package headscale import ( "encoding/json" "errors" "fmt" "io" "net/netip" "os" "path/filepath" "strconv" "strings" "time" "github.com/rs/zerolog/log" "github.com/samber/lo" "github.com/tailscale/hujson" "go4.org/netipx" "gopkg.in/yaml.v3" "tailscale.com/envknob" "tailscale.com/tailcfg" ) const ( errEmptyPolicy = Error("empty policy") errInvalidAction = Error("invalid action") errInvalidGroup = Error("invalid group") errInvalidTag = Error("invalid tag") errInvalidPortFormat = Error("invalid port format") errWildcardIsNeeded = Error("wildcard as port is required for the protocol") ) const ( Base8 = 8 Base10 = 10 BitSize16 = 16 BitSize32 = 32 BitSize64 = 64 portRangeBegin = 0 portRangeEnd = 65535 expectedTokenItems = 2 ) // For some reason golang.org/x/net/internal/iana is an internal package. const ( protocolICMP = 1 // Internet Control Message protocolIGMP = 2 // Internet Group Management protocolIPv4 = 4 // IPv4 encapsulation protocolTCP = 6 // Transmission Control protocolEGP = 8 // Exterior Gateway Protocol protocolIGP = 9 // any private interior gateway (used by Cisco for their IGRP) protocolUDP = 17 // User Datagram protocolGRE = 47 // Generic Routing Encapsulation protocolESP = 50 // Encap Security Payload protocolAH = 51 // Authentication Header protocolIPv6ICMP = 58 // ICMP for IPv6 protocolSCTP = 132 // Stream Control Transmission Protocol ProtocolFC = 133 // Fibre Channel ) var featureEnableSSH = envknob.RegisterBool("HEADSCALE_EXPERIMENTAL_FEATURE_SSH") // LoadACLPolicy loads the ACL policy from the specify path, and generates the ACL rules. func (h *Headscale) LoadACLPolicy(path string) error { log.Debug(). Str("func", "LoadACLPolicy"). Str("path", path). Msg("Loading ACL policy from path") policyFile, err := os.Open(path) if err != nil { return err } defer policyFile.Close() var policy ACLPolicy policyBytes, err := io.ReadAll(policyFile) if err != nil { return err } switch filepath.Ext(path) { case ".yml", ".yaml": log.Debug(). Str("path", path). Bytes("file", policyBytes). Msg("Loading ACLs from YAML") err := yaml.Unmarshal(policyBytes, &policy) if err != nil { return err } log.Trace(). Interface("policy", policy). Msg("Loaded policy from YAML") default: ast, err := hujson.Parse(policyBytes) if err != nil { return err } ast.Standardize() policyBytes = ast.Pack() err = json.Unmarshal(policyBytes, &policy) if err != nil { return err } } if policy.IsZero() { return errEmptyPolicy } h.aclPolicy = &policy return h.UpdateACLRules() } func (h *Headscale) UpdateACLRules() error { machines, err := h.ListMachines() if err != nil { return err } if h.aclPolicy == nil { return errEmptyPolicy } rules, err := h.aclPolicy.generateFilterRules(machines, h.cfg.OIDC.StripEmaildomain) if err != nil { return err } log.Trace().Interface("ACL", rules).Msg("ACL rules generated") h.aclRules = rules // Precompute a map of which sources can reach each destination, this is // to provide quicker lookup when we calculate the peerlist for the map // response to nodes. aclPeerCacheMap := generateACLPeerCacheMap(rules) h.aclPeerCacheMapRW.Lock() h.aclPeerCacheMap = aclPeerCacheMap h.aclPeerCacheMapRW.Unlock() if featureEnableSSH() { sshRules, err := h.generateSSHRules() if err != nil { return err } log.Trace().Interface("SSH", sshRules).Msg("SSH rules generated") if h.sshPolicy == nil { h.sshPolicy = &tailcfg.SSHPolicy{} } h.sshPolicy.Rules = sshRules } else if h.aclPolicy != nil && len(h.aclPolicy.SSHs) > 0 { log.Info().Msg("SSH ACLs has been defined, but HEADSCALE_EXPERIMENTAL_FEATURE_SSH is not enabled, this is a unstable feature, check docs before activating") } return nil } // generateACLPeerCacheMap takes a list of Tailscale filter rules and generates a map // of which Sources ("*" and IPs) can access destinations. This is to speed up the // process of generating MapResponses when deciding which Peers to inform nodes about. func generateACLPeerCacheMap(rules []tailcfg.FilterRule) map[string][]string { aclCachePeerMap := make(map[string][]string) for _, rule := range rules { for _, srcIP := range rule.SrcIPs { for _, ip := range expandACLPeerAddr(srcIP) { if data, ok := aclCachePeerMap[ip]; ok { for _, dstPort := range rule.DstPorts { data = append(data, dstPort.IP) } aclCachePeerMap[ip] = data } else { dstPortsMap := make([]string, 0) for _, dstPort := range rule.DstPorts { dstPortsMap = append(dstPortsMap, dstPort.IP) } aclCachePeerMap[ip] = dstPortsMap } } } } log.Trace().Interface("ACL Cache Map", aclCachePeerMap).Msg("ACL Peer Cache Map generated") return aclCachePeerMap } // expandACLPeerAddr takes a "tailcfg.FilterRule" "IP" and expands it into // something our cache logic can look up, which is "*" or single IP addresses. // This is probably quite inefficient, but it is a result of // "make it work, then make it fast", and a lot of the ACL stuff does not // work, but people have tried to make it fast. func expandACLPeerAddr(srcIP string) []string { if ip, err := netip.ParseAddr(srcIP); err == nil { return []string{ip.String()} } if cidr, err := netip.ParsePrefix(srcIP); err == nil { addrs := []string{} ipRange := netipx.RangeOfPrefix(cidr) from := ipRange.From() too := ipRange.To() if from == too { return []string{from.String()} } for from != too && from.Less(too) { addrs = append(addrs, from.String()) from = from.Next() } addrs = append(addrs, too.String()) // Add the last IP address in the range return addrs } // probably "*" or other string based "IP" return []string{srcIP} } // generateFilterRules takes a set of machines and an ACLPolicy and generates a // set of Tailscale compatible FilterRules used to allow traffic on clients. func (pol *ACLPolicy) generateFilterRules( machines []Machine, stripEmaildomain bool, ) ([]tailcfg.FilterRule, error) { rules := []tailcfg.FilterRule{} for index, acl := range pol.ACLs { if acl.Action != "accept" { return nil, errInvalidAction } srcIPs := []string{} for srcIndex, src := range acl.Sources { srcs, err := pol.getIPsFromSource(src, machines, stripEmaildomain) if err != nil { log.Error(). Interface("src", src). Int("ACL index", index). Int("Src index", srcIndex). Msgf("Error parsing ACL") return nil, err } srcIPs = append(srcIPs, srcs...) } protocols, needsWildcard, err := parseProtocol(acl.Protocol) if err != nil { log.Error(). Msgf("Error parsing ACL %d. protocol unknown %s", index, acl.Protocol) return nil, err } destPorts := []tailcfg.NetPortRange{} for destIndex, dest := range acl.Destinations { dests, err := pol.getNetPortRangeFromDestination( dest, machines, needsWildcard, stripEmaildomain, ) if err != nil { log.Error(). Interface("dest", dest). Int("ACL index", index). Int("dest index", destIndex). Msgf("Error parsing ACL") return nil, err } destPorts = append(destPorts, dests...) } rules = append(rules, tailcfg.FilterRule{ SrcIPs: srcIPs, DstPorts: destPorts, IPProto: protocols, }) } return rules, nil } func (h *Headscale) generateSSHRules() ([]*tailcfg.SSHRule, error) { rules := []*tailcfg.SSHRule{} if h.aclPolicy == nil { return nil, errEmptyPolicy } machines, err := h.ListMachines() if err != nil { return nil, err } acceptAction := tailcfg.SSHAction{ Message: "", Reject: false, Accept: true, SessionDuration: 0, AllowAgentForwarding: false, HoldAndDelegate: "", AllowLocalPortForwarding: true, } rejectAction := tailcfg.SSHAction{ Message: "", Reject: true, Accept: false, SessionDuration: 0, AllowAgentForwarding: false, HoldAndDelegate: "", AllowLocalPortForwarding: false, } for index, sshACL := range h.aclPolicy.SSHs { action := rejectAction switch sshACL.Action { case "accept": action = acceptAction case "check": checkAction, err := sshCheckAction(sshACL.CheckPeriod) if err != nil { log.Error(). Msgf("Error parsing SSH %d, check action with unparsable duration '%s'", index, sshACL.CheckPeriod) } else { action = *checkAction } default: log.Error(). Msgf("Error parsing SSH %d, unknown action '%s'", index, sshACL.Action) return nil, err } principals := make([]*tailcfg.SSHPrincipal, 0, len(sshACL.Sources)) for innerIndex, rawSrc := range sshACL.Sources { expandedSrcs, err := h.aclPolicy.expandAlias( machines, rawSrc, h.cfg.OIDC.StripEmaildomain, ) if err != nil { log.Error(). Msgf("Error parsing SSH %d, Source %d", index, innerIndex) return nil, err } for _, expandedSrc := range expandedSrcs { principals = append(principals, &tailcfg.SSHPrincipal{ NodeIP: expandedSrc, }) } } userMap := make(map[string]string, len(sshACL.Users)) for _, user := range sshACL.Users { userMap[user] = "=" } rules = append(rules, &tailcfg.SSHRule{ RuleExpires: nil, Principals: principals, SSHUsers: userMap, Action: &action, }) } return rules, nil } func sshCheckAction(duration string) (*tailcfg.SSHAction, error) { sessionLength, err := time.ParseDuration(duration) if err != nil { return nil, err } return &tailcfg.SSHAction{ Message: "", Reject: false, Accept: true, SessionDuration: sessionLength, AllowAgentForwarding: false, HoldAndDelegate: "", AllowLocalPortForwarding: true, }, nil } // getIPsFromSource returns a set of Source IPs that would be associated // with the given src alias. func (pol *ACLPolicy) getIPsFromSource( src string, machines []Machine, stripEmaildomain bool, ) ([]string, error) { return pol.expandAlias(machines, src, stripEmaildomain) } // getNetPortRangeFromDestination returns a set of tailcfg.NetPortRange // which are associated with the dest alias. func (pol *ACLPolicy) getNetPortRangeFromDestination( dest string, machines []Machine, needsWildcard bool, stripEmaildomain bool, ) ([]tailcfg.NetPortRange, error) { var tokens []string log.Trace().Str("destination", dest).Msg("generating policy destination") // Check if there is a IPv4/6:Port combination, IPv6 has more than // three ":". tokens = strings.Split(dest, ":") if len(tokens) < expectedTokenItems || len(tokens) > 3 { port := tokens[len(tokens)-1] maybeIPv6Str := strings.TrimSuffix(dest, ":"+port) log.Trace().Str("maybeIPv6Str", maybeIPv6Str).Msg("") if maybeIPv6, err := netip.ParseAddr(maybeIPv6Str); err != nil && !maybeIPv6.Is6() { log.Trace().Err(err).Msg("trying to parse as IPv6") return nil, fmt.Errorf( "failed to parse destination, tokens %v: %w", tokens, errInvalidPortFormat, ) } else { tokens = []string{maybeIPv6Str, port} } } log.Trace().Strs("tokens", tokens).Msg("generating policy destination") var alias string // We can have here stuff like: // git-server:* // 192.168.1.0/24:22 // fd7a:115c:a1e0::2:22 // fd7a:115c:a1e0::2/128:22 // tag:montreal-webserver:80,443 // tag:api-server:443 // example-host-1:* if len(tokens) == expectedTokenItems { alias = tokens[0] } else { alias = fmt.Sprintf("%s:%s", tokens[0], tokens[1]) } expanded, err := pol.expandAlias( machines, alias, stripEmaildomain, ) if err != nil { return nil, err } ports, err := expandPorts(tokens[len(tokens)-1], needsWildcard) if err != nil { return nil, err } dests := []tailcfg.NetPortRange{} for _, d := range expanded { for _, p := range *ports { pr := tailcfg.NetPortRange{ IP: d, Ports: p, } dests = append(dests, pr) } } return dests, nil } // parseProtocol reads the proto field of the ACL and generates a list of // protocols that will be allowed, following the IANA IP protocol number // https://www.iana.org/assignments/protocol-numbers/protocol-numbers.xhtml // // If the ACL proto field is empty, it allows ICMPv4, ICMPv6, TCP, and UDP, // as per Tailscale behaviour (see tailcfg.FilterRule). // // Also returns a boolean indicating if the protocol // requires all the destinations to use wildcard as port number (only TCP, // UDP and SCTP support specifying ports). func parseProtocol(protocol string) ([]int, bool, error) { switch protocol { case "": return nil, false, nil case "igmp": return []int{protocolIGMP}, true, nil case "ipv4", "ip-in-ip": return []int{protocolIPv4}, true, nil case "tcp": return []int{protocolTCP}, false, nil case "egp": return []int{protocolEGP}, true, nil case "igp": return []int{protocolIGP}, true, nil case "udp": return []int{protocolUDP}, false, nil case "gre": return []int{protocolGRE}, true, nil case "esp": return []int{protocolESP}, true, nil case "ah": return []int{protocolAH}, true, nil case "sctp": return []int{protocolSCTP}, false, nil case "icmp": return []int{protocolICMP, protocolIPv6ICMP}, true, nil default: protocolNumber, err := strconv.Atoi(protocol) if err != nil { return nil, false, err } needsWildcard := protocolNumber != protocolTCP && protocolNumber != protocolUDP && protocolNumber != protocolSCTP return []int{protocolNumber}, needsWildcard, nil } } // expandalias has an input of either // - a user // - a group // - a tag // - a host // - an ip // - a cidr // and transform these in IPAddresses. func (pol *ACLPolicy) expandAlias( machines Machines, alias string, stripEmailDomain bool, ) ([]string, error) { if alias == "*" { return []string{"*"}, nil } log.Debug(). Str("alias", alias). Msg("Expanding") // if alias is a group if strings.HasPrefix(alias, "group:") { return pol.getIPsFromGroup(alias, machines, stripEmailDomain) } // if alias is a tag if strings.HasPrefix(alias, "tag:") { return pol.getIPsFromTag(alias, machines, stripEmailDomain) } // if alias is a user if ips := pol.getIPsForUser(alias, machines, stripEmailDomain); len(ips) > 0 { return ips, nil } // if alias is an host if h, ok := pol.Hosts[alias]; ok { log.Trace().Str("host", h.String()).Msg("expandAlias got hosts entry") return pol.expandAlias(machines, h.String(), stripEmailDomain) } // if alias is an IP if ip, err := netip.ParseAddr(alias); err == nil { return pol.getIPsFromSingleIP(ip, machines) } // if alias is an IP Prefix (CIDR) if prefix, err := netip.ParsePrefix(alias); err == nil { return pol.getIPsFromIPPrefix(prefix, machines) } log.Warn().Msgf("No IPs found with the alias %v", alias) return []string{}, nil } // excludeCorrectlyTaggedNodes will remove from the list of input nodes the ones // that are correctly tagged since they should not be listed as being in the user // we assume in this function that we only have nodes from 1 user. func excludeCorrectlyTaggedNodes( aclPolicy *ACLPolicy, nodes []Machine, user string, stripEmailDomain bool, ) []Machine { out := []Machine{} tags := []string{} for tag := range aclPolicy.TagOwners { owners, _ := getTagOwners(aclPolicy, user, stripEmailDomain) ns := append(owners, user) if contains(ns, user) { tags = append(tags, tag) } } // for each machine if tag is in tags list, don't append it. for _, machine := range nodes { hi := machine.GetHostInfo() found := false for _, t := range hi.RequestTags { if contains(tags, t) { found = true break } } if len(machine.ForcedTags) > 0 { found = true } if !found { out = append(out, machine) } } return out } func expandPorts(portsStr string, needsWildcard bool) (*[]tailcfg.PortRange, error) { if portsStr == "*" { return &[]tailcfg.PortRange{ {First: portRangeBegin, Last: portRangeEnd}, }, nil } if needsWildcard { return nil, errWildcardIsNeeded } ports := []tailcfg.PortRange{} for _, portStr := range strings.Split(portsStr, ",") { log.Trace().Msgf("parsing portstring: %s", portStr) rang := strings.Split(portStr, "-") switch len(rang) { case 1: port, err := strconv.ParseUint(rang[0], Base10, BitSize16) if err != nil { return nil, err } ports = append(ports, tailcfg.PortRange{ First: uint16(port), Last: uint16(port), }) case expectedTokenItems: start, err := strconv.ParseUint(rang[0], Base10, BitSize16) if err != nil { return nil, err } last, err := strconv.ParseUint(rang[1], Base10, BitSize16) if err != nil { return nil, err } ports = append(ports, tailcfg.PortRange{ First: uint16(start), Last: uint16(last), }) default: return nil, errInvalidPortFormat } } return &ports, nil } func filterMachinesByUser(machines []Machine, user string) []Machine { out := []Machine{} for _, machine := range machines { if machine.User.Name == user { out = append(out, machine) } } return out } // getTagOwners will return a list of user. An owner can be either a user or a group // a group cannot be composed of groups. func getTagOwners( pol *ACLPolicy, tag string, stripEmailDomain bool, ) ([]string, error) { var owners []string ows, ok := pol.TagOwners[tag] if !ok { return []string{}, fmt.Errorf( "%w. %v isn't owned by a TagOwner. Please add one first. https://tailscale.com/kb/1018/acls/#tag-owners", errInvalidTag, tag, ) } for _, owner := range ows { if strings.HasPrefix(owner, "group:") { gs, err := pol.getUsersInGroup(owner, stripEmailDomain) if err != nil { return []string{}, err } owners = append(owners, gs...) } else { owners = append(owners, owner) } } return owners, nil } // getUsersInGroup will return the list of user inside the group // after some validation. func (pol *ACLPolicy) getUsersInGroup( group string, stripEmailDomain bool, ) ([]string, error) { users := []string{} log.Trace().Caller().Interface("pol", pol).Msg("test") aclGroups, ok := pol.Groups[group] if !ok { return []string{}, fmt.Errorf( "group %v isn't registered. %w", group, errInvalidGroup, ) } for _, group := range aclGroups { if strings.HasPrefix(group, "group:") { return []string{}, fmt.Errorf( "%w. A group cannot be composed of groups. https://tailscale.com/kb/1018/acls/#groups", errInvalidGroup, ) } grp, err := NormalizeToFQDNRules(group, stripEmailDomain) if err != nil { return []string{}, fmt.Errorf( "failed to normalize group %q, err: %w", group, errInvalidGroup, ) } users = append(users, grp) } return users, nil } func (pol *ACLPolicy) getIPsFromGroup( group string, machines Machines, stripEmailDomain bool, ) ([]string, error) { ips := []string{} users, err := pol.getUsersInGroup(group, stripEmailDomain) if err != nil { return ips, err } for _, n := range users { nodes := filterMachinesByUser(machines, n) for _, node := range nodes { ips = append(ips, node.IPAddresses.ToStringSlice()...) } } return ips, nil } func (pol *ACLPolicy) getIPsFromTag( alias string, machines Machines, stripEmailDomain bool, ) ([]string, error) { ips := []string{} // check for forced tags for _, machine := range machines { if contains(machine.ForcedTags, alias) { ips = append(ips, machine.IPAddresses.ToStringSlice()...) } } // find tag owners owners, err := getTagOwners(pol, alias, stripEmailDomain) if err != nil { if errors.Is(err, errInvalidTag) { if len(ips) == 0 { return ips, fmt.Errorf( "%w. %v isn't owned by a TagOwner and no forced tags are defined", errInvalidTag, alias, ) } return ips, nil } else { return ips, err } } // filter out machines per tag owner for _, user := range owners { machines := filterMachinesByUser(machines, user) for _, machine := range machines { hi := machine.GetHostInfo() if contains(hi.RequestTags, alias) { ips = append(ips, machine.IPAddresses.ToStringSlice()...) } } } return ips, nil } func (pol *ACLPolicy) getIPsForUser( user string, machines Machines, stripEmailDomain bool, ) []string { ips := []string{} nodes := filterMachinesByUser(machines, user) nodes = excludeCorrectlyTaggedNodes(pol, nodes, user, stripEmailDomain) for _, n := range nodes { ips = append(ips, n.IPAddresses.ToStringSlice()...) } return ips } func (pol *ACLPolicy) getIPsFromSingleIP( ip netip.Addr, machines Machines, ) ([]string, error) { log.Trace().Str("ip", ip.String()).Msg("expandAlias got ip") ips := []string{ip.String()} matches := machines.FilterByIP(ip) for _, machine := range matches { ips = append(ips, machine.IPAddresses.ToStringSlice()...) } return lo.Uniq(ips), nil } func (pol *ACLPolicy) getIPsFromIPPrefix( prefix netip.Prefix, machines Machines, ) ([]string, error) { log.Trace().Str("prefix", prefix.String()).Msg("expandAlias got prefix") val := []string{prefix.String()} // This is suboptimal and quite expensive, but if we only add the prefix, we will miss all the relevant IPv6 // addresses for the hosts that belong to tailscale. This doesnt really affect stuff like subnet routers. for _, machine := range machines { for _, ip := range machine.IPAddresses { // log.Trace(). // Msgf("checking if machine ip (%s) is part of prefix (%s): %v, is single ip prefix (%v), addr: %s", ip.String(), prefix.String(), prefix.Contains(ip), prefix.IsSingleIP(), prefix.Addr().String()) if prefix.Contains(ip) { val = append(val, machine.IPAddresses.ToStringSlice()...) } } } return lo.Uniq(val), nil }