Guard: watch logs, ban misbehaving addresses in Housecarl

Files and the journal, patterns with one <IP>, bounties within a window,
bans in a Housecarl ban list with repeat offenders' bans growing; presets
for sshd, jarl, housecarl, steward, bind-xfr, postfix-sasl, dovecot,
nginx-auth and housecarl-drops; guard run/check/test/status/bans/ban/unban;
a Debian package with a sandboxed unit; an end-to-end test with Housecarl.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
abrown38andClaude Opus 5.5 committed 2026-10-22 16:29:37 +01:00
commit d06bf8d531
47 files changed
+4614

No files matched your search

+541
View File
@@ -0,0 +1,541 @@
// Package daemon is Guard at work: each watch's source feeds its lines
// to its patterns; an address that matches often enough is banned in
// Housecarl; bans Housecarl can't take now are sent again later.
package daemon
import (
"context"
"encoding/json"
"errors"
"fmt"
"log/slog"
"math"
"net"
"net/http"
"net/netip"
"os"
"path/filepath"
"sort"
"strings"
"sync"
"sync/atomic"
"time"
"guard/internal/bounty"
"guard/internal/config"
"guard/internal/housecarl"
"guard/internal/match"
"guard/internal/source"
"guard/internal/state"
)
// Banner is what bans go to: Housecarl (a fake in tests).
type Banner interface {
Ban(ctx context.Context, list string, in housecarl.BanInput) (*housecarl.BanResult, error)
BanList(ctx context.Context, list string) (*housecarl.BanList, error)
}
// Daemon is a running Guard.
type Daemon struct {
cfg *config.Config
log *slog.Logger
hc Banner
state *state.State
never *NeverBan
watches []*Watch
started time.Time
mu sync.Mutex
groups map[string]int // ban list → IPv6 grouping
// pending are the bans Housecarl hasn't taken yet, by list and group.
pending map[string]state.Ban
wake chan struct{}
// refused remembers addresses Housecarl refused, so each is logged
// once an hour.
refused map[string]time.Time
// Now and RetryEvery can be changed by tests.
Now func() time.Time
RetryEvery time.Duration
}
// Watch is one watch at work.
type Watch struct {
Cfg *config.Watch
Matcher *match.Matcher
Source source.Source
Tracker *bounty.Tracker
lines, matches, bans atomic.Uint64
}
// New makes a Daemon. hc may be a fake; sources come from the watches'
// configuration unless a test sets them (Watch.Source).
func New(cfg *config.Config, hc Banner, logs source.LogReader, st *state.State, log *slog.Logger) (*Daemon, error) {
d := &Daemon{cfg: cfg, log: log, hc: hc, state: st, never: NewNeverBan(cfg.NeverBan, log), groups: map[string]int{},
pending: map[string]state.Ban{}, wake: make(chan struct{}, 1), refused: map[string]time.Time{},
Now: time.Now, RetryEvery: 30 * time.Second}
for _, b := range st.Pending {
d.pending[b.List+" "+b.Group+" "+b.Protocol+" "+b.Ports] = b
}
for _, w := range cfg.Watch {
if w.Disabled {
continue
}
m, err := match.Compile(w.Match, w.Ignore)
if err != nil {
return nil, fmt.Errorf("watch %s: %v", w.Name, err)
}
wt := &Watch{Cfg: w, Matcher: m, Tracker: bounty.New(w.Hits, w.Within.D())}
switch w.Kind() {
case "file":
wt.Source = &source.File{Pattern: w.File, Log: log.With("watch", w.Name)}
case "journal":
wt.Source = &source.Journal{Units: w.Journal.Unit, Identifiers: w.Journal.Identifier, Log: log.With("watch", w.Name)}
case config.SourceHousecarlDrops:
wt.Source = &source.HousecarlDrops{Client: logs, Protocol: w.Protocol, Shadow: w.Shadow, Log: log.With("watch", w.Name)}
}
d.watches = append(d.watches, wt)
}
if len(d.watches) == 0 {
return nil, errors.New("every watch is disabled: nothing to do")
}
return d, nil
}
// Watches are the watches at work (tests set their sources).
func (d *Daemon) Watches() []*Watch { return d.watches }
// Run watches until ctx ends.
func (d *Daemon) Run(ctx context.Context) error {
d.started = d.Now()
d.loadGroups(ctx)
go d.never.Run(ctx)
var wg sync.WaitGroup
for _, w := range d.watches {
lines := make(chan source.Line, 256)
wg.Add(2)
go func() {
defer wg.Done()
defer close(lines)
for {
err := w.Source.Run(ctx, lines)
if ctx.Err() != nil {
return
}
d.log.Error("the watch's source stopped; starting it again in 10 seconds", "watch", w.Cfg.Name, "err", err)
select {
case <-ctx.Done():
return
case <-time.After(10 * time.Second):
}
}
}()
go func() {
defer wg.Done()
for l := range lines {
d.Line(ctx, w, l)
}
}()
d.log.Info("watching", "watch", w.Cfg.Name, "source", Describe(w.Cfg), "patterns", len(w.Cfg.Match), "hits", w.Cfg.Hits, "within", w.Cfg.Within.String(), "list", w.Cfg.Ban.List)
}
wg.Add(1)
go func() {
defer wg.Done()
d.sender(ctx)
}()
if d.cfg.Socket != "" {
if err := d.serveStatus(ctx); err != nil {
d.log.Warn("no status socket (guard status won't work)", "socket", d.cfg.Socket, "err", err)
}
}
<-ctx.Done()
wg.Wait()
d.savePending()
return nil
}
// Describe says where a watch's lines come from.
func Describe(w *config.Watch) string {
switch w.Kind() {
case "journal":
var parts []string
if len(w.Journal.Unit) > 0 {
parts = append(parts, "unit "+strings.Join(w.Journal.Unit, ", "))
}
if len(w.Journal.Identifier) > 0 {
parts = append(parts, "identifier "+strings.Join(w.Journal.Identifier, ", "))
}
return "journal (" + strings.Join(parts, "; ") + ")"
case config.SourceHousecarlDrops:
return "Housecarl's dropped packets (" + w.Protocol + ")"
}
return "file " + w.File
}
// loadGroups asks Housecarl how each ban list groups IPv6 addresses, so
// bounties are counted as the list bans. A list that can't be read now
// counts by /64, its default.
func (d *Daemon) loadGroups(ctx context.Context) {
for _, w := range d.watches {
list := w.Cfg.Ban.List
d.mu.Lock()
_, done := d.groups[list]
d.mu.Unlock()
if done {
continue
}
c, cancel := context.WithTimeout(ctx, 10*time.Second)
bl, err := d.hc.BanList(c, list)
cancel()
group := 64
switch {
case err != nil:
d.log.Warn("can't read the ban list from Housecarl (yet); counting IPv6 addresses by /64", "list", list, "err", err)
case bl.Disabled:
d.log.Warn("the ban list is disabled in Housecarl: its bans won't be in force", "list", list)
group = bl.V6Group()
default:
group = bl.V6Group()
}
d.mu.Lock()
d.groups[list] = group
d.mu.Unlock()
}
}
func (d *Daemon) group(list string) int {
d.mu.Lock()
defer d.mu.Unlock()
if g, ok := d.groups[list]; ok {
return g
}
return 64
}
// Line handles one line of a watch.
func (d *Daemon) Line(ctx context.Context, w *Watch, l source.Line) {
w.lines.Add(1)
addr := l.Addr
if !addr.IsValid() {
var ok bool
if addr, ok = w.Matcher.Match(l.Text); !ok {
return
}
}
w.matches.Add(1)
if why := d.never.Why(addr); why != "" {
d.log.Debug("match, but never banned", "watch", w.Cfg.Name, "address", addr, "why", why)
return
}
g := match.Group(addr, d.group(w.Cfg.Ban.List))
n, reached := w.Tracker.Add(g, l.Time)
d.log.Debug("match", "watch", w.Cfg.Name, "address", addr, "bounty", n, "hits", w.Cfg.Hits, "line", clip(l.Text, 300))
if !reached {
return
}
w.bans.Add(1)
d.ban(w, addr, g, l.Text)
}
// BanTime is how long a group's next ban lasts: the watch's time,
// times the repeat factor for each earlier ban within forget, up to the
// most.
func (d *Daemon) BanTime(w *config.Watch, group string) time.Duration {
r := d.cfg.Defaults.Repeat
base := w.Ban.Time.D()
n := d.state.Repeats(group, d.Now().Add(-r.Forget.D()))
t := time.Duration(float64(base) * math.Pow(r.Factor, float64(n)))
if t > r.Max.D() || t < 0 {
t = r.Max.D()
}
if t < base {
t = base
}
return t
}
func (d *Daemon) ban(w *Watch, addr netip.Addr, g netip.Prefix, line string) {
group := g.String()
t := d.BanTime(w.Cfg, group)
b := state.Ban{Time: d.Now(), Watch: w.Cfg.Name, List: w.Cfg.Ban.List, Address: addr.String(), Group: group,
Protocol: strings.ToLower(w.Cfg.Ban.Protocol), Ports: strings.Join(w.Cfg.Ban.Port, ","), Seconds: int(t / time.Second), Reason: clip(line, 300)}
d.log.Info("bounty reached; banning", "watch", w.Cfg.Name, "address", addr, "list", b.List, "ports", portsText(b), "for", config.FormatDuration(t))
d.mu.Lock()
d.pending[b.List+" "+b.Group+" "+b.Protocol+" "+b.Ports] = b
d.mu.Unlock()
select {
case d.wake <- struct{}{}:
default:
}
}
func portsText(b state.Ban) string {
if b.Ports == "" {
return "all"
}
if b.Protocol == "" {
return b.Ports
}
return b.Protocol + "/" + b.Ports
}
// sender sends the pending bans to Housecarl: at once, and again every
// RetryEvery while Housecarl can't be reached.
func (d *Daemon) sender(ctx context.Context) {
t := time.NewTicker(d.RetryEvery)
defer t.Stop()
failing := false
for {
select {
case <-ctx.Done():
return
case <-d.wake:
case <-t.C:
}
d.mu.Lock()
var todo []state.Ban
for _, b := range d.pending {
todo = append(todo, b)
}
d.mu.Unlock()
sort.Slice(todo, func(i, j int) bool { return todo[i].Time.Before(todo[j].Time) })
for _, b := range todo {
err := d.send(ctx, b)
if err != nil && !housecarl.Refused(err) {
if !failing {
d.log.Warn("can't reach Housecarl; the bans wait and are sent again", "err", err, "waiting", len(todo))
failing = true
}
break
}
if failing {
d.log.Info("Housecarl is back; sending the bans that waited")
failing = false
}
d.mu.Lock()
if cur, ok := d.pending[b.List+" "+b.Group+" "+b.Protocol+" "+b.Ports]; ok && cur.Time.Equal(b.Time) {
delete(d.pending, b.List+" "+b.Group+" "+b.Protocol+" "+b.Ports)
}
d.mu.Unlock()
}
d.savePending()
}
}
// send makes one ban. A ban Housecarl refuses (never banned, no
// permission) isn't sent again.
func (d *Daemon) send(ctx context.Context, b state.Ban) error {
c, cancel := context.WithTimeout(ctx, 15*time.Second)
defer cancel()
// What waited is sent for the time it has left.
secs := b.Seconds - int(d.Now().Sub(b.Time)/time.Second)
if secs <= 0 {
return nil
}
res, err := d.hc.Ban(c, b.List, housecarl.BanInput{Address: b.Address, Protocol: b.Protocol, Ports: b.Ports, Time: secs,
Source: "guard/" + b.Watch, Reason: b.Reason})
if err != nil {
if housecarl.Refused(err) {
b.Result = err.Error()
d.state.Failed(b)
d.mu.Lock()
last, seen := d.refused[b.Group]
d.refused[b.Group] = d.Now()
d.mu.Unlock()
if !seen || d.Now().Sub(last) > time.Hour {
d.log.Warn("Housecarl refused the ban", "watch", b.Watch, "address", b.Address, "list", b.List, "err", err)
}
}
return err
}
b.Result = "banned"
for _, x := range res.Bans {
if x.Existing {
b.Result = "renewed"
}
}
d.state.Banned(b)
d.log.Info("banned", "watch", b.Watch, "address", b.Address, "list", b.List, "ports", portsText(b), "for", config.FormatDuration(time.Duration(secs)*time.Second),
"result", b.Result, "connections_closed", res.Closed)
return nil
}
func (d *Daemon) savePending() {
d.mu.Lock()
var p []state.Ban
for _, b := range d.pending {
p = append(p, b)
}
d.mu.Unlock()
d.state.SetPending(p)
d.state.Forget(d.Now().Add(-d.cfg.Defaults.Repeat.Forget.D()))
if err := d.state.Save(); err != nil {
d.log.Warn("saving the state failed", "file", d.cfg.State, "err", err)
}
}
func clip(s string, n int) string {
if len(s) > n {
return s[:n]
}
return s
}
// ---- status
// Status is what guard status shows.
type Status struct {
Started time.Time `json:"started"`
Watches []WatchStatus `json:"watches"`
Recent []state.Ban `json:"recent"`
Pending []state.Ban `json:"pending"`
}
// WatchStatus is a watch's.
type WatchStatus struct {
Name string `json:"name"`
Source string `json:"source"`
List string `json:"list"`
Hits int `json:"hits"`
Within string `json:"within"`
Lines uint64 `json:"lines"`
Matches uint64 `json:"matches"`
Bans uint64 `json:"bans"`
Top []bounty.Bounty `json:"top"`
}
// Status reports the watches, their highest bounties and the recent bans.
func (d *Daemon) Status() Status {
s := Status{Started: d.started}
for _, w := range d.watches {
s.Watches = append(s.Watches, WatchStatus{Name: w.Cfg.Name, Source: Describe(w.Cfg), List: w.Cfg.Ban.List, Hits: w.Cfg.Hits,
Within: w.Cfg.Within.String(), Lines: w.lines.Load(), Matches: w.matches.Load(), Bans: w.bans.Load(), Top: w.Tracker.Top(10, d.Now())})
}
s.Recent, s.Pending = d.state.Snapshot()
if len(s.Recent) > 50 {
s.Recent = s.Recent[:50]
}
return s
}
// serveStatus answers GET /status on the local socket (root and the
// guard user only: the socket is 0660 in /run/guard).
func (d *Daemon) serveStatus(ctx context.Context) error {
path := d.cfg.Socket
if err := os.MkdirAll(filepath.Dir(path), 0o750); err != nil {
return err
}
os.Remove(path)
ln, err := net.Listen("unix", path)
if err != nil {
return err
}
os.Chmod(path, 0o660)
mux := http.NewServeMux()
mux.HandleFunc("GET /status", func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
json.NewEncoder(w).Encode(d.Status())
})
srv := &http.Server{Handler: mux, ReadHeaderTimeout: 5 * time.Second}
go srv.Serve(ln)
go func() {
<-ctx.Done()
srv.Close()
os.Remove(path)
}()
return nil
}
// ---- never banned
// NeverBan holds the addresses never banned: prefixes, and host names
// looked up every ten minutes.
type NeverBan struct {
log *slog.Logger
prefixes []netip.Prefix
names []string
mu sync.RWMutex
resolved map[string][]netip.Addr
// Lookup resolves names (tests).
Lookup func(ctx context.Context, host string) ([]netip.Addr, error)
}
// NewNeverBan reads never_ban (checked by config).
func NewNeverBan(list []string, log *slog.Logger) *NeverBan {
n := &NeverBan{log: log, resolved: map[string][]netip.Addr{},
Lookup: func(ctx context.Context, host string) ([]netip.Addr, error) {
return net.DefaultResolver.LookupNetIP(ctx, "ip", host)
}}
for _, e := range list {
if p, err := netip.ParsePrefix(e); err == nil {
n.prefixes = append(n.prefixes, p.Masked())
} else if a, err := netip.ParseAddr(e); err == nil {
n.prefixes = append(n.prefixes, netip.PrefixFrom(a.Unmap(), a.Unmap().BitLen()))
} else {
n.names = append(n.names, e)
}
}
return n
}
// Run looks the names up now and every ten minutes.
func (n *NeverBan) Run(ctx context.Context) {
if len(n.names) == 0 {
return
}
for {
n.resolve(ctx)
select {
case <-ctx.Done():
return
case <-time.After(10 * time.Minute):
}
}
}
func (n *NeverBan) resolve(ctx context.Context) {
for _, h := range n.names {
c, cancel := context.WithTimeout(ctx, 10*time.Second)
addrs, err := n.Lookup(c, h)
cancel()
if err != nil {
// Keep what it was: a name that can't be looked up now is still
// never banned as it was.
if n.log != nil {
n.log.Warn("can't look up a never_ban name; keeping its last addresses", "name", h, "err", err)
}
continue
}
for i := range addrs {
addrs[i] = addrs[i].Unmap()
}
n.mu.Lock()
n.resolved[h] = addrs
n.mu.Unlock()
}
}
// Why says why an address is never banned ("" when it may be).
func (n *NeverBan) Why(a netip.Addr) string {
a = a.Unmap()
switch {
case a.IsLoopback(), a.IsUnspecified(), a.IsMulticast(), a.IsLinkLocalUnicast():
return "loopback, link-local, multicast or unspecified"
}
for _, p := range n.prefixes {
if p.Contains(a) {
return "in never_ban " + p.String()
}
}
n.mu.RLock()
defer n.mu.RUnlock()
for h, addrs := range n.resolved {
for _, x := range addrs {
if x == a {
return "never_ban " + h
}
}
}
return ""
}
+179
View File
@@ -0,0 +1,179 @@
package daemon
import (
"context"
"errors"
"io"
"log/slog"
"path/filepath"
"strings"
"sync"
"testing"
"time"
"guard/internal/config"
"guard/internal/housecarl"
"guard/internal/source"
"guard/internal/state"
)
type fakeHC struct {
mu sync.Mutex
bans []housecarl.BanInput
fail error
}
func (f *fakeHC) Ban(_ context.Context, list string, in housecarl.BanInput) (*housecarl.BanResult, error) {
f.mu.Lock()
defer f.mu.Unlock()
if f.fail != nil {
return nil, f.fail
}
f.bans = append(f.bans, in)
return &housecarl.BanResult{Bans: []housecarl.Ban{{Address: in.Address, List: list}}}, nil
}
func (f *fakeHC) BanList(_ context.Context, list string) (*housecarl.BanList, error) {
return &housecarl.BanList{Name: list, GroupV6: 64}, nil
}
func (f *fakeHC) setFail(err error) {
f.mu.Lock()
f.fail = err
f.mu.Unlock()
}
func (f *fakeHC) count() int {
f.mu.Lock()
defer f.mu.Unlock()
return len(f.bans)
}
func (f *fakeHC) last() housecarl.BanInput {
f.mu.Lock()
defer f.mu.Unlock()
return f.bans[len(f.bans)-1]
}
type chanSource struct{ ch chan source.Line }
func (c *chanSource) Run(ctx context.Context, out chan<- source.Line) error {
for {
select {
case <-ctx.Done():
return nil
case l := <-c.ch:
out <- l
}
}
}
func waitFor(t *testing.T, what string, cond func() bool) {
t.Helper()
for i := 0; i < 200; i++ {
if cond() {
return
}
time.Sleep(10 * time.Millisecond)
}
t.Fatalf("waited in vain: %s", what)
}
// TestDaemon: hits within the window ban, by the watch's ports and time;
// a repeat offender's ban is longer; IPv6 counts by /64; never_ban is
// never banned; bans wait while Housecarl can't be reached, and a
// refused one isn't sent again.
func TestDaemon(t *testing.T) {
dir := t.TempDir()
cfg, err := config.Parse([]byte(`housecarl: {url: https://fw:8070, list: web}
never_ban: [198.51.100.0/24]
defaults: {hits: 3, within: 10m, ban: 1h, repeat: {factor: 2, max: 3h}}
watch:
- {name: ssh, preset: sshd, file: /var/log/auth.log, ban: {protocol: tcp, port: 22}}
state: ` + filepath.Join(dir, "state.json") + `
socket: ` + filepath.Join(dir, "guard.sock") + `
`))
if err != nil {
t.Fatal(err)
}
st, _ := state.Load(cfg.State)
hc := &fakeHC{}
d, err := New(cfg, hc, nil, st, slog.New(slog.NewTextHandler(io.Discard, nil)))
if err != nil {
t.Fatal(err)
}
d.RetryEvery = 50 * time.Millisecond
src := &chanSource{ch: make(chan source.Line)}
d.Watches()[0].Source = src
ctx, cancel := context.WithCancel(context.Background())
done := make(chan struct{})
go func() { d.Run(ctx); close(done) }()
fail := func(addr string, n int) {
for i := 0; i < n; i++ {
src.ch <- source.Line{Text: "Failed password for root from " + addr + " port 4000 ssh2", Time: time.Now()}
}
}
fail("203.0.113.9", 2)
time.Sleep(50 * time.Millisecond)
if hc.count() != 0 {
t.Fatal("banned below the hits")
}
fail("203.0.113.9", 1)
waitFor(t, "the first ban", func() bool { return hc.count() == 1 })
if b := hc.last(); b.Address != "203.0.113.9" || b.Protocol != "tcp" || b.Ports != "22" || b.Time < 3590 || b.Time > 3600 || b.Source != "guard/ssh" ||
!strings.Contains(b.Reason, "Failed password") {
t.Errorf("ban %+v", b)
}
fail("203.0.113.9", 3)
waitFor(t, "the repeat ban", func() bool { return hc.count() == 2 })
if b := hc.last(); b.Time < 7190 || b.Time > 7200 {
t.Errorf("a repeat offender's second ban: %ds, want 2h", b.Time)
}
fail("203.0.113.9", 3)
waitFor(t, "the third ban", func() bool { return hc.count() == 3 })
if b := hc.last(); b.Time < 10790 || b.Time > 10800 {
t.Errorf("capped at max: %ds, want 3h", b.Time)
}
// IPv6, by /64.
for _, a := range []string{"2001:db8:1:2::1", "2001:db8:1:2::2", "2001:db8:1:2::3"} {
fail(a, 1)
}
waitFor(t, "the IPv6 ban", func() bool { return hc.count() == 4 })
// never_ban.
fail("198.51.100.7", 5)
time.Sleep(50 * time.Millisecond)
if hc.count() != 4 {
t.Errorf("never_ban banned: %+v", hc.last())
}
// Housecarl down: the ban waits, then goes.
hc.setFail(errors.New("connection refused"))
fail("203.0.113.20", 3)
time.Sleep(150 * time.Millisecond)
if _, pending := st.Snapshot(); hc.count() != 4 || len(pending) != 1 {
t.Errorf("while Housecarl is down: %d bans, %d pending", hc.count(), len(pending))
}
hc.setFail(nil)
waitFor(t, "the ban that waited", func() bool { return hc.count() == 5 && hc.last().Address == "203.0.113.20" })
// Refused: recorded, not sent again.
hc.setFail(&housecarl.Error{Status: 422, Code: "never_ban", Message: "203.0.113.30 isn't banned: it is inside"})
fail("203.0.113.30", 3)
waitFor(t, "the refusal recorded", func() bool {
recent, _ := st.Snapshot()
return len(recent) > 0 && strings.Contains(recent[0].Result, "inside")
})
time.Sleep(150 * time.Millisecond)
if _, pending := st.Snapshot(); len(pending) != 0 {
t.Errorf("a refused ban waits: %+v", pending)
}
s := d.Status()
if len(s.Watches) != 1 || s.Watches[0].Bans != 6 || s.Watches[0].Matches != 23 {
t.Errorf("status %+v", s.Watches)
}
cancel()
<-done
// The state outlives the daemon.
st2, err := state.Load(cfg.State)
if err != nil || st2.Repeats("203.0.113.9/32", time.Now().Add(-time.Hour)) != 3 {
t.Errorf("saved state: %v %+v", err, st2.Offences)
}
}