| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/ed25519" |
| 6 | "database/sql" |
| 7 | "encoding/json" |
| 8 | "errors" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/carlosframework/platform/internal/hostbox" |
| 12 | "github.com/carlosframework/platform/internal/hostguard" |
| 13 | ) |
| 14 | |
| 15 | var ErrClaimStopControllerAlive = errors.New("claim stop controller is still alive") |
| 16 | |
| 17 | type stopRecoveryKernel interface { |
| 18 | RecoverStopped(context.Context, hostguard.FenceTarget, StopInventory) (StopKernelEvidence, DrainAcknowledgement, error) |
| 19 | } |
| 20 | type exactStopInventory interface { |
| 21 | ReadExactStopInventory(context.Context, hostguard.FenceTarget) (StopInventory, error) |
| 22 | } |
| 23 | |
| 24 | func (c ClaimStopCoordinator) verifyUnjournaled(ctx context.Context, target hostguard.FenceTarget, in StopInventory, cause error) (StopKernelEvidence, error) { |
| 25 | kernel, ok := c.Kernel.(interface { |
| 26 | VerifyUnjournaled(context.Context, hostguard.FenceTarget, StopInventory) (StopKernelEvidence, error) |
| 27 | }) |
| 28 | if !ok || c.Guard.Root == nil || !errors.Is(cause, sql.ErrNoRows) { |
| 29 | return StopKernelEvidence{}, cause |
| 30 | } |
| 31 | if err := c.Guard.ResumeFence(ctx, target); err != nil { |
| 32 | return StopKernelEvidence{}, err |
| 33 | } |
| 34 | return kernel.VerifyUnjournaled(ctx, target, in) |
| 35 | } |
| 36 | |
| 37 | func (c ClaimStopCoordinator) recoverKernel(ctx context.Context, kernel stopRecoveryKernel, target hostguard.FenceTarget, in StopInventory) (StopKernelEvidence, DrainAcknowledgement, error) { |
| 38 | evidence, ack, err := kernel.RecoverStopped(ctx, target, in) |
| 39 | if err != nil { |
| 40 | evidence, err = c.verifyUnjournaled(ctx, target, in, err) |
| 41 | if err == nil { |
| 42 | ack = DrainClaimAcknowledgement(target.FenceID, target.ControllerID, "unlaunched", nil, hostbox.ProducerCompletion{}) |
| 43 | } |
| 44 | } |
| 45 | return evidence, ack, err |
| 46 | } |
| 47 | |
| 48 | func (c ClaimStopCoordinator) RecoverStopped(ctx context.Context, target hostguard.FenceTarget) (StopResult, error) { |
| 49 | kernel, ok := c.Kernel.(stopRecoveryKernel) |
| 50 | if !ok || c.Now == nil || c.NewID == nil || len(c.EvidenceSigner) != ed25519.PrivateKeySize || c.RequestLimit != hostbox.RequestLimit || !stopOperationTuple(target.Op, target.Phase, target.ObligationID) { |
| 51 | return StopResult{}, ErrClaimStopUnavailable |
| 52 | } |
| 53 | inventory, ok := c.Inventory.(exactStopInventory) |
| 54 | if !ok { |
| 55 | return StopResult{}, ErrClaimStopInventory |
| 56 | } |
| 57 | // A fence persisted without a roster numeral cannot produce a receipt the |
| 58 | // console accepts; refuse it before any kill (roster binding N1). |
| 59 | numeral, err := stopReceiptNumeral(target) |
| 60 | if err != nil { |
| 61 | return StopResult{}, err |
| 62 | } |
| 63 | if err := c.Guard.ResumeFence(ctx, target); err != nil { |
| 64 | return StopResult{}, err |
| 65 | } |
| 66 | in, err := inventory.ReadExactStopInventory(ctx, target) |
| 67 | if err != nil { |
| 68 | return StopResult{}, err |
| 69 | } |
| 70 | if in.Host != target.Host || in.Incarnation != target.Incarnation || in.InstallationStamp != target.Stamp || in.Box != target.Box || in.Enrollment != target.Enrollment || in.DirectoryRevision != target.DirectoryRevision { |
| 71 | return StopResult{}, ErrClaimStopInventory |
| 72 | } |
| 73 | evidence, ack, err := c.recoverKernel(ctx, kernel, target, in) |
| 74 | if err != nil { |
| 75 | return StopResult{}, err |
| 76 | } |
| 77 | if ack.FenceID != target.FenceID || ack.ControllerID != target.ControllerID || ack.ProcessStart == "" || !stopDigest(ack.Digest) { |
| 78 | return StopResult{}, ErrClaimStopIdentity |
| 79 | } |
| 80 | id := c.NewID() |
| 81 | if !stopComponent(id) { |
| 82 | return StopResult{}, ErrClaimStopUnavailable |
| 83 | } |
| 84 | at := c.Now().UTC() |
| 85 | if at.IsZero() || !stopDigest(evidence.InventorySHA256) || !stopDigest(evidence.LocalJournalSHA256) || len(evidence.CgroupIdentities) == 0 { |
| 86 | return StopResult{}, ErrClaimStopReceipt |
| 87 | } |
| 88 | receipt := StopReceipt{RequestLimitNS: int64(c.RequestLimit), Region: target.Region, ClaimETag: target.ClaimETag, ObservedBootID: evidence.ObservedBootID, V: 1, ID: id, ClaimGen: target.ClaimGen, Host: target.Host, Incarnation: target.Incarnation, InstallationStamp: target.Stamp, WriterSession: target.Session, OperationID: target.Op, Phase: target.Phase, ObligationID: target.ObligationID, Box: target.Box, Enrollment: target.Enrollment, BootID: target.BootID, ControllerID: target.ControllerID, DirectoryRevision: numeral, Binding: target.DirectoryRevision, Kind: "stopped", EvidenceAt: at, FenceID: target.FenceID, InventorySHA256: evidence.InventorySHA256, LocalJournalSHA256: evidence.LocalJournalSHA256, ProducerDrainDigest: ack.Digest, CgroupIdentities: evidence.CgroupIdentities, UnitInvocationIDs: evidence.UnitInvocationIDs} |
| 89 | body, err := json.Marshal(receipt) |
| 90 | if err != nil { |
| 91 | return StopResult{}, err |
| 92 | } |
| 93 | if err := c.Guard.CompleteFence(ctx, target, body, at, func(context.Context, hostguard.FenceTarget, []byte, time.Time) error { return nil }); err != nil { |
| 94 | return StopResult{}, err |
| 95 | } |
| 96 | return StopResult{Target: target, Key: stopEvidenceKey(target.Host, target.Op, target.Box, id), Digest: digestRawStop(body), Bytes: body, EvidenceAt: at}, nil |
| 97 | } |
| 98 | func (h ClaimStopHTTP) RecoverPending(ctx context.Context) error { |
| 99 | targets, err := h.Coordinator.Guard.PendingStopFences(ctx) |
| 100 | if err != nil { |
| 101 | return err |
| 102 | } |
| 103 | for _, target := range targets { |
| 104 | if _, err := h.Coordinator.RecoverStopped(ctx, target); err != nil && !errors.Is(err, ErrClaimStopControllerAlive) { |
| 105 | return err |
| 106 | } |
| 107 | } |
| 108 | return nil |
| 109 | } |
| 110 | |
amadan