/*

 * 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 <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 "dpif-offload.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);

COVERAGE_DEFINE(netdev_linux_invalid_l4_csum);

COVERAGE_DEFINE(netdev_linux_unknown_l4_csum);

#ifndef HAVE_IFLA_IF_NETNSID

#define IFLA_IF_NETNSID 46

#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 3.2 introduced "unknown" speed and duplex. */

#ifndef SPEED_UNKNOWN

#define SPEED_UNKNOWN -1

#endif

#ifndef DUPLEX_UNKNOWN

#define DUPLEX_UNKNOWN 0xff

#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 3.19 introduced speed for 40G. */

#ifndef SPEED_40000

#define SPEED_40000 40000

#endif

/* Linux 4.2 introduced speed for 100G. */

#ifndef SPEED_100000

#define SPEED_100000 100000

#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,

};

/* Linux 4.4 introduced the ability to skip the internal stats gathering

 * that netlink does via an external filter mask that can be passed into

 * a netlink request.

 */

#ifndef RTEXT_FILTER_SKIP_STATS

#define RTEXT_FILTER_SKIP_STATS (1 << 3)

#endif

/* 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,

};

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(uint64_t rate, unsigned int ticks);

static unsigned int tc_bytes_to_ticks(uint64_t rate, unsigned int size);

static unsigned int tc_buffer_per_jiffy(uint64_t 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 *, uint64_t kbits_rate,

                          uint32_t kbits_burst, uint32_t kpkts_rate,

                          uint32_t kpkts_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);

static void tc_policer_init(struct tc_police *tc_police, uint64_t kbits_rate,

                            uint64_t kbits_burst);

void

tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate,

            uint64_t rate64);

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(uint64_t Bps, int mtu, uint64_t burst_bytes);

/* 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);

/* Very old kernels from the 2.6 era don't support vnet headers with the tun

 * device. We can detect this while constructing a netdev, but need this for

 * packet rx/tx. */

static bool tap_supports_vnet_hdr = true;

static int netdev_linux_parse_vnet_hdr(struct dp_packet *b);

static int 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 void netdev_linux_set_ol(struct netdev *netdev);

static bool

is_tap_netdev(const struct netdev *netdev)

{

    return netdev_get_class(netdev) == &netdev_tap_class;

}

/* Ensures that our own network namespace has a self-referential nsid mapping

 * and records its value through netnsid_set_self() so that netnsid_is_local()

 * treats it as the local namespace.

 *

 * NETLINK_LISTEN_ALL_NSID workaround: OVS enables this option on its RTNL

 * notification socket so that it can receive events from remote namespaces.

 * A side-effect of this option is that the kernel tags every broadcast

 * (including locally-originated RTM events) with the sender nsid as looked up

 * in the receiver nsid table.  Normally local events carry no nsid cmsg

 * (which OVS interprets as NETNSID_LOCAL), but if a self-referential nsid

 * mapping exists for the local namespace an older kernel may tag them with

 * that nsid instead, causing OVS to silently reject them.  Some container

 * runtimes create such a mapping as a side-effect of cross-namespace link

 * queries.

 *

 * To make this deterministic, we create the mapping ourselves (or read back

 * the existing one) at startup, before the notification socket exists and

 * record the resulting nsid as local.  A self-referential nsid mapping is

 * permanent once created, the kernel only removes nsid entries when the

 * peer namespace is destroyed, which can never happen for our own namespace.

 * Therefore, the value never changes and no runtime monitoring is required.

 *

 * The namespace is identified by our own pid (NETNSA_PID).

 *

 * This kernel behavior was fixed in commit 88b126b39f97 ("net: netlink: don't

 * set nsid on local notifications"), which stops tagging local events with the

 * self-referential nsid.  This workaround (and the self-nsid check in

 * netnsid_is_local()) is only needed for older kernels that lack that fix and

 * can be removed once such kernels are no longer supported. */

static void

netdev_linux_init_self_nsid(void)

{

#ifdef HAVE_LINUX_NET_NAMESPACE_H

    const int rta_offset = NLMSG_ALIGN(sizeof(struct rtgenmsg));

    struct ofpbuf request;

    struct ofpbuf *reply = NULL;

    uint32_t pid = getpid();

    int error;

    /* Create a self-referential nsid mapping for our own namespace.  Passing

     * NETNSA_NSID == -1 lets the kernel allocate a free id. If the mapping

     * already exists the request fails with EEXIST, which is harmless. */

    ofpbuf_init(&request, 0);

    nl_msg_put_nlmsghdr(&request,

                        rta_offset + 2 * NL_ATTR_SIZE(sizeof(uint32_t)),

                        RTM_NEWNSID, NLM_F_REQUEST | NLM_F_ACK);

    ofpbuf_put_zeros(&request, rta_offset);

    nl_msg_put_u32(&request, NETNSA_PID, pid);

    nl_msg_put_u32(&request, NETNSA_NSID, NETNSID_LOCAL);

    nl_transact(NETLINK_ROUTE, &request, NULL);

    ofpbuf_uninit(&request);

    /* Read back the nsid that the kernel assigned to our namespace. */

    ofpbuf_init(&request, 0);

    nl_msg_put_nlmsghdr(&request, rta_offset + NL_ATTR_SIZE(sizeof(uint32_t)),

                        RTM_GETNSID, NLM_F_REQUEST);

    ofpbuf_put_zeros(&request, rta_offset);

    nl_msg_put_u32(&request, NETNSA_PID, pid);

    error = nl_transact(NETLINK_ROUTE, &request, &reply);

    if (!error && reply) {

        const struct nlattr *a;

        a = nl_attr_find(reply, NLMSG_HDRLEN + rta_offset, NETNSA_NSID);

        if (a) {

            int nsid = nl_attr_get_u32(a);

            if (nsid >= 0) {

                netnsid_set_self(nsid);

                VLOG_DBG("local network namespace has nsid %d", nsid);

            }

        }

    } else {

        VLOG_WARN("could not query local network namespace nsid: %s",

                  ovs_strerror(error));

    }

    ofpbuf_uninit(&request);

    ofpbuf_delete(reply);

#endif

}

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)) {

        const char *dpif_type = netdev_get_dpif_type(&netdev->up);

        if (netdev_get_class(&netdev->up) == &netdev_tap_class

            || (dpif_type && strcmp(dpif_type, "system"))) {

            /* vport netlink lookup makes no sense for

             * non-system dpif types, set nsid to local. */

            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;

        /* Discover (creating it if necessary) the nsid that the kernel uses

         * for our own namespace, before the notification socket starts

         * receiving namespace-tagged events. */

        netdev_linux_init_self_nsid();

        error = nl_sock_create(NETLINK_ROUTE, &sock);

        if (!error) {

            size_t i;

            nl_sock_listen_all_nsid(sock, true);

            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;

                }

            }

        }

        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 *primary_netdev;

            char primary_name[IFNAMSIZ];

            uint32_t block_id;

            int error = 0;

            if (!if_indextoname(change->master_ifindex, primary_name)) {

                return;

            }

            primary_netdev = netdev_from_name(primary_name);

            if (!primary_netdev) {

                return;

            }

            /* If LAG primary member is not attached to ovs,

             * ingress block on LAG members should not be updated. */

            if (!primary_netdev->auto_classified &&

                is_netdev_linux_class(primary_netdev->netdev_class)) {

                block_id = netdev_get_block_id(primary_netdev);

                if (!block_id) {

                    netdev_close(primary_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 primary 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(primary_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) && !change.irrelevant) {

                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_primary = 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);