received* again on the tap device.  This doesn't occur
 * on other interface types because we attach a socket filter to the rx
 * socket. */
static int
netdev_linux_tap_batch_send(struct netdev *netdev_, bool tso, int mtu,
                            struct dp_packet_batch *batch)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct dp_packet *packet;

    /* The Linux tap driver returns EIO if the device is not up,
     * so if the device is not up, don't waste time sending it.
     * However, if the device is in another network namespace
     * then OVS can't retrieve the state. In that case, send the
     * packets anyway. */
    if (netdev->present && !(netdev->ifi_flags & IFF_UP)) {
        netdev->tx_dropped += dp_packet_batch_size(batch);
        return 0;
    }

    DP_PACKET_BATCH_FOR_EACH (i, packet, batch) {
        size_t size;
        ssize_t retval;
        int error;

        if (tso) {
            netdev_linux_prepend_vnet_hdr(packet, mtu);
        }

        size = dp_packet_size(packet);
        do {
            retval = write(netdev->tap_fd, dp_packet_data(packet), size);
            error = retval < 0 ? errno : 0;
        } while (error == EINTR);

        if (error) {
            /* The Linux tap driver returns EIO if the device is not up.  From
             * the OVS side this is not an error, so we ignore it; otherwise,
             * return the erro. */
            if (error != EIO) {
                return error;
            }
        } else if (retval != size) {
            VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" "
                         "bytes of %"PRIuSIZE") on %s",
                         retval, size, netdev_get_name(netdev_));
            return EMSGSIZE;
        }
    }
    return 0;
}

static int
netdev_linux_get_numa_id__(struct netdev_linux *netdev)
    OVS_REQUIRES(netdev->mutex)
{
    char *numa_node_path;
    const char *name;
    int node_id;
    FILE *stream;

    if (netdev->cache_valid & VALID_NUMA_ID) {
        return netdev->numa_id;
    }

    netdev->numa_id = 0;
    netdev->cache_valid |= VALID_NUMA_ID;

    if (ovs_numa_get_n_numas() < 2) {
        /* No need to check on system with a single NUMA node. */
        return 0;
    }

    name = netdev_get_name(&netdev->up);
    if (strpbrk(name, "/\\")) {
        VLOG_ERR_RL(&rl, "\"%s\" is not a valid name for a port. "
                    "A valid name must not include '/' or '\\'."
                    "Using numa_id 0", name);
        return 0;
    }

    numa_node_path = xasprintf("/sys/class/net/%s/device/numa_node", name);

    stream = fopen(numa_node_path, "r");
    if (!stream) {
        /* Virtual device does not have this info. */
        VLOG_INFO_RL(&rl, "%s: Can't open '%s': %s, using numa_id 0",
                     name, numa_node_path, ovs_strerror(errno));
        free(numa_node_path);
        return 0;
    }

    if (fscanf(stream, "%d", &node_id) != 1
        || !ovs_numa_numa_id_is_valid(node_id))  {
        VLOG_WARN_RL(&rl, "%s: Can't detect NUMA node, using numa_id 0", name);
        node_id = 0;
    }

    netdev->numa_id = node_id;
    fclose(stream);
    free(numa_node_path);
    return node_id;
}

static int OVS_UNUSED
netdev_linux_get_numa_id(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int numa_id;

    ovs_mutex_lock(&netdev->mutex);
    numa_id = netdev_linux_get_numa_id__(netdev);
    ovs_mutex_unlock(&netdev->mutex);

    return numa_id;
}

/* Sends 'batch' on 'netdev'.  Returns 0 if successful, otherwise a positive
 * errno value.  Returns EAGAIN without blocking if the packet cannot be queued
 * immediately.  Returns EMSGSIZE if a partial packet was transmitted or if
 * the packet is too big or too small to transmit on the device.
 *
 * The kernel maintains a packet transmission queue, so the caller is not
 * expected to do additional queuing of packets. */
static int
netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED,
                  struct dp_packet_batch *batch,
                  bool concurrent_txq OVS_UNUSED)
{
    bool tso = userspace_tso_enabled();
    int mtu = ETH_PAYLOAD_MAX;
    int error = 0;
    int sock = 0;

    if (tso) {
        netdev_linux_get_mtu__(netdev_linux_cast(netdev_), &mtu);
    }

    if (!is_tap_netdev(netdev_)) {
        if (netdev_linux_netnsid_is_remote(netdev_linux_cast(netdev_))) {
            error = EOPNOTSUPP;
            goto free_batch;
        }

        sock = af_packet_sock();
        if (sock < 0) {
            error = -sock;
            goto free_batch;
        }

        int ifindex = netdev_get_ifindex(netdev_);
        if (ifindex < 0) {
            error = -ifindex;
            goto free_batch;
        }

        error = netdev_linux_sock_batch_send(sock, ifindex, tso, mtu, batch);
    } else {
        error = netdev_linux_tap_batch_send(netdev_, tso, mtu, batch);
    }
    if (error) {
        if (error == ENOBUFS) {
            /* The Linux AF_PACKET implementation never blocks waiting
             * for room for packets, instead returning ENOBUFS.
             * Translate this into EAGAIN for the caller. */
            error = EAGAIN;
        } else {
            VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
                         netdev_get_name(netdev_), ovs_strerror(error));
        }
    }

free_batch:
    dp_packet_delete_batch(batch, true);
    return error;
}

/* Registers with the poll loop to wake up from the next call to poll_block()
 * when the packet transmission queue has sufficient room to transmit a packet
 * with netdev_send().
 *
 * The kernel maintains a packet transmission queue, so the client is not
 * expected to do additional queuing of packets.  Thus, this function is
 * unlikely to ever be used.  It is included for completeness. */
static void
netdev_linux_send_wait(struct netdev *netdev, int qid OVS_UNUSED)
{
    if (is_tap_netdev(netdev)) {
        /* TAP device always accepts packets.*/
        poll_immediate_wake();
    }
}

/* Attempts to set 'netdev''s MAC address to 'mac'.  Returns 0 if successful,
 * otherwise a positive errno value. */
static int
netdev_linux_set_etheraddr(struct netdev *netdev_, const struct eth_addr mac)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    enum netdev_flags old_flags = 0;
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    if (netdev->cache_valid & VALID_ETHERADDR) {
        error = netdev->ether_addr_error;
        if (error || eth_addr_equals(netdev->etheraddr, mac)) {
            goto exit;
        }
        netdev->cache_valid &= ~VALID_ETHERADDR;
    }

    /* Tap devices must be brought down before setting the address. */
    if (is_tap_netdev(netdev_)) {
        update_flags(netdev, NETDEV_UP, 0, &old_flags);
    }
    error = set_etheraddr(netdev_get_name(netdev_), mac);
    if (!error || error == ENODEV) {
        netdev->ether_addr_error = error;
        netdev->cache_valid |= VALID_ETHERADDR;
        if (!error) {
            netdev->etheraddr = mac;
        }
    }

    if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) {
        update_flags(netdev, 0, NETDEV_UP, &old_flags);
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

/* Copies 'netdev''s MAC address to 'mac' which is passed as param. */
static int
netdev_linux_get_etheraddr(const struct netdev *netdev_, struct eth_addr *mac)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (!(netdev->cache_valid & VALID_ETHERADDR)) {
        netdev_linux_update_via_netlink(netdev);
    }

    if (!(netdev->cache_valid & VALID_ETHERADDR)) {
        /* Fall back to ioctl if netlink fails */
        netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_),
                                                 &netdev->etheraddr);
        netdev->cache_valid |= VALID_ETHERADDR;
    }

    error = netdev->ether_addr_error;
    if (!error) {
        *mac = netdev->etheraddr;
    }
    ovs_mutex_unlock(&netdev->mutex);

    return error;
}

static int
netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup)
{
    int error;

    if (!(netdev->cache_valid & VALID_MTU)) {
        netdev_linux_update_via_netlink(netdev);
    }

    if (!(netdev->cache_valid & VALID_MTU)) {
        /* Fall back to ioctl if netlink fails */
        struct ifreq ifr;

        netdev->netdev_mtu_error = af_inet_ifreq_ioctl(
            netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU");
        netdev->mtu = ifr.ifr_mtu;
        netdev->cache_valid |= VALID_MTU;
    }

    error = netdev->netdev_mtu_error;
    if (!error) {
        *mtup = netdev->mtu;
    }

    return error;
}

/* Returns the maximum size of transmitted (and received) packets on 'netdev',
 * in bytes, not including the hardware header; thus, this is typically 1500
 * bytes for Ethernet devices. */
static int
netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    error = netdev_linux_get_mtu__(netdev, mtup);
    ovs_mutex_unlock(&netdev->mutex);

    return error;
}

/* Sets the maximum size of transmitted (MTU) for given device using linux
 * networking ioctl interface.
 */
static int
netdev_linux_set_mtu(struct netdev *netdev_, int mtu)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct ifreq ifr;
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

#ifdef HAVE_AF_XDP
    if (netdev_get_class(netdev_) == &netdev_afxdp_class) {
        error = netdev_afxdp_verify_mtu_size(netdev_, mtu);
        if (error) {
            goto exit;
        }
    }
#endif

    if (netdev->cache_valid & VALID_MTU) {
        error = netdev->netdev_mtu_error;
        if (error || netdev->mtu == mtu) {
            goto exit;
        }
        netdev->cache_valid &= ~VALID_MTU;
    }
    ifr.ifr_mtu = mtu;
    error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr,
                                SIOCSIFMTU, "SIOCSIFMTU");
    if (!error || error == ENODEV) {
        netdev->netdev_mtu_error = error;
        netdev->mtu = ifr.ifr_mtu;
        netdev->cache_valid |= VALID_MTU;
    }
exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

/* Returns the ifindex of 'netdev', if successful, as a positive number.
 * On failure, returns a negative errno value. */
static int
netdev_linux_get_ifindex(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int ifindex, error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }
    error = get_ifindex(netdev_, &ifindex);

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error ? -error : ifindex;
}

static int
netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    ovs_mutex_lock(&netdev->mutex);
    if (netdev->miimon_interval > 0) {
        *carrier = netdev->miimon;
    } else {
        *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0;
    }
    ovs_mutex_unlock(&netdev->mutex);

    return 0;
}

static long long int
netdev_linux_get_carrier_resets(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    long long int carrier_resets;

    ovs_mutex_lock(&netdev->mutex);
    carrier_resets = netdev->carrier_resets;
    ovs_mutex_unlock(&netdev->mutex);

    return carrier_resets;
}

static int
netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
                       struct mii_ioctl_data *data)
{
    struct ifreq ifr;
    int error;

    memset(&ifr, 0, sizeof ifr);
    memcpy(&ifr.ifr_data, data, sizeof *data);
    error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name);
    memcpy(data, &ifr.ifr_data, sizeof *data);

    return error;
}

static int
netdev_linux_get_miimon(const char *name, bool *miimon)
{
    struct mii_ioctl_data data;
    int error;

    *miimon = false;

    memset(&data, 0, sizeof data);
    error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
    if (!error) {
        /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
        data.reg_num = MII_BMSR;
        error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
                                       &data);

        if (!error) {
            *miimon = !!(data.val_out & BMSR_LSTATUS);
        }
    }
    if (error) {
        struct ethtool_cmd ecmd;

        VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
                    name);

        COVERAGE_INC(netdev_get_ethtool);
        memset(&ecmd, 0, sizeof ecmd);
        error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
                                        "ETHTOOL_GLINK");
        if (!error) {
            struct ethtool_value eval;

            memcpy(&eval, &ecmd, sizeof eval);
            *miimon = !!eval.data;
        } else {
            VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
        }
    }

    return error;
}

static int
netdev_linux_set_miimon_interval(struct netdev *netdev_,
                                 long long int interval)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    ovs_mutex_lock(&netdev->mutex);
    interval = interval > 0 ? MAX(interval, 100) : 0;
    if (netdev->miimon_interval != interval) {
        if (interval && !netdev->miimon_interval) {
            atomic_count_inc(&miimon_cnt);
        } else if (!interval && netdev->miimon_interval) {
            atomic_count_dec(&miimon_cnt);
        }

        netdev->miimon_interval = interval;
        timer_set_expired(&netdev->miimon_timer);
    }
    ovs_mutex_unlock(&netdev->mutex);

    return 0;
}

static void
netdev_linux_miimon_run(void)
{
    struct shash device_shash;
    struct shash_node *node;

    shash_init(&device_shash);
    netdev_get_devices(&netdev_linux_class, &device_shash);
    SHASH_FOR_EACH (node, &device_shash) {
        struct netdev *netdev = node->data;
        struct netdev_linux *dev = netdev_linux_cast(netdev);
        bool miimon;

        ovs_mutex_lock(&dev->mutex);
        if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) {
            netdev_linux_get_miimon(dev->up.name, &miimon);
            if (miimon != dev->miimon) {
                dev->miimon = miimon;
                netdev_linux_changed(dev, dev->ifi_flags, 0);
            }

            timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
        }
        ovs_mutex_unlock(&dev->mutex);
        netdev_close(netdev);
    }

    shash_destroy(&device_shash);
}

static void
netdev_linux_miimon_wait(void)
{
    struct shash device_shash;
    struct shash_node *node;

    shash_init(&device_shash);
    netdev_get_devices(&netdev_linux_class, &device_shash);
    SHASH_FOR_EACH (node, &device_shash) {
        struct netdev *netdev = node->data;
        struct netdev_linux *dev = netdev_linux_cast(netdev);

        ovs_mutex_lock(&dev->mutex);
        if (dev->miimon_interval > 0) {
            timer_wait(&dev->miimon_timer);
        }
        ovs_mutex_unlock(&dev->mutex);
        netdev_close(netdev);
    }
    shash_destroy(&device_shash);
}

static void
swap_uint64(uint64_t *a, uint64_t *b)
{
    uint64_t tmp = *a;
    *a = *b;
    *b = tmp;
}

/* Copies 'src' into 'dst', performing format conversion in the process.
 *
 * 'src' is allowed to be misaligned. */
static void
netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst,
                                  const struct ovs_vport_stats *src)
{
    dst->rx_packets = get_32aligned_u64(&src->rx_packets);
    dst->tx_packets = get_32aligned_u64(&src->tx_packets);
    dst->rx_bytes = get_32aligned_u64(&src->rx_bytes);
    dst->tx_bytes = get_32aligned_u64(&src->tx_bytes);
    dst->rx_errors = get_32aligned_u64(&src->rx_errors);
    dst->tx_errors = get_32aligned_u64(&src->tx_errors);
    dst->rx_dropped = get_32aligned_u64(&src->rx_dropped);
    dst->tx_dropped = get_32aligned_u64(&src->tx_dropped);
    dst->multicast = 0;
    dst->collisions = 0;
    dst->rx_length_errors = 0;
    dst->rx_over_errors = 0;
    dst->rx_crc_errors = 0;
    dst->rx_frame_errors = 0;
    dst->rx_fifo_errors = 0;
    dst->rx_missed_errors = 0;
    dst->tx_aborted_errors = 0;
    dst->tx_carrier_errors = 0;
    dst->tx_fifo_errors = 0;
    dst->tx_heartbeat_errors = 0;
    dst->tx_window_errors = 0;
}

static int
get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats)
{
    struct dpif_netlink_vport reply;
    struct ofpbuf *buf;
    int error;

    error = dpif_netlink_vport_get(netdev_get_name(netdev), &reply, &buf);
    if (error) {
        return error;
    } else if (!reply.stats) {
        ofpbuf_delete(buf);
        return EOPNOTSUPP;
    }

    netdev_stats_from_ovs_vport_stats(stats, reply.stats);

    ofpbuf_delete(buf);

    return 0;
}

static void
get_stats_via_vport(const struct netdev *netdev_,
                    struct netdev_stats *stats)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    if (!netdev->vport_stats_error ||
        !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) {
        int error;

        error = get_stats_via_vport__(netdev_, stats);
        if (error && error != ENOENT && error != ENODEV) {
            VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
                         "(%s)",
                         netdev_get_name(netdev_), ovs_strerror(error));
        }
        netdev->vport_stats_error = error;
        netdev->cache_valid |= VALID_VPORT_STAT_ERROR;
    }
}

/* Retrieves current device stats for 'netdev-linux'. */
static int
netdev_linux_get_stats(const struct netdev *netdev_,
                       struct netdev_stats *stats)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct netdev_stats dev_stats;
    int error;

    ovs_mutex_lock(&netdev->mutex);
    get_stats_via_vport(netdev_, stats);
    error = get_stats_via_netlink(netdev_, &dev_stats);
    if (error) {
        if (!netdev->vport_stats_error) {
            error = 0;
        }
    } else if (netdev->vport_stats_error) {
        /* stats not available from OVS then use netdev stats. */
        *stats = dev_stats;
    } else {
        stats->multicast           += dev_stats.multicast;
        stats->collisions          += dev_stats.collisions;
        stats->rx_length_errors    += dev_stats.rx_length_errors;
        stats->rx_over_errors      += dev_stats.rx_over_errors;
        stats->rx_crc_errors       += dev_stats.rx_crc_errors;
        stats->rx_frame_errors     += dev_stats.rx_frame_errors;
        stats->rx_fifo_errors      += dev_stats.rx_fifo_errors;
        stats->rx_missed_errors    += dev_stats.rx_missed_errors;
        stats->tx_aborted_errors   += dev_stats.tx_aborted_errors;
        stats->tx_carrier_errors   += dev_stats.tx_carrier_errors;
        stats->tx_fifo_errors      += dev_stats.tx_fifo_errors;
        stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
        stats->tx_window_errors    += dev_stats.tx_window_errors;
    }
    ovs_mutex_unlock(&netdev->mutex);

    return error;
}

/* Retrieves current device stats for 'netdev-tap' netdev or
 * netdev-internal. */
static int
netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct netdev_stats dev_stats;
    int error;

    ovs_mutex_lock(&netdev->mutex);
    get_stats_via_vport(netdev_, stats);
    error = get_stats_via_netlink(netdev_, &dev_stats);
    if (error) {
        if (!netdev->vport_stats_error) {
            error = 0;
        }
    } else if (netdev->vport_stats_error) {
        /* Transmit and receive stats will appear to be swapped relative to the
         * other ports since we are the one sending the data, not a remote
         * computer.  For consistency, we swap them back here. This does not
         * apply if we are getting stats from the vport layer because it always
         * tracks stats from the perspective of the switch. */

        *stats = dev_stats;
        swap_uint64(&stats->rx_packets, &stats->tx_packets);
        swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
        swap_uint64(&stats->rx_errors, &stats->tx_errors);
        swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
        stats->rx_length_errors = 0;
        stats->rx_over_errors = 0;
        stats->rx_crc_errors = 0;
        stats->rx_frame_errors = 0;
        stats->rx_fifo_errors = 0;
        stats->rx_missed_errors = 0;
        stats->tx_aborted_errors = 0;
        stats->tx_carrier_errors = 0;
        stats->tx_fifo_errors = 0;
        stats->tx_heartbeat_errors = 0;
        stats->tx_window_errors = 0;
    } else {
        /* Use kernel netdev's packet and byte counts since vport counters
         * do not reflect packet counts on the wire when GSO, TSO or GRO
         * are enabled. */
        stats->rx_packets = dev_stats.tx_packets;
        stats->rx_bytes = dev_stats.tx_bytes;
        stats->tx_packets = dev_stats.rx_packets;
        stats->tx_bytes = dev_stats.rx_bytes;

        stats->rx_dropped          += dev_stats.tx_dropped;
        stats->tx_dropped          += dev_stats.rx_dropped;

        stats->rx_errors           += dev_stats.tx_errors;
        stats->tx_errors           += dev_stats.rx_errors;

        stats->multicast           += dev_stats.multicast;
        stats->collisions          += dev_stats.collisions;
    }
    stats->tx_dropped += netdev->tx_dropped;
    stats->rx_dropped += netdev->rx_dropped;
    ovs_mutex_unlock(&netdev->mutex);

    return error;
}

static int
netdev_internal_get_stats(const struct netdev *netdev_,
                          struct netdev_stats *stats)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    get_stats_via_vport(netdev_, stats);
    error = netdev->vport_stats_error;
    ovs_mutex_unlock(&netdev->mutex);

    return error;
}

static void
netdev_linux_read_features(struct netdev_linux *netdev)
{
    struct ethtool_cmd ecmd;
    uint32_t speed;
    int error;

    if (netdev->cache_valid & VALID_FEATURES) {
        return;
    }

    COVERAGE_INC(netdev_get_ethtool);
    memset(&ecmd, 0, sizeof ecmd);
    error = netdev_linux_do_ethtool(netdev->up.name, &ecmd,
                                    ETHTOOL_GSET, "ETHTOOL_GSET");
    if (error) {
        goto out;
    }

    /* Supported features. */
    netdev->supported = 0;
    if (ecmd.supported & SUPPORTED_10baseT_Half) {
        netdev->supported |= NETDEV_F_10MB_HD;
    }
    if (ecmd.supported & SUPPORTED_10baseT_Full) {
        netdev->supported |= NETDEV_F_10MB_FD;
    }
    if (ecmd.supported & SUPPORTED_100baseT_Half)  {
        netdev->supported |= NETDEV_F_100MB_HD;
    }
    if (ecmd.supported & SUPPORTED_100baseT_Full) {
        netdev->supported |= NETDEV_F_100MB_FD;
    }
    if (ecmd.supported & SUPPORTED_1000baseT_Half) {
        netdev->supported |= NETDEV_F_1GB_HD;
    }
    if ((ecmd.supported & SUPPORTED_1000baseT_Full) ||
        (ecmd.supported & SUPPORTED_1000baseKX_Full)) {
        netdev->supported |= NETDEV_F_1GB_FD;
    }
    if ((ecmd.supported & SUPPORTED_10000baseT_Full) ||
        (ecmd.supported & SUPPORTED_10000baseKX4_Full) ||
        (ecmd.supported & SUPPORTED_10000baseKR_Full) ||
        (ecmd.supported & SUPPORTED_10000baseR_FEC)) {
        netdev->supported |= NETDEV_F_10GB_FD;
    }
    if ((ecmd.supported & SUPPORTED_40000baseKR4_Full) ||
        (ecmd.supported & SUPPORTED_40000baseCR4_Full) ||
        (ecmd.supported & SUPPORTED_40000baseSR4_Full) ||
        (ecmd.supported & SUPPORTED_40000baseLR4_Full)) {
        netdev->supported |= NETDEV_F_40GB_FD;
    }
    if (ecmd.supported & SUPPORTED_TP) {
        netdev->supported |= NETDEV_F_COPPER;
    }
    if (ecmd.supported & SUPPORTED_FIBRE) {
        netdev->supported |= NETDEV_F_FIBER;
    }
    if (ecmd.supported & SUPPORTED_Autoneg) {
        netdev->supported |= NETDEV_F_AUTONEG;
    }
    if (ecmd.supported & SUPPORTED_Pause) {
        netdev->supported |= NETDEV_F_PAUSE;
    }
    if (ecmd.supported & SUPPORTED_Asym_Pause) {
        netdev->supported |= NETDEV_F_PAUSE_ASYM;
    }

    /* Advertised features. */
    netdev->advertised = 0;
    if (ecmd.advertising & ADVERTISED_10baseT_Half) {
        netdev->advertised |= NETDEV_F_10MB_HD;
    }
    if (ecmd.advertising & ADVERTISED_10baseT_Full) {
        netdev->advertised |= NETDEV_F_10MB_FD;
    }
    if (ecmd.advertising & ADVERTISED_100baseT_Half) {
        netdev->advertised |= NETDEV_F_100MB_HD;
    }
    if (ecmd.advertising & ADVERTISED_100baseT_Full) {
        netdev->advertised |= NETDEV_F_100MB_FD;
    }
    if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
        netdev->advertised |= NETDEV_F_1GB_HD;
    }
    if ((ecmd.advertising & ADVERTISED_1000baseT_Full) ||
        (ecmd.advertising & ADVERTISED_1000baseKX_Full)) {
        netdev->advertised |= NETDEV_F_1GB_FD;
    }
    if ((ecmd.advertising & ADVERTISED_10000baseT_Full) ||
        (ecmd.advertising & ADVERTISED_10000baseKX4_Full) ||
        (ecmd.advertising & ADVERTISED_10000baseKR_Full) ||
        (ecmd.advertising & ADVERTISED_10000baseR_FEC)) {
        netdev->advertised |= NETDEV_F_10GB_FD;
    }
    if ((ecmd.advertising & ADVERTISED_40000baseKR4_Full) ||
        (ecmd.advertising & ADVERTISED_40000baseCR4_Full) ||
        (ecmd.advertising & ADVERTISED_40000baseSR4_Full) ||
        (ecmd.advertising & ADVERTISED_40000baseLR4_Full)) {
        netdev->advertised |= NETDEV_F_40GB_FD;
    }
    if (ecmd.advertising & ADVERTISED_TP) {
        netdev->advertised |= NETDEV_F_COPPER;
    }
    if (ecmd.advertising & ADVERTISED_FIBRE) {
        netdev->advertised |= NETDEV_F_FIBER;
    }
    if (ecmd.advertising & ADVERTISED_Autoneg) {
        netdev->advertised |= NETDEV_F_AUTONEG;
    }
    if (ecmd.advertising & ADVERTISED_Pause) {
        netdev->advertised |= NETDEV_F_PAUSE;
    }
    if (ecmd.advertising & ADVERTISED_Asym_Pause) {
        netdev->advertised |= NETDEV_F_PAUSE_ASYM;
    }

    /* Current settings. */
    speed = ethtool_cmd_speed(&ecmd);
    if (speed == SPEED_10) {
        netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD;
    } else if (speed == SPEED_100) {
        netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD;
    } else if (speed == SPEED_1000) {
        netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD;
    } else if (speed == SPEED_10000) {
        netdev->current = NETDEV_F_10GB_FD;
    } else if (speed == 40000) {
        netdev->current = NETDEV_F_40GB_FD;
    } else if (speed == 100000) {
        netdev->current = NETDEV_F_100GB_FD;
    } else if (speed == 1000000) {
        netdev->current = NETDEV_F_1TB_FD;
    } else {
        netdev->current = 0;
    }

    if (ecmd.port == PORT_TP) {
        netdev->current |= NETDEV_F_COPPER;
    } else if (ecmd.port == PORT_FIBRE) {
        netdev->current |= NETDEV_F_FIBER;
    }

    if (ecmd.autoneg) {
        netdev->current |= NETDEV_F_AUTONEG;
    }

out:
    netdev->cache_valid |= VALID_FEATURES;
    netdev->get_features_error = error;
}

/* Stores the features supported by 'netdev' into of '*current', '*advertised',
 * '*supported', and '*peer'.  Each value is a bitmap of NETDEV_* bits.
 * Returns 0 if successful, otherwise a positive errno value. */
static int
netdev_linux_get_features(const struct netdev *netdev_,
                          enum netdev_features *current,
                          enum netdev_features *advertised,
                          enum netdev_features *supported,
                          enum netdev_features *peer)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    netdev_linux_read_features(netdev);
    if (!netdev->get_features_error) {
        *current = netdev->current;
        *advertised = netdev->advertised;
        *supported = netdev->supported;
        *peer = 0;              /* XXX */
    }
    error = netdev->get_features_error;

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

/* Set the features advertised by 'netdev' to 'advertise'. */
static int
netdev_linux_set_advertisements(struct netdev *netdev_,
                                enum netdev_features advertise)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct ethtool_cmd ecmd;
    int error;

    ovs_mutex_lock(&netdev->mutex);

    COVERAGE_INC(netdev_get_ethtool);

    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    memset(&ecmd, 0, sizeof ecmd);
    error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd,
                                    ETHTOOL_GSET, "ETHTOOL_GSET");
    if (error) {
        goto exit;
    }

    ecmd.advertising = 0;
    if (advertise & NETDEV_F_10MB_HD) {
        ecmd.advertising |= ADVERTISED_10baseT_Half;
    }
    if (advertise & NETDEV_F_10MB_FD) {
        ecmd.advertising |= ADVERTISED_10baseT_Full;
    }
    if (advertise & NETDEV_F_100MB_HD) {
        ecmd.advertising |= ADVERTISED_100baseT_Half;
    }
    if (advertise & NETDEV_F_100MB_FD) {
        ecmd.advertising |= ADVERTISED_100baseT_Full;
    }
    if (advertise & NETDEV_F_1GB_HD) {
        ecmd.advertising |= ADVERTISED_1000baseT_Half;
    }
    if (advertise & NETDEV_F_1GB_FD) {
        ecmd.advertising |= ADVERTISED_1000baseT_Full;
    }
    if (advertise & NETDEV_F_10GB_FD) {
        ecmd.advertising |= ADVERTISED_10000baseT_Full;
    }
    if (advertise & NETDEV_F_COPPER) {
        ecmd.advertising |= ADVERTISED_TP;
    }
    if (advertise & NETDEV_F_FIBER) {
        ecmd.advertising |= ADVERTISED_FIBRE;
    }
    if (advertise & NETDEV_F_AUTONEG) {
        ecmd.advertising |= ADVERTISED_Autoneg;
    }
    if (advertise & NETDEV_F_PAUSE) {
        ecmd.advertising |= ADVERTISED_Pause;
    }
    if (advertise & NETDEV_F_PAUSE_ASYM) {
        ecmd.advertising |= ADVERTISED_Asym_Pause;
    }
    COVERAGE_INC(netdev_set_ethtool);
    error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd,
                                    ETHTOOL_SSET, "ETHTOOL_SSET");

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static struct tc_police
tc_matchall_fill_police(uint32_t kbits_rate, uint32_t kbits_burst)
{
    unsigned int bsize = MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 64;
    unsigned int bps = ((uint64_t) kbits_rate * 1000) / 8;
    struct tc_police police;
    struct tc_ratespec rate;
    int mtu = 65535;

    memset(&rate, 0, sizeof rate);
    rate.rate = bps;
    rate.cell_log = tc_calc_cell_log(mtu);
    rate.mpu = ETH_TOTAL_MIN;

    memset(&police, 0, sizeof police);
    police.burst = tc_bytes_to_ticks(bps, bsize);
    police.action = TC_POLICE_SHOT;
    police.rate = rate;
    police.mtu = mtu;

    return police;
}

static void
nl_msg_put_act_police(struct ofpbuf *request, struct tc_police police)
{
    size_t offset;

    nl_msg_put_string(request, TCA_ACT_KIND, "police");
    offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
    nl_msg_put_unspec(request, TCA_POLICE_TBF, &police, sizeof police);
    tc_put_rtab(request, TCA_POLICE_RATE, &police.rate);
    nl_msg_put_u32(request, TCA_POLICE_RESULT, TC_ACT_UNSPEC);
    nl_msg_end_nested(request, offset);
}

static int
tc_add_matchall_policer(struct netdev *netdev, uint32_t kbits_rate,
                        uint32_t kbits_burst)
{
    uint16_t eth_type = (OVS_FORCE uint16_t) htons(ETH_P_ALL);
    size_t basic_offset, action_offset, inner_offset;
    uint16_t prio = TC_RESERVED_PRIORITY_POLICE;
    int ifindex, err = 0;
    struct tc_police pol_act;
    struct ofpbuf request;
    struct ofpbuf *reply;
    struct tcmsg *tcmsg;
    uint32_t handle = 1;

    err = get_ifindex(netdev, &ifindex);
    if (err) {
        return err;
    }

    tcmsg = tc_make_request(ifindex, RTM_NEWTFILTER, NLM_F_CREATE | NLM_F_ECHO,
                            &request);
    tcmsg->tcm_parent = TC_INGRESS_PARENT;
    tcmsg->tcm_info = tc_make_handle(prio, eth_type);
    tcmsg->tcm_handle = handle;

    pol_act = tc_matchall_fill_police(kbits_rate, kbits_burst);
    nl_msg_put_string(&request, TCA_KIND, "matchall");
    basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    action_offset = nl_msg_start_nested(&request, TCA_MATCHALL_ACT);
    inner_offset = nl_msg_start_nested(&request, 1);
    nl_msg_put_act_police(&request, pol_act);
    nl_msg_end_nested(&request, inner_offset);
    nl_msg_end_nested(&request, action_offset);
    nl_msg_end_nested(&request, basic_offset);

    err = tc_transact(&request, &reply);
    if (!err) {
        struct tcmsg *tc =
            ofpbuf_at_assert(reply, NLMSG_HDRLEN, sizeof *tc);
        ofpbuf_delete(reply);
    }

    return err;
}

static int
tc_del_matchall_policer(struct netdev *netdev)
{
    int prio = TC_RESERVED_PRIORITY_POLICE;
    uint32_t block_id = 0;
    struct tcf_id id;
    int ifindex;
    int err;

    err = get_ifindex(netdev, &ifindex);
    if (err) {
        return err;
    }

    id = tc_make_tcf_id(ifindex, block_id, prio, TC_INGRESS);
    err = tc_del_filter(&id);
    if (err) {
        return err;
    }

    return 0;
}

/* Attempts to set input rate limiting (policing) policy.  Returns 0 if
 * successful, otherwise a positive errno value. */
static int
netdev_linux_set_policing(struct netdev *netdev_,
                          uint32_t kbits_rate, uint32_t kbits_burst)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    const char *netdev_name = netdev_get_name(netdev_);
    int ifindex;
    int error;

    kbits_burst = (!kbits_rate ? 0       /* Force to 0 if no rate specified. */
                   : !kbits_burst ? 8000 /* Default to 8000 kbits if 0. */
                   : kbits_burst);       /* Stick with user-specified value. */

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto out;
    }

    if (netdev->cache_valid & VALID_POLICING) {
        error = netdev->netdev_policing_error;
        if (error || (netdev->kbits_rate == kbits_rate &&
                      netdev->kbits_burst == kbits_burst)) {
            /* Assume that settings haven't changed since we last set them. */
            goto out;
        }
        netdev->cache_valid &= ~VALID_POLICING;
    }

    COVERAGE_INC(netdev_set_policing);

    /* Use matchall for policing when offloadling ovs with tc-flower. */
    if (netdev_is_flow_api_enabled()) {
        error = tc_del_matchall_policer(netdev_);
        if (kbits_rate) {
            error = tc_add_matchall_policer(netdev_, kbits_rate, kbits_burst);
        }
        ovs_mutex_unlock(&netdev->mutex);
        return error;
    }

    error = get_ifindex(netdev_, &ifindex);
    if (error) {
        goto out;
    }

    /* Remove any existing ingress qdisc. */
    error = tc_add_del_qdisc(ifindex, false, 0, TC_INGRESS);
    if (error) {
        VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
                     netdev_name, ovs_strerror(error));
        goto out;
    }

    if (kbits_rate) {
        error = tc_add_del_qdisc(ifindex, true, 0, TC_INGRESS);
        if (error) {
            VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
                         netdev_name, ovs_strerror(error));
            goto out;
        }

        error = tc_add_policer(netdev_, kbits_rate, kbits_burst);
        if (error){
            VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
                    netdev_name, ovs_strerror(error));
            goto out;
        }
    }

    netdev->kbits_rate = kbits_rate;
    netdev->kbits_burst = kbits_burst;

out:
    if (!error || error == ENODEV) {
        netdev->netdev_policing_error = error;
        netdev->cache_valid |= VALID_POLICING;
    }
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
                           struct sset *types)
{
    const struct tc_ops *const *opsp;
    for (opsp = tcs; *opsp != NULL; opsp++) {
        const struct tc_ops *ops = *opsp;
        if (ops->tc_install && ops->ovs_name[0] != '\0') {
            sset_add(types, ops->ovs_name);
        }
    }
    return 0;
}

static const struct tc_ops *
tc_lookup_ovs_name(const char *name)
{
    const struct tc_ops *const *opsp;

    for (opsp = tcs; *opsp != NULL; opsp++) {
        const struct tc_ops *ops = *opsp;
        if (!strcmp(name, ops->ovs_name)) {
            return ops;
        }
    }
    return NULL;
}

static const struct tc_ops *
tc_lookup_linux_name(const char *name)
{
    const struct tc_ops *const *opsp;

    for (opsp = tcs; *opsp != NULL; opsp++) {
        const struct tc_ops *ops = *opsp;
        if (ops->linux_name && !strcmp(name, ops->linux_name)) {
            return ops;
        }
    }
    return NULL;
}

static struct tc_queue *
tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id,
                size_t hash)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct tc_queue *queue;

    HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) {
        if (queue->queue_id == queue_id) {
            return queue;
        }
    }
    return NULL;
}

static struct tc_queue *
tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
{
    return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
}

static int
netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
                                  const char *type,
                                  struct netdev_qos_capabilities *caps)
{
    const struct tc_ops *ops = tc_lookup_ovs_name(type);
    if (!ops) {
        return EOPNOTSUPP;
    }
    caps->n_queues = ops->n_queues;
    return 0;
}

static int
netdev_linux_get_qos(const struct netdev *netdev_,
                     const char **typep, struct smap *details)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        *typep = netdev->tc->ops->ovs_name;
        error = (netdev->tc->ops->qdisc_get
                 ? netdev->tc->ops->qdisc_get(netdev_, details)
                 : 0);
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_set_qos(struct netdev *netdev_,
                     const char *type, const struct smap *details)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    const struct tc_ops *new_ops;
    int error;

    new_ops = tc_lookup_ovs_name(type);
    if (!new_ops || !new_ops->tc_install) {
        return EOPNOTSUPP;
    }

    if (new_ops == &tc_ops_noop) {
        return new_ops->tc_install(netdev_, details);
    }

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (error) {
        goto exit;
    }

    if (new_ops == netdev->tc->ops) {
        error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0;
    } else {
        /* Delete existing qdisc. */
        error = tc_del_qdisc(netdev_);
        if (error) {
            goto exit;
        }
        ovs_assert(netdev->tc == NULL);

        /* Install new qdisc. */
        error = new_ops->tc_install(netdev_, details);
        ovs_assert((error == 0) == (netdev->tc != NULL));
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_get_queue(const struct netdev *netdev_,
                       unsigned int queue_id, struct smap *details)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
        error = (queue
                ? netdev->tc->ops->class_get(netdev_, queue, details)
                : ENOENT);
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_set_queue(struct netdev *netdev_,
                       unsigned int queue_id, const struct smap *details)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        error = (queue_id < netdev->tc->ops->n_queues
                 && netdev->tc->ops->class_set
                 ? netdev->tc->ops->class_set(netdev_, queue_id, details)
                 : EINVAL);
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        if (netdev->tc->ops->class_delete) {
            struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
            error = (queue
                     ? netdev->tc->ops->class_delete(netdev_, queue)
                     : ENOENT);
        } else {
            error = EINVAL;
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_get_queue_stats(const struct netdev *netdev_,
                             unsigned int queue_id,
                             struct netdev_queue_stats *stats)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        if (netdev->tc->ops->class_get_stats) {
            const struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
            if (queue) {
                stats->created = queue->created;
                error = netdev->tc->ops->class_get_stats(netdev_, queue,
                                                         stats);
            } else {
                error = ENOENT;
            }
        } else {
            error = EOPNOTSUPP;
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

struct queue_dump_state {
    struct nl_dump dump;
    struct ofpbuf buf;
};

static bool
start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state)
{
    struct ofpbuf request;
    struct tcmsg *tcmsg;

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
    if (!tcmsg) {
        return false;
    }
    tcmsg->tcm_parent = 0;
    nl_dump_start(&state->dump, NETLINK_ROUTE, &request);
    ofpbuf_uninit(&request);

    ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE);
    return true;
}

static int
finish_queue_dump(struct queue_dump_state *state)
{
    ofpbuf_uninit(&state->buf);
    return nl_dump_done(&state->dump);
}

struct netdev_linux_queue_state {
    unsigned int *queues;
    size_t cur_queue;
    size_t n_queues;
};

static int
netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        if (netdev->tc->ops->class_get) {
            struct netdev_linux_queue_state *state;
            struct tc_queue *queue;
            size_t i;

            *statep = state = xmalloc(sizeof *state);
            state->n_queues = hmap_count(&netdev->tc->queues);
            state->cur_queue = 0;
            state->queues = xmalloc(state->n_queues * sizeof *state->queues);

            i = 0;
            HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) {
                state->queues[i++] = queue->queue_id;
            }
        } else {
            error = EOPNOTSUPP;
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_,
                             unsigned int *queue_idp, struct smap *details)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct netdev_linux_queue_state *state = state_;
    int error = EOF;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    while (state->cur_queue < state->n_queues) {
        unsigned int queue_id = state->queues[state->cur_queue++];
        struct tc_queue *queue = tc_find_queue(netdev_, queue_id);

        if (queue) {
            *queue_idp = queue_id;
            error = netdev->tc->ops->class_get(netdev_, queue, details);
            break;
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED,
                             void *state_)
{
    struct netdev_linux_queue_state *state = state_;

    free(state->queues);
    free(state);
    return 0;
}

static int
netdev_linux_dump_queue_stats(const struct netdev *netdev_,
                              netdev_dump_queue_stats_cb *cb, void *aux)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = tc_query_qdisc(netdev_);
    if (!error) {
        struct queue_dump_state state;

        if (!netdev->tc->ops->class_dump_stats) {
            error = EOPNOTSUPP;
        } else if (!start_queue_dump(netdev_, &state)) {
            error = ENODEV;
        } else {
            struct ofpbuf msg;
            int retval;

            while (nl_dump_next(&state.dump, &msg, &state.buf)) {
                retval = netdev->tc->ops->class_dump_stats(netdev_, &msg,
                                                           cb, aux);
                if (retval) {
                    error = retval;
                }
            }

            retval = finish_queue_dump(&state);
            if (retval) {
                error = retval;
            }
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static int
netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
                     struct in_addr netmask)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
    if (!error) {
        if (address.s_addr != INADDR_ANY) {
            error = do_set_addr(netdev_, SIOCSIFNETMASK,
                                "SIOCSIFNETMASK", netmask);
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address.
 * Otherwise, sets '*in6' to 'in6addr_any' and returns the corresponding
 * error. */
static int
netdev_linux_get_addr_list(const struct netdev *netdev_,
                          struct in6_addr **addr, struct in6_addr **mask, int *n_cnt)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error;

    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
        goto exit;
    }

    error = netdev_get_addrs(netdev_get_name(netdev_), addr, mask, n_cnt);

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

static void
make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
{
    struct sockaddr_in sin;
    memset(&sin, 0, sizeof sin);
    sin.sin_family = AF_INET;
    sin.sin_addr = addr;
    sin.sin_port = 0;

    memset(sa, 0, sizeof *sa);
    memcpy(sa, &sin, sizeof sin);
}

static int
do_set_addr(struct netdev *netdev,
            int ioctl_nr, const char *ioctl_name, struct in_addr addr)
{
    struct ifreq ifr;

    make_in4_sockaddr(&ifr.ifr_addr, addr);
    return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
                               ioctl_name);
}

/* Adds 'router' as a default IP gateway. */
static int
netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
{
    struct in_addr any = { INADDR_ANY };
    struct rtentry rt;
    int error;

    memset(&rt, 0, sizeof rt);
    make_in4_sockaddr(&rt.rt_dst, any);
    make_in4_sockaddr(&rt.rt_gateway, router);
    make_in4_sockaddr(&rt.rt_genmask, any);
    rt.rt_flags = RTF_UP | RTF_GATEWAY;
    error = af_inet_ioctl(SIOCADDRT, &rt);
    if (error) {
        VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error));
    }
    return error;
}

static int
netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
                          char **netdev_name)
{
    static const char fn[] = "/proc/net/route";
    FILE *stream;
    char line[256];
    int ln;

    *netdev_name = NULL;
    stream = fopen(fn, "r");
    if (stream == NULL) {
        VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno));
        return errno;
    }

    ln = 0;
    while (fgets(line, sizeof line, stream)) {
        if (++ln >= 2) {
            char iface[17];
            ovs_be32 dest, gateway, mask;
            int refcnt, metric, mtu;
            unsigned int flags, use, window, irtt;

            if (!ovs_scan(line,
                          "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
                          " %d %u %u\n",
                          iface, &dest, &gateway, &flags, &refcnt,
                          &use, &metric, &mask, &mtu, &window, &irtt)) {
                VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
                        fn, ln, line);
                continue;
            }
            if (!(flags & RTF_UP)) {
                /* Skip routes that aren't up. */
                continue;
            }

            /* The output of 'dest', 'mask', and 'gateway' were given in
             * network byte order, so we don't need need any endian
             * conversions here. */
            if ((dest & mask) == (host->s_addr & mask)) {
                if (!gateway) {
                    /* The host is directly reachable. */
                    next_hop->s_addr = 0;
                } else {
                    /* To reach the host, we must go through a gateway. */
                    next_hop->s_addr = gateway;
                }
                *netdev_name = xstrdup(iface);
                fclose(stream);
                return 0;
            }
        }
    }

    fclose(stream);
    return ENXIO;
}

static int
netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error = 0;

    ovs_mutex_lock(&netdev->mutex);
    if (!(netdev->cache_valid & VALID_DRVINFO)) {
        struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo;

        COVERAGE_INC(netdev_get_ethtool);
        memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo);
        error = netdev_linux_do_ethtool(netdev->up.name,
                                        cmd,
                                        ETHTOOL_GDRVINFO,
                                        "ETHTOOL_GDRVINFO");
        if (!error) {
            netdev->cache_valid |= VALID_DRVINFO;
        }
    }

    if (!error) {
        smap_add(smap, "driver_name", netdev->drvinfo.driver);
        smap_add(smap, "driver_version", netdev->drvinfo.version);
        smap_add(smap, "firmware_version", netdev->drvinfo.fw_version);
    }
    ovs_mutex_unlock(&netdev->mutex);

    return error;
}

static int
netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED,
                           struct smap *smap)
{
    smap_add(smap, "driver_name", "openvswitch");
    return 0;
}

static uint32_t
netdev_linux_get_block_id(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    uint32_t block_id = 0;

    ovs_mutex_lock(&netdev->mutex);
    /* Ensure the linux netdev has had its fields populated. */
    if (!(netdev->cache_valid & VALID_IFINDEX)) {
        netdev_linux_update_via_netlink(netdev);
    }

    /* Only assigning block ids to linux netdevs that are LAG masters. */
    if (netdev->is_lag_master) {
        block_id = netdev->ifindex;
    }
    ovs_mutex_unlock(&netdev->mutex);

    return block_id;
}

/* Looks up the ARP table entry for 'ip' on 'netdev'.  If one exists and can be
 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
 * returns 0.  Otherwise, it returns a positive errno value; in particular,
 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
static int
netdev_linux_arp_lookup(const struct netdev *netdev,
                        ovs_be32 ip, struct eth_addr *mac)
{
    struct arpreq r;
    struct sockaddr_in sin;
    int retval;

    memset(&r, 0, sizeof r);
    memset(&sin, 0, sizeof sin);
    sin.sin_family = AF_INET;
    sin.sin_addr.s_addr = ip;
    sin.sin_port = 0;
    memcpy(&r.arp_pa, &sin, sizeof sin);
    r.arp_ha.sa_family = ARPHRD_ETHER;
    r.arp_flags = 0;
    ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
    COVERAGE_INC(netdev_arp_lookup);
    retval = af_inet_ioctl(SIOCGARP, &r);
    if (!retval) {
        memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
    } else if (retval != ENXIO) {
        VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
                     netdev_get_name(netdev), IP_ARGS(ip),
                     ovs_strerror(retval));
    }
    return retval;
}

static unsigned int
nd_to_iff_flags(enum netdev_flags nd)
{
    unsigned int iff = 0;
    if (nd & NETDEV_UP) {
        iff |= IFF_UP;
    }
    if (nd & NETDEV_PROMISC) {
        iff |= IFF_PROMISC;
    }
    if (nd & NETDEV_LOOPBACK) {
        iff |= IFF_LOOPBACK;
    }
    return iff;
}

static int
iff_to_nd_flags(unsigned int iff)
{
    enum netdev_flags nd = 0;
    if (iff & IFF_UP) {
        nd |= NETDEV_UP;
    }
    if (iff & IFF_PROMISC) {
        nd |= NETDEV_PROMISC;
    }
    if (iff & IFF_LOOPBACK) {
        nd |= NETDEV_LOOPBACK;
    }
    return nd;
}

static int
update_flags(struct netdev_linux *netdev, enum netdev_flags off,
             enum netdev_flags on, enum netdev_flags *old_flagsp)
    OVS_REQUIRES(netdev->mutex)
{
    unsigned int old_flags, new_flags;
    int error = 0;

    old_flags = netdev->ifi_flags;
    *old_flagsp = iff_to_nd_flags(old_flags);
    new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
    if (new_flags != old_flags) {
        error = set_flags(netdev_get_name(&netdev->up), new_flags);
        get_flags(&netdev->up, &netdev->ifi_flags);
    }

    return error;
}

static int
netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off,
                          enum netdev_flags on, enum netdev_flags *old_flagsp)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    int error = 0;

    ovs_mutex_lock(&netdev->mutex);
    if (on || off) {
        /* Changing flags over netlink isn't support yet. */
        if (netdev_linux_netnsid_is_remote(netdev)) {
            error = EOPNOTSUPP;
            goto exit;
        }
        error = update_flags(netdev, off, on, old_flagsp);
    } else {
        /* Try reading flags over netlink, or fall back to ioctl. */
        if (!netdev_linux_update_via_netlink(netdev)) {
            *old_flagsp = iff_to_nd_flags(netdev->ifi_flags);
        } else {
            error = update_flags(netdev, off, on, old_flagsp);
        }
    }

exit:
    ovs_mutex_unlock(&netdev->mutex);
    return error;
}

#define NETDEV_LINUX_CLASS_COMMON                               \
    .run = netdev_linux_run,                                    \
    .wait = netdev_linux_wait,                                  \
    .alloc = netdev_linux_alloc,                                \
    .dealloc = netdev_linux_dealloc,                            \
    .send_wait = netdev_linux_send_wait,                        \
    .set_etheraddr = netdev_linux_set_etheraddr,                \
    .get_etheraddr = netdev_linux_get_etheraddr,                \
    .get_mtu = netdev_linux_get_mtu,                            \
    .set_mtu = netdev_linux_set_mtu,                            \
    .get_ifindex = netdev_linux_get_ifindex,                    \
    .get_carrier = netdev_linux_get_carrier,                    \
    .get_carrier_resets = netdev_linux_get_carrier_resets,      \
    .set_miimon_interval = netdev_linux_set_miimon_interval,    \
    .set_advertisements = netdev_linux_set_advertisements,      \
    .set_policing = netdev_linux_set_policing,                  \
    .get_qos_types = netdev_linux_get_qos_types,                \
    .get_qos_capabilities = netdev_linux_get_qos_capabilities,  \
    .get_qos = netdev_linux_get_qos,                            \
    .set_qos = netdev_linux_set_qos,                            \
    .get_queue = netdev_linux_get_queue,                        \
    .set_queue = netdev_linux_set_queue,                        \
    .delete_queue = netdev_linux_delete_queue,                  \
    .get_queue_stats = netdev_linux_get_queue_stats,            \
    .queue_dump_start = netdev_linux_queue_dump_start,          \
    .queue_dump_next = netdev_linux_queue_dump_next,            \
    .queue_dump_done = netdev_linux_queue_dump_done,            \
    .dump_queue_stats = netdev_linux_dump_queue_stats,          \
    .set_in4 = netdev_linux_set_in4,                            \
    .get_addr_list = netdev_linux_get_addr_list,                \
    .add_router = netdev_linux_add_router,                      \
    .get_next_hop = netdev_linux_get_next_hop,                  \
    .arp_lookup = netdev_linux_arp_lookup,                      \
    .update_flags = netdev_linux_update_flags,                  \
    .rxq_alloc = netdev_linux_rxq_alloc,                        \
    .rxq_dealloc = netdev_linux_rxq_dealloc,                    \
    .rxq_wait = netdev_linux_rxq_wait,                          \
    .rxq_drain = netdev_linux_rxq_drain

const struct netdev_class netdev_linux_class = {
    NETDEV_LINUX_CLASS_COMMON,
    .type = "system",
    .is_pmd = false,
    .construct = netdev_linux_construct,
    .destruct = netdev_linux_destruct,
    .get_stats = netdev_linux_get_stats,
    .get_features = netdev_linux_get_features,
    .get_status = netdev_linux_get_status,
    .get_block_id = netdev_linux_get_block_id,
    .send = netdev_linux_send,
    .rxq_construct = netdev_linux_rxq_construct,
    .rxq_destruct = netdev_linux_rxq_destruct,
    .rxq_recv = netdev_linux_rxq_recv,
};

const struct netdev_class netdev_tap_class = {
    NETDEV_LINUX_CLASS_COMMON,
    .type = "tap",
    .is_pmd = false,
    .construct = netdev_linux_construct_tap,
    .destruct = netdev_linux_destruct,
    .get_stats = netdev_tap_get_stats,
    .get_features = netdev_linux_get_features,
    .get_status = netdev_linux_get_status,
    .send = netdev_linux_send,
    .rxq_construct = netdev_linux_rxq_construct,
    .rxq_destruct = netdev_linux_rxq_destruct,
    .rxq_recv = netdev_linux_rxq_recv,
};

const struct netdev_class netdev_internal_class = {
    NETDEV_LINUX_CLASS_COMMON,
    .type = "internal",
    .is_pmd = false,
    .construct = netdev_linux_construct,
    .destruct = netdev_linux_destruct,
    .get_stats = netdev_internal_get_stats,
    .get_status = netdev_internal_get_status,
    .send = netdev_linux_send,
    .rxq_construct = netdev_linux_rxq_construct,
    .rxq_destruct = netdev_linux_rxq_destruct,
    .rxq_recv = netdev_linux_rxq_recv,
};

#ifdef HAVE_AF_XDP
#define NETDEV_AFXDP_CLASS_COMMON                               \
    .init = netdev_afxdp_init,                                  \
    .construct = netdev_afxdp_construct,                        \
    .destruct = netdev_afxdp_destruct,                          \
    .get_stats = netdev_afxdp_get_stats,                        \
    .get_custom_stats = netdev_afxdp_get_custom_stats,          \
    .get_status = netdev_linux_get_status,                      \
    .set_config = netdev_afxdp_set_config,                      \
    .get_config = netdev_afxdp_get_config,                      \
    .reconfigure = netdev_afxdp_reconfigure,                    \
    .get_numa_id = netdev_linux_get_numa_id,                    \
    .send = netdev_afxdp_batch_send,                            \
    .rxq_construct = netdev_afxdp_rxq_construct,                \
    .rxq_destruct = netdev_afxdp_rxq_destruct,                  \
    .rxq_recv = netdev_afxdp_rxq_recv

const struct netdev_class netdev_afxdp_class = {
    NETDEV_LINUX_CLASS_COMMON,
    NETDEV_AFXDP_CLASS_COMMON,
    .type = "afxdp",
    .is_pmd = true,
};

const struct netdev_class netdev_afxdp_nonpmd_class = {
    NETDEV_LINUX_CLASS_COMMON,
    NETDEV_AFXDP_CLASS_COMMON,
    .type = "afxdp-nonpmd",
    .is_pmd = false,
};
#endif
\f

#define CODEL_N_QUEUES 0x0000

/* In sufficiently new kernel headers these are defined as enums in
 * <linux/pkt_sched.h>.  Define them here as macros to help out with older
 * kernels.  (This overrides any enum definition in the header file but that's
 * harmless.) */
#define TCA_CODEL_TARGET   1
#define TCA_CODEL_LIMIT    2
#define TCA_CODEL_INTERVAL 3

struct codel {
    struct tc tc;
    uint32_t target;
    uint32_t limit;
    uint32_t interval;
};

static struct codel *
codel_get__(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    return CONTAINER_OF(netdev->tc, struct codel, tc);
}

static void
codel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit,
                uint32_t interval)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct codel *codel;

    codel = xmalloc(sizeof *codel);
    tc_init(&codel->tc, &tc_ops_codel);
    codel->target = target;
    codel->limit = limit;
    codel->interval = interval;

    netdev->tc = &codel->tc;
}

static int
codel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit,
                    uint32_t interval)
{
    size_t opt_offset;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    uint32_t otarget, olimit, ointerval;
    int error;

    tc_del_qdisc(netdev);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    otarget = target ? target : 5000;
    olimit = limit ? limit : 10240;
    ointerval = interval ? interval : 100000;

    nl_msg_put_string(&request, TCA_KIND, "codel");
    opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    nl_msg_put_u32(&request, TCA_CODEL_TARGET, otarget);
    nl_msg_put_u32(&request, TCA_CODEL_LIMIT, olimit);
    nl_msg_put_u32(&request, TCA_CODEL_INTERVAL, ointerval);
    nl_msg_end_nested(&request, opt_offset);

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
        "target %u, limit %u, interval %u error %d(%s)",
        netdev_get_name(netdev),
        otarget, olimit, ointerval,
        error, ovs_strerror(error));
    }
    return error;
}

static void
codel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED,
                            const struct smap *details, struct codel *codel)
{
    codel->target = smap_get_ullong(details, "target", 0);
    codel->limit = smap_get_ullong(details, "limit", 0);
    codel->interval = smap_get_ullong(details, "interval", 0);

    if (!codel->target) {
        codel->target = 5000;
    }
    if (!codel->limit) {
        codel->limit = 10240;
    }
    if (!codel->interval) {
        codel->interval = 100000;
    }
}

static int
codel_tc_install(struct netdev *netdev, const struct smap *details)
{
    int error;
    struct codel codel;

    codel_parse_qdisc_details__(netdev, details, &codel);
    error = codel_setup_qdisc__(netdev, codel.target, codel.limit,
                                codel.interval);
    if (!error) {
        codel_install__(netdev, codel.target, codel.limit, codel.interval);
    }
    return error;
}

static int
codel_parse_tca_options__(struct nlattr *nl_options, struct codel *codel)
{
    static const struct nl_policy tca_codel_policy[] = {
        [TCA_CODEL_TARGET] = { .type = NL_A_U32 },
        [TCA_CODEL_LIMIT] = { .type = NL_A_U32 },
        [TCA_CODEL_INTERVAL] = { .type = NL_A_U32 }
    };

    struct nlattr *attrs[ARRAY_SIZE(tca_codel_policy)];

    if (!nl_parse_nested(nl_options, tca_codel_policy,
                         attrs, ARRAY_SIZE(tca_codel_policy))) {
        VLOG_WARN_RL(&rl, "failed to parse CoDel class options");
        return EPROTO;
    }

    codel->target = nl_attr_get_u32(attrs[TCA_CODEL_TARGET]);
    codel->limit = nl_attr_get_u32(attrs[TCA_CODEL_LIMIT]);
    codel->interval = nl_attr_get_u32(attrs[TCA_CODEL_INTERVAL]);
    return 0;
}

static int
codel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
{
    struct nlattr *nlattr;
    const char * kind;
    int error;
    struct codel codel;

    error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
    if (error != 0) {
        return error;
    }

    error = codel_parse_tca_options__(nlattr, &codel);
    if (error != 0) {
        return error;
    }

    codel_install__(netdev, codel.target, codel.limit, codel.interval);
    return 0;
}


static void
codel_tc_destroy(struct tc *tc)
{
    struct codel *codel = CONTAINER_OF(tc, struct codel, tc);
    tc_destroy(tc);
    free(codel);
}

static int
codel_qdisc_get(const struct netdev *netdev, struct smap *details)
{
    const struct codel *codel = codel_get__(netdev);
    smap_add_format(details, "target", "%u", codel->target);
    smap_add_format(details, "limit", "%u", codel->limit);
    smap_add_format(details, "interval", "%u", codel->interval);
    return 0;
}

static int
codel_qdisc_set(struct netdev *netdev, const struct smap *details)
{
    struct codel codel;

    codel_parse_qdisc_details__(netdev, details, &codel);
    codel_install__(netdev, codel.target, codel.limit, codel.interval);
    codel_get__(netdev)->target = codel.target;
    codel_get__(netdev)->limit = codel.limit;
    codel_get__(netdev)->interval = codel.interval;
    return 0;
}

static const struct tc_ops tc_ops_codel = {
    .linux_name = "codel",
    .ovs_name = "linux-codel",
    .n_queues = CODEL_N_QUEUES,
    .tc_install = codel_tc_install,
    .tc_load = codel_tc_load,
    .tc_destroy = codel_tc_destroy,
    .qdisc_get = codel_qdisc_get,
    .qdisc_set = codel_qdisc_set,
};
\f
/* FQ-CoDel traffic control class. */

#define FQCODEL_N_QUEUES 0x0000

/* In sufficiently new kernel headers these are defined as enums in
 * <linux/pkt_sched.h>.  Define them here as macros to help out with older
 * kernels.  (This overrides any enum definition in the header file but that's
 * harmless.) */
#define TCA_FQ_CODEL_TARGET     1
#define TCA_FQ_CODEL_LIMIT      2
#define TCA_FQ_CODEL_INTERVAL   3
#define TCA_FQ_CODEL_ECN        4
#define TCA_FQ_CODEL_FLOWS      5
#define TCA_FQ_CODEL_QUANTUM    6

struct fqcodel {
    struct tc tc;
    uint32_t target;
    uint32_t limit;
    uint32_t interval;
    uint32_t flows;
    uint32_t quantum;
};

static struct fqcodel *
fqcodel_get__(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    return CONTAINER_OF(netdev->tc, struct fqcodel, tc);
}

static void
fqcodel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit,
                  uint32_t interval, uint32_t flows, uint32_t quantum)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct fqcodel *fqcodel;

    fqcodel = xmalloc(sizeof *fqcodel);
    tc_init(&fqcodel->tc, &tc_ops_fqcodel);
    fqcodel->target = target;
    fqcodel->limit = limit;
    fqcodel->interval = interval;
    fqcodel->flows = flows;
    fqcodel->quantum = quantum;

    netdev->tc = &fqcodel->tc;
}

static int
fqcodel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit,
                      uint32_t interval, uint32_t flows, uint32_t quantum)
{
    size_t opt_offset;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    uint32_t otarget, olimit, ointerval, oflows,  oquantum;
    int error;

    tc_del_qdisc(netdev);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    otarget = target ? target : 5000;
    olimit = limit ? limit : 10240;
    ointerval = interval ? interval : 100000;
    oflows = flows ? flows : 1024;
    oquantum = quantum ? quantum : 1514; /* fq_codel default quantum is 1514
                                            not mtu */

    nl_msg_put_string(&request, TCA_KIND, "fq_codel");
    opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    nl_msg_put_u32(&request, TCA_FQ_CODEL_TARGET, otarget);
    nl_msg_put_u32(&request, TCA_FQ_CODEL_LIMIT, olimit);
    nl_msg_put_u32(&request, TCA_FQ_CODEL_INTERVAL, ointerval);
    nl_msg_put_u32(&request, TCA_FQ_CODEL_FLOWS, oflows);
    nl_msg_put_u32(&request, TCA_FQ_CODEL_QUANTUM, oquantum);
    nl_msg_end_nested(&request, opt_offset);

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
        "target %u, limit %u, interval %u, flows %u, quantum %u error %d(%s)",
        netdev_get_name(netdev),
        otarget, olimit, ointerval, oflows, oquantum,
        error, ovs_strerror(error));
    }
    return error;
}

static void
fqcodel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED,
                          const struct smap *details, struct fqcodel *fqcodel)
{
    fqcodel->target = smap_get_ullong(details, "target", 0);
    fqcodel->limit = smap_get_ullong(details, "limit", 0);
    fqcodel->interval = smap_get_ullong(details, "interval", 0);
    fqcodel->flows = smap_get_ullong(details, "flows", 0);
    fqcodel->quantum = smap_get_ullong(details, "quantum", 0);

    if (!fqcodel->target) {
        fqcodel->target = 5000;
    }
    if (!fqcodel->limit) {
        fqcodel->limit = 10240;
    }
    if (!fqcodel->interval) {
        fqcodel->interval = 1000000;
    }
    if (!fqcodel->flows) {
        fqcodel->flows = 1024;
    }
    if (!fqcodel->quantum) {
        fqcodel->quantum = 1514;
    }
}

static int
fqcodel_tc_install(struct netdev *netdev, const struct smap *details)
{
    int error;
    struct fqcodel fqcodel;

    fqcodel_parse_qdisc_details__(netdev, details, &fqcodel);
    error = fqcodel_setup_qdisc__(netdev, fqcodel.target, fqcodel.limit,
                                  fqcodel.interval, fqcodel.flows,
                                  fqcodel.quantum);
    if (!error) {
        fqcodel_install__(netdev, fqcodel.target, fqcodel.limit,
                          fqcodel.interval, fqcodel.flows, fqcodel.quantum);
    }
    return error;
}

static int
fqcodel_parse_tca_options__(struct nlattr *nl_options, struct fqcodel *fqcodel)
{
    static const struct nl_policy tca_fqcodel_policy[] = {
        [TCA_FQ_CODEL_TARGET] = { .type = NL_A_U32 },
        [TCA_FQ_CODEL_LIMIT] = { .type = NL_A_U32 },
        [TCA_FQ_CODEL_INTERVAL] = { .type = NL_A_U32 },
        [TCA_FQ_CODEL_FLOWS] = { .type = NL_A_U32 },
        [TCA_FQ_CODEL_QUANTUM] = { .type = NL_A_U32 }
    };

    struct nlattr *attrs[ARRAY_SIZE(tca_fqcodel_policy)];

    if (!nl_parse_nested(nl_options, tca_fqcodel_policy,
                         attrs, ARRAY_SIZE(tca_fqcodel_policy))) {
        VLOG_WARN_RL(&rl, "failed to parse FQ_CoDel class options");
        return EPROTO;
    }

    fqcodel->target = nl_attr_get_u32(attrs[TCA_FQ_CODEL_TARGET]);
    fqcodel->limit = nl_attr_get_u32(attrs[TCA_FQ_CODEL_LIMIT]);
    fqcodel->interval =nl_attr_get_u32(attrs[TCA_FQ_CODEL_INTERVAL]);
    fqcodel->flows = nl_attr_get_u32(attrs[TCA_FQ_CODEL_FLOWS]);
    fqcodel->quantum = nl_attr_get_u32(attrs[TCA_FQ_CODEL_QUANTUM]);
    return 0;
}

static int
fqcodel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
{
    struct nlattr *nlattr;
    const char * kind;
    int error;
    struct fqcodel fqcodel;

    error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
    if (error != 0) {
        return error;
    }

    error = fqcodel_parse_tca_options__(nlattr, &fqcodel);
    if (error != 0) {
        return error;
    }

    fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval,
                      fqcodel.flows, fqcodel.quantum);
    return 0;
}

static void
fqcodel_tc_destroy(struct tc *tc)
{
    struct fqcodel *fqcodel = CONTAINER_OF(tc, struct fqcodel, tc);
    tc_destroy(tc);
    free(fqcodel);
}

static int
fqcodel_qdisc_get(const struct netdev *netdev, struct smap *details)
{
    const struct fqcodel *fqcodel = fqcodel_get__(netdev);
    smap_add_format(details, "target", "%u", fqcodel->target);
    smap_add_format(details, "limit", "%u", fqcodel->limit);
    smap_add_format(details, "interval", "%u", fqcodel->interval);
    smap_add_format(details, "flows", "%u", fqcodel->flows);
    smap_add_format(details, "quantum", "%u", fqcodel->quantum);
    return 0;
}

static int
fqcodel_qdisc_set(struct netdev *netdev, const struct smap *details)
{
    struct fqcodel fqcodel;

    fqcodel_parse_qdisc_details__(netdev, details, &fqcodel);
    fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval,
                      fqcodel.flows, fqcodel.quantum);
    fqcodel_get__(netdev)->target = fqcodel.target;
    fqcodel_get__(netdev)->limit = fqcodel.limit;
    fqcodel_get__(netdev)->interval = fqcodel.interval;
    fqcodel_get__(netdev)->flows = fqcodel.flows;
    fqcodel_get__(netdev)->quantum = fqcodel.quantum;
    return 0;
}

static const struct tc_ops tc_ops_fqcodel = {
    .linux_name = "fq_codel",
    .ovs_name = "linux-fq_codel",
    .n_queues = FQCODEL_N_QUEUES,
    .tc_install = fqcodel_tc_install,
    .tc_load = fqcodel_tc_load,
    .tc_destroy = fqcodel_tc_destroy,
    .qdisc_get = fqcodel_qdisc_get,
    .qdisc_set = fqcodel_qdisc_set,
};
\f
/* SFQ traffic control class. */

#define SFQ_N_QUEUES 0x0000

struct sfq {
    struct tc tc;
    uint32_t quantum;
    uint32_t perturb;
};

static struct sfq *
sfq_get__(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    return CONTAINER_OF(netdev->tc, struct sfq, tc);
}

static void
sfq_install__(struct netdev *netdev_, uint32_t quantum, uint32_t perturb)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct sfq *sfq;

    sfq = xmalloc(sizeof *sfq);
    tc_init(&sfq->tc, &tc_ops_sfq);
    sfq->perturb = perturb;
    sfq->quantum = quantum;

    netdev->tc = &sfq->tc;
}

static int
sfq_setup_qdisc__(struct netdev *netdev, uint32_t quantum, uint32_t perturb)
{
    struct tc_sfq_qopt opt;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    int mtu;
    int mtu_error, error;
    mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);

    tc_del_qdisc(netdev);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    memset(&opt, 0, sizeof opt);
    if (!quantum) {
        if (!mtu_error) {
            opt.quantum = mtu; /* if we cannot find mtu, use default */
        }
    } else {
        opt.quantum = quantum;
    }

    if (!perturb) {
        opt.perturb_period = 10;
    } else {
        opt.perturb_period = perturb;
    }

    nl_msg_put_string(&request, TCA_KIND, "sfq");
    nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
                     "quantum %u, perturb %u error %d(%s)",
                     netdev_get_name(netdev),
                     opt.quantum, opt.perturb_period,
                     error, ovs_strerror(error));
    }
    return error;
}

static void
sfq_parse_qdisc_details__(struct netdev *netdev,
                          const struct smap *details, struct sfq *sfq)
{
    sfq->perturb = smap_get_ullong(details, "perturb", 0);
    sfq->quantum = smap_get_ullong(details, "quantum", 0);

    if (!sfq->perturb) {
        sfq->perturb = 10;
    }

    if (!sfq->quantum) {
        int mtu;
        if (!netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) {
            sfq->quantum = mtu;
        } else {
            VLOG_WARN_RL(&rl, "when using SFQ, you must specify quantum on a "
                         "device without mtu");
        }
    }
}

static int
sfq_tc_install(struct netdev *netdev, const struct smap *details)
{
    int error;
    struct sfq sfq;

    sfq_parse_qdisc_details__(netdev, details, &sfq);
    error = sfq_setup_qdisc__(netdev, sfq.quantum, sfq.perturb);
    if (!error) {
        sfq_install__(netdev, sfq.quantum, sfq.perturb);
    }
    return error;
}

static int
sfq_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
{
    const struct tc_sfq_qopt *sfq;
    struct nlattr *nlattr;
    const char * kind;
    int error;

    error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
    if (error == 0) {
        sfq = nl_attr_get(nlattr);
        sfq_install__(netdev, sfq->quantum, sfq->perturb_period);
        return 0;
    }

    return error;
}

static void
sfq_tc_destroy(struct tc *tc)
{
    struct sfq *sfq = CONTAINER_OF(tc, struct sfq, tc);
    tc_destroy(tc);
    free(sfq);
}

static int
sfq_qdisc_get(const struct netdev *netdev, struct smap *details)
{
    const struct sfq *sfq = sfq_get__(netdev);
    smap_add_format(details, "quantum", "%u", sfq->quantum);
    smap_add_format(details, "perturb", "%u", sfq->perturb);
    return 0;
}

static int
sfq_qdisc_set(struct netdev *netdev, const struct smap *details)
{
    struct sfq sfq;

    sfq_parse_qdisc_details__(netdev, details, &sfq);
    sfq_install__(netdev, sfq.quantum, sfq.perturb);
    sfq_get__(netdev)->quantum = sfq.quantum;
    sfq_get__(netdev)->perturb = sfq.perturb;
    return 0;
}

static const struct tc_ops tc_ops_sfq = {
    .linux_name = "sfq",
    .ovs_name = "linux-sfq",
    .n_queues = SFQ_N_QUEUES,
    .tc_install = sfq_tc_install,
    .tc_load = sfq_tc_load,
    .tc_destroy = sfq_tc_destroy,
    .qdisc_get = sfq_qdisc_get,
    .qdisc_set = sfq_qdisc_set,
};
\f
/* netem traffic control class. */

struct netem {
    struct tc tc;
    uint32_t latency;
    uint32_t limit;
    uint32_t loss;
};

static struct netem *
netem_get__(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    return CONTAINER_OF(netdev->tc, struct netem, tc);
}

static void
netem_install__(struct netdev *netdev_, uint32_t latency,
                uint32_t limit, uint32_t loss)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct netem *netem;

    netem = xmalloc(sizeof *netem);
    tc_init(&netem->tc, &tc_ops_netem);
    netem->latency = latency;
    netem->limit = limit;
    netem->loss = loss;

    netdev->tc = &netem->tc;
}

static int
netem_setup_qdisc__(struct netdev *netdev, uint32_t latency,
                    uint32_t limit, uint32_t loss)
{
    struct tc_netem_qopt opt;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    int error;

    tc_del_qdisc(netdev);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    memset(&opt, 0, sizeof opt);

    if (!limit) {
        opt.limit = 1000;
    } else {
        opt.limit = limit;
    }

    if (loss) {
        if (loss > 100) {
            VLOG_WARN_RL(&rl,
                         "loss should be a percentage value between 0 to 100, "
                         "loss was %u", loss);
            return EINVAL;
        }
        opt.loss = floor(UINT32_MAX * (loss / 100.0));
    }

    opt.latency = tc_time_to_ticks(latency);

    nl_msg_put_string(&request, TCA_KIND, "netem");
    nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
                          "latency %u, limit %u, loss %u error %d(%s)",
                     netdev_get_name(netdev),
                     opt.latency, opt.limit, opt.loss,
                     error, ovs_strerror(error));
    }
    return error;
}

static void
netem_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED,
                          const struct smap *details, struct netem *netem)
{
    netem->latency = smap_get_ullong(details, "latency", 0);
    netem->limit = smap_get_ullong(details, "limit", 0);
    netem->loss = smap_get_ullong(details, "loss", 0);

    if (!netem->limit) {
        netem->limit = 1000;
    }
}

static int
netem_tc_install(struct netdev *netdev, const struct smap *details)
{
    int error;
    struct netem netem;

    netem_parse_qdisc_details__(netdev, details, &netem);
    error = netem_setup_qdisc__(netdev, netem.latency,
                                netem.limit, netem.loss);
    if (!error) {
        netem_install__(netdev, netem.latency, netem.limit, netem.loss);
    }
    return error;
}

static int
netem_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
{
    const struct tc_netem_qopt *netem;
    struct nlattr *nlattr;
    const char *kind;
    int error;

    error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
    if (error == 0) {
        netem = nl_attr_get(nlattr);
        netem_install__(netdev, netem->latency, netem->limit, netem->loss);
        return 0;
    }

    return error;
}

static void
netem_tc_destroy(struct tc *tc)
{
    struct netem *netem = CONTAINER_OF(tc, struct netem, tc);
    tc_destroy(tc);
    free(netem);
}

static int
netem_qdisc_get(const struct netdev *netdev, struct smap *details)
{
    const struct netem *netem = netem_get__(netdev);
    smap_add_format(details, "latency", "%u", netem->latency);
    smap_add_format(details, "limit", "%u", netem->limit);
    smap_add_format(details, "loss", "%u", netem->loss);
    return 0;
}

static int
netem_qdisc_set(struct netdev *netdev, const struct smap *details)
{
    struct netem netem;

    netem_parse_qdisc_details__(netdev, details, &netem);
    netem_install__(netdev, netem.latency, netem.limit, netem.loss);
    netem_get__(netdev)->latency = netem.latency;
    netem_get__(netdev)->limit = netem.limit;
    netem_get__(netdev)->loss = netem.loss;
    return 0;
}

static const struct tc_ops tc_ops_netem = {
    .linux_name = "netem",
    .ovs_name = "linux-netem",
    .n_queues = 0,
    .tc_install = netem_tc_install,
    .tc_load = netem_tc_load,
    .tc_destroy = netem_tc_destroy,
    .qdisc_get = netem_qdisc_get,
    .qdisc_set = netem_qdisc_set,
};
\f
/* HTB traffic control class. */

#define HTB_N_QUEUES 0xf000
#define HTB_RATE2QUANTUM 10

struct htb {
    struct tc tc;
    unsigned int max_rate;      /* In bytes/s. */
};

struct htb_class {
    struct tc_queue tc_queue;
    unsigned int min_rate;      /* In bytes/s. */
    unsigned int max_rate;      /* In bytes/s. */
    unsigned int burst;         /* In bytes. */
    unsigned int priority;      /* Lower values are higher priorities. */
};

static struct htb *
htb_get__(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    return CONTAINER_OF(netdev->tc, struct htb, tc);
}

static void
htb_install__(struct netdev *netdev_, uint64_t max_rate)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct htb *htb;

    htb = xmalloc(sizeof *htb);
    tc_init(&htb->tc, &tc_ops_htb);
    htb->max_rate = max_rate;

    netdev->tc = &htb->tc;
}

/* Create an HTB qdisc.
 *
 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
static int
htb_setup_qdisc__(struct netdev *netdev)
{
    size_t opt_offset;
    struct tc_htb_glob opt;
    struct ofpbuf request;
    struct tcmsg *tcmsg;

    tc_del_qdisc(netdev);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    nl_msg_put_string(&request, TCA_KIND, "htb");

    memset(&opt, 0, sizeof opt);
    opt.rate2quantum = HTB_RATE2QUANTUM;
    opt.version = 3;
    opt.defcls = 1;

    opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
    nl_msg_end_nested(&request, opt_offset);

    return tc_transact(&request, NULL);
}

/* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
static int
htb_setup_class__(struct netdev *netdev, unsigned int handle,
                  unsigned int parent, struct htb_class *class)
{
    size_t opt_offset;
    struct tc_htb_opt opt;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    int error;
    int mtu;

    error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
    if (error) {
        VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
                     netdev_get_name(netdev));
        return error;
    }

    memset(&opt, 0, sizeof opt);
    tc_fill_rate(&opt.rate, class->min_rate, mtu);
    tc_fill_rate(&opt.ceil, class->max_rate, mtu);
    /* Makes sure the quantum is at least MTU.  Setting quantum will
     * make htb ignore the r2q for this class. */
    if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) {
        opt.quantum = mtu;
    }
    opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
    opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
    opt.prio = class->priority;

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE,
                                         &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = handle;
    tcmsg->tcm_parent = parent;

    nl_msg_put_string(&request, TCA_KIND, "htb");
    opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
    tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
    tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
    nl_msg_end_nested(&request, opt_offset);

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
                     "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
                     netdev_get_name(netdev),
                     tc_get_major(handle), tc_get_minor(handle),
                     tc_get_major(parent), tc_get_minor(parent),
                     class->min_rate, class->max_rate,
                     class->burst, class->priority, ovs_strerror(error));
    }
    return error;
}

/* Parses Netlink attributes in 'options' for HTB parameters and stores a
 * description of them into 'details'.  The description complies with the
 * specification given in the vswitch database documentation for linux-htb
 * queue details. */
static int
htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
{
    static const struct nl_policy tca_htb_policy[] = {
        [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
                            .min_len = sizeof(struct tc_htb_opt) },
    };

    struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
    const struct tc_htb_opt *htb;

    if (!nl_parse_nested(nl_options, tca_htb_policy,
                         attrs, ARRAY_SIZE(tca_htb_policy))) {
        VLOG_WARN_RL(&rl, "failed to parse HTB class options");
        return EPROTO;
    }

    htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
    class->min_rate = htb->rate.rate;
    class->max_rate = htb->ceil.rate;
    class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
    class->priority = htb->prio;
    return 0;
}

static int
htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
                  struct htb_class *options,
                  struct netdev_queue_stats *stats)
{
    struct nlattr *nl_options;
    unsigned int handle;
    int error;

    error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
    if (!error && queue_id) {
        unsigned int major = tc_get_major(handle);
        unsigned int minor = tc_get_minor(handle);
        if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
            *queue_id = minor - 1;
        } else {
            error = EPROTO;
        }
    }
    if (!error && options) {
        error = htb_parse_tca_options__(nl_options, options);
    }
    return error;
}

static void
htb_parse_qdisc_details__(struct netdev *netdev_,
                          const struct smap *details, struct htb_class *hc)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    hc->max_rate = smap_get_ullong(details, "max-rate", 0) / 8;
    if (!hc->max_rate) {
        enum netdev_features current;

        netdev_linux_read_features(netdev);
        current = !netdev->get_features_error ? netdev->current : 0;
        hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
    }
    hc->min_rate = hc->max_rate;
    hc->burst = 0;
    hc->priority = 0;
}

static int
htb_parse_class_details__(struct netdev *netdev,
                          const struct smap *details, struct htb_class *hc)
{
    const struct htb *htb = htb_get__(netdev);
    int mtu, error;
    unsigned long long int max_rate_bit;

    error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
    if (error) {
        VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
                     netdev_get_name(netdev));
        return error;
    }

    /* HTB requires at least an mtu sized min-rate to send any traffic even
     * on uncongested links. */
    hc->min_rate = smap_get_ullong(details, "min-rate", 0) / 8;
    hc->min_rate = MAX(hc->min_rate, mtu);
    hc->min_rate = MIN(hc->min_rate, htb->max_rate);

    /* max-rate */
    max_rate_bit = smap_get_ullong(details, "max-rate", 0);
    hc->max_rate = max_rate_bit ? max_rate_bit / 8 : htb->max_rate;
    hc->max_rate = MAX(hc->max_rate, hc->min_rate);
    hc->max_rate = MIN(hc->max_rate, htb->max_rate);

    /* burst
     *
     * According to hints in the documentation that I've read, it is important
     * that 'burst' be at least as big as the largest frame that might be
     * transmitted.  Also, making 'burst' a bit bigger than necessary is OK,
     * but having it a bit too small is a problem.  Since netdev_get_mtu()
     * doesn't include the Ethernet header, we need to add at least 14 (18?) to
     * the MTU.  We actually add 64, instead of 14, as a guard against
     * additional headers get tacked on somewhere that we're not aware of. */
    hc->burst = smap_get_ullong(details, "burst", 0) / 8;
    hc->burst = MAX(hc->burst, mtu + 64);

    /* priority */
    hc->priority = smap_get_ullong(details, "priority", 0);

    return 0;
}

static int
htb_query_class__(const struct netdev *netdev, unsigned int handle,
                  unsigned int parent, struct htb_class *options,
                  struct netdev_queue_stats *stats)
{
    struct ofpbuf *reply;
    int error;

    error = tc_query_class(netdev, handle, parent, &reply);
    if (!error) {
        error = htb_parse_tcmsg__(reply, NULL, options, stats);
        ofpbuf_delete(reply);
    }
    return error;
}

static int
htb_tc_install(struct netdev *netdev, const struct smap *details)
{
    int error;

    error = htb_setup_qdisc__(netdev);
    if (!error) {
        struct htb_class hc;

        htb_parse_qdisc_details__(netdev, details, &hc);
        error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
                                  tc_make_handle(1, 0), &hc);
        if (!error) {
            htb_install__(netdev, hc.max_rate);
        }
    }
    return error;
}

static struct htb_class *
htb_class_cast__(const struct tc_queue *queue)
{
    return CONTAINER_OF(queue, struct htb_class, tc_queue);
}

static void
htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
                   const struct htb_class *hc)
{
    struct htb *htb = htb_get__(netdev);
    size_t hash = hash_int(queue_id, 0);
    struct tc_queue *queue;
    struct htb_class *hcp;

    queue = tc_find_queue__(netdev, queue_id, hash);
    if (queue) {
        hcp = htb_class_cast__(queue);
    } else {
        hcp = xmalloc(sizeof *hcp);
        queue = &hcp->tc_queue;
        queue->queue_id = queue_id;
        queue->created = time_msec();
        hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
    }

    hcp->min_rate = hc->min_rate;
    hcp->max_rate = hc->max_rate;
    hcp->burst = hc->burst;
    hcp->priority = hc->priority;
}

static int
htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
    struct ofpbuf msg;
    struct queue_dump_state state;
    struct htb_class hc;

    /* Get qdisc options. */
    hc.max_rate = 0;
    htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
    htb_install__(netdev, hc.max_rate);

    /* Get queues. */
    if (!start_queue_dump(netdev, &state)) {
        return ENODEV;
    }
    while (nl_dump_next(&state.dump, &msg, &state.buf)) {
        unsigned int queue_id;

        if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
            htb_update_queue__(netdev, queue_id, &hc);
        }
    }
    finish_queue_dump(&state);

    return 0;
}

static void
htb_tc_destroy(struct tc *tc)
{
    struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
    struct htb_class *hc;

    HMAP_FOR_EACH_POP (hc, tc_queue.hmap_node, &htb->tc.queues) {
        free(hc);
    }
    tc_destroy(tc);
    free(htb);
}

static int
htb_qdisc_get(const struct netdev *netdev, struct smap *details)
{
    const struct htb *htb = htb_get__(netdev);
    smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate);
    return 0;
}

static int
htb_qdisc_set(struct netdev *netdev, const struct smap *details)
{
    struct htb_class hc;
    int error;

    htb_parse_qdisc_details__(netdev, details, &hc);
    error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
                              tc_make_handle(1, 0), &hc);
    if (!error) {
        htb_get__(netdev)->max_rate = hc.max_rate;
    }
    return error;
}

static int
htb_class_get(const struct netdev *netdev OVS_UNUSED,
              const struct tc_queue *queue, struct smap *details)
{
    const struct htb_class *hc = htb_class_cast__(queue);

    smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
    if (hc->min_rate != hc->max_rate) {
        smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
    }
    smap_add_format(details, "burst", "%llu", 8ULL * hc->burst);
    if (hc->priority) {
        smap_add_format(details, "priority", "%u", hc->priority);
    }
    return 0;
}

static int
htb_class_set(struct netdev *netdev, unsigned int queue_id,
              const struct smap *details)
{
    struct htb_class hc;
    int error;

    error = htb_parse_class_details__(netdev, details, &hc);
    if (error) {
        return error;
    }

    error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
                              tc_make_handle(1, 0xfffe), &hc);
    if (error) {
        return error;
    }

    htb_update_queue__(netdev, queue_id, &hc);
    return 0;
}

static int
htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
{
    struct htb_class *hc = htb_class_cast__(queue);
    struct htb *htb = htb_get__(netdev);
    int error;

    error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
    if (!error) {
        hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
        free(hc);
    }
    return error;
}

static int
htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
                    struct netdev_queue_stats *stats)
{
    return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
                             tc_make_handle(1, 0xfffe), NULL, stats);
}

static int
htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
                     const struct ofpbuf *nlmsg,
                     netdev_dump_queue_stats_cb *cb, void *aux)
{
    struct netdev_queue_stats stats;
    unsigned int handle, major, minor;
    int error;

    error = tc_parse_class(nlmsg, &handle, NULL, &stats);
    if (error) {
        return error;
    }

    major = tc_get_major(handle);
    minor = tc_get_minor(handle);
    if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
        (*cb)(minor - 1, &stats, aux);
    }
    return 0;
}

static const struct tc_ops tc_ops_htb = {
    .linux_name = "htb",
    .ovs_name = "linux-htb",
    .n_queues = HTB_N_QUEUES,
    .tc_install = htb_tc_install,
    .tc_load = htb_tc_load,
    .tc_destroy = htb_tc_destroy,
    .qdisc_get = htb_qdisc_get,
    .qdisc_set = htb_qdisc_set,
    .class_get = htb_class_get,
    .class_set = htb_class_set,
    .class_delete = htb_class_delete,
    .class_get_stats = htb_class_get_stats,
    .class_dump_stats = htb_class_dump_stats
};
\f
/* "linux-hfsc" traffic control class. */

#define HFSC_N_QUEUES 0xf000

struct hfsc {
    struct tc tc;
    uint32_t max_rate;
};

struct hfsc_class {
    struct tc_queue tc_queue;
    uint32_t min_rate;
    uint32_t max_rate;
};

static struct hfsc *
hfsc_get__(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    return CONTAINER_OF(netdev->tc, struct hfsc, tc);
}

static struct hfsc_class *
hfsc_class_cast__(const struct tc_queue *queue)
{
    return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
}

static void
hfsc_install__(struct netdev *netdev_, uint32_t max_rate)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct hfsc *hfsc;

    hfsc = xmalloc(sizeof *hfsc);
    tc_init(&hfsc->tc, &tc_ops_hfsc);
    hfsc->max_rate = max_rate;
    netdev->tc = &hfsc->tc;
}

static void
hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
                    const struct hfsc_class *hc)
{
    size_t hash;
    struct hfsc *hfsc;
    struct hfsc_class *hcp;
    struct tc_queue *queue;

    hfsc = hfsc_get__(netdev);
    hash = hash_int(queue_id, 0);

    queue = tc_find_queue__(netdev, queue_id, hash);
    if (queue) {
        hcp = hfsc_class_cast__(queue);
    } else {
        hcp             = xmalloc(sizeof *hcp);
        queue           = &hcp->tc_queue;
        queue->queue_id = queue_id;
        queue->created  = time_msec();
        hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
    }

    hcp->min_rate = hc->min_rate;
    hcp->max_rate = hc->max_rate;
}

static int
hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
{
    const struct tc_service_curve *rsc, *fsc, *usc;
    static const struct nl_policy tca_hfsc_policy[] = {
        [TCA_HFSC_RSC] = {
            .type      = NL_A_UNSPEC,
            .optional  = false,
            .min_len   = sizeof(struct tc_service_curve),
        },
        [TCA_HFSC_FSC] = {
            .type      = NL_A_UNSPEC,
            .optional  = false,
            .min_len   = sizeof(struct tc_service_curve),
        },
        [TCA_HFSC_USC] = {
            .type      = NL_A_UNSPEC,
            .optional  = false,
            .min_len   = sizeof(struct tc_service_curve),
        },
    };
    struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];

    if (!nl_parse_nested(nl_options, tca_hfsc_policy,
                         attrs, ARRAY_SIZE(tca_hfsc_policy))) {
        VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
        return EPROTO;
    }

    rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
    fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
    usc = nl_attr_get(attrs[TCA_HFSC_USC]);

    if (rsc->m1 != 0 || rsc->d != 0 ||
        fsc->m1 != 0 || fsc->d != 0 ||
        usc->m1 != 0 || usc->d != 0) {
        VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
                     "Non-linear service curves are not supported.");
        return EPROTO;
    }

    if (rsc->m2 != fsc->m2) {
        VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
                     "Real-time service curves are not supported ");
        return EPROTO;
    }

    if (rsc->m2 > usc->m2) {
        VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
                     "Min-rate service curve is greater than "
                     "the max-rate service curve.");
        return EPROTO;
    }

    class->min_rate = fsc->m2;
    class->max_rate = usc->m2;
    return 0;
}

static int
hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
                   struct hfsc_class *options,
                   struct netdev_queue_stats *stats)
{
    int error;
    unsigned int handle;
    struct nlattr *nl_options;

    error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
    if (error) {
        return error;
    }

    if (queue_id) {
        unsigned int major, minor;

        major = tc_get_major(handle);
        minor = tc_get_minor(handle);
        if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
            *queue_id = minor - 1;
        } else {
            return EPROTO;
        }
    }

    if (options) {
        error = hfsc_parse_tca_options__(nl_options, options);
    }

    return error;
}

static int
hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
                   unsigned int parent, struct hfsc_class *options,
                   struct netdev_queue_stats *stats)
{
    int error;
    struct ofpbuf *reply;

    error = tc_query_class(netdev, handle, parent, &reply);
    if (error) {
        return error;
    }

    error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
    ofpbuf_delete(reply);
    return error;
}

static void
hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details,
                           struct hfsc_class *class)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    uint32_t max_rate = smap_get_ullong(details, "max-rate", 0) / 8;
    if (!max_rate) {
        enum netdev_features current;

        netdev_linux_read_features(netdev);
        current = !netdev->get_features_error ? netdev->current : 0;
        max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
    }

    class->min_rate = max_rate;
    class->max_rate = max_rate;
}

static int
hfsc_parse_class_details__(struct netdev *netdev,
                           const struct smap *details,
                           struct hfsc_class * class)
{
    const struct hfsc *hfsc;
    uint32_t min_rate, max_rate;

    hfsc       = hfsc_get__(netdev);

    min_rate = smap_get_ullong(details, "min-rate", 0) / 8;
    min_rate = MAX(min_rate, 1);
    min_rate = MIN(min_rate, hfsc->max_rate);

    max_rate = smap_get_ullong(details, "max-rate", hfsc->max_rate * 8) / 8;
    max_rate = MAX(max_rate, min_rate);
    max_rate = MIN(max_rate, hfsc->max_rate);

    class->min_rate = min_rate;
    class->max_rate = max_rate;

    return 0;
}

/* Create an HFSC qdisc.
 *
 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
static int
hfsc_setup_qdisc__(struct netdev * netdev)
{
    struct tcmsg *tcmsg;
    struct ofpbuf request;
    struct tc_hfsc_qopt opt;

    tc_del_qdisc(netdev);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);

    if (!tcmsg) {
        return ENODEV;
    }

    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    memset(&opt, 0, sizeof opt);
    opt.defcls = 1;

    nl_msg_put_string(&request, TCA_KIND, "hfsc");
    nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);

    return tc_transact(&request, NULL);
}

/* Create an HFSC class.
 *
 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
 * sc rate <min_rate> ul rate <max_rate>" */
static int
hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
                   unsigned int parent, struct hfsc_class *class)
{
    int error;
    size_t opt_offset;
    struct tcmsg *tcmsg;
    struct ofpbuf request;
    struct tc_service_curve min, max;

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE,
                                         &request);

    if (!tcmsg) {
        return ENODEV;
    }

    tcmsg->tcm_handle = handle;
    tcmsg->tcm_parent = parent;

    min.m1 = 0;
    min.d  = 0;
    min.m2 = class->min_rate;

    max.m1 = 0;
    max.d  = 0;
    max.m2 = class->max_rate;

    nl_msg_put_string(&request, TCA_KIND, "hfsc");
    opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
    nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
    nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
    nl_msg_end_nested(&request, opt_offset);

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
                     "min-rate %ubps, max-rate %ubps (%s)",
                     netdev_get_name(netdev),
                     tc_get_major(handle), tc_get_minor(handle),
                     tc_get_major(parent), tc_get_minor(parent),
                     class->min_rate, class->max_rate, ovs_strerror(error));
    }

    return error;
}

static int
hfsc_tc_install(struct netdev *netdev, const struct smap *details)
{
    int error;
    struct hfsc_class class;

    error = hfsc_setup_qdisc__(netdev);

    if (error) {
        return error;
    }

    hfsc_parse_qdisc_details__(netdev, details, &class);
    error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
                               tc_make_handle(1, 0), &class);

    if (error) {
        return error;
    }

    hfsc_install__(netdev, class.max_rate);
    return 0;
}

static int
hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
    struct ofpbuf msg;
    struct queue_dump_state state;
    struct hfsc_class hc;

    hc.max_rate = 0;
    hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
    hfsc_install__(netdev, hc.max_rate);

    if (!start_queue_dump(netdev, &state)) {
        return ENODEV;
    }

    while (nl_dump_next(&state.dump, &msg, &state.buf)) {
        unsigned int queue_id;

        if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
            hfsc_update_queue__(netdev, queue_id, &hc);
        }
    }

    finish_queue_dump(&state);
    return 0;
}

static void
hfsc_tc_destroy(struct tc *tc)
{
    struct hfsc *hfsc;
    struct hfsc_class *hc, *next;

    hfsc = CONTAINER_OF(tc, struct hfsc, tc);

    HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
        hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
        free(hc);
    }

    tc_destroy(tc);
    free(hfsc);
}

static int
hfsc_qdisc_get(const struct netdev *netdev, struct smap *details)
{
    const struct hfsc *hfsc;
    hfsc = hfsc_get__(netdev);
    smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate);
    return 0;
}

static int
hfsc_qdisc_set(struct netdev *netdev, const struct smap *details)
{
    int error;
    struct hfsc_class class;

    hfsc_parse_qdisc_details__(netdev, details, &class);
    error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
                               tc_make_handle(1, 0), &class);

    if (!error) {
        hfsc_get__(netdev)->max_rate = class.max_rate;
    }

    return error;
}

static int
hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
              const struct tc_queue *queue, struct smap *details)
{
    const struct hfsc_class *hc;

    hc = hfsc_class_cast__(queue);
    smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
    if (hc->min_rate != hc->max_rate) {
        smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
    }
    return 0;
}

static int
hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
               const struct smap *details)
{
    int error;
    struct hfsc_class class;

    error = hfsc_parse_class_details__(netdev, details, &class);
    if (error) {
        return error;
    }

    error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
                               tc_make_handle(1, 0xfffe), &class);
    if (error) {
        return error;
    }

    hfsc_update_queue__(netdev, queue_id, &class);
    return 0;
}

static int
hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
{
    int error;
    struct hfsc *hfsc;
    struct hfsc_class *hc;

    hc   = hfsc_class_cast__(queue);
    hfsc = hfsc_get__(netdev);

    error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
    if (!error) {
        hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
        free(hc);
    }
    return error;
}

static int
hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
                     struct netdev_queue_stats *stats)
{
    return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
                             tc_make_handle(1, 0xfffe), NULL, stats);
}

static int
hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
                      const struct ofpbuf *nlmsg,
                      netdev_dump_queue_stats_cb *cb, void *aux)
{
    struct netdev_queue_stats stats;
    unsigned int handle, major, minor;
    int error;

    error = tc_parse_class(nlmsg, &handle, NULL, &stats);
    if (error) {
        return error;
    }

    major = tc_get_major(handle);
    minor = tc_get_minor(handle);
    if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
        (*cb)(minor - 1, &stats, aux);
    }
    return 0;
}

static const struct tc_ops tc_ops_hfsc = {
    .linux_name = "hfsc",
    .ovs_name = "linux-hfsc",
    .n_queues = HFSC_N_QUEUES,              /* n_queues */
    .tc_install = hfsc_tc_install,
    .tc_load = hfsc_tc_load,
    .tc_destroy = hfsc_tc_destroy,
    .qdisc_get = hfsc_qdisc_get,
    .qdisc_set = hfsc_qdisc_set,
    .class_get = hfsc_class_get,
    .class_set = hfsc_class_set,
    .class_delete = hfsc_class_delete,
    .class_get_stats = hfsc_class_get_stats,
    .class_dump_stats = hfsc_class_dump_stats,
};
\f
/* "linux-noop" traffic control class. */

static void
noop_install__(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default);

    netdev->tc = CONST_CAST(struct tc *, &tc);
}

static int
noop_tc_install(struct netdev *netdev,
                   const struct smap *details OVS_UNUSED)
{
    noop_install__(netdev);
    return 0;
}

static int
noop_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
    noop_install__(netdev);
    return 0;
}

static const struct tc_ops tc_ops_noop = {
    .ovs_name = "linux-noop",               /* ovs_name */
    .tc_install = noop_tc_install,
    .tc_load = noop_tc_load,
};
\f
/* "linux-default" traffic control class.
 *
 * This class represents the default, unnamed Linux qdisc.  It corresponds to
 * the "" (empty string) QoS type in the OVS database. */

static void
default_install__(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default);

    /* Nothing but a tc class implementation is allowed to write to a tc.  This
     * class never does that, so we can legitimately use a const tc object. */
    netdev->tc = CONST_CAST(struct tc *, &tc);
}

static int
default_tc_install(struct netdev *netdev,
                   const struct smap *details OVS_UNUSED)
{
    default_install__(netdev);
    return 0;
}

static int
default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
    default_install__(netdev);
    return 0;
}

static const struct tc_ops tc_ops_default = {
    .ovs_name = "",                         /* ovs_name */
    .tc_install = default_tc_install,
    .tc_load = default_tc_load,
};
\f
/* "linux-other" traffic control class.
 *
 * */

static int
other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other);

    /* Nothing but a tc class implementation is allowed to write to a tc.  This
     * class never does that, so we can legitimately use a const tc object. */
    netdev->tc = CONST_CAST(struct tc *, &tc);
    return 0;
}

static const struct tc_ops tc_ops_other = {
    .ovs_name = "linux-other",
    .tc_load = other_tc_load,
};
\f
/* Traffic control. */

/* Number of kernel "tc" ticks per second. */
static double ticks_per_s;

/* Number of kernel "jiffies" per second.  This is used for the purpose of
 * computing buffer sizes.  Generally kernel qdiscs need to be able to buffer
 * one jiffy's worth of data.
 *
 * There are two possibilities here:
 *
 *    - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
 *      approximate range of 100 to 1024.  That means that we really need to
 *      make sure that the qdisc can buffer that much data.
 *
 *    - 'buffer_hz' is an absurdly large number.  That means that the kernel
 *      has finely granular timers and there's no need to fudge additional room
 *      for buffers.  (There's no extra effort needed to implement that: the
 *      large 'buffer_hz' is used as a divisor, so practically any number will
 *      come out as 0 in the division.  Small integer results in the case of
 *      really high dividends won't have any real effect anyhow.)
 */
static unsigned int buffer_hz;

static struct tcmsg *
netdev_linux_tc_make_request(const struct netdev *netdev, int type,
                             unsigned int flags, struct ofpbuf *request)
{
    int ifindex;
    int error;

    error = get_ifindex(netdev, &ifindex);
    if (error) {
        return NULL;
    }

    return tc_make_request(ifindex, type, flags, request);
}

/* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
 * of 'kbits_burst'.
 *
 * This function is equivalent to running:
 *     /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
 *              basic police rate <kbits_rate>kbit burst <kbits_burst>k
 *              mtu 65535 drop
 *
 * The configuration and stats may be seen with the following command:
 *     /sbin/tc -s filter show dev <devname> parent ffff:
 *
 * Returns 0 if successful, otherwise a positive errno value.
 */
static int
tc_add_policer(struct netdev *netdev,
               uint32_t kbits_rate, uint32_t kbits_burst)
{
    struct tc_police tc_police;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    size_t basic_offset;
    size_t police_offset;
    int error;
    int mtu = 65535;

    memset(&tc_police, 0, sizeof tc_police);
    tc_police.action = TC_POLICE_SHOT;
    tc_police.mtu = mtu;
    tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu);

    /* The following appears wrong in one way: In networking a kilobit is
     * usually 1000 bits but this uses 1024 bits.
     *
     * However if you "fix" those problems then "tc filter show ..." shows
     * "125000b", meaning 125,000 bits, when OVS configures it for 1000 kbit ==
     * 1,000,000 bits, whereas this actually ends up doing the right thing from
     * tc's point of view.  Whatever. */
    tc_police.burst = tc_bytes_to_ticks(
        tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 8);

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTFILTER,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
    tcmsg->tcm_info = tc_make_handle(49,
                                     (OVS_FORCE uint16_t) htons(ETH_P_ALL));

    nl_msg_put_string(&request, TCA_KIND, "basic");
    basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
    nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
    tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
    nl_msg_end_nested(&request, police_offset);
    nl_msg_end_nested(&request, basic_offset);

    error = tc_transact(&request, NULL);
    if (error) {
        return error;
    }

    return 0;
}

static void
read_psched(void)
{
    /* The values in psched are not individually very meaningful, but they are
     * important.  The tables below show some values seen in the wild.
     *
     * Some notes:
     *
     *   - "c" has always been a constant 1000000 since at least Linux 2.4.14.
     *     (Before that, there are hints that it was 1000000000.)
     *
     *   - "d" can be unrealistically large, see the comment on 'buffer_hz'
     *     above.
     *
     *                        /proc/net/psched
     *     -----------------------------------
     * [1] 000c8000 000f4240 000f4240 00000064
     * [2] 000003e8 00000400 000f4240 3b9aca00
     * [3] 000003e8 00000400 000f4240 3b9aca00
     * [4] 000003e8 00000400 000f4240 00000064
     * [5] 000003e8 00000040 000f4240 3b9aca00
     * [6] 000003e8 00000040 000f4240 000000f9
     *
     *           a         b          c             d ticks_per_s     buffer_hz
     *     ------- --------- ---------- ------------- ----------- -------------
     * [1] 819,200 1,000,000  1,000,000           100     819,200           100
     * [2]   1,000     1,024  1,000,000 1,000,000,000     976,562 1,000,000,000
     * [3]   1,000     1,024  1,000,000 1,000,000,000     976,562 1,000,000,000
     * [4]   1,000     1,024  1,000,000           100     976,562           100
     * [5]   1,000        64  1,000,000 1,000,000,000  15,625,000 1,000,000,000
     * [6]   1,000        64  1,000,000           249  15,625,000           249
     *
     * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
     * [2] 2.6.26-1-686-bigmem from Debian lenny
     * [3] 2.6.26-2-sparc64 from Debian lenny
     * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
     * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
     * [6] 2.6.34 from kernel.org on KVM
     */
    static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
    static const char fn[] = "/proc/net/psched";
    unsigned int a, b, c, d;
    FILE *stream;

    if (!ovsthread_once_start(&once)) {
        return;
    }

    ticks_per_s = 1.0;
    buffer_hz = 100;

    stream = fopen(fn, "r");
    if (!stream) {
        VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno));
        goto exit;
    }

    if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
        VLOG_WARN("%s: read failed", fn);
        fclose(stream);
        goto exit;
    }
    VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
    fclose(stream);

    if (!a || !b || !c) {
        VLOG_WARN("%s: invalid scheduler parameters", fn);
        goto exit;
    }

    ticks_per_s = (double) a * c / b;
    if (c == 1000000) {
        buffer_hz = d;
    } else {
        VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
                  fn, a, b, c, d);
    }
    VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);

exit:
    ovsthread_once_done(&once);
}

/* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
 * rate of 'rate' bytes per second. */
static unsigned int
tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
{
    read_psched();
    return (rate * ticks) / ticks_per_s;
}

/* Returns the number of ticks that it would take to transmit 'size' bytes at a
 * rate of 'rate' bytes per second. */
static unsigned int
tc_bytes_to_ticks(unsigned int rate, unsigned int size)
{
    read_psched();
    return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
}

/* Returns the number of bytes that need to be reserved for qdisc buffering at
 * a transmission rate of 'rate' bytes per second. */
static unsigned int
tc_buffer_per_jiffy(unsigned int rate)
{
    read_psched();
    return rate / buffer_hz;
}

static uint32_t
tc_time_to_ticks(uint32_t time) {
    read_psched();
    return time * (ticks_per_s / 1000000);
}

/* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
 * e.g. "htb", into '*kind' (if it is nonnull).  If 'options' is nonnull,
 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
 * stores NULL into it if it is absent.
 *
 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
 * 'msg'.
 *
 * Returns 0 if successful, otherwise a positive errno value. */
static int
tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
               struct nlattr **options)
{
    static const struct nl_policy tca_policy[] = {
        [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
        [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
    };
    struct nlattr *ta[ARRAY_SIZE(tca_policy)];

    if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
                         tca_policy, ta, ARRAY_SIZE(ta))) {
        VLOG_WARN_RL(&rl, "failed to parse qdisc message");
        goto error;
    }

    if (kind) {
        *kind = nl_attr_get_string(ta[TCA_KIND]);
    }

    if (options) {
        *options = ta[TCA_OPTIONS];
    }

    return 0;

error:
    if (kind) {
        *kind = NULL;
    }
    if (options) {
        *options = NULL;
    }
    return EPROTO;
}

/* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
 * into '*options', and its queue statistics into '*stats'.  Any of the output
 * arguments may be null.
 *
 * Returns 0 if successful, otherwise a positive errno value. */
static int
tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
               struct nlattr **options, struct netdev_queue_stats *stats)
{
    static const struct nl_policy tca_policy[] = {
        [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
        [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
    };
    struct nlattr *ta[ARRAY_SIZE(tca_policy)];

    if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
                         tca_policy, ta, ARRAY_SIZE(ta))) {
        VLOG_WARN_RL(&rl, "failed to parse class message");
        goto error;
    }

    if (handlep) {
        struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
        *handlep = tc->tcm_handle;
    }

    if (options) {
        *options = ta[TCA_OPTIONS];
    }

    if (stats) {
        const struct gnet_stats_queue *gsq;
        struct gnet_stats_basic gsb;

        static const struct nl_policy stats_policy[] = {
            [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
                                  .min_len = sizeof gsb },
            [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
                                  .min_len = sizeof *gsq },
        };
        struct nlattr *sa[ARRAY_SIZE(stats_policy)];

        if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
                             sa, ARRAY_SIZE(sa))) {
            VLOG_WARN_RL(&rl, "failed to parse class stats");
            goto error;
        }

        /* Alignment issues screw up the length of struct gnet_stats_basic on
         * some arch/bitsize combinations.  Newer versions of Linux have a
         * struct gnet_stats_basic_packed, but we can't depend on that.  The
         * easiest thing to do is just to make a copy. */
        memset(&gsb, 0, sizeof gsb);
        memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
               MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
        stats->tx_bytes = gsb.bytes;
        stats->tx_packets = gsb.packets;

        gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
        stats->tx_errors = gsq->drops;
    }

    return 0;

error:
    if (options) {
        *options = NULL;
    }
    if (stats) {
        memset(stats, 0, sizeof *stats);
    }
    return EPROTO;
}

/* Queries the kernel for class with identifier 'handle' and parent 'parent'
 * on 'netdev'. */
static int
tc_query_class(const struct netdev *netdev,
               unsigned int handle, unsigned int parent,
               struct ofpbuf **replyp)
{
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    int error;

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO,
                                         &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = handle;
    tcmsg->tcm_parent = parent;

    error = tc_transact(&request, replyp);
    if (error) {
        VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
                     netdev_get_name(netdev),
                     tc_get_major(handle), tc_get_minor(handle),
                     tc_get_major(parent), tc_get_minor(parent),
                     ovs_strerror(error));
    }
    return error;
}

/* Equivalent to "tc class del dev <name> handle <handle>". */
static int
tc_delete_class(const struct netdev *netdev, unsigned int handle)
{
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    int error;

    tcmsg = netdev_linux_tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = handle;
    tcmsg->tcm_parent = 0;

    error = tc_transact(&request, NULL);
    if (error) {
        VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
                     netdev_get_name(netdev),
                     tc_get_major(handle), tc_get_minor(handle),
                     ovs_strerror(error));
    }
    return error;
}

/* Equivalent to "tc qdisc del dev <name> root". */
static int
tc_del_qdisc(struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct ofpbuf request;
    struct tcmsg *tcmsg;
    int error;

    tcmsg = netdev_linux_tc_make_request(netdev_, RTM_DELQDISC, 0, &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(1, 0);
    tcmsg->tcm_parent = TC_H_ROOT;

    error = tc_transact(&request, NULL);
    if (error == EINVAL) {
        /* EINVAL probably means that the default qdisc was in use, in which
         * case we've accomplished our purpose. */
        error = 0;
    }
    if (!error && netdev->tc) {
        if (netdev->tc->ops->tc_destroy) {
            netdev->tc->ops->tc_destroy(netdev->tc);
        }
        netdev->tc = NULL;
    }
    return error;
}

static bool
getqdisc_is_safe(void)
{
    static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
    static bool safe = false;

    if (ovsthread_once_start(&once)) {
        struct utsname utsname;
        int major, minor;

        if (uname(&utsname) == -1) {
            VLOG_WARN("uname failed (%s)", ovs_strerror(errno));
        } else if (!ovs_scan(utsname.release, "%d.%d", &major, &minor)) {
            VLOG_WARN("uname reported bad OS release (%s)", utsname.release);
        } else if (major < 2 || (major == 2 && minor < 35)) {
            VLOG_INFO("disabling unsafe RTM_GETQDISC in Linux kernel %s",
                      utsname.release);
        } else {
            safe = true;
        }
        ovsthread_once_done(&once);
    }
    return safe;
}

/* If 'netdev''s qdisc type and parameters are not yet known, queries the
 * kernel to determine what they are.  Returns 0 if successful, otherwise a
 * positive errno value. */
static int
tc_query_qdisc(const struct netdev *netdev_)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    struct ofpbuf request, *qdisc;
    const struct tc_ops *ops;
    struct tcmsg *tcmsg;
    int load_error;
    int error;

    if (netdev->tc) {
        return 0;
    }

    /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
     * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
     * 2.6.35 without that fix backported to it.
     *
     * To avoid the OOPS, we must not make a request that would attempt to dump
     * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
     * few others.  There are a few ways that I can see to do this, but most of
     * them seem to be racy (and if you lose the race the kernel OOPSes).  The
     * technique chosen here is to assume that any non-default qdisc that we
     * create will have a class with handle 1:0.  The built-in qdiscs only have
     * a class with handle 0:0.
     *
     * On Linux 2.6.35+ we use the straightforward method because it allows us
     * to handle non-builtin qdiscs without handle 1:0 (e.g. codel).  However,
     * in such a case we get no response at all from the kernel (!) if a
     * builtin qdisc is in use (which is later caught by "!error &&
     * !qdisc->size"). */
    tcmsg = netdev_linux_tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO,
                                         &request);
    if (!tcmsg) {
        return ENODEV;
    }
    tcmsg->tcm_handle = tc_make_handle(getqdisc_is_safe() ? 0 : 1, 0);
    tcmsg->tcm_parent = getqdisc_is_safe() ? TC_H_ROOT : 0;

    /* Figure out what tc class to instantiate. */
    error = tc_transact(&request, &qdisc);
    if (!error && qdisc->size) {
        const char *kind;

        error = tc_parse_qdisc(qdisc, &kind, NULL);
        if (error) {
            ops = &tc_ops_other;
        } else {
            ops = tc_lookup_linux_name(kind);
            if (!ops) {
                static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
                VLOG_DBG_RL(&rl2, "unknown qdisc \"%s\"", kind);

                ops = &tc_ops_other;
            }
        }
    } else if ((!error && !qdisc->size) || error == ENOENT) {
        /* Either it's a built-in qdisc, or (on Linux pre-2.6.35) it's a qdisc
         * set up by some other entity that doesn't have a handle 1:0.  We will
         * assume that it's the system default qdisc. */
        ops = &tc_ops_default;
        error = 0;
    } else {
        /* Who knows?  Maybe the device got deleted. */
        VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
                     netdev_get_name(netdev_), ovs_strerror(error));
        ops = &tc_ops_other;
    }

    /* Instantiate it. */
    load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc);
    ovs_assert((load_error == 0) == (netdev->tc != NULL));
    ofpbuf_delete(qdisc);

    return error ? error : load_error;
}

/* Linux traffic control uses tables with 256 entries ("rtab" tables) to
   approximate the time to transmit packets of various lengths.  For an MTU of
   256 or less, each entry is exact; for an MTU of 257 through 512, each entry
   represents two possible packet lengths; for a MTU of 513 through 1024, four
   possible lengths; and so on.

   Returns, for the specified 'mtu', the number of bits that packet lengths
   need to be shifted right to fit within such a 256-entry table. */
static int
tc_calc_cell_log(unsigned int mtu)
{
    int cell_log;

    if (!mtu) {
        mtu = ETH_PAYLOAD_MAX;
    }
    mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;

    for (cell_log = 0; mtu >= 256; cell_log++) {
        mtu >>= 1;
    }

    return cell_log;
}

/* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
 * of 'mtu'. */
static void
tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
{
    memset(rate, 0, sizeof *rate);
    rate->cell_log = tc_calc_cell_log(mtu);
    /* rate->overhead = 0; */           /* New in 2.6.24, not yet in some */
    /* rate->cell_align = 0; */         /* distro headers. */
    rate->mpu = ETH_TOTAL_MIN;
    rate->rate = Bps;
}

/* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
 * attribute of the specified "type".
 *
 * See tc_calc_cell_log() above for a description of "rtab"s. */
void
tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
{
    uint32_t *rtab;
    unsigned int i;

    rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
    for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
        unsigned packet_size = (i + 1) << rate->cell_log;
        if (packet_size < rate->mpu) {
            packet_size = rate->mpu;
        }
        rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
    }
}

/* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
 * burst size of 'burst_bytes'.  (If no value was requested, a 'burst_bytes' of
 * 0 is fine.) */
static int
tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
{
    unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
    return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
}
\f
/* Linux-only functions declared in netdev-linux.h  */

/* Modifies the 'flag' bit in ethtool's flags field for 'netdev'.  If
 * 'enable' is true, the bit is set.  Otherwise, it is cleared. */
int
netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
                              const char *flag_name, bool enable)
{
    const char *netdev_name = netdev_get_name(netdev);
    struct ethtool_value evalue;
    uint32_t new_flags;
    int error;

    COVERAGE_INC(netdev_get_ethtool);
    memset(&evalue, 0, sizeof evalue);
    error = netdev_linux_do_ethtool(netdev_name,
                                    (struct ethtool_cmd *)&evalue,
                                    ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
    if (error) {
        return error;
    }

    COVERAGE_INC(netdev_set_ethtool);
    new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
    if (new_flags == evalue.data) {
        return 0;
    }
    evalue.data = new_flags;
    error = netdev_linux_do_ethtool(netdev_name,
                                    (struct ethtool_cmd *)&evalue,
                                    ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
    if (error) {
        return error;
    }

    COVERAGE_INC(netdev_get_ethtool);
    memset(&evalue, 0, sizeof evalue);
    error = netdev_linux_do_ethtool(netdev_name,
                                    (struct ethtool_cmd *)&evalue,
                                    ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
    if (error) {
        return error;
    }

    if (new_flags != evalue.data) {
        VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
                     "device %s failed", enable ? "enable" : "disable",
                     flag_name, netdev_name);
        return EOPNOTSUPP;
    }

    return 0;
}
\f
/* Utility functions. */

/* Copies 'src' into 'dst', performing format conversion in the process. */
static void
netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
                                  const struct rtnl_link_stats *src)
{
    dst->rx_packets = src->rx_packets;
    dst->tx_packets = src->tx_packets;
    dst->rx_bytes = src->rx_bytes;
    dst->tx_bytes = src->tx_bytes;
    dst->rx_errors = src->rx_errors;
    dst->tx_errors = src->tx_errors;
    dst->rx_dropped = src->rx_dropped;
    dst->tx_dropped = src->tx_dropped;
    dst->multicast = src->multicast;
    dst->collisions = src->collisions;
    dst->rx_length_errors = src->rx_length_errors;
    dst->rx_over_errors = src->rx_over_errors;
    dst->rx_crc_errors = src->rx_crc_errors;
    dst->rx_frame_errors = src->rx_frame_errors;
    dst->rx_fifo_errors = src->rx_fifo_errors;
    dst->rx_missed_errors = src->rx_missed_errors;
    dst->tx_aborted_errors = src->tx_aborted_errors;
    dst->tx_carrier_errors = src->tx_carrier_errors;
    dst->tx_fifo_errors = src->tx_fifo_errors;
    dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
    dst->tx_window_errors = src->tx_window_errors;
}

/* Copies 'src' into 'dst', performing format conversion in the process. */
static void
netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst,
                                    const struct rtnl_link_stats64 *src)
{
    dst->rx_packets = src->rx_packets;
    dst->tx_packets = src->tx_packets;
    dst->rx_bytes = src->rx_bytes;
    dst->tx_bytes = src->tx_bytes;
    dst->rx_errors = src->rx_errors;
    dst->tx_errors = src->tx_errors;
    dst->rx_dropped = src->rx_dropped;
    dst->tx_dropped = src->tx_dropped;
    dst->multicast = src->multicast;
    dst->collisions = src->collisions;
    dst->rx_length_errors = src->rx_length_errors;
    dst->rx_over_errors = src->rx_over_errors;
    dst->rx_crc_errors = src->rx_crc_errors;
    dst->rx_frame_errors = src->rx_frame_errors;
    dst->rx_fifo_errors = src->rx_fifo_errors;
    dst->rx_missed_errors = src->rx_missed_errors;
    dst->tx_aborted_errors = src->tx_aborted_errors;
    dst->tx_carrier_errors = src->tx_carrier_errors;
    dst->tx_fifo_errors = src->tx_fifo_errors;
    dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
    dst->tx_window_errors = src->tx_window_errors;
}

int
get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats)
{
    struct ofpbuf request;
    struct ofpbuf *reply;
    int error;

    /* Filtering all counters by default */
    memset(stats, 0xFF, sizeof(struct netdev_stats));

    ofpbuf_init(&request, 0);
    nl_msg_put_nlmsghdr(&request,
                        sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ),
                        RTM_GETLINK, NLM_F_REQUEST);
    ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
    nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_));
    error = nl_transact(NETLINK_ROUTE, &request, &reply);
    ofpbuf_uninit(&request);
    if (error) {
        return error;
    }

    if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) {
        const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS64);
        if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats64)) {
            netdev_stats_from_rtnl_link_stats64(stats, nl_attr_get(a));
            error = 0;
        } else {
            a = nl_attr_find(reply, 0, IFLA_STATS);
            if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) {
                netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a));
                error = 0;
            } else {
                VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
                error = EPROTO;
            }
        }
    } else {
        VLOG_WARN_RL(&rl, "short RTM_GETLINK reply");
        error = EPROTO;
    }


    ofpbuf_delete(reply);
    return error;
}

static int
get_flags(const struct netdev *dev, unsigned int *flags)
{
    struct ifreq ifr;
    int error;

    *flags = 0;
    error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS");
    if (!error) {
        *flags = ifr.ifr_flags;
    }
    return error;
}

static int
set_flags(const char *name, unsigned int flags)
{
    struct ifreq ifr;

    ifr.ifr_flags = flags;
    return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS");
}

int
linux_get_ifindex(const char *netdev_name)
{
    struct ifreq ifr;
    int error;

    ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
    COVERAGE_INC(netdev_get_ifindex);

    error = af_inet_ioctl(SIOCGIFINDEX, &ifr);
    if (error) {
        /* ENODEV probably means that a vif disappeared asynchronously and
         * hasn't been removed from the database yet, so reduce the log level
         * to INFO for that case. */
        VLOG_RL(&rl, error == ENODEV ? VLL_INFO : VLL_ERR,
                "ioctl(SIOCGIFINDEX) on %s device failed: %s",
                netdev_name, ovs_strerror(error));
        return -error;
    }
    return ifr.ifr_ifindex;
}

static int
get_ifindex(const struct netdev *netdev_, int *ifindexp)
{
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);

    if (!(netdev->cache_valid & VALID_IFINDEX)) {
        netdev_linux_update_via_netlink(netdev);
    }

    if (!(netdev->cache_valid & VALID_IFINDEX)) {
        /* Fall back to ioctl if netlink fails */
        int ifindex = linux_get_ifindex(netdev_get_name(netdev_));

        if (ifindex < 0) {
            netdev->get_ifindex_error = -ifindex;
            netdev->ifindex = 0;
        } else {
            netdev->get_ifindex_error = 0;
            netdev->ifindex = ifindex;
        }
        netdev->cache_valid |= VALID_IFINDEX;
    }

    *ifindexp = netdev->ifindex;
    return netdev->get_ifindex_error;
}

static int
netdev_linux_update_via_netlink(struct netdev_linux *netdev)
{
    struct ofpbuf request;
    struct ofpbuf *reply;
    struct rtnetlink_change chg;
    struct rtnetlink_change *change = &chg;
    int error;

    ofpbuf_init(&request, 0);
    nl_msg_put_nlmsghdr(&request,
                        sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ) +
                        NL_A_U32_SIZE, RTM_GETLINK, NLM_F_REQUEST);
    ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));

    /* The correct identifiers for a Linux device are netnsid and ifindex,
     * but ifindex changes as the port is moved to another network namespace
     * and the interface name statically stored in ovsdb. */
    nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(&netdev->up));
    if (netdev_linux_netnsid_is_remote(netdev)) {
        nl_msg_put_u32(&request, IFLA_IF_NETNSID, netdev->netnsid);
    }
    error = nl_transact(NETLINK_ROUTE, &request, &reply);
    ofpbuf_uninit(&request);
    if (error) {
        ofpbuf_delete(reply);
        return error;
    }

    if (rtnetlink_parse(reply, change)
        && change->nlmsg_type == RTM_NEWLINK) {
        bool changed = false;
        error = 0;

        /* Update netdev from rtnl msg and increment its seq if needed. */
        if ((change->ifi_flags ^ netdev->ifi_flags) & IFF_RUNNING) {
            netdev->carrier_resets++;
            changed = true;
        }
        if (change->ifi_flags != netdev->ifi_flags) {
            netdev->ifi_flags = change->ifi_flags;
            changed = true;
        }
        if (change->mtu && change->mtu != netdev->mtu) {
            netdev->mtu = change->mtu;
            netdev->cache_valid |= VALID_MTU;
            netdev->netdev_mtu_error = 0;
            changed = true;
        }
        if (!eth_addr_is_zero(change->mac)
            && !eth_addr_equals(change->mac, netdev->etheraddr)) {
            netdev->etheraddr = change->mac;
            netdev->cache_valid |= VALID_ETHERADDR;
            netdev->ether_addr_error = 0;
            changed = true;
        }
        if (change->if_index != netdev->ifindex) {
            netdev->ifindex = change->if_index;
            netdev->cache_valid |= VALID_IFINDEX;
            netdev->get_ifindex_error = 0;
            changed = true;
        }
        if (change->primary && netdev_linux_kind_is_lag(change->primary)) {
            netdev->is_lag_master = true;
        }
        if (changed) {
            netdev_change_seq_changed(&netdev->up);
        }
    } else {
        error = EINVAL;
    }

    ofpbuf_delete(reply);
    return error;
}

static int
get_etheraddr(const char *netdev_name, struct eth_addr *ea)
{
    struct ifreq ifr;
    int hwaddr_family;
    int error;

    memset(&ifr, 0, sizeof ifr);
    ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
    COVERAGE_INC(netdev_get_hwaddr);
    error = af_inet_ioctl(SIOCGIFHWADDR, &ifr);
    if (error) {
        /* ENODEV probably means that a vif disappeared asynchronously and
         * hasn't been removed from the database yet, so reduce the log level
         * to INFO for that case. */
        VLOG(error == ENODEV ? VLL_INFO : VLL_ERR,
             "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
             netdev_name, ovs_strerror(error));
        return error;
    }
    hwaddr_family = ifr.ifr_hwaddr.sa_family;
    if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER &&
        hwaddr_family != ARPHRD_NONE) {
        VLOG_INFO("%s device has unknown hardware address family %d",
                  netdev_name, hwaddr_family);
        return EINVAL;
    }
    memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
    return 0;
}

static int
set_etheraddr(const char *netdev_name, const struct eth_addr mac)
{
    struct ifreq ifr;
    int error;

    memset(&ifr, 0, sizeof ifr);
    ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
    ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
    memcpy(ifr.ifr_hwaddr.sa_data, &mac, ETH_ADDR_LEN);
    COVERAGE_INC(netdev_set_hwaddr);
    error = af_inet_ioctl(SIOCSIFHWADDR, &ifr);
    if (error) {
        VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
                 netdev_name, ovs_strerror(error));
    }
    return error;
}

static int
netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
                        int cmd, const char *cmd_name)
{
    struct ifreq ifr;
    int error;

    memset(&ifr, 0, sizeof ifr);
    ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
    ifr.ifr_data = (caddr_t) ecmd;

    ecmd->cmd = cmd;
    error = af_inet_ioctl(SIOCETHTOOL, &ifr);
    if (error) {
        if (error != EOPNOTSUPP) {
            VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
                         "failed: %s", cmd_name, name, ovs_strerror(error));
        } else {
            /* The device doesn't support this operation.  That's pretty
             * common, so there's no point in logging anything. */
        }
    }
    return error;
}

/* Returns an AF_PACKET raw socket or a negative errno value. */
static int
af_packet_sock(void)
{
    static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
    static int sock;

    if (ovsthread_once_start(&once)) {
        sock = socket(AF_PACKET, SOCK_RAW, 0);
        if (sock >= 0) {
            int error = set_nonblocking(sock);
            if (error) {
                close(sock);
                sock = -error;
            } else if (userspace_tso_enabled()) {
                int val = 1;
                error = setsockopt(sock, SOL_PACKET, PACKET_VNET_HDR, &val,
                                   sizeof val);
                if (error) {
                    error = errno;
                    VLOG_ERR("failed to enable vnet hdr in raw socket: %s",
                             ovs_strerror(errno));
                    close(sock);
                    sock = -error;
                }
            }
        } else {
            sock = -errno;
            VLOG_ERR("failed to create packet socket: %s",
                     ovs_strerror(errno));
        }
        ovsthread_once_done(&once);
    }

    return sock;
}

static int
netdev_linux_parse_l2(struct dp_packet *b, uint16_t *l4proto)
{
    struct eth_header *eth_hdr;
    ovs_be16 eth_type;
    int l2_len;

    eth_hdr = dp_packet_at(b, 0, ETH_HEADER_LEN);
    if (!eth_hdr) {
        return -EINVAL;
    }

    l2_len = ETH_HEADER_LEN;
    eth_type = eth_hdr->eth_type;
    if (eth_type_vlan(eth_type)) {
        struct vlan_header *vlan = dp_packet_at(b, l2_len, VLAN_HEADER_LEN);

        if (!vlan) {
            return -EINVAL;
        }

        eth_type = vlan->vlan_next_type;
        l2_len += VLAN_HEADER_LEN;
    }

    if (eth_type == htons(ETH_TYPE_IP)) {
        struct ip_header *ip_hdr = dp_packet_at(b, l2_len, IP_HEADER_LEN);

        if (!ip_hdr) {
            return -EINVAL;
        }

        *l4proto = ip_hdr->ip_proto;
        dp_packet_hwol_set_tx_ipv4(b);
    } else if (eth_type == htons(ETH_TYPE_IPV6)) {
        struct ovs_16aligned_ip6_hdr *nh6;

        nh6 = dp_packet_at(b, l2_len, IPV6_HEADER_LEN);
        if (!nh6) {
            return -EINVAL;
        }

        *l4proto = nh6->ip6_ctlun.ip6_un1.ip6_un1_nxt;
        dp_packet_hwol_set_tx_ipv6(b);
    }

    return 0;
}

static int
netdev_linux_parse_vnet_hdr(struct dp_packet *b)
{
    struct virtio_net_hdr *vnet = dp_packet_pull(b, sizeof *vnet);
    uint16_t l4proto = 0;

    if (OVS_UNLIKELY(!vnet)) {
        return -EINVAL;
    }

    if (vnet->flags == 0 && vnet->gso_type == VIRTIO_NET_HDR_GSO_NONE) {
        return 0;
    }

    if (netdev_linux_parse_l2(b, &l4proto)) {
        return -EINVAL;
    }

    if (vnet->flags == VIRTIO_NET_HDR_F_NEEDS_CSUM) {
        if (l4proto == IPPROTO_TCP) {
            dp_packet_hwol_set_csum_tcp(b);
        } else if (l4proto == IPPROTO_UDP) {
            dp_packet_hwol_set_csum_udp(b);
        } else if (l4proto == IPPROTO_SCTP) {
            dp_packet_hwol_set_csum_sctp(b);
        }
    }

    if (l4proto && vnet->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
        uint8_t allowed_mask = VIRTIO_NET_HDR_GSO_TCPV4
                                | VIRTIO_NET_HDR_GSO_TCPV6
                                | VIRTIO_NET_HDR_GSO_UDP;
        uint8_t type = vnet->gso_type & allowed_mask;

        if (type == VIRTIO_NET_HDR_GSO_TCPV4
            || type == VIRTIO_NET_HDR_GSO_TCPV6) {
            dp_packet_hwol_set_tcp_seg(b);
        }
    }

    return 0;
}

static void
netdev_linux_prepend_vnet_hdr(struct dp_packet *b, int mtu)
{
    struct virtio_net_hdr *vnet = dp_packet_push_zeros(b, sizeof *vnet);

    if (dp_packet_hwol_is_tso(b)) {
        uint16_t hdr_len = ((char *)dp_packet_l4(b) - (char *)dp_packet_eth(b))
                            + TCP_HEADER_LEN;

        vnet->hdr_len = (OVS_FORCE __virtio16)hdr_len;
        vnet->gso_size = (OVS_FORCE __virtio16)(mtu - hdr_len);
        if (dp_packet_hwol_is_ipv4(b)) {
            vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
        } else {
            vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
        }

    } else {
        vnet->flags = VIRTIO_NET_HDR_GSO_NONE;
    }

    if (dp_packet_hwol_l4_mask(b)) {
        vnet->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
        vnet->csum_start = (OVS_FORCE __virtio16)((char *)dp_packet_l4(b)
                                                  - (char *)dp_packet_eth(b));

        if (dp_packet_hwol_l4_is_tcp(b)) {
            vnet->csum_offset = (OVS_FORCE __virtio16) __builtin_offsetof(
                                    struct tcp_header, tcp_csum);
        } else if (dp_packet_hwol_l4_is_udp(b)) {
            vnet->csum_offset = (OVS_FORCE __virtio16) __builtin_offsetof(
                                    struct udp_header, udp_csum);
        } else if (dp_packet_hwol_l4_is_sctp(b)) {
            vnet->csum_offset = (OVS_FORCE __virtio16) __builtin_offsetof(
                                    struct sctp_header, sctp_csum);
        } else {
            VLOG_WARN_RL(&rl, "Unsupported L4 protocol");
        }
    }
}