ovs-vswitchd.c的main函數最終會進入一個while循環,在這個無限循環中,裏面最重要的兩個函數是bridge_run()和netdev_run()。
bridge_init(remote);
free(remote);
exiting = false;
while (!exiting) { //while循環,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(); //處理controller交互,ovs-ofctl命令
unixctl_server_run(unixctl); //處理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,當沒有請求處理的時候,阻塞在此處
if (should_service_stop()) {
exiting = true;
}
}
bridge_exit();
unixctl_server_destroy(unixctl);
service_stop();
Openvswitch主要管理兩種類型的設備,一個是創建的虛擬網橋,一個是連接到虛擬網橋上的設備。
其中bridge_run就是初始化數據庫中已經創建的虛擬網橋。
在具體介紹代碼調用之前,需要回顧一下OVS 的代碼架構:
ovs-vswitchd是ovs中最核心的組件,openflow的相關邏輯都在vswitchd裏實現,一般來說ovs分爲datapath, vswitchd以及ovsdb三個部分,datapath一般是和具體是數據面平臺相關的,比如白盒交換機,或者linux內核等。ovsdb用於存儲vswitch本身的配置信息,比如端口,拓撲,規則等。vswitchd本身是分層的結構,最上層daemon主要用於和ovsdb通信,做配置的下發和更新等,中間層ofproto,用於和openflow控制器通信,以及通過ofproto_class暴露了ofproto provider接口(目前只支持一個ofproto provider,即ofproto-dpif),不同平臺上openflow的具體實現就通過ofproto_class統一。
ovs bridge的創建流程:
1.通過ovs-vsctl 創建網橋,將創建參數發送給ovsdb-server,ovsdb-server將數據寫入數據庫。
2.ovs-vswitchd從ovsdb-server中讀取創建網橋的信息,在ovs-vswithd層創建一個bridge結構體信息
3.然後將brdige信息應用到ofproto層,在ofproto層通過ofproto_create創建網橋,
ofproto_create(br->name, br->type, &br->ofproto)根據配置創建一個交換機,其中type是交換機類型。這個ofproto對象對外和OpenFlow Controller打交道,對內通過netdev和ofproto-dpif和下面打交道。ofproto結構體(該結構體也表示一個bridge)通過ofproto provider 構造函數創建ofproto provider的私有信息(ofproto_class的定義在ofproto-provider.h中,它的實現在ofproto-dpif.c的ofproto_dpif_class中)。
ofproto_class的作用:不同的交換形態(軟交換、硬件加速),目前只有一種,軟交換。
4.ofproto-dpif通過backer關聯dpif。struct dpif_class是datapath interface實現的工廠接口類,用於和實際的datapath(比如openvswitch.ko, 或者userspace datapath)交互。目前已有的兩個dpif的實現是dpif-netlink和dpif-netdev,前者是基於內核datapath的dpif實現(linux),後者基於用戶態datapath(dpdk):
1)lib/dpif-netlink.c: 特定Linux實現的dpif,該dpif與Open vSwith實現的內核模塊通信。內核模塊執行所有的交換工作,將內核態不匹配的數據包發送到用戶態。dpif封裝調用內核接口。
2)lib/dpif-netdev.c :是一種通用的 dpif 實現。該dpif就是Open vSwith在用戶態的實現。數據包的交換不會進入內核。
dpif_class的作用:不同的轉發平面(用戶態、內核態)
用戶態dp:基於dpdk技術,輪訓轉發,適用NFV領域;
內核態dp:內核態交換流表方式,適用於管理平面或IT領域
一、虛擬網橋的初始化bridge_run
bridge_run會調用bridge_run__,bridge_run__中最重要的是對於所有的網橋,都調用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)會調用error = p->ofproto_class->run§;
ofproto_class的定義在ofproto-provider.h中,它的實現在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中註釋裏是這樣說的,這裏定義了四類數據結構:
- Struct ofproto表示一個交換機
- Struct ofport表示交換機上的一個端口
- Struct rule表示交換機上的一條flow規則
- Struct ofgroup表示一個flow規則組
上面說到啓動的過程中,會調用ofproto_class->run,也即會調用ofproto-dpif.c中的static int run(struct ofproto *ofproto_)函數。在這個函數中,會初始化netflow, sflow, ipfix,stp, rstp, mac address learning等一系列操作。
bridge_run還會調用static void bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg),其中ovs_cfg是從ovsdb-server裏面讀取出來的配置。
在這個函數裏面,包括以下調用:
// ofproto_create根據配置創建一個交換機
ofproto_create(br->name, br->type, &br->ofproto)
↓
ofproto->ofproto_class->construct(ofproto)
//set config
ofproto_bridge_set_config
// 對於每一個網橋,將網卡添加進去:
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);
}
然後依次調用:
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);
然後ofproto-dpif.c中的ofproto_dpif_class的 port_add函數會通過 dpif_port_add函數調用
error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
用戶態dp會調用dpif_netdev_class的dpif_netdev_port_add函數,之後便會啓動pmd轉發線程,詳見
https://blog.csdn.net/qq_20817327/article/details/106761936
內核態dp會調用dpif_netlink_class的dpif_netlink_port_add,會調用 dpif_netlink_port_add__,在這個函數裏面,會調用netlink的API,命令爲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);
這裏會調用內核模塊openvswitch.ko,在內核中添加虛擬網卡。
二、虛擬網卡的初始化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);
}
依次循環調用netdev_classes中的每一個run。
對於不同類型的虛擬網卡,都有對應的netdev_class。
例如對於dpdk的網卡有:
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);
對於物理網卡,也需要有相應的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);
對於連接到KVM的tap網卡:
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);
對於虛擬的軟網卡,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是一個宏,不是所有的參數都需要全部填寫。
#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()調用的是netdev-linux.c下的netdev_linux_run。
netdev_linux_run會調用netlink的sock得到虛擬網卡的狀態,並且更新狀態:
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_);
}
}
總結:
bridge_run__ 循環處理一些必要的操作,如stp、rstp、mcast處理等,同時ovs-ofctl 下發openflow,也是在這裏處理。bridge_reconfigure 主要完成根據數據以及當前進程信息,創建、更新、刪除必要的網橋、接口、端口以及其他協議的配置等。最終這些操作爲應用到ofproto 層、ofproto dpif 層、run_stats_update、run_status_update、run_system_stats 更新openvswitch數據庫狀態信息。netdev_run監控網卡狀態並更新。
補充:
ovs+dpdk場景下,對應的netdev_dpdk_run爲NULL,前面介紹過,用戶態dp會調用dpif_netdev_class的dpif_netdev_port_add函數,在其中啓動pmd線程,採用同步輪詢機制來接收和發送報文,而不是中斷,以此提高報文的處理效率。
原文鏈接:https://blog.csdn.net/qq_15437629/article/details/79598556