Compare commits

..
Author SHA1 Message Date
sneak 2284398c81 rproxy tests: leave room for a busy host in the 408 test
check / check (push) Waiting to run
TestAVisitorTooSlowToSendTheRestIsAnswered408 still used a 3-second
request time limit. Under load, rproxy had not yet started waiting for
the rest of the body when it ran out, and answered 504 instead of 408,
as reported on the issue. It now uses 5 seconds, like the other
time-limit tests in this PR.

Model: opus-5-5
2026-11-19 18:45:49 +00:00
sneak 7002a5bfb0 rproxy tests: leave room for a busy host in five timing tests (closes #451)
check / check (push) Waiting to run
Three tests that wait out a 2-second request or response time limit now
use 5 seconds, so the request reaches the app before it runs out under
load. The WebSocket one also renews its connection's deadline after the
wait.

The two breaker tests keep sending requests until one is kept from
asking waf, instead of expecting the sixth to be. On a busy host the five
asks can spread over more than five seconds, or the sixth can arrive
after the open breaker's one-second wait. Their stand-ins for waf now
fail every ask, so a request that asks never closes the breaker.

Model: opus-5-5
2026-11-19 18:34:21 +00:00
clawbot 11ac8aee50 config: export the setting names waf logs (closes #448)
check / check (push) Waiting to run
waf wrote eight setting names as string literals into log lines and
ban notes, and config wrote each again where it reads the setting. A
rename that missed one would log a setting that does not exist. config
now exports a constant for each name, used by its reader and by waf.

Model: opus-5-5
2026-11-19 19:29:49 +01:00
clawbot e7cc0adbdb waf: end the asks to GeoJS, the DNSBL zones, AbuseIPDB and routewatch at the stop (closes #439)
check / check (push) Waiting to run
Each of the four keeps the cancel function of every ask it started, and a
wait group of their goroutines. Its new method (GeoJS.StopAsking,
DNSBL.StopAsking, AbuseIPDB.StopChecking, routewatch Client.StopAsking)
refuses further asks, ends those under way, and returns once their
goroutines have. An ask ended that way is no failure: no alert, no failure
count, no wait before the next. Each provider in internal/sww adds the
method as a stop hook, so the last write of the state files follows it.

Model: opus-5-5
2026-11-19 19:02:12 +01:00
11 changed files with 447 additions and 122 deletions

No files matched your search

+47 -14
View File
@@ -99,8 +99,13 @@ type GeoJS struct {
// waiting are the clients without an answer: those to ask GeoJS about,
// and those it is being asked about.
waiting map[netip.Prefix]*wait
// asking is true while a request to GeoJS is under way.
asking bool
// asking is true while a request to GeoJS is under way, and endAsking
// ends it. stopped is true once StopAsking has been called, and
// askGoroutines waits for the goroutine that makes the requests.
asking bool
endAsking context.CancelFunc
stopped bool
askGoroutines sync.WaitGroup
// retryDelay is how long GeoJS is left alone after its last failure,
// zero after an answer; retryAt is when it may be asked again.
retryDelay time.Duration
@@ -210,6 +215,23 @@ func (g *GeoJS) LookUpWithoutWaiting(
return answer
}
// StopAsking ends the request to GeoJS under way, and returns once the
// goroutine that made it has. A request it ends is no failure: Failed is not
// called, and GeoJS is not left alone after it. After StopAsking, GeoJS is
// asked nothing.
func (g *GeoJS) StopAsking() {
g.mu.Lock()
g.stopped = true
if g.asking {
g.endAsking()
}
g.mu.Unlock()
g.askGoroutines.Wait()
}
// Kept returns client's answer, if one is kept, without asking GeoJS.
func (g *GeoJS) Kept(client netip.Prefix) (Answer, bool) {
g.mu.Lock()
@@ -294,7 +316,8 @@ func (g *GeoJS) answerOrWait(
}
if !g.asking {
// GeoJS is left alone after a failure, so no answer can come.
// GeoJS is left alone after a failure, or asked nothing after
// StopAsking, so no answer can come.
w.late = true
}
@@ -323,22 +346,28 @@ func (g *GeoJS) kept(client netip.Prefix) (Answer, bool) {
}
// ask starts asking GeoJS about the waiting clients, unless a request to
// it is under way or it is left alone after a failure. The requests to
// GeoJS are for every client waiting, so they go on when the caller whose
// ctx is given stops waiting.
// it is under way, it is left alone after a failure, or StopAsking has
// been called. The requests to GeoJS are for every client waiting, so they
// go on when the caller whose ctx is given stops waiting, until StopAsking.
func (g *GeoJS) ask(ctx context.Context) {
if g.asking || g.now().Before(g.retryAt) {
if g.asking || g.stopped || g.now().Before(g.retryAt) {
return
}
g.asking = true
askCtx, end := context.WithCancel(context.WithoutCancel(ctx))
g.asking, g.endAsking = true, end
go g.askAboutWaiting(context.WithoutCancel(ctx))
g.askGoroutines.Go(func() {
defer end()
g.askAboutWaiting(askCtx)
})
}
// askAboutWaiting asks GeoJS about the waiting clients, one request at a
// time, until none is left or GeoJS fails. Each answer kept is given to
// Answered, outside the lock, since Answered takes locks of its own.
// time, until none is left, GeoJS fails, or StopAsking ends ctx. Each
// answer kept is given to Answered, outside the lock, since Answered takes
// locks of its own.
func (g *GeoJS) askAboutWaiting(ctx context.Context) {
for {
clients := g.nextClients()
@@ -347,6 +376,10 @@ func (g *GeoJS) askAboutWaiting(ctx context.Context) {
}
given, err := g.request(ctx, clients)
if err != nil && ctx.Err() != nil {
continue // ended by StopAsking, which is no failure
}
kept, answered := g.keep(clients, given, err)
if g.answered != nil {
@@ -362,13 +395,13 @@ func (g *GeoJS) askAboutWaiting(ctx context.Context) {
}
// nextClients returns up to maxPerRequest of the waiting clients. When
// none is waiting, it returns none and notes that no request to GeoJS is
// under way.
// none is waiting, or StopAsking has been called, it returns none and
// notes that no request to GeoJS is under way.
func (g *GeoJS) nextClients() []netip.Prefix {
g.mu.Lock()
defer g.mu.Unlock()
if len(g.waiting) == 0 {
if len(g.waiting) == 0 || g.stopped {
g.asking = false
return nil
+38
View File
@@ -318,6 +318,44 @@ func TestCallerWhoseContextEndsEarlyDoesNotStopTheRequest(t *testing.T) {
})
}
func TestStopAskingEndsTheRequestUnderWayAsNoFailureAndAsksNoMore(
t *testing.T,
) {
t.Parallel()
synctest.Test(t, func(t *testing.T) {
service, _, g, reports := startWithReports()
clients := newClients()
service.set(hanging)
g.LookUpWithoutWaiting(t.Context(), clients())
waitForRequests(t, service, 1)
// The request is ended, not waited out, and is no failure.
began := time.Now()
g.StopAsking()
if waited := time.Since(began); waited != 0 {
t.Errorf("StopAsking returned after %s, want at once", waited)
}
reports.wantFailures(t)
// A new client is then not asked about, and its lookup does not
// wait.
service.set(answering)
wantCountry(t, g, clients(), "")
if waited := time.Since(began); waited != 0 {
t.Errorf("the lookup returned after %s, want at once", waited)
}
waitForRequests(t, service, 1)
reports.wantFailures(t)
})
}
func TestRequestOver10SecondsIsAFailure(t *testing.T) {
t.Parallel()
+48 -16
View File
@@ -110,8 +110,12 @@ type AbuseIPDB struct {
// scores are by client. Each is added as it is fetched and never moved
// up, so that the one fetched longest ago is the first dropped.
scores *simplelru.LRU[netip.Prefix, Score]
// checking are the clients whose check is under way.
checking map[netip.Prefix]bool
// checking are the clients whose check is under way, each with the
// function that ends it. stopped is true once StopChecking has been
// called, and checkGoroutines waits for the goroutine of each check.
checking map[netip.Prefix]context.CancelFunc
stopped bool
checkGoroutines sync.WaitGroup
// day is the day, in UTC, of the checks spent counts.
day civil.Date
spent int
@@ -134,7 +138,7 @@ func NewAbuseIPDB(params AbuseIPDBParams) *AbuseIPDB {
params: params,
httpClient: &http.Client{},
scores: scores,
checking: map[netip.Prefix]bool{},
checking: map[netip.Prefix]context.CancelFunc{},
}
}
@@ -144,11 +148,11 @@ func NewAbuseIPDB(params AbuseIPDBParams) *AbuseIPDB {
// addresses its request comes from. A client without one is checked in
// the background, by addr, the address its request came from, if
// offender, if it has committed an offence, unless its check is under
// way, a check failed less than failureDelay ago, or the day's checks
// have used up DailyBudget; Hit never waits for a check. The check that
// uses the budget up is logged and raised as a source_failure alert. ctx
// is the context of the client's request, and a check goes on after the
// request ends.
// way, a check failed less than failureDelay ago, the day's checks have
// used up DailyBudget, or StopChecking has been called; Hit never waits
// for a check. The check that uses the budget up is logged and raised as a
// source_failure alert. ctx is the context of the client's request, and a
// check goes on after the request ends, until StopChecking.
func (a *AbuseIPDB) Hit(
ctx context.Context, client netip.Prefix, addr netip.Addr, offender bool,
) (int64, bool) {
@@ -167,14 +171,21 @@ func (a *AbuseIPDB) Hit(
a.day, a.spent = today, 0
}
check := offender && !a.checking[client] && !now.Before(a.retryAt) &&
a.spent < a.params.DailyBudget
_, checking := a.checking[client]
check := offender && !checking && !a.stopped &&
!now.Before(a.retryAt) && a.spent < a.params.DailyBudget
if check {
a.checking[client] = true
checkCtx, end := context.WithCancel(context.WithoutCancel(ctx))
a.checking[client] = end
a.checks++
a.spent++
go a.check(context.WithoutCancel(ctx), client, addr)
a.checkGoroutines.Go(func() {
defer end()
a.check(checkCtx, client, addr)
})
}
usedUp := check && a.spent == a.params.DailyBudget
@@ -189,6 +200,24 @@ func (a *AbuseIPDB) Hit(
return 0, false
}
// StopChecking ends the checks under way, and returns once their goroutines
// have. A check it ends gives no score and is no failure: it is neither
// counted, logged nor raised as an alert, and checks are not left alone
// after it. The ended check stays spent from the day's budget. After
// StopChecking, Hit checks no client.
func (a *AbuseIPDB) StopChecking() {
a.mu.Lock()
a.stopped = true
for _, end := range a.checking {
end()
}
a.mu.Unlock()
a.checkGoroutines.Wait()
}
// Checked returns how many checks were made.
func (a *AbuseIPDB) Checked() int {
a.mu.Lock()
@@ -315,7 +344,8 @@ func (a *AbuseIPDB) RunBlacklist(ctx context.Context) {
// check checks client with AbuseIPDB by addr, one of its addresses, keeps
// the score as client's, and notes the check as no longer under way. A
// check that fails gives no score: it is counted, logged and raised as a
// source_failure alert, and no client is checked for failureDelay.
// source_failure alert, and no client is checked for failureDelay. A check
// ended as ctx ends, by StopChecking, is no failure.
func (a *AbuseIPDB) check(
ctx context.Context,
client netip.Prefix,
@@ -323,24 +353,26 @@ func (a *AbuseIPDB) check(
) {
score, err := a.ask(ctx, addr)
now := a.params.Now()
isFailure := err != nil && ctx.Err() == nil
a.mu.Lock()
delete(a.checking, client)
if err == nil {
switch {
case err == nil:
a.scores.Add(
client,
Score{Client: client, Score: score, Fetched: now},
)
} else {
case isFailure:
a.failures++
a.retryAt = now.Add(failureDelay)
}
a.mu.Unlock()
if err != nil {
if isFailure {
a.alert("checking a client with AbuseIPDB failed", err)
}
}
+37
View File
@@ -394,6 +394,43 @@ func TestCheckNotAnsweredWithinTheTimeoutFailsAndHitNeverWaits(
})
}
func TestStopCheckingEndsTheCheckUnderWayAsNoFailureAndChecksNoMore(
t *testing.T,
) {
t.Parallel()
synctest.Test(t, func(t *testing.T) {
queue := newQueue()
p := abuseIPDBParams()
p.Alerts = queue
abuseIPDB := &abuseIPDBStandIn{hanging: true}
checker := newAbuseIPDB(abuseIPDB, p)
wantScore(t, checker, suspect, true, 0, false)
synctest.Wait()
// The check is ended, not waited out, and is no failure.
began := time.Now()
checker.StopChecking()
if waited := time.Since(began); waited != 0 {
t.Errorf("StopChecking returned after %s, want at once", waited)
}
wantFailures(t, checker, 0)
if got := waiting(queue); len(got) != 0 {
t.Errorf("alerts waiting %+v, want none", got)
}
// An offender without a score is then not checked.
wantScore(t, checker, other, true, 0, false)
synctest.Wait()
wantChecked(t, abuseIPDB, suspect)
})
}
func TestKeyIsSentInTheKeyHeaderAndNeverShown(t *testing.T) {
t.Parallel()
+44 -13
View File
@@ -82,8 +82,12 @@ type DNSBL struct {
// verdicts are by query. Each is added as it is fetched and never moved
// up, so that the one fetched longest ago is the first dropped.
verdicts *simplelru.LRU[query, Verdict]
// asking are the queries under way.
asking map[query]bool
// asking are the queries under way, each with the function that ends
// it. stopped is true once StopAsking has been called, and
// queryGoroutines waits for the goroutine of each query.
asking map[query]context.CancelFunc
stopped bool
queryGoroutines sync.WaitGroup
// queries and failures count, by zone, the queries made and those that
// failed, and retryAt is when a zone whose last query failed may be
// asked again.
@@ -120,7 +124,7 @@ func NewDNSBL(params DNSBLParams) *DNSBL {
params: params,
resolver: resolver,
verdicts: verdicts,
asking: map[query]bool{},
asking: map[query]context.CancelFunc{},
queries: map[string]int{},
failures: map[string]int{},
retryAt: map[string]time.Time{},
@@ -138,9 +142,10 @@ func (d *DNSBL) Zones() []string {
// log, the alerts and the metrics. A verdict is used until CacheTTL has
// passed since it was fetched. Each zone without one is asked about addr
// in the background, unless a query about addr to it is under way, the
// zone is left alone after a failure, or maxQueries are under way;
// ListedBy never waits for a query. ctx is the context of the client's
// request, and a query goes on after the request ends.
// zone is left alone after a failure, maxQueries are under way, or
// StopAsking has been called; ListedBy never waits for a query. ctx is the
// context of the client's request, and a query goes on after the request
// ends, until StopAsking.
func (d *DNSBL) ListedBy(ctx context.Context, addr netip.Addr) []string {
d.mu.Lock()
defer d.mu.Unlock()
@@ -153,23 +158,47 @@ func (d *DNSBL) ListedBy(ctx context.Context, addr netip.Addr) []string {
q := query{zone: zone, client: addr}
kept, found := d.verdicts.Peek(q)
_, asking := d.asking[q]
switch {
case found && now.Sub(kept.Fetched) < d.params.CacheTTL:
if kept.Listed {
listedBy = append(listedBy, config.MaskZoneKey(zone))
}
case !d.asking[q] && !now.Before(d.retryAt[zone]) && len(d.asking) < maxQueries:
d.asking[q] = true
case !asking && !d.stopped && !now.Before(d.retryAt[zone]) &&
len(d.asking) < maxQueries:
queryCtx, end := context.WithCancel(context.WithoutCancel(ctx))
d.asking[q] = end
d.queries[zone]++
go d.ask(context.WithoutCancel(ctx), q)
d.queryGoroutines.Go(func() {
defer end()
d.ask(queryCtx, q)
})
}
}
return listedBy
}
// StopAsking ends the queries under way, and returns once their
// goroutines have. A query it ends gives no verdict and is no failure: it
// is neither counted, logged nor raised as an alert, and the zone is not
// left alone after it. After StopAsking, ListedBy asks no zone.
func (d *DNSBL) StopAsking() {
d.mu.Lock()
d.stopped = true
for _, end := range d.asking {
end()
}
d.mu.Unlock()
d.queryGoroutines.Wait()
}
// Queries returns how many queries were made to zone.
func (d *DNSBL) Queries(zone string) int {
d.mu.Lock()
@@ -238,27 +267,29 @@ func (d *DNSBL) Load(verdicts []Verdict) {
// query as no longer under way. A query that fails gives no verdict: it
// is counted, logged and raised as a source_failure alert, which show the
// zone with its key masked, and the zone is not asked again for
// failureDelay.
// failureDelay. A query ended as ctx ends, by StopAsking, is no failure.
func (d *DNSBL) ask(ctx context.Context, q query) {
listed, err := d.lookUp(ctx, q)
now := d.params.Now()
isFailure := err != nil && ctx.Err() == nil
d.mu.Lock()
delete(d.asking, q)
if err == nil {
switch {
case err == nil:
d.verdicts.Add(q, Verdict{
Zone: q.zone, Client: q.client, Listed: listed, Fetched: now,
})
} else {
case isFailure:
d.failures[q.zone]++
d.retryAt[q.zone] = now.Add(failureDelay)
}
d.mu.Unlock()
if err != nil {
if isFailure {
const failed = "asking a DNSBL zone failed"
shown := config.MaskZoneKey(q.zone)
+39
View File
@@ -210,6 +210,45 @@ func TestQueryNotAnsweredWithinTheTimeoutFails(t *testing.T) {
})
}
//
//nolint:paralleltest // one at a time, as the comment at the top of this file says
func TestStopAskingEndsTheQueryUnderWayAsNoFailureAndAsksNoMore(
t *testing.T,
) {
synctest.Test(t, func(t *testing.T) {
queue := newQueue()
p := dnsblParams(zone)
p.Alerts = queue
resolver := &resolverStandIn{hanging: true}
dnsbl := newDNSBL(resolver, p)
wantZones(t, dnsbl, listed)
synctest.Wait()
// The query is ended, not waited out, and is no failure.
began := time.Now()
dnsbl.StopAsking()
if waited := time.Since(began); waited != 0 {
t.Errorf("StopAsking returned after %s, want at once", waited)
}
wantQueries(t, dnsbl, 1, 0)
if got := waiting(queue); len(got) != 0 {
t.Errorf("alerts waiting %+v, want none", got)
}
// A client without a verdict is then asked about in no zone.
wantZones(t, dnsbl, unlisted)
synctest.Wait()
wantQueries(t, dnsbl, 1, 0)
wantAsked(t, resolver, listedName)
waitForTheResolver()
})
}
//
//nolint:paralleltest // one at a time, as the comment at the top of this file says
func TestZoneThatFailsOrRefusesGivesNoVerdictAndIsLeftAloneForAMinute(
+43 -12
View File
@@ -99,8 +99,12 @@ type Client struct {
// answers are by client. Each is added as it comes and never moved up,
// so that the one that came longest ago is the first dropped.
answers *simplelru.LRU[netip.Prefix, Answer]
// asking are the clients whose query is under way.
asking map[netip.Prefix]bool
// asking are the clients whose query is under way, each with the
// function that ends it. stopped is true once StopAsking has been
// called, and queryGoroutines waits for the goroutine of each query.
asking map[netip.Prefix]context.CancelFunc
stopped bool
queryGoroutines sync.WaitGroup
// retryAt is when routewatch may be asked again after a query failed.
retryAt time.Time
}
@@ -116,7 +120,7 @@ func New(params Params) *Client {
params: params,
httpClient: &http.Client{},
answers: answers,
asking: map[netip.Prefix]bool{},
asking: map[netip.Prefix]context.CancelFunc{},
}
}
@@ -124,8 +128,9 @@ func New(params Params) *Client {
// IPv6 group, or why it has no route. It never waits for a query. A client
// without an answer is asked about in the background, by addr, the address its
// request came from, unless its query is under way, a query failed less than
// failureKept ago, or maxQueries are under way. ctx is the context of the
// client's request, and a query goes on after the request ends.
// failureKept ago, maxQueries are under way, or StopAsking has been called. ctx
// is the context of the client's request, and a query goes on after the request
// ends, until StopAsking.
func (c *Client) Route(
ctx context.Context, client netip.Prefix, addr netip.Addr,
) (Route, Missing) {
@@ -147,15 +152,38 @@ func (c *Client) Route(
return Route{}, RequestFailed
}
if !c.asking[client] && len(c.asking) < maxQueries {
c.asking[client] = true
_, asking := c.asking[client]
if !asking && !c.stopped && len(c.asking) < maxQueries {
queryCtx, end := context.WithCancel(context.WithoutCancel(ctx))
c.asking[client] = end
go c.ask(context.WithoutCancel(ctx), client, addr)
c.queryGoroutines.Go(func() {
defer end()
c.ask(queryCtx, client, addr)
})
}
return Route{}, NoAnswerYet
}
// StopAsking ends the queries under way, and returns once their goroutines
// have. A query it ends gives no answer and is no failure: it is not logged,
// and routewatch is not left alone after it. After StopAsking, Route asks
// routewatch nothing.
func (c *Client) StopAsking() {
c.mu.Lock()
c.stopped = true
for _, end := range c.asking {
end()
}
c.mu.Unlock()
c.queryGoroutines.Wait()
}
// Snapshot returns every answer still in use, sorted by client, as
// routewatch.json lists them.
func (c *Client) Snapshot() []Answer {
@@ -205,7 +233,8 @@ func (c *Client) Load(answers []Answer) {
// ask asks routewatch about addr, keeps the answer as client's, and notes the
// query as no longer under way. A query that fails gives no answer, and
// routewatch is not asked again for failureKept. Only the first failure of
// those under way is logged.
// those under way is logged. A query ended as ctx ends, by StopAsking, is no
// failure.
func (c *Client) ask(
ctx context.Context,
client netip.Prefix,
@@ -213,18 +242,20 @@ func (c *Client) ask(
) {
route, err := c.query(ctx, addr)
now := c.params.Now()
isFailure := err != nil && ctx.Err() == nil
c.mu.Lock()
delete(c.asking, client)
firstFailure := err != nil && !now.Before(c.retryAt)
firstFailure := isFailure && !now.Before(c.retryAt)
if err == nil {
switch {
case err == nil:
c.answers.Add(client, Answer{
Client: client, Netblock: route.Netblock, ASN: route.ASN, Answered: now,
})
} else {
case isFailure:
c.retryAt = now.Add(failureKept)
}
+48
View File
@@ -361,6 +361,54 @@ func TestHangingRoutewatchNeverDelaysAndFailsTheQuery(t *testing.T) {
})
}
func TestStopAskingEndsTheQueryUnderWayAsNoFailureAndAsksNoMore(
t *testing.T,
) {
t.Parallel()
// In a synctest bubble, as TestHangingRoutewatchNeverDelaysAndFailsTheQuery
// says.
synctest.Test(t, func(t *testing.T) {
f := &fixture{t: t, log: &syncBuffer{}}
f.client = routewatch.New(routewatch.Params{
URL: silentURL, Now: time.Now,
ProcessLog: slog.New(slog.NewTextHandler(f.log, nil)),
})
f.route()
synctest.Wait()
// The query is ended, not waited out.
began := time.Now()
f.client.StopAsking()
if waited := time.Since(began); waited != 0 {
t.Errorf("StopAsking returned after %s, want at once", waited)
}
// A query started now would fail within waitLimit. None is, and the
// one ended is no failure: nothing is logged, and routewatch is not
// left alone.
for range 2 {
if _, missing := f.route(); missing != routewatch.NoAnswerYet {
t.Errorf(
"Route gave %q, want %q",
missing,
routewatch.NoAnswerYet,
)
}
time.Sleep(waitLimit)
synctest.Wait()
}
if f.log.String() != "" {
t.Errorf("process log %q, want nothing", f.log.String())
}
})
}
func TestRepeatedRequestsMakeOneQueryAndTheAnswerExpires(t *testing.T) {
t.Parallel()
+13 -4
View File
@@ -870,10 +870,14 @@ func TestAnAnswerRProxyGivesItselfHasTenSecondsToBeSentAtLeast(
func TestAVisitorTooSlowToSendTheRestIsAnswered408(t *testing.T) {
t.Parallel()
// rproxy is waiting for the rest of the body well within this limit, even on
// a busy host, and answers when it runs out, well before dial's deadline.
const limit = 5 * time.Second
standIn := startWAF(t, allow)
app := startApp(t, greet)
addr, _ := startRProxy(t, standIn.URL, app.URL, map[string]string{
maxRequestTime: "3000",
maxRequestTime: strconv.FormatInt(limit.Milliseconds(), 10),
})
conn := dial(t, addr)
@@ -1262,7 +1266,9 @@ func TestAVisitorThatStopsALittleShortIsReadUntilTheRequestTimeLimit(
) {
t.Parallel()
const limit = 2 * time.Second
// The app's answer arrives well within this limit, even on a busy host, and
// the connection closes when it runs out, well before dial's deadline.
const limit = 5 * time.Second
standIn := startWAF(t, allow)
app := startApp(t, answerBeforeTheBody)
@@ -1427,17 +1433,20 @@ func TestTheRequestTimeLimitDoesNotCutAnAnswerOnceTheAppHasTheRequest(
const answer = "the end of a long answer"
// The request reaches the app well within this limit, even on a busy host.
const limit = 5 * time.Second
standIn := startWAF(t, allow)
app := startApp(t, func(w http.ResponseWriter, r *http.Request) {
_, _ = io.Copy(io.Discard, r.Body)
_ = http.NewResponseController(w).Flush()
time.Sleep(2100 * time.Millisecond)
time.Sleep(limit + 100*time.Millisecond)
_, _ = io.WriteString(w, answer)
})
addr, _ := startRProxy(t, standIn.URL, app.URL, map[string]string{
maxRequestTime: "2000",
maxRequestTime: strconv.FormatInt(limit.Milliseconds(), 10),
})
got := visit(
+54 -51
View File
@@ -1412,26 +1412,12 @@ func TestFiveFailedAsksKeepTheNextRequestFromAskingWAF(t *testing.T) {
t.Run(name, func(t *testing.T) {
t.Parallel()
var answering atomic.Bool
standIn := startWAF(
t,
func(w http.ResponseWriter, r *http.Request) {
if answering.Load() {
allow(w, r)
return
}
failure(w, r)
},
)
nextID := make(chan string, 1)
standIn := startWAF(t, failure)
// The app sends back the request ID, so the test knows which
// request rproxy kept from asking.
app := startApp(t, func(w http.ResponseWriter, r *http.Request) {
if answering.Load() {
nextID <- r.Header.Get(requestIDHeader)
}
id := r.Header.Get(requestIDHeader)
w.Header().Set(requestIDHeader, id)
greet(w, r)
})
addr, out := startRProxy(
@@ -1455,19 +1441,35 @@ func TestFiveFailedAsksKeepTheNextRequestFromAskingWAF(t *testing.T) {
t.Error("the circuit breaker opened before five asks failed")
}
answering.Store(true)
wantGreeting(t, get(t, addr, "/"))
out.waitFor(t, "the circuit breaker is open")
// Every ask fails, so a request that asks waf keeps the breaker open
// or opens it. On a busy host, more requests can ask: the five asks
// can take longer than five seconds, and an open breaker lets one
// request ask each second.
deadline := time.Now().Add(waitLimit)
// Had the next request asked waf, its question would have reached
// the stand-in before the request reached the app.
var id string
select {
case id = <-nextID:
default:
t.Fatal("the app did not get the next request")
var got answered
for !out.has("the circuit breaker is open") {
if time.Now().After(deadline) {
t.Fatal(
`rproxy did not write "the circuit breaker is open"`,
)
}
// The stand-in waits while its channel of questions is full, so
// it is emptied first.
for len(standIn.questions) > 0 {
<-standIn.questions
}
got = get(t, addr, "/")
wantGreeting(t, got)
}
// No question the stand-in got since it was last emptied came from
// the request kept from asking.
id := got.header.Get(requestIDHeader)
for len(standIn.questions) > 0 {
asked := <-standIn.questions
if asked.header.Get(requestIDHeader) == id {
@@ -1519,18 +1521,7 @@ func TestVisitorsThatGoAwayWhileWAFIsAskedLeaveTheBreakerClosed(
func TestAnswersThatCannotBePassedOnOpenTheBreaker(t *testing.T) {
t.Parallel()
var answering atomic.Bool
tooLarge := overOneMiB(dispositionDeny)
standIn := startWAF(t, func(w http.ResponseWriter, r *http.Request) {
if answering.Load() {
allow(w, r)
return
}
tooLarge(w, r)
})
standIn := startWAF(t, overOneMiB(dispositionDeny))
app := startApp(t, greet)
addr, out := startRProxy(t, standIn.URL, app.URL, map[string]string{
wafBackendTimeout: "60000",
@@ -1544,15 +1535,20 @@ func TestAnswersThatCannotBePassedOnOpenTheBreaker(t *testing.T) {
standIn.question(t)
}
answering.Store(true)
wantGreeting(t, get(t, addr, "/"))
out.waitFor(t, "the circuit breaker is open")
// A request that asks waf is answered 502, and one kept from asking gets the
// app's greeting. On a busy host, more requests can ask: the five asks can
// take longer than five seconds, and an open breaker lets one request ask
// each second.
deadline := time.Now().Add(waitLimit)
select {
case <-standIn.questions:
t.Error("waf was asked, want it not to be")
case <-time.After(10 * pollInterval):
got := get(t, addr, "/")
for got.status == http.StatusBadGateway && time.Now().Before(deadline) {
standIn.question(t)
got = get(t, addr, "/")
}
wantGreeting(t, got)
out.waitFor(t, "the circuit breaker is open")
}
func TestAPathUnderOwnPrefixNeverReachesTheApp(t *testing.T) {
@@ -1809,11 +1805,15 @@ func TestWaitForReportsWaitsForAReportBeingSent(t *testing.T) {
func TestReportOfAWebSocketCountsWhatItCarried(t *testing.T) {
t.Parallel()
// The app's answer to the upgrade arrives well within these limits, even on a
// busy host.
const limit = 5 * time.Second
standIn := startWAF(t, allow)
app := startApp(t, echoAfterUpgrade)
addr, out := startRProxy(t, standIn.URL, app.URL, map[string]string{
maxRequestTime: "2000",
maxResponseTime: "2000",
maxRequestTime: strconv.FormatInt(limit.Milliseconds(), 10),
maxResponseTime: strconv.FormatInt(limit.Milliseconds(), 10),
})
conn := dial(t, addr)
@@ -1838,7 +1838,10 @@ func TestReportOfAWebSocketCountsWhatItCarried(t *testing.T) {
}
// Past both time limits, the connection still carries what is sent.
time.Sleep(2100 * time.Millisecond)
time.Sleep(limit + 100*time.Millisecond)
// dial's deadline ran from before the upgrade and this wait.
_ = conn.SetDeadline(time.Now().Add(waitLimit))
const message = "still here\n"
+36 -12
View File
@@ -118,7 +118,8 @@ func RunWAF(ctx context.Context, params Params) int {
"settings", cfg)
// The last write follows the stop of every hook, and so the end of every
// loop that changes the state or raises an alert. A hook of the state
// loop that changes the state or raises an alert, and of every ask of
// GeoJS, the DNSBL zones, AbuseIPDB and routewatch. A hook of the state
// files would stop before the loops of the parts built ahead of them.
return run(ctx, app, failed, processLog, func() error {
err := files.WriteAll()
@@ -297,11 +298,13 @@ func newLists(
return lists
}
// newDNSBL returns what asks the DNSBL zones, none asked yet.
// newDNSBL returns what asks the DNSBL zones, none asked yet. At the stop it
// ends the queries under way, as reputation.DNSBL.StopAsking says.
func newDNSBL(
cfg *config.WAF, processLog *slog.Logger, alertQueue *alerts.Queue,
lc fx.Lifecycle, cfg *config.WAF, processLog *slog.Logger,
alertQueue *alerts.Queue,
) *reputation.DNSBL {
return reputation.NewDNSBL(reputation.DNSBLParams{
dnsbl := reputation.NewDNSBL(reputation.DNSBLParams{
Zones: cfg.DNSBLZones,
Resolver: cfg.DNSBLResolver,
CacheTTL: cfg.IPReputationCacheTTL,
@@ -310,12 +313,17 @@ func newDNSBL(
ProcessLog: processLog,
Alerts: alertQueue,
})
lc.Append(fx.StopHook(dnsbl.StopAsking))
return dnsbl
}
// newAbuseIPDB returns what checks clients with AbuseIPDB while
// SWW_WAF_ABUSEIPDB_KEY is set, none checked yet. Until waf stops, it
// fetches AbuseIPDB's blacklist as it is due, and bans the clients on it in
// ledger, as waf.BanAbuseIPDBBlacklist says.
// ledger, as waf.BanAbuseIPDBBlacklist says. At the stop it ends the checks
// under way, as reputation.AbuseIPDB.StopChecking says.
func newAbuseIPDB(
lc fx.Lifecycle, cfg *config.WAF, processLog *slog.Logger,
alertQueue *alerts.Queue, ledger *bans.Ledger,
@@ -341,24 +349,30 @@ func newAbuseIPDB(
})
lc.Append(background(abuseIPDB.RunBlacklist))
lc.Append(fx.StopHook(abuseIPDB.StopChecking))
return abuseIPDB
}
// newRouteWatch returns what asks routewatch about clients, or nil while
// SWW_WAF_ROUTEWATCH_URL is unset.
// SWW_WAF_ROUTEWATCH_URL is unset. At the stop it ends the queries under
// way, as routewatch.Client.StopAsking says.
func newRouteWatch(
cfg *config.WAF, processLog *slog.Logger,
lc fx.Lifecycle, cfg *config.WAF, processLog *slog.Logger,
) *routewatch.Client {
if cfg.RouteWatchURL == "" {
return nil
}
return routewatch.New(routewatch.Params{
routeWatch := routewatch.New(routewatch.Params{
URL: cfg.RouteWatchURL,
Now: utcNow,
ProcessLog: processLog,
})
lc.Append(fx.StopHook(routeWatch.StopAsking))
return routeWatch
}
// newMetrics returns the metrics, with those read from the other parts as
@@ -396,12 +410,18 @@ func newMetrics(
// newGeoJS returns what looks clients up at GeoJS, none looked up yet. Each
// answer goes to the limiter and the ledger, as waf.AddLookup says, and each
// failure is handled as waf.GeoJSFailed says.
// failure is handled as waf.GeoJSFailed says. At the stop it ends the request
// under way, as geojs.GeoJS.StopAsking says.
func newGeoJS(
cfg *config.WAF, m *metrics.Metrics, limiter *ratelimit.Limiter,
ledger *bans.Ledger, alertQueue *alerts.Queue, processLog *slog.Logger,
lc fx.Lifecycle,
cfg *config.WAF,
m *metrics.Metrics,
limiter *ratelimit.Limiter,
ledger *bans.Ledger,
alertQueue *alerts.Queue,
processLog *slog.Logger,
) *geojs.GeoJS {
return geojs.New(geojs.Params{
geoJS := geojs.New(geojs.Params{
URL: geojs.URL,
Answered: func(answer geojs.Answer) {
waf.AddLookup(cfg, limiter, ledger, answer)
@@ -413,6 +433,10 @@ func newGeoJS(
},
WentWithoutAnswer: m.RequestsWithoutGeoJSAnswer.Inc,
})
lc.Append(fx.StopHook(geoJS.StopAsking))
return geoJS
}
// newWAFServer returns the server waf runs, with its parts and the time