| 1 | package hostguard |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "time" |
| 7 | |
| 8 | "github.com/carlosframework/platform/internal/hostclaim" |
| 9 | "github.com/carlosframework/platform/internal/release" |
| 10 | ) |
| 11 | |
| 12 | var ( |
| 13 | ErrInvalidInput = errors.New("hostguard: invalid input") |
| 14 | ErrNoClaims = errors.New("hostguard: no claim reader") |
| 15 | ErrNoClaim = errors.New("hostguard: claim absent") |
| 16 | ErrMalformedClaim = errors.New("hostguard: malformed claim observation") |
| 17 | ErrNoHolder = errors.New("hostguard: claim has no holder") |
| 18 | ErrClaimIncarnation = errors.New("hostguard: captured incarnation differs from claim") |
| 19 | ErrRecordIncarnation = errors.New("hostguard: local incarnation differs") |
| 20 | ErrInactive = errors.New("hostguard: installation inactive") |
| 21 | ErrStampMismatch = errors.New("hostguard: captured stamp is not active") |
| 22 | ErrCheckRequired = errors.New("hostguard: identity check required") |
| 23 | ErrNotParticipant = errors.New("hostguard: box is not authorized to activate") |
| 24 | ErrRetiredStamp = errors.New("hostguard: installation stamp retired") |
| 25 | ErrOldStamp = errors.New("hostguard: installation stamp predates local history") |
| 26 | ErrFenced = errors.New("hostguard: execution denied by durable fence") |
| 27 | ) |
| 28 | |
| 29 | type ClaimReader interface { |
| 30 | Read(context.Context, string) (hostclaim.Observed, error) |
| 31 | } |
| 32 | |
| 33 | type StoreReader struct{ Store release.Store } |
| 34 | |
| 35 | func (r StoreReader) Read(ctx context.Context, host string) (hostclaim.Observed, error) { |
| 36 | if r.Store == nil { |
| 37 | return hostclaim.Observed{}, ErrNoClaims |
| 38 | } |
| 39 | return hostclaim.Read(ctx, r.Store, host) |
| 40 | } |
| 41 | |
| 42 | type Guard struct { |
| 43 | Root *RootAuthority |
| 44 | FenceView FenceViewProvider |
| 45 | DataDir string |
| 46 | LockDir string |
| 47 | Claims ClaimReader |
| 48 | openRecords func(string) (*recordDir, error) |
| 49 | } |
| 50 | |
| 51 | func (g Guard) locked(ctx context.Context, host string, fn func(*recordDir) error) error { |
| 52 | if err := validateFileHost(host); err != nil { |
| 53 | return err |
| 54 | } |
| 55 | if g.DataDir == "" || ctx == nil { |
| 56 | return ErrInvalidInput |
| 57 | } |
| 58 | if g.Root != nil { |
| 59 | if g.Root.store == nil { |
| 60 | return ErrInvalidInput |
| 61 | } |
| 62 | unlock, err := g.Root.store.Lock(ctx, host+".lock") |
| 63 | if err != nil { |
| 64 | return err |
| 65 | } |
| 66 | defer unlock() |
| 67 | } |
| 68 | dir := g.LockDir |
| 69 | if dir == "" { |
| 70 | dir = "/run/carlos/hostguard" |
| 71 | } |
| 72 | lockCtx := ctx |
| 73 | cancel := func() {} |
| 74 | if g.Root != nil { |
| 75 | lockCtx, cancel = context.WithTimeout(ctx, 250*time.Millisecond) |
| 76 | } |
| 77 | defer cancel() |
| 78 | unlock, err := acquire(lockCtx, dir, host, g.Root != nil) |
| 79 | if err != nil { |
| 80 | // Contention and I/O errors are retryable, not hostile evidence. |
| 81 | // Only a positively unsafe lock can quarantine an installation. |
| 82 | if g.Root != nil && errors.Is(err, ErrUnsafeFile) { |
| 83 | r, readErr := g.Root.read(host) |
| 84 | if readErr == nil { |
| 85 | r.quarantined, r.settled = true, false |
| 86 | return errors.Join(err, g.Root.write(r)) |
| 87 | } |
| 88 | if !errors.Is(readErr, ErrNoRecord) { |
| 89 | return errors.Join(err, readErr) |
| 90 | } |
| 91 | } |
| 92 | return err |
| 93 | } |
| 94 | defer unlock() |
| 95 | if g.Root != nil { |
| 96 | return fn(&recordDir{root: g.Root}) |
| 97 | } |
| 98 | open := g.openRecords |
| 99 | if open == nil { |
| 100 | open = openRecordDir |
| 101 | } |
| 102 | d, err := open(g.DataDir) |
| 103 | if err != nil { |
| 104 | return err |
| 105 | } |
| 106 | defer d.close() |
| 107 | return fn(d) |
| 108 | } |
| 109 | |
| 110 | func (g Guard) fresh(ctx context.Context, host string) (*hostclaim.Claim, error) { |
| 111 | obs, err := g.freshObserved(ctx, host) |
| 112 | if err != nil { |
| 113 | return nil, err |
| 114 | } |
| 115 | return obs.Claim, nil |
| 116 | } |
| 117 | |
| 118 | func (g Guard) freshObserved(ctx context.Context, host string) (hostclaim.Observed, error) { |
| 119 | if err := ctx.Err(); err != nil { |
| 120 | return hostclaim.Observed{}, err |
| 121 | } |
| 122 | if g.Claims == nil { |
| 123 | return hostclaim.Observed{}, ErrNoClaims |
| 124 | } |
| 125 | obs, err := g.Claims.Read(ctx, host) |
| 126 | if err != nil { |
| 127 | return hostclaim.Observed{}, err |
| 128 | } |
| 129 | if err := ctx.Err(); err != nil { |
| 130 | return hostclaim.Observed{}, err |
| 131 | } |
| 132 | if obs.Host != host { |
| 133 | return hostclaim.Observed{}, ErrMalformedClaim |
| 134 | } |
| 135 | if obs.Absent { |
| 136 | return hostclaim.Observed{}, ErrNoClaim |
| 137 | } |
| 138 | if obs.Claim == nil || obs.Claim.Host != host { |
| 139 | return hostclaim.Observed{}, ErrMalformedClaim |
| 140 | } |
| 141 | if obs.Claim.Holder == nil { |
| 142 | return hostclaim.Observed{}, ErrNoHolder |
| 143 | } |
| 144 | if obs.Claim.Holder.Incarnation == "" { |
| 145 | return hostclaim.Observed{}, ErrMalformedClaim |
| 146 | } |
| 147 | return obs, nil |
| 148 | } |
| 149 | |
| 150 | func (g Guard) With(ctx context.Context, host string, captured Captured, check func(*hostclaim.Claim) error, fn func() error) error { |
| 151 | return g.locked(ctx, host, func(d *recordDir) error { |
| 152 | c, err := g.fresh(ctx, host) |
| 153 | if err != nil { |
| 154 | return err |
| 155 | } |
| 156 | if (g.Root != nil || g.FenceView != nil) && c.Holder.Stamp != captured.Stamp { |
| 157 | return ErrStampMismatch |
| 158 | } |
| 159 | if g.FenceView != nil { |
| 160 | v, err := g.FenceView(ctx, host, captured) |
| 161 | if err != nil { |
| 162 | return err |
| 163 | } |
| 164 | if err := v.admit(host, captured, c.Holder.Session); err != nil { |
| 165 | return err |
| 166 | } |
| 167 | } |
| 168 | if c.Holder.Incarnation != captured.Incarnation { |
| 169 | return ErrClaimIncarnation |
| 170 | } |
| 171 | if check == nil { |
| 172 | return ErrCheckRequired |
| 173 | } |
| 174 | if err := check(c); err != nil { |
| 175 | return err |
| 176 | } |
| 177 | r, err := d.read(host) |
| 178 | if err != nil { |
| 179 | return err |
| 180 | } |
| 181 | if err := r.admits(c, captured); err != nil { |
| 182 | return err |
| 183 | } |
| 184 | if fn == nil { |
| 185 | return ErrInvalidInput |
| 186 | } |
| 187 | if err := ctx.Err(); err != nil { |
| 188 | return err |
| 189 | } |
| 190 | if g.FenceView != nil { |
| 191 | v, err := g.FenceView(ctx, host, captured) |
| 192 | if err != nil { |
| 193 | return err |
| 194 | } |
| 195 | if err := v.admit(host, captured, c.Holder.Session); err != nil { |
| 196 | return err |
| 197 | } |
| 198 | } |
| 199 | return fn() |
| 200 | }) |
| 201 | } |
| 202 | |
| 203 | // admits is With's reading of the local record: nil only when the record |
| 204 | // lets the captured installation run claim c's session. AdmissionFenced |
| 205 | // reads the same, so the two cannot disagree. |
| 206 | func (r record) admits(c *hostclaim.Claim, captured Captured) error { |
| 207 | if r.quarantined || r.pendingInstallation != nil || r.barrier != nil { |
| 208 | return ErrFenced |
| 209 | } |
| 210 | if r.root && (r.installation.Account != c.Holder.Account || r.installation.App != c.Holder.App) { |
| 211 | return ErrRecordIncarnation |
| 212 | } |
| 213 | if r.Incarnation != captured.Incarnation { |
| 214 | return ErrRecordIncarnation |
| 215 | } |
| 216 | if r.ActiveStamp == 0 { |
| 217 | return ErrInactive |
| 218 | } |
| 219 | if captured.Stamp != r.ActiveStamp { |
| 220 | return ErrStampMismatch |
| 221 | } |
| 222 | if migrationAdmissionDenied(r.migrationDenies, r.FenceJournal, captured, c.Holder.Session) { |
| 223 | return ErrFenced |
| 224 | } |
| 225 | if r.ActiveFence != nil && r.ActiveFence.matchesSession(c, captured) { |
| 226 | return ErrFenced |
| 227 | } |
| 228 | for _, receipt := range r.FenceJournal { |
| 229 | if receipt.Target.matchesSession(c, captured) { |
| 230 | return ErrFenced |
| 231 | } |
| 232 | } |
| 233 | return nil |
| 234 | } |
| 235 | |
| 236 | func (r *record) retire() { |
| 237 | if r.ActiveStamp == 0 { |
| 238 | return |
| 239 | } |
| 240 | if r.root { |
| 241 | r.rootRetired = append(r.rootRetired, Captured{Incarnation: r.Incarnation, Stamp: r.ActiveStamp}) |
| 242 | } |
| 243 | r.RetiredStamps = append(r.RetiredStamps, r.ActiveStamp) |
| 244 | if !r.root && len(r.RetiredStamps) > 32 { |
| 245 | r.RetiredStamps = r.RetiredStamps[len(r.RetiredStamps)-32:] |
| 246 | } |
| 247 | r.ActiveStamp = 0 |
| 248 | } |
| 249 | |
| 250 | func (g Guard) Activate(ctx context.Context, host, box string, participant func(*hostclaim.Claim) bool) error { |
| 251 | if g.Root != nil { |
| 252 | return g.locked(ctx, host, func(d *recordDir) error { |
| 253 | c, err := g.fresh(ctx, host) |
| 254 | if err != nil { |
| 255 | return err |
| 256 | } |
| 257 | r, err := d.read(host) |
| 258 | if err != nil { |
| 259 | return err |
| 260 | } |
| 261 | if r.quarantined || r.pendingInstallation != nil || r.barrier != nil || r.Incarnation != c.Holder.Incarnation || r.ActiveStamp != c.Holder.Stamp || r.ActiveStamp == 0 || r.installation.Box != box { |
| 262 | return ErrFenced |
| 263 | } |
| 264 | return d.sync(host) |
| 265 | }) |
| 266 | } |
| 267 | return g.locked(ctx, host, func(d *recordDir) error { |
| 268 | c, err := g.fresh(ctx, host) |
| 269 | if err != nil { |
| 270 | return err |
| 271 | } |
| 272 | if box == "" || (c.Op.Exec != box && (participant == nil || !participant(c))) { |
| 273 | return ErrNotParticipant |
| 274 | } |
| 275 | stamp := c.Holder.Stamp |
| 276 | if stamp == 0 { |
| 277 | return ErrMalformedClaim |
| 278 | } |
| 279 | r, err := d.read(host) |
| 280 | if err != nil && !errors.Is(err, ErrNoRecord) { |
| 281 | return err |
| 282 | } |
| 283 | if errors.Is(err, ErrNoRecord) { |
| 284 | r = record{V: 1, Host: host, Incarnation: c.Holder.Incarnation, RetiredStamps: []uint64{}} |
| 285 | } else if r.Incarnation != c.Holder.Incarnation { |
| 286 | // Fence evidence belongs to the machine, not the successor holder. |
| 287 | // Preserve it so delayed cleanup can prove and retire only its target. |
| 288 | r = record{V: 1, Host: host, Incarnation: c.Holder.Incarnation, RetiredStamps: []uint64{}, ActiveFence: r.ActiveFence, FenceJournal: r.FenceJournal} |
| 289 | } else { |
| 290 | for _, s := range r.RetiredStamps { |
| 291 | if stamp == s { |
| 292 | return ErrRetiredStamp |
| 293 | } |
| 294 | } |
| 295 | if stamp < r.ActiveStamp { |
| 296 | return ErrOldStamp |
| 297 | } |
| 298 | // Keep refusing trimmed stamps even after retirement sets active to zero. |
| 299 | if n := len(r.RetiredStamps); n != 0 && stamp < r.RetiredStamps[n-1] { |
| 300 | return ErrOldStamp |
| 301 | } |
| 302 | if stamp == r.ActiveStamp { |
| 303 | return d.sync(host) |
| 304 | } |
| 305 | r.retire() |
| 306 | } |
| 307 | r.ActiveStamp = stamp |
| 308 | if err := ctx.Err(); err != nil { |
| 309 | return err |
| 310 | } |
| 311 | return d.write(r) |
| 312 | }) |
| 313 | } |
| 314 | |
| 315 | func (g Guard) RetireInstallation(ctx context.Context, host string, check func(*hostclaim.Claim) error) error { |
| 316 | return g.locked(ctx, host, func(d *recordDir) error { |
| 317 | c, err := g.fresh(ctx, host) |
| 318 | if err != nil { |
| 319 | return err |
| 320 | } |
| 321 | if check == nil { |
| 322 | return ErrCheckRequired |
| 323 | } |
| 324 | if err := check(c); err != nil { |
| 325 | return err |
| 326 | } |
| 327 | r, err := d.read(host) |
| 328 | if err != nil { |
| 329 | return err |
| 330 | } |
| 331 | if r.Incarnation != c.Holder.Incarnation { |
| 332 | return ErrRecordIncarnation |
| 333 | } |
| 334 | if r.ActiveStamp == 0 { |
| 335 | return d.sync(host) |
| 336 | } |
| 337 | r.retire() |
| 338 | if err := ctx.Err(); err != nil { |
| 339 | return err |
| 340 | } |
| 341 | return d.write(r) |
| 342 | }) |
| 343 | } |
| 344 | |
| 345 | func (g Guard) readLocked(ctx context.Context, host string, fn func(*recordDir) error) error { |
| 346 | if g.Root == nil { |
| 347 | return g.locked(ctx, host, fn) |
| 348 | } |
| 349 | if validateFileHost(host) != nil || g.Root.store == nil { |
| 350 | return ErrInvalidInput |
| 351 | } |
| 352 | unlock, err := g.Root.store.Lock(ctx, host+".lock") |
| 353 | if err != nil { |
| 354 | return err |
| 355 | } |
| 356 | defer unlock() |
| 357 | return fn(&recordDir{root: g.Root}) |
| 358 | } |
| 359 | |
amadan