[client] Unify peer and route ACL filtering with multi-source rules (#6322)

* Unify peer and route ACL filtering with multi-source peer rules

* Remove partial userspace firewall mode and open foreign chains via a table-less allower

* Snapshot iptables rule maps before persisting state

* Scope userspace firewall wildcard source rules per address family

* Install nftables peer filter and mangle rules in a single transaction

* Share the iptables jump rule spec between install and cleanup

* Fix legacy ACL source wildcard and keep rollback tracking on delete failure

* Fix CI: recognize multi-value port set lookups in tests and correct PeerIP lint suppression

* Fall back to per-prefix filter rules when ipset is unavailable

* Annotate legacy PeerIP usages in ACL tests and fix import formatting

* Keep firewall rule bookkeeping in step with the kernel on replace and teardown

* Release the routing reference when the route manager shuts down

* Keep set references and rule tracking consistent when a routing rule fails

client/firewall/nftables/acl_linux.go

@@ -1,713 +0,0 @@
-package nftables
-
-import (
-	"bytes"
-	"fmt"
-	"net"
-	"slices"
-	"strconv"
-	"strings"
-	"time"
-
-	"github.com/google/nftables"
-	"github.com/google/nftables/binaryutil"
-	"github.com/google/nftables/expr"
-	log "github.com/sirupsen/logrus"
-	"golang.org/x/sys/unix"
-
-	firewall "github.com/netbirdio/netbird/client/firewall/manager"
-	nbnet "github.com/netbirdio/netbird/client/net"
-)
-
-const (
-
-	// rules chains contains the effective ACL rules
-	chainNameInputRules = "netbird-acl-input-rules"
-
-	// filter chains contains the rules that jump to the rules chains
-	chainNameInputFilter       = "netbird-acl-input-filter"
-	chainNameForwardFilter     = "netbird-acl-forward-filter"
-	chainNameManglePrerouting  = "netbird-mangle-prerouting"
-	chainNameManglePostrouting = "netbird-mangle-postrouting"
-)
-
-const flushError = "flush: %w"
-
-type AclManager struct {
-	rConn              *nftables.Conn
-	sConn              *nftables.Conn
-	wgIface            iFaceMapper
-	routingFwChainName string
-	af                 addrFamily
-
-	workTable       *nftables.Table
-	chainInputRules *nftables.Chain
-	chainPrerouting *nftables.Chain
-
-	ipsetStore *ipsetStore
-	rules      map[string]*Rule
-}
-
-func newAclManager(table *nftables.Table, wgIface iFaceMapper, routingFwChainName string) (*AclManager, error) {
-	// sConn is used for creating sets and adding/removing elements from them
-	// it's differ then rConn (which does create new conn for each flush operation)
-	// and is permanent. Using same connection for both type of operations
-	// overloads netlink with high amount of rules ( > 10000)
-	sConn, err := nftables.New(nftables.AsLasting())
-	if err != nil {
-		return nil, fmt.Errorf("create nf conn: %w", err)
-	}
-
-	return &AclManager{
-		rConn:              &nftables.Conn{},
-		sConn:              sConn,
-		wgIface:            wgIface,
-		workTable:          table,
-		routingFwChainName: routingFwChainName,
-		af:                 familyForAddr(table.Family == nftables.TableFamilyIPv4),
-
-		ipsetStore: newIpsetStore(),
-		rules:      make(map[string]*Rule),
-	}, nil
-}
-
-func (m *AclManager) init(workTable *nftables.Table) error {
-	m.workTable = workTable
-	return m.createDefaultChains()
-}
-
-// AddPeerFiltering rule to the firewall
-//
-// If comment argument is empty firewall manager should set
-// rule ID as comment for the rule
-func (m *AclManager) AddPeerFiltering(
-	id []byte,
-	ip net.IP,
-	proto firewall.Protocol,
-	sPort *firewall.Port,
-	dPort *firewall.Port,
-	action firewall.Action,
-	ipsetName string,
-) ([]firewall.Rule, error) {
-	var ipset *nftables.Set
-	if ipsetName != "" {
-		var err error
-		ipset, err = m.addIpToSet(ipsetName, ip)
-		if err != nil {
-			return nil, err
-		}
-	}
-
-	newRules := make([]firewall.Rule, 0, 2)
-	ioRule, err := m.addIOFiltering(ip, proto, sPort, dPort, action, ipset)
-	if err != nil {
-		return nil, err
-	}
-
-	newRules = append(newRules, ioRule)
-	return newRules, nil
-}
-
-// DeletePeerRule from the firewall by rule definition
-func (m *AclManager) DeletePeerRule(rule firewall.Rule) error {
-	r, ok := rule.(*Rule)
-	if !ok {
-		return fmt.Errorf("invalid rule type")
-	}
-
-	if r.nftSet == nil {
-		if err := m.rConn.DelRule(r.nftRule); err != nil {
-			log.Errorf("failed to delete rule: %v", err)
-		}
-		if r.mangleRule != nil {
-			if err := m.rConn.DelRule(r.mangleRule); err != nil {
-				log.Errorf("failed to delete mangle rule: %v", err)
-			}
-		}
-		delete(m.rules, r.ID())
-		return m.rConn.Flush()
-	}
-
-	ips, ok := m.ipsetStore.ips(r.nftSet.Name)
-	if !ok {
-		if err := m.rConn.DelRule(r.nftRule); err != nil {
-			log.Errorf("failed to delete rule: %v", err)
-		}
-		if r.mangleRule != nil {
-			if err := m.rConn.DelRule(r.mangleRule); err != nil {
-				log.Errorf("failed to delete mangle rule: %v", err)
-			}
-		}
-		delete(m.rules, r.ID())
-		return m.rConn.Flush()
-	}
-
-	if _, ok := ips[r.ip.String()]; ok {
-		err := m.sConn.SetDeleteElements(r.nftSet, []nftables.SetElement{{Key: ipToBytes(r.ip, m.af)}})
-		if err != nil {
-			log.Errorf("delete elements for set %q: %v", r.nftSet.Name, err)
-		}
-		if err := m.sConn.Flush(); err != nil {
-			log.Debugf("flush error of set delete element, %s", r.nftSet.Name)
-			return err
-		}
-		m.ipsetStore.DeleteIpFromSet(r.nftSet.Name, r.ip)
-	}
-
-	// if after delete, set still contains other IPs,
-	// no need to delete firewall rule and we should exit here
-	if len(ips) > 0 {
-		return nil
-	}
-
-	if err := m.rConn.DelRule(r.nftRule); err != nil {
-		log.Errorf("failed to delete rule: %v", err)
-	}
-	if r.mangleRule != nil {
-		if err := m.rConn.DelRule(r.mangleRule); err != nil {
-			log.Errorf("failed to delete mangle rule: %v", err)
-		}
-	}
-
-	if err := m.rConn.Flush(); err != nil {
-		return err
-	}
-
-	delete(m.rules, r.ID())
-	m.ipsetStore.DeleteReferenceFromIpSet(r.nftSet.Name)
-
-	if m.ipsetStore.HasReferenceToSet(r.nftSet.Name) {
-		return nil
-	}
-
-	// we delete last IP from the set, that means we need to delete
-	// set itself and associated firewall rule too
-	m.rConn.FlushSet(r.nftSet)
-	m.rConn.DelSet(r.nftSet)
-	m.ipsetStore.deleteIpset(r.nftSet.Name)
-	return nil
-}
-
-// createDefaultAllowRules creates default allow rules for the input and output chains
-func (m *AclManager) createDefaultAllowRules() error {
-	expIn := []expr.Any{
-		&expr.Verdict{
-			Kind: expr.VerdictAccept,
-		},
-	}
-
-	_ = m.rConn.InsertRule(&nftables.Rule{
-		Table:    m.workTable,
-		Chain:    m.chainInputRules,
-		Position: 0,
-		Exprs:    expIn,
-	})
-
-	if err := m.rConn.Flush(); err != nil {
-		return fmt.Errorf(flushError, err)
-	}
-	return nil
-}
-
-// Flush rule/chain/set operations from the buffer
-//
-// Method also get all rules after flush and refreshes handle values in the rulesets
-func (m *AclManager) Flush() error {
-	if err := m.flushWithBackoff(); err != nil {
-		return err
-	}
-
-	if err := m.refreshRuleHandles(m.chainInputRules, false); err != nil {
-		log.Errorf("failed to refresh rule handles ipv4 input chain: %v", err)
-	}
-	if err := m.refreshRuleHandles(m.chainPrerouting, true); err != nil {
-		log.Errorf("failed to refresh rule handles prerouting chain: %v", err)
-	}
-
-	return nil
-}
-
-func (m *AclManager) addIOFiltering(
-	ip net.IP,
-	proto firewall.Protocol,
-	sPort *firewall.Port,
-	dPort *firewall.Port,
-	action firewall.Action,
-	ipset *nftables.Set,
-) (*Rule, error) {
-	ruleId := generatePeerRuleId(ip, proto, sPort, dPort, action, ipset)
-	if r, ok := m.rules[ruleId]; ok {
-		return &Rule{
-			nftRule:    r.nftRule,
-			mangleRule: r.mangleRule,
-			nftSet:     r.nftSet,
-			ruleID:     r.ruleID,
-			ip:         ip,
-		}, nil
-	}
-
-	var expressions []expr.Any
-
-	if proto != firewall.ProtocolALL {
-		expressions = append(expressions, &expr.Payload{
-			DestRegister: 1,
-			Base:         expr.PayloadBaseNetworkHeader,
-			Offset:       m.af.protoOffset,
-			Len:          uint32(1),
-		})
-
-		protoData, err := m.af.protoNum(proto)
-		if err != nil {
-			return nil, fmt.Errorf("convert protocol to number: %v", err)
-		}
-
-		expressions = append(expressions, &expr.Cmp{
-			Register: 1,
-			Op:       expr.CmpOpEq,
-			Data:     []byte{protoData},
-		})
-	}
-
-	rawIP := ipToBytes(ip, m.af)
-	// check if rawIP contains zeroed IPv4 0.0.0.0 value
-	// in that case not add IP match expression into the rule definition
-	if slices.ContainsFunc(rawIP, func(v byte) bool { return v != 0 }) {
-		expressions = append(expressions,
-			&expr.Payload{
-				DestRegister: 1,
-				Base:         expr.PayloadBaseNetworkHeader,
-				Offset:       m.af.srcAddrOffset,
-				Len:          m.af.addrLen,
-			},
-		)
-		// add individual IP for match if no ipset defined
-		if ipset == nil {
-			expressions = append(expressions,
-				&expr.Cmp{
-					Op:       expr.CmpOpEq,
-					Register: 1,
-					Data:     rawIP,
-				},
-			)
-		} else {
-			expressions = append(expressions,
-				&expr.Lookup{
-					SourceRegister: 1,
-					SetName:        ipset.Name,
-					SetID:          ipset.ID,
-				},
-			)
-		}
-	}
-
-	expressions = append(expressions, applyPort(sPort, true)...)
-	expressions = append(expressions, applyPort(dPort, false)...)
-
-	mainExpressions := slices.Clone(expressions)
-
-	switch action {
-	case firewall.ActionAccept:
-		mainExpressions = append(mainExpressions, &expr.Verdict{Kind: expr.VerdictAccept})
-	case firewall.ActionDrop:
-		mainExpressions = append(mainExpressions, &expr.Verdict{Kind: expr.VerdictDrop})
-	}
-
-	userData := []byte(ruleId)
-
-	chain := m.chainInputRules
-	rule := &nftables.Rule{
-		Table:    m.workTable,
-		Chain:    chain,
-		Exprs:    mainExpressions,
-		UserData: userData,
-	}
-
-	// Insert DROP rules at the beginning, append ACCEPT rules at the end
-	var nftRule *nftables.Rule
-	if action == firewall.ActionDrop {
-		nftRule = m.rConn.InsertRule(rule)
-	} else {
-		nftRule = m.rConn.AddRule(rule)
-	}
-
-	if err := m.rConn.Flush(); err != nil {
-		return nil, fmt.Errorf("flush input rule %s: %v", ruleId, err)
-	}
-
-	ruleStruct := &Rule{
-		nftRule: nftRule,
-		// best effort mangle rule
-		mangleRule: m.createPreroutingRule(expressions, userData),
-		nftSet:     ipset,
-		ruleID:     ruleId,
-		ip:         ip,
-	}
-	m.rules[ruleId] = ruleStruct
-	if ipset != nil {
-		m.ipsetStore.AddReferenceToIpset(ipset.Name)
-	}
-
-	return ruleStruct, nil
-}
-
-func (m *AclManager) createPreroutingRule(expressions []expr.Any, userData []byte) *nftables.Rule {
-	if m.chainPrerouting == nil {
-		log.Warn("prerouting chain is not created")
-		return nil
-	}
-
-	preroutingExprs := slices.Clone(expressions)
-
-	// interface
-	preroutingExprs = append([]expr.Any{
-		&expr.Meta{
-			Key:      expr.MetaKeyIIFNAME,
-			Register: 1,
-		},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     ifname(m.wgIface.Name()),
-		},
-	}, preroutingExprs...)
-
-	// local destination and mark
-	preroutingExprs = append(preroutingExprs,
-		&expr.Fib{
-			Register:       1,
-			ResultADDRTYPE: true,
-			FlagDADDR:      true,
-		},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     binaryutil.NativeEndian.PutUint32(unix.RTN_LOCAL),
-		},
-
-		&expr.Immediate{
-			Register: 1,
-			Data:     binaryutil.NativeEndian.PutUint32(nbnet.PreroutingFwmarkRedirected),
-		},
-		&expr.Meta{
-			Key:            expr.MetaKeyMARK,
-			Register:       1,
-			SourceRegister: true,
-		},
-	)
-
-	nfRule := m.rConn.AddRule(&nftables.Rule{
-		Table:    m.workTable,
-		Chain:    m.chainPrerouting,
-		Exprs:    preroutingExprs,
-		UserData: userData,
-	})
-
-	if err := m.rConn.Flush(); err != nil {
-		log.Errorf("failed to flush mangle rule %s: %v", string(userData), err)
-		return nil
-	}
-
-	return nfRule
-}
-
-func (m *AclManager) createDefaultChains() (err error) {
-	// chainNameInputRules
-	chain := m.createChain(chainNameInputRules)
-	err = m.rConn.Flush()
-	if err != nil {
-		log.Debugf("failed to create chain (%s): %s", chain.Name, err)
-		return fmt.Errorf(flushError, err)
-	}
-	m.chainInputRules = chain
-
-	// netbird-acl-input-filter
-	// type filter hook input priority filter; policy accept;
-	chain = m.createFilterChainWithHook(chainNameInputFilter, nftables.ChainHookInput)
-	m.addJumpRule(chain, m.chainInputRules.Name, expr.MetaKeyIIFNAME) // to netbird-acl-input-rules
-	m.addDropExpressions(chain, expr.MetaKeyIIFNAME)
-	err = m.rConn.Flush()
-	if err != nil {
-		log.Debugf("failed to create chain (%s): %s", chain.Name, err)
-		return err
-	}
-
-	// netbird-acl-forward-filter
-	chainFwFilter := m.createFilterChainWithHook(chainNameForwardFilter, nftables.ChainHookForward)
-	m.addJumpRulesToRtForward(chainFwFilter) // to netbird-rt-fwd
-	m.addDropExpressions(chainFwFilter, expr.MetaKeyIIFNAME)
-
-	err = m.rConn.Flush()
-	if err != nil {
-		log.Debugf("failed to create chain (%s): %s", chainNameForwardFilter, err)
-		return fmt.Errorf(flushError, err)
-	}
-
-	if err := m.allowRedirectedTraffic(chainFwFilter); err != nil {
-		log.Errorf("failed to allow redirected traffic: %s", err)
-	}
-
-	return nil
-}
-
-// Makes redirected traffic originally destined for the host itself (now subject to the forward filter)
-// go through the input filter as well. This will enable e.g. Docker services to keep working by accessing the
-// netbird peer IP.
-func (m *AclManager) allowRedirectedTraffic(chainFwFilter *nftables.Chain) error {
-	// Chain is created by route manager
-	// TODO: move creation to a common place
-	m.chainPrerouting = &nftables.Chain{
-		Name:     chainNameManglePrerouting,
-		Table:    m.workTable,
-		Type:     nftables.ChainTypeFilter,
-		Hooknum:  nftables.ChainHookPrerouting,
-		Priority: nftables.ChainPriorityMangle,
-	}
-
-	m.addFwmarkToForward(chainFwFilter)
-
-	if err := m.rConn.Flush(); err != nil {
-		return fmt.Errorf(flushError, err)
-	}
-
-	return nil
-}
-
-func (m *AclManager) addFwmarkToForward(chainFwFilter *nftables.Chain) {
-	m.rConn.InsertRule(&nftables.Rule{
-		Table: m.workTable,
-		Chain: chainFwFilter,
-		Exprs: []expr.Any{
-			&expr.Meta{
-				Key:      expr.MetaKeyMARK,
-				Register: 1,
-			},
-			&expr.Cmp{
-				Op:       expr.CmpOpEq,
-				Register: 1,
-				Data:     binaryutil.NativeEndian.PutUint32(nbnet.PreroutingFwmarkRedirected),
-			},
-			&expr.Verdict{
-				Kind: expr.VerdictAccept,
-			},
-		},
-	})
-}
-
-func (m *AclManager) addJumpRulesToRtForward(chainFwFilter *nftables.Chain) {
-	expressions := []expr.Any{
-		&expr.Meta{Key: expr.MetaKeyIIFNAME, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     ifname(m.wgIface.Name()),
-		},
-		&expr.Verdict{
-			Kind:  expr.VerdictJump,
-			Chain: m.routingFwChainName,
-		},
-	}
-
-	_ = m.rConn.AddRule(&nftables.Rule{
-		Table: m.workTable,
-		Chain: chainFwFilter,
-		Exprs: expressions,
-	})
-}
-
-func (m *AclManager) createChain(name string) *nftables.Chain {
-	chain := &nftables.Chain{
-		Name:  name,
-		Table: m.workTable,
-	}
-
-	chain = m.rConn.AddChain(chain)
-
-	insertReturnTrafficRule(m.rConn, m.workTable, chain)
-
-	return chain
-}
-
-func (m *AclManager) createFilterChainWithHook(name string, hookNum *nftables.ChainHook) *nftables.Chain {
-	polAccept := nftables.ChainPolicyAccept
-	chain := &nftables.Chain{
-		Name:     name,
-		Table:    m.workTable,
-		Hooknum:  hookNum,
-		Priority: nftables.ChainPriorityFilter,
-		Type:     nftables.ChainTypeFilter,
-		Policy:   &polAccept,
-	}
-
-	return m.rConn.AddChain(chain)
-}
-
-func (m *AclManager) addDropExpressions(chain *nftables.Chain, ifaceKey expr.MetaKey) []expr.Any {
-	expressions := []expr.Any{
-		&expr.Meta{Key: ifaceKey, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     ifname(m.wgIface.Name()),
-		},
-		&expr.Verdict{Kind: expr.VerdictDrop},
-	}
-	_ = m.rConn.AddRule(&nftables.Rule{
-		Table: m.workTable,
-		Chain: chain,
-		Exprs: expressions,
-	})
-	return nil
-}
-
-func (m *AclManager) addJumpRule(chain *nftables.Chain, to string, ifaceKey expr.MetaKey) {
-	expressions := []expr.Any{
-		&expr.Meta{Key: ifaceKey, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     ifname(m.wgIface.Name()),
-		},
-		&expr.Verdict{
-			Kind:  expr.VerdictJump,
-			Chain: to,
-		},
-	}
-
-	_ = m.rConn.AddRule(&nftables.Rule{
-		Table: chain.Table,
-		Chain: chain,
-		Exprs: expressions,
-	})
-}
-
-func (m *AclManager) addIpToSet(ipsetName string, ip net.IP) (*nftables.Set, error) {
-	ipset, err := m.rConn.GetSetByName(m.workTable, ipsetName)
-	rawIP := ipToBytes(ip, m.af)
-	if err != nil {
-		if ipset, err = m.createSet(m.workTable, ipsetName); err != nil {
-			return nil, fmt.Errorf("get set name: %v", err)
-		}
-
-		m.ipsetStore.newIpset(ipset.Name)
-	}
-
-	if m.ipsetStore.IsIpInSet(ipset.Name, ip) {
-		return ipset, nil
-	}
-
-	if err := m.sConn.SetAddElements(ipset, []nftables.SetElement{{Key: rawIP}}); err != nil {
-		return nil, fmt.Errorf("add set element for the first time: %v", err)
-	}
-
-	m.ipsetStore.AddIpToSet(ipset.Name, ip)
-
-	if err := m.sConn.Flush(); err != nil {
-		return nil, fmt.Errorf("flush add elements: %v", err)
-	}
-
-	return ipset, nil
-}
-
-// createSet in given table by name
-func (m *AclManager) createSet(table *nftables.Table, name string) (*nftables.Set, error) {
-	ipset := &nftables.Set{
-		Name:    name,
-		Table:   table,
-		Dynamic: true,
-		KeyType: m.af.setKeyType,
-	}
-
-	if err := m.rConn.AddSet(ipset, nil); err != nil {
-		return nil, fmt.Errorf("create set: %v", err)
-	}
-
-	if err := m.rConn.Flush(); err != nil {
-		return nil, fmt.Errorf("flush created set: %v", err)
-	}
-
-	return ipset, nil
-}
-
-func (m *AclManager) flushWithBackoff() (err error) {
-	backoff := 4
-	backoffTime := 1000 * time.Millisecond
-	for i := 0; ; i++ {
-		err = m.rConn.Flush()
-		if err != nil {
-			log.Debugf("failed to flush nftables: %v", err)
-			if !strings.Contains(err.Error(), "busy") {
-				return
-			}
-			log.Error("failed to flush nftables, retrying...")
-			if i == backoff-1 {
-				return err
-			}
-			time.Sleep(backoffTime)
-			backoffTime *= 2
-			continue
-		}
-		break
-	}
-	return
-}
-
-func (m *AclManager) refreshRuleHandles(chain *nftables.Chain, mangle bool) error {
-	if m.workTable == nil || chain == nil {
-		return nil
-	}
-
-	list, err := m.rConn.GetRules(m.workTable, chain)
-	if err != nil {
-		return err
-	}
-
-	for _, rule := range list {
-		if len(rule.UserData) == 0 {
-			continue
-		}
-		split := bytes.Split(rule.UserData, []byte(" "))
-		r, ok := m.rules[string(split[0])]
-		if ok {
-			if mangle {
-				*r.mangleRule = *rule
-			} else {
-				*r.nftRule = *rule
-			}
-		}
-	}
-
-	return nil
-}
-
-func generatePeerRuleId(ip net.IP, proto firewall.Protocol, sPort *firewall.Port, dPort *firewall.Port, action firewall.Action, ipset *nftables.Set) string {
-	rulesetID := ":" + string(proto) + ":"
-	if sPort != nil {
-		rulesetID += sPort.String()
-	}
-	rulesetID += ":"
-	if dPort != nil {
-		rulesetID += dPort.String()
-	}
-	rulesetID += ":"
-	rulesetID += strconv.Itoa(int(action))
-	if ipset == nil {
-		return "ip:" + ip.String() + rulesetID
-	}
-	return "set:" + ipset.Name + rulesetID
-}
-
-func ifname(n string) []byte {
-	b := make([]byte, 16)
-	copy(b, n+"\x00")
-	return b
-}
-
-
-// ipToBytes converts net.IP to the correct byte length for the address family.
-func ipToBytes(ip net.IP, af addrFamily) []byte {
-	if af.addrLen == 4 {
-		return ip.To4()
-	}
-	return ip.To16()
-}
-