| 1 | package console |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/sha256" |
| 6 | "encoding/hex" |
| 7 | "encoding/json" |
| 8 | "errors" |
| 9 | "fmt" |
| 10 | "io" |
| 11 | "log" |
| 12 | "os" |
| 13 | "path/filepath" |
| 14 | "strings" |
| 15 | "sync" |
| 16 | "time" |
| 17 | |
| 18 | "github.com/carlosframework/platform/internal/geoip" |
| 19 | "github.com/carlosframework/platform/internal/release" |
| 20 | ) |
| 21 | |
| 22 | // geoipProbeAddrs is the country probe (spec §4.1 step 4): at least three |
| 23 | // must answer with a country. A variable so tests can use addresses the |
| 24 | // MaxMind fixtures hold (FACTS.md). |
| 25 | var geoipProbeAddrs = []string{"8.8.8.8", "1.1.1.1", "9.9.9.9", "208.67.222.222", "2001:4860:4860::8888"} |
| 26 | |
| 27 | const geoipProbeNeed = 3 |
| 28 | |
| 29 | var ( |
| 30 | errGeoipBusy = errors.New("a geoip publish is already in progress") |
| 31 | errGeoipRefused = errors.New("geoip database refused") |
| 32 | ) |
| 33 | |
| 34 | func refused(category string) error { return fmt.Errorf("%w: %s", errGeoipRefused, category) } |
| 35 | |
| 36 | // geoipPublisher is the console's single writer of control/geoip/ (spec §3). |
| 37 | // mu is held through cleanup by every publish, off, sweep and expiry, and a |
| 38 | // publish that finds it held answers errGeoipBusy rather than queueing. |
| 39 | type geoipPublisher struct { |
| 40 | mu sync.Mutex |
| 41 | store release.Store |
| 42 | tmpRoot string |
| 43 | maxBytes int64 |
| 44 | deadline time.Duration |
| 45 | now func() time.Time |
| 46 | // expired is D1's rule, a field so a test can apply it to the GeoIP2 |
| 47 | // fixtures (which never expire) without a GeoLite2 file in the repo. |
| 48 | expired func(dbType string, build, now time.Time) bool |
| 49 | // storeTimeout bounds a single small bucket read made while holding, |
| 50 | // or about to hold, the publish slot. |
| 51 | storeTimeout time.Duration |
| 52 | } |
| 53 | |
| 54 | type geoipPublishOpts struct { |
| 55 | Origin, Edition string |
| 56 | Force bool |
| 57 | // RequireType refuses a database of any other type (the feed's check, |
| 58 | // spec §10.3, including on the first publish). Monotonic refuses a |
| 59 | // build that is not newer than the current pointer's when that pointer |
| 60 | // came from the same Origin. The feed sets both and never Force. |
| 61 | RequireType string |
| 62 | Monotonic bool |
| 63 | } |
| 64 | |
| 65 | // newGeoipPublisher clears tmpRoot (the dedicated geoip-staging directory) |
| 66 | // before anything can stage into it: a console that crashed mid-upload or |
| 67 | // mid-job must not leave its 1 GiB behind for good (spec §4.2, codex R6-2). |
| 68 | func newGeoipPublisher(store release.Store, tmpRoot string) *geoipPublisher { |
| 69 | if tmpRoot != "" { |
| 70 | if err := os.RemoveAll(tmpRoot); err != nil { |
| 71 | log.Printf("console: geoip staging directory could not be cleared: %v", err) |
| 72 | } |
| 73 | if err := os.MkdirAll(tmpRoot, 0o700); err != nil { |
| 74 | log.Printf("console: geoip staging directory could not be created: %v", err) |
| 75 | } |
| 76 | } |
| 77 | return &geoipPublisher{store: store, tmpRoot: tmpRoot, maxBytes: geoip.MaxBlobBytes, |
| 78 | deadline: 15 * time.Minute, now: time.Now, expired: geoip.Expired, storeTimeout: 30 * time.Second} |
| 79 | } |
| 80 | |
| 81 | func marshalPointer(p geoip.Pointer) ([]byte, error) { return json.MarshalIndent(p, "", " ") } |
| 82 | |
| 83 | // Current reads the pointer. present=false means no pointer (feature off). |
| 84 | func (p *geoipPublisher) Current(ctx context.Context) (geoip.Pointer, string, bool, error) { |
| 85 | raw, etag, err := readPointerCapped(ctx, p.store) |
| 86 | if errors.Is(err, os.ErrNotExist) { |
| 87 | return geoip.Pointer{}, "", false, nil |
| 88 | } |
| 89 | if errors.Is(err, errPointerTooLarge) { |
| 90 | // Present but not a pointer anyone wrote: replaceable, never decoded. |
| 91 | return geoip.Pointer{}, etag, true, err |
| 92 | } |
| 93 | if err != nil { |
| 94 | return geoip.Pointer{}, "", false, err |
| 95 | } |
| 96 | ptr, err := geoip.ParsePointer(raw) |
| 97 | if err != nil { |
| 98 | return geoip.Pointer{}, etag, true, err |
| 99 | } |
| 100 | return ptr, etag, true, nil |
| 101 | } |
| 102 | |
| 103 | // geoipStaged is an upload on disk awaiting Commit. Whoever holds one owns |
| 104 | // the publish slot (p.mu) and the temporary directory: Commit or Abandon |
| 105 | // releases both (spec §4.2, codex R5-1). |
| 106 | type geoipStaged struct { |
| 107 | dir, path, sha string |
| 108 | size int64 |
| 109 | // The pointer as it stood when the slot was taken (spec §3 step 1): |
| 110 | // every comparison is against this read, and the pointer write is |
| 111 | // conditional on its etag, so a foreign write during the upload is |
| 112 | // refused rather than overwritten. |
| 113 | cur geoip.Pointer |
| 114 | curEtag string |
| 115 | curPresent bool |
| 116 | curErr error |
| 117 | } |
| 118 | |
| 119 | // Stage reserves the one publish slot BEFORE reading any of src, then |
| 120 | // streams src to a temporary file, hashing and capping as it copies. On |
| 121 | // success the slot stays held for Commit; on failure it is released. |
| 122 | func (p *geoipPublisher) Stage(ctx context.Context, src io.Reader) (*geoipStaged, error) { |
| 123 | if !p.mu.TryLock() { |
| 124 | return nil, errGeoipBusy |
| 125 | } |
| 126 | dir, err := os.MkdirTemp(p.tmpRoot, "geoip-publish-*") |
| 127 | if err != nil { |
| 128 | p.mu.Unlock() |
| 129 | return nil, err |
| 130 | } |
| 131 | st := &geoipStaged{dir: dir, path: filepath.Join(dir, "db.mmdb")} |
| 132 | // The pointer read is bounded on its own: the caller's context may be |
| 133 | // an HTTP request whose only deadline is on reading its body (code |
| 134 | // review R4-1). |
| 135 | rctx, rcancel := context.WithTimeout(ctx, p.storeTimeout) |
| 136 | st.cur, st.curEtag, st.curPresent, st.curErr = p.Current(rctx) |
| 137 | rcancel() |
| 138 | if st.curErr != nil && !st.curPresent { |
| 139 | p.Abandon(st) |
| 140 | return nil, st.curErr // the store itself failed |
| 141 | } |
| 142 | st.sha, st.size, err = copyCapped(ctx, st.path, src, p.maxBytes) |
| 143 | if err != nil { |
| 144 | p.Abandon(st) |
| 145 | return nil, err |
| 146 | } |
| 147 | return st, nil |
| 148 | } |
| 149 | |
| 150 | // Abandon releases a staged upload that will not be committed. |
| 151 | func (p *geoipPublisher) Abandon(st *geoipStaged) { |
| 152 | os.RemoveAll(st.dir) |
| 153 | p.mu.Unlock() |
| 154 | } |
| 155 | |
| 156 | var errPointerTooLarge = errors.New("the object at the geoip pointer key is larger than a pointer can be") |
| 157 | |
| 158 | // readPointerCapped reads control/geoip/current.json streamed and capped at |
| 159 | // geoip.MaxPointerBytes, so an oversized object there can never be pulled |
| 160 | // into the console's memory (code review R1-3). The etag is the store's; |
| 161 | // DirStore cannot report one from Open, and its etag is defined as the |
| 162 | // content's sha256 (release.DirStore, etagOf), so it is computed here from |
| 163 | // the bytes, which are complete whenever they are under the cap. |
| 164 | func readPointerCapped(ctx context.Context, s release.Store) ([]byte, string, error) { |
| 165 | return readCappedKey(ctx, s, geoip.PointerKey, geoip.MaxPointerBytes) |
| 166 | } |
| 167 | |
| 168 | // readCappedKey is readPointerCapped for any small control object. |
| 169 | func readCappedKey(ctx context.Context, s release.Store, key string, max int) ([]byte, string, error) { |
| 170 | rc, etag, err := release.OpenObject(ctx, s, key) |
| 171 | if err != nil { |
| 172 | return nil, "", err |
| 173 | } |
| 174 | defer rc.Close() |
| 175 | b, err := io.ReadAll(io.LimitReader(rc, int64(max)+1)) |
| 176 | if err != nil { |
| 177 | return nil, "", err |
| 178 | } |
| 179 | if len(b) > max { |
| 180 | if etag == "" { |
| 181 | // Without the store's etag a conditional replace cannot be |
| 182 | // aimed at it; only off (unconditional delete) clears it. |
| 183 | return nil, "", errPointerTooLarge |
| 184 | } |
| 185 | return nil, etag, errPointerTooLarge |
| 186 | } |
| 187 | if etag == "" { |
| 188 | sum := sha256.Sum256(b) |
| 189 | etag = hex.EncodeToString(sum[:]) |
| 190 | } |
| 191 | return b, etag, nil |
| 192 | } |
| 193 | |
| 194 | // Publish is Stage then Commit, under one deadline. |
| 195 | func (p *geoipPublisher) Publish(ctx context.Context, src io.Reader, o geoipPublishOpts) (geoip.Pointer, error) { |
| 196 | ctx, cancel := context.WithTimeout(ctx, p.deadline) |
| 197 | defer cancel() |
| 198 | st, err := p.Stage(ctx, src) |
| 199 | if err != nil { |
| 200 | return geoip.Pointer{}, err |
| 201 | } |
| 202 | return p.Commit(ctx, st, o) |
| 203 | } |
| 204 | |
| 205 | // Commit is spec §4.1 steps 3-6 on a staged upload. It always removes the |
| 206 | // staged directory and releases the slot. |
| 207 | func (p *geoipPublisher) Commit(ctx context.Context, st *geoipStaged, o geoipPublishOpts) (geoip.Pointer, error) { |
| 208 | defer p.Abandon(st) |
| 209 | if _, ok := ctx.Deadline(); !ok { |
| 210 | var cancel context.CancelFunc |
| 211 | ctx, cancel = context.WithTimeout(ctx, p.deadline) |
| 212 | defer cancel() |
| 213 | } |
| 214 | // The file is untrusted: it is parsed only in a child process (code |
| 215 | // review R9), never here. |
| 216 | db, err := verifyInChild(ctx, st.path, geoipProbeAddrs) |
| 217 | switch { |
| 218 | case ctx.Err() != nil: |
| 219 | return geoip.Pointer{}, ctx.Err() |
| 220 | case errors.Is(err, errGeoipVerifyTimeout): |
| 221 | return geoip.Pointer{}, refused("the database took too long to verify") |
| 222 | case errors.Is(err, errGeoipVerifyTree): |
| 223 | return geoip.Pointer{}, refused("the database's search tree is malformed") |
| 224 | case err != nil: |
| 225 | return geoip.Pointer{}, refused("not a valid MaxMind-format database") |
| 226 | } |
| 227 | if db.Countries < geoipProbeNeed { |
| 228 | return geoip.Pointer{}, refused("the database has no country data") |
| 229 | } |
| 230 | dbBuild := time.Unix(db.BuildEpoch, 0).UTC() |
| 231 | // D1 at the door (codex R7-1): an already-expired GeoLite2 would replace |
| 232 | // a usable database with one every edge must reject. force never |
| 233 | // overrides it; the check runs again just before the pointer write. |
| 234 | if p.expired(db.Source, dbBuild, p.now()) { |
| 235 | return geoip.Pointer{}, refused("the database is already expired (GeoLite2, 30 days after build)") |
| 236 | } |
| 237 | if o.RequireType != "" && db.Source != o.RequireType { |
| 238 | return geoip.Pointer{}, refused("the feed returned a " + db.Source + " database, not " + o.RequireType) |
| 239 | } |
| 240 | cur, etag, present, curErr := st.cur, st.curEtag, st.curPresent, st.curErr |
| 241 | if o.Monotonic && present && curErr == nil && cur.Origin == o.Origin && !dbBuild.After(cur.Build()) { |
| 242 | return geoip.Pointer{}, refused("not newer than the published edition") |
| 243 | } |
| 244 | switch { |
| 245 | case curErr != nil: |
| 246 | // A pointer exists but does not parse: replace it (etag still guards). |
| 247 | log.Printf("console: geoip publish: the current pointer is unreadable and will be replaced") |
| 248 | case present && !o.Force && cur.DatabaseType == db.Source && db.NodeCount < cur.NodeCount/2: |
| 249 | return geoip.Pointer{}, refused("fewer than half the current database's nodes (use -force if intended)") |
| 250 | } |
| 251 | next := geoip.Pointer{V: 1, SHA256: st.sha, Key: geoip.BlobKey(st.sha), Size: st.size, |
| 252 | DatabaseType: db.Source, BuildEpoch: db.BuildEpoch, NodeCount: db.NodeCount, |
| 253 | Published: p.now().UTC().Truncate(time.Second), Origin: o.Origin, Edition: o.Edition} |
| 254 | f, err := os.Open(st.path) |
| 255 | if err != nil { |
| 256 | return geoip.Pointer{}, err |
| 257 | } |
| 258 | defer f.Close() |
| 259 | // A CAS conflict on the blob means an object already sits at this |
| 260 | // content-addressed name. It is only the same bytes if it hashes to the |
| 261 | // name: a truncated or foreign object there must never be published |
| 262 | // (spec §3 step 2, plan review R1-6). |
| 263 | if _, err := release.PutObjectFrom(ctx, p.store, next.Key, f, st.size, ""); errors.Is(err, release.ErrCASConflict) { |
| 264 | if err := p.verifyExistingBlob(ctx, next.Key, st.sha, st.size); err != nil { |
| 265 | return geoip.Pointer{}, err |
| 266 | } |
| 267 | } else if err != nil { |
| 268 | return geoip.Pointer{}, err |
| 269 | } |
| 270 | body, err := marshalPointer(next) |
| 271 | if err != nil { |
| 272 | return geoip.Pointer{}, err |
| 273 | } |
| 274 | if len(body) > geoip.MaxPointerBytes { |
| 275 | // Every reader caps the pointer at this size: one larger would |
| 276 | // publish a database no edge can ever read (code review R7-1). |
| 277 | return geoip.Pointer{}, refused("the pointer would be larger than readers accept") |
| 278 | } |
| 279 | if !present { |
| 280 | etag = "" |
| 281 | } |
| 282 | if p.expired(next.DatabaseType, next.Build(), p.now()) { |
| 283 | return geoip.Pointer{}, refused("the database expired while it was being published (GeoLite2, 30 days after build)") |
| 284 | } |
| 285 | if _, err := p.store.Put(ctx, geoip.PointerKey, body, etag); err != nil { |
| 286 | return geoip.Pointer{}, fmt.Errorf("pointer write: %w", err) |
| 287 | } |
| 288 | if err := p.sweepLocked(ctx, next.Key); err != nil { |
| 289 | log.Printf("console: geoip sweep after publish: %v", err) |
| 290 | } |
| 291 | log.Printf("console: geoip published %s (%s, %s)", st.sha[:8], next.DatabaseType, next.Origin) |
| 292 | return next, nil |
| 293 | } |
| 294 | |
| 295 | // copyCapped streams src into path, hashing, refusing more than max bytes |
| 296 | // as it copies (codex R1-7). |
| 297 | func copyCapped(ctx context.Context, path string, src io.Reader, max int64) (string, int64, error) { |
| 298 | f, err := os.OpenFile(path, os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) |
| 299 | if err != nil { |
| 300 | return "", 0, err |
| 301 | } |
| 302 | defer f.Close() |
| 303 | h := sha256.New() |
| 304 | n, err := io.Copy(io.MultiWriter(f, h), io.LimitReader(geoipCtxReader{ctx, src}, max+1)) |
| 305 | if err != nil { |
| 306 | return "", 0, err |
| 307 | } |
| 308 | if n > max { |
| 309 | return "", 0, refused(fmt.Sprintf("larger than %d bytes", max)) |
| 310 | } |
| 311 | if n == 0 { |
| 312 | return "", 0, refused("empty upload") |
| 313 | } |
| 314 | if err := f.Sync(); err != nil { |
| 315 | return "", 0, err |
| 316 | } |
| 317 | return hex.EncodeToString(h.Sum(nil)), n, nil |
| 318 | } |
| 319 | |
| 320 | type geoipCtxReader struct { |
| 321 | ctx context.Context |
| 322 | r io.Reader |
| 323 | } |
| 324 | |
| 325 | func (c geoipCtxReader) Read(b []byte) (int, error) { |
| 326 | if err := c.ctx.Err(); err != nil { |
| 327 | return 0, err |
| 328 | } |
| 329 | return c.r.Read(b) |
| 330 | } |
| 331 | |
| 332 | // verifyExistingBlob streams the object at key and checks it is exactly |
| 333 | // sha/size. Rare (a republish of the same file), so the read is acceptable. |
| 334 | func (p *geoipPublisher) verifyExistingBlob(ctx context.Context, key, sha string, size int64) error { |
| 335 | rc, _, err := release.OpenObject(ctx, p.store, key) |
| 336 | if err != nil { |
| 337 | return fmt.Errorf("existing blob unreadable: %w", err) |
| 338 | } |
| 339 | defer rc.Close() |
| 340 | h := sha256.New() |
| 341 | n, err := io.Copy(h, io.LimitReader(rc, size+1)) |
| 342 | if err != nil { |
| 343 | return fmt.Errorf("existing blob unreadable: %w", err) |
| 344 | } |
| 345 | if n != size || hex.EncodeToString(h.Sum(nil)) != sha { |
| 346 | return refused("a different object already sits at this database's blob key") |
| 347 | } |
| 348 | return nil |
| 349 | } |
| 350 | |
| 351 | // Off deletes the pointer, then every blob (spec §3). Unconditional delete: |
| 352 | // safe under the single-writer mutex (codex R2-6). |
| 353 | func (p *geoipPublisher) Off(ctx context.Context) error { |
| 354 | // Never waits behind a publish (codex R9-2): a slow upload or job can |
| 355 | // hold the slot for minutes, longer than any caller's timeout. The |
| 356 | // operator gets a 409 and runs off again when the publish finishes. |
| 357 | if !p.mu.TryLock() { |
| 358 | return errGeoipBusy |
| 359 | } |
| 360 | defer p.mu.Unlock() |
| 361 | ctx, cancel := context.WithTimeout(ctx, 2*time.Minute) |
| 362 | defer cancel() |
| 363 | if err := p.store.Delete(ctx, geoip.PointerKey); err != nil && !errors.Is(err, os.ErrNotExist) { |
| 364 | return err |
| 365 | } |
| 366 | return p.sweepLocked(ctx, "") |
| 367 | } |
| 368 | |
| 369 | // Sweep deletes every blob the pointer does not name (orphans from a crash |
| 370 | // between blob upload and pointer write). Run at startup and daily. |
| 371 | func (p *geoipPublisher) Sweep(ctx context.Context) error { |
| 372 | p.mu.Lock() |
| 373 | defer p.mu.Unlock() |
| 374 | cur, _, present, err := p.Current(ctx) |
| 375 | if err != nil { |
| 376 | return err |
| 377 | } |
| 378 | keep := "" |
| 379 | if present && !cur.Expired { |
| 380 | keep = cur.Key |
| 381 | } |
| 382 | return p.sweepLocked(ctx, keep) |
| 383 | } |
| 384 | |
| 385 | func (p *geoipPublisher) sweepLocked(ctx context.Context, keep string) error { |
| 386 | keys, err := p.store.List(ctx, geoip.Prefix) |
| 387 | if err != nil { |
| 388 | return err |
| 389 | } |
| 390 | var firstErr error |
| 391 | for _, k := range keys { |
| 392 | if !strings.HasSuffix(k, ".mmdb") || k == keep { |
| 393 | continue |
| 394 | } |
| 395 | if err := p.store.Delete(ctx, k); err != nil && !errors.Is(err, os.ErrNotExist) && firstErr == nil { |
| 396 | firstErr = err |
| 397 | } |
| 398 | } |
| 399 | return firstErr |
| 400 | } |
| 401 | |
| 402 | // ExpireIfDue applies D1 to the bucket: a GeoLite2 pointer past expiry is |
| 403 | // rewritten with expired=true and its blob destroyed. |
| 404 | func (p *geoipPublisher) ExpireIfDue(ctx context.Context, now time.Time) (bool, error) { |
| 405 | p.mu.Lock() |
| 406 | defer p.mu.Unlock() |
| 407 | cur, etag, present, err := p.Current(ctx) |
| 408 | if err != nil || !present || cur.Expired || !geoip.Expired(cur.DatabaseType, cur.Build(), now) { |
| 409 | return false, err |
| 410 | } |
| 411 | cur.Expired = true |
| 412 | body, err := marshalPointer(cur) |
| 413 | if err != nil { |
| 414 | return false, err |
| 415 | } |
| 416 | if _, err := p.store.Put(ctx, geoip.PointerKey, body, etag); err != nil { |
| 417 | return false, err |
| 418 | } |
| 419 | log.Printf("console: geoip database %s expired (GeoLite2, 30 days after build) and was removed", cur.SHA256[:8]) |
| 420 | return true, p.sweepLocked(ctx, "") |
| 421 | } |
| 422 | |
amadan