Openvswitch原理与代码分析(2): ovs-vswitchd的启动

ovs-vswitchd.c的main函数最终会进入一个while循环，在这个无限循环中，里面最重要的两个函数是bridge_run()和netdev_run()。

 

 

Openvswitch主要管理两种类型的设备，一个是创建的虚拟网桥，一个是连接到虚拟网桥上的设备。

 

其中bridge_run就是初始化数据库中已经创建的虚拟网桥。

 

一、虚拟网桥的初始化bridge_run

 

bridge_run会调用bridge_run__，bridge_run__中最重要的是对于所有的网桥，都调用ofproto_run

 

1. static void
2. bridge_run__(void)
3. {
4. ……
5.     /* Let each bridge do the work that it needs to do. */
6.     HMAP_FOR_EACH (br, node, &all_bridges) {
7.         ofproto_run(br->ofproto);
8.     }
9. }

 

Int ofproto_run(struct ofproto *p)会调用error = p->ofproto_class->run(p);

 

ofproto_class的定义在ofproto-provider.h中，它的实现定义在ofproto-dpif.c中，这里面的所有的函数，在这个文件中都有定义。

 

1. const struct ofproto_class ofproto_dpif_class = {
2.     init,
3.     enumerate_types,
4.     enumerate_names,
5.     del,
6.     port_open_type,
7.     type_run,
8.     type_wait,
9.     alloc,
10.     construct,
11.     destruct,
12.     dealloc,
13.     run,
14.     wait,
15.     NULL, /* get_memory_usage. */
16.     type_get_memory_usage,
17.     flush,
18.     query_tables,
19.     set_tables_version,
20.     port_alloc,
21.     port_construct,
22.     port_destruct,
23.     port_dealloc,
24.     port_modified,
25.     port_reconfigured,
26.     port_query_by_name,
27.     port_add,
28.     port_del,
29.     port_get_stats,
30.     port_dump_start,
31.     port_dump_next,
32.     port_dump_done,
33.     port_poll,
34.     port_poll_wait,
35.     port_is_lacp_current,
36.     port_get_lacp_stats,
37.     NULL, /* rule_choose_table */
38.     rule_alloc,
39.     rule_construct,
40.     rule_insert,
41.     rule_delete,
42.     rule_destruct,
43.     rule_dealloc,
44.     rule_get_stats,
45.     rule_execute,
46.     set_frag_handling,
47.     packet_out,
48.     set_netflow,
49.     get_netflow_ids,
50.     set_sflow,
51.     set_ipfix,
52.     set_cfm,
53.     cfm_status_changed,
54.     get_cfm_status,
55.     set_lldp,
56.     get_lldp_status,
57.     set_aa,
58.     aa_mapping_set,
59.     aa_mapping_unset,
60.     aa_vlan_get_queued,
61.     aa_vlan_get_queue_size,
62.     set_bfd,
63.     bfd_status_changed,
64.     get_bfd_status,
65.     set_stp,
66.     get_stp_status,
67.     set_stp_port,
68.     get_stp_port_status,
69.     get_stp_port_stats,
70.     set_rstp,
71.     get_rstp_status,
72.     set_rstp_port,
73.     get_rstp_port_status,
74.     set_queues,
75.     bundle_set,
76.     bundle_remove,
77.     mirror_set__,
78.     mirror_get_stats__,
79.     set_flood_vlans,
80.     is_mirror_output_bundle,
81.     forward_bpdu_changed,
82.     set_mac_table_config,
83.     set_mcast_snooping,
84.     set_mcast_snooping_port,
85.     set_realdev,
86.     NULL, /* meter_get_features */
87.     NULL, /* meter_set */
88.     NULL, /* meter_get */
89.     NULL, /* meter_del */
90.     group_alloc, /* group_alloc */
91.     group_construct, /* group_construct */
92.     group_destruct, /* group_destruct */
93.     group_dealloc, /* group_dealloc */
94.     group_modify, /* group_modify */
95.     group_get_stats, /* group_get_stats */
96.     get_datapath_version, /* get_datapath_version */
97. };

 

在ofproto-provider.h中注释里是这样说的。

这里定义了四类数据结构

Struct ofproto表示一个交换机

Struct ofport表示交换机上的一个端口

Struct rule表示交换机上的一条flow规则

Struct ofgroup表示一个flow规则组

 

上面说到启动的过程中，会调用ofproto_class->run，也即会调用ofproto-dpif.c中的static int run(struct ofproto *ofproto_)函数。

 

在这个函数中，会初始化netflow, sflow, ipfix，stp, rstp, mac address learning等一系列操作。

 

bridge_run还会调用static void bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)，其中ovs_cfg是从ovsdb-server里面读取出来的配置。

 

在这个函数里面，对于每一个网桥，将网卡添加进去

1. HMAP_FOR_EACH (br, node, &all_bridges) {
2.     bridge_add_ports(br, &br->wanted_ports);
3.     shash_destroy(&br->wanted_ports);
4. }

 

1. static void
2. bridge_add_ports(struct bridge *br, const struct shash *wanted_ports)
3. {
4.     /* First add interfaces that request a particular port number. */
5.     bridge_add_ports__(br, wanted_ports, true);
6.  
7.     /* Then add interfaces that want automatic port number assignment.
8.      * We add these afterward to avoid accidentally taking a specifically
9.      * requested port number. */
10.     bridge_add_ports__(br, wanted_ports, false);
11. }

 

static void bridge_add_ports__(struct bridge *br, const struct shash *wanted_ports, bool with_requested_port)会调用

static bool iface_create(struct bridge *br, const struct ovsrec_interface *iface_cfg, const struct ovsrec_port *port_cfg)会调用

static int iface_do_create(const struct bridge *br, const struct ovsrec_interface *iface_cfg, const struct ovsrec_port *port_cfg, ofp_port_t *ofp_portp, struct netdev **netdevp, char **errp)会调用

int ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev, ofp_port_t *ofp_portp)会调用

 

1. error = ofproto->ofproto_class->port_add(ofproto, netdev);

 

会调用ofproto-dpif.c中的ofproto_dpif_class的static int port_add(struct ofproto *ofproto_, struct netdev *netdev)函数。

 

会调用int dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop)会调用

 

1. error = dpif->dpif_class->port_add(dpif, netdev, &port_no);

 

会调用dpif_netlink_class的port_add函数，也即dpif_netlink_port_add，也即

static int dpif_netlink_port_add(struct dpif *dpif_, struct netdev *netdev,odp_port_t *port_nop)会调用

static int dpif_netlink_port_add__(struct dpif_netlink *dpif, struct netdev *netdev, odp_port_t *port_nop)

 

在这个函数里面，会调用netlink的API，命令为OVS_VPORT_CMD_NEW

 

1. const char *name = netdev_vport_get_dpif_port(netdev,
2.                                                   namebuf, sizeof namebuf);
3. struct dpif_netlink_vport request, reply;
4. struct nl_sock **socksp = NULL;
5.  
6. if (dpif->handlers) {
7.     socksp = vport_create_socksp(dpif, &error);
8.     if (!socksp) {
9.         return error;
10.     }
11. }
12.  
13. dpif_netlink_vport_init(&request);
14. request.cmd = OVS_VPORT_CMD_NEW;
15. request.dp_ifindex = dpif->dp_ifindex;
16. request.type = netdev_to_ovs_vport_type(netdev);
17.  
18. request.name = name;
19.  
20. upcall_pids = vport_socksp_to_pids(socksp, dpif->n_handlers);
21. request.n_upcall_pids = socksp ? dpif->n_handlers : 1;
22. request.upcall_pids = upcall_pids;
23. error = dpif_netlink_vport_transact(&request, &reply, &buf);

 

这里会调用内核模块openvswitch.ko，在内核中添加虚拟网卡。这部分详细的过程将在下一节分析。

二、虚拟网卡的初始化netdev_run()

 

1. void
2. netdev_run(void)
3.     OVS_EXCLUDED(netdev_class_mutex, netdev_mutex)
4. {
5.     struct netdev_registered_class *rc;
6.  
7.     netdev_initialize();
8.     ovs_mutex_lock(&netdev_class_mutex);
9.     HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
10.         if (rc->class->run) {
11.             rc->class->run();
12.         }
13.     }
14.     ovs_mutex_unlock(&netdev_class_mutex);
15. }

 

依次循环调用netdev_classes中的每一个run。

 

对于不同类型的虚拟网卡，都有对应的netdev_class。

 

例如对于dpdk的网卡有

 

1. static const struct netdev_class dpdk_class =
2.     NETDEV_DPDK_CLASS(
3.         "dpdk",
4.         NULL,
5.         netdev_dpdk_construct,
6.         netdev_dpdk_destruct,
7.         netdev_dpdk_set_multiq,
8.         netdev_dpdk_eth_send,
9.         netdev_dpdk_get_carrier,
10.         netdev_dpdk_get_stats,
11.         netdev_dpdk_get_features,
12.         netdev_dpdk_get_status,
13.         netdev_dpdk_rxq_recv);

 

对于物理网卡，也需要有相应的netdev_class

 

1. const struct netdev_class netdev_linux_class =
2.     NETDEV_LINUX_CLASS(
3.         "system",
4.         netdev_linux_construct,
5.         netdev_linux_get_stats,
6.         netdev_linux_get_features,
7.         netdev_linux_get_status);

 

对于连接到KVM的tap网卡

1. const struct netdev_class netdev_tap_class =
2.     NETDEV_LINUX_CLASS(
3.         "tap",
4.         netdev_linux_construct_tap,
5.         netdev_tap_get_stats,
6.         netdev_linux_get_features,
7.         netdev_linux_get_status);

 

对于虚拟的软网卡，比如veth pair

1. const struct netdev_class netdev_internal_class =
2.     NETDEV_LINUX_CLASS(
3.         "internal",
4.         netdev_linux_construct,
5.         netdev_internal_get_stats,
6.         NULL, /* get_features */
7.         netdev_internal_get_status);

 

其中NETDEV_LINUX_CLASS是一个宏，不是所有的参数都需要全部填写。

1. #define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS,
2.                            GET_FEATURES, GET_STATUS)
3. {
4.     NAME,
5.                                                                 
6.     NULL,
7.     netdev_linux_run,
8.     netdev_linux_wait,
9.                                                                 
10.     netdev_linux_alloc,
11.     CONSTRUCT,
12.     netdev_linux_destruct,
13.     netdev_linux_dealloc,
14.     NULL, /* get_config */
15.     NULL, /* set_config */
16.     NULL, /* get_tunnel_config */
17.     NULL, /* build header */
18.     NULL, /* push header */
19.     NULL, /* pop header */
20.     NULL, /* get_numa_id */
21.     NULL, /* set_multiq */
22.                                                                 
23.     netdev_linux_send,
24.     netdev_linux_send_wait,
25.                                                                 
26.     netdev_linux_set_etheraddr,
27.     netdev_linux_get_etheraddr,
28.     netdev_linux_get_mtu,
29.     netdev_linux_set_mtu,
30.     netdev_linux_get_ifindex,
31.     netdev_linux_get_carrier,
32.     netdev_linux_get_carrier_resets,
33.     netdev_linux_set_miimon_interval,
34.     GET_STATS,
35.                                                                 
36.     GET_FEATURES,
37.     netdev_linux_set_advertisements,
38.                                                                 
39.     netdev_linux_set_policing,
40.     netdev_linux_get_qos_types,
41.     netdev_linux_get_qos_capabilities,
42.     netdev_linux_get_qos,
43.     netdev_linux_set_qos,
44.     netdev_linux_get_queue,
45.     netdev_linux_set_queue,
46.     netdev_linux_delete_queue,
47.     netdev_linux_get_queue_stats,
48.     netdev_linux_queue_dump_start,
49.     netdev_linux_queue_dump_next,
50.     netdev_linux_queue_dump_done,
51.     netdev_linux_dump_queue_stats,
52.                                                                 
53.     netdev_linux_get_in4,
54.     netdev_linux_set_in4,
55.     netdev_linux_get_in6,
56.     netdev_linux_add_router,
57.     netdev_linux_get_next_hop,
58.     GET_STATUS,
59.     netdev_linux_arp_lookup,
60.                                                                 
61.     netdev_linux_update_flags,
62.                                                                 
63.     netdev_linux_rxq_alloc,
64.     netdev_linux_rxq_construct,
65.     netdev_linux_rxq_destruct,
66.     netdev_linux_rxq_dealloc,
67.     netdev_linux_rxq_recv,
68.     netdev_linux_rxq_wait,
69.     netdev_linux_rxq_drain,
70. }

 

rc->class->run()调用的是netdev-linux.c下的netdev_linux_run

 

netdev_linux_run会调用netlink的sock得到虚拟网卡的状态，并且更新状态。

 

1. error = nl_sock_recv(sock, &buf, false);
2. if (!error) {
3.     struct rtnetlink_change change;
4.     if (rtnetlink_parse(&buf, &change)) {
5.         struct netdev *netdev_ = netdev_from_name(change.ifname);
6.         if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) {
7.            struct netdev_linux *netdev = netdev_linux_cast(netdev_);
8.            ovs_mutex_lock(&netdev->mutex);
9.            netdev_linux_update(netdev, &change);
10.            ovs_mutex_unlock(&netdev->mutex);
11.         }
12.         netdev_close(netdev_);
13.      }
14. }