本文链接： https://blog.csdn.net/zhonglinzhang/article/details/98183625

源码路径： github.com/projectcalico/felix/routetable/route_table.go

RouteTable 结构体定义如下：

type RouteTable struct {
   logCxt *log.Entry

   ipVersion      uint8
   netlinkFamily  int
   netlinkTimeout time.Duration
   // numConsistentNetlinkFailures counts the number of repeated netlink connection failures.
   // reset on successful connection.
   numConsistentNetlinkFailures int
   // Current netlink handle, or nil if we need to reconnect.
   cachedNetlinkHandle HandleIface

   dirtyIfaces set.Set

   ifacePrefixes     set.Set
   ifacePrefixRegexp *regexp.Regexp

   ifaceNameToTargets          map[string][]Target
   ifaceNameToL2Targets        map[string][]L2Target
   ifaceNameToFirstSeen        map[string]time.Time
   pendingIfaceNameToTargets   map[string][]Target
   pendingIfaceNameToL2Targets map[string][]L2Target

   pendingConntrackCleanups map[ip.Addr]chan struct{}

   inSync bool

   // Whether this route table is managing vxlan routes.
   vxlan bool

   // Testing shims, swapped with mock versions for UT
   newNetlinkHandle  func() (HandleIface, error)
   addStaticARPEntry func(cidr ip.CIDR, destMAC net.HardwareAddr, ifaceName string) error
   conntrack         conntrackIface
   time              timeIface
}

NewIntDataplaneDriver

--> routetable.New

1. 实例化 RouteTable

interfacePrefixes 接口前缀为 cali
ipVersion 为 4
NetlinkTimeout 为10000000000，大约 10s

func New(interfacePrefixes []string, ipVersion uint8, vxlan bool, netlinkTimeout time.Duration) *RouteTable {
	return NewWithShims(
		interfacePrefixes,
		ipVersion,
		newNetlinkHandle,
		vxlan,
		netlinkTimeout,
		addStaticARPEntry,
		conntrack.New(),
		realTime{},
	)
}

loopUpdatingDataplane

--> processIfaceUpdate

--> onIfaceStateChanged

2. onIfaceStateChanged 函数

如果 iface 接口没有按照规则 ^cali.* 匹配，则无需处理，这个函数就是处理 calico 接口

如果 iface 为 up 状态的，则加入到 dirtyInfaces 等待处理这个接口

func (r *RouteTable) OnIfaceStateChanged(ifaceName string, state ifacemonitor.State) {
	logCxt := r.logCxt.WithField("ifaceName", ifaceName)
	if !r.ifacePrefixRegexp.MatchString(ifaceName) {
		logCxt.Debug("Ignoring interface state change, not a Calico interface.")
		return
	}
	if state == ifacemonitor.StateUp {
		logCxt.Debug("Interface up, marking for route sync")
		r.dirtyIfaces.Add(ifaceName)
		r.onIfaceSeen(ifaceName)
	}
}

loopUpdatingDataplane

--> apply

--> RouteTable Apply

3. RouteTable Apply 函数

func (r *RouteTable) Apply() error {
	if !r.inSync {
		listStartTime := time.Now()

		nl, err := r.getNetlinkHandle()
		if err != nil {
			r.logCxt.WithError(err).Error("Failed to connect to netlink, retrying...")
			return ConnectFailed
		}
		links, err := nl.LinkList()
		if err != nil {
			r.logCxt.WithError(err).Error("Failed to list interfaces, retrying...")
			r.closeNetlinkHandle() // Defensive: force a netlink reconnection next time.
			return ListFailed
		}

根据 inSync 已经设置为 false，getNetlinkHandle 根据初始化时为 route，

func (r *RouteTable) QueueResync() {
	r.logCxt.Info("Queueing a resync of routing table.")
	r.inSync = false
}

3.1 将 calico 接口名加入到 dirtyIfaces 中

会对 dirtyIfaces 迭代更新操作，后文接续分析

// Clear the dirty set; there's no point trying to update non-existent interfaces.
r.dirtyIfaces = set.New()
for _, link := range links {
	attrs := link.Attrs()
	if attrs == nil {
		continue
	}
	ifaceName := attrs.Name
	if r.ifacePrefixRegexp.MatchString(ifaceName) {
		r.logCxt.WithField("ifaceName", ifaceName).Debug(
			"Resync: found calico-owned interface")
		r.dirtyIfaces.Add(ifaceName)
		r.onIfaceSeen(ifaceName)
	}
}

迭代所有 dirtyIfaces 进行处理，一步步分析，最大重试次数为 2

3.2 如果设置为 vxlan 模式

if r.vxlan {
	// Sync L2 routes first.
	err := r.syncL2RoutesForLink(ifaceName)
	if err == IfaceNotPresent {
		logCxt.Info("Interface missing, will retry if it appears.")
		break
	} else if err == IfaceDown {
		logCxt.Info("Interface down, will retry if it goes up.")
		break
	} else if err != nil {
		logCxt.WithError(err).Warn("Failed to syncronise routes.")
		retries--
		continue
	}
	logCxt.Debug("Synchronised L2 routes on interface")
}

3.2.1 syncL2RoutesForLink

上一次加入 pendingIfaceNameToL2Targets，本轮将从 pendingIfaceNameToL2Targets 加入到 ifaceNameToL2Targets，即本轮期望加入的

func (r *RouteTable) syncL2RoutesForLink(ifaceName string) error {
	logCxt := r.logCxt.WithField("ifaceName", ifaceName)
	logCxt.Debug("Syncing interface L2 routes")
	if updatedTargets, ok := r.pendingIfaceNameToL2Targets[ifaceName]; ok {
		logCxt.Debug("Have updated targets.")
		if updatedTargets == nil {
			delete(r.ifaceNameToL2Targets, ifaceName)
		} else {
			r.ifaceNameToL2Targets[ifaceName] = updatedTargets
		}
		delete(r.pendingIfaceNameToL2Targets, ifaceName)
	}
	expectedTargets := r.ifaceNameToL2Targets[ifaceName]

3.2.1.1 arp 表和 fdb 表

ARP表：IP和MAC的对应关系；
FDB表：MAC+VLAN和PORT的对应关系

ARP是三层转发，FDB是用于二层转发。两个设备不在一个网段或者没配IP，只要两者之间的链路层是连通的，就可以通过FDB表进行数据的转发

FDB表的作用就在于告诉设备从某个端口出去就可以到某个目的MAC

// Build maps based on desired target state, used below to clean up
// stale entries. Each L2 target results in an ARP entry as well as
// a FDB entry.
expectedARPEntries := map[string]net.HardwareAddr{}
expectedFDBEntries := map[string]net.HardwareAddr{}
for _, target := range expectedTargets {
	expectedARPEntries[target.GW.String()] = target.VTEPMAC
	expectedFDBEntries[target.IP.String()] = target.VTEPMAC
}

3.2.1.2 查看 arp 地址表，相当于 ip neighbor show

# ip neighbor show
192.168.1.1 dev enp0s3 lladdr f4:b8:a7:4b:57:da STALE
192.168.1.6 dev enp0s3 lladdr 74:c6:3b:61:51:fb REACHABLE
fe80::1 dev enp0s3 lladdr f4:b8:a7:4b:57:da router STALE

// Get the current set of neighbors on this interface.
existingNeigh, err := netlink.NeighList(linkAttrs.Index, netlink.FAMILY_V4)
if err != nil {
	return err
}

3.2.1.3 对于所有的 arp 表，如果不再期待中，则删除，调用命令为 ip neighbor del

// For each existing neighbor, if it is not present in the expected set, then remove it.
var updatesFailed bool
for _, existing := range existingNeigh {
	if existing.Family == syscall.AF_BRIDGE {
		// FDB entries have family set to bridge.
		if _, ok := expectedFDBEntries[existing.IP.String()]; !ok {
			logCxt.WithField("neighbor", existing).Info("Removing old neighbor entry (FDB)")
			if err := netlink.NeighDel(&existing); err != nil {
				updatesFailed = true
				continue
			}
		}
	} else {
		if _, ok := expectedARPEntries[existing.IP.String()]; !ok {
			logCxt.WithField("neighbor", existing).Info("Removing old neighbor entry (ARP)")
			if err := netlink.NeighDel(&existing); err != nil {
				updatesFailed = true
				continue
			}
		}

	}
}

3.3 syncRoutesForLink 同步设置 L3 路由

// Sync L3 routes.
err := r.syncRoutesForLink(ifaceName)
if err == IfaceNotPresent {
	logCxt.Info("Interface missing, will retry if it appears.")
	break
} else if err == IfaceDown {
	logCxt.Info("Interface down, will retry if it goes up.")
	break
} else if err == IfaceGrace {
	logCxt.Info("Interface in cleanup grace period, will retry after.")
	graceIfaces++
	return nil
} else if err != nil {
	logCxt.WithError(err).Warn("Failed to syncronise routes.")
	retries--
	continue
}

3.3.1 syncRoutesForLink 函数

inGracePeriod 主要用于优雅的删除，不是立即删除，给一段时间进行清理工作

func (r *RouteTable) syncRoutesForLink(ifaceName string) error {
	startTime := time.Now()
	defer func() {
		perIfaceSyncTime.Observe(time.Since(startTime).Seconds())
	}()
	logCxt := r.logCxt.WithField("ifaceName", ifaceName)
	logCxt.Debug("Syncing interface routes")

	// In order to allow Calico to run without Felix in an emergency, the CNI plugin pre-adds
	// the route to the interface.  To avoid flapping the route when Felix sees the interface
	// before learning about the endpoint, we give each interface a grace period after we first
	// see it before we remove routes that we're not expecting.  Check whether the grace period
	// applies to this interface.
	inGracePeriod := r.time.Since(r.ifaceNameToFirstSeen[ifaceName]) < cleanupGracePeriod
	leaveDirty := false

3.3.3.1 列出所有路由

// Got the link; try to sync its routes.  Note: We used to check if the interface
// was oper down before we tried to do the sync but that prevented us from removing
// routes from an interface in some corner cases (such as being admin up but oper
// down).
linkAttrs := link.Attrs()
oldRoutes, err := nl.RouteList(link, r.netlinkFamily)

3.3.3.2 对于旧的路由，如果在期待加入的路由中无需处理，如果优雅删除期间则暂时先不清理，删除一些其他路由

for _, route := range oldRoutes {

	seenCIDRs.Add(dest)
	if expectedCIDRs.Contains(dest) {
		logCxt.Debug("Syncing routes: Found expected route.")
		continue
	}
	if !r.vxlan && inGracePeriod {
		continue
	}
	logCxt.Info("Syncing routes: removing old route.")

}

3.3.3.3 对于期待的不在 seenCIDRs 中则调用 Route add 添加路由

for _, target := range expectedTargets {
	cidr := target.CIDR
	if !seenCIDRs.Contains(cidr) {
		logCxt := logCxt.WithField("targetCIDR", target.CIDR)
		logCxt.Info("Syncing routes: adding new route.")
		ipNet := cidr.ToIPNet()
		route := netlink.Route{
			LinkIndex: linkAttrs.Index,
			Dst:       &ipNet,
			Type:      syscall.RTN_UNICAST,
			Protocol:  syscall.RTPROT_BOOT,
			Scope:     netlink.SCOPE_LINK,
		}

		if target.GW != nil {
			route.Gw = target.GW.AsNetIP()
		}

		if target.Type == TargetTypeVXLAN {
			route.Scope = netlink.SCOPE_UNIVERSE
			route.SetFlag(syscall.RTNH_F_ONLINK)
		}

		// In case this IP is being re-used, wait for any previous conntrack entry
		// to be cleaned up.  (No-op if there are no pending deletes.)
		r.waitForPendingConntrackDeletion(cidr.Addr())
		if err := nl.RouteAdd(&route); err != nil {
			logCxt.WithError(err).Warn("Failed to add route")
			updatesFailed = true
		}
	}

3.3.3.4 addStaticARPEntry

如果 ipv4 版本且目的存在 MAC 则添加静态 arp，arp -i eth0 -s 192.168.*** ff:ee:ee:ee:ee:ee 如下：

# arp -a
? (192.168.72.134) at d4:bb:c8:a6:60:c1 [ether] on enp0s3
? (192.168.72.125) at <incomplete> on enp0s3
? (192.168.73.122) at dc:72:9b:42:02:b9 [ether] on enp0s3
gateway (192.168.72.1) at 3c:15:fb:09:8b:8c [ether] on enp0s3
? (192.168.72.140) at fc:2a:9c:ec:63:77 [ether] on enp0s3
? (192.170.77.147) at 7a:ae:8c:b8:09:e9 [ether] on cali691e7cb552c
? (192.168.74.100) at 20:ab:37:5e:82:5b [ether] on enp0s3
? (192.168.72.132) at 5c:03:39:9d:c8:c3 [ether] on enp0s3
? (192.168.73.64) at 74:c6:3b:61:51:fb [ether] on enp0s3
? (192.168.73.100) at f8:38:80:3e:b9:aa [ether] on enp0s3
? (192.168.73.232) at 84:be:52:ae:f1:fc [ether] on enp0s3
? (192.168.72.62) at c8:14:51:4e:39:66 [ether] on enp0s3
? (192.168.73.50) at 4c:32:75:08:84:34 [ether] on enp0s3
? (192.168.72.212) at 34:7c:25:53:db:45 [ether] on enp0s3
? (192.170.77.149) at 7e:24:34:c6:10:99 [ether] on calif4ae9085e4a
? (192.170.77.148) at 7a:cf:86:10:b6:33 [ether] on cali8bbfee2ad9e

if r.ipVersion == 4 && target.DestMAC != nil {
	// TODO(smc) clean up/sync old ARP entries
	err := r.addStaticARPEntry(cidr, target.DestMAC, ifaceName)
	if err != nil {
		logCxt.WithError(err).Warn("Failed to set ARP entry")
		updatesFailed = true
	}
}

4. 主要由更新 iface 或者 iface addr 变化时调用，存入 pendingIfaceNameToTargets

func (r *RouteTable) SetRoutes(ifaceName string, targets []Target) {
	r.pendingIfaceNameToTargets[ifaceName] = targets
	r.dirtyIfaces.Add(ifaceName)
}

func (r *RouteTable) SetL2Routes(ifaceName string, targets []L2Target) {
	r.pendingIfaceNameToL2Targets[ifaceName] = targets
	r.dirtyIfaces.Add(ifaceName)
}

总结：

本文分析的简单，主要是根据 iface，l2 或者 l3 变更路由

【kubernetes/k8s源码分析】calico node felix routetable 源码分析