mesh-controller/internal/link/enrolment.go

package link

import (
	"context"
	"crypto/ed25519"
	"crypto/rand"
	"encoding/base64"
	"errors"
	"fmt"
	"sort"

	"github.com/novox/mesh-control/internal/broker"
	"github.com/novox/mesh-control/internal/identity"
	"github.com/novox/mesh-control/internal/inventory"
)

// Enrolment is what actually happens when a node presents a token: the token is spent, the key is
// recorded, and the node gets its own queue.
//
// It reaches across two contexts and reads neither one's store from the other (novox/hq ADR 0008)
// — it holds both grants and asks each for its part, which is what the process running them is
// for.
type Enrolment struct {
	Inventory  *inventory.Inventory
	Identity   *identity.Identity
	Management *broker.Management
	Broker     broker.Broker
}

// Enrol spends the token and records what the node presented.
//
// Order matters and it is the order things become irreversible. The token is spent first, in a
// single statement that both finds and marks it, so two machines racing on one secret produce one
// winner. Only then is a key recorded — because recording a key for a node whose token turned out
// to be spent would leave the mesh believing a machine that never had the right to join.
func (e Enrolment) Enrol(ctx context.Context, request EnrolRequest) (EnrolReply, error) {
	secret, public, profile := request.Secret, ed25519.PublicKey(request.PublicKey), request.Profile

	if len(public) != ed25519.PublicKeySize {
		return EnrolReply{}, fmt.Errorf("a node presented a %d-byte key, and an identity is %d",
			len(public), ed25519.PublicKeySize)
	}

	node, err := e.Inventory.Redeem(ctx, secret)
	if err != nil {
		return EnrolReply{}, err
	}

	// From here the token is gone whatever happens next, so anything that fails leaves a node
	// record with no live key — which is visible and fixable with a new token, where a spent
	// token believed to be unspent is neither.
	if _, err := e.Identity.RecordNodeKey(ctx, node.ID, public); err != nil {
		return EnrolReply{}, fmt.Errorf(
			"the token was spent and the key could not be recorded, so %s has no identity and "+
				"needs a new token: %w", node.Name, err)
	}

	key, err := e.Identity.Active(ctx)
	if err != nil {
		return EnrolReply{}, err
	}

	reply := EnrolReply{
		Accepted:    true,
		Node:        node.Name,
		Queue:       QueueFor(node.Name),
		Broker:      e.Broker.Address,
		Fingerprint: e.Broker.Fingerprint,
		Signer:      key.Public,
	}

	// The token's secret was the broker password up to this moment, which is what let this
	// connection exist at all. It is replaced now, so the one-time thing stays one-time and the
	// credential the node keeps for years is not the one that was pasted into a terminal.
	if e.Management != nil {
		password, err := freshPassword()
		if err != nil {
			return EnrolReply{}, err
		}
		if err := e.Management.CreateNodeAccount(ctx, node.Name, password); err != nil {
			return EnrolReply{}, fmt.Errorf(
				"the token was spent and %s's broker password could not be replaced: %w",
				node.Name, err)
		}
		reply.Password = password
	}

	// Recorded before the profile because the overlay is the first declaration this node will
	// receive, and without this key the mesh cannot compose one. A node enrolled with no overlay
	// key is a node the graph skips — an ordinary in-between state, and one worth leaving as
	// briefly as possible.
	// And the key its secrets are sealed to. Same reasoning as the overlay key below and one step
	// stronger: without it the mesh cannot send this node a credential at all, and a node that
	// enrolled without one will be refused a sealed file rather than quietly given none.
	// And the key it serves TLS with, so the mesh can certify its internal name. Public, so it is
	// recorded rather than sealed — the node keeps the half that matters.
	if request.ServingKey != "" {
		if err := e.Identity.RecordServingKey(ctx, node.ID, request.ServingKey); err != nil {
			return EnrolReply{}, fmt.Errorf(
				"the token was spent and %s's serving key could not be recorded: %w",
				node.Name, err)
		}
	}
	if request.SealingKey != "" {
		if err := e.Inventory.RecordSealingKey(ctx, node.ID, request.SealingKey); err != nil {
			return EnrolReply{}, fmt.Errorf(
				"the token was spent and %s's sealing key could not be recorded: %w",
				node.Name, err)
		}
	}
	if request.OverlayKey != "" {
		if err := e.Inventory.RecordOverlayKey(ctx, node.ID, request.OverlayKey); err != nil {
			return EnrolReply{}, fmt.Errorf(
				"the token was spent and %s's overlay key could not be recorded: %w", node.Name, err)
		}
	}

	if profile != nil {
		// Not fatal if it fails. The profile is what the control plane needs in order to decide
		// what this machine should run, and it is reported again on every connection — so losing
		// it here costs a decision that can be made later, not the enrolment.
		_ = e.Inventory.RecordProfile(ctx, node.ID, profile)
	}
	return reply, nil
}

// freshPassword is the node's own broker credential from enrolment onward.
func freshPassword() (string, error) {
	raw := make([]byte, 32)
	if _, err := rand.Read(raw); err != nil {
		return "", fmt.Errorf("cannot generate a broker password: %w", err)
	}
	return base64.RawURLEncoding.EncodeToString(raw), nil
}

var _ Enroller = Enrolment{}

// ErrNoBrokerManagement is returned when an account cannot be made because nothing was configured.
var ErrNoBrokerManagement = errors.New("no broker management configured")

// Heard records what a node reported about itself.
//
// A node states; the owning context writes (novox/hq ADR 0006). What a node says it applied is
// its own account of its own machine, kept as a copy for recovery — so this writes it down and
// decides nothing from it.
func (e Enrolment) Heard(ctx context.Context, report Report) error {
	if report.Node == "" {
		return errors.New("a report named no node")
	}
	node, err := e.Inventory.NodeByName(ctx, report.Node)
	if err != nil {
		return err
	}
	// What it did is kept whichever way it went. Until this, a refusal or a failure moved
	// last_seen and the reason went to a log line, so "which machine is not doing what it was
	// told" had no answer the next morning — which is the question a mesh exists to answer.
	doing := inventory.Doing{
		Outcome: inventory.OutcomeApplied,
		Refused: report.Refused,
		Applied: len(report.Applied),
	}
	switch {
	case report.Refused != "":
		doing.Outcome = inventory.OutcomeRefused
	case len(report.Failed) > 0:
		doing.Outcome = inventory.OutcomeFailed
	}
	for id, why := range report.Failed {
		doing.Failed = append(doing.Failed, inventory.FailedResource{ID: id, Error: why})
	}
	// Ordered, so two readings of one failure are the same reading.
	sort.Slice(doing.Failed, func(i, j int) bool { return doing.Failed[i].ID < doing.Failed[j].ID })
	if err := e.Inventory.RecordDoing(ctx, node.ID, doing); err != nil {
		return err
	}

	// A refusal, a failure, or a bare word that the node is there — none of them is an account of
	// what the machine holds, so each moves last_seen and nothing else. Recording a partial list
	// as though it were the whole would tell a rebuilding node to remove what it still has.
	if report.Refused != "" || len(report.Failed) > 0 || report.Applied == nil {
		return e.Inventory.Seen(ctx, node.ID)
	}
	return e.Inventory.RecordOwned(ctx, node.ID, report.Applied)
}