checker-alias/checker/rules_chain.go

package checker

import (
	"context"
	"fmt"

	sdk "git.happydns.org/checker-sdk-go/checker"
)

type chainLoopRule struct{}

func (chainLoopRule) Name() string { return "chain_loop" }
func (chainLoopRule) Description() string {
	return "Detects CNAME/DNAME cycles in the resolution chain."
}

func (chainLoopRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, _ sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	if data.ChainTerminated.Reason != TermLoop {
		return okState(data.Owner, "no loop in the alias chain")
	}
	return []sdk.CheckState{withHint(sdk.CheckState{
		Status:  sdk.StatusCrit,
		Subject: data.ChainTerminated.Subject,
		Message: fmt.Sprintf("chain loops back to %s", data.ChainTerminated.Subject),
	}, "Break the loop by pointing the last CNAME at an A/AAAA-bearing name.")}
}

type chainLengthRule struct{}

func (chainLengthRule) Name() string { return "chain_length" }
func (chainLengthRule) Description() string {
	return "Flags alias chains longer than the configured maximum (most resolvers give up around 8-16 hops)."
}

func (chainLengthRule) Options() sdk.CheckerOptionsDocumentation {
	return sdk.CheckerOptionsDocumentation{
		UserOpts: []sdk.CheckerOptionDocumentation{
			{
				Id:          "maxChainLength",
				Type:        "uint",
				Label:       "Maximum chain length",
				Description: "Above this number of hops the chain is reported as critical.",
				Default:     float64(defaultMaxChainLength),
			},
		},
	}
}

func (chainLengthRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, opts sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	maxLen := sdk.GetIntOption(opts, "maxChainLength", defaultMaxChainLength)
	if data.ChainTerminated.Reason != TermTooLong {
		return okState(data.Owner, fmt.Sprintf("chain has %d hop(s), within limit of %d", len(data.Chain), maxLen))
	}
	return []sdk.CheckState{withHint(sdk.CheckState{
		Status:  sdk.StatusCrit,
		Subject: data.ChainTerminated.Subject,
		Message: fmt.Sprintf("chain exceeds %d hops; many resolvers will give up", maxLen),
	}, "Flatten intermediate CNAMEs so that the chain is at most a few hops long.")}
}

type chainQueryErrorRule struct{}

func (chainQueryErrorRule) Name() string { return "chain_query_error" }
func (chainQueryErrorRule) Description() string {
	return "Flags DNS query failures encountered while walking the alias chain."
}

func (chainQueryErrorRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, _ sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	if data.ChainTerminated.Reason != TermQueryErr {
		return okState(data.Owner, "all chain queries succeeded")
	}
	// A transport failure (connection refused, timeout, network unreachable)
	// means we could not observe the alias, not that the alias is broken. Report
	// it as Unknown so an intermittent reachability glitch does not flap the
	// check into Warn/Crit; only definitive DNS evidence drives those statuses.
	return []sdk.CheckState{withHint(sdk.CheckState{
		Status:  sdk.StatusUnknown,
		Subject: data.ChainTerminated.Subject,
		Message: fmt.Sprintf("CNAME query for %s could not be completed: %s", data.ChainTerminated.Subject, data.ChainTerminated.Detail),
	}, "Check authoritative-server reachability and firewall rules; the alias state could not be determined while queries fail.")}
}

type chainRcodeRule struct{}

func (chainRcodeRule) Name() string { return "chain_rcode" }
func (chainRcodeRule) Description() string {
	return "Flags NXDOMAIN/SERVFAIL/other rcodes encountered mid-chain or on the final target lookup."
}

func (chainRcodeRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, _ sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	var out []sdk.CheckState
	if data.ChainTerminated.Reason == TermRcode {
		out = append(out, withHint(sdk.CheckState{
			Status:  sdk.StatusCrit,
			Subject: data.ChainTerminated.Subject,
			Message: fmt.Sprintf("server answered %s mid-chain", data.ChainTerminated.Rcode),
		}, "Ensure the zone publishes the expected record; NXDOMAIN/SERVFAIL mid-chain breaks the alias."))
	}
	if data.FinalRcode != "" && data.FinalRcode != "NOERROR" {
		out = append(out, withHint(sdk.CheckState{
			Status:  sdk.StatusWarn,
			Subject: data.FinalTarget,
			Message: fmt.Sprintf("final A lookup for %s returned %s", data.FinalTarget, data.FinalRcode),
		}, "Check the upstream zone's A/AAAA publication."))
	}
	if len(out) == 0 {
		return okState(data.Owner, "all chain and final lookups returned NOERROR")
	}
	return out
}

type hopTTLRule struct{}

func (hopTTLRule) Name() string { return "hop_ttl" }
func (hopTTLRule) Description() string {
	return "Flags chain hops whose TTL is below the configured minimum."
}

func (hopTTLRule) Options() sdk.CheckerOptionsDocumentation {
	return sdk.CheckerOptionsDocumentation{
		UserOpts: []sdk.CheckerOptionDocumentation{
			{
				Id:          "minTargetTTL",
				Type:        "uint",
				Label:       "Minimum TTL (seconds)",
				Description: "Hops with a TTL below this threshold are flagged as a warning.",
				Default:     float64(defaultMinTargetTTL),
			},
		},
	}
}

func (hopTTLRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, opts sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	if len(data.Chain) == 0 {
		return skipped("chain is empty")
	}
	minTTL := uint32(sdk.GetIntOption(opts, "minTargetTTL", defaultMinTargetTTL))
	var out []sdk.CheckState
	for _, h := range data.Chain {
		if h.Kind == KindTarget || h.TTL == 0 {
			continue
		}
		if h.TTL < minTTL {
			out = append(out, withHint(sdk.CheckState{
				Status:  sdk.StatusWarn,
				Subject: h.Owner,
				Message: fmt.Sprintf("hop %s → %s has TTL %ds (< %d)", h.Owner, h.Target, h.TTL, minTTL),
			}, "Raise the CNAME TTL to improve cache efficiency (5-15 minutes is a common floor)."))
		}
	}
	if len(out) == 0 {
		return okState(data.Owner, fmt.Sprintf("all chain hops have TTL ≥ %ds", minTTL))
	}
	return out
}

type targetResolvableRule struct{}

func (targetResolvableRule) Name() string { return "target_resolvable" }
func (targetResolvableRule) Description() string {
	return "Verifies that the final target of the alias chain exists in DNS (returns NOERROR, not NXDOMAIN)."
}

func (targetResolvableRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, _ sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	if data.ChainTerminated.Reason != TermOK {
		return skipped("chain did not terminate normally")
	}
	if data.FinalRcode != "NXDOMAIN" {
		return okState(data.FinalTarget, fmt.Sprintf("target %s exists in DNS", data.FinalTarget))
	}
	return []sdk.CheckState{withHint(sdk.CheckState{
		Status:  sdk.StatusCrit,
		Subject: data.FinalTarget,
		Message: fmt.Sprintf("final target %s does not exist (NXDOMAIN)", data.FinalTarget),
	}, "The alias points at a name that does not exist; create the missing record or update the alias target.")}
}

type multipleRecordsRule struct{}

func (multipleRecordsRule) Name() string { return "multiple_records" }
func (multipleRecordsRule) Description() string {
	return "Flags owners that carry more than one CNAME/DNAME record; only one is legal per owner."
}

func (multipleRecordsRule) Evaluate(ctx context.Context, obs sdk.ObservationGetter, _ sdk.CheckerOptions) []sdk.CheckState {
	data, errState := loadAlias(ctx, obs)
	if errState != nil {
		return errState
	}
	if !apexKnown(data) {
		return skipped("apex lookup failed")
	}
	seen := map[string]int{}
	for _, h := range data.Chain {
		if h.Kind == KindCNAME || h.Kind == KindDNAME {
			seen[h.Owner]++
		}
	}
	var out []sdk.CheckState
	for owner, n := range seen {
		if n > 1 {
			out = append(out, withHint(sdk.CheckState{
				Status:  sdk.StatusCrit,
				Subject: owner,
				Message: fmt.Sprintf("%s carries %d CNAME/DNAME records in the chain", owner, n),
			}, "Keep a single CNAME per name; remove duplicates at the authoritative zone."))
		}
	}
	if len(out) == 0 {
		return okState(data.Owner, "every chain owner carries a single CNAME/DNAME")
	}
	return out
}