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()。
bridge_init(remote);
free(remote);
exiting = false;
while (!exiting) { //while循环,直到退出
memory_run();
if (memory_should_report()) {
struct simap usage;
simap_init(&usage);
bridge_get_memory_usage(&usage);
memory_report(&usage);
simap_destroy(&usage);
}
bridge_run(); //处理controller交互,ovs-ofctl命令
unixctl_server_run(unixctl); //处理ovs-vsctl命令
netdev_run();
memory_wait();
bridge_wait();
unixctl_server_wait(unixctl);
netdev_wait();
if (exiting) {
poll_immediate_wake();
}
poll_block(); //阻塞,当没有请求处理的时候,阻塞在此处
if (should_service_stop()) {
exiting = true;
}
}
bridge_exit();
unixctl_server_destroy(unixctl);
service_stop();
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。
在具体介绍代码调用之前,需要回顾一下OVS 的代码架构:
ovs-vswitchd是ovs中最核心的组件,openflow的相关逻辑都在vswitchd里实现,一般来说ovs分为datapath, vswitchd以及ovsdb三个部分,datapath一般是和具体是数据面平台相关的,比如白盒交换机,或者linux内核等。ovsdb用于存储vswitch本身的配置信息,比如端口,拓扑,规则等。vswitchd本身是分层的结构,最上层daemon主要用于和ovsdb通信,做配置的下发和更新等,中间层ofproto,用于和openflow控制器通信,以及通过ofproto_class暴露了ofproto provider接口(目前只支持一个ofproto provider,即ofproto-dpif),不同平台上openflow的具体实现就通过ofproto_class统一。
ovs bridge的创建流程:
1.通过ovs-vsctl 创建网桥,将创建参数发送给ovsdb-server,ovsdb-server将数据写入数据库。
2.ovs-vswitchd从ovsdb-server中读取创建网桥的信息,在ovs-vswithd层创建一个bridge结构体信息
3.然后将brdige信息应用到ofproto层,在ofproto层通过ofproto_create创建网桥,
ofproto_create(br->name, br->type, &br->ofproto)根据配置创建一个交换机,其中type是交换机类型。这个ofproto对象对外和OpenFlow Controller打交道,对内通过netdev和ofproto-dpif和下面打交道。ofproto结构体(该结构体也表示一个bridge)通过ofproto provider 构造函数创建ofproto provider的私有信息(ofproto_class的定义在ofproto-provider.h中,它的实现在ofproto-dpif.c的ofproto_dpif_class中)。
ofproto_class的作用:不同的交换形态(软交换、硬件加速),目前只有一种,软交换。
4.ofproto-dpif通过backer关联dpif。struct dpif_class是datapath interface实现的工厂接口类,用于和实际的datapath(比如openvswitch.ko, 或者userspace datapath)交互。目前已有的两个dpif的实现是dpif-netlink和dpif-netdev,前者是基于内核datapath的dpif实现(linux),后者基于用户态datapath(dpdk):
1)lib/dpif-netlink.c: 特定Linux实现的dpif,该dpif与Open vSwith实现的内核模块通信。内核模块执行所有的交换工作,将内核态不匹配的数据包发送到用户态。dpif封装调用内核接口。
2)lib/dpif-netdev.c :是一种通用的 dpif 实现。该dpif就是Open vSwith在用户态的实现。数据包的交换不会进入内核。
dpif_class的作用:不同的转发平面(用户态、内核态)
用户态dp:基于dpdk技术,轮训转发,适用NFV领域;
内核态dp:内核态交换流表方式,适用于管理平面或IT领域
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:
static void
bridge_run__(void)
{
/* Let each bridge do the work that it needs to do. */
HMAP_FOR_EACH (br, node, &all_bridges) {
ofproto_run(br->ofproto);
}
}
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,它的实现在ofproto-dpif.c的ofproto_dpif_class中:
const struct ofproto_class ofproto_dpif_class = {
init,
enumerate_types,
enumerate_names,
del,
port_open_type,
type_run,
type_wait,
alloc,
construct,
destruct,
dealloc,
run,
wait,
NULL, /* get_memory_usage. */
type_get_memory_usage,
flush,
query_tables,
set_tables_version,
port_alloc,
port_construct,
port_destruct,
port_dealloc,
port_modified,
port_reconfigured,
port_query_by_name,
port_add,
port_del,
port_get_stats,
port_dump_start,
port_dump_next,
port_dump_done,
port_poll,
port_poll_wait,
port_is_lacp_current,
port_get_lacp_stats,
NULL, /* rule_choose_table */
rule_alloc,
rule_construct,
rule_insert,
rule_delete,
rule_destruct,
rule_dealloc,
rule_get_stats,
rule_execute,
set_frag_handling,
packet_out,
set_netflow,
get_netflow_ids,
set_sflow,
set_ipfix,
set_cfm,
cfm_status_changed,
get_cfm_status,
set_lldp,
get_lldp_status,
set_aa,
aa_mapping_set,
aa_mapping_unset,
aa_vlan_get_queued,
aa_vlan_get_queue_size,
set_bfd,
bfd_status_changed,
get_bfd_status,
set_stp,
get_stp_status,
set_stp_port,
get_stp_port_status,
get_stp_port_stats,
set_rstp,
get_rstp_status,
set_rstp_port,
get_rstp_port_status,
set_queues,
bundle_set,
bundle_remove,
mirror_set__,
mirror_get_stats__,
set_flood_vlans,
is_mirror_output_bundle,
forward_bpdu_changed,
set_mac_table_config,
set_mcast_snooping,
set_mcast_snooping_port,
set_realdev,
NULL, /* meter_get_features */
NULL, /* meter_set */
NULL, /* meter_get */
NULL, /* meter_del */
group_alloc, /* group_alloc */
group_construct, /* group_construct */
group_destruct, /* group_destruct */
group_dealloc, /* group_dealloc */
group_modify, /* group_modify */
group_get_stats, /* group_get_stats */
get_datapath_version, /* get_datapath_version */
};
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,包括以下调用:
// ofproto_create根据配置创建一个交换机
ofproto_create(br->name, br->type, &br->ofproto)
↓
ofproto->ofproto_class->construct(ofproto)
//set config
ofproto_bridge_set_config
// 对于每一个网桥,将网卡添加进去:
HMAP_FOR_EACH (br, node, &all_bridges) {
bridge_add_ports(br, &br->wanted_ports);
shash_destroy(&br->wanted_ports);
}
static void
bridge_add_ports(struct bridge *br, const struct shash *wanted_ports)
{
/* First add interfaces that request a particular port number. */
bridge_add_ports__(br, wanted_ports, true);
/* Then add interfaces that want automatic port number assignment.
* We add these afterward to avoid accidentally taking a specifically
* requested port number. */
bridge_add_ports__(br, wanted_ports, false);
}
然后依次调用:
static void bridge_add_ports__(struct bridge *br, const struct shash *wanted_ports, bool with_requested_port)
↓
static bool iface_create(struct bridge *br, const struct ovsrec_interface *iface_cfg, const struct ovsrec_port *port_cfg)
↓
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)
↓
int ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev, ofp_port_t *ofp_portp)
↓
error = ofproto->ofproto_class->port_add(ofproto, netdev);
然后ofproto-dpif.c中的ofproto_dpif_class的 port_add函数会通过 dpif_port_add函数调用
error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
用户态dp会调用dpif_netdev_class的dpif_netdev_port_add函数;内核态dp会调用dpif_netlink_class的dpif_netlink_port_add,会调用 dpif_netlink_port_add__,在这个函数里面,会调用netlink的API,命令为OVS_VPORT_CMD_NEW
const char *name = netdev_vport_get_dpif_port(netdev,
namebuf, sizeof namebuf);
struct dpif_netlink_vport request, reply;
struct nl_sock **socksp = NULL;
if (dpif->handlers) {
socksp = vport_create_socksp(dpif, &error);
if (!socksp) {
return error;
}
}
dpif_netlink_vport_init(&request);
request.cmd = OVS_VPORT_CMD_NEW;
request.dp_ifindex = dpif->dp_ifindex;
request.type = netdev_to_ovs_vport_type(netdev);
request.name = name;
upcall_pids = vport_socksp_to_pids(socksp, dpif->n_handlers);
request.n_upcall_pids = socksp ? dpif->n_handlers : 1;
request.upcall_pids = upcall_pids;
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。
二、虚拟网卡的初始化netdev_run()
void
netdev_run(void)
OVS_EXCLUDED(netdev_class_mutex, netdev_mutex)
{
struct netdev_registered_class *rc;
netdev_initialize();
ovs_mutex_lock(&netdev_class_mutex);
HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
if (rc->class->run) {
rc->class->run();
}
}
ovs_mutex_unlock(&netdev_class_mutex);
}
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:
static const struct netdev_class dpdk_class =
NETDEV_DPDK_CLASS(
"dpdk",
NULL,
netdev_dpdk_construct,
netdev_dpdk_destruct,
netdev_dpdk_set_multiq,
netdev_dpdk_eth_send,
netdev_dpdk_get_carrier,
netdev_dpdk_get_stats,
netdev_dpdk_get_features,
netdev_dpdk_get_status,
netdev_dpdk_rxq_recv);
对于物理网卡,也需要有相应的netdev_class:
const struct netdev_class netdev_linux_class =
NETDEV_LINUX_CLASS(
"system",
netdev_linux_construct,
netdev_linux_get_stats,
netdev_linux_get_features,
netdev_linux_get_status);
对于连接到KVM的tap网卡:
const struct netdev_class netdev_tap_class =
NETDEV_LINUX_CLASS(
"tap",
netdev_linux_construct_tap,
netdev_tap_get_stats,
netdev_linux_get_features,
netdev_linux_get_status);
对于虚拟的软网卡,比如veth pair:
const struct netdev_class netdev_internal_class =
NETDEV_LINUX_CLASS(
"internal",
netdev_linux_construct,
netdev_internal_get_stats,
NULL, /* get_features */
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。
#define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, \
GET_FEATURES, GET_STATUS) \
{ \
NAME, \
\
NULL, \
netdev_linux_run, \
netdev_linux_wait, \
\
netdev_linux_alloc, \
CONSTRUCT, \
netdev_linux_destruct, \
netdev_linux_dealloc, \
NULL, /* get_config */ \
NULL, /* set_config */ \
NULL, /* get_tunnel_config */ \
NULL, /* build header */ \
NULL, /* push header */ \
NULL, /* pop header */ \
NULL, /* get_numa_id */ \
NULL, /* set_multiq */ \
\
netdev_linux_send, \
netdev_linux_send_wait, \
\
netdev_linux_set_etheraddr, \
netdev_linux_get_etheraddr, \
netdev_linux_get_mtu, \
netdev_linux_set_mtu, \
netdev_linux_get_ifindex, \
netdev_linux_get_carrier, \
netdev_linux_get_carrier_resets, \
netdev_linux_set_miimon_interval, \
GET_STATS, \
\
GET_FEATURES, \
netdev_linux_set_advertisements, \
\
netdev_linux_set_policing, \
netdev_linux_get_qos_types, \
netdev_linux_get_qos_capabilities, \
netdev_linux_get_qos, \
netdev_linux_set_qos, \
netdev_linux_get_queue, \
netdev_linux_set_queue, \
netdev_linux_delete_queue, \
netdev_linux_get_queue_stats, \
netdev_linux_queue_dump_start, \
netdev_linux_queue_dump_next, \
netdev_linux_queue_dump_done, \
netdev_linux_dump_queue_stats, \
\
netdev_linux_get_in4, \
netdev_linux_set_in4, \
netdev_linux_get_in6, \
netdev_linux_add_router, \
netdev_linux_get_next_hop, \
GET_STATUS, \
netdev_linux_arp_lookup, \
\
netdev_linux_update_flags, \
\
netdev_linux_rxq_alloc, \
netdev_linux_rxq_construct, \
netdev_linux_rxq_destruct, \
netdev_linux_rxq_dealloc, \
netdev_linux_rxq_recv, \
netdev_linux_rxq_wait, \
netdev_linux_rxq_drain, \
}
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:
error = nl_sock_recv(sock, &buf, false);
if (!error) {
struct rtnetlink_change change;
if (rtnetlink_parse(&buf, &change)) {
struct netdev *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, &change);
ovs_mutex_unlock(&netdev->mutex);
}
netdev_close(netdev_);
}
}
总结:
bridge_run__ 循环处理一些必要的操作,如stp、rstp、mcast处理等,同时ovs-ofctl 下发openflow,也是在这里处理。bridge_reconfigure 主要完成根据数据以及当前进程信息,创建、更新、删除必要的网桥、接口、端口以及其他协议的配置等。最终这些操作为应用到ofproto 层、ofproto dpif 层、run_stats_update、run_status_update、run_system_stats 更新openvswitch数据库状态信息。netdev_run监控网卡状态并更新。
补充:
ovs+dpdk场景下,对应的netdev_dpdk_run为NULL,前面介绍过,用户态dp会调用dpif_netdev_class的dpif_netdev_port_add函数,在其中启动pmd线程,采用同步轮询机制来接收和发送报文,而不是中断,以此提高报文的处理效率。