Add routing peer support (#441)

Handle routes updates from management

Manage routing firewall rules

Manage peer RIB table

Add get peer and get notification channel from the status recorder

Update interface peers allowed IPs
This commit is contained in:
Maycon Santos
2022-09-05 09:06:35 +02:00
committed by GitHub
parent 788bb00ef1
commit 1012172f04
26 changed files with 3030 additions and 35 deletions

View File

@ -0,0 +1,384 @@
package routemanager
import (
"context"
"fmt"
"github.com/google/nftables/binaryutil"
"github.com/google/nftables/expr"
log "github.com/sirupsen/logrus"
"net"
"net/netip"
"sync"
)
import "github.com/google/nftables"
//
const (
nftablesTable = "netbird-rt"
nftablesRoutingForwardingChain = "netbird-rt-fwd"
nftablesRoutingNatChain = "netbird-rt-nat"
)
// constants needed to create nftable rules
const (
ipv4Len = 4
ipv4SrcOffset = 12
ipv4DestOffset = 16
ipv6Len = 16
ipv6SrcOffset = 8
ipv6DestOffset = 24
exprDirectionSource = "source"
exprDirectionDestination = "destination"
)
// some presets for building nftable rules
var (
zeroXor = binaryutil.NativeEndian.PutUint32(0)
zeroXor6 = append(binaryutil.NativeEndian.PutUint64(0), binaryutil.NativeEndian.PutUint64(0)...)
exprAllowRelatedEstablished = []expr.Any{
&expr.Ct{
Register: 1,
SourceRegister: false,
Key: 0,
},
&expr.Bitwise{
DestRegister: 1,
SourceRegister: 1,
Len: 4,
Mask: []uint8{0x6, 0x0, 0x0, 0x0},
Xor: zeroXor,
},
&expr.Cmp{
Register: 1,
Data: binaryutil.NativeEndian.PutUint32(0),
},
&expr.Counter{},
&expr.Verdict{
Kind: expr.VerdictAccept,
},
}
exprCounterAccept = []expr.Any{
&expr.Counter{},
&expr.Verdict{
Kind: expr.VerdictAccept,
},
}
)
type nftablesManager struct {
ctx context.Context
stop context.CancelFunc
conn *nftables.Conn
tableIPv4 *nftables.Table
tableIPv6 *nftables.Table
chains map[string]map[string]*nftables.Chain
rules map[string]*nftables.Rule
mux sync.Mutex
}
// CleanRoutingRules cleans existing nftables rules from the system
func (n *nftablesManager) CleanRoutingRules() {
n.mux.Lock()
defer n.mux.Unlock()
log.Debug("flushing tables")
n.conn.FlushTable(n.tableIPv6)
n.conn.FlushTable(n.tableIPv4)
log.Debugf("flushing tables result in: %v error", n.conn.Flush())
}
// RestoreOrCreateContainers restores existing nftables containers (tables and chains)
// if they don't exist, we create them
func (n *nftablesManager) RestoreOrCreateContainers() error {
n.mux.Lock()
defer n.mux.Unlock()
if n.tableIPv6 != nil && n.tableIPv4 != nil {
log.Debugf("nftables: containers already restored, skipping")
return nil
}
tables, err := n.conn.ListTables()
if err != nil {
return fmt.Errorf("nftables: unable to list tables: %v", err)
}
for _, table := range tables {
if table.Name == nftablesTable {
if table.Family == nftables.TableFamilyIPv4 {
n.tableIPv4 = table
continue
}
n.tableIPv6 = table
}
}
if n.tableIPv4 == nil {
n.tableIPv4 = n.conn.AddTable(&nftables.Table{
Name: nftablesTable,
Family: nftables.TableFamilyIPv4,
})
}
if n.tableIPv6 == nil {
n.tableIPv6 = n.conn.AddTable(&nftables.Table{
Name: nftablesTable,
Family: nftables.TableFamilyIPv6,
})
}
chains, err := n.conn.ListChains()
if err != nil {
return fmt.Errorf("nftables: unable to list chains: %v", err)
}
n.chains[ipv4] = make(map[string]*nftables.Chain)
n.chains[ipv6] = make(map[string]*nftables.Chain)
for _, chain := range chains {
switch {
case chain.Table.Name == nftablesTable && chain.Table.Family == nftables.TableFamilyIPv4:
n.chains[ipv4][chain.Name] = chain
case chain.Table.Name == nftablesTable && chain.Table.Family == nftables.TableFamilyIPv6:
n.chains[ipv6][chain.Name] = chain
}
}
if _, found := n.chains[ipv4][nftablesRoutingForwardingChain]; !found {
n.chains[ipv4][nftablesRoutingForwardingChain] = n.conn.AddChain(&nftables.Chain{
Name: nftablesRoutingForwardingChain,
Table: n.tableIPv4,
Hooknum: nftables.ChainHookForward,
Priority: nftables.ChainPriorityNATDest + 1,
Type: nftables.ChainTypeFilter,
})
}
if _, found := n.chains[ipv4][nftablesRoutingNatChain]; !found {
n.chains[ipv4][nftablesRoutingNatChain] = n.conn.AddChain(&nftables.Chain{
Name: nftablesRoutingNatChain,
Table: n.tableIPv4,
Hooknum: nftables.ChainHookPostrouting,
Priority: nftables.ChainPriorityNATSource - 1,
Type: nftables.ChainTypeNAT,
})
}
if _, found := n.chains[ipv6][nftablesRoutingForwardingChain]; !found {
n.chains[ipv6][nftablesRoutingForwardingChain] = n.conn.AddChain(&nftables.Chain{
Name: nftablesRoutingForwardingChain,
Table: n.tableIPv6,
Hooknum: nftables.ChainHookForward,
Priority: nftables.ChainPriorityNATDest + 1,
Type: nftables.ChainTypeFilter,
})
}
if _, found := n.chains[ipv6][nftablesRoutingNatChain]; !found {
n.chains[ipv6][nftablesRoutingNatChain] = n.conn.AddChain(&nftables.Chain{
Name: nftablesRoutingNatChain,
Table: n.tableIPv6,
Hooknum: nftables.ChainHookPostrouting,
Priority: nftables.ChainPriorityNATSource - 1,
Type: nftables.ChainTypeNAT,
})
}
err = n.refreshRulesMap()
if err != nil {
return err
}
n.checkOrCreateDefaultForwardingRules()
err = n.conn.Flush()
if err != nil {
return fmt.Errorf("nftables: unable to initialize table: %v", err)
}
return nil
}
// refreshRulesMap refreshes the rule map with the latest rules. this is useful to avoid
// duplicates and to get missing attributes that we don't have when adding new rules
func (n *nftablesManager) refreshRulesMap() error {
for _, registeredChains := range n.chains {
for _, chain := range registeredChains {
rules, err := n.conn.GetRules(chain.Table, chain)
if err != nil {
return fmt.Errorf("nftables: unable to list rules: %v", err)
}
for _, rule := range rules {
if len(rule.UserData) > 0 {
n.rules[string(rule.UserData)] = rule
}
}
}
}
return nil
}
// checkOrCreateDefaultForwardingRules checks if the default forwarding rules are enabled
func (n *nftablesManager) checkOrCreateDefaultForwardingRules() {
_, foundIPv4 := n.rules[ipv4Forwarding]
if !foundIPv4 {
n.rules[ipv4Forwarding] = n.conn.AddRule(&nftables.Rule{
Table: n.tableIPv4,
Chain: n.chains[ipv4][nftablesRoutingForwardingChain],
Exprs: exprAllowRelatedEstablished,
UserData: []byte(ipv4Forwarding),
})
}
_, foundIPv6 := n.rules[ipv6Forwarding]
if !foundIPv6 {
n.rules[ipv6Forwarding] = n.conn.AddRule(&nftables.Rule{
Table: n.tableIPv6,
Chain: n.chains[ipv6][nftablesRoutingForwardingChain],
Exprs: exprAllowRelatedEstablished,
UserData: []byte(ipv6Forwarding),
})
}
}
// InsertRoutingRules inserts a nftable rule pair to the forwarding chain and if enabled, to the nat chain
func (n *nftablesManager) InsertRoutingRules(pair routerPair) error {
n.mux.Lock()
defer n.mux.Unlock()
prefix := netip.MustParsePrefix(pair.source)
sourceExp := generateCIDRMatcherExpressions("source", pair.source)
destExp := generateCIDRMatcherExpressions("destination", pair.destination)
forwardExp := append(sourceExp, append(destExp, exprCounterAccept...)...)
fwdKey := genKey(forwardingFormat, pair.ID)
if prefix.Addr().Unmap().Is4() {
n.rules[fwdKey] = n.conn.InsertRule(&nftables.Rule{
Table: n.tableIPv4,
Chain: n.chains[ipv4][nftablesRoutingForwardingChain],
Exprs: forwardExp,
UserData: []byte(fwdKey),
})
} else {
n.rules[fwdKey] = n.conn.InsertRule(&nftables.Rule{
Table: n.tableIPv6,
Chain: n.chains[ipv6][nftablesRoutingForwardingChain],
Exprs: forwardExp,
UserData: []byte(fwdKey),
})
}
if pair.masquerade {
natExp := append(sourceExp, append(destExp, &expr.Counter{}, &expr.Masq{})...)
natKey := genKey(natFormat, pair.ID)
if prefix.Addr().Unmap().Is4() {
n.rules[natKey] = n.conn.InsertRule(&nftables.Rule{
Table: n.tableIPv4,
Chain: n.chains[ipv4][nftablesRoutingNatChain],
Exprs: natExp,
UserData: []byte(natKey),
})
} else {
n.rules[natKey] = n.conn.InsertRule(&nftables.Rule{
Table: n.tableIPv6,
Chain: n.chains[ipv6][nftablesRoutingNatChain],
Exprs: natExp,
UserData: []byte(natKey),
})
}
}
err := n.conn.Flush()
if err != nil {
return fmt.Errorf("nftables: unable to insert rules for %s: %v", pair.destination, err)
}
return nil
}
// RemoveRoutingRules removes a nftable rule pair from forwarding and nat chains
func (n *nftablesManager) RemoveRoutingRules(pair routerPair) error {
n.mux.Lock()
defer n.mux.Unlock()
err := n.refreshRulesMap()
if err != nil {
return err
}
fwdKey := genKey(forwardingFormat, pair.ID)
natKey := genKey(natFormat, pair.ID)
fwdRule, found := n.rules[fwdKey]
if found {
err = n.conn.DelRule(fwdRule)
if err != nil {
return fmt.Errorf("nftables: unable to remove forwarding rule for %s: %v", pair.destination, err)
}
log.Debugf("nftables: removing forwarding rule for %s", pair.destination)
delete(n.rules, fwdKey)
}
natRule, found := n.rules[natKey]
if found {
err = n.conn.DelRule(natRule)
if err != nil {
return fmt.Errorf("nftables: unable to remove nat rule for %s: %v", pair.destination, err)
}
log.Debugf("nftables: removing nat rule for %s", pair.destination)
delete(n.rules, natKey)
}
err = n.conn.Flush()
if err != nil {
return fmt.Errorf("nftables: received error while applying rule removal for %s: %v", pair.destination, err)
}
log.Debugf("nftables: removed rules for %s", pair.destination)
return nil
}
// getPayloadDirectives get expression directives based on ip version and direction
func getPayloadDirectives(direction string, isIPv4 bool, isIPv6 bool) (uint32, uint32, []byte) {
switch {
case direction == exprDirectionSource && isIPv4:
return ipv4SrcOffset, ipv4Len, zeroXor
case direction == exprDirectionDestination && isIPv4:
return ipv4DestOffset, ipv4Len, zeroXor
case direction == exprDirectionSource && isIPv6:
return ipv6SrcOffset, ipv6Len, zeroXor6
case direction == exprDirectionDestination && isIPv6:
return ipv6DestOffset, ipv6Len, zeroXor6
default:
panic("no matched payload directive")
}
}
// generateCIDRMatcherExpressions generates nftables expressions that matches a CIDR
func generateCIDRMatcherExpressions(direction string, cidr string) []expr.Any {
ip, network, _ := net.ParseCIDR(cidr)
ipToAdd, _ := netip.AddrFromSlice(ip)
add := ipToAdd.Unmap()
offSet, packetLen, zeroXor := getPayloadDirectives(direction, add.Is4(), add.Is6())
return []expr.Any{
// fetch src add
&expr.Payload{
DestRegister: 1,
Base: expr.PayloadBaseNetworkHeader,
Offset: offSet,
Len: packetLen,
},
// net mask
&expr.Bitwise{
DestRegister: 1,
SourceRegister: 1,
Len: packetLen,
Mask: network.Mask,
Xor: zeroXor,
},
// net address
&expr.Cmp{
Register: 1,
Data: add.AsSlice(),
},
}
}