| 1 | package hostfiles |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/rand" |
| 6 | "encoding/hex" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "io/fs" |
| 10 | "os" |
| 11 | "path" |
| 12 | "path/filepath" |
| 13 | "slices" |
| 14 | "sort" |
| 15 | "strings" |
| 16 | |
| 17 | "github.com/carlosframework/platform/internal/release" |
| 18 | ) |
| 19 | |
| 20 | // syncDir makes a directory's entries durable. A test seam. |
| 21 | // |
| 22 | // The open is descriptor-relative (os.Root) with restoreDirFlags: |
| 23 | // O_DIRECTORY|O_NONBLOCK|O_NOFOLLOW on unix, so whatever was swapped in at |
| 24 | // rel since realParents looked (a FIFO, a file) is refused at once rather |
| 25 | // than waited on: a plain open of a FIFO with no writer never returns, and |
| 26 | // no context can interrupt it. |
| 27 | var syncDir = func(root *os.Root, rel string) error { |
| 28 | d, err := root.OpenFile(filepath.FromSlash(rel), restoreDirFlags, 0) |
| 29 | if err != nil { |
| 30 | return err |
| 31 | } |
| 32 | defer d.Close() |
| 33 | return d.Sync() |
| 34 | } |
| 35 | |
| 36 | type RestoreStats struct { |
| 37 | Written, Removed int |
| 38 | Skipped []string // paths the manifest holds as present but unpublished |
| 39 | Superseded []string // live paths not written: a live entry above or below them is newer |
| 40 | } |
| 41 | |
| 42 | // Restore writes into dir what the manifest under prefix names and removes |
| 43 | // what it tombstones: every tombstoned entry is removed, every published one |
| 44 | // written, and the skipped ones reported. It does not delete files the |
| 45 | // manifest does not name. It reads nothing but the manifest and the objects |
| 46 | // the manifest names, and verifies each object against the hash the manifest |
| 47 | // records. |
| 48 | // |
| 49 | // Every path is validated before anything is touched, so a manifest with |
| 50 | // one bad path changes nothing. Every file operation goes through an |
| 51 | // os.Root opened on dir, so nothing is written or removed outside it even |
| 52 | // if dir already holds a symlink pointing elsewhere. |
| 53 | // |
| 54 | // When Restore returns nil, what it did is durable: each file was synced |
| 55 | // before its rename, and every directory in which a name was created, |
| 56 | // replaced or removed has been synced since. A caller may record the |
| 57 | // installation as complete on that basis, and on no weaker one. |
| 58 | // |
| 59 | // origins are the generation prefixes whose objects this restore may read: |
| 60 | // the set the caller's pin admits. nil or empty means prefix alone. Every |
| 61 | // entry's object must be exactly "<origin>objects/<seq>-<sha256>" for one |
| 62 | // of them, with the entry's own Seq and SHA256; any other entry refuses the |
| 63 | // whole manifest before anything is touched (ErrOrigin when the key is |
| 64 | // well formed but its origin is not admitted). |
| 65 | func Restore(ctx context.Context, s release.Store, prefix, dir string, origins []string) (Manifest, RestoreStats, error) { |
| 66 | return restore(ctx, s, prefix, origins, func() (*os.Root, func(), error) { |
| 67 | root, err := os.OpenRoot(dir) |
| 68 | if err != nil { |
| 69 | return nil, nil, fmt.Errorf("hostfiles: restore into %s: %w", dir, err) |
| 70 | } |
| 71 | return root, func() { root.Close() }, nil |
| 72 | }) |
| 73 | } |
| 74 | |
| 75 | // RestoreRoot is Restore into a directory the caller has already opened as |
| 76 | // root, and does not close. os.OpenRoot resolves Restore's dir as a path, |
| 77 | // following a link anywhere in it and blocking on a FIFO at its last |
| 78 | // component; a caller restoring into ground someone else can write opens |
| 79 | // the root itself without either (instancefs.OpenRootBeneath) and restores |
| 80 | // here. Everything else is Restore's: the one implementation both share. |
| 81 | func RestoreRoot(ctx context.Context, s release.Store, prefix string, root *os.Root, origins []string) (Manifest, RestoreStats, error) { |
| 82 | if root == nil { |
| 83 | return Manifest{}, RestoreStats{}, errors.New("hostfiles: restore into a nil root") |
| 84 | } |
| 85 | return restore(ctx, s, prefix, origins, func() (*os.Root, func(), error) { return root, func() {}, nil }) |
| 86 | } |
| 87 | |
| 88 | // restore is Restore and RestoreRoot. open is called only once the |
| 89 | // manifest is validated and names something to do, so a bad manifest |
| 90 | // touches no directory and an empty one needs none. |
| 91 | func restore(ctx context.Context, s release.Store, prefix string, origins []string, open func() (*os.Root, func(), error)) (Manifest, RestoreStats, error) { |
| 92 | var st RestoreStats |
| 93 | if err := checkPrefix(prefix); err != nil { |
| 94 | return Manifest{}, st, err |
| 95 | } |
| 96 | origins, err := admitted(prefix, origins) |
| 97 | if err != nil { |
| 98 | return Manifest{}, st, err |
| 99 | } |
| 100 | m, _, _, err := load(ctx, s, prefix) |
| 101 | if err != nil { |
| 102 | return Manifest{}, st, err |
| 103 | } |
| 104 | if err := checkManifest(m, origins); err != nil { |
| 105 | return Manifest{}, st, err |
| 106 | } |
| 107 | var live, dead []string |
| 108 | for p, e := range m.Entries { |
| 109 | if e.Deleted { |
| 110 | dead = append(dead, p) |
| 111 | } else { |
| 112 | live = append(live, p) |
| 113 | } |
| 114 | } |
| 115 | if len(live)+len(dead) == 0 { |
| 116 | return m, st, nil |
| 117 | } |
| 118 | root, done, err := open() |
| 119 | if err != nil { |
| 120 | return Manifest{}, st, err |
| 121 | } |
| 122 | defer done() |
| 123 | |
| 124 | // Tombstones first, deepest first, so a directory that became a file |
| 125 | // is emptied before the file is written over it. |
| 126 | // Every directory on the way to a path the manifest speaks of, live or |
| 127 | // tombstoned, is a directory whose entries this restore may have |
| 128 | // changed: now, or in an earlier attempt that removed or created the |
| 129 | // name and then failed before its sync. So the set does not depend on |
| 130 | // what this call happened to find already done. |
| 131 | touched := map[string]bool{".": true} |
| 132 | mark := func(p string) { |
| 133 | for d := path.Dir(p); d != "."; d = path.Dir(d) { |
| 134 | touched[d] = true |
| 135 | } |
| 136 | } |
| 137 | sort.Slice(dead, func(i, j int) bool { |
| 138 | di, dj := strings.Count(dead[i], "/"), strings.Count(dead[j], "/") |
| 139 | if di != dj { |
| 140 | return di > dj |
| 141 | } |
| 142 | return dead[i] < dead[j] |
| 143 | }) |
| 144 | for _, p := range dead { |
| 145 | mark(p) |
| 146 | removed, err := removeInside(root, p) |
| 147 | if err != nil { |
| 148 | return Manifest{}, st, fmt.Errorf("hostfiles: remove %s: %w", p, err) |
| 149 | } |
| 150 | if removed { |
| 151 | st.Removed++ |
| 152 | } |
| 153 | } |
| 154 | sort.Strings(live) |
| 155 | loser := superseded(m) |
| 156 | for _, p := range live { |
| 157 | if loser[p] { |
| 158 | st.Superseded = append(st.Superseded, p) |
| 159 | } |
| 160 | } |
| 161 | for _, p := range live { |
| 162 | if loser[p] { |
| 163 | continue |
| 164 | } |
| 165 | e := m.Entries[p] |
| 166 | if e.Skipped != "" { |
| 167 | st.Skipped = append(st.Skipped, p) |
| 168 | continue |
| 169 | } |
| 170 | mark(p) |
| 171 | body, err := fetch(ctx, s, e, origins) |
| 172 | if err != nil { |
| 173 | return Manifest{}, st, fmt.Errorf("hostfiles: %s: %w", p, err) |
| 174 | } |
| 175 | if err := writeInside(root, p, body, os.FileMode(e.Mode&0o777)); err != nil { |
| 176 | return Manifest{}, st, fmt.Errorf("hostfiles: write %s: %w", p, err) |
| 177 | } |
| 178 | st.Written++ |
| 179 | } |
| 180 | // Sync the directories that survive, deepest first and the root last, |
| 181 | // so a directory is durable before the entry that names it is. One |
| 182 | // that no longer exists as a real directory (it was emptied and then |
| 183 | // replaced by a file, or it sat under a path that became a file) has |
| 184 | // nothing to sync: the directory above it records its going. |
| 185 | dirs := make([]string, 0, len(touched)) |
| 186 | for d := range touched { |
| 187 | dirs = append(dirs, d) |
| 188 | } |
| 189 | sort.Slice(dirs, func(i, j int) bool { |
| 190 | if (dirs[i] == ".") != (dirs[j] == ".") { |
| 191 | return dirs[j] == "." |
| 192 | } |
| 193 | di, dj := strings.Count(dirs[i], "/"), strings.Count(dirs[j], "/") |
| 194 | if di != dj { |
| 195 | return di > dj |
| 196 | } |
| 197 | return dirs[i] < dirs[j] |
| 198 | }) |
| 199 | for _, d := range dirs { |
| 200 | if d != "." { |
| 201 | if ok, err := realParents(root, d+"/x"); err != nil { |
| 202 | return Manifest{}, st, fmt.Errorf("hostfiles: sync directory %s: %w", d, err) |
| 203 | } else if !ok { |
| 204 | continue |
| 205 | } |
| 206 | } |
| 207 | if err := syncDir(root, d); err != nil { |
| 208 | return Manifest{}, st, fmt.Errorf("hostfiles: sync directory %s: %w", d, err) |
| 209 | } |
| 210 | } |
| 211 | return m, st, nil |
| 212 | } |
| 213 | |
| 214 | // superseded is the one overlap rule for live entries: a live file and a |
| 215 | // live entry beneath it cannot both exist, so the newer one (higher Seq) |
| 216 | // wins and the other is superseded; equal sequences keep the deeper one. |
| 217 | // Restore leaves a superseded entry unwritten and Present does not look |
| 218 | // for it, so what Restore produces is exactly what Present accepts |
| 219 | // (stage B final review A1). Tombstones take no part. |
| 220 | func superseded(m Manifest) map[string]bool { |
| 221 | loser := map[string]bool{} |
| 222 | for p, e := range m.Entries { |
| 223 | if e.Deleted { |
| 224 | continue |
| 225 | } |
| 226 | for a := path.Dir(p); a != "."; a = path.Dir(a) { |
| 227 | above, ok := m.Entries[a] |
| 228 | if !ok || above.Deleted { |
| 229 | continue |
| 230 | } |
| 231 | if above.Seq > e.Seq { |
| 232 | loser[p] = true |
| 233 | } else { |
| 234 | loser[a] = true |
| 235 | } |
| 236 | } |
| 237 | } |
| 238 | return loser |
| 239 | } |
| 240 | |
| 241 | // ReadPlatform returns a platform-managed item's bytes, verified. A name |
| 242 | // the manifest does not hold is os.ErrNotExist. prefix is where m was |
| 243 | // loaded from; origins are as for Restore (nil or empty: prefix alone). |
| 244 | func ReadPlatform(ctx context.Context, s release.Store, prefix string, m Manifest, name string, origins []string) ([]byte, error) { |
| 245 | if err := checkPrefix(prefix); err != nil { |
| 246 | return nil, err |
| 247 | } |
| 248 | origins, err := admitted(prefix, origins) |
| 249 | if err != nil { |
| 250 | return nil, err |
| 251 | } |
| 252 | e, ok := m.Platform[name] |
| 253 | if !ok { |
| 254 | return nil, fmt.Errorf("hostfiles: platform item %q: %w", name, os.ErrNotExist) |
| 255 | } |
| 256 | return fetch(ctx, s, e, origins) |
| 257 | } |
| 258 | |
| 259 | // checkManifest validates every path and every entry before a caller acts |
| 260 | // on any of them. |
| 261 | func checkManifest(m Manifest, origins []string) error { |
| 262 | for name, e := range m.Platform { |
| 263 | if err := checkEntry(e, origins); err != nil { |
| 264 | return fmt.Errorf("hostfiles: platform item %q: %w", name, err) |
| 265 | } |
| 266 | } |
| 267 | for p, e := range m.Entries { |
| 268 | if _, err := CleanPath(p); err != nil { |
| 269 | return err |
| 270 | } |
| 271 | if err := withinTreeCap(p); err != nil { |
| 272 | return err // no restore could recreate it: refused before anything is touched |
| 273 | } |
| 274 | if err := checkEntry(e, origins); err != nil { |
| 275 | return fmt.Errorf("hostfiles: %s: %w", p, err) |
| 276 | } |
| 277 | } |
| 278 | return nil |
| 279 | } |
| 280 | |
| 281 | // checkEntry refuses a published entry whose object is not one this |
| 282 | // package would have written for it in an admitted origin, or whose size |
| 283 | // it could never have published. Tombstones and skipped entries name no |
| 284 | // object. origins are already validated prefixes. |
| 285 | func checkEntry(e Entry, origins []string) error { |
| 286 | if e.Deleted || e.Skipped != "" { |
| 287 | return nil |
| 288 | } |
| 289 | if len(e.SHA256) != 64 || strings.Trim(e.SHA256, "0123456789abcdef") != "" { |
| 290 | return fmt.Errorf("sha256 %q is not 64 lowercase hex characters", e.SHA256) |
| 291 | } |
| 292 | if _, err := CleanPath(e.Object); err != nil { |
| 293 | return fmt.Errorf("object %q is not a clean bucket key", e.Object) |
| 294 | } |
| 295 | want := objectKey(e.Seq, e.SHA256) |
| 296 | origin, ok := strings.CutSuffix(e.Object, want) |
| 297 | if !ok { |
| 298 | return fmt.Errorf("object %q does not end in %q", e.Object, want) |
| 299 | } |
| 300 | if !slices.Contains(origins, origin) { |
| 301 | return fmt.Errorf("%w: object %q lies under %q, not one of %q", ErrOrigin, e.Object, origin, origins) |
| 302 | } |
| 303 | if e.Size < 0 || e.Size > MaxFileBytes { |
| 304 | return fmt.Errorf("size %d is outside 0 to %d", e.Size, MaxFileBytes) |
| 305 | } |
| 306 | return nil |
| 307 | } |
| 308 | |
| 309 | // fetch reads e's object at exactly the key the entry names. |
| 310 | func fetch(ctx context.Context, s release.Store, e Entry, origins []string) ([]byte, error) { |
| 311 | if err := checkEntry(e, origins); err != nil { |
| 312 | return nil, err |
| 313 | } |
| 314 | body, _, err := s.Get(ctx, e.Object) |
| 315 | if err != nil { |
| 316 | return nil, err |
| 317 | } |
| 318 | if got := sum(body); got != e.SHA256 { |
| 319 | return nil, fmt.Errorf("object %s has sha256 %s, the manifest says %s", e.Object, got, e.SHA256) |
| 320 | } |
| 321 | if int64(len(body)) != e.Size { |
| 322 | return nil, fmt.Errorf("object %s has size %d, the manifest says %d", e.Object, len(body), e.Size) |
| 323 | } |
| 324 | return body, nil |
| 325 | } |
| 326 | |
| 327 | // realParents reports whether every ancestor of rel inside root is a real |
| 328 | // directory: not a symlink, not a file, not missing. |
| 329 | func realParents(root *os.Root, rel string) (bool, error) { |
| 330 | parts := strings.Split(rel, "/") |
| 331 | cur := "" |
| 332 | for _, part := range parts[:len(parts)-1] { |
| 333 | cur = path.Join(cur, part) |
| 334 | fi, err := root.Lstat(filepath.FromSlash(cur)) |
| 335 | if errors.Is(err, fs.ErrNotExist) { |
| 336 | return false, nil |
| 337 | } |
| 338 | if err != nil { |
| 339 | return false, err |
| 340 | } |
| 341 | if !fi.IsDir() { |
| 342 | return false, nil |
| 343 | } |
| 344 | } |
| 345 | return true, nil |
| 346 | } |
| 347 | |
| 348 | // removeInside removes rel if it is a file of ours. A path whose ancestor |
| 349 | // is a symlink, a file or missing is not ours to remove and is left alone: |
| 350 | // that is how a tombstone for "a/b" behaves once "a" has become a file, |
| 351 | // and why a symlinked ancestor never lets a delete reach outside root. |
| 352 | func removeInside(root *os.Root, rel string) (bool, error) { |
| 353 | ok, err := realParents(root, rel) |
| 354 | if err != nil || !ok { |
| 355 | return false, err |
| 356 | } |
| 357 | name := filepath.FromSlash(rel) |
| 358 | fi, err := root.Lstat(name) |
| 359 | if errors.Is(err, fs.ErrNotExist) { |
| 360 | return false, nil |
| 361 | } |
| 362 | if err != nil { |
| 363 | return false, err |
| 364 | } |
| 365 | if fi.IsDir() { |
| 366 | // The path is a directory now: a later entry lives under it, and |
| 367 | // the tombstone is for a file that no longer exists. |
| 368 | return false, nil |
| 369 | } |
| 370 | if err := root.Remove(name); err != nil && !errors.Is(err, fs.ErrNotExist) { |
| 371 | return false, err |
| 372 | } |
| 373 | return true, nil |
| 374 | } |
| 375 | |
| 376 | // writeInside writes rel atomically inside root. Anything in the way that |
| 377 | // is not a real directory (a file where a directory must be, a symlink) |
| 378 | // is removed first, and a directory where the file must be is removed |
| 379 | // with its contents: the manifest is the truth about this tree. |
| 380 | func writeInside(root *os.Root, rel string, body []byte, mode os.FileMode) error { |
| 381 | if mode == 0 { |
| 382 | mode = 0o600 |
| 383 | } |
| 384 | parent := path.Dir(rel) |
| 385 | if parent != "." { |
| 386 | cur := "" |
| 387 | for _, part := range strings.Split(parent, "/") { |
| 388 | cur = path.Join(cur, part) |
| 389 | fi, err := root.Lstat(filepath.FromSlash(cur)) |
| 390 | if errors.Is(err, fs.ErrNotExist) { |
| 391 | break |
| 392 | } |
| 393 | if err != nil { |
| 394 | return err |
| 395 | } |
| 396 | if !fi.IsDir() { |
| 397 | if err := root.Remove(filepath.FromSlash(cur)); err != nil { |
| 398 | return err |
| 399 | } |
| 400 | break |
| 401 | } |
| 402 | } |
| 403 | if err := root.MkdirAll(filepath.FromSlash(parent), 0o755); err != nil { |
| 404 | return err |
| 405 | } |
| 406 | } |
| 407 | name := filepath.FromSlash(rel) |
| 408 | if fi, err := root.Lstat(name); err == nil && fi.IsDir() { |
| 409 | if err := root.RemoveAll(name); err != nil { |
| 410 | return err |
| 411 | } |
| 412 | } |
| 413 | var rnd [8]byte |
| 414 | if _, err := rand.Read(rnd[:]); err != nil { |
| 415 | return err |
| 416 | } |
| 417 | tmp := filepath.FromSlash(path.Join(parent, ".carlos-restore-"+hex.EncodeToString(rnd[:]))) |
| 418 | f, err := root.OpenFile(tmp, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) |
| 419 | if err != nil { |
| 420 | return err |
| 421 | } |
| 422 | done := false |
| 423 | defer func() { |
| 424 | if !done { |
| 425 | f.Close() |
| 426 | root.Remove(tmp) |
| 427 | } |
| 428 | }() |
| 429 | if _, err := f.Write(body); err != nil { |
| 430 | return err |
| 431 | } |
| 432 | if err := f.Chmod(mode); err != nil { |
| 433 | return err |
| 434 | } |
| 435 | if err := f.Sync(); err != nil { |
| 436 | return err |
| 437 | } |
| 438 | if err := f.Close(); err != nil { |
| 439 | return err |
| 440 | } |
| 441 | if err := root.Rename(tmp, name); err != nil { |
| 442 | return err |
| 443 | } |
| 444 | done = true |
| 445 | return nil |
| 446 | } |
| 447 | |
amadan