| 1 | // Package archive is the files archive of the move-any-app design §8: a |
| 2 | // deterministic tar of an instance's files directory and its <host>.env, |
| 3 | // compressed with zstd, then sealed under the deployment's config keypair. |
| 4 | // |
| 5 | // The source box builds and seals it; the target box opens, verifies and |
| 6 | // extracts it. The target treats every archive as hostile. The config |
| 7 | // public key is widely held, so anyone who can write the staging bucket can |
| 8 | // seal an archive that opens: authentication proves who it was sealed for, |
| 9 | // not who sealed it. Every limit Build obeys, Verify and Extract enforce |
| 10 | // again, and Build obeys every limit Verify and Extract enforce, so an |
| 11 | // archive Build wrote always lands. |
| 12 | package archive |
| 13 | |
| 14 | import ( |
| 15 | "archive/tar" |
| 16 | "crypto/sha256" |
| 17 | "encoding/hex" |
| 18 | "errors" |
| 19 | "fmt" |
| 20 | "hash" |
| 21 | "io" |
| 22 | "io/fs" |
| 23 | "os" |
| 24 | "path/filepath" |
| 25 | "sort" |
| 26 | |
| 27 | "github.com/klauspost/compress/zstd" |
| 28 | ) |
| 29 | |
| 30 | // ManifestVersion is the Manifest.V this package writes and reads. |
| 31 | const ManifestVersion = 1 |
| 32 | |
| 33 | // Errors from Build (ErrSpecialFile, ErrOverCap, ErrTooManyEntries, |
| 34 | // ErrChanged and the long-path ErrBadPath) reach the operator through the |
| 35 | // source report's S1; their texts are rows MV-N77 to MV-N81 and MV-N86 of |
| 36 | // the copy record. The rest are the target's refusals. |
| 37 | var ( |
| 38 | ErrSpecialFile = errors.New("archive: only regular files and directories move") |
| 39 | ErrOverCap = errors.New("archive: the files directory is over 256 MiB") |
| 40 | ErrTooManyEntries = fmt.Errorf("archive: the files directory holds more than %d entries", MaxEntries) |
| 41 | ErrChanged = errors.New("archive: a file changed while it was being archived") |
| 42 | ErrBadPath = errors.New("archive: refused path") |
| 43 | ErrTooLarge = errors.New("archive: the decoded archive is over its size limit") |
| 44 | ErrNotEmpty = errors.New("archive: the destination files directory is not empty") |
| 45 | // ErrPastTreeCap is Build's refusal of a path longer than TreePathLen |
| 46 | // (hostfiles' cap) but within MaxPathLen; the source says MV-N502. |
| 47 | ErrPastTreeCap = errors.New("archive: a path is past the files tree's length cap") |
| 48 | ) |
| 49 | |
| 50 | // PathError is a Build refusal that names a file: Err is the sentinel |
| 51 | // (errors.Is sees it), Path the file as Build names it (relative to the |
| 52 | // files directory, or the env file's base name). It lets a caller say |
| 53 | // which file in its own words without reading the error's text; Error() is |
| 54 | // the text the sentinel's rows record (MV-N77, MV-N80, MV-N81, MV-N86). |
| 55 | type PathError struct { |
| 56 | Err error |
| 57 | Path string |
| 58 | msg string |
| 59 | } |
| 60 | |
| 61 | func (e *PathError) Error() string { |
| 62 | if e.msg != "" { |
| 63 | return e.msg |
| 64 | } |
| 65 | return e.Err.Error() + ": " + e.Path |
| 66 | } |
| 67 | func (e *PathError) Unwrap() error { return e.Err } |
| 68 | |
| 69 | // The refusals that name a file, each one whole recorded sentence (copy |
| 70 | // record MV-N77, MV-N80, MV-N81 and MV-N86) so the copy gate can read it. |
| 71 | // Err is the sentinel errors.Is sees; the text is the sentinel's, a colon |
| 72 | // and the path, as the rows say. |
| 73 | func specialFileError(path string) error { |
| 74 | return &PathError{Err: ErrSpecialFile, Path: path, msg: fmt.Sprintf("archive: only regular files and directories move: %s", path)} |
| 75 | } |
| 76 | |
| 77 | func hardLinkError(path string) error { |
| 78 | return &PathError{Err: ErrSpecialFile, Path: path, msg: fmt.Sprintf("archive: only regular files and directories move: %s is a hard link", path)} |
| 79 | } |
| 80 | |
| 81 | func changedError(path string) error { |
| 82 | return &PathError{Err: ErrChanged, Path: path, msg: fmt.Sprintf("archive: a file changed while it was being archived: %s", path)} |
| 83 | } |
| 84 | |
| 85 | // The row says 4096, so the sentence does; this fails to compile if |
| 86 | // MaxPathLen is ever changed without the row. |
| 87 | const _ = uint(MaxPathLen-4096) + uint(4096-MaxPathLen) |
| 88 | |
| 89 | // treePathError is a path past TreePathLen but within MaxPathLen: the |
| 90 | // archive format could carry it, the target's files manifest could not. |
| 91 | // The operator's sentence is the caller's (MV-N502, from the sentinel); |
| 92 | // this text reaches the log only. |
| 93 | func treePathError(path string) error { |
| 94 | return &PathError{Err: ErrPastTreeCap, Path: path, msg: fmt.Sprintf("archive: refused path: %s is longer than the files tree's %d-byte cap", path, TreePathLen)} |
| 95 | } |
| 96 | |
| 97 | func longPathError(path string) error { |
| 98 | return &PathError{Err: ErrBadPath, Path: path, msg: fmt.Sprintf("archive: refused path: %s is longer than 4096 bytes", path)} |
| 99 | } |
| 100 | |
| 101 | // The limits as enforced. Variables only so tests can lower them; nothing |
| 102 | // else writes them. |
| 103 | var ( |
| 104 | entryLimit = MaxEntries |
| 105 | decodedLimit = MaxDecoded |
| 106 | ) |
| 107 | |
| 108 | // Entry is one regular file in the manifest: its archive path ("env" or |
| 109 | // "files/<path>"), size and sha256. |
| 110 | type Entry struct { |
| 111 | Path string `json:"path"` |
| 112 | Size int64 `json:"size"` |
| 113 | SHA256 string `json:"sha256"` |
| 114 | } |
| 115 | |
| 116 | // Manifest travels beside the sealed object (§8.2). SealedDataKey is the |
| 117 | // archive's data key sealed to the deployment's config keypair (a v2 |
| 118 | // envelope bound to the app); it is not an object key (F29). |
| 119 | type Manifest struct { |
| 120 | V int `json:"v"` |
| 121 | Op string `json:"op"` |
| 122 | Incarnation string `json:"incarnation"` |
| 123 | Plain string `json:"plain"` |
| 124 | Sealed string `json:"sealed"` |
| 125 | Bytes int64 `json:"bytes"` |
| 126 | Entries []Entry `json:"entries"` |
| 127 | SealedDataKey string `json:"sealed_data_key"` |
| 128 | Chunk int `json:"chunk"` |
| 129 | } |
| 130 | |
| 131 | // item is one listed entry: its archive name, where to open it, and its |
| 132 | // Lstat at listing time, which the open must match. |
| 133 | type item struct { |
| 134 | name string // "env", "files/<rel>" |
| 135 | rel string // path inside the files root; "" for env |
| 136 | root *os.Root // the files root; nil for env |
| 137 | path string // envPath, for env |
| 138 | info fs.FileInfo // as listed |
| 139 | } |
| 140 | |
| 141 | // walk lists the files directory and envPath in archive order, refusing |
| 142 | // anything that is not a regular file or a directory and excluding the |
| 143 | // marker. The files directory is read through an os.Root, so nothing a |
| 144 | // tenant does inside it (a directory swapped for a symlink mid-walk) can |
| 145 | // make the walk or a later open leave it. |
| 146 | // |
| 147 | // skip, when non-nil, is asked each path relative to the files directory |
| 148 | // (slash-separated, as the archive names it) before anything else is read |
| 149 | // about it: a path it accepts is left out as the marker is, a directory |
| 150 | // with everything under it, and is neither typed nor counted, so a |
| 151 | // platform artifact of any kind (a socket, a FIFO) cannot fail the walk. |
| 152 | // envPath is never asked. |
| 153 | func walk(root *os.Root, envPath string, skip func(rel string) bool) ([]item, error) { |
| 154 | var items []item |
| 155 | add := func(it item) error { |
| 156 | if len(items) >= entryLimit { |
| 157 | return ErrTooManyEntries |
| 158 | } |
| 159 | items = append(items, it) |
| 160 | return nil |
| 161 | } |
| 162 | if root != nil { |
| 163 | err := fs.WalkDir(root.FS(), ".", func(p string, d fs.DirEntry, err error) error { |
| 164 | if err != nil { |
| 165 | return err |
| 166 | } |
| 167 | if p == "." { |
| 168 | return nil |
| 169 | } |
| 170 | if p == MarkerName || (skip != nil && skip(p)) { |
| 171 | // The marker is never archived, nor anything under it if |
| 172 | // it is (wrongly) a directory; nor is a path the caller |
| 173 | // skips. |
| 174 | if d.IsDir() { |
| 175 | return fs.SkipDir |
| 176 | } |
| 177 | return nil |
| 178 | } |
| 179 | // Through the root only (census ruling moves-overcap-cut-236, |
| 180 | // binding): a read that fails refuses the walk, never a retry by |
| 181 | // path. |
| 182 | info, err := root.Lstat(p) |
| 183 | if err != nil { |
| 184 | return err |
| 185 | } |
| 186 | if !info.Mode().IsRegular() && !info.IsDir() { |
| 187 | return specialFileError(p) |
| 188 | } |
| 189 | if len(p) > MaxPathLen { |
| 190 | return longPathError(p) |
| 191 | } |
| 192 | if len(p) > TreePathLen { |
| 193 | return treePathError(p) |
| 194 | } |
| 195 | return add(item{name: FilesPrefix + p, rel: p, root: root, info: info}) |
| 196 | }) |
| 197 | if err != nil { |
| 198 | return nil, err |
| 199 | } |
| 200 | } |
| 201 | if envPath != "" { |
| 202 | info, err := os.Lstat(envPath) |
| 203 | switch { |
| 204 | case errors.Is(err, fs.ErrNotExist): |
| 205 | case err != nil: |
| 206 | return nil, err |
| 207 | case !info.Mode().IsRegular(): |
| 208 | return nil, specialFileError(filepath.Base(envPath)) |
| 209 | default: |
| 210 | if err := add(item{name: EnvEntry, path: envPath, info: info}); err != nil { |
| 211 | return nil, err |
| 212 | } |
| 213 | } |
| 214 | } |
| 215 | sort.Slice(items, func(i, j int) bool { return items[i].name < items[j].name }) |
| 216 | return items, nil |
| 217 | } |
| 218 | |
| 219 | // openItem opens a listed regular file. A variable so a test can stand in |
| 220 | // a file that grows after it was listed. |
| 221 | var openItem = openItemChecked |
| 222 | |
| 223 | // openItemChecked opens it.rel inside the files root (or the env path) and |
| 224 | // requires the opened file to be the one listed: a regular file with the |
| 225 | // listed identity and one name. A file swapped after the walk (for a FIFO, another file, |
| 226 | // or a symlink the root would refuse to leave by) fails here. The open is |
| 227 | // non-blocking where the platform has the flag, so a swapped-in FIFO |
| 228 | // returns at once instead of waiting for a writer. |
| 229 | func openItemChecked(it item) (io.ReadCloser, error) { |
| 230 | var f *os.File |
| 231 | var err error |
| 232 | if it.root != nil { |
| 233 | f, err = it.root.OpenFile(it.rel, os.O_RDONLY|openNonblock, 0) |
| 234 | } else { |
| 235 | f, err = os.OpenFile(it.path, os.O_RDONLY|openNonblock, 0) |
| 236 | } |
| 237 | if err != nil { |
| 238 | return nil, err |
| 239 | } |
| 240 | got, err := f.Stat() |
| 241 | if err != nil { |
| 242 | f.Close() |
| 243 | return nil, err |
| 244 | } |
| 245 | if !got.Mode().IsRegular() || !os.SameFile(it.info, got) { |
| 246 | f.Close() |
| 247 | return nil, changedError(displayName(it)) |
| 248 | } |
| 249 | // Phase 1 does not carry hard links. os.Root constrains names, not |
| 250 | // inodes: a pre-existing link to a file outside the instance directory |
| 251 | // is a regular file with the listed identity, and reading it would |
| 252 | // carry the outside inode to the target. Counted on the opened file. |
| 253 | if linkCount(got) > 1 { |
| 254 | f.Close() |
| 255 | return nil, hardLinkError(displayName(it)) |
| 256 | } |
| 257 | return f, nil |
| 258 | } |
| 259 | |
| 260 | func displayName(it item) string { |
| 261 | if it.root == nil { |
| 262 | return filepath.Base(it.path) |
| 263 | } |
| 264 | return it.rel |
| 265 | } |
| 266 | |
| 267 | // hashWriter hashes and counts the tar stream on its way to the |
| 268 | // compressor. It enforces no limit: content is capped where sizes are |
| 269 | // admitted, and the stream's length follows from the entry limits. |
| 270 | type hashWriter struct { |
| 271 | w io.Writer |
| 272 | h hash.Hash |
| 273 | n int64 |
| 274 | } |
| 275 | |
| 276 | func (c *hashWriter) Write(p []byte) (int, error) { |
| 277 | c.h.Write(p) |
| 278 | c.n += int64(len(p)) |
| 279 | return c.w.Write(p) |
| 280 | } |
| 281 | |
| 282 | func digestString(h hash.Hash) string { |
| 283 | return "sha256:" + hex.EncodeToString(h.Sum(nil)) |
| 284 | } |
| 285 | |
| 286 | // built is one pass over the tree. |
| 287 | type built struct { |
| 288 | plain string |
| 289 | entries []Entry |
| 290 | content int64 |
| 291 | tarBytes int64 |
| 292 | count int // tar entries: files, directories and env |
| 293 | } |
| 294 | |
| 295 | // openFilesRoot opens filesDir as an os.Root, requiring it to be a real |
| 296 | // directory (not a symlink to one) at both the Lstat and the open. A |
| 297 | // missing directory is an empty one: the archive then carries env only. |
| 298 | func openFilesRoot(filesDir string) (*os.Root, error) { |
| 299 | if filesDir == "" { |
| 300 | return nil, errors.New("archive: no files directory named") |
| 301 | } |
| 302 | info, err := os.Lstat(filesDir) |
| 303 | if errors.Is(err, fs.ErrNotExist) { |
| 304 | return nil, nil |
| 305 | } |
| 306 | if err != nil { |
| 307 | return nil, err |
| 308 | } |
| 309 | if !info.IsDir() { |
| 310 | // The files root itself is not a directory: untyped, so the caller |
| 311 | // reports it as the files failing to archive, never as a <path>. |
| 312 | return nil, fmt.Errorf("%w: %s", ErrSpecialFile, filesDir) |
| 313 | } |
| 314 | root, err := os.OpenRoot(filesDir) |
| 315 | if err != nil { |
| 316 | return nil, err |
| 317 | } |
| 318 | got, err := root.Stat(".") |
| 319 | if err != nil { |
| 320 | root.Close() |
| 321 | return nil, err |
| 322 | } |
| 323 | if !os.SameFile(info, got) { |
| 324 | root.Close() |
| 325 | // The root itself, not a file in it: untyped, as above. |
| 326 | return nil, fmt.Errorf("%w: %s", ErrChanged, filesDir) |
| 327 | } |
| 328 | return root, nil |
| 329 | } |
| 330 | |
| 331 | // build writes the tar of filesDir and envPath to out and returns what |
| 332 | // Build and PlainDigest report. |
| 333 | func build(filesDir, envPath string, out io.Writer) (built, error) { |
| 334 | return buildFiltered(filesDir, envPath, nil, out) |
| 335 | } |
| 336 | |
| 337 | // buildFiltered is build over the paths skip leaves in (walk). |
| 338 | func buildFiltered(filesDir, envPath string, skip func(rel string) bool, out io.Writer) (built, error) { |
| 339 | root, err := openFilesRoot(filesDir) |
| 340 | if err != nil { |
| 341 | return built{}, err |
| 342 | } |
| 343 | if root != nil { |
| 344 | defer root.Close() |
| 345 | } |
| 346 | items, err := walk(root, envPath, skip) |
| 347 | if err != nil { |
| 348 | return built{}, err |
| 349 | } |
| 350 | var b built |
| 351 | hw := &hashWriter{w: out, h: sha256.New()} |
| 352 | tw := tar.NewWriter(hw) |
| 353 | for _, it := range items { |
| 354 | hdr := &tar.Header{Name: it.name, Mode: int64(it.info.Mode().Perm()), Format: tar.FormatPAX} |
| 355 | if it.info.IsDir() { |
| 356 | hdr.Typeflag = tar.TypeDir |
| 357 | hdr.Name += "/" |
| 358 | } else { |
| 359 | hdr.Typeflag = tar.TypeReg |
| 360 | hdr.Size = it.info.Size() |
| 361 | // Admitted against the cap before the header commits the size. |
| 362 | if hdr.Size > Cap-b.content { |
| 363 | return built{}, ErrOverCap |
| 364 | } |
| 365 | b.content += hdr.Size |
| 366 | } |
| 367 | if err := tw.WriteHeader(hdr); err != nil { |
| 368 | return built{}, err |
| 369 | } |
| 370 | b.count++ |
| 371 | if hdr.Typeflag != tar.TypeReg { |
| 372 | continue |
| 373 | } |
| 374 | e, err := copyItem(tw, it, hdr.Size) |
| 375 | if err != nil { |
| 376 | return built{}, err |
| 377 | } |
| 378 | b.entries = append(b.entries, e) |
| 379 | } |
| 380 | if err := tw.Close(); err != nil { |
| 381 | return built{}, err |
| 382 | } |
| 383 | b.plain, b.tarBytes = digestString(hw.h), hw.n |
| 384 | return b, nil |
| 385 | } |
| 386 | |
| 387 | // copyItem streams exactly the admitted size of one file into the tar. |
| 388 | // This is the cap while streaming: a file is never read past the size its |
| 389 | // header committed (and the cap admitted), so a file that grows after it |
| 390 | // was listed cannot carry the archive past Cap. It is refused as changed, |
| 391 | // as is one that shrank, rather than archived torn. |
| 392 | func copyItem(tw io.Writer, it item, size int64) (Entry, error) { |
| 393 | rc, err := openItem(it) |
| 394 | if err != nil { |
| 395 | return Entry{}, err |
| 396 | } |
| 397 | defer rc.Close() |
| 398 | fh := sha256.New() |
| 399 | n, err := io.CopyN(io.MultiWriter(tw, fh), rc, size) |
| 400 | if errors.Is(err, io.EOF) { |
| 401 | return Entry{}, changedError(displayName(it)) |
| 402 | } |
| 403 | if err != nil { |
| 404 | return Entry{}, err |
| 405 | } |
| 406 | var one [1]byte |
| 407 | m, err := io.ReadFull(rc, one[:]) |
| 408 | if m > 0 { |
| 409 | return Entry{}, changedError(displayName(it)) |
| 410 | } |
| 411 | if err != io.EOF { |
| 412 | return Entry{}, err |
| 413 | } |
| 414 | return Entry{Path: it.name, Size: n, SHA256: digestString(fh)}, nil |
| 415 | } |
| 416 | |
| 417 | // Build writes the archive (tar then zstd) to w and returns the plaintext |
| 418 | // digest (of the tar stream), the regular-file entries and the content |
| 419 | // byte count. Owners and times are zeroed, modes kept, so equal content |
| 420 | // gives equal plaintext (§8.1). The files directory's entries live under |
| 421 | // "files/" and <host>.env is "env", in separate namespaces, so an |
| 422 | // application file or directory named "env" is carried as "files/env" and |
| 423 | // can never collide with the overlay. A missing envPath (or "") means the |
| 424 | // archive has no env entry; a missing filesDir is an empty one. |
| 425 | func Build(filesDir, envPath string, w io.Writer) (plain string, entries []Entry, bytes int64, err error) { |
| 426 | return BuildFiltered(filesDir, envPath, nil, w) |
| 427 | } |
| 428 | |
| 429 | // BuildFiltered is Build over the files directory without the paths skip |
| 430 | // accepts. skip is asked each path relative to filesDir, slash-separated |
| 431 | // (a top-level entry is its bare name); a skipped directory is skipped |
| 432 | // whole, and a skipped entry of any type is never read, typed or counted. |
| 433 | // The env entry is never asked. A nil skip is Build. Nothing calls it with |
| 434 | // a filter today: the offline move's sealer archives a nested instance's |
| 435 | // files directory (<instance dir>/<host>, which holds nothing of the |
| 436 | // platform's but the marker) with Build, and refuses the seal over any |
| 437 | // entry of the app's beside it. |
| 438 | func BuildFiltered(filesDir, envPath string, skip func(rel string) bool, w io.Writer) (plain string, entries []Entry, bytes int64, err error) { |
| 439 | zw, err := zstd.NewWriter(w, zstd.WithEncoderConcurrency(1), zstd.WithWindowSize(zstdWindow)) |
| 440 | if err != nil { |
| 441 | return "", nil, 0, err |
| 442 | } |
| 443 | b, err := buildFiltered(filesDir, envPath, skip, zw) |
| 444 | if err != nil { |
| 445 | zw.Close() |
| 446 | return "", nil, 0, err |
| 447 | } |
| 448 | if err := zw.Close(); err != nil { |
| 449 | return "", nil, 0, err |
| 450 | } |
| 451 | return b.plain, b.entries, b.content, nil |
| 452 | } |
| 453 | |
| 454 | // PlainDigest runs Build's walk and tar without compressing or writing |
| 455 | // anything, and returns the same plaintext digest Build would, the content |
| 456 | // byte count and the number of regular-file entries (env included). It is |
| 457 | // the baseline the source records at `stopped` before any archive exists |
| 458 | // (§8.3), and the comparison every recovery makes. |
| 459 | func PlainDigest(filesDir, envPath string) (plain string, bytes int64, entries int, err error) { |
| 460 | b, err := build(filesDir, envPath, io.Discard) |
| 461 | if err != nil { |
| 462 | return "", 0, 0, err |
| 463 | } |
| 464 | return b.plain, b.content, len(b.entries), nil |
| 465 | } |
| 466 | |
amadan