Files
coredns/plugin/cache/handler.go
Nitin Nizhawan 87ccb6f90e plugin/cache: add prefer_positive stale policy (#8378)
* plugin/cache: add prefer_positive stale policy

Add an opt-in serve_stale_policy that prefers an eligible success-cache
answer over denial-cache entries while serve_stale is enabled. Preserve the
existing ncache-first behavior when the policy is absent.

Also classify SOA-backed CNAME NODATA responses in the cache so incomplete
answers cannot be selected as positive stale responses.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 25da81ab-92dd-4663-b480-efd6262090c6
Signed-off-by: Nitin Nizhawan <nnizhawan@microsoft.com>

* plugin/cache: retain last-known-good positive answers

Keep an answering success-cache item reachable when a later NOERROR or
referral response overwrites the visible cache key without answering the
question. This lets prefer_positive survive empty responses, referrals, and
additional-only data while leaving policy-off lookup behavior unchanged.

Return the exact accepted verify refresh item instead of re-reading an
ambiguous cache key, avoiding expired TTL wraparound for uncacheable replies.
Add regression coverage for non-answer refreshes, NODATA, SERVFAIL, NOTIMP,
stale-window expiry, and bounded verify reply shaping.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 25da81ab-92dd-4663-b480-efd6262090c6
Signed-off-by: Nitin Nizhawan <nnizhawan@microsoft.com>

* plugin/cache: validate preferred stale answers

Reject truncated, DNSSEC-expired, mismatched-class, unrelated ANY, and ambiguous CNAME refreshes before replacing a stale last-known-good answer. Precompute answer eligibility when cache items are created so prefer_positive hits avoid repeated CNAME walks.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: 25da81ab-92dd-4663-b480-efd6262090c6
Signed-off-by: Nitin Nizhawan <nnizhawan@microsoft.com>

---------

Signed-off-by: Nitin Nizhawan <nnizhawan@microsoft.com>
Co-authored-by: Nitin Nizhawan <nnizhawan@microsoft.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 25da81ab-92dd-4663-b480-efd6262090c6
2026-08-14 01:18:39 -07:00

284 lines
9.5 KiB
Go

package cache
import (
"context"
"math"
"time"
"github.com/coredns/coredns/plugin"
"github.com/coredns/coredns/plugin/metadata"
"github.com/coredns/coredns/plugin/metrics"
"github.com/coredns/coredns/request"
"github.com/miekg/dns"
)
// ServeDNS implements the plugin.Handler interface.
func (c *Cache) ServeDNS(ctx context.Context, w dns.ResponseWriter, r *dns.Msg) (int, error) {
rc := r.Copy() // We potentially modify r, to prevent other plugins from seeing this (r is a pointer), copy r into rc.
state := request.Request{W: w, Req: rc}
do := state.Do()
cd := r.CheckingDisabled
ad := r.AuthenticatedData
zone := plugin.Zones(c.Zones).Matches(state.Name())
if zone == "" {
return plugin.NextOrFailure(c.Name(), c.Next, ctx, w, rc)
}
now := c.now()
server := metrics.WithServer(ctx)
// On cache refresh, we will just use the DO bit from the incoming query for the refresh since we key our cache
// with the query DO bit. That means two separate cache items for the query DO bit true or false. In the situation
// in which upstream doesn't support DNSSEC, the two cache items will effectively be the same. Regardless, any
// DNSSEC RRs in the response are written to cache with the response.
i := c.getIfNotStale(now, state, server)
if i == nil {
crr := &ResponseWriter{ResponseWriter: w, Cache: c, state: state, server: server, do: do, ad: ad, cd: cd,
nexcept: c.nexcept, pexcept: c.pexcept, wildcardFunc: wildcardFunc(ctx)}
return c.doRefresh(ctx, state, crr)
}
ttl := i.ttl(now)
stale := ttl <= 0
if stale {
// serve stale behavior
if c.verifyStale {
failureRecheck := c.staleRecheck
nowFunc := c.now
trackRefresh := failureRecheck > 0
if !trackRefresh || i.beginRefresh(now, failureRecheck) {
crr := &ResponseWriter{ResponseWriter: w, Cache: c, state: state, server: server, do: do, cd: cd}
if c.verifyStaleTimeout > 0 {
// Background verify: cache the response but do not write to the wire.
// On timeout, we serve the stale entry below and let the goroutine continue.
crr.prefetch = true
}
cw := newVerifyStaleResponseWriter(crr)
if c.verifyStaleTimeout == 0 {
ret, err := c.doRefresh(ctx, state, cw)
if trackRefresh {
i.endRefresh(nowFunc(), failureRecheck, cw.refreshed)
}
if cw.refreshed {
return ret, err
}
} else if served, ret, err := c.verifyWithTimeout(ctx, state, w, cw, r, do, ad, i, failureRecheck, nowFunc); served {
return ret, err
}
}
}
// Adjust the time to get a 0 TTL in the reply built from a stale item.
now = now.Add(time.Duration(ttl) * time.Second)
if !c.verifyStale {
c.tryPrefetch(ctx, i, server, rc, do, cd, now, true)
}
servedStale.WithLabelValues(server, c.zonesMetricLabel, c.viewMetricLabel).Inc()
} else if c.shouldPrefetch(i, now) {
c.tryPrefetch(ctx, i, server, rc, do, cd, now, false)
}
if i.wildcard != "" {
// Set wildcard source record name to metadata
metadata.SetValueFunc(ctx, "zone/wildcard", func() string {
return i.wildcard
})
}
if c.keepttl {
// If keepttl is enabled we fake the current time to the stored
// one so that we always get the original TTL
now = i.stored
}
var resp *dns.Msg
if stale && c.staleTTL > 0 {
staleTTL := uint32(c.staleTTL / time.Second) // #nosec G115 -- configuration parsing bounds this to the DNS TTL range.
resp = i.toMsgWithTTL(r, staleTTL, do, ad)
} else {
resp = i.toMsg(r, now, do, ad)
}
w.WriteMsg(resp)
return dns.RcodeSuccess, nil
}
func wildcardFunc(ctx context.Context) func() string {
return func() string {
// Get wildcard source record name from metadata
if f := metadata.ValueFunc(ctx, "zone/wildcard"); f != nil {
return f()
}
return ""
}
}
// tryPrefetch dispatches a background prefetch for i if one is not already in
// flight. The CAS on i.refreshing ensures at most one prefetch goroutine per
// item, so prefetch load scales with distinct stale keys rather than QPS.
func (c *Cache) tryPrefetch(ctx context.Context, i *item, server string, req *dns.Msg, do, cd bool, now time.Time, stale bool) {
failureRecheck := time.Duration(0)
if stale {
failureRecheck = c.staleRecheck
}
nowFunc := c.now
if !i.beginRefresh(nowFunc(), failureRecheck) {
return
}
cw := newPrefetchResponseWriter(server, req, do, cd, c)
go func() {
refreshed := c.doPrefetch(ctx, cw, i, now, stale)
i.endRefresh(nowFunc(), failureRecheck, refreshed)
}()
}
func (c *Cache) doPrefetch(ctx context.Context, cw *ResponseWriter, i *item, now time.Time, stale bool) bool {
// Use a fresh metadata map to avoid concurrent writes to the original request's metadata.
ctx = metadata.ContextWithMetadata(ctx)
cachePrefetches.WithLabelValues(cw.server, c.zonesMetricLabel, c.viewMetricLabel).Inc()
refreshed := true
if stale {
refreshWriter := newVerifyStaleResponseWriter(cw)
c.doRefresh(ctx, cw.state, refreshWriter)
refreshed = refreshWriter.refreshed
} else {
c.doRefresh(ctx, cw.state, cw)
}
if !refreshed {
return false
}
// When prefetching we loose the item i, and with it the frequency
// that we've gathered sofar. See we copy the frequencies info back
// into the new item that was stored in the cache.
if i1 := c.exists(cw.state.Name(), cw.state.QType(), cw.state.QClass(), cw.do, cw.cd); i1 != nil {
i1.Reset(now, i.Hits())
}
return true
}
func (c *Cache) doRefresh(ctx context.Context, state request.Request, cw dns.ResponseWriter) (int, error) {
return plugin.NextOrFailure(c.Name(), c.Next, ctx, cw, state.Req)
}
// verifyWithTimeout runs the upstream verify in a background goroutine and races it
// against verifyStaleTimeout. If the verify completes within the timeout and the
// response is accepted by verifyStaleResponseWriter, the freshly cached entry is
// served to the client and served is true. Otherwise served is false and the caller
// falls through to serve stale; the goroutine continues to run and any cacheable
// response updates the cache without writing to the detached client connection.
func (c *Cache) verifyWithTimeout(ctx context.Context, state request.Request, w dns.ResponseWriter, cw *verifyStaleResponseWriter, r *dns.Msg, do, ad bool, i *item, failureRecheck time.Duration, now func() time.Time) (served bool, code int, err error) {
type result struct {
code int
err error
}
done := make(chan result, 1)
refreshCtx := metadata.ContextWithMetadata(ctx)
for label, valueFunc := range metadata.ValueFuncs(ctx) {
metadata.SetValueFunc(refreshCtx, label, valueFunc)
}
go func() {
rc, re := c.doRefresh(refreshCtx, state, cw)
if failureRecheck > 0 {
i.endRefresh(now(), failureRecheck, cw.refreshed)
}
done <- result{rc, re}
}()
timer := time.NewTimer(c.verifyStaleTimeout)
defer timer.Stop()
select {
case res := <-done:
if !cw.refreshed {
return false, 0, nil
}
if cw.response == nil {
return true, res.code, res.err
}
response := cw.response
if cw.item != nil {
now := c.now()
if c.keepttl {
now = cw.item.stored
}
response = cw.item.toMsg(r, now, do, ad)
}
if err := w.WriteMsg(response); err != nil {
return true, dns.RcodeServerFailure, err
}
return true, dns.RcodeSuccess, nil
case <-timer.C:
return false, 0, nil
}
}
func (c *Cache) shouldPrefetch(i *item, now time.Time) bool {
if c.prefetch <= 0 {
return false
}
i.Update(c.duration, now)
threshold := int(math.Ceil(float64(c.percentage) / 100 * float64(i.origTTL)))
return i.Hits() >= c.prefetch && i.ttl(now) <= threshold
}
// Name implements the Handler interface.
func (c *Cache) Name() string { return "cache" }
// getIfNotStale returns an item if it exists in the cache and has not expired.
func (c *Cache) getIfNotStale(now time.Time, state request.Request, server string) *item {
k := hash(state.Name(), state.QType(), state.QClass(), state.Do(), state.Req.CheckingDisabled)
cacheRequests.WithLabelValues(server, c.zonesMetricLabel, c.viewMetricLabel).Inc()
if c.preferPositive && c.staleUpTo > 0 {
if i, ok := c.pcache.Get(k); ok {
i = i.answeringItem(state)
if i != nil {
ttl := i.ttl(now)
if ttl > 0 || -ttl < int(c.staleUpTo.Seconds()) {
cacheHits.WithLabelValues(server, Success, c.zonesMetricLabel, c.viewMetricLabel).Inc()
return i
}
}
}
}
if i, ok := c.ncache.Get(k); ok {
ttl := i.ttl(now)
if i.matches(state) && (ttl > 0 || (c.staleUpTo > 0 && -ttl < int(c.staleUpTo.Seconds()))) {
// SERVFAIL is transient; prefer a valid positive cache entry if one
// exists, so a cached SERVFAIL does not shadow a previously good answer.
if i.Rcode == dns.RcodeServerFailure {
if p, pok := c.pcache.Get(k); pok {
pttl := p.ttl(now)
if p.matches(state) && (pttl > 0 || (c.staleUpTo > 0 && -pttl < int(c.staleUpTo.Seconds()))) {
cacheHits.WithLabelValues(server, Success, c.zonesMetricLabel, c.viewMetricLabel).Inc()
return p
}
}
}
cacheHits.WithLabelValues(server, Denial, c.zonesMetricLabel, c.viewMetricLabel).Inc()
return i
}
}
if i, ok := c.pcache.Get(k); ok {
ttl := i.ttl(now)
if i.matches(state) && (ttl > 0 || (c.staleUpTo > 0 && -ttl < int(c.staleUpTo.Seconds()))) {
cacheHits.WithLabelValues(server, Success, c.zonesMetricLabel, c.viewMetricLabel).Inc()
return i
}
}
cacheMisses.WithLabelValues(server, c.zonesMetricLabel, c.viewMetricLabel).Inc()
return nil
}
// exists unconditionally returns an item if it exists in the cache.
func (c *Cache) exists(name string, qtype, qclass uint16, do, cd bool) *item {
k := hash(name, qtype, qclass, do, cd)
if i, ok := c.ncache.Get(k); ok {
return i
}
if i, ok := c.pcache.Get(k); ok {
return i
}
return nil
}