| 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 | "mime/multipart" |
| 13 | "net/http" |
| 14 | netmail "net/mail" |
| 15 | "os" |
| 16 | "time" |
| 17 | |
| 18 | "github.com/carlosframework/platform/internal/hostclaim" |
| 19 | "github.com/carlosframework/platform/internal/provision" |
| 20 | "github.com/carlosframework/platform/internal/registry" |
| 21 | "github.com/carlosframework/platform/internal/release" |
| 22 | "github.com/carlosframework/platform/internal/runs" |
| 23 | "github.com/carlosframework/platform/internal/svcrole" |
| 24 | ) |
| 25 | |
| 26 | // The run handlers' refusal codes (service runs spec §3): every refusal a |
| 27 | // run handler writes carries one beside its sentence, so a caller with no |
| 28 | // person reading the answer can act on it. The shared gates before the |
| 29 | // handler (not signed in, the role mismatch, the membership 404) keep their |
| 30 | // bodies exactly as every other door gives them. |
| 31 | const ( |
| 32 | runCodeNotFound = "not_found" |
| 33 | runCodeHostClaim = "host_claim" |
| 34 | runCodeRateLimited = "rate_limited" |
| 35 | runCodeUnavailable = "unavailable" |
| 36 | runCodeRecordFailed = "record_failed" |
| 37 | runCodeReadFailed = "read_failed" |
| 38 | runCodeNotRunnable = "not_runnable" |
| 39 | runCodeMoving = "moving" |
| 40 | runCodeRunOpen = "run_open" |
| 41 | runCodeMalformed = "malformed" |
| 42 | runCodeInvalidRequest = "invalid_request" |
| 43 | runCodeFilesTooLarge = "files_too_large" |
| 44 | runCodeImportUncommitted = "import_uncommitted" |
| 45 | runCodeCommandNotGranted = "command_not_granted" |
| 46 | runCodeOnBehalfOfInvalid = "on_behalf_of_invalid" |
| 47 | runCodeRunStarted = "run_started" |
| 48 | runCodeRunEnded = "run_ended" |
| 49 | runCodeRunNotFinished = "run_not_finished" |
| 50 | runCodeLastSeqMismatch = "last_seq_mismatch" |
| 51 | runCodeKeyInvalid = "idempotency_key_invalid" |
| 52 | runCodeKeyReused = "idempotency_key_reused" |
| 53 | runCodeKeysNotReady = "keys_not_ready" |
| 54 | runCodeFileHashMismatch = "file_hash_mismatch" |
| 55 | ) |
| 56 | |
| 57 | // runRefusalCodes is every code above, the one list the docs and the tests |
| 58 | // read. |
| 59 | var runRefusalCodes = []string{ |
| 60 | runCodeNotFound, runCodeHostClaim, runCodeRateLimited, runCodeUnavailable, |
| 61 | runCodeRecordFailed, runCodeReadFailed, runCodeNotRunnable, runCodeMoving, |
| 62 | runCodeRunOpen, runCodeMalformed, runCodeInvalidRequest, runCodeFilesTooLarge, |
| 63 | runCodeImportUncommitted, runCodeCommandNotGranted, runCodeOnBehalfOfInvalid, |
| 64 | runCodeRunStarted, runCodeRunEnded, runCodeRunNotFinished, runCodeLastSeqMismatch, |
| 65 | runCodeKeyInvalid, runCodeKeyReused, runCodeKeysNotReady, runCodeFileHashMismatch, |
| 66 | } |
| 67 | |
| 68 | // runCaller reports whether p may use the run doors at all: a person (whose |
| 69 | // membership the handler checks next), or a run service with a credential |
| 70 | // id to own its runs by. |
| 71 | func runCaller(p Principal) bool { |
| 72 | switch p.Kind { |
| 73 | case PrincipalUser: |
| 74 | return true |
| 75 | case PrincipalService: |
| 76 | return p.Role == svcrole.Run && p.CredentialID != "" |
| 77 | } |
| 78 | return false |
| 79 | } |
| 80 | |
| 81 | // callerScope names a caller for every comparison between callers: the |
| 82 | // principal's KIND and its stable identity, so a person's address can never |
| 83 | // equal a credential's id. "" for a principal with no identity, which |
| 84 | // matches nothing. |
| 85 | func callerScope(p Principal) string { |
| 86 | switch { |
| 87 | case p.Kind == PrincipalService && p.CredentialID != "": |
| 88 | return "svc:" + p.CredentialID |
| 89 | case p.Kind == PrincipalUser && p.Address != "": |
| 90 | return "user:" + p.Address |
| 91 | } |
| 92 | return "" |
| 93 | } |
| 94 | |
| 95 | // runVisibleTo reports whether p may see and act on req. A person sees every |
| 96 | // run on an account they are a member of (the gate has checked that); a |
| 97 | // service sees only the runs its own credential started, by the credential's |
| 98 | // id, never its name, so a name minted again after a revoke inherits |
| 99 | // nothing. |
| 100 | func (s *Server) runVisibleTo(p Principal, req runs.Request) bool { |
| 101 | switch p.Kind { |
| 102 | case PrincipalUser: |
| 103 | return true |
| 104 | case PrincipalService: |
| 105 | return req.ByCredential != "" && callerScope(p) == "svc:"+req.ByCredential |
| 106 | } |
| 107 | return false |
| 108 | } |
| 109 | |
| 110 | // maxOnBehalfOf is the longest on_behalf_of an address may be. |
| 111 | const maxOnBehalfOf = 254 |
| 112 | |
| 113 | const onBehalfOfPersonMsg = "on_behalf_of is for service credentials. a person runs as themselves" |
| 114 | |
| 115 | const onBehalfOfShapeMsg = "on_behalf_of must be an email address of at most 254 bytes" |
| 116 | |
| 117 | // onBehalfOfRefusal is "" when meta's on_behalf_of may stand: empty, or an |
| 118 | // address of the right shape sent by a run service. Its shape is checked, |
| 119 | // never its membership: it is recorded on the service's word and grants |
| 120 | // nothing. |
| 121 | func onBehalfOfRefusal(p Principal, v string) string { |
| 122 | switch { |
| 123 | case v == "": |
| 124 | return "" |
| 125 | case p.Kind != PrincipalService: |
| 126 | return onBehalfOfPersonMsg |
| 127 | case len(v) > maxOnBehalfOf || !ValidAddress(v): |
| 128 | return onBehalfOfShapeMsg |
| 129 | } |
| 130 | if parsed, err := netmail.ParseAddress(v); err != nil || parsed.Address != v || parsed.Name != "" { |
| 131 | return onBehalfOfShapeMsg |
| 132 | } |
| 133 | return "" |
| 134 | } |
| 135 | |
| 136 | // claimRunUnsupportedMsg is every run route's answer on a host-claim |
| 137 | // console (index item 12, RUN-C42). |
| 138 | const claimRunUnsupportedMsg = "carlos run is not available on this console yet. it needs the host-claim launch protocol" |
| 139 | |
| 140 | const runTooManyMsg = "too many run requests. wait a minute and try again" |
| 141 | |
| 142 | // runOpenMsg is refusal 7 (RUN-C24), at the door and at admission. req is |
| 143 | // the open run's request when it could be read: a service's run is named in |
| 144 | // the house wording (runs.Request.ServiceBy), never by its audit address. |
| 145 | func runOpenMsg(host string, run provision.RunState, req *runs.Request) string { |
| 146 | if req != nil { |
| 147 | if by := req.ServiceBy(); by != "" { |
| 148 | return fmt.Sprintf("%s already has a run open: %s, started %s", host, run.ID, by) |
| 149 | } |
| 150 | } |
| 151 | return fmt.Sprintf("%s already has a run open: %s, started by %s", host, run.ID, run.By) |
| 152 | } |
| 153 | |
| 154 | // runOpenRedactedMsg is refusal 7 for a service whose open run is not its |
| 155 | // own: neither that run's id nor who started it. |
| 156 | func runOpenRedactedMsg(host string) string { |
| 157 | return fmt.Sprintf("%s already has a run open. try again once it has finished", host) |
| 158 | } |
| 159 | |
| 160 | // runOpenRefusal is refusal 7 as p may read it. A service is told the open |
| 161 | // run's id and requester only when the run is its own; a run whose request |
| 162 | // cannot be read is not shown to be its own. |
| 163 | func (s *Server) runOpenRefusal(ctx context.Context, p Principal, sqid, app, host string, run provision.RunState) string { |
| 164 | req, err := runs.ReadRequest(ctx, s.cfg.Releases, sqid, app, run.ID) |
| 165 | if err != nil { |
| 166 | req = nil |
| 167 | } |
| 168 | if p.Kind == PrincipalUser { |
| 169 | return runOpenMsg(host, run, req) |
| 170 | } |
| 171 | if req != nil && s.runVisibleTo(p, *req) { |
| 172 | return runOpenMsg(host, run, req) |
| 173 | } |
| 174 | return runOpenRedactedMsg(host) |
| 175 | } |
| 176 | |
| 177 | // runMovingMsg is refusal 6 (RUN-C23), at the door and at admission. |
| 178 | func runMovingMsg(host string) string { |
| 179 | return fmt.Sprintf("%s is moving. run the command once the move has finished", host) |
| 180 | } |
| 181 | |
| 182 | // runImportMsg is refusal 9 (RUN-C33), at the door and at admission. |
| 183 | func runImportMsg(host string) string { |
| 184 | return fmt.Sprintf("%s is an import that is not committed yet. finish or abort the import first", host) |
| 185 | } |
| 186 | |
| 187 | // runFilesTooLargeMsg is refusal 8 for the file total (RUN-C30), whether meta |
| 188 | // declares it or the body grows past it. |
| 189 | const runFilesTooLargeMsg = "the files come to more than 64 MB" |
| 190 | |
| 191 | const runMalformedMsg = "the run request is malformed" |
| 192 | |
| 193 | const runRecordFailedMsg = "could not record the run. try again" |
| 194 | |
| 195 | const runCommandNotGrantedMsg = "this credential may not run that command. it runs only the commands it was minted with" |
| 196 | |
| 197 | // runKeyHeader is the name of a create's idempotency key (service runs spec |
| 198 | // § 3). |
| 199 | const runKeyHeader = "Idempotency-Key" |
| 200 | |
| 201 | const runKeysNotReadyMsg = "this console does not take an Idempotency-Key on runs yet. try again later" |
| 202 | |
| 203 | const runKeyInvalidMsg = "an Idempotency-Key is 1 to 128 characters: letters, digits, dots, underscores, colons and hyphens" |
| 204 | |
| 205 | const runKeyReusedMsg = "this Idempotency-Key was used for a different run request. use a new key for a new request" |
| 206 | |
| 207 | const runFileHashRequiredMsg = "a run created with an Idempotency-Key must declare the sha256 of each file" |
| 208 | |
| 209 | func runFileHashShapeMsg(name string) string { |
| 210 | return fmt.Sprintf("the sha256 declared for %s is not 64 hex digits", name) |
| 211 | } |
| 212 | |
| 213 | func runFileHashMismatchMsg(name string) string { |
| 214 | return fmt.Sprintf("%s does not match the sha256 declared for it", name) |
| 215 | } |
| 216 | |
| 217 | func runNoKeyMsg(host string) string { |
| 218 | return fmt.Sprintf("%s has no run started with that Idempotency-Key", host) |
| 219 | } |
| 220 | |
| 221 | // runKeyOf reads the create's Idempotency-Key. sent reports whether the |
| 222 | // header was there at all, so an empty, repeated or malformed key is |
| 223 | // refused (ok false), never taken for an unkeyed create. |
| 224 | func runKeyOf(r *http.Request) (key string, sent, ok bool) { |
| 225 | vals := r.Header.Values(runKeyHeader) |
| 226 | switch { |
| 227 | case len(vals) == 0: |
| 228 | return "", false, false |
| 229 | case len(vals) != 1 || !runs.ValidKey(vals[0]): |
| 230 | return "", true, false |
| 231 | } |
| 232 | return vals[0], true, true |
| 233 | } |
| 234 | |
| 235 | // runFileHashRefusal is "" when meta's declared sha256s may stand: each one |
| 236 | // declared is 64 hex digits, and a keyed create declares them all. |
| 237 | func runFileHashRefusal(m runs.Meta, keyed bool) string { |
| 238 | for _, f := range m.Files { |
| 239 | switch { |
| 240 | case f.SHA256 == "" && keyed: |
| 241 | return runFileHashRequiredMsg |
| 242 | case f.SHA256 != "" && !runs.ValidFileHash(f.SHA256): |
| 243 | return runFileHashShapeMsg(f.Name) |
| 244 | } |
| 245 | } |
| 246 | return "" |
| 247 | } |
| 248 | |
| 249 | // runReplayed answers a create whose key already admitted run id: 202 with |
| 250 | // that run, replayed, and its state while the record still names it. |
| 251 | func runReplayed(w http.ResponseWriter, rec *provision.Record, host, id string) { |
| 252 | state := "" |
| 253 | if rec != nil && rec.ActiveRun != nil && rec.ActiveRun.ID == id { |
| 254 | state = rec.ActiveRun.State |
| 255 | } |
| 256 | writeJSON(w, http.StatusAccepted, runs.Created{ID: id, Host: host, State: state, Replayed: true}) |
| 257 | } |
| 258 | |
| 259 | // runMetaRefusalCode is the code for a refusal runMetaRefusal named. |
| 260 | func runMetaRefusalCode(msg string) string { |
| 261 | if msg == runFilesTooLargeMsg { |
| 262 | return runCodeFilesTooLarge |
| 263 | } |
| 264 | return runCodeInvalidRequest |
| 265 | } |
| 266 | |
| 267 | // runClassRefusal is spec §5.1 refusal 5 as a sentence, "" for a |
| 268 | // platform-run, hibernating exec instance on a platform box. |
| 269 | func runClassRefusal(rec provision.Record) string { |
| 270 | switch { |
| 271 | case rec.Kind != "" && rec.Kind != "instance": |
| 272 | return fmt.Sprintf("%s is a %s route. carlos run works on instances", rec.Host, rec.Kind) |
| 273 | case rec.Addr != "" || len(rec.Addrs) > 0: |
| 274 | return fmt.Sprintf("%s is served from its own address, so the platform does not run its process", rec.Host) |
| 275 | case !rec.HibernateOrDefault(): |
| 276 | return fmt.Sprintf("%s does not sleep, so the platform does not run its process", rec.Host) |
| 277 | case rec.Pool != "": |
| 278 | return fmt.Sprintf("%s is a pooled instance. carlos run works on instances with a process of their own", rec.Host) |
| 279 | } |
| 280 | if fleet, ok := registry.FleetPlacement(rec.Region); ok { |
| 281 | return fmt.Sprintf("%s runs on your fleet %s. carlos run works on platform boxes only", rec.Host, fleet) |
| 282 | } |
| 283 | return "" |
| 284 | } |
| 285 | |
| 286 | // runMetaRefusal is refusal 8 as a sentence, "" when meta is valid. |
| 287 | func runMetaRefusal(m runs.Meta) string { |
| 288 | switch err := runs.CheckArgv(m.Argv); { |
| 289 | case errors.Is(err, runs.ErrNoArgs): |
| 290 | return "name the command to run after --" |
| 291 | case errors.Is(err, runs.ErrTooManyArgs): |
| 292 | return "a run takes at most 256 arguments" |
| 293 | case errors.Is(err, runs.ErrArgvTooLarge): |
| 294 | return "the arguments come to more than 64 KB" |
| 295 | case errors.Is(err, runs.ErrArgNUL): |
| 296 | return "an argument holds a NUL byte" |
| 297 | } |
| 298 | if runs.CheckTimeout(time.Duration(m.TimeoutS)*time.Second) != nil { |
| 299 | return "the timeout must be between 1 second and 6 hours" |
| 300 | } |
| 301 | err := runs.CheckFiles(m.Files) |
| 302 | var fe *runs.FileNameError |
| 303 | switch { |
| 304 | case errors.Is(err, runs.ErrFilesTooLarge): |
| 305 | return runFilesTooLargeMsg |
| 306 | case errors.As(err, &fe) && fe.Twice: |
| 307 | return fmt.Sprintf("two files are both named %s", fe.Name) |
| 308 | case errors.As(err, &fe): |
| 309 | return fmt.Sprintf("%s is not a plain file name", fe.Name) |
| 310 | } |
| 311 | return "" |
| 312 | } |
| 313 | |
| 314 | // spooled is one uploaded file on the console's disk. |
| 315 | type spooled struct { |
| 316 | runs.File |
| 317 | path string |
| 318 | } |
| 319 | |
| 320 | // spoolRunFiles reads the file parts, one per meta file in meta's order, |
| 321 | // into temp files, checking each name and size and hashing it. The caller |
| 322 | // removes the files. A short, long, misnamed or missing part is |
| 323 | // malformed; a body past the cap is the 64 MB refusal. |
| 324 | func spoolRunFiles(mr *multipart.Reader, files []runs.MetaFile) ([]spooled, error) { |
| 325 | var out []spooled |
| 326 | fail := func(err error) ([]spooled, error) { |
| 327 | for _, f := range out { |
| 328 | os.Remove(f.path) |
| 329 | } |
| 330 | return nil, err |
| 331 | } |
| 332 | for _, want := range files { |
| 333 | part, err := mr.NextPart() |
| 334 | if err != nil { |
| 335 | return fail(err) |
| 336 | } |
| 337 | if part.FormName() != "file" || part.FileName() != want.Name { |
| 338 | return fail(errRunMalformed) |
| 339 | } |
| 340 | tmp, err := os.CreateTemp("", "carlos-run-*") |
| 341 | if err != nil { |
| 342 | return fail(err) |
| 343 | } |
| 344 | h := sha256.New() |
| 345 | n, err := io.Copy(io.MultiWriter(tmp, h), io.LimitReader(part, want.Size+1)) |
| 346 | cerr := tmp.Close() |
| 347 | out = append(out, spooled{File: runs.File{Name: want.Name, Size: n, SHA256: hex.EncodeToString(h.Sum(nil))}, path: tmp.Name()}) |
| 348 | if err != nil { |
| 349 | return fail(err) |
| 350 | } |
| 351 | if cerr != nil { |
| 352 | return fail(cerr) |
| 353 | } |
| 354 | if n != want.Size { |
| 355 | return fail(errRunMalformed) |
| 356 | } |
| 357 | } |
| 358 | if _, err := mr.NextPart(); err != io.EOF { |
| 359 | return fail(errRunMalformed) |
| 360 | } |
| 361 | return out, nil |
| 362 | } |
| 363 | |
| 364 | var errRunMalformed = errors.New("console: run request disagrees with its meta") |
| 365 | |
| 366 | // handleAPIRunCreate is POST …/instances/{host}/runs (run spec §5.1). The |
| 367 | // refusals come in the spec's order; nothing is written before every one |
| 368 | // that needs no body has passed. |
| 369 | // |
| 370 | // A create that carries an Idempotency-Key (service runs spec § 3) reads its |
| 371 | // meta straight after the record, and answers a key the record already |
| 372 | // holds before any other refusal and without reading a file: the same |
| 373 | // request is the run that key admitted, a different one is refused. Only |
| 374 | // then come the record's refusals, the command check and the files, whose |
| 375 | // bytes must match the sha256 meta declares. The admission records the key |
| 376 | // in the same write. An unkeyed create keeps the order it had before keys. |
| 377 | func (s *Server) handleAPIRunCreate(w http.ResponseWriter, r *http.Request) { |
| 378 | p, ok := principalFrom(r.Context()) |
| 379 | if !ok || !runCaller(p) { |
| 380 | http.NotFound(w, r) |
| 381 | return |
| 382 | } |
| 383 | actor, acct, app, ok := s.apiAccountApp(w, r) |
| 384 | if !ok { |
| 385 | return |
| 386 | } |
| 387 | if s.cfg.HostClaim { |
| 388 | jsonErrorCode(w, http.StatusConflict, runCodeHostClaim, claimRunUnsupportedMsg) |
| 389 | return |
| 390 | } |
| 391 | if !s.cli.instances.allow(hashToken(bearer(r)), time.Now()) { |
| 392 | jsonErrorCode(w, http.StatusTooManyRequests, runCodeRateLimited, runTooManyMsg) |
| 393 | return |
| 394 | } |
| 395 | key, keyed, keyOK := runKeyOf(r) |
| 396 | if keyed && !s.cfg.RunKeysEnabled { |
| 397 | // Never run as unkeyed: the caller would retry it into a second run. |
| 398 | jsonErrorCode(w, http.StatusServiceUnavailable, runCodeKeysNotReady, runKeysNotReadyMsg) |
| 399 | return |
| 400 | } |
| 401 | scope := callerScope(p) |
| 402 | if keyed && (!keyOK || scope == "") { |
| 403 | jsonErrorCode(w, http.StatusBadRequest, runCodeKeyInvalid, runKeyInvalidMsg) |
| 404 | return |
| 405 | } |
| 406 | if s.cfg.Releases == nil { |
| 407 | jsonErrorCode(w, http.StatusServiceUnavailable, runCodeUnavailable, runRecordFailedMsg) |
| 408 | return |
| 409 | } |
| 410 | ctx := r.Context() |
| 411 | host := r.PathValue("host") |
| 412 | notInstance := fmt.Sprintf("%s is not an instance of %s", host, app) |
| 413 | if !provision.ValidHostShape(host) { |
| 414 | jsonErrorCode(w, http.StatusNotFound, runCodeNotFound, notInstance) |
| 415 | return |
| 416 | } |
| 417 | rec, err := provision.GetRecord(ctx, s.cfg.Releases, acct.Sqid, app, host) |
| 418 | if errors.Is(err, os.ErrNotExist) { |
| 419 | jsonErrorCode(w, http.StatusNotFound, runCodeNotFound, notInstance) |
| 420 | return |
| 421 | } |
| 422 | if err != nil { |
| 423 | log.Printf("console: api run %s/%s/%s: read record: %v", acct.Sqid, app, host, err) |
| 424 | jsonErrorCode(w, http.StatusBadGateway, runCodeRecordFailed, runRecordFailedMsg) |
| 425 | return |
| 426 | } |
| 427 | // recordRefused writes refusals 5 to 7 from the record (the class, a move |
| 428 | // in flight, an open run) and reports whether it wrote one. |
| 429 | recordRefused := func() bool { |
| 430 | if msg := runClassRefusal(*rec); msg != "" { |
| 431 | jsonErrorCode(w, http.StatusConflict, runCodeNotRunnable, msg) |
| 432 | return true |
| 433 | } |
| 434 | if rec.Move != nil { |
| 435 | jsonErrorCode(w, http.StatusConflict, runCodeMoving, runMovingMsg(host)) |
| 436 | return true |
| 437 | } |
| 438 | if rec.ActiveRun.Open() { |
| 439 | jsonErrorCode(w, http.StatusConflict, runCodeRunOpen, s.runOpenRefusal(ctx, p, acct.Sqid, app, host, *rec.ActiveRun)) |
| 440 | return true |
| 441 | } |
| 442 | return false |
| 443 | } |
| 444 | if !keyed && recordRefused() { |
| 445 | return |
| 446 | } |
| 447 | r.Body = http.MaxBytesReader(w, r.Body, runs.MaxFileBytes+runs.MetaAllowance) |
| 448 | mr, err := r.MultipartReader() |
| 449 | if err != nil { |
| 450 | jsonErrorCode(w, http.StatusBadRequest, runCodeMalformed, runMalformedMsg) |
| 451 | return |
| 452 | } |
| 453 | var meta runs.Meta |
| 454 | part, err := mr.NextPart() |
| 455 | if err == nil && part.FormName() == "meta" { |
| 456 | err = json.NewDecoder(io.LimitReader(part, runs.MetaAllowance)).Decode(&meta) |
| 457 | } else if err == nil { |
| 458 | err = errRunMalformed |
| 459 | } |
| 460 | if err != nil { |
| 461 | jsonErrorCode(w, http.StatusBadRequest, runCodeMalformed, runMalformedMsg) |
| 462 | return |
| 463 | } |
| 464 | // A declared sha256 is compared and digested in the lowercase form the |
| 465 | // spool computes, so either case declares the same file. |
| 466 | for i := range meta.Files { |
| 467 | meta.Files[i].SHA256 = runs.NormalFileHash(meta.Files[i].SHA256) |
| 468 | } |
| 469 | if msg := runMetaRefusal(meta); msg != "" { |
| 470 | jsonErrorCode(w, http.StatusBadRequest, runMetaRefusalCode(msg), msg) |
| 471 | return |
| 472 | } |
| 473 | if msg := onBehalfOfRefusal(p, meta.OnBehalfOf); msg != "" { |
| 474 | jsonErrorCode(w, http.StatusBadRequest, runCodeOnBehalfOfInvalid, msg) |
| 475 | return |
| 476 | } |
| 477 | if msg := runFileHashRefusal(meta, keyed); msg != "" { |
| 478 | jsonErrorCode(w, http.StatusBadRequest, runCodeInvalidRequest, msg) |
| 479 | return |
| 480 | } |
| 481 | var scoped, digest, caller string |
| 482 | if keyed { |
| 483 | scoped, digest, caller = runs.ScopedKey(scope, key), runs.MetaDigest(meta), runs.CallerOf(scope) |
| 484 | if k, ok := rec.RunKeyFor(scoped, time.Now()); ok { |
| 485 | if k.Digest != digest { |
| 486 | jsonErrorCode(w, http.StatusConflict, runCodeKeyReused, runKeyReusedMsg) |
| 487 | return |
| 488 | } |
| 489 | // The files are never read: the run that key admitted has them. |
| 490 | runReplayed(w, rec, host, k.RunID) |
| 491 | return |
| 492 | } |
| 493 | if recordRefused() { |
| 494 | return |
| 495 | } |
| 496 | if rec.ImportUncommitted() { |
| 497 | jsonErrorCode(w, http.StatusConflict, runCodeImportUncommitted, runImportMsg(host)) |
| 498 | return |
| 499 | } |
| 500 | } |
| 501 | // A run service runs only what it was granted, refused before any |
| 502 | // object is written. A person runs any command. |
| 503 | if p.Kind == PrincipalService && !runs.CommandGranted(meta.Argv, p.Commands) { |
| 504 | jsonErrorCode(w, http.StatusForbidden, runCodeCommandNotGranted, runCommandNotGrantedMsg) |
| 505 | return |
| 506 | } |
| 507 | if !keyed && rec.ImportUncommitted() { |
| 508 | jsonErrorCode(w, http.StatusConflict, runCodeImportUncommitted, runImportMsg(host)) |
| 509 | return |
| 510 | } |
| 511 | files, err := spoolRunFiles(mr, meta.Files) |
| 512 | defer func() { |
| 513 | for _, f := range files { |
| 514 | os.Remove(f.path) |
| 515 | } |
| 516 | }() |
| 517 | var tooLarge *http.MaxBytesError |
| 518 | switch { |
| 519 | case errors.As(err, &tooLarge): |
| 520 | jsonErrorCode(w, http.StatusRequestEntityTooLarge, runCodeFilesTooLarge, runFilesTooLargeMsg) |
| 521 | return |
| 522 | case err != nil: |
| 523 | jsonErrorCode(w, http.StatusBadRequest, runCodeMalformed, runMalformedMsg) |
| 524 | return |
| 525 | } |
| 526 | for i, f := range files { |
| 527 | if want := meta.Files[i].SHA256; want != "" && f.SHA256 != want { |
| 528 | jsonErrorCode(w, http.StatusBadRequest, runCodeFileHashMismatch, runFileHashMismatchMsg(f.Name)) |
| 529 | return |
| 530 | } |
| 531 | } |
| 532 | id, err := runs.MintID() |
| 533 | if err != nil { |
| 534 | jsonErrorCode(w, http.StatusBadGateway, runCodeRecordFailed, runRecordFailedMsg) |
| 535 | return |
| 536 | } |
| 537 | inc, err := hostclaim.Incarnation(rec.Account, rec.App, rec.Host, rec.CreatedAt) |
| 538 | if err != nil { |
| 539 | jsonErrorCode(w, http.StatusBadGateway, runCodeRecordFailed, runRecordFailedMsg) |
| 540 | return |
| 541 | } |
| 542 | now := time.Now().UTC() |
| 543 | req := runs.Request{Schema: 1, ID: id, Account: acct.Sqid, App: app, Host: host, Incarnation: inc, |
| 544 | Argv: meta.Argv, TimeoutS: meta.TimeoutS, By: actor, RequestedAt: now, IdempotencyKey: scoped} |
| 545 | if p.Kind == PrincipalService { |
| 546 | req.ByService, req.ByCredential, req.OnBehalfOf = p.Name, p.CredentialID, meta.OnBehalfOf |
| 547 | } |
| 548 | for _, f := range files { |
| 549 | req.Files = append(req.Files, f.File) |
| 550 | } |
| 551 | if err := s.writeRunObjects(ctx, req, files); err != nil { |
| 552 | log.Printf("console: api run %s/%s/%s: write %s: %v", acct.Sqid, app, host, id, err) |
| 553 | _ = runs.DeleteRun(context.WithoutCancel(ctx), s.cfg.Releases, acct.Sqid, app, id) |
| 554 | jsonErrorCode(w, http.StatusBadGateway, runCodeRecordFailed, runRecordFailedMsg) |
| 555 | return |
| 556 | } |
| 557 | // The pickup window opens at admission, not at the first byte: a slow |
| 558 | // upload must not eat the box's 60 seconds. |
| 559 | // |
| 560 | // The admission runs to its end even if the caller has gone: a conditional |
| 561 | // Put that is cancelled may or may not have landed, and deleting the |
| 562 | // objects of a run the record holds would leave a queued run no box can |
| 563 | // read. The outcome is then settled from the record, not from the error. |
| 564 | wctx, cancel := context.WithTimeout(context.WithoutCancel(ctx), 30*time.Second) |
| 565 | defer cancel() |
| 566 | state, err := s.admitRun(wctx, acct.Sqid, app, host, id, actor, scoped, digest, caller) |
| 567 | // twinWon answers with the run a twin (the same key and request) had |
| 568 | // admitted first. This request's own objects are no one's. |
| 569 | twinWon := func(cur *provision.Record, twin string) { |
| 570 | if derr := runs.DeleteRun(context.WithoutCancel(ctx), s.cfg.Releases, acct.Sqid, app, id); derr != nil { |
| 571 | log.Printf("console: api run %s/%s/%s: delete %s after its twin %s won: %v", acct.Sqid, app, host, id, twin, derr) |
| 572 | } |
| 573 | runReplayed(w, cur, host, twin) |
| 574 | } |
| 575 | var replay *provision.RunKeyReplay |
| 576 | if errors.As(err, &replay) { |
| 577 | // admitRun answered a replay of this request's own id as its |
| 578 | // admission, so this one names a twin. |
| 579 | cur, _ := provision.GetRecord(wctx, s.cfg.Releases, acct.Sqid, app, host) |
| 580 | twinWon(cur, replay.RunID) |
| 581 | return |
| 582 | } |
| 583 | if err != nil && !runAdmitRefused(err) { |
| 584 | // The outcome is unknown (a lost response, a cancelled request, a |
| 585 | // transport error after the write, or both CAS attempts lost): ask |
| 586 | // the record whose run it holds. |
| 587 | cur, rerr := provision.GetRecord(wctx, s.cfg.Releases, acct.Sqid, app, host) |
| 588 | switch { |
| 589 | case rerr != nil: |
| 590 | // Still unknown. Keep the objects: if the run was admitted a box |
| 591 | // needs them, and if not, the orphan sweep ages them out by |
| 592 | // request.json (Task 5). |
| 593 | log.Printf("console: api run %s/%s/%s: admit %s: %v; record unreadable: %v", acct.Sqid, app, host, id, err, rerr) |
| 594 | jsonErrorCode(w, http.StatusBadGateway, runCodeRecordFailed, runRecordFailedMsg) |
| 595 | return |
| 596 | case keyed: |
| 597 | // A keyed admission is settled by its key, which commits with it |
| 598 | // or not at all, never by ActiveRun, which a box or an ack may |
| 599 | // already have moved on: a committed run keeps its objects. |
| 600 | k, found := cur.RunKeyFor(scoped, time.Now()) |
| 601 | switch { |
| 602 | case found && k.RunID == id: |
| 603 | state, err = "", nil |
| 604 | if cur.ActiveRun != nil && cur.ActiveRun.ID == id { |
| 605 | state = cur.ActiveRun.State |
| 606 | } |
| 607 | case found && k.Digest == digest: |
| 608 | twinWon(cur, k.RunID) |
| 609 | return |
| 610 | case found: |
| 611 | err = provision.ErrRunKeyReused |
| 612 | case cur.ActiveRun != nil && cur.ActiveRun.ID == id: |
| 613 | // The key is gone (a record writer dropped run_keys between |
| 614 | // the admit and now) but the record still holds this run: |
| 615 | // it is admitted, and deleting its objects would leave a |
| 616 | // queued run no box can read. |
| 617 | state, err = cur.ActiveRun.State, nil |
| 618 | } |
| 619 | case cur.ActiveRun != nil && cur.ActiveRun.ID == id: |
| 620 | state, err = cur.ActiveRun.State, nil |
| 621 | } |
| 622 | } |
| 623 | if err != nil { |
| 624 | _ = runs.DeleteRun(context.WithoutCancel(ctx), s.cfg.Releases, acct.Sqid, app, id) |
| 625 | var open *provision.RunOpenError |
| 626 | switch { |
| 627 | case errors.As(err, &open): |
| 628 | jsonErrorCode(w, http.StatusConflict, runCodeRunOpen, s.runOpenRefusal(wctx, p, acct.Sqid, app, host, open.Run)) |
| 629 | case errors.Is(err, provision.ErrRunMoving): |
| 630 | jsonErrorCode(w, http.StatusConflict, runCodeMoving, runMovingMsg(host)) |
| 631 | case errors.Is(err, provision.ErrRunImport): |
| 632 | jsonErrorCode(w, http.StatusConflict, runCodeImportUncommitted, runImportMsg(host)) |
| 633 | case errors.Is(err, provision.ErrRunKeyReused): |
| 634 | jsonErrorCode(w, http.StatusConflict, runCodeKeyReused, runKeyReusedMsg) |
| 635 | case errors.Is(err, os.ErrNotExist): |
| 636 | jsonErrorCode(w, http.StatusNotFound, runCodeNotFound, notInstance) |
| 637 | default: |
| 638 | log.Printf("console: api run %s/%s/%s: admit %s: %v", acct.Sqid, app, host, id, err) |
| 639 | jsonErrorCode(w, http.StatusBadGateway, runCodeRecordFailed, runRecordFailedMsg) |
| 640 | } |
| 641 | return |
| 642 | } |
| 643 | // The serial bump and the run-start row belong to the admission, so they |
| 644 | // run on its context too: a caller that left after it must not cost the |
| 645 | // box its prompt pickup or the row its version. |
| 646 | s.bumpSerial(wctx) |
| 647 | s.auditActivity(acct, app, actor, Activity{Action: "run-start", TargetType: "instance", Target: host, |
| 648 | Detail: runStartDetail(req, s.runVersion(wctx, acct.Sqid, app, rec.Channel))}) |
| 649 | writeJSON(w, http.StatusAccepted, runs.Created{ID: id, Host: host, State: state}) |
| 650 | } |
| 651 | |
| 652 | // admitRun is the admission CAS, with the caller's scoped key, request |
| 653 | // digest and runs.CallerOf when the create is keyed. A RunOpenError naming id itself means an |
| 654 | // earlier attempt of this same call committed (updateRecord retries after a |
| 655 | // lost-response conflict, and finds its own run queued): that is admitted. |
| 656 | // So is a RunKeyReplay naming id, the same retry finding its own key; any |
| 657 | // other RunKeyReplay is returned for the caller to answer. |
| 658 | func (s *Server) admitRun(ctx context.Context, sqid, app, host, id, by, key, digest, caller string) (string, error) { |
| 659 | got, err := provision.SetRecordRun(ctx, s.cfg.Releases, sqid, app, host, |
| 660 | provision.RunAdmit(provision.RunAdmission{ID: id, By: by, Deadline: time.Now().UTC().Add(runs.PickupWindow), Key: key, Digest: digest, Caller: caller})) |
| 661 | var open *provision.RunOpenError |
| 662 | var replay *provision.RunKeyReplay |
| 663 | switch { |
| 664 | case err == nil: |
| 665 | return got.ActiveRun.State, nil |
| 666 | case errors.As(err, &open) && open.Run.ID == id: |
| 667 | return open.Run.State, nil |
| 668 | case errors.As(err, &replay) && replay.RunID == id: |
| 669 | // This request's own admission committed; delete nothing. The state |
| 670 | // is the record's, "" if it cannot be read now. |
| 671 | state := "" |
| 672 | if cur, rerr := provision.GetRecord(ctx, s.cfg.Releases, sqid, app, host); rerr == nil && cur.ActiveRun != nil && cur.ActiveRun.ID == id { |
| 673 | state = cur.ActiveRun.State |
| 674 | } |
| 675 | return state, nil |
| 676 | } |
| 677 | return "", err |
| 678 | } |
| 679 | |
| 680 | // runAdmitRefused reports whether err is the record saying no (another run, |
| 681 | // a move, an uncommitted import, a key reused, no record) rather than a |
| 682 | // failure whose outcome the record has to settle. |
| 683 | func runAdmitRefused(err error) bool { |
| 684 | var open *provision.RunOpenError |
| 685 | return errors.As(err, &open) || errors.Is(err, provision.ErrRunMoving) || |
| 686 | errors.Is(err, provision.ErrRunImport) || errors.Is(err, provision.ErrRunKeyReused) || |
| 687 | errors.Is(err, os.ErrNotExist) |
| 688 | } |
| 689 | |
| 690 | // handleAPIRunKeyLookup is GET …/instances/{host}/runs?key=<key> (service |
| 691 | // runs spec § 3): the run the caller's own key admitted on this instance, |
| 692 | // from the record, or 404. Another caller's key is never found, because the |
| 693 | // record holds keys scoped to their caller. It is the run doors' gate, then |
| 694 | // a read. |
| 695 | func (s *Server) handleAPIRunKeyLookup(w http.ResponseWriter, r *http.Request) { |
| 696 | p, ok := principalFrom(r.Context()) |
| 697 | if !ok || !runCaller(p) { |
| 698 | http.NotFound(w, r) |
| 699 | return |
| 700 | } |
| 701 | _, acct, app, ok := s.apiAccountApp(w, r) |
| 702 | if !ok { |
| 703 | return |
| 704 | } |
| 705 | if s.cfg.HostClaim { |
| 706 | jsonErrorCode(w, http.StatusConflict, runCodeHostClaim, claimRunUnsupportedMsg) |
| 707 | return |
| 708 | } |
| 709 | if !s.cli.runs.allow(hashToken(bearer(r)), time.Now()) { |
| 710 | jsonErrorCode(w, http.StatusTooManyRequests, runCodeRateLimited, runTooManyMsg) |
| 711 | return |
| 712 | } |
| 713 | keys, scope := r.URL.Query()["key"], callerScope(p) |
| 714 | if len(keys) != 1 || !runs.ValidKey(keys[0]) || scope == "" { |
| 715 | jsonErrorCode(w, http.StatusBadRequest, runCodeKeyInvalid, runKeyInvalidMsg) |
| 716 | return |
| 717 | } |
| 718 | if s.cfg.Releases == nil { |
| 719 | jsonErrorCode(w, http.StatusServiceUnavailable, runCodeUnavailable, runReadFailedMsg) |
| 720 | return |
| 721 | } |
| 722 | host := r.PathValue("host") |
| 723 | if !provision.ValidHostShape(host) { |
| 724 | jsonErrorCode(w, http.StatusNotFound, runCodeNotFound, fmt.Sprintf("%s is not an instance of %s", host, app)) |
| 725 | return |
| 726 | } |
| 727 | rec, err := provision.GetRecord(r.Context(), s.cfg.Releases, acct.Sqid, app, host) |
| 728 | switch { |
| 729 | case errors.Is(err, os.ErrNotExist): |
| 730 | jsonErrorCode(w, http.StatusNotFound, runCodeNotFound, fmt.Sprintf("%s is not an instance of %s", host, app)) |
| 731 | return |
| 732 | case err != nil: |
| 733 | jsonErrorCode(w, http.StatusBadGateway, runCodeReadFailed, runReadFailedMsg) |
| 734 | return |
| 735 | } |
| 736 | k, found := rec.RunKeyFor(runs.ScopedKey(scope, keys[0]), time.Now()) |
| 737 | if !found { |
| 738 | jsonErrorCode(w, http.StatusNotFound, runCodeNotFound, runNoKeyMsg(host)) |
| 739 | return |
| 740 | } |
| 741 | writeJSON(w, http.StatusOK, runs.KeyAnswer{RunID: k.RunID}) |
| 742 | } |
| 743 | |
| 744 | // writeRunObjects writes request.json first (the orphan sweep ages a run |
| 745 | // directory by it, Task 5), then the files. The record CAS comes after |
| 746 | // both, so no box ever sees a run whose objects are partial. |
| 747 | func (s *Server) writeRunObjects(ctx context.Context, req runs.Request, files []spooled) error { |
| 748 | body, err := json.Marshal(req) |
| 749 | if err != nil { |
| 750 | return err |
| 751 | } |
| 752 | if _, err := s.cfg.Releases.Put(ctx, runs.RequestKey(req.Account, req.App, req.ID), body, ""); err != nil { |
| 753 | return err |
| 754 | } |
| 755 | for _, f := range files { |
| 756 | fh, err := os.Open(f.path) |
| 757 | if err != nil { |
| 758 | return err |
| 759 | } |
| 760 | _, err = release.PutObjectFrom(ctx, s.cfg.Releases, runs.InputKey(req.Account, req.App, req.ID, f.Name), fh, f.Size, "") |
| 761 | fh.Close() |
| 762 | if err != nil { |
| 763 | return err |
| 764 | } |
| 765 | } |
| 766 | return nil |
| 767 | } |
| 768 | |
| 769 | // runVersion is the release the instance's channel serves, "" when it |
| 770 | // cannot be read: the audit row's version is a record, not a gate. |
| 771 | func (s *Server) runVersion(ctx context.Context, sqid, app, channel string) string { |
| 772 | ptr, err := release.Channel(ctx, s.cfg.Releases, s.cfg.ReleasePub, sqid, app, channel) |
| 773 | if err != nil || ptr == nil { |
| 774 | return "" |
| 775 | } |
| 776 | return ptr.Version |
| 777 | } |
| 778 | |
| 779 | // runStartDetail is the run-start row's detail (run spec §6): the run id, |
| 780 | // argv, the file names and sizes and the version, as JSON. Output never |
| 781 | // goes in the audit log; argv does, and the docs say so. |
| 782 | func runStartDetail(req runs.Request, version string) string { |
| 783 | type file struct { |
| 784 | Name string `json:"name"` |
| 785 | Size int64 `json:"size"` |
| 786 | } |
| 787 | d := struct { |
| 788 | ID string `json:"id"` |
| 789 | Argv []string `json:"argv"` |
| 790 | Files []file `json:"files"` |
| 791 | Version string `json:"version"` |
| 792 | // Requester is a service's run's line (runs.Request.ServiceLine), |
| 793 | // absent for a person's: the row's actor already names them. |
| 794 | Requester string `json:"requester,omitempty"` |
| 795 | }{ID: req.ID, Argv: req.Argv, Version: version, Requester: req.ServiceLine()} |
| 796 | for _, f := range req.Files { |
| 797 | d.Files = append(d.Files, file{f.Name, f.Size}) |
| 798 | } |
| 799 | raw, _ := json.Marshal(d) |
| 800 | return string(raw) |
| 801 | } |
| 802 | |
amadan